Image classification standard update method, program, and image
Contents
1. Yes GENERATE AND DISPLAY HISTOGRAM 309 S310 No Yes UPDATE CLASSIFICATION STANDARD DATA S311 End US 8 625 906 B2 Sheet 6 of 13 Jan 7 2014 U S Patent 906 S06 Amen 3508 A sores sois wawao 02 Z 06 os 56 LL 001 26 lx aig amna A a DII S 7 Lw 1 WE 1 139 Ca jJ D H 09 a TI lg ya MOI A LHOHS NH3LIVd 29 LYOHS NH3 LILVd go 0qv enuey ziz G12 cl Ar Woo HE 202 002 2007 8 51 Wi U S Patent Jan 7 2014 Sheet 7 of 13 US 8 625 906 B2 202 C3 PATTERN SHORT EP WANN 200b 200 Manual C1 PARTICLE FIG 9 201 U S Patent Jan 7 2014 Sheet 8 of 13 US 8 625 906 B2 202 C3 PATTERN SHORT 9 CIRCULARITY 200c 200 C1 PARTICLE FIG 10 a amp m 201 s FLATNESS U S Patent Jan 7 2014 Sheet 9 of 13 US 8 625 906 B2 FIG 11 IMAGE CAPTURING 401 AND ADC PROCESSING READING OF DEFECT IMAGE DATA 402 CONFIRMATION S403 AND RECLASSIFICATION BY UER REFLECTION INTO 404 CLASSIFICATION DATA GENERATION AND DISPLAY S405 OF CONFUSION MATRIX S406 ADEQUATE gt Yes COPYING AS IMAGE 407 FOR ADDITIONAL LEARNING UPDATING OF CLASSIFICATION 408 STANDARD DATA ADC PROCESSING 409 U S Patent Jan 7 2014 Sheet 10 of 13 US 8 625 906 B2 FIG 12 CHECK PROCESS IMAGE COMPARISON PROCESS AN2. 17 121 and only simulation is performed on what a generated confusion matrix 211a will be like if the data of the defect image 722a is added to the classification standard data 121 As a result of simulation defect image data to be added category confusion matrix 211a temporary classification standard data 121 temporary classification data 122 etc may be held by the display processing section 111 in the storage section 12 as history In this case if the user selects a history via a history selection pull down menu 734 then the display processing section 111 displays the state at that time defect image data to be added category confusion matrix 211a temporary classification standard data 121 temporary classification data 122 etc on the check screen 500 Referring to the confusion matrix 211a updated as a result of the simulation the user determines whether or not it is appropriate to add the defect image 722a displayed in the simulate execution area 730 to the classification standard data 121 If determined to be appropriate a copy button 701 is selected and entered via the input section 13 That is the display processing section 111 determines whether or not the copy button 701 has been selected and entered and thereby determines whether or not to copy the defect image 722a to the classification standard data 121 S508 in FIG 12 The process in S508 corresponds to S406 in FIG 11 If determined not to copy S508 No
3. In recognizing a defect portion and extracting a characteristic amount the characteristic amount extraction section 114 compares nor mal image data with defect image data and then extracts the defect portion In this state in order not to erroneously extract a noise on an image as a defect portion the characteristic amount extraction section 114 removes a noise at a certain level and in order not to erroneously extract a portion that appears bright due to light as a defect portion the character istic amount extraction section 114 adjusts the detection sen sitivity The process in S203 is a technology described in JP 2007 198968 A and others and accordingly description will be omitted Then after extracting defect potions the characteristic amount extraction section 114 extracts the characteristic amounts of these defect patterns and determines how to weight the extracted characteristic amounts when performing ADC S204 Extracting the characteristic amounts means calculating physical characteristics having been set and quan tified in advance for each defect image data As the physical characteristics flatness brightness circularity size as described above and in addition height shape color texture defect background and the like can be considered The char acteristic amount extraction section 114 stores the extracted characteristic amounts in the storage section 12 as standard characteristic amount data 124 Subsequent
4. and displays bars 412 at positions representing the respective characteristic amounts corre sponding to the values obtained by the display processing section 111 In a separation degree list display area 431 the above described separation degrees are listed in the descending order Radio buttons 421 422 are used to indicate characteristic amounts which are currently used in performing classifica tion by ADC processing In the example in FIG 10 flat ness brightness and circularity for which the radio buttons 421 422 are ON are characteristic amounts which are currently used in the ADC process By switching On Off of the radio buttons 421 422 the user can set effective ness ineffectiveness of characteristic amounts to be used in the ADC process For example when the user determines that brightness and circularity are ineffective characteristic amounts the user can set the usage of these characteristic amounts in the ADC process to be ineffective by selecting and entering the corresponding radio buttons 412 422 Further on the contrary when the user determines that size is valid characteristic amount the user can set the usage of this char acteristic amount inthe ADC process to be effective by select ing and entering the corresponding radio buttons 421 422 By determining whether or not a move button 332 FIG 9 has been selected and entered the display processing section 111 d
5. display processing section 111 thereafter determines whether or not an execution button 732 has been selected and thereby determines whether or not to execute simulation S506 in FIG 12 As a result of S506 if simulation is not to be executed S506 No the display processing section 111 proceeds the process to S508 As a result of S506 if simulation is to be executed S506 Yes the characteristic amount extraction section 114 performs simulation on what a generated confusion matrix 211a will be like if the defect image 722a displayed in the simulate execution area 730 is added to the classification standard data 121 and displays a result in the display section 14 S507 in FIG 12 More specifically the automatic defect classification section 113 uses temporary classification stan dard data 121 to which the defect image 722a is added to perform reclassification of the defect image data group 123 with the newly added defect image data and thereby creates a temporary classification data 122 Then the display pro cessing section 111 generates a confusion matrix 211a from the temporary classification data 122 by a method similar to the above described procedure and then displays this confu sion matrix 211a in the confusion matrix display area 201a thereby updating the confusion matrix 211a At the moment of S507 the data of the defect image 722a has not actually been added to the classification standard data US 8 625 906 B2
6. 202a One defect image 512a among the defect images 511 is copied by drag and drop into an object image display area 330a and displayed as a defect image 5126 A move button 332a in FIG 13 is a button for moving the data of the selected defect image 5125 to another category in the same classification data 122 FIG 14 is a diagram showing an example of a check screen characteristic amount comparison according to the present embodiment In the check screen 5005 500 the symbols 201a 202a 203a 205a 206a 211a 214a 400a 402a 403a 411a 412a 421a 422a and 431a are similar to the symbols 201 202 203 205 206 211 214 400 402 403 411 412 421 422 and 431 in FIG 10 except that the symbols in the check screen 5005 500 are created based on classification data 122 while the symbols in FIG 10 are created based on classification standard data 121 and description will be accordingly omitted In the example in FIG 14 one symbol 601a among defect images displayed in the defect image list display area 202a is copied by drag and drop into a characteristic amount display area 400a and displayed as a defect image 6015 Returning to FIG 12 subsequent to S501 the display processing section 111 determines whether or not analysis 2 tab in FIG 13 or 14 has been selected and entered whether or not selected S502 As a result of S502 if analysis 2 tab has not been selected S502 No the display processing s
7. 309535 A 20 35 40 45 60 65 2 DISCLOSURE OF INVENTION Technical Problem In the technology described in Patent Document 1 a clas sification standard for automatic defect classification using a neural network learns based on visual classification by human eyes Consequently if an inspector makes a classification error there may be a contradiction in the classification stan dard resulting in a drop in the classification performance for automatic defect classification That is learning is performed via a neural network based on a classification standard as a result of visual classification which causes problems that a classification standard with errors is created and a learning result outputs a result with errors Furthermore in some cases a desirable classification can not be performed since a defect having one type of charac teristic may have another type of characteristic if a semicon ductor manufacturing process varies after a currently effective classification standard was created That is it is necessary to perform learning each time when a defect of a type that has not been registered in a classification standard is detected The technology described in Patent Document 2 does not include a technology for updating a once created classifica tion standard and thereby improving the classification stan dard The present invention has been developed in view of the foregoing background and an object of th
