Data classification method and system for machine

Abstract

The invention relates to the technical field of wafer data processing, in particular to a data classification method and system for a machine, and the method comprises the steps: obtaining data points reflecting the working indexes of the machine, fitting a plurality of first numerical curves according to the data points, and carrying out the data processing of the first numerical curves to obtain a plurality of second numerical curves; a first-class curve is selected from the multiple second numerical value curves, and the similarity between the first-class curve and the first reference curve is smaller than or equal to a first similarity threshold value; obtaining a plurality of fault types, and determining a plurality of corresponding data change types according to the fault types; and performing a second classification operation on the plurality of first-class curves according to the plurality of data change types so as to divide the first-class curves into a plurality of curve data sets. The invention provides a multi-stage classification method based on curve change characteristics, which is used for carrying out preprocessing and key classification on massive data points, so that a wafer engineer can conveniently carry out targeted intervention analysis on the data points.

Description

Technical Field

[0001] The present invention relates to the technical field of wafer manufacturing, and in particular to a data classification method and system for a machine. Background Art

[0002] During the manufacturing process of semiconductor devices, each process may cause some unexpected structures on the wafer. Among them, the defects that cause the circuits on the chip to not work properly are called wafer defects. In order to ensure the wafer yield and production capacity, the maintenance of wafer processing equipment becomes very important.

[0003] In wafer processing, the quality of the wafer needs to be controlled. When defective products appear, the wafer defect data needs to be analyzed and processed. In the process of quality control, the operating status of the machine is usually monitored. For example, when a certain working parameter of the machine exceeds the set upper threshold or falls below the lower threshold, it means that the machine may have an operating failure. However, on the one hand, the traditional threshold monitoring method has huge data monitoring pressure, and often requires traversing and comparing millions of data points; on the other hand, this threshold monitoring method also has great limitations in monitoring capabilities. It can often only be detected after the fault is generated, that is, there will be a certain delay in detection.

[0004] In order to improve the accuracy of data monitoring, a method of processing raw data by using frequency domain algorithm or statistical probability distribution is proposed in the prior art.

[0005] For example, patent application CN111223799A discloses a process control method, device, system and storage medium, which obtain the original data of the sensor; perform feature extraction on the original data to obtain feature data; perform correlation analysis on the feature data and the measurement data of the detection equipment to obtain target data; input the target data into an error detection and classification FDC system so that the control strategy is output through the FDC system; and perform process control of the machine according to the control strategy.

[0006] For example, patent application CN113255840A discloses a fault detection and classification method, device, system and storage medium. The method attempts to apply frequency domain algorithm and / or statistical probability distribution algorithm to extract features from the segmented data to obtain feature data that is correlated with at least one characteristic of the product; and based on the feature data, perform fault detection and classification on the process used for the product.

[0007] However, when faced with millions or even hundreds of millions of data points, this data processing method will face tremendous data processing pressure.

[0008] Therefore, there is an urgent need for a method that can reduce data processing pressure and processing costs. Summary of the invention

[0009] The purpose of the present invention is to provide a data classification method and system for a machine, which partially solves or alleviates the above-mentioned deficiencies in the prior art and can pre-classify the massive data of the machine so that wafer engineers can intervene in the processing in a targeted manner to more promptly and accurately discover possible fault problems that may occur during the operation of the machine.

[0010] In order to solve the above-mentioned technical problems, the present invention specifically adopts the following technical solutions: A first aspect of the present invention is to provide a data classification method for a machine, comprising the steps of: S100, acquiring a plurality of data points from at least one machine, wherein the plurality of data points are used to reflect at least one working indicator of the machine; S101, fitting a plurality of first numerical curves corresponding to the machine according to the plurality of data points; S102, performing data processing on the plurality of first numerical curves to obtain a plurality of second numerical curves correspondingly, wherein the data processing method comprises: a first processing method, the first processing method is used to calculate the change trend of the first numerical curve, and / or a second processing method, the second processing method is used to amplify the change trend of the first numerical curve; S103, performing a first classification operation on the plurality of second value curves to select a type of curves from the plurality of second value curves, wherein the similarity between the type of curves and the pre-stored first reference curve is less than or equal to a set first similarity threshold; S104, obtaining a plurality of fault types, and determining a plurality of corresponding data change types respectively according to the fault types, wherein the fault types include: a first typical fault type, the data change type is used to describe a first characteristic trend of data changes in the first type of curve, and one of the first typical fault types corresponds to at least one of the data change types; S105, performing a second classification operation on the plurality of the first-class curves according to the plurality of data change types to classify the first-class curves into a plurality of curve data sets, wherein one curve data set includes: a plurality of the first-class curves, and the plurality of the first-class curves have the same or similar data change trends.

