A semiconductor big data traceability analysis method and system

Abstract

The application relates to the technical field of semiconductor data analysis, in particular to a semiconductor big data traceability analysis method and system, which comprises the following steps: acquiring semiconductor production line data, wherein the semiconductor production line data at least comprises: (1) a plurality of processing equipment information experienced by a plurality of semiconductor products, (2) process parameters adopted when the semiconductor products experience at least one processing equipment and / or category information of the semiconductor products, and (3) measurement values; calling a data analysis scheme for the semiconductor production line data; the data analysis scheme comprises a fault prediction algorithm; pre-processing the semiconductor production line data according to the data analysis scheme to generate a pre-processing result; inputting the pre-processing result and corresponding semiconductor production line data into a preset big data platform; and the big data platform outputs an analysis result correspondingly. The big data traceability analysis method can improve the data analysis efficiency, reduce the requirement for computing power in the analysis process and reduce the implementation cost.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of semiconductor manufacturing, in particular to a semiconductor big data traceability analysis method and system. BACKGROUND

[0002] In the field of semiconductor manufacturing, especially in 12-inch wafer manufacturing plants (hereinafter referred to as Fab plants), the root cause analysis (RCA) method is an important means of quality management and process control. When there are problems such as yield reduction, product defects or customer complaints in the production process, systematic analysis is needed to identify the root cause of the problem and take measures to prevent it from happening again.

[0003] Currently, the RCA method commonly used in Fab plants mainly relies on the experience of engineers and manual data collection. The specific process usually includes: extracting relevant data from multiple independent data systems such as FDC (fault detection and classification), SPC (statistical process control), WAT (electrical testing), Inline (online measurement), etc., and then comparing and analyzing.

[0004] For example, patent application CN202310893076.0 discloses a FDC traceability analysis method based on distributed parallel computing and a storage medium, which includes obtaining various information of failed devices; obtaining FDC data and configuration files in the semiconductor wafer production and manufacturing process; preprocessing the FDC data set to obtain data for algorithm calculation structure; performing traceability analysis to find out influencing factors; based on the traceability analysis results, summarizing and extracting the top N influencing factors of each group; comparing the test results with known failure modes to determine possible failure causes; based on the feedback of multiple possible failure causes, further testing and diagnosis operations are performed to verify and determine the root cause of failure; and taking appropriate measures based on the root cause of failure.

[0005] However, due to the following problems of semiconductor production data: 1. Data sources are scattered and difficult to integrate: the data format, structure and storage method of existing systems are different, and there is a lack of unified data platform support. Different data sources are difficult to achieve efficient interconnection and fusion, resulting in low data analysis efficiency and easy omission of key information. 2. Long analysis period and slow response speed: the traditional root cause analysis process highly depends on manual participation, engineers need to manually export, clean, process and analyze data, the whole process takes a long time, usually it takes several weeks or even a month to complete a complete RCA analysis, which cannot meet the demand of quickly positioning abnormalities. Therefore, there is an urgent need for a more efficient semiconductor root cause traceability method. SUMMARY

[0006] The purpose of the present application is to provide a big data traceability analysis method, which partially solves or alleviates the above-mentioned deficiencies in the prior art, and can improve the efficiency and accuracy of data analysis. In order to solve the above-mentioned technical problems, the present application specifically adopts the following technical solutions:

[0007] The first aspect of the present application is to provide a semiconductor big data traceability analysis method, comprising the steps of:

[0008] S101, acquiring semiconductor production line data, the semiconductor production line data at least including: (1) a plurality of processing equipment information experienced by a plurality of semiconductor products, (2) process parameters adopted when the semiconductor products experience at least one of the processing equipment, and / or category information of the semiconductor products, (3) measurement values, and the processing equipment information includes at least one of the following: site information, machine table information, chamber information, the process parameters include at least one of the following: formula, sensor type, and the category information includes at least one of the following: raw material source, semiconductor model; S102, calling a data analysis scheme for the semiconductor production line data, the data analysis scheme including: a fault prediction algorithm; S103, preprocessing the semiconductor production line data according to the data analysis scheme to generate a preprocessing result, the preprocessing result including: an abnormal root cause predicted and / or an abnormal weight of the abnormal root cause; wherein S103 includes the steps of: S1031, acquiring an extraction rule, the extraction rule including: device type and configuration type, and the device type including at least one processing equipment information, the configuration type including at least one process parameter information and / or at least one of the category information; S1032, identifying a plurality of production sequences from the semiconductor production line data according to the extraction rule, and the production sequences corresponding to records of the measurement values; S1033, predicting an abnormal root cause according to the difference between at least one production sequence and other production sequences by using the fault prediction algorithm; S104, inputting the preprocessing result and the corresponding semiconductor production line data into a preset big data platform, and the big data platform corresponding to output an analysis result.

[0009] In some embodiments, the fault prediction algorithm includes: a single factor variance analysis algorithm, a history analysis algorithm and / or a similarity analysis algorithm. In some embodiments, the big data platform is an Apache Spark big data platform. In some embodiments, the production sequence at least records the processing sequence of at least two processing equipment experienced by the semiconductor product and the corresponding measurement values; correspondingly, S1033 further includes: S10331, obtaining at least one production sequence group, and the production sequences in the same group have the same device type and configuration type; S10332, calculating the difference between one production sequence and other production sequences in the same group by using the fault prediction algorithm to predict the abnormal root cause.

[0010] In some embodiments, when the failure prediction algorithm is the history analysis algorithm, S10332 comprises the steps of: identifying the data type of the measurement value, wherein the data type comprises discrete type and / or continuous type; selecting a corresponding mutual information algorithm according to the data type; and calculating the mutual information of the production sequence according to the mutual information algorithm , wherein the mutual information is used to represent the amount of information about the feature contained in the random variable Y; and calculating the root cause weight of at least one processing equipment in the production sequence according to the mutual information index.

[0011] In some embodiments, before S10332, there is further included the step of: obtaining the number of sequences within a group of product sequences; and determining whether the number of sequences is greater than a preset number, and if so, proceeding to step S10332.

[0012] In some embodiments, before S1031, there is further included the step of: selecting at least one set of pre-extraction rules from a rule library; generating a plurality of pre-production sequences according to at least a part of the semiconductor production data using the pre-extraction rules; dividing the plurality of pre-production sequences into a plurality of sub-production sequences according to the plurality of pre-production sequences, the sub-production sequences having a preset length; obtaining analysis results of the plurality of sub-production sequences and actual results input by a user; and scoring the pre-extraction rules according to the difference between the analysis results and the actual results.