8. As shown in FIG 4 the classification standard data 121 and the classification data 122 have a field for categories categorized by user a field for categories categorized by ADC and a field for the names of defect image data The categories by user refer to categories classified by a user in 202 in FIG 6 or later described S403 in FIG 11 The categories by ADC are those classified by ADC processing in later described S205 in FIG 6 or later described S401 in FIG 11 In the example shown in FIG 4 it is shown that defect image data having been determined to be C1 par ticle by a user and determined to be C1 particle also by ADC processing are A1 jpg A2 jpg A3 jpg Further it is shown that there is no defect image data that has been determined to be C1 particle by the user and determined to be C2 scratch by ADC processing Further it is shown that defect image data that has been determined to be C2 scratch by the user and determined to be C1 particle by ADC processing is A10 jpg Herein C1 C2 and the like are identification numbers assigned to categories In the present embodiment C1 rep resents particle C2 represents scratch C3 represents pattern short C4 represents pattern open and C5 repre sents no defect In the present embodiment these identifica tion numbers will be used as appropriate instead of category names In addition to t
9. MODES FOR CARRYING OUT THE INVENTION Modes for carrying out the present invention referred to as embodiments will be described below referring to the drawings as appropriate In the preset embodiment an example will be described where an image classification device is applied to a semiconductor wafer manufacturing system Z however the invention is not limited thereto and is applicable to systems that perform defect inspection using images such as image inspection of foods FIG 1 is a diagram showing an example of the configura tion of a semiconductor wafer manufacturing system in the present embodiment Manufacturing devices 4 for manufacturing semiconduc tor wafers are normally set in a clean room 7 where clean environment is maintained Further semiconductor wafers manufactured on the line of the manufacturing devices 4 are subjected to a conduction test by a probe inspection device 5 Inthe clean room 7 there are provided appearance inspec tion devices 2 for detecting appearance defects of produced wafers and review devices 1 image classification device for observation of the appearance defects in another word reviewing based on data from the appearance inspection devices 2 Further outside the clean room 7 provided is a data processing device 3 that performs processing of image data having been obtained by the appearance inspection devices 2 or the review devices 1 The appearance inspection device 2 the review dev
10. The appearance inspec tion device 2 captures the images of the appearance of the semiconductor wafers and if an appearance defect is detected the appearance inspection device 2 obtains the coor dinates of the position of the detected appearance defect as defect data and transmits the obtained defect data to the data processing device 3 S101 Because the amount of defect data that the appearance inspection device 2 outputs is huge the data processing device 3 transmits defect data having been filtered by a filter function to an optical review devices 1a or an SEM review devices 15 via the communication line 6 S102 S103 The filtering function includes for example extraction of a pre determined number of pieces of detect information at ran dom The optical review devices 1a or the SEM review devices 1b capture the images at the coordinate positions according to the transmitted defect information by using an optical micro scope or an electronic microscope and obtain the images of the semiconductor wafers at the portions of the detected defects defect image data of the semiconductor wafers The optical review devices la and the SEM review devices 1b perform classification of defects by using an ADC function implemented therein Information on results of such defect classification is transmitted as ADR ADC information via the communication line 6 to the data processing device 3 S104 S105 In the present embodiment described is a
11. amount data 200 200a 2005 200c self check screen 211 211a confusion matrix association information between categories Z semiconductor wafer manufacturing system The invention claimed is 1 A method for updating an image classification standard by an image classification device automatically classifying image data comprising the steps of storing in a storage section of the image classification device classification standard data which includes information on image data used as a standard for automatically classi fying the image data and classification data which includes information on image data of newly input image data automatically classified using the classification standard data and updating the classification standard data by adding infor mation on image data selected by a user from the image data included in the classification data into the classifi cation standard data US 8 625 906 B2 19 when an instruction is input via an input section of the image classification device to add the information on image data selected by the user from the image data included in the classification data into the classification standard data 2 The method for updating an image classification stan dard according to claim 1 further comprising the step carried out by the image classification device of displaying an image corresponding to the information on image data in the classification standard data and an image co
12. and FIGS 8 to 10 FIG 7 is a flowchart showing the process for self checking in the present embodiment The process shown in FIG 7 is a process corresponding to S207 to 209 in FIG 6 First the display processing section 111 generates a con fusion matrix 211 association information between catego ries and displays a self check screen 200a FIG 8 includ ing the generated confusion matrix 211 S301 FIG 8 is a diagram showing an example of a self check screen initial screen in the present embodiment The self check screen 200a 200 includes a confusion matrix display area 201 a defect image list display area 202 and a defect image confirmation area 203 which are dis played in the same window In the confusion matrix display area 201 a confusion matrix 211 is displayed The confusion matrix 211 is a table that indicates the numbers of images in the respective catego ries according to the classification by the user represented by 20 25 30 35 40 45 50 55 60 65 8 Manual in FIG 8 and the respective categories according to the classification by ADC represented by ADC in FIG 8 Symbols C1 to C5 are as described above category iden tification numbers wherein C1 represents particle C2 represents scratch C3 represents pattern short C4 rep resents pattern open and C5 represents no defect In the example in FIG 8 the vertical axis repr
13. counts the number of defect image data with the same characteristic amount on each indi vidual characteristic For example in the example in FIG 5 assuming that A1 jpg and A2 jpg are objects of process ing the display processing section 111 first refers to record A1 jpg and counts flatness 50 by 1 and counts bright ness 60 by 1 The display processing section 111 likewise counts circularity and size of this record as well Then the display processing section 111 refers to record A2 jpg and counts flatness 40 by 1 and counts bright ness 60 by 1 brightness 60 thereby becomes 2 The display processing section 111 likewise counts circularity and size of this record as well The display processing section 111 performs this process on all the obtained names of defect image data thereafter further performs the same process also on the category selected via the characteristic amount selection pull down menu 403 and calculates the histograms of characteristic amounts for the respective characteristics Further bars 412 in the characteristic amount distribution display area 411 represent the respective values of the char acteristic amounts of the defect image 401b The display processing section 111 obtains the values of characteristic amounts from the standard characteristic amount data 124 in FIG 5 by using the name of the defect image data of the defect image 4015 as a key
14. egories Further symbol 213 represents the ratio of the num ber of defect image data in which classification by user conforms with classification by ADC to the total number of all defect image data When a matrix button 214 in the input section 13 is selected and entered the display processing section 111 counts the numbers of defect image data in the respective categories referring to the classification standard data 121 shown in FIG 4 and displays the counted numbers in the confusion matrix 211 Herein the display processing section 111 monitors whether or not a cell of the confusion matrix 211 has been selected S302 in FIG 7 When no cell is selected S302 No the display process ing section 111 forwards the process to S304 In FIG 7 and in FIG 12 if the process moves forward to the step No Sm when no selection input is made in step No Sn it means that the processing section 11 determines nothing and executes the process of the step Sm This is because the steps in FIG 7 and FIG 12 are actually image processing steps and each step is executed when an instruction is input regardless ofthe order of the steps shown in the drawings If the user selects one of the cells in the confusion matrix 211 S302 Yes in FIG 7 then a defect image correspond ing to the selected cell is displayed in the defect image list display area 202 S303 in FIG 7 For example if a cell 215 whose category is C3 pattern short ac
