A defect file and image number association method based on multi-format analysis

Abstract

The application discloses a defect file and image number association method based on multi-format analysis, relates to the technical field of semiconductor detection and data analysis, and comprises the following steps: constructing a multi-format rule library, receiving batch Klarf files and corresponding defect picture folders, and extracting picture numbers in defect picture filenames; reading the field names of a preset number of Defect data of the Klarf files, matching the field names with the multi-format rule library, and selecting an analysis strategy according to a matching result by using an adaptive analysis algorithm; pre-processing the analyzed Defect data, calculating an association confidence by using a three-layer verification algorithm, and generating an association result table. The application realizes the automation, standardization and high reliability of semiconductor defect detection data processing.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of semiconductor detection and data analysis, and particularly relates to a defect file and image number association method based on multi-format analysis. BACKGROUND

[0002] In the semiconductor manufacturing process, as the process size of integrated circuits continues to shrink, the defect detection precision and efficiency of wafers and chips are particularly important. AOI (automatic optical inspection) and SEM (scanning electron microscope) detection equipment is widely used in the rapid detection and fine analysis of wafer surface defects. The generated detection data is usually stored in the form of Klarf files as a standardized storage format. Klarf files contain key parameters such as defect coordinate position, topographic features, type information, etc. At the same time, the detection system generates image files corresponding to the defects and assigns them unique numbers, so that the defect morphology can be observed directly through the images in the subsequent process, thereby assisting in process defect tracing and yield improvement. However, due to the fact that different equipment manufacturers follow the SEMI standard while extending or modifying the Klarf file fields according to their own product characteristics, there are significant differences in field naming rules, number formats, and data organization structures in the actual generated Klarf files. The existing technology relies on a single fixed format analysis method and lacks effective fault tolerance mechanisms for abnormal data, often failing to parse when encountering non-standardized fields or missing number fields, and cannot reliably associate Defect and image numbers. Especially in batch detection scenarios, when the number of Klarf files is as high as hundreds and the Defect entries in a single file are thousands, the traditional parsing method is not only inefficient, but also prone to errors in the manual review process due to fatigue, affecting the accuracy and timeliness of defect analysis. Therefore, how to realize multi-format adaptive parsing of Klarf files and maintain the continuity and high accuracy of batch processing in the presence of abnormal data has become a key problem in current semiconductor detection data management.

[0003] CN1398348A Defect Source Identifier proposes a defect tracing method based on defect detection information and server-side interaction, which can display defect solutions on the client side and adjust the operation of the wafer processing system. This method has certain practicality in the utilization of defect identification and solution information, but its focus is mainly on defect tracing and solution strategy presentation, and does not involve the automatic association of Defect and image numbers in Klarf files.

[0004] CN1392954A Defect knowledge base, proposes to create and store research case database containing defect image information to support subsequent defect information calling and sharing. The scheme can enrich the knowledge storage of defect data, but its processing mode focuses on the establishment and access of case library, and does not provide a solution to the automatic discrimination of defect image numbers and multi-format Klarf file analysis. Although the related industry standards such as SEMIE142-00 define the basic structure of Klarf file, they do not solve the compatibility problem caused by manufacturer extension field. Therefore, the existing technology still has the problems of poor format compatibility, low correlation efficiency, insufficient fault tolerance mechanism and the like, and it is difficult to meet the needs of large-scale automatic detection data management.

[0005] In summary, the existing Klarf file analysis and Defect and picture number association method has the common problems of insufficient adaptability, limited efficiency and accuracy, and poor fault tolerance. The present application provides a defect file and image number association method based on multi-format analysis, which realizes compatible analysis of Klarf files output by different devices by constructing a multi-format rule library and combining an adaptive analysis algorithm; uses multi-feature fusion and a three-layer verification mechanism to improve the association accuracy of Defect and picture number; and designs a fault tolerance processing flow for abnormal data to ensure the continuity and stability of batch file processing. This method effectively solves the shortcomings of existing technology in multi-format adaptation and large-scale batch processing. SUMMARY

[0006] This section is intended to summarize some aspects of the embodiments of the present application and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the abstract and title of the specification to avoid obscuring the purpose of this section, the abstract and the title, and such simplifications or omissions cannot be used to limit the scope of the present application.

[0007] In view of the problems of poor format compatibility, low correlation efficiency and lack of abnormal fault tolerance processing in the existing Klarf file analysis method, the present application is proposed.

[0008] Therefore, the problem to be solved by the present application is how to realize high adaptability analysis of Klarf files output by different detection devices, and improve the association accuracy of Defect and picture number while ensuring the continuity of batch processing.

[0009] To solve the above technical problems, the present application provides the following technical solutions:

[0010] In a first aspect, the embodiments of the present application provide a defect file and image number association method based on multi-format analysis, which comprises,

[0011] Constructing a multi-format rule library, receiving batch Klarf files and corresponding defect picture folders, extracting picture numbers in the defect picture file names;

[0012] Reading the field name of a preset number of Defect data of the Klarf file, matching with the multi-format rule library, and selecting a parsing strategy according to the matching result by using an adaptive parsing algorithm;

[0013] Pretreating the parsed Defect data, calculating the correlation confidence by using a three-layer verification algorithm, and generating a correlation result table.