[0013] In some embodiments, when the data type is discrete, the mutual information is calculated as follows:

[0014] ; wherein, represents the value of the feature , represents the measurement value of the category Y; is the joint probability distribution of the feature and the category ; is the marginal probability distribution of the feature ; is the marginal probability distribution of the category Y; wherein the feature refers to the feature in the i th production sequence, and the feature refers to the equipment type or configuration type; and the category refers to the category of the measurement value of a production sequence.

[0015] In some embodiments, when the data type is continuous, the mutual information index is calculated as follows: ; wherein, represents the value of the feature , represents the measurement value of the category Y; joint probability distribution of features and categories ; marginal probability distribution of features ; marginal probability distribution of categories Y; wherein the features refer to features in the ith production sequence, the features refer to equipment types or configuration types; the categories refer to categories of measurement values of a production sequence.

[0016] In some embodiments, the root cause weight is calculated as follows: ; represents the features of the jth sub-production sequence in the ith production sequence, is the proportion of the number of wafers that have experienced the ith production sequence in the total number of wafers, is the number of production sequences in a product sequence group.

[0017] Another aspect of the present application also provides a semiconductor big data traceability analysis system, comprising:

[0018] a data acquisition module, configured to acquire semiconductor production line data, the semiconductor production line data at least including: (1) a plurality of processing equipment information experienced by a plurality of semiconductor products, (2) process parameters used when the semiconductor products experience at least one of the processing equipment, and / or category information of the semiconductor products, (3) measurement values, and the processing equipment information includes at least one of the following: site information, machine information, chamber information, the process parameters include at least one of the following: recipe, sensor type, and the category information includes at least one of the following: raw material source, semiconductor model;

[0019] a scheme calling module, configured to call a data analysis scheme for the semiconductor production line data, the data analysis scheme including: a fault prediction algorithm;

[0020] a preprocessing module, configured to preprocess the semiconductor production line data according to the data analysis scheme to generate a preprocessing result, the preprocessing result including: an abnormal root cause predicted and / or an abnormal weight of the abnormal root cause; wherein the preprocessing module includes: a rule acquisition unit, configured to acquire extraction rules, the extraction rules including: equipment types and configuration types, and the equipment types including at least one processing equipment information, and the configuration types including at least one process parameter information and / or at least one of the category information; a sequence identification unit, configured to identify a plurality of production sequences from the semiconductor production line data according to the extraction rules; a fault prediction unit, configured to predict an abnormal root cause according to differences between at least one production sequence and other production sequences by using the fault prediction algorithm;

[0021] an analysis module configured to input the pre-processing result and corresponding semiconductor production line data into a preset big data platform, and the big data platform outputs an analysis result.

[0022] Beneficial technical effects: It should be noted that in the process of semiconductor manufacturing, it may need to go through thousands of processes, and the parameters of each process (such as the formula set by the engineer or the working state of the equipment itself) may also differ. Therefore, a semiconductor will record a large amount of production line data. Moreover, once the yield of the semiconductor decreases, it is very difficult to analyze the root cause from the massive production line data, even experienced engineers may need to spend several days or even weeks to locate the fault root cause.

[0023] To this end, the present application proposes a pre-processing and AI processing coordinated big data traceability analysis method (i.e. root cause analysis method), that is, first, through pre-processing, the root cause is roughly located (such as screening out the area where the root cause is highly likely to exist) to obtain the pre-processing result (such as the possible abnormal root cause or the abnormal weight of the abnormal root cause) and the corresponding detailed information (i.e. production line data) input into the preset AI platform, and the production line data is analyzed in depth by the AI platform under the guidance of the pre-processing result. This way of coordinated analysis based on pre-processing result and production line data can effectively improve the efficiency of AI analysis.

[0024] Moreover, it should be noted that the traditional root cause analysis often uses an analysis scheme based on complete process parameters, such as combining complete process parameters, such as FDC data set (which often contains relatively complete device parameters such as temperature, pressure, flow, voltage, and current) to trace the root cause. That is to say, the traditional root cause analysis is a parameter-based analysis.

[0025] On the contrary, the present application uses a non-parametric analysis scheme, such as the present application only needs to extract the production sequence (which does not need to contain detailed process parameters, such as only needs to record the process sequence and process category) from the FDC data set, and predicts the possible root cause type according to the production sequence. This non-parametric analysis scheme helps to achieve faster preliminary analysis and improves analysis efficiency to some extent.

[0026] Further, the preliminary result obtained by the non-parametric analysis means is combined with the corresponding complete data set, and the possible root cause can be analyzed in detail based on the AI platform, thereby efficiently providing the user with a relatively reliable and complete analysis result.

[0027] Specifically, for the non-parametric analysis means, the present application actually provides a piecewise data preprocessing method, specifically, the present application acquires a sub-production sequence with analysis value in a grouping and segmenting manner, and the grouping and segmenting manner comprises the following steps: 1) grouping the semiconductor production line data according to the equipment type and the configuration type, and extracting a production sequence; 2) cutting the production sequence to acquire a plurality of pieces (i.e. sub-production sequences). BRIEF DESCRIPTION OF DRAWINGS

[0028] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, hereinafter, the drawings needed to be used in the embodiments or the prior art description will be briefly introduced. In all the drawings, similar elements or parts are generally identified by similar reference signs. In the drawings, the elements or parts are not necessarily drawn according to the actual proportions. Obviously, the drawings described below are some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0029] Figure 1 Method flowchart in an exemplary embodiment of the present application;

[0030] Figure 2 Production line correlation graph in an exemplary embodiment;

[0031] Figure 3 Product flow direction graph in another exemplary embodiment;

[0032] Figure 4 System module architecture diagram in an exemplary embodiment of the present application. DETAILED DESCRIPTION

[0033] In order to make the purpose, technical solutions and advantages of the embodiments of the present application more clear, hereinafter, the technical solutions in the embodiments of the present application will be clearly and completely described with the drawings in the embodiments of the present application. Obviously, the described embodiments are some of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0034] Herein, the suffix such as "module", "component" or "unit" used to represent an element is only for the convenience of the description of the present application, and has no specific meaning. Therefore, "module", "component" or "unit" can be mixedly used.

[0035] In this document, the terms "upper", "lower", "inner", "outer", "front", "back", "one end", "the other end" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the present application. In addition, the terms "first", "second" are only for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0036] In this document, unless otherwise explicitly specified and limited, the terms "mounting", "provided with", "connection" and the like should be understood broadly, for example, "connection" can be fixed connection, can also be detachable connection, or integral connection; can be mechanical connection, can be direct connection, can also be indirect connection through intermediate medium, can be internal communication of two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0037] In this document, "and / or" includes any and all combinations of one or more of the listed related items.

[0038] In this document, "a plurality of" means two or more, that is, it includes two, three, four, five, etc.