[0011] In some embodiments, one of the first typical fault types corresponds to at least one test result, and the method further includes: S106, calculating a plurality of correlation number sets between at least one of the curve data sets and at least one of the test results; wherein one of the curve data sets forms one of the correlation number sets corresponding to one of the test results, and the correlation number set includes: a plurality of correlations between a plurality of the one type of curves and the test results respectively calculated by a correlation calculation method; S107, calculating the difference between at least two of the correlations in the correlation data set; S108: When the difference is greater than a first set difference, a first prompt signal is output to prompt the user to perform a reclassification operation on the curve data set.

[0012] In some embodiments, one of the first typical fault types corresponds to at least one test result, comprising the steps of: Calculating the correlation between the type of curve and the test result; wherein, when the correlation is lower than a set first correlation value, it is considered that the type of curve is not correlated with the test result, otherwise, it is considered that the type of curve is correlated with the test result; When the number of a type of curves related to the test result is less than a set first correlation amount, the corresponding test result is marked as an unknown result, and a second prompt signal is issued accordingly to prompt the user to correct the first numerical curve.

[0013] In some embodiments, the step of correcting the first numerical curve includes: (1) obtaining the first numerical curve processed by the first type of processing method, and processing the first numerical curve by using the second type of processing method to obtain a new second numerical curve; (2) determining whether the similarity between the new second numerical curve and the second reference curve is less than or equal to a set second similarity threshold; If not, proceed to (3); (3) calculating the correlation between the second numerical curve and the test result marked as an unknown result; (4) If the correlation is greater than a set second correlation value, the second numerical curve is classified as a new curve data set.

[0014] In some embodiments, the steps include: The fault type of the new curve data set is obtained, and the fault type is marked as a second typical fault type.

[0015] In some embodiments, the steps include: The data change type corresponding to the second typical fault type is recorded to update the fault type and the data change type in S104.

[0016] In some embodiments, the method includes the step of selecting a first typical fault type from a historical database of the machine.

[0017] In some embodiments, the first type of processing methods includes one or more of the following: calculus processing; and / or, the second type of processing methods includes one or more of the following: exponential processing.

[0018] In some embodiments, the steps include: A corresponding data processing method is selected according to a data label of a type of curve; wherein the data label includes: a machine type, or a typical fault type of a machine; and the machine type or the typical fault type is correspondingly marked with at least one preferred data processing method.

[0019] The present invention also provides a data classification system for a machine, comprising: A data acquisition module, used to acquire a plurality of data points from at least one machine, wherein the plurality of data points are used to reflect at least one working indicator of the machine; A data fitting module, used for fitting a plurality of first numerical curves corresponding to the machine according to the plurality of data points; a data processing module, configured to perform data processing on the plurality of first numerical curves to obtain a plurality of second numerical curves correspondingly, wherein the data processing method comprises: a first type of processing method, the first type of processing method is used to calculate the change trend of the first numerical curve, and / or a second type of processing method, the second type of processing method is used to amplify the change trend of the first numerical curve; a first classification module, configured to perform a first classification operation on the plurality of the second value curves, so as to select a type of curves from the plurality of the second value curves, wherein the similarity between the type of curves and the pre-stored first reference curve is less than or equal to a set first similarity threshold; A type determination module, used to obtain multiple fault types, and determine corresponding multiple data change types according to the fault types, wherein the fault types include: a first typical fault type, the data change type is used to describe a first characteristic trend of data changes in the first type of curve, and one of the first typical fault types corresponds to at least one of the data change types; The second classification module is used to perform a second classification operation on the multiple first-class curves according to the multiple data change types to divide the first-class curves into multiple curve data sets, and one curve data set includes: multiple first-class curves, and the multiple first-class curves have the same or similar data change trends.

[0020] Beneficial technical effects: In view of the massive data points outputted from the wafer production line, the present invention proposes a multi-level verification and classification mechanism for the data based on one or two types of processing methods.