[0014] As a preferred scheme of the defect file and image number correlation method based on multi-format parsing, wherein the three-layer verification algorithm comprises:

[0015] Performing a first-layer number matching verification to compare the standardized picture number with the standardized picture number stored in the picture number database by using string comparison;

[0016] Based on the record passing the first-layer verification, performing a second-layer coordinate deviation verification to calculate the coordinate deviation value of the standardized X coordinate and the picture shooting X coordinate and the coordinate deviation value of the standardized Y coordinate and the picture shooting Y coordinate, and judging whether both coordinate deviation values are less than a preset deviation threshold value;

[0017] Based on the record passing the second-layer verification, performing a third-layer type correlation verification to query a preset defect type and picture type mapping table according to the defect type code, to obtain an expected picture type identifier and compare it with an actual picture type identifier.

[0018] As a preferred scheme of the defect file and image number correlation method based on multi-format parsing, wherein the correlation confidence value is calculated according to the verification result of the three-layer verification algorithm, comprising:

[0019] If only the first-layer verification result is in a passing state, the correlation confidence value is set as a low confidence level;

[0020] If only the first-layer verification result and the second-layer verification result are in a passing state, the correlation confidence value is set as a medium confidence level;

[0021] If the three-layer verification results are all in a passing state, the correlation confidence value is set as a high confidence level;

[0022] If any layer of the three-layer verification algorithm does not pass, the correlation confidence value is set as a confidence level according to the level of the layer that does not pass, and the verification result is marked as not passing, and an abnormal processing process is entered.

[0023] As a preferred scheme of the defect file and image number association method based on multi-format parsing provided in the present application, wherein: the association state of each Defect record is determined according to the association confidence value, and the association confidence value is compared with a preset confidence threshold value:

[0024] When the association confidence value is higher than the high confidence threshold value, the Defect record is marked as a certain association state;

[0025] When the association confidence value is between the medium confidence threshold value and the high confidence threshold value, the Defect record is marked as a possible association state;

[0026] When the association confidence value is lower than the medium confidence threshold value, the Defect record is marked as an association failure state;

[0027] Meanwhile, the Defect record without an original picture number or with a missing field is marked as a data exception state, and the association state determination result is recorded in the association record table together with the Defect ID, the standardized picture number and the association confidence value.

[0028] As a preferred scheme of the defect file and image number association method based on multi-format parsing provided in the present application, wherein: an adaptive parsing algorithm is used to select a parsing strategy according to a matching result, including:

[0029] The Defect data segment of the current file to be processed is read, the starting position of the first Defect record is located, and a preset number of Defect records are parsed line by line according to the standard structure of the Klarf file;

[0030] The field name and field value of each record in the Defect record are extracted to form a field information matrix;

[0031] The field name in the field information matrix is matched and compared with a multi-format rule library for multiple rounds to identify the picture number field corresponding to different device types, wherein the first round of matching is for the KLA device format rule in the multi-format rule library, and whether there is a field completely consistent with the field name Image_Num is checked;

[0032] According to the matching result, a corresponding parsing strategy is selected to generate a parsing configuration parameter.

[0033] As a preferred scheme of the defect file and image number association method based on multi-format parsing provided in the present application, wherein: the field name in the field information matrix is matched and compared with a multi-format rule library for multiple rounds, including:

[0034] If the first round of matching is successful, the matched device type is recorded as KLA device type, and the corresponding field position index and numbering format rule are obtained; if the first round of matching fails, the field name in the field information matrix is matched with the ASML device format rule in the multi-format rule library in the second round of matching, and it is checked whether the Pic_ID field name corresponding to the ASML device format rule is contained in the field name;

[0035] If the second round of matching is successful, the matched device type is recorded as ASML device type, and the corresponding field position index and numbering format rule are obtained; if the second round of matching fails, a keyword fuzzy matching mechanism is started, the field name in the field information matrix is matched with the pre-stored other device format rule in the multi-format rule library in the third round of matching, and it is checked whether any one of the Image prefix field, the Pic prefix field or the Photo prefix field is contained in the field name;

[0036] If the third round of matching is successful, the matched device type is recorded as other device type; if the third round of matching fails, a field length analysis strategy is used as an emergency analysis scheme, the field value corresponding to the field name is traversed, the field whose field value length is in a preset range and contains a digital character is selected as a picture numbering candidate field, and the field name and position index of the picture numbering candidate field are recorded as an analysis rule of the unknown device type.

[0037] As a preferred scheme of the defect file and image numbering association method based on multi-format analysis provided by the application, according to the matching result, a corresponding analysis strategy is selected, and analysis configuration parameters are generated, including:

[0038] If the current file to be processed is marked as a KLA device format type, a KLA analysis strategy is used to directly extract the Image_Num field value;

[0039] If the current file to be processed is marked as an ASML device format type, an ASML analysis strategy is used to extract the Pic_ID field value and remove the device prefix;

[0040] When the adaptive analysis algorithm is started, a general analysis strategy is used to extract the digital part of the target picture numbering field, and analysis configuration parameters of the current file to be processed are established, wherein the analysis configuration parameters include a selected analysis strategy type, a target field name, a field position index and a data preprocessing rule.

[0041] As a preferred scheme of the defect file and image numbering association method based on multi-format analysis provided by the application, the construction method of the multi-format rule library is,

[0042] Initialize a multi-format rule library, in which KLA equipment format rules, ASML equipment format rules and general equipment format rules of detection equipment are pre-stored;

[0043] Receive a batch of Klarf files uploaded by a user, and receive a defect picture folder corresponding to the batch of Klarf files;

[0044] Traverse the defect picture folder, and extract a continuous digital string of a picture file name as a picture numbering rule by using a regular expression;

[0045] Establish a picture numbering database, store a standardized picture number as a primary key, and store corresponding defect picture file paths, file names and picture format information as associated attributes.