[0039] In this specification, the term "about" typically means + / - 5% of the stated value, more typically + / - 4% of the stated value, more typically + / - 3% of the stated value, more typically + / - 2% of the stated value, even more typically + / - 1% of the stated value, even more typically + / - 0.5% of the stated value.

[0040] In this specification, certain embodiments can be disclosed in a format that is a range. It is to be understood that such a "range" format is merely used for convenience and brevity and should be interpreted in a flexible manner to include not only the numerical values explicitly recited as the limits of the range, but also to include all the individual numerical values or sub-ranges within that range as if each numerical value and sub-range is explicitly recited. For example, a range of 1-6 should be interpreted to include the explicitly recited ranges of, for example, from 1 to 3, from 1 to 4, from 1 to 5, from 2 to 4, from 2 to 6, from 3 to 6, and the like, as well as the individual numbers 1, 2, 3, 4, 5, and 6. This same principle applies to ranges derived from two or more ranges, for example, a range of 1-6 and a range of 4-8 can be interpreted as a range of 1-8.

[0041] In this document, "wafer" refers to a silicon wafer used for the manufacture of silicon semiconductor integrated circuits. Because its shape is usually circular, it is also called wafer, also known as wafer; various circuit element structures can be processed and manufactured on the silicon wafer to become IC products with specific electrical properties.

[0042] Herein, "lot" refers to a basic batch of wafer sets. For example, usually 1 lot is 12 wafers.

[0043] Embodiment One

[0044] Referring to Figure 1 As shown, the present application provides a semiconductor big data traceability analysis method, comprising steps of:

[0045] S101, acquiring semiconductor production line data, wherein the semiconductor production line data at least includes: (1) a plurality of processing equipment information experienced by a plurality of semiconductor products, (2) process parameters adopted by the semiconductor products when experiencing at least one of the processing equipment, and / or category information of the semiconductor products, (3) measurement values, and the processing equipment information includes at least one of the following: site information, machine table information, chamber information, the process parameters include at least one of the following: formula, sensor type, and the category information includes at least one of the following: raw material source, semiconductor model;

[0046] For example, in some embodiments, the semiconductor production line data can come from the FDC system, or can come from the WAT system. Among them, the semiconductor production line data can record the processing information experienced by a semiconductor product in the whole process from entering the production line to leaving the production line, such as the model, category, number, etc. of the processing equipment (such as site, machine table, chamber) experienced, and the process parameters (equivalent to process parameters) of each processing equipment, such as temperature, pressure, voltage, current, etc.

[0047] For example, in some embodiments, the measurement value can be the result of measuring the yield indicators such as structure, size or performance of the semiconductor product after the semiconductor product passes through any one processing equipment.

[0048] Alternatively, in other embodiments, the measurement value can also be the process parameters of the semiconductor product measured when the semiconductor product is processed by any one processing equipment, such as voltage, current and other parameters, which can reflect to some extent whether the equipment is in a healthy running state, or reflect the probability of the existence of defects or faults on the semiconductor product.

[0049] S102, calling a data analysis scheme for the semiconductor production line data, wherein the data analysis scheme includes: a fault prediction algorithm;

[0050] S103, pre-processing the semiconductor production line data according to the data analysis scheme to generate a pre-processing result, wherein the pre-processing result includes: an abnormal root cause predicted, and / or an abnormal weight of the abnormal root cause;

[0051] S104, inputting the preprocessing result and corresponding semiconductor production line data into a preset big data platform, and the big data platform outputs an analysis result.

[0052] In some embodiments, the failure prediction algorithm comprises a single factor variance analysis algorithm, a history analysis algorithm, and / or a similarity analysis algorithm.

[0053] In some embodiments, the big data platform is configured with one or more pre-trained failure prediction AI models (also referred to as an AI platform), which can analyze abnormal points from the semiconductor production line data, thereby predicting the root cause of the failure.

[0054] In some embodiments, the big data platform is an Apache Spark big data platform.

[0055] In this embodiment, the preprocessing result and corresponding semiconductor production line data are synchronously input into the big data platform, so that the big data platform can perform localized focused analysis on the massive semiconductor production line data based on the preprocessing result, thereby improving the efficiency of failure prediction and reducing the requirement for analysis computing power of the massive semiconductor production line data.

[0056] Specifically, the present application proposes a preprocessing and AI processing coordinated big data traceability analysis method (i.e. root cause analysis method), that is, first, the root cause is roughly located (such as filtering out the area where the root cause is highly likely to exist) through preprocessing to obtain a preprocessing result (such as a possible abnormal root cause or an abnormal weight of an abnormal root cause) and corresponding detailed information (i.e. production line data) input into a preset AI platform, and the AI platform performs in-depth analysis on the production line data under the guidance of the preprocessing result. This coordinated analysis based on the preprocessing result and the production line data can effectively improve the efficiency of AI analysis.

[0057] Moreover, it is worth noting that the traditional root cause analysis often adopts an analysis scheme based on complete process parameters, such as combining complete process parameters, such as FDC data set (which often contains relatively complete device parameters such as temperature, pressure, flow, voltage, and current) for root cause tracing. That is, the traditional root cause analysis is a parameter-based analysis.

[0058] On the contrary, the present application adopts a non-parametric analysis scheme, such as the present application only needs to extract the production sequence (which does not need to contain detailed process parameters, such as only needs to record the process sequence and process category) from the FDC data set, and predicts the possible root cause type according to the production sequence. This non-parametric analysis scheme helps to achieve faster preliminary analysis and improves analysis efficiency to some extent.

[0059] Further, the preliminary results obtained by the non-parametric analysis means are combined with the corresponding complete data set, and the AI platform can be used to analyze the possible root causes in detail, thereby efficiently providing relatively reliable and complete analysis results for the user.

[0060] In the following, the non-parametric analysis means proposed by the present application will be described in detail:

[0061] In some embodiments, S103 includes the following steps:

[0062] S1031, obtaining extraction rules, wherein the extraction rules include: device type and / or configuration type, the device type includes at least one piece of processing equipment information, and the configuration type includes at least one piece of process parameter information and / or at least one piece of the category information;

[0063] Preferably, the extraction rules include: device type and configuration type.

[0064] S1032, identifying a plurality of production sequences from the semiconductor production line data according to the extraction rules, wherein the production sequences correspond to the recorded measurement values;

[0065] S1033, using a fault prediction algorithm to predict the abnormal root cause according to the difference between at least one production sequence and other production sequences.

[0066] In this embodiment, the extraction rules can also be regarded as a grouping rule. Data satisfying the same extraction rules is recorded in a production sequence, and a plurality of production sequences satisfying the same rules are regarded as a production sequence group.