15. for example a case where the image of a defect is not correctly captured image for creating the classification standard data 121 then the delete button 206 is selected and entered via the input section 13 and thereupon the display processing section 111 can delete the name of this defect image data from the classifica tion standard data 121 When the user has moved a defect image data name to be used for learning to another category by selecting and enter ing the move button 332 FIG 9 when the user has deleted defect image data from the classification standard data 121 by selecting and entering the delete button 206 or when the user has switched effectiveness ineffectiveness of a characteristic amount to be used for ADC processing it is possible to update the confusion matrix 211 by selecting and entering the matrix button 214 each time Further the classification standard data 121 in this state can be overwritten or saved with another name by selecting and entering the save button 205 The classification standard data 121 having been created in such a manner is used as a classification standard for ADC processing in the review device 1 and the review device 1 automatically classifies defects on semiconductor wafers and transfers identification numbers of categories in respective results to the data processing device 3 On the other hand the defect image data group 123 determined to be defect images by ADC is stored in the storage s
16. in FIG 10 the characteristic amount distribution of C1 particle is displayed by a hollow histogram and the characteristic amount distribution of C2 scratch is displayed by a hatched histogram The percentage displayed in the right top portion of a characteristic amount distribution display area 411 represents the separation degree that is the ratio of the non overlapped distribution portion to the entire characteristic distribution in two categories That is the percentage represents the ratio of the histograms which are not black solid to the entire characteristic amount distri US 8 625 906 B2 11 bution It is shown that if the separation degree is larger the difference is the greater between the characteristic amount distributions of two categories A histogram representing characteristic amount distribu tion is created in the following procedure First the display processing section 111 searches a category by the user in the classification standard data 121 in FIG 4 by using the cat egory name selected via the characteristic amount selection pull down menu 402 as a key and obtains the names of defect image data included in all corresponding records Then the display processing section 111 searches in the standard characteristic amount data 124 in FIG 5 by using the obtained defect image data names as a key refers to the values ofthe respective characteristic amounts corresponding to the defect image data names and
17. person Accordingly classification of detects of observation objects is biased depending on an inspector In order to solve this problem in recent years technologies for ADR automatic defect review and ADC automatic defect classification in which a device automatically performs determination of the size the shape the kind and the like of a defect using an image processing technology have come to be introduced For example in order to observe in another word review inspected parts for example patterns formed on wafers by using an SEM scanning electron microscopy review device a system that efficiently performs a task while reducing the workload of a user is proposed As a method for extracting information included in an inspection image as characteristic amounts and performing automatic classification based on the characteristic amounts a method using a neural network is disclosed for example refer to Patent Document 1 Further in order to reduce effects of inappropriate characteristic amounts on the classi fication performance in learning weighting of respective characteristic amounts for creating a classification standard for performing automatic classification a method that auto matically selects characteristic amounts that are effective for classification is disclosed for example refer to Patent Docu ment 2 PRIOR ART DOCUMENTS Patent Documents Patent Document 1 JP H08 021803 A Patent Document 2 JP 2005
18. the processing sec tion 11 terminates the process If determined to copy S508 Yes the display processing section 111 copies the corresponding defect image 722a to the classification standard data 121 S509 in FIG 12 Actu ally the display processing section 111 copies a defect image data name corresponding to the defect image 722a displayed in the simulate execution area 730 to the classification stan dard data 121 The process in S509 is corresponding to S407 in FIG 11 Subsequently the display processing section 111 deter mines whether or not an application button 733 has been selected and entered and thereby determines whether or not to apply copy S510 in FIG 12 If determined not to apply copy S510 No the process ing section 11 terminates the process A case of not applying copy refers to a case that the application button 733 has not been selected and entered for a certain time after application of the copy button 701 a case that a delete button 206a has been selected and entered a case that the check screen 500 has been closed and other cases In a case of applying copy S510 Yes the display pro cessing section 111 additionally registers the data name ofthe defect image 722a displayed in the simulate execution area 730 into the classification standard data 121 and updates the classification standard data 121 fixes the classification stan dard data 121 511 in FIG 12 Alternatively by retrieving t
19. to S204 the automatic defect classification section 113 performs ADC processing S205 by using the characteristic amounts extracted in S204 and the determined weight and classifies the defect image data by ADC The US 8 625 906 B2 7 ADC process is a technology described in JP H09 101970 and the like and description in detail will be omitted Then the automatic defect classification section 113 reflects a result of the ADC process into the classification standard data 121 FIG 4 S206 The automatic defect classification section 113 registers the result of the classifi cation in S205 into the column which was blank at the step of S202 as a category by ADC More specifically the automatic defect classification section 113 further classifies the classi fication made at the step of S202 for more details For example it will be assumed that A20 jpg A21 jpg and A22 jpg not shown in the drawings had been determined to be C3 pattern short in S202 classification by the user however in S205 classification by ADC A20 jpg has been determined to be C1 particle and A21 jpg and A22 jpg have been determined to be C3 pattern short In this case A20 jpg is classified to be C3 pattern short as category determined by the user and to be C1 particle as category by ADC while A21 jpg and A22 jpg are classi fiedto be C3 pattern short as category by the user and to be C3 pat
20. 007 Kanda etal 382 149 FOREIGN PATENT DOCUMENTS JP 08 021803 A 1 1996 JP 2001 156135 A 6 2001 JP 2005 185560 A 7 2005 JP 2005 309535 A 11 2005 OTHER PUBLICATIONS Written Opinion of the International Searching Authority PCT ISA 237 issued in PCT JP2009 071774 with English Translation dated Mar 23 2010 International Search Report issued in PCT JP2009 071774 dated Mar 23 2010 with English Translation cited by examiner Primary Examiner Daniel Mariam 74 Attorney Agent or Firm McDermott Will amp Emery LLP 57 ABSTRACT The objective is to improve a classification standard Classi fication standard data in which is registered image data infor mation that is the standard when image data is classified and classification data in which is registered image data informa tion that is the result when newly input image data is classified using the classification standard data are stored in a storage unit An image classification device is characterized in that when any image data information of the image data that is registered in the classification data is selected by means of an input unit and an instruction to additionally register the selected image data information in the classification standard data is input by means of the input unit the selected image data information is additionally registered in the classifica tion standard data 18 Claims 13 Drawing Sheets U S P
21. D CHARACTERISTIC 9901 AMOUNT COMPARISON PROCESS 502 No ANALYSIS 2 TAB SELECTED Yes 550 CATEGORY SELECTED Yes GENERATION AND DISPLAY OF HISTOGRAM S904 905 No DEFECT IMAGE SELECTED Yes S506 Yes SIMULATION EXECUTION AND RESULT DISPLAY 907 508 ne es COPYING OF CORRESPONDING DEFECT IMAGE S909 S510 x es UPDATING OF CLASSIFICATION STANDARD DATA S511 US 8 625 906 B2 Sheet 11 of 13 Jan 7 2014 U S Patent e90z 4171313134 LOG Ad09 e602 JAVYS Z 4 44 YA Baye ejpe l A 3101 Vd 3A0W A 19075 wania ions nyawad eo egoz Loze SUL BOGE ELEC e YA NSH 20 10 CO 149 h 1334409 1HOHS NILI Yo 19 3lonuvd 10 av enue 2202 e 006 2006 21514 i eiz XIHLYW US 8 625 906 B2 Sheet 12 of 13 Jan 7 2014 U S Patent 2902 106 Ad09 LHOHS NILI Vd Yo al ody TIOLLEVd 19 enue 006 4006 vi DI T BLUZ eb1Z XIHIVN US 8 625 906 B2 Sheet 13 of 13 Jan 7 2014 U S Patent 313730 Les A AOL VA 10 A 1 3581 l xMo1SH VeL 31V WIS JHOHS NH3llVd O TIOILHVd 19 2qv enuey ezag 006 9006 SG UDIH 2102 NOLLVZHO CUVONVLS NOUVSTISSWI JO SILL LY OQUIVW US 8 625 906 B2 1 IMAGE CLASSIFICATION STANDARD UPDATE METHOD PROGRAM AND IMAGE CLASSIFICATION DEVICE RELATED APPLICATIONS This applicatio
22. a United States Patent Isomae et al US008625906B2 US 8 625 906 B2 Jan 7 2014 10 Patent No 45 Date of Patent 54 IMAGE CLASSIFICATION STANDARD UPDATE METHOD PROGRAM AND IMAGE CLASSIFICATION DEVICE 75 Inventors Yuya Isomae Hitachinaka JP Fumiaki Endo Hitachinaka JP Tomohiro Funakoshi Hitachinaka JP Junko Konishi Hitachinaka JP Tsunehiro Sakai Mito JP 73 Assignee Hitachi High Technologies Corporation Tokyo JP Notice Subject to any disclaimer the term of this patent is extended or adjusted under 35 U S C 154 b by 148 days 21 Appl No 13 142 812 22 PCT Filed Dec 28 2009 86 PCT No 371 c 1 2 4 Date PCT JP2009 071774 Jun 29 2011 87 PCT Pub No WO2010 076882 PCT Pub Date Jul 8 2010 65 Prior Publication Data US 2011 0274362 Al Nov 10 2011 30 Foreign Application Priority Data Dec 29 2008 JP sse 2008 335779 51 Int CI G06K 9 62 2006 01 G06K 9 54 2006 01 52 U S CI USPC itti 382 224 382 305 58 Field of Classification Search USPC teet dns 382 141 149 224 294 305 See application file for complete search history 56 References Cited U S PATENT DOCUMENTS 5 526 258 A 6 1996 Bacus ween 382 129 7 113 628 Bl 9 2006 Obara et al 7 634 141 B2 12 2009 Hayashi etal 382 224 8 176 050 B2 5 2012 Inakoshiet al e 707 737 2005 0152592 Al 7 2005 Kasai 2007 0025611 Al 2 2