[0046] In a second aspect, an embodiment of the present application provides a computer device, including a memory and a processor, and the memory stores a computer program, wherein the processor implements any step of the defect file and image number association method based on multi-format analysis when executing the computer program.

[0047] In a third aspect, an embodiment of the present application provides a computer readable storage medium, which stores a computer program, wherein the computer program is executed by a processor to implement any step of the defect file and image number association method based on multi-format analysis.

[0048] Compared with the prior art, the present application has the beneficial effects that: by constructing a multi-format rule library containing KLA equipment, ASML equipment and general equipment format rules, the present application realizes the unified processing capability of Klarf files generated by different semiconductor detection equipment, solves the technical problems of the traditional method that can only process a single equipment format and needs manual configuration of parsing rules, thereby achieving the beneficial effects of improving system compatibility and reducing manual intervention; by using multi-round matching and adaptive parsing algorithms, the present application realizes intelligent identification of Klarf file field structure and dynamic selection of parsing strategies, can automatically adapt to data format differences and field naming changes of different equipment, effectively solves the problems of fixed parsing strategies and poor adaptability to new equipment formats in the prior art, and achieves the beneficial effects of improving parsing accuracy and system flexibility; by establishing a three-layer verification algorithm of numbering matching, coordinate deviation and type association, the present application realizes multi-dimensional verification and confidence quantification evaluation of the association results of Defect data and defect pictures, not only improves the accuracy of association matching, but also identifies and filters mis-matching results, solves the technical defects of traditional single matching methods that are prone to false association and lack of reliability evaluation, and achieves the beneficial effects of significantly improving defect data association quality, reducing misjudgment rate and providing reliable quantification indicators. BRIEF DESCRIPTION OF DRAWINGS

[0049] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor. Among them:

[0050] Figure 1 Flow chart of the defect file and image number association method based on multi-format parsing;

[0051] Figure 2 Flow chart of three-layer verification algorithm of the defect file and image number association method based on multi-format parsing. DETAILED DESCRIPTION

[0052] In order to make the above-mentioned purposes, features and advantages of the present application more apparent and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments.

[0053] Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should be within the scope of protection of the present application.

[0054] In the following description, many specific details are set forth in order to provide a thorough understanding of the present application, but the present application can also be implemented in other ways different from those described herein, and those skilled in the art can make similar generalizations without departing from the connotation of the present application, therefore the present application is not limited to the specific embodiments disclosed below.

[0055] As described in the above background, the existing Klarf file parsing and Defect and picture number association method generally has poor format compatibility, low association efficiency and insufficient fault tolerance in batch processing scene, etc., which is difficult to meet the demand of large-scale automatic processing of semiconductor detection data. In view of the above problems, the present application provides a defect file and image number association method based on multi-format parsing.

[0056] Reference Figure 1 , Figure 1 Flow chart of a defect file and image number association method based on multi-format parsing according to an embodiment of the present application. As shown in Figure 1 , in a defect file and image number association method based on multi-format parsing, comprising:

[0057] S1: Construct a multi-format rule library, receive batch Klarf files and corresponding defect picture folders, extract picture numbers in defect picture file names;

[0058] S2: read the field name of the preset number Defect data of the Klarf file, match with the multi-format rule library, and select a parsing strategy according to the matching result by using an adaptive parsing algorithm;

[0059] S3: pre-process the parsed Defect data, calculate the correlation confidence by using a three-layer verification algorithm, and generate a correlation result table.

[0060] In the embodiment of the present application, the above step S1 includes:

[0061] S1.1: initialize the multi-format rule library, and pre-store the KLA equipment format rule, the ASML equipment format rule and the general equipment format rule of the detection equipment in the multi-format rule library;

[0062] It should be noted that the KLA equipment format rule defines the picture number field name as Image_Num and the numbering format as pure numbers; the ASML equipment format rule defines the picture number field name as Pic_ID and the numbering format as the equipment prefix plus numbers; and the general equipment format rule defines the field name with Image, Pic and Photo as the prefix field as the candidate picture number field.

[0063] For example, the rule of the KLA equipment is defined as: the picture number field name is Image_Num, and the value is 3 pure numbers (such as 001); the rule of the ASML equipment is defined as: the field name is Pic_ID, and the value format is equipment number-number (such as AS-0001); and the general equipment rule identifies the candidate field through the keyword prefix (such as Image, Pic and Photo), and extracts 3-6 characters after the prefix as the number candidate value.

[0064] S1.2: receive the batch Klarf file uploaded by the user, and receive the defect picture folder corresponding to the batch Klarf file;

[0065] It should be noted that the batch Klarf file supports text files with txt suffix, 000 suffix and trf suffix; and the defect picture folder includes picture files in jpg format, tiff format and png format.

[0066] S1.3: traverse the defect picture folder, and extract the continuous number string of the picture file name as the picture numbering rule by using a regular expression;

[0067] For example, "001" is extracted from the file name "Defect_001.jpg", and "0005" is extracted from "SEM_Pic_0005.tiff"; for the file name with prefix (such as "KLA_Image_123.png"), the "KLA_Image_" part is removed, and "123" is reserved as the standardized number, ensuring the uniformity of the number format.

[0068] S1.4: Establish a picture number database, taking the standardized picture number as the primary key, and storing the corresponding defect picture file path, file name and picture format information as the associated attribute;

[0069] Preferably, based on the picture number extraction rule, the corresponding picture number is extracted from the file name of each defect picture file, for the file name containing the device prefix, the prefix character is removed to reserve the pure digital part as the standardized picture number, and the mapping relationship between the standardized picture number and the corresponding defect picture file path is established.