[0067] Referring to FIG. 1, Figure 2 When the extraction rules include: site, machine, and chamber, the history path of wafer 1 products that have experienced the same site, machine, and chamber can be extracted as a production sequence, and the history path of wafer 2 products can be extracted as another production sequence. That is, the site, machine, and chamber information also serve as a classification factor.

[0068] It can be understood that the more rules (or classification factors) in the extraction rules, the finer the granularity of the grouping. The extraction rules can be adaptively selected according to actual analysis requirements.

[0069] In some embodiments, the production sequence records at least the processing sequence of at least two processing equipment experienced by the semiconductor product and the corresponding measurement values; correspondingly, S1033 further includes:

[0070] S10331, obtaining at least one production sequence group, and the production sequences in the same group have the same device type and configuration type.

[0071] S10332, calculating a difference between a production sequence and other production sequences in the same group using a fault prediction algorithm to predict an abnormal root cause.

[0072] For example, in some embodiments, a group of production sequences includes abnormal sequences and normal sequences, where an abnormal sequence refers to a production sequence corresponding to a semiconductor product that is not qualified in terms of yield or measurement, and a normal sequence refers to a production sequence corresponding to a semiconductor product that is qualified in terms of yield or measurement. By calculating the difference between the abnormal sequence and the normal sequence in the production condition, the root cause of the problem that may occur can be preliminarily predicted.

[0073] For example, in some embodiments, a production sequence includes a semiconductor product passing through n same type of processing equipment (that is, at least two processing equipment are configured under at least one equipment type). For example, a process can include multiple same stations, such as station 1, station 2, … (that is, station 1 and station 2 can be used to perform the same process). For example, a station can include multiple same parallel chambers, such as chamber 1, chamber 2, and the chamber number of different chambers needs to be recorded. The parameterized analysis method provided by the present application can directly analyze the root cause area (such as the machine that may cause the fault) from the equipment processing sequence, equipment type, and difference, without falling into specific parameters such as voltage and current abnormality identification. The step-by-step analysis method of non-parametric root cause area prediction in cooperation with the AI platform is beneficial to more quickly and accurately analyze the fault.

[0074] In some embodiments, when the fault prediction algorithm is the history analysis algorithm, S10332 includes the following steps:

[0075] (1) identifying the data type of the measurement value, wherein the data type includes discrete type and / or continuous type;

[0076] (2) selecting a corresponding mutual information algorithm according to the data type;

[0077] (3) calculating mutual information of the production sequence according to the mutual information algorithm , wherein the mutual information is used to represent the amount of information about contained in the random variable Y;

[0078] , wherein refers to the characteristics of the production sequence, such as may represent the equipment type or configuration type of the i-th production sequence.

[0079] (4) calculating the root cause weight of at least one processing equipment in the production sequence according to the mutual information.

[0080] For example, in some embodiments, the root cause weight of a production sequence can be calculated according to mutual information, or the root cause weight of a specific processing device in a production sequence can also be calculated according to mutual information.

[0081] In some embodiments, before S10332, further comprising steps of:

[0082] Obtaining the number of sequences in a group of product sequences;

[0083] Determining whether the number of sequences is greater than a preset number, and if so, entering S10332.

[0084] For example, in some embodiments, the preset number is 1.

[0085] In some embodiments, before S1031, further comprising steps of:

[0086] Selecting at least one group of pre-extraction rules from a rule library;

[0087] Generating a plurality of pre-production sequences according to at least part of the semiconductor production data using the pre-extraction rules;

[0088] Dividing the plurality of pre-production sequences into a plurality of sub-production sequences, the sub-production sequences having a preset length;

[0089] Obtaining analysis results of the plurality of sub-production sequences and actual results input by a user;

[0090] Scoring the pre-extraction rules according to differences between the analysis results and the actual results.

[0091] For example, in some embodiments, the preset length can also be selected / updated, wherein the selection / update of the preset length comprises:

[0092] (1) Cutting the pre-production sequences using the preset length and a test length respectively to obtain first and second cutting groups correspondingly; wherein the cutting groups include a plurality of sub-production sequences after cutting; for example, in some embodiments, the preset length can be a default length preset by a user, or the preset length can be a cutting length used in the last cutting of the pre-production sequences. The test length can be a new cutting length obtained by increasing or decreasing the preset length, which can be manually input by a user, or can be randomly generated according to the preset length.

[0093] It should be noted that the cutting of the production sequence in this embodiment is actually equivalent to dividing the production process into a plurality of smaller stages.

[0094] (2) Obtain the first prediction accuracy of the first segmentation group and the second prediction accuracy of the second segmentation group; wherein, the prediction accuracy refers to the degree of difference between the analysis results based on the fault prediction algorithm and the actual results input by the user;

[0095] (3) Calculate the prediction difference between the first prediction accuracy and the second prediction accuracy;

[0096] (4) When the prediction difference is greater than the preset prediction threshold (which can be set or adjusted by the user), proceed to step (5).

[0097] (5) Obtain the first prediction duration corresponding to the first segmentation group and the second prediction duration of the second segmentation group;

[0098] (6) Calculate the duration difference between the first prediction duration and the second prediction duration;

[0099] (7) When the duration difference is less than the preset duration threshold, it is recommended to update the test length to the new preset length.

[0100] It should be noted that, in order to improve the overall efficiency of fault analysis, this invention introduces a segmentation and grouping analysis mode for the early non-parametric analysis process. That is, by dividing a large production sequence into multiple small sequences (i.e., sub-production sequences), the root cause region can be quickly located through the parallel operation of multiple small sequences.

[0101] Meanwhile, this embodiment also provides a restrictive cutting scale adjustment scheme for non-parametric analysis. This size adjustment updates / adjusts the cutting length in a restrictive manner based on prediction time and prediction accuracy. This allows for rapid prediction of areas suspected of having faults within a relatively limited time, while ensuring that the prediction results have a certain degree of reliability.

[0102] For example, for the same type of wafer manufacturing task or the same type of wafer production line, a pre-testing method can be used to screen for suitable dicing sizes. Similarly, before the formal application of this step-by-step analysis method, a suitable dicing size can be screened through an initial trial phase.

[0103] In some embodiments, when the data type is discrete, the mutual information index is calculated as follows: ;

[0104] in, Representation of features Values ​​(such as site category, machine category, etc.). Represents the measured value of category Y; Features and categories a joint probability distribution of the features and the classes, which is used to describe or represent the joint probability distribution of the random variables and the classes the probability law of the simultaneous occurrence. is the marginal probability distribution of the features and the marginal probability distribution is used to describe or represent the prior probability of the specific feature in all possible states after integrating (or ignoring) the influence of all other features and class information. is the marginal probability distribution of the classes Y, and the marginal probability distribution is used to describe or represent the prior distribution or base rate of each class in the population after integrating (or ignoring) the specific values of all other features;

[0105] wherein the feature refers to the feature in the i th production sequence; represents the measurement value of the class Y.