[0021] First, the present invention proposes a collaborative classification method based on curve types and typical fault types to reasonably set the classification degree. Therefore, the reasonable setting of the classification degree can avoid over-classification, that is, avoid prematurely ignoring the possible associations between different curve data sets, increase the difficulty for later wafer engineers to accurately locate faults, or prolong the judgment time of related faults; on the other hand, it can also avoid too low a classification degree (equivalent to invalid classification), resulting in wafer engineers still needing a large degree of manual intervention to judge the correlation between data and perform manual classification, which will increase the work pressure of wafer engineers and put higher requirements on the engineering experience of wafer engineers. For example, for some relatively junior wafer engineers, it is very likely that they will produce wrong conclusions in data processing.

[0022] Furthermore, the present invention also provides a classification mechanism with autonomous updating of classification degree. For example, the present invention can evaluate the correlation of test results within the curve data set to determine whether the preliminary classification results need to be disassembled again, such as dividing the curve data set into multiple sub-curve data sets. For another example, the present invention can also evaluate the correlation between all curves of a class (i.e., all curve data sets) and the test results to determine whether there are any omissions. In other words, the present invention provides a collaborative classification scheme based on three dimensions, namely, fault type, curve change type, and fault test results, to comprehensively evaluate the classification degree and classification accuracy.

[0023] Furthermore, the present invention can also analyze the accuracy and applicability of the real-time classification degree, that is, obtain the latest typical fault types, so as to timely adjust the classification degree (that is, the number of curve data sets divided in S104) in the next classification process, thereby ensuring that the classification mechanism has a high degree of adaptability to the actual operation of the wafer production line. In other words, this classification mechanism with autonomously updated classification degree also makes the method more versatile when facing different production lines or different production stages of the same production line. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, each element or part is not necessarily drawn according to the actual scale. Obviously, the drawings described below are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings without paying creative labor.

[0025] Figure 1 is a flow chart of a classification method in a first exemplary embodiment of the present invention; Figure 2 is a flow chart of a classification method in a second exemplary embodiment of the present invention; Figure 3 is a schematic diagram of a method flow chart of a reprocessing step in a third exemplary embodiment of the present invention; Figure 4 It is a schematic diagram of the module structure of a classification system in an exemplary embodiment of the present invention. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0027] Herein, suffixes such as "module", "component" or "unit" used to represent elements are only used to facilitate the description of the present invention, and have no specific meanings by themselves. Therefore, "module", "component" or "unit" can be used mixedly.

[0028] In this document, the terms "upper", "lower", "inner", "outer", "front", "back", "one end", "the other end" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a specific orientation, be constructed and operate in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance.

[0029] In this document, unless otherwise clearly specified and limited, the terms "installed", "provided with", "connected", etc. should be understood in a broad sense. For example, "connected" can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection, a direct connection, or an indirect connection through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0030] Herein "and / or" includes any and all combinations of one or more of the associated listed items.

[0031] Herein, "plurality" means two or more than two, ie, it includes two, three, four, five, etc.

[0032] As used 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, and even more typically + / - 0.5% of the stated value.

[0033] In this specification, some embodiments may be disclosed in a format of being in a range. It should be understood that this description of "being in a range" is only for convenience and brevity, and should not be interpreted as a rigid limitation on the disclosed range. Therefore, the description of the range should be considered to have specifically disclosed all possible sub-ranges and independent numerical values ​​within this range. For example, the description of the range 1-6 should be considered to have specifically disclosed sub-ranges such as from 1 to 3, from 1 to 4, from 1 to 5, from 2 to 4, from 2 to 6, from 3 to 6, etc., as well as individual numbers within this range, such as 1, 2, 3, 4, 5 and 6. Regardless of the breadth of the range, the above rules apply.

[0034] In this article, "machine (or processing machine)" is also referred to as equipment (or processing equipment), which can refer to any one or more processing modules or processing devices on the wafer production processing line.

[0035] In this article, "operating index" refers to the state parameter output (or collected) by the machine during the process of processing the wafer. For example, the state parameter can be the operating current and operating voltage of the machine itself. Correspondingly, the "first numerical curve" can be a numerical curve of the current of the machine changing over time, or the "first numerical curve" can be a numerical curve of the voltage of the machine changing over time. For another example, the state parameter can be the pressure applied to the wafer when the machine processes the wafer. For another example, the state parameter can also refer to the operating temperature of the machine, etc. These operating indicators are usually the original data of the machine operation.

[0036] In this article, "failure" can also be referred to as machine failure or equipment failure, which refers to equipment or process problems that may affect the processing, testing and other processes of the wafer, and thus affect the yield or production capacity of the wafer.