[0070] Specifically, the first Klarf file in the batch Klarf file is selected as the current to-be-processed file, the file header information and the field structure of the first N Defect records of the current to-be-processed file are read; all field names in the first N Defect records are extracted to form a field name set; the field name set is compared with the picture number field names of the device format rules in the multi-format rule library.

[0071] It should be noted that when there is a field in the field name set that completely matches the picture number field name of a device format rule, it is determined that the current to-be-processed file corresponds to the device format rule, otherwise the general device format rule is started for fuzzy matching, and the candidate picture number field name is determined through the keyword prefix and the field value feature; according to the determined device format rule or the candidate picture number field name, the parsing rule mapping of the current to-be-processed file is established.

[0072] Further, the integrity of the picture number database is checked, the number of defect picture files from which the picture number is successfully extracted is counted, the defect picture files for which the picture number extraction fails due to abnormal file name format are identified, and the information of the abnormal defect picture files is recorded to the abnormal log; the batch file processing queue is initialized, the batch Klarf files are arranged in the processing queue in the order of file creation time, and the processing state identifier is set to track the parsing progress of each Klarf file.

[0073] In the embodiment of the present application, the above step S2 comprises:

[0074] S2.1: read the Defect data section of the current file to be processed, locate to the start position of the first Defect record, and parse a preset number of Defect records line by line according to the standard structure of the Klarf file;

[0075] S2.2: extract the field name and field value of each record in the Defect record to form a field information matrix;

[0076] It should be noted that the field information matrix contains field name, field position and field data type information.

[0077] S2.3: match the field name in the field information matrix with the multi-format rule library for multiple rounds to identify the picture number field corresponding to different device types, wherein the first round of matching is for KLA device format rules in the multi-format rule library, and whether there is a field completely consistent with the field name Image_Num is checked.

[0078] In an optional embodiment, if the first round of matching is successful, the matching device type is recorded as KLA device type, and the corresponding field position index and number format rule are obtained; if the first round of matching fails, the second round of matching is performed between the field name in the field information matrix and the ASML device format rule in the multi-format rule library, and whether the field name contains the Pic_ID field name corresponding to the ASML device format rule is checked.

[0079] In an optional embodiment, if the second round of matching is successful, the matching device type is recorded as ASML device type, and the corresponding field position index and number format rule are obtained; if the second round of matching fails, a keyword fuzzy matching mechanism is started, the third round of matching is performed between the field name in the field information matrix and the pre-stored other device format rules in the multi-format rule library, and whether the field name contains any one of the Image prefix field, the Pic prefix field or the Photo prefix field is checked.

[0080] In an optional embodiment, if the third round of matching is successful, the matching device type is recorded as other device type; if the third round of matching fails, a field length analysis strategy is used as an emergency analysis scheme, the field value corresponding to the field name is traversed, the field whose field value length is in a preset range and contains a digital character is selected as a picture number candidate field, and the field name and position index of the picture number candidate field are recorded as an analysis rule of an unknown device type.

[0081] For example, check if there is an "Image_Num" field (KLA device), if not, check the "Pic_ID" field (ASML device), if still not matched, start fuzzy matching, find the field containing the prefix "Image", "Pic", "Photo"; if all matching fails, enable emergency strategy: traverse all field values, filter the field with length of 3-6 and containing numbers as candidate number field; if there is a field "Photo_No" in a certain file and its value is "005", it is identified as a candidate field.

[0082] S2.4: Select the corresponding analysis strategy according to the matching result, and generate analysis configuration parameters;

[0083] In an optional embodiment, if the current file to be processed is marked as KLA device format type, the KLA analysis strategy is adopted to directly extract the Image_Num field value; if the current file to be processed is marked as ASML device format type, the ASML analysis strategy is adopted to extract the Pic_ID field value and remove the device prefix; when the adaptive analysis algorithm is started, the general analysis strategy is adopted to extract the digital part of the target picture number field, and the analysis configuration parameters of the current file to be processed are established, wherein the analysis configuration parameters include the selected analysis strategy type, the target field name, the field position index and the data preprocessing rule.

[0084] In the embodiment of the present application, the above step S3 comprises:

[0085] S3.1: Based on the analysis configuration parameters, extract DefectID, original picture number, X coordinate value, Y coordinate value and defect type code from each Defect data record of the single Klarf file;

[0086] S3.2: Take the original picture number as the number data to be processed, the X coordinate value and the Y coordinate value as the defect position coordinates, and the defect type code as the defect classification identifier;

[0087] S3.3: Standardize the original picture number, and detect whether the original picture number contains non-numeric prefix characters;

[0088] In an optional embodiment, if the original picture number contains prefix characters, the non-numeric prefix characters are removed by string truncation to retain the pure digital part to form the standardized picture number; if the original picture number is pure digital, it is directly taken as the standardized picture number.

[0089] For example, if the picture number of a certain Defect is "AS-0001", the prefix "AS-" is removed and "0001" is retained as the standardized number; if the number is "001", it is directly retained.

[0090] S3.4: Unit conversion processing is performed on the defect position coordinates to identify the current measurement units of the X coordinate value and the Y coordinate value;

[0091] In an optional embodiment, if the measurement units are microns, the pixel conversion ratio is converted into pixel coordinate units according to the preset pixel conversion ratio; if the measurement units are already pixel units, the original values are kept to obtain the standardized X coordinate and the standardized Y coordinate in the unified pixel units.