[0106] In some embodiments, when the data type is continuous, the mutual information is calculated as follows:

[0107] ;

[0108] wherein, represents the value of the feature , represents the measurement value; is a joint probability distribution of the features and the classes ; is the marginal probability distribution of the features ; is the marginal probability distribution of the classes Y;

[0109] wherein the feature refers to the feature in the i th production sequence; represents the measurement value of the class Y.

[0110] In some embodiments, the root cause weight is calculated as follows:

[0111] ;

[0112] represents the feature of the j th sub-production sequence (or the j th stage) in the i th production sequence, refers to the mutual information of the j th sub-production sequence in the i th production sequence, is a penalty factor, wherein, represents the wafer proportion (i.e., the proportion of the number of wafers that have undergone the i th production sequence in the total number of wafers), The number of production sequences in a product sequence group.

[0113] It should be noted that the above probability-based failure prediction algorithm is only an exemplary analysis method provided by the present application, and other non-parametric failure analysis methods can also be used according to specific analysis needs.

[0114] In some embodiments, the probability density is estimated using distance information of K-nearest neighbors, and exemplary calculation steps are as follows:

[0115] Given a sample point , the distance of the kth nearest neighbor of the point is calculated;

[0116] The number of other points in this distance range is counted (for estimating local density);

[0117] Using these local density information, the mutual information contribution of each sample point is estimated;

[0118] The final mutual information estimate is obtained by averaging all sample points.

[0119] For example, assume that the collected data has m site groups, each group has s stages (e.g., a sub-production sequence can be divided based on a stage), and each stage has site chamber combinations, the ratio of the number of wafers in the current group to the total number of wafers wafer_ratio, and by calling the mutual information algorithm I, when =1, i.e., all wafers have passed through a chamber, and at this time, it is impossible to separate, .

[0120] First exemplary embodiment:

[0121] Collecting production line data based on user input information, and adding a column value: corresponding to the value information of the wafer user.

[0122] Grouping based on user input grouping fields, each group is analyzed separately, for example, in the case of history chamber analysis, the results of a group are as follows.

[0123]

[0124] Or

[0125]

[0126] History analysis of a group, each wafer is sorted in time from front to back, divided into different stages, and the grouping results about Figure 2 can be shown in the following table:

[0127]

[0128] Furthermore, based on the above grouping results, a difference analysis is performed: If, in any given stage, for continuous values ​​(i.e., values), those with high values ​​appear in one chamber and those with low values ​​in another, and for discrete values, such as 0 values ​​appearing in one chamber and 1 values ​​in another, these values ​​can be completely separated, the greater the likelihood that this situation will be identified as the root cause. This embodiment introduces a mutual information analysis method to measure the difference in the history analysis.

[0129] Second exemplary embodiment:

[0130] The following section uses another example of production line data to illustrate the implementation of the history analysis algorithm:

[0131] Resume Information Form

[0132]

[0133] The path diagram corresponding to the above resume information table is as follows: Figure 3 As shown. STAGE1 contains machine tool M1, and STAGE2 contains all the chambers of machine tool M2. The path from STAGE1 to STAGE2 is shown as follows. Figure 3 The lines in the diagram, VALUE (i.e., the measured value) are discrete here. 0 is GOOD and is marked in green, and 1 is BAD and is marked in red. The flow diagram can intuitively show the path flow of GOOD and BAD wafers.

[0134] Step 1: Before calculating the mutual information between STAGE1 and VALUE, because the STAGE1 column is text type, it needs to be encoded using natural numbers (a simple encoding that maps categorical data to a sequence of natural numbers (0, 1, 2, 3,...)).

[0135]

[0136] The encoded STAGE1 column is as follows:

[0137]

[0138] Correspondingly, = STAGE1_encoding;

[0139] Similarly, calculate =STAGE2_encoding.

[0140]

[0141] The encoded STAGE2 column is listed as:

[0142]

[0143] Second step: Calculate mutual information, y is the VALUE column information, which is a numerical type.

[0144]

[0145]

[0146] Third step: Calculate the weight of the site, assuming that all wafer quantities are 16 pieces, the wafer ratio column of the group is 0.5, and based on the history table, the wafer chamber combination of each stage can be calculated

[0147] According to the weight calculation formula, the weight of the site is as follows:

[0148]

[0149] Finally, all the weight parameters of all sites of all groups are summarized from large to small, and the root cause weight ranking from large to small is obtained. The output result is as follows:

[0150]

[0151] It should be noted that for a 12-inch fab, it involves tb-level data every month. With the help of Apache Spark, a big data analysis engine, batch processing, stream processing, graph computing and machine learning can be supported. The core advantage of Spark is in-memory computing, which is faster than Hadoop, while providing high fault tolerance and easy-to-use API (supporting Scala, Java, Python and R).

[0152] The main modules of the Spark big data platform include SparkCore, SparkSQL, SparkStreaming, MLlib (machine learning) and GraphX (graph computing). In this application, three kinds of traceability analysis algorithms are designed and run on spark, which are suitable for application in FDC, WAT, Inline, wafer history information and other data sources, so that large-scale data analysis becomes possible, and the result generation speed of the fastest minute and the slowest hour is realized.

[0153] ​​​​That is to say, one of the purposes of the present application is to provide a root cause analysis method based on 12-inch semiconductor Fab manufacturing data, to solve the problems of scattered data sources, low analysis efficiency, dependence on manual judgment, limited analysis granularity and the like in the existing technology in the traceability analysis process.

[0154] To achieve the above-mentioned purpose, the present application constructs three kinds of key algorithm models based on the Apache Spark big data platform, including: data grouping analysis algorithm, history analysis algorithm and similarity analysis algorithm, which are respectively applicable to different types of input data (such as discrete labels, continuous output values, abnormal characteristic parameters, etc.), so as to realize unified analysis and efficient traceability of various data sources such as FDC, WAT, Inline and wafer history.

[0155] The present application can complete the automatic root cause identification of large-scale manufacturing data in the time range of minutes to hours, improve the speed and accuracy of root cause positioning, reduce the work intensity of engineers, and provide strong support for process optimization and quality control.

[0156] Embodiment two

[0157] Next, taking the data grouping analysis algorithm as an example, an exemplary analysis method provided by the present application is introduced and described as follows:

[0158] The user gives a batch of WAFER GOOD, BAD labels, and locates the root cause from the selected time range, and the positioning granularity can be adjusted, such as (product + site + machine + Sensor_Name) or (product + site + machine + chamber + Sensor_Name). The specific analysis process includes the following steps:

[0159] Step 1: Select the data source to be analyzed, such as FDC, WAT, Inline, upload wafer information, mark "GOOD", "BAD", select the time range for traceability, and if further refinement is needed, filter through product, site, machine, chamber, and Lot information.