23. age data The data obtaining section 115 has a function of obtaining data from the transmitting receiv ing section 15 The processing section 11 and the respective sections 111 to 115 are realized by executing a program stored in a ROM read only memory not shown or a HD hard disk not shown is loaded into a RAM random access memory not shown by a CPU central processing unit not shown The storage section 12 stores classification standard data 121 classification data 122 a defect image data group 123 standard characteristic amount data 124 and characteristic amount data 125 The classification standard data 121 the classification data 122 the standard characteristic amount data 124 and the characteristic amount data 125 will be described later referring to FIGS 4 and 5 The defect image data group 123 are defect image data captured by the review device 1 Various Data FIG 4 is a diagram showing an example of classification standard data and classification data Herein the classifica tion standard data 121 are data created by a process which will be described later with reference to FIG 6 and the classification data 122 are data created by a process which will be described later with reference to FIG 11 Although the classification standard data 121 and the classification data 122 are different in terms of stored data the formats are similar to each other and will be commonly described below referring to FIG 4
24. assification data 122 are superimposed with each other the characteristic amount type selection pull down menu 712 may display the characteristic amount types in ascending order of lower agreement ratio i e in descending order of higher separation degree Then the display processing section 111 determines whether or not a defect image displayed in the defect image display area 720 has been dragged and dropped into the simulate execution area 730 via the input section 13 and thereby determines whether or not a defect image has been selected S505 in FIG 12 As a result of S505 if a defect image has not been selected S505 No the display processing section 111 proceeds the process to S508 in FIG 12 As a result of 505 if a defect image has been selected S505 Yes the display processing section 111 displays the dragged and dropped defect image in the simulate execution area 730 in FIG 15 In the example in FIG 15 among the defect images 721 722 displayed in the defect image display area 720 the defect image 722 has been dragged and dropped selected and is displayedin the simulate execution area 730 as a defect image 722a however it is possible to select a plurality of defect images Then the user selects as to which category category by user in the classification standard data 121 the defect image 722a displayed in the simulate execution area 730 is to be moved via a category selection pull down menu 731 and the
25. atent Jan 7 2014 Sheet 1 of 13 US 8 625 906 B2 FIG 2 U S Patent Jan 7 2014 Sheet 2 of 13 US 8 625 906 B2 FIG 3 U S Patent Jan 7 2014 Sheet 3 of 13 US 8 625 906 B2 FIG 4 121 CLASSIFICATION STANDARD DATA 122 CLASSIFICATION DATA CATEGORY CATEGORY BY USER BY ADC DEFECT IMAGE DATA NAME C1 PARTICLE Ci PARTICLE Aljpg A2jpg A3 jpg CT PARTICLE C2 SCRATCH C2 SCRATCH FIG 5 124 STANDARD CHARACTERISTIC AMOUNT DATA 125 CHARACTERISTIC AMOUNT DATA DEFECT IMAGE DATA NAME Al jpg A2jpg U S Patent Jan 7 2014 Sheet 4 of 13 US 8 625 906 B2 FIG 6 READING OF DEFECT IMAGE DATA S201 FOR CREATING CLASSIFICATION STANDARD CLASSIFICATION BY USER 202 AUTOMATIC ADJUSTMENT OF DEFECT S203 RECOGNITION PARAMETERS EXTRACTION OF CHARACTERISTIC AMOUNTS S204 AND DETERMINATION OF WEIGHTING ADC PROCESSING 9205 REFLECTION INTO CLASSIFICATION STANDARD DATA 9206 DISPLAY OF SELF CHECK SCREEN 207 SELF CHECK S208 SELF CHECKED ADEQUATELY Yes End U S Patent Jan 7 2014 Sheet 5 of 13 US 8 625 906 B2 FIG 7 SELF CHECK PROCESS GENERATE AND DISPLAY CONFUSION MATRIX 301 S302 No Yes DISPLAY DEFECT IMAGE LIST DISPLAY AREA 9303 S304 No DEFECT IMAGE SELECTED Yes s305 CATEGORY SELECTED Yes DISPLAY COMPARISON IMAGE IN CORRESPONDING CATEGORY 306 5307 No ANALYSIS TAB SELECTED Yes 6300 CATEGORY SELECTED
26. cording to classification by user Manual and is also C3 pattern short according to Classification by ADC is selected and entered via the input section 13 then the display processing section 111 obtains from the classification stan dard data 121 in FIG 4 the names of defect image data stored in the records of both categories by the user and ADC C3 pattern short Then the display processing section 111 US 8 625 906 B2 9 obtains from the defect image data group 123 FIG 3 in the storage section 12 image data corresponding to the obtained name of defect image data and displays the obtained image data in the defect image list display area 202 Incidentally in FIG 8 as 12 is described in cell 215 the number of images displayed in the defect image list display area 202 is also 12 The usercanreferto 12 images by moving the slide bar in the defect image list display area 202 shown in FIG 8 In the defect image confirmation area 203 nothing is dis played at the step of S303 A save button 205 and a delete button 206 will be described later In order to create accurate classification standard data 121 and thereby improve the accuracy of classification by ADC it is necessary to improve the purity ratio and the correct result ratio in the confusion matrix 211 A method for updating the classification standard data 121 for improving the purity ratio and the correct result ratio will be described below referring t
27. data from the defect image data group 123 in the storage section 12 based on the information on update date and time and the like S402 At this moment the display processing section 111 also reads the classification data 122 in FIG 4 and displays a result of classification by ADC in which a result of ADC processing and defect images are associated with each other on the display section 14 The user refers to the displayed classification result and confirms whether the classification has been correctly made by the ADC processing and ifthere are errors in classification the user performs reclassification into appropriate categories S403 The input processing sec tion 112 reflects a result of the reclassification into the clas sification data 122 in FIG 4 S404 The order of reflection is similar to that in S206 in FIG 6 except that the order clas sification by user classification by ADC has changed to classification by ADC classification by user and detailed description will be accordingly omitted Then the display processing section 111 generates a con fusion matrix 211a FIG 13 based on the classification data 122 and displays a check screen 500 FIG 13 including this confusion matrix 211a S405 The procedure of generating the confusion matrix 211a based on the classification data 122 is similar to the procedure described above with refer ence to FIG 8 and detailed description will be accordingly omi
28. e characteristic amount selection pull down menu 402 403 S308 Yes in FIG 7 then the display processing section 111 generates a histogram representing the distribution of the respective characteristic amounts in the selected category and displays the generated histogram in a characteristic amount distribution display area 411 S309 in FIG 7 A category selected via the character istic amount selection pull down menu 402 403 is a category classified by user In graphs displayed in the characteristic amount distribu tion display area 411 the horizontal axis represents the values ofrespective characteristic amounts and the vertical axis rep resents the numbers of defect image data with the respective values In the characteristic amount distribution display area 411 the characteristic amount distribution of a category selected via the characteristic amount selection pull down menu 402 is displayed as a hollow histogram and the char acteristic amount distribution of a category selected via the characteristic amount selection pull down menu 403 is dis played as a hatched histogram Further a portion where the characteristic amount distribution of a category selected via the characteristic amount selection pull down menu 402 and the characteristic amount distribution of a category selected via the characteristic amount selection pull down menu 403 overlap with each other is displayed as a black solid histo gram In the example shown
29. e invention is to improve a classification standard Technical Solution In order to solve the above described problem the present invention is a method for updating an image classification standard by using an image classification device for classify ing image data wherein a storage section stores classification standard data in which information on image data to be a standard for classifying image data is registered and classi fication data in which information on image data as a result of classification of newly input image data using the classifica tion standard data is registered and wherein when informa tion on arbitrary image data is selected via an input section from the image data registered in the classification data and an instruction is input via the input section to additionally register the selected information on image data into the clas sification standard data the image classification device addi tionally registers the selected information on image data into the classification standard data Other solutions will be described later in embodiments Advantageous Effects The present invention can improve a classification stan dard BRIEF DESCRIPTION OF THE DRAWINGS FIG 1 is a diagram showing an example of the configura tion of a semiconductor wafer manufacturing system in the present embodiment FIG 2 is a diagram showing the flow of data in a semicon ductor wafer manufacturing system in the present embod