[0092] S3.5: The first layer number matching verification is performed to compare the standardized picture number with the standardized picture number stored in the picture number database by string comparison.

[0093] Preferably, the overall logic flow of the three-layer verification algorithm is as shown in Figure 2

[0094] It should be noted that, according to the matching quality of the standardized picture number and the database number, three different processing schemes are dynamically enabled, and the core logic flow is as follows: if the number can be accurately matched, enter the efficient confident process (scheme a); if it cannot be accurately matched but is highly similar, enter the fuzzy matching and recommendation process (scheme b); if the format is abnormal or completely unmatched, enter the abnormal processing and reconstruction process (scheme c).

[0095] In an optional embodiment, when the first layer number matching verification scheme a is performed, if the string comparison result is a completely accurate matching state, a straight-through verification confirmation mode is adopted, a one-time accurate positioning comparison of the standardized picture number is performed by a high-efficiency hash index engine, three-dimensional verification confirmation functions are added to the character sequence, the number length and the data integrity; if a potential duplicate number conflict event occurs, a uniqueness enhancement verification mode is adopted, a full database scanning comparison of the number uniqueness is performed by a distributed deduplication detector, two-dimensional cross verification functions are added to the character sequence, the number length and the data integrity; if the index performance of the picture number database fluctuates, an intelligent cache optimization algorithm is used to adaptively adjust the query strategy, an efficient matching search mechanism is constructed, and query delay nodes are prevented.

[0096] ​It should be noted that the three-dimensional verification confirmation function includes bit-by-bit character consistency confirmation function, number bit length check function and coding format integrity verification function, the bit-by-bit character consistency confirmation function performs ASCII code value accurate comparison for each character bit, the number bit length check function performs matching verification on the total number of bits and the standard length of the database, and the coding format integrity verification function checks the purity and format specification of the number characters; The two-dimensional cross-checking function includes number uniqueness verification function and timestamp association verification function, the number characteristic information is dynamically maintained and fed back through real-time index updating and historical matching record online mode, and the number matching success rate is statistically evaluated.

[0097] When the first layer number matching verification scheme b is executed, the string partial matching or fuzzy similarity event occurs, the similarity quantization analysis verification mode is adopted, the number difference is finely analyzed and positioned by the multi-algorithm fusion string similarity calculator, the four-dimensional progressive verification function is added to the character sequence, number length and data integrity; If the similarity threshold exceeds the preset critical point, the number verification system carries out candidate matching mark feedback, automatically triggers the artificial intelligence auxiliary judgment mechanism, adds fuzzy matching index construction function to the number database, the number verification system automatically receives the character difference characteristic information fed back by the similarity calculator, processes the similar number with priority, and automatically calls the best matching recommendation algorithm; If the recommended matching is confirmed, the number verification system stores the successful fuzzy matching strategy parameters, automatically updates the similarity threshold dynamic adjustment strategy, and identifies the key character feature difference between the successful fuzzy matching and the failed matching by comparison analysis method, and establishes the character fault tolerance matching knowledge graph.

[0098] It should be noted that the four-dimensional progressive verification function includes bit-by-bit character consistency confirmation function, number bit length check function, coding format integrity verification function and string similarity quantization analysis function, and the multi-algorithm similarity calculator comprehensively evaluates the string difference by Levenshtein distance, Jaccard coefficient and cosine similarity.

[0099] When the first layer number matching verification scheme c is executed, the multi-level exception handling and reconstruction verification mode is adopted in the face of string complete mismatch or format exception event, the abnormal number is analyzed and reconstructed in depth by the intelligent number parsing reconstructor, the five-dimensional cyclic diagnostic verification function is added to the character sequence, number length and data integrity; after the abnormal number reconstruction is completed, the number verification system carries out secondary integrity verification on the repaired number, automatically generates abnormal processing report and repair suggestion, continuously optimizes the number quality control through the hierarchical progressive abnormal processing mechanism, and establishes the adaptive abnormal number processing knowledge base; when all abnormal processing strategies cannot repair the number, the system automatically marks the number as a state to be manually audited, and triggers the abnormal number statistical analysis process, and ends the multi-level abnormal processing verification process.

[0100] It should be noted that the five-dimensional cyclic diagnostic verification function includes bit-by-bit character consistency confirmation function, number bit length verification function, coding format integrity verification function, string similarity quantitative analysis function and number format standardization reconstruction function, the intelligent number parsing reconstructor analyzes the abnormal number layer by layer through the sequential diagnostic strategy, and establishes the number format exception pattern recognition mechanism; when the repairable format exception is found, the system automatically triggers the number standardization reconstruction process, intelligently cleans and reconstructs the abnormal characters through the cyclic iteration method, and establishes the abnormal number automatic repair rule library.

[0101] For example, when the standardized number "001" is completely consistent with the number in the database, scheme a is started, the hash index is quickly located, and the character bit, length and format are verified three times to ensure accurate matching. If the number "001" does not exist in the database, but "002" or "000" exists, scheme b is started, the string similarity is calculated, and if the similarity exceeds the threshold, it is marked as a candidate match. If the number format is abnormal (such as "1A3B"), scheme c is started to try to reconstruct the number format, and if it cannot be repaired, it is marked as to be manually audited.

[0102] S3.6: Based on the record passing the first layer verification, perform second layer coordinate deviation verification, calculate the coordinate deviation value of the standardized X coordinate and the picture shooting X coordinate, and the coordinate deviation value of the standardized Y coordinate and the picture shooting Y coordinate, and judge whether the two coordinate deviation values are less than the preset deviation threshold value;

[0103] In an optional implementation, when the two coordinate deviation values are less than the preset deviation threshold value, the second layer verification result is recorded as deviation verification passing.