[0160] Step 2: Select the grouping analysis granularity to determine the degree of positioning. For example, for FDC data source, the traceability granularity is selected as three kinds: 1. site + machine + chamber 2. site 3. site + machine, and Parameter is the smallest parameter for traceability. For example, FDC Parameter is SensorName#Window#stats (the statistical characteristics of a group of consecutive points in the specified window, such as'max': maximum value,'min': minimum value,'min': average value). Click to run and submit the analysis task.

[0161] Step 3: The platform submits the task information to the spark big data analysis engine of the platform, calls the data grouping analysis algorithm, and after running, outputs the algorithm output result to the database.

[0162] Step 4: The platform displays the results of the task in the platform result view.

[0163] The core algorithm part is as follows:

[0164] Step 1: Collect data sets: select data from user-specified data sources based on user-provided waferlist, good, and bad label information. Assign values to the 'label' column based on the wafer column, with 0 representing 'GOOD' and 1 representing 'BAD'.

[0165] Step 2: Select product, machine, chamber, site, Parameter, wafer, and 'label' label column data, and remove rows with missing values.

[0166] Step 3: Group based on product + grouping information, score all Parameter for each group, and mark as "weight". For example, group 1, group field is site, machine, and chamber.

[0167]

[0168] Step 4: If there is a significant difference in label grouping and parameter, parameter is more likely to be the root cause. One case of root cause: because a sensor corresponding to a device accessory at a certain moment has an abnormality, causing the collected data to be too high or too low. There is a large separation phenomenon in the form. In post-analysis, single factor variance analysis is used to judge the difference in parameter between good and bad groups of wafer.

[0169] Single factor variance analysis, also known as one-dimensional variance analysis, is used to analyze whether the mean of the dependent variable exists significant difference when different levels of a single control factor are taken. Different levels correspond to the two groups of GOOD and BAD, and the control factor is a single parameter.

[0170] Specifically, the core calculation process of step 4 is as follows:

[0171] First, give the hypothesis definition:

[0172] H0 (Null Hypothesis): The means of all groups are equal (μ1=μ2=...=μ k ), k is the number of groups.

[0173] I1 (Alternative Hypothesis): At least one group mean is different from the others.

[0174] Total Sum of Squares (SST): Sum of squared deviations of all observations from the overall mean:

[0175]

[0176] Xij: jth observation in the ith group.

[0177] X: Overall mean of all data.

[0178] Sum of Squares Between (SSB): Measures the difference between group means and the overall mean:

[0179]

[0180] ni: Number of samples in the ith group.

[0181] mi: Mean of the ith group.

[0182] Sum of Squares Within (SSW): Measures the difference between each data point within a group and the group mean:

[0183]

[0184] Degrees of Freedom:

[0185] Total Degrees of Freedom: (N is the total number of samples).

[0186] Degrees of Freedom Between: (k is the number of groups).

[0187] Degrees of Freedom Within: .

[0188] Mean Square Error (MSE):

[0189] Mean Square Between (MSB): ;

[0190] Mean Square Within (MSW): ;

[0191] Calculate the F-statistic to compare the variance between groups and within groups:

[0192]

[0193] Determine the significance level, set at 0.05, and calculate the p-value, which is calculated based on the F-distribution.

[0194] Cumulative distribution function of F distribution, represents the probability of less than or equal to the calculated F statistics under the given inter-group degrees of freedom, intra-group degrees of freedom:

[0195] ;

[0196] The p value is the probability of the opposite event.

[0197] ;

[0198] Significance determination:

[0199] When , the difference is significant, reject the original hypothesis, and select the alternative hypothesis.

[0200] Otherwise, the difference is not significant, accept the original hypothesis.

[0201] The smaller the p value is in the case of less than 0.05, the greater the difference, and the F statistics value is the greater the better.

[0202] Specifically, in step 3, if the number of rows of "GOOD" or "BAD" in the grouping + Parameter is less than 3, the sample size is too small, the analysis is not reliable, and the root cause weight is set to 0.

[0203] Otherwise, the variance analysis algorithm is called, the total number of groupings is given as m, the total number of parameters in the grouping is n, and the F statistics of all parameters in the grouping is calculated, The F statistics of the i-th grouping and the j-th parameter is as follows:

[0204]

[0205] The proportion of the number of badwafers in the grouping to the total number of badwafers, indicating that the more the number of badwafers in the grouping, the greater the possibility of being a root cause.

[0206] Summarize the weight parameters of all parameters in all groupings, arrange them from large to small, and get the root cause weight ranking from large to small. The output result is as follows:

[0207]

[0208] Example three

[0209] Next, taking the similarity analysis algorithm as an example, the analysis method provided by the present application is described again:

[0210] The analysis method of the embodiment is applicable to a set of abnormal parameters known about the wafer, to other data sources such as FDC, WAT, Inline, to find a set of parameters related to the wafer, whether there is a linear, square, root relationship, as long as one of the three exists, which can be used for the setting of the root cause weight. The algorithm designed here uses a linear regression algorithm, and then evaluates the R2 of the fitting optimization degree as the root cause weight.

[0211] The similarity analysis service analysis steps are as follows:

[0212] Step 1: Select the data source that needs to be compared for similarity analysis, such as FDC, WAT, Inline, select a set of wafer abnormal characteristics, such as: FDC, WAT, Inline, select the time range of the trace, if further refinement is needed, can be filtered through product, site, machine, chamber, Lot information.

[0213] Step 2: Select grouping: the granularity of the analysis, determine the degree of positioning. For example, the FDC data source, the trace granularity is selected as three: 1. Site + machine + chamber 2. Site 3. Site + machine, and Parameter is the smallest parameter of the trace, for example, FDC Parameter is SensorName#Window#stats (the statistical characteristics of a set of consecutive points calculated in the specified set window of the sensor, such as'max': maximum value,'min': minimum value,'min': average value). Click to run, submit the analysis task.

[0214] Step 3: The platform submits the task information to the spark big data analysis engine of the platform, calls the data similarity analysis algorithm, and after the running is completed, outputs the algorithm output result to the database.

[0215] Step 4: The platform displays the results of the task to the platform result view.