30. ection 111 proceeds the process to S505 As a result of S502 if analysis 2 tab has been selected S502 Yes the display processing section 111 displays a check screen 500c 500 shown in FIG 15 FIG 15 is a diagram showing an example ofa check screen simulation according to the present embodiment In FIG 15 elements similar to those in FIGS 13 and 14 are given with the same symbols and description will be omitted In the check screen 500c 500 a simulate area 700 is displayed in the defect image confirmation area 203a The simulate area 700 includes a characteristic amount display area 710 a defect image display area 720 and a simulate execution area 730 The display processing section 111 determines whether or not a category has been selected via a category selection pull down menu 711 in the characteristic amount display area 710 S503 in FIG 12 If not selected S503 No the dis play processing section 111 proceeds the process to S505 in FIG 12 If a category has been selected via the category selection pull down menu 711 S503 Yes and further a characteristic US 8 625 906 B2 15 amount type is selected via a characteristic amount type selection pull down menu 712 then the display processing section 111 generates a histogram representing characteristic amount distribution based on the classification standard data 121 and the classification data 122 and displays the generated histogram in a character
31. ection 12 of the review device 1 for respective wafers Through the process up to here a classification standard data 121 has been created and adjusted for classification of defect images by ADC A process to be performed when new defect image data is transmitted to the review device 1 after the classification standard data 121 is created will be described below referring to FIGS 10 to 14 The transmitted new defect image data is stored in the defect image data group 123 in the storage section 12 FIG 11 is a flowchart showing the process executed when defect image data is obtained anew First when the review device 1 captures new defect image data after the process in FIG 6 the characteristic amount extraction section 114 automatically adjusts the defect rec ognition parameters thereafter extracts characteristic amounts and the automatic defect classification section 113 performs ADC processing of the captured defect image data S401 and thereby classifies defect images A result of ADC processing is registered in the classification data 122 shownin US 8 625 906 B2 13 FIG 4 Automatic adjustment of defect recognition param eters and extraction of characteristic amounts are similar to the processes in S203 S204 in FIG 6 and description will be accordingly omitted Then when an instruction of classification by the user is entered via the input section 13 the input processing section 112 reads the newly input defect image
32. egory according to classification by user If the user wishes to compare images in a category with images in another category the user selects said another category by using the category selection pull down menu 341 via the input section 13 More specifically the display processing section 111 deter mines whether or not a category has been selected by the category selection pull down menu 341 S305 in FIG 7 If no category has been selected S305 No then the display processing section 111 proceeds the process to S307 If a category is selected by the category selection pull down menu 341 S305 Yes in FIG 7 then the display processing section 111 refers to the classification standard data 121 in FIG 4 with a key of the selected category and thereby obtains the names of defect image data stored in all records corresponding to the selected category with classifi cation by user Then the display processing section 111 obtains defect image data corresponding to the obtained names of defect image data from the defect image data group 123 FIG 3 in the storage section 12 and displays the 20 25 30 35 40 45 50 55 60 65 10 obtained defect image data in the comparison image display area 340 as comparison images in the corresponding category S306 in FIG 7 A characteristic amount display area 400 FIG 10 is hid den at the back of the image comparison area 320 and the characteristic amount disp
33. en 500a 500 shown in FIG 13 and a check screen 5005 500 shown in FIG 14 and performs an image comparison process characteristic amount comparison process S501 S501 is a process similar to the process shown in FIG 6 except that classification data 122 is used instead of classifi cation standard data 121 and detailed description will be accordingly omitted 20 25 30 35 40 45 50 55 65 14 FIG 13 is a diagram showing an example ofa check screen image comparison according to the present embodiment In the check screen 500a 500 the symbols 201a 202a 203a 205a 206a 211a 214a 320a 330a 331a 332a 340a 341a and 342a are similar to the symbols 201 202 203 205 206 211 214 320 330 331 332 340 341 and 342 in FIG 9 except that the symbols in the check screen 500a 500 are created based on classification data 122 while the symbols in FIG 9 are created based on classification standard data 121 and description will be accordingly omitted Further although analysis tab in FIG 9 is replaced by analysis 1 tab in the check screen 500a 500 these functions are similar A copy button 501 will be described later Inthe example in FIG 13 when a cell 502 in the confusion matrix 211a created based on the classification data 122 is selected via the input section 13 the display processing sec tion 111 displays corresponding defect images 511 in the defect image list display area
34. esents cat egories according to classification by the user Manual and the horizontal axis represents categories according to classification by ADC For example regarding line 1 of the confusion matrix 211 38 34 4 defect image data are determined to be C1 par ticle according to the classification Manual by the user while 34 defect image data among them are determined to be C1 particle and 4 defect image data among them are deter mined to be C3 pattern short according to the classification by ADC A correct result ratio is the ratio of a classification result by ADC that agrees with a classification result by user to the classification result by the user For example the cor rect result ratio of line 1 is 34 38x100289 Yo Likewise regarding row 1 in the confusion matrix 211 in FIG 8 it is recognized that 37 344142 defect image data are determined to be C1 particle by ADC while one defect image data is determined to be C2 scratch and two defect image data are determined to be C5 no defect by the user A purity ratio is the agreement ratio of a result of classification by user to a result of classification by ADC For example the purity ratio of row 1 is 34 37x100292 Yo The elements having a reference symbol 212 the central oblique line in the confusion matrix 211 represent the num bers of defect image data in which classification by user conforms with classification by ADC for the respective cat
35. etermines whether or not to move corresponding defect image data from the current category to another category 20 25 30 35 40 45 50 55 60 65 12 S310 in FIG 7 and if not to move S310 No the pro cessing section 11 terminates the process Herein not to move refers to a case for example where a delete button 206 is selected and entered or the user closes the self check screen 200 in a state that the move button 332 has not been selected and entered Incidentally S310 corresponds to the process in S209 in FIG 6 If the user intends to move the current defect image from the current category to another category the user selects a moving destination category via a moving destination cat egory selection pull down menu 331 and selects and enters the move button 332 FIG 9 S310 Yes in FIG 7 and thereupon the display processing section 111 moves the data name ofthe defect image 3115 displayed in the object image display area 330 FIG 9 to the selected moving destination category More specifically the display processing section 111 moves the name of the defect image data corresponding to the defect image 3114 to a record of the selected moving destination category selected by the user in the classification standard data 121 and thereby updates the classification stan dard data 121 S311 in FIG 7 Further if a defect image displayed in the object image display area 330 is an inappropriate
36. ge of a semiconductor wafer 8 A non transitory computer readable medium storing a program for executing on a computer the method for updat ing an image classification standard according to claim 1 9 A method for updating an image classification standard by an image classification device automatically classifying image data comprising the steps of storing in a storage section of the image classification device classification standard data which includes infor mation on image data used as a standard for automati cally classifying image data and the image data classified in respective categories and updating the classification standard data by moving image data selected by a user from the image data in the clas sification standard data from a category in which the image data selected by the user is classified to a different category in the classification standard data when an instruction is inputted via an input section of the image classification device to move the image data selected by the user from the category in which the image data selected by the user is classified to the dif ferent category in the classification standard data 10 The method for updating an image classification stan dard according to claim 9 further comprising the step carried out by the image classification device of displaying image data in the different category in the clas sification standard data in the same window with the image data