[0104] Preferably, the specific formula of the coordinate deviation value is as follows:

[0105] ;

[0106] wherein, is a coordinate deviation value, is a normalized X coordinate, the X coordinate value in the Defect record, is a picture shooting X coordinate, the X coordinate extracted from the picture EXIF information or file name, is a normalized Y coordinate, the Y coordinate value in the Defect record, is a picture shooting Y coordinate, the Y coordinate extracted from the picture EXIF information or file name.

[0107] It should be noted that the formula calculates the Euclidean distance between the coordinate point in the Defect record and the picture shooting coordinate point, that is, the straight line distance deviation (in pixels) of the two points in the two-dimensional plane; the preset deviation threshold is determined based on the imaging accuracy of the detection equipment, the picture resolution and the feature size of the process node; usually, the threshold is set to 3 to 5 pixels to adapt to the coordinate measurement error of different detection equipment (such as AOI, SEM), and to ensure accurate association of Defect and corresponding picture within the allowable positioning tolerance range.

[0108] For example, the normalized coordinate of a certain Defect is (X = 100, Y = 200), and the picture shooting coordinate is (X = 102, Y = 198), and the deviation value is about 2.83 pixels; if the preset threshold is 5 pixels, the record passes the second layer of verification.

[0109] S3.7: Based on the record passing the second layer of verification, perform a third layer type association verification, query the preset defect type and picture type mapping table according to the defect type code, obtain the expected picture type identifier and compare it with the actual picture type identifier;

[0110] In an optional embodiment, a hierarchical judgment strategy is adopted for accurate matching verification; if the expected picture type identifier and the actual picture type identifier are completely consistent, it is directly marked as type matching success, and the association confidence of the record is improved to a high confidence level; if there is a partial matching situation, an intelligent compatibility analysis module will be started, and whether it belongs to an acceptable type variant is judged through a preset type compatibility rule table, such as the compatibility of TIFF format and TIF format, the corresponding relationship of high resolution and standard resolution pictures, etc.; if type mismatch or abnormal situation is detected, an interactive manual confirmation mechanism will be automatically triggered, and a detailed type difference report will be pushed to the operator, including expected type, actual type, possible cause analysis and suggested processing scheme, the operator can choose to accept the system suggestion, manually adjust the mapping relationship or mark the record as a state to be further audited.

[0111] For example, if the type code of a Defect is "SCRATCH", the expected picture type should be "AOI"; if the actual picture type is "SEM", the type is not matched and the artificial confirmation mechanism is triggered to prompt the operator to check whether it is a mislabeling or a special case. If the type is "PARTICLE" and the picture type is "SEM", it is directly passed, and the association confidence is promoted to a high level.

[0112] S3.8: Calculate the association confidence value according to the verification result of the three-layer verification algorithm;

[0113] In an optional embodiment, if only the first layer verification result is a pass state, the association confidence value is set to a low confidence level, and the specific formula is as follows:

[0114] ;

[0115] wherein, is the association confidence value when only the first layer verification is passed, is the weight coefficient of the number matching layer, which is an empirical constant that can be configured, and is usually , is the number of matching characters, which is the number of characters that are completely consistent in the character sequence comparison between the standardized picture number and the database number, is the source number length, which is the total number of characters of the standardized picture number in the current Defect record, is the target number length, which is the total number of characters of the corresponding standardized picture number in the picture number database, is the maximum number length allowed by the system, which is a preset system constant for length normalization.

[0116] It should be noted that, the value range of the association confidence value when only the first layer verification is passed is [0, ], when , it means that the number is completely matched and the length is consistent, reaching the highest confidence in this case; when , it means that the number is not completely matched or the length difference is too large, and the confidence is the lowest; The closer the value of to , the higher the credibility of the number matching layer alone. The size of the weight coefficient reflects the degree of trust of the system in the single evidence of number matching, which can usually be set to 0.6, indicating that even if the number is perfectly matched, the highest confidence will not exceed 0.6, which belongs to the low confidence level.

[0117] In an optional embodiment, if only the first layer and the second layer are passed, the associated confidence value is set to a medium confidence level, and the specific formula is as follows:

[0118] ;

[0119] wherein, is the associated confidence value when the first layer and the second layer are passed, is the weight coefficient of the number matching layer, is the weight coefficient of the coordinate matching layer, satisfying 0 <1, and + <1, D is the coordinate deviation value, i.e., the Euclidean distance between the defect record coordinate and the picture shooting coordinate, is the coordinate deviation tolerance parameter, used to control the rate of Gaussian decay.

[0120] It should be noted that, the value range of the associated confidence value when the first layer and the second layer are passed is [0, + ], when approaches the upper limit + , it means that the number is completely matched and the coordinate deviation is extremely small, reaching the highest level of medium confidence level; when approaches 0, it means that the number matching degree is extremely poor and the coordinate deviation is extremely large, and the confidence is extremely low. Generally, the value of + is about 0.8 (for example, =0.5, =0.3), indicating that in the absence of third layer type verification evidence, the highest confidence given by the system is at a medium level; the value of needs to be compared with the confidence of passing only one layer of verification and the confidence of passing three layers of verification in a unified dimension and scale.