[0216] Among them, the specific similarity analysis algorithm can include the following steps:

[0217] Step 1: Data collection: collect the required columns from the data source to be compared for similarity, such as the abnormal characteristic column Inline of a certain Parameter, compared with the FDC data source, for example, the results are as follows:

[0218]

[0219] Step 2: Based on the set comparison data source, such as FDC, locate the grouping field as site+tool+chamber, compare Parameter with abnormal characterization value, group the data of step 2 according to site+tool+chamber, and detect whether there is a linear relationship, a square relationship, or a square root relationship. Train a linear regression on the Parameter and abnormal characterization value column to evaluate the goodness of fit between the true value and the predicted value as the similarity.

[0220] The principle of step 2 is as follows:

[0221] Linear relationship detection: using linear regression, given , the last column is the intercept column, the sample size is N, the number of features is d, the slope vector (including intercept), the target vector , is the intercept term, and the model is constructed as follows: ;

[0222] where is the linear regression model, X is the FDC / WAT / Inline parameter to be detected (such as the parameter column of FDC in the group), and R is a real number.

[0223] The loss function is: ;

[0224] The optimal solution is obtained by minimizing the loss: ;

[0225] where Y represents the selected group of wafer abnormal characterization values, and the value column in the reference table.

[0226] Square relationship detection:

[0227] Consider the simplest case, d=1, only one-dimensional feature case x={ };

[0228] Then bring it into the linear regression to get the optimal solution.

[0229] Square root relationship detection: is equivalent to and exists in a linear case. If x is less than 0, transform x by subtracting the minimum value of x from each element of x, so that x is greater than or equal to 0. Then bring it into the linear regression model.

[0230] Method for obtaining the weight of root cause analysis:

[0231] Given one-dimensional input x and one-dimensional y.

[0232] Detect linear relationship, build linear regression model of y and x;

[0233] Detect square relationship: build regression model of y and x, ;

[0234] Detect square root relationship: build regression model of and x.

[0235] Calculate goodness of fit of predicted value and true value in turn, which represents the degree of explanation of the change of target variable by the model

[0236] ;

[0237] Wherein, is the predicted value of the model, is the mean of the target variable. The determination coefficient is better, the determination coefficient Equal to 0, equivalent to mean prediction, the determination coefficient Less than 0, indicating that the model performs worse than "predicting with mean".

[0238] For the determination coefficient less than 0, change to 0, then find the maximum Among the three, as the similarity.

[0239] Assume that the total data has m groups, each group has Parameter, call the algorithm based on linear, square, square root relationship, one of the three meets the abnormal,

[0240] If the sample size is less than or equal to 3, the sample size is too small at this time, which is not reliable, 0;

[0241] Otherwise, call the similarity algorithm to calculate:

[0242] ;

[0243] is the linear regression model, square regression model, square root regression model of training, is the goodness of fit formula.

[0244] Step 3: aggregate the weight parameters of all sites of all groups from large to small, and get the root cause weight ranking from large to small. For example, the output result is as follows:

[0245]

[0246] It needs to be explained that the Fab factory produces a large amount of data in the manufacturing process, and the types are various; on the other hand, the traditional analysis method and technical architecture are difficult to effectively cope with such high-dimensional, heterogeneous and real-time data challenges.

[0247] The present application is just for such high-dimensional, heterogeneous and real-time data challenges, and provides an improved solution to overcome the many shortcomings in the prior art, and provides a large-scale data traceability analysis method suitable for 12-inch Fab factory.

[0248] Or, the present application provides a root cause analysis method based on 12-inch semiconductor Fab factory manufacturing data to solve the problems of scattered data sources, low analysis efficiency, dependence on manual judgment and limited analysis granularity in the prior art in the traceability analysis process.

[0249] To achieve the above-mentioned purpose, the present application constructs three kinds of key algorithm models based on ApacheSpark big data platform, including: data grouping analysis algorithm, history analysis algorithm and similarity analysis algorithm, which are respectively suitable for different types of input data (such as discrete labels, continuous output values, abnormal characteristic parameters, etc.), so as to realize the unified analysis and efficient traceability of FDC, WAT, Inline and wafer history and other data sources.

[0250] The present application can complete the automatic root cause identification of large-scale manufacturing data in the time range of minutes to hours, improve the speed and accuracy of root cause positioning, reduce the work intensity of engineers, and provide strong support for process optimization and quality control.

[0251] Embodiment four

[0252] Referring to Figure 4 The present application also provides a semiconductor big data traceability analysis system, which comprises:

[0253] The data acquisition module 101 is used for acquiring semiconductor production line data, and the semiconductor production line data at least includes: (1) a plurality of processing equipment information experienced by a plurality of semiconductor products, (2) process parameters adopted when the semiconductor products experience at least one processing equipment, and / or category information of the semiconductor products, (3) measurement values, and the processing equipment information includes at least one of the following: site information, machine table information, chamber information, the process parameters include at least one of the following: formula, sensor type, and the category information includes at least one of the following: raw material source, semiconductor model;

[0254] The scheme calling module 102 is used for calling a data analysis scheme for the semiconductor production line data, and the data analysis scheme includes a fault prediction algorithm.

[0255] The preprocessing module 103 is configured to preprocess the semiconductor production line data according to a data analysis scheme to generate a preprocessing result, wherein the preprocessing result comprises a predicted abnormal root cause and / or an abnormal weight of the abnormal root cause; and the preprocessing module comprises:

[0256] The rule acquisition unit 1031 is configured to acquire an extraction rule, wherein the extraction rule comprises a device type and a configuration type, the device type comprises at least one piece of processing device information, and the configuration type comprises at least one piece of process parameter information and / or at least one piece of the category information;

[0257] The sequence identification unit 1032 is configured to identify a plurality of production sequences from the semiconductor production line data according to the extraction rule;

[0258] The fault prediction unit 1033 is configured to predict an abnormal root cause according to a difference between at least one production sequence and other production sequences by using a fault prediction algorithm.

[0259] The analysis module 104 is configured to input the preprocessing result and corresponding semiconductor production line data into a preset big data platform, and the big data platform outputs an analysis result.

[0260] It should be noted that the system in the embodiment can implement the method or steps in any one of the above embodiments, which will not be described here.

[0261] It should be noted that in this document, the term "comprising" or "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or apparatus including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such a process, method, article or apparatus. Without more limitations, the element defined by the statement "comprising a" does not exclude the presence of additional identical elements in the process, method, article or apparatus including the element.

[0262] From the above description of the embodiments, those skilled in the art can clearly understand that the above-mentioned embodiment method can be realized by means of software and necessary general hardware platform, of course, it can also be realized by hardware, but in many cases, the former is a better embodiment. Based on such understanding, the technical solutions of the present application can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes a plurality of instructions for making a computer terminal (which can be a mobile phone, computer, server, or network device, etc.) execute the method described in each embodiment of the present application.