37. he history of simulation via the history selection pull down menu 734 and selecting and entering the application button 733 a state corresponding to the history of simulation may be reflected and fixed into the classification standard data 121 Incidentally the copy button 701 may be omitted and the process in S508 may be omitted In this case if the application button 733 15 selected and entered then the defect image data name is copied into the classification standard data 121 and simultaneously subjected to applying processing In the present invention a histogram showing characteris tic amount distribution is displayed however the invention is mot limited thereto For example it is also possible to display characteristic amount distribution by a scatter diagram or the like 20 25 30 35 40 45 50 55 60 65 18 According to the present invention the classification stan dard data 121 used for ADC can be created or updated by comparing respective defect images comparing characteris tic amount distributions moving image data names in the classification standard data 121 as a result of these compari sons and copying defect image data names in classification data 122 into the classification standard data 121 As a result it is possible to prevent the classification performance of a classification standard from being reduced due to inappropri ate learning and improve the classification accuracy of ADC F
38. hese it is possible to freely set catego ries such as to be critical foreign matter non critical foreign matter and false information without particularly consider ing image processing or a characteristic amount That is the user can freely set category names The classification standard data 121 and the classification data 122 may include data related to every possible combi nation of categories by a user and by ADC or may include only data related to combinations of categories in which corresponding defect image data actually exists That is for example as shown in line 2 in FIG 4 records having no corresponding defect image may be omitted Further each defect image data may have a format to which information of a category by user and information of a category by ADC are added a 5 30 40 45 6 FIG 5 is a diagram showing an example of standard char acteristic amount data and characteristic amount data Herein the standard characteristic amount data 124 is data created by a later described process with reference to FIG 6 and the characteristic amount data 125 is data created by a later described process with reference to FIG 11 Although the standard characteristic amount data 124 and the characteristic amount data 125 are different from each other in terms of data to be stored their formats are the same therefore these data will be explained referring to FIG 5 As shown in FIG 5 the standard characteris
39. i ment FIG 3 is a diagram showing an example of the configura tion of a review device in the present embodiment US 8 625 906 B2 3 FIG 4 is a diagram showing an example of classification standard data and classification data FIG 5 is a diagram showing an example of standard char acteristic amount data and characteristic amount data FIG 6 is a flowchart showing the procedure of a process for creating a classification standard in the present embodiment FIG 7 is a flowchart showing the procedure of a process for self checking in the present embodiment FIG 8 is a diagram showing an example of a self check screen initial screen in the present embodiment FIG 9 is a diagram showing an example of a self check screen image comparison in the present embodiment FIG 10 is a diagram showing an example of a self check screen characteristic comparison in the present embodi ment FIG 11 is a flowchart showing the procedure of a process executed upon newly obtaining defect image data FIG 12 is a flowchart showing the procedure of a check process in the present embodiment FIG 13 is a diagram showing an example of a check screen image comparison in the present embodiment FIG 14 is a diagram showing an example of a check screen characteristic amount comparison in the present embodi ment and FIG 15 is a diagram showing an example of a check screen simulation in accordance with the present embodiment BEST
40. ice 1 the probe inspection device 5 and the data processing device 3 are connected with each other via a communication line 6 FIG 2 is a diagram showing the flow of data in the semi conductor wafer manufacturing system in the present embodiment In FIG 2 elements same as those in FIG 1 are given with the same symbols and description will be omitted The review device 1 includes a plurality of optical review devices 1a and a plurality of SEM review devices 1b The optical review devices 1a obtain defect image data that are data of defect images on semiconductor wafers obtained by a digital camera connected with an optical microscope and 20 25 30 35 40 45 50 55 60 65 4 analyze the defect image data The SEM review devices 15 obtain defect image data captured by an electronic micro scope and analyze the defect image data The appearance inspection devices 2 the optical review devices 1a and the SEM review devices 15 are respectively arranged in plural number and plural defect image data can be simultaneously obtained Semiconductor wafers which are to become products flow by lot unit through a plurality of manufacturing devices 4 FIG 1 After completion ofa process in which appearance inspection of semiconductor wafers is scheduled in advance a worker or a conveying machine conveys the semiconductor wafers to the appearance inspection device 2 and appearance inspection processing is performed
41. ing to each of the categories classified by a user in the classi fication standard data is matched with image data belonging to each of the categories automatically clas sified by the image classification device in the classifi cation standard data and displays the generated association information between categories on a display section of the image classifica tion device 18 The image classification device according to claim 17 wherein the association information between categories is dis played on the display section in a matrix having a plu rality ofcells each cell showing a number of image data in the association information between categories and the processing section displays image data in the associa tion information between categories on the display sec tion when one of the cells corresponding to the image data
42. instruction is input via an input section of the image classification device to add the information on the image data selected by the user into the classification standard data 16 An image classification device automatically classify ing image data comprising a storage section that stores classification standard data which includes image data being classified in respective categories and information on the image data used as a standard for auto matically classifying image data and a processing section that updates the classification standard data by moving image data selected by a user from the image data in the classification standard data from a category in which the image data selected by the user is classified to a different category in the classification standard data 20 22 when an instruction is input via an input section of the image classification device to move the image data selected by the user from the category in which the image data selected by the user is classified to the dif ferent category 17 The image classification device according to claim 16 wherein the image data is stored in the classification standard data in each of categories into which the image data is classified by a user and each of categories into which the image data is automatically classified using the classification standard data and processing section generates association information between categories in which image data belong
43. istic amount distribution display area 713 S504 The histogram displayed in the characteristic amount dis tribution display area 713 is different from the histogram displayed in FIG 10 or 14 and represents the characteristic amount distribution of the classification standard data 121 and the characteristic amount distribution of the newly added defect image data group 123 with respect to the same cat egory and characteristic amount type In the example in FIG 15 the histogram at the time of creation of the classification standard namely the classifica tion standard data 121 for the category C1 particle and the characteristic amount type brightness is displayed as a hol low histogram and the histogram of the group of additional images namely the classification data 122 is displayed as a hatched histogram The portion where the two histograms are overlapped with each other is black solid The procedure of creating histograms is as follows First the display processing section 111 refers to the categories selected by user of the classification standard data 121 by using the category name selected via the category selection pull down menu 711 as a key and obtains all corresponding defect image data names Then the display processing sec tion 111 refers to the characteristic amount type in the stan dard characteristic amount data 124 selected via the charac teristic amount type selection pull down menu 712 by using a