[0037] For example, the fault may be a mechanical fault, such as a scratch defect on the wafer due to the looseness of the machine's robotic arm. For example, due to the poor voltage stabilization effect of the machine, the voltage fluctuation of the wafer may be too large, which in turn causes corresponding defects in the circuit structure formed on the wafer. For example, due to the failure of the cleaning machine, the cleaning agent may remain, which will form white impurities on the wafer.

[0038] For another example, the failure may also be a process failure, such as insufficient wafer production capacity due to a problem in the processing recipe setting of a machine.

[0039] Taking current as an example, the processing of work indicators is explained: in a wafer production line, the wafer may go through about 800 process steps, and each process step is executed by a corresponding machine. Correspondingly, after the wafer is processed by the production line, a machine will output at least tens of thousands of current data points, and the entire production line will output at least millions of data points. When these data points are abnormal, it may indicate that the machine has failed, which will also cause corresponding defects on the wafer, affecting the yield. Therefore, the analysis of data points helps wafer engineers understand the operating status of the production line. However, millions or even hundreds of millions of data points will bring great data processing pressure to wafer engineers.

[0040] In this regard, the present invention proposes a data pre-processing method, that is, to pre-classify and process the massive data in a reasonable manner. Specifically, the present invention classifies the massive data points generated during the wafer processing process according to the actual operation of the wafer plant through classification, matching and other mechanisms, so that subsequent engineers can, under the guidance of preliminary classification, use professional experience to perform more refined data analysis and processing on each data set (such as using the wafer data analysis platform or manual experience to conduct targeted analysis of the causes and repair solutions of the faults), so as to reduce the data processing pressure of engineers and reduce the monitoring cost of the wafer production line.

[0041] In other words, the present invention proposes a method for simplifying the processing of massive data points.

[0042] Embodiment 1 First, see Figure 1 As shown, the present invention provides a data classification method for a machine, comprising the steps of: S100, acquiring a plurality of data points from at least one machine, wherein the plurality of data points are used to reflect at least one working indicator of the machine.

[0043] In some embodiments, the data points may be collected via FDC (Failure Date Collection).

[0044] S101, fitting a plurality of first value curves corresponding to the machine according to a plurality of the data points.

[0045] S102, performing data processing on the plurality of first numerical curves to obtain a plurality of second numerical curves correspondingly, wherein the data processing method comprises: a first type of processing method, the first type of processing method is used to calculate the change trend of the first numerical curve, and / or a second type of processing method, the second type of processing method is used to amplify the change trend of the first numerical curve.

[0046] S103: Perform a first classification operation on the plurality of second value curves to select a type of curve from the plurality of second value curves.

[0047] In some embodiments, the second numerical curve is classified into a first-category curve and a second-category curve by comparing the similarity between the second numerical curve and the reference value.

[0048] For example, in some embodiments, the similarity between the second numerical curve and the reference value is calculated, and the second numerical curve whose similarity is less than or equal to the set similarity threshold is classified into one type of curve. Conversely, when the similarity is greater than the similarity threshold, it is classified into two types of curves.

[0049] For example, in some embodiments, the reference value can be set to a first reference curve, which is used to reflect the fluctuation trend of conventional working indicators, and can be a reference data pre-set by wafer engineers in combination with the actual operation of the wafer production line. Correspondingly, the similarity between the first type of curve and the pre-stored first reference curve is less than or equal to the set first similarity threshold; the similarity between the second type of curve and the first reference curve is greater than the first similarity threshold.

[0050] In this embodiment, the second numerical curve can be preferably classified into a first category according to the first reference curve, that is, a first category curve with abnormal fluctuation and a second category curve with normal fluctuation. For example, the second category curve refers to a second numerical curve that has no fluctuation or a very small fluctuation difference compared with the first reference curve.

[0051] Alternatively, in some embodiments, the reference value may be set to one or more detection value points at one or more times, and the curve classification is performed by comparing the similarity between the multiple detection value points and the second value curve (such as the difference between the value point and the corresponding value on the second value curve). The similarity may be a statistical result of the similarity calculated from the multiple detection data points, such as the statistical result may be the mean, maximum value, median, mode, total number, etc. of multiple data (such as multiple similarities).

[0052] S104, obtaining a plurality of fault types, and determining a plurality of corresponding data change types respectively according to the fault types, wherein the fault types include: a first typical fault type, the data change type is used to describe a first