[0263] The embodiments of the present application are described above with reference to the accompanying drawings, but the present application is not limited to the above-described specific embodiments, and the above-described specific embodiments are merely illustrative, but not restrictive, and a person of ordinary skill in the art can make many forms under the inspiration of the present application without departing from the purpose of the present application and the scope protected by the claims, and these all belong to the protection of the present application.

Claims

1. A semiconductor big data traceability and analysis method, characterized in that, Including the following steps: S101, acquire semiconductor production line data, wherein the semiconductor production line data includes at least: (1) information on multiple processing equipment that multiple semiconductor products have undergone, (2) process parameters used when the semiconductor products undergo at least one of the processing equipment, and / or category information of the semiconductor products, (3) measurement values, and the processing equipment information includes at least one of the following: site information, machine information, chamber information, the process parameters include at least one of the following: formula, sensor type, and the category information includes at least one of the following: raw material source, semiconductor model; S102, invoke a data analysis scheme for the semiconductor production line data, the data analysis scheme including: a fault prediction algorithm; S103, preprocess the semiconductor production line data according to the data analysis scheme to generate preprocessing results, the preprocessing results including: predicted root causes of anomalies and anomaly weights of the root causes of anomalies; wherein, S103 includes the following steps: S1031, Obtain extraction rules, the extraction rules include: equipment type and configuration type, wherein the equipment type includes at least one piece of processing equipment information, and the configuration type includes at least one piece of process parameter information and / or at least one piece of category information; S1032, Multiple production sequences are identified from the semiconductor production line data according to the extraction rules, and the production sequence is correspondingly recorded with the measured value; S1033, a fault prediction algorithm is used to predict the root cause of the anomaly based on the difference between at least one production sequence and multiple other production sequences; the production sequence records the processing order of at least two processing devices through which the semiconductor product has undergone, and the corresponding measurement values; correspondingly, S1033 further includes: S10331, Obtain at least one production sequence group, and the production sequences in the same group have the same equipment type and configuration type; S10332, A fault prediction algorithm is used to calculate the difference between a production sequence and multiple other production sequences in the same group to predict the root cause of the anomaly; S104, the preprocessing results and the corresponding semiconductor production line data are input into a preset big data platform, and the big data platform outputs the corresponding analysis results.

2. The method according to claim 1, characterized in that, The fault prediction algorithm includes: one-way ANOVA algorithm, history analysis algorithm and / or similarity analysis algorithm; And / or, the big data platform is the Apache Spark big data platform.

3. The method according to claim 2, characterized in that, When the fault prediction algorithm is a history analysis algorithm, S10332 includes the following steps: Identify the data type of the measured value, wherein the data type includes: discrete and / or continuous; Select the corresponding mutual information algorithm based on the data type; The mutual information of the production sequence is calculated based on the mutual information algorithm. The mutual information is used to represent the information contained in the random variable Y about the features. The amount of information; The root cause weight of at least one processing device in the production sequence is calculated based on the mutual information index.

4. The method according to claim 2, characterized in that, Before S10332, the following steps are also included: Get the number of sequences within a set of product sequences; Determine whether the number of sequences is greater than a preset number. If so, proceed to step S10332.

5. The method according to claim 1, characterized in that, Before S1031, the following steps are also included: Select at least one set of pre-extracted rules from the grouping rule base; Multiple pre-production sequences are generated based on at least a portion of the semiconductor generation data using the pre-extraction rules. Multiple pre-production sequences are divided into multiple sub-production sequences, and each sub-production sequence has a preset length. Obtain the analysis results of multiple sub-production sequences and the actual results input by the user; The pre-extraction rules are scored based on the difference between the analysis results and the actual results.

6. The method according to claim 3, characterized in that, When the data type is discrete, the mutual information is calculated as follows: ； in, Representation of features The value, Represents the measured value of category Y; Features and categories The joint probability distribution of ; Features The marginal probability distribution; The marginal probability distribution of category Y; Among them, features This refers to a feature in the i-th production sequence, where the feature refers to the device type or configuration type; category This refers to the category of measurements for a production sequence.

7. The method according to claim 3, characterized in that, When the data type is continuous, the mutual information index is calculated as follows: ； in, Representation of features The value, Represents the measured value of category Y; Features and categories The joint probability distribution of ; Features The marginal probability distribution; The marginal probability distribution of category Y; Among them, features This refers to a feature in the i-th production sequence, where the feature refers to the device type or configuration type; category This refers to the category of measurements for a production sequence.

8. The method according to claim 6 or 7, characterized in that, The root cause weights are calculated as follows: ； This represents the characteristics of the j-th sub-production sequence within the i-th production sequence. This represents the percentage of wafers that have undergone the i-th production sequence out of the total number of wafers. The number of production sequences in a product sequence group.

9. A semiconductor big data traceability and analysis system, characterized in that, include: The data acquisition module is used to acquire semiconductor production line data, which includes at least: (1) information on multiple processing equipment that multiple semiconductor products have undergone, (2) process parameters used when the semiconductor products undergo at least one of the processing equipment, and / or category information of the semiconductor products, (3) measurement values, and the processing equipment information includes at least one of the following: site information, machine information, chamber information, the process parameters include at least one of the following: formula, sensor type, and the category information includes at least one of the following: raw material source, semiconductor model; The scheme invocation module is used to invoke a data analysis scheme for the semiconductor production line data, and the data analysis scheme includes: a fault prediction algorithm; A preprocessing module is used to preprocess the semiconductor production line data according to a data analysis scheme to generate preprocessing results. The preprocessing results include: predicted root causes of anomalies and anomaly weights for these root causes. The preprocessing module includes: The rule acquisition unit is used to acquire extraction rules, the extraction rules including: device type and configuration type, wherein the device type includes at least one piece of processing equipment information, and the configuration type includes at least one piece of process parameter information and / or at least one piece of category information; A sequence recognition unit is used to identify multiple production sequences from the semiconductor production line data according to extraction rules; The fault prediction unit is used to predict the root cause of anomalies based on the differences between at least one production sequence and multiple other production sequences using a fault prediction algorithm. The production sequence records at least the processing order of at least two processing devices through which the semiconductor product undergoes processing, as well as the corresponding measurement values; correspondingly, the step of using a fault prediction algorithm to predict the root cause of the anomaly based on the difference between at least one production sequence and multiple other production sequences includes: obtaining at least one production sequence group, wherein the production sequences in the same group have the same equipment type and configuration type; and using a fault prediction algorithm to calculate the difference between one production sequence and multiple other production sequences in the same group to predict the root cause of the anomaly. The analysis module is used to input the preprocessing results and the corresponding semiconductor production line data into a preset big data platform, and the big data platform outputs the analysis results accordingly.

Images