44. lay area 400 FIG 10 is displayed in front by selectively inputting analysis tab via the input section 13 That is the display processing section 111 determines whether or not analysis tab has been selectively input se lected S307 in FIG 7 and if the analysis tab has not been selected S307 No the display processing section 111 pro ceeds the process to S310 in FIG 7 If the analysis tab has been selected S307 Yes then the display processing section 111 displays the characteristic amount display area 400 shown in FIG 10 FIG 10 is a diagram showing an example of a self check screen for comparing characteristic amounts according to the present embodiment If the user drags and drops an arbitrary defect image 401a displayed in the defect image list display area 202 on the self check screen 200c 200 to the characteristic amount display area 400 then the dragged and dropped defect image 401a is copied and displayed in the characteristic amount display area 400 as a defect image 401 The display processing section 111 determines whether or not a category whose characteristic amounts the user intends to display has been selected via a characteristic amount selec tion pull down menu 402 403 S308 in FIG 7 and if not selected S308 No the display processing section 111 pro ceeds the process to S310 in FIG 7 Ifa category whose characteristic amounts the user intends to display has been selected via th
45. ll the obtained defect image data names as a key and counts the numbers of defect image data for the respective values of characteristic amounts The display processing section 111 performs a similar process on the classification data 122 By histograms created in this manner the user can visually recognize the difference between the characteristic amount distribution of the classification standard data 121 and the characteristic amount distribution of the classification data 122 For example from the characteristic amount distribution display area 713 regarding the category the category classi fied by the user of C1 particle and the characteristic amount type of brightness it is observed that the charac teristic amount distribution of the group of additional images concentrate in lower values the left side of the graph com pared with that at the time of creation of the classification standard classification standard data 121 Therefore it is found to be appropriate to supplement data having the lower values in the characteristic amount distribution to the classi fication standard data 121 Then if a bar 714 is moved via the input section 13 the display processing section 111 obtains defect image data of the value at which this bar 714 is located and displays the defect image data in the defect image display area 720 In the example in FIG 15 as the bar 714 indicates the value at the leftmost side of the histogram defect i
46. mages 721 722 with this value of brightness and of category C1 particle are displayed More specifically upon reading the value of the character istic amount indicated by the bar 714 the display processing section 111 searches the value of the characteristic amount type selected via the characteristic amount type selection pull down menu 712 in the characteristic amount data 125 in FIG 5 by using the read value as a key and obtains defect image data names having this value The display processing section 111 refers to the user categories of the classification data 122 by using the obtained defect image date names as a 0 jai 5 35 40 45 50 60 65 16 key and obtains image data names corresponding to the cat egory selected via the category selection pull down menu 711 from the obtained defect image data names Then the display processing section 111 obtains defect image data from the defect image data group 123 by using the obtained defect image data names as a key and displays the obtained defect image data in the defect image display area 720 Incidentally as an example of a default arrangement the value at the leftmost of the histogram may be selected unless the user moves the bar 714 When the characteristic amount distribution at the time of creating the classification standard classification standard data 121 and the characteristic amount distribution of the group of the additional images cl
47. n image data newly input into the classification data into categories into which informa tion on image data is classified by a user and categories into which information on image data is automatically classified using the classification standard data generating association information between categories in which image data belonging to each of the categories classified by the user in the classification data is matched with image data belonging to each of the categories automatically classified by the image classification device in the classification data and displaying the generated association information between categories on a display section of the image classifica tion device 6 The method for updating an image classification stan dard according to claim 5 the method further comprising the steps carried out by the image classification device of displaying the association information between categories on the display section in a matrix having a plurality of cells each cell showing a number of image data in the association information between categories displaying image data on the display section when one of the cells showing the corresponding number of the image data in the association information between cat egories is selected 5 0 25 30 40 45 50 55 60 65 20 7 The method for updating an image classification stan dard according to claim 1 wherein the image data is a defect ima
48. n is the U S National Phase under 35 U S C 371 of International Application No PCT JP2009 071774 filed on Dec 28 2009 which in turn claims the benefit of Japanese Application No 2008 335779 filed on Dec 29 2008 the disclosures of which Applications are incorporated by reference herein TECHNICAL FIELD The present invention relates to a technology of a method and a program for updating an image classification standard and relates to an image classification device BACKGROUND ART During a process of manufacturing semiconductor prod ucts it is concerned that short circuit may occur on a formed circuit pattern because foreign matter or the like is generated or a defect such as breaking of wire and a defect due to a problematic of conditions ofa manufacturing process and the like In order to improve the product yield ratio it is necessary to identify the root cause of such a defect at an early stage and to take countermeasures For this purpose it is necessary to inspect the semiconductor wafer for foreign matter adhered on a wafer surface and pattern defects formed on the wafer surface by using a device for inspecting foreign matter on semiconductor wafers or a visual inspection device for semi conductor wafers and thereby continuously monitor occur rence of such defects and take measures to find the causes of such defects based on inspection results Conventionally such inspection has been carried out visu ally by a
49. o FIGS 9 and 10 Incidentally in FIGS 9 and 10 elements that are similar to those in FIG 8 are given with the same symbols and description will be omitted FIG 9 is a diagram showing an example of a self check screen image comparison according to the present embodi ment In FIG 9 an example is shown where in the confusion matrix 211 a cell 301 whose classification by user Manual is Cl particle and whose classification by ADC ADC is C3 pattern short is selectively input The display processing section 111 determines whether or not a defect image displayed in the defect image list display area 202 has been selected S304 in FIG 7 If no defect image has been selected S304 No then the display pro cessing section 111 proceeds the process to S307 in FIG 7 Ifthe user drags and drops a defect image 311a from defect images displayed in the defect image list display area 202 to the defect image confirmation area 203 S304 Yes in FIG 7 then the dragged and dropped defect image 311a is copied to an object image display area 330 in a image comparison area 320 in the defect image confirmation area 203 and enlarged and displayed as a defect image 311 Further in a comparison image display area 340 defect images 342 which belong to a category selected via a cat egory selection pull down menu 341 are displayed A cat egory which is selected via the category selection pull down menu 341 is a cat
50. on information between categories on a display section of the image classifica tion device 13 The method for updating an image classification stan dard according to claim 12 further comprising the steps carried out by the image classification device of displaying the association information between categories on the display section in a matrix having a plurality of cells each cell showing a number of image data in the association information between categories US 8 625 906 B2 21 selecting one of the cells showing a number of image data in the association information between categories and displaying the corresponding image data on the display section 14 The method for updating an image classification stan dard according to claim 9 wherein the image data are defect images of semiconductor wafers 15 An image classification device automatically classify ing image data comprising a storage section that stores classification standard data which includes information on image data used as a standard for automatically classifying the image data and classification data which includes information on image data of newly input image data automatically classified using the classification standard data and a processing section that updates the classification standard data by adding information on image data selected by a user from image data included in the classification data into the classification standard data when an