[0121] In an optional embodiment, if all three layers of verification are passed, the associated confidence value is set to a high confidence level, and the specific formula is as follows:

[0122] ;

[0123] wherein, is the associated confidence value when all three layers are passed, is the weight coefficient of the type matching layer, satisfying 0 1, and ; For type consistency function, the specific formula is as follows:

[0124] ;

[0125] wherein, is a set of expected picture type identifiers, is a set of actual picture type identifiers, is a type compatibility coefficient, when the set of expected picture type identifiers and the set of actual picture type identifiers exist in the preset compatibility rule table, 1.0 is taken, otherwise 0.8 is taken.

[0126] It should be noted that the value range of the association confidence value when passing all three layers of verification is [0, 1], when , it means that the number is completely matched, the coordinate deviation is zero, the type is completely consistent and compatible, and the highest confidence is reached; when , it means a high confidence level, and the association result is highly reliable; when , although it passes the three layers of verification, the confidence does not reach the highest level due to a small flaw in some evidence.

[0127] For example, if a Defect only passes the first layer of number matching, the confidence is low (such as 60%); if it passes the first two layers (number + coordinate), the confidence is medium (such as 80%); if all three layers pass, the confidence is high (such as 85%); the confidence calculation formula comprehensively considers number matching degree, coordinate deviation, type consistency and other multi-dimensional features, ensuring that the evaluation result is scientific and reliable; according to the confidence threshold, the association state is automatically determined, and an association result table containing DefectID, picture number, confidence and other fields is generated.

[0128] Specifically, if any layer in the three-layer verification algorithm does not pass, the association confidence value is set to the confidence level according to the level of the layer that does not pass, and the verification result is marked as not passing and enters the exception handling process; if the first layer does not pass, skip the second and third layers of verification, and directly mark it as pending manual review; if the second or third layer does not pass, record the specific level and reason of the layer that does not pass, and mark it as pending manual review.

[0129] Further, according to the association confidence value, the association state of each Defect record is determined by comparing the association confidence value with the preset confidence threshold, and an association result table is generated.

[0130] It should be noted that the association result table contains the ID of Defect, the standardized picture number, the association confidence value, the verification state identifier and the association result field of the corresponding defect picture file path, and each record in the association result table is indexed and associated with the original Defect data record.

[0131] In an optional embodiment, when the association confidence value is higher than the high confidence threshold, the Defect record is marked as a certain association state; when the association confidence value is between the medium confidence threshold and the high confidence threshold, the Defect record is marked as a possible association state; when the association confidence value is lower than the medium confidence threshold, the Defect record is marked as an association failure state; and the Defect record without an original picture number or with a missing field is marked as a data anomaly state. The association state determination result is recorded in the association result table together with the Defect ID, the standardized picture number and the association confidence value.

[0132] To sum up, the present application realizes the unified processing capability of Klarf files generated by different semiconductor detection devices by constructing a multi-format rule library containing the KLA device, the ASML device and the general device format rule, solves the technical problems of the traditional method that can only process a single device format and needs manual configuration of parsing rules, and thus achieves the beneficial effects of improving system compatibility and reducing manual intervention. By adopting the multi-round matching and adaptive parsing algorithm, the present application realizes the intelligent identification of the Klarf file field structure and the dynamic selection of the parsing strategy, can automatically adapt to the data format difference and field naming change of different devices, effectively solves the problems of fixed parsing strategy and poor adaptability to new device formats in the prior art, and achieves the beneficial effects of improving parsing accuracy and system flexibility. By establishing the three-layer verification algorithm of number matching, coordinate deviation and type association, the present application realizes the multi-dimensional verification and confidence quantification evaluation of the association result of Defect data and defect pictures, not only improves the accuracy of association matching, but also identifies and filters the mis-matching result, solves the technical defects of the traditional single matching method that is easy to produce false association and lacks reliability evaluation, and achieves the beneficial effects of significantly improving the defect data association quality, reducing the misjudgment rate and providing a reliable quantification index.

[0133] The present embodiment also provides a computer device suitable for the case of the defect file and image number association method based on multi-format parsing, which comprises a memory and a processor; the memory is used for storing computer executable instructions, and the processor is used for executing the computer executable instructions to realize the defect file and image number association method based on multi-format parsing proposed in the above embodiments.

[0134] The computer device can be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for running the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is configured to perform wired or wireless communication with an external terminal. The wireless communication can be achieved through WIFI, an operator network, NFC (Near Field Communication) or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computer device, or an external keyboard, touchpad or mouse, etc.

[0135] The embodiment also provides a storage medium having a computer program stored thereon, and the computer program is executed by a processor to implement the method for associating a defect file with an image number based on multi-format analysis.

[0136] The storage medium provided by the embodiment belongs to the same inventive concept as the data storage method provided by the above embodiment, and the technical details not described in detail in the embodiment can be referred to the above embodiment, and the embodiment has the same beneficial effects as the above embodiment.

[0137] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application but not limit the present application. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present application, and all of them should be covered in the scope of the claims of the present application.