51. rresponding to the information on image data in the classification data in the same window 3 The method for updating an image classification stan dard according to claim 1 further comprising the steps car ried out by the image classification device of storing in the storage section standard characteristic amount data that is data of charac teristic amounts related to the information on image data in the classification standard data and characteristic amount data that is data of characteristic amounts related to the information on image data in the classification data and displaying distribution of characteristic amounts in the standard characteristic amount data and distribution of characteristic amounts in the characteristic amount data in the same window 4 The method for updating an image classification stan dard according to claim 1 the method further comprising the steps carried out by the image classification device of simulating reclassification of the information on the newly input image data by using the classification standard data which the information on the image data selected by the user has been added and displaying the result of the simulation on a display section ofthe image classification device 5 The method for updating an image classification stan dard according to claim 1 the method further comprising the steps carried out by the image classification device of adding information o
52. selected by the user on a display section of the image classification device 11 The method for updating an image classification stan dard according to claim 9 further comprising the steps car ried out by the image classification device of storing in each of the categories in the storage section standard characteristic amount data that is data of char acteristic amounts of the image data in the classification standard data and displaying distributions of the characteristic amounts of two categories selected by a user from the standard characteristic amount data in distinguishable figure in the same graph on a display section of the image classi fication device 12 The method for updating an image classification stan dard according to claim 9 further comprising the steps car ried out by the image classification device of storing the image data in the classification standard data in each of categories into which the image data is classified by a user and each of categories into which the image data is automatically classified using the classification standard data generating association information between categories in which image data belonging to each of the categories classified by the user in the classification standard data is matched with image data belonging to each of the cat egories automatically classified by the image classifica tion device in the classification standard data and displaying the generated associati
53. technology related to a review device 1 optical review devices 1a SEM review devices 15 FIG 3 is a diagram showing an example of the configura tion of a review device in the present embodiment In the present embodiment an example is shown where an SEM review device 15 is assumed to be a review device 1 however the invention is not limited thereto and may be applied to an optical review device 1a A review device 1 includes an input section 13 such as a keyboard or a mouse a display section 14 such as a display atransmitting receiving section 15 such as a communication interface card a processing section 11 for processing infor mation and a storage section 12 for storing information The processing section 11 includes a display processing section 111 an input processing section 112 an automatic defect classification section 113 a characteristic amount extraction section 114 and a data obtaining section 115 The display processing section 111 has a function of processing information and display the processed information on the display section 14 The input processing section 112 has a function of processing the information having been input from the input section 13 The automatic defect classification section 113 has a function of classifying defect image data by using ADC The characteristic amount extraction section 114 has a function of extracting the characteristic amounts of US 8 625 906 B2 5 respective defect im
54. tern short as category by ADC Incidentally in case that the classification standard data 121 is not a type as shown in FIG 4 but a type in which a category names are given to each image data it is merely required to add the category names as a result of classification in S205 to the corresponding defect image data Then the display processing section 111 displays a self check screen 200 shown in FIG 8 on the display section 14 S207 and the user performs self check via the self check screen 200 S208 Self check will be described later refer ring to FIGS 8 to 10 Then from a result of the self check the user determines as to whether or not the classification of the classification stan dard data 121 is adequate S209 As a result of S209 if it is determined to be inadequate determined that the classification of the classification stan dard data 121 is inappropriate S209 No a change is reflected into the classification standard data 121 and then the process returns to S203 to extract characteristic amounts Then the display processing section 111 again performs ADC processing and displays a result as a self check screen As a result of S209 if it is determined to be adequate determined that the classification of the classification stan dard data 121 is appropriate S209 Yes then the process is terminated The procedure of a self check process will be described below based on FIG 7 and referring to FIG 3
55. tic amount data 124 and the characteristic amount data 125 each having a defect image data name have characteristic amounts of flatness brightness circularity size etc The process for creating a classification standard will be explained below based on FIG 6 referring to FIGS 3 4 and 5 FIG 6 is a flowchart showing the process for creating a classification standard in the present embodiment In the pro cess for creating a classification standard as shown in FIG 6 is a process for creating the classification standard data 121 First the processing section 11 reads defect image data for creating a classification standard from the defect image data group 123 in the storage section 12 S201 Then the user classifies the defect image data obtained via the input section 13 S202 For example the user visually classifies the defect image data one by one into kinds such as particle scratch and the like By the process in 202 initial classification standard data 121 is created In the step of S202 only the column of categories by the user is filled in while the column of categories by ADC is blank Subsequent to S202 the characteristic amount extraction section 114 automatically adjusts defect recognition param eters for example detection sensitivity noise removing threshold protrusion recession threshold for extracting characteristic amounts from the defect image data S203 Herein the following operation is performed
56. tted Then the user determines whether or not the current clas sification result is appropriate referring to the displayed check screen 500 FIG 13 S406 As a result of S406 if it is determined to be adequate i e the classification result is appropriate S406 Yes then the process is terminated Asa result of S406 if it is determined to be inadequate i e the classification result is inappropriate S406 No then the display processing section 111 copies defect image data selected via the check screen 500 as image data for additional learning S407 Then the display processing section 111 updates the clas sification standard data 121 S408 the characteristic amount extraction section 114 performs automatic adjustment of the defect reorganization parameters and extraction of character istic amounts and the automatic defect classification section 113 thereafter performs ADC processing based on the updated classification standard data 121 S409 S407 and S408 will be described later Subsequently the processing section 11 returns the process to S405 The procedure of a check process will be described below based on FIG 12 and referring to FIGS 3 and 13 to 15 as appropriate The process shown in FIG 12 corresponds to S405 to S408 in FIG 11 FIG 12 is a flowchart showing the procedure of a check process according to the present embodiment First the display processing section 111 displays a check scre
57. urthermore a user can effectively determine appropriate ness inappropriateness of classification of defect images because it is possible to effectively display and modify char acteristic amounts between categories to be objects of classi fication such as displaying defect images and characteristic amount distributions That is as a device for automatic clas sification of detects it is possible to improve the operability in creating a classification standard improve the user friendli ness and more quickly and accurately create and adjust a classification standard Accordingly it is possible to more accurately feed back the occurrence state of detect type to which attention is paid to a line and thereby improve the yield ratio of the line In addition even when a defect of a type which is not registered in a classification standard is detected flexible measures can be taken REFERENCE NUMERALS 1 review device image classification device la optical review device 15 SEM review device 11 processing section 12 storage section 13 input section 14 display section 15 transmitting receiving section 111 display processing section 112 input processing section 113 automatic defect classification section 114 characteristic amount extraction section 115 data obtaining section 121 classification standard data 122 classification data 123 defect image data group 124 standard characteristic amount data 125 characteristic
Download Pdf Manuals
Related Search