Claims

1. A method for associating a defect file with an image number based on multi-format parsing, the method comprising: Comprising, ​ Constructing a multi-format rule library, receiving batch Klarf files and corresponding defect picture folders, extracting picture numbers in the defect picture file names; Reading the field names of a preset number of Defect data of the Klarf file, matching with the multi-format rule library, and selecting a parsing strategy according to the matching result by using an adaptive parsing algorithm; Pretreating the parsed Defect data, calculating the correlation confidence by using a three-layer verification algorithm, and generating a correlation result table; The three-layer verification algorithm comprises: Performing first-layer number matching verification to compare the standardized picture numbers with the standardized picture numbers stored in the picture number database by using string comparison; Based on the records passed by the first-layer verification, performing second-layer coordinate deviation verification to calculate the coordinate deviation values of the standardized X coordinate and the picture shooting X coordinate and the coordinate deviation values of the standardized Y coordinate and the picture shooting Y coordinate, and judging whether both coordinate deviation values are less than a preset deviation threshold; Based on the records passed by the second-layer verification, performing third-layer type correlation verification to query the preset defect type and picture type mapping table according to the defect type code, to obtain an expected picture type identifier and compare it with an actual picture type identifier; Calculating the correlation confidence value according to the verification result of the three-layer verification algorithm, comprising: If only the first-layer verification result is in the pass state, setting the correlation confidence value as a low confidence level; If only the first-layer verification result and the second-layer verification result are in the pass state, setting the correlation confidence value as a medium confidence level; If the three-layer verification results are all in the pass state, setting the correlation confidence value as a high confidence level; If any layer in the three-layer verification algorithm fails, setting the correlation confidence value as a confidence level according to the failed layer, marking the verification result as failed, and entering an abnormal processing process; Selecting a parsing strategy according to the matching result by using an adaptive parsing algorithm, comprising: Reading the Defect data segment of the current file to be processed, positioning to the start position of the first Defect record, and parsing a preset number of Defect records row by row according to the standard structure of the Klarf file; Extracting the field name and field value of each record in the Defect record to form a field information matrix; Matching and comparing the field names in the field information matrix with the multi-format rule library for multiple rounds to identify the picture number fields corresponding to different device types, wherein the first round of matching is for the KLA device format rule in the multi-format rule library to check whether there is a field completely consistent with the field name Image_Num; Selecting a corresponding parsing strategy according to the matching result to generate parsing configuration parameters.

2. The method for associating a defect file with an image number based on multi-format parsing according to claim 1, wherein: Determining the correlation state of each Defect record according to the correlation confidence value by comparing the correlation confidence value with a preset confidence threshold: When the correlation confidence value is higher than the high confidence threshold, the Defect record is marked as a determined correlation state; When the correlation confidence value is between the medium confidence threshold and the high confidence threshold, the Defect record is marked as a possible correlation state; When the association confidence value is lower than the medium confidence threshold, the Defect record is marked as a failed association state; At the same time, the Defect record without an original picture number or with a missing field is marked as a data exception state, and the association state determination result is recorded in the association record table together with the Defect ID, the standardized picture number, and the association confidence value.

3. The method for associating a defect file with an image number based on multi-format parsing according to claim 1, wherein: The field name in the field information matrix is matched and compared with the multi-format rule library in multiple rounds, including: If the first round of matching is successful, the matched equipment type is recorded as a KLA equipment type, and the corresponding field position index and number format rule are obtained; if the first round of matching fails, the field name in the field information matrix is matched with the ASML equipment format rule in the multi-format rule library in the second round of matching, and it is checked whether the Pic_ID field name corresponding to the ASML equipment format rule is included in the field name; If the second round of matching is successful, the matched equipment type is recorded as an ASML equipment type, and the corresponding field position index and number format rule are obtained; if the second round of matching fails, a keyword fuzzy matching mechanism is started, the field name in the field information matrix is matched with the pre-stored other equipment format rule in the multi-format rule library in the third round of matching, and it is checked whether any of the Image prefix field, the Pic prefix field, or the Photo prefix field is included in the field name; If the third round of matching is successful, the matched equipment type is recorded as another equipment type; if the third round of matching fails, a field length analysis strategy is used as an emergency analysis scheme, the field values corresponding to the field name are traversed, the fields with a length within a preset range and containing a digital character are selected as picture number candidate fields, and the field name and position index of the picture number candidate field are recorded as an analysis rule of an unknown equipment type.

4. The method for associating a defect file with an image number based on multi-format parsing according to claim 1, wherein: According to the matching result, a corresponding analysis strategy is selected, and an analysis configuration parameter is generated, including: If the current file to be processed is marked as a KLA equipment format type, the KLA analysis strategy is used to directly extract the Image_Num field value; If the current file to be processed is marked as an ASML equipment format type, the ASML analysis strategy is used to extract the Pic_ID field value and remove the equipment prefix; When the adaptive analysis algorithm is started, the digital part of the target picture number field is extracted by using the general analysis strategy, and the analysis configuration parameter of the current file to be processed is established, wherein the analysis configuration parameter includes the selected analysis strategy type, the target field name, the field position index, and the data preprocessing rule.

5. The method for associating a defect file with an image number based on multi-format parsing according to claim 3, wherein: The construction method of the multi-format rule library is as follows: The multi-format rule library is initialized, and the KLA equipment format rule, the ASML equipment format rule, and the general equipment format rule of the detection equipment are pre-stored in the multi-format rule library; A batch of Klarf files uploaded by a user are received, and a defect picture folder corresponding to the batch of Klarf files is received; The defect picture folder is traversed, and a continuous digital string of a picture file name is extracted as a picture number rule by using a regular expression; A picture number database is established, and a standardized picture number is taken as a primary key, and corresponding defect picture file path, file name and picture format information are taken as associated attributes for storage.

6. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that: The processor implements the steps of the defect file and image number association method based on multi-format analysis according to any one of claims 1-5 when executing the computer program.

7. A computer readable storage medium having stored thereon a computer program, characterized in that: The computer program is executed by the processor to implement the steps of the defect file and image number association method based on multi-format analysis according to any one of claims 1-5.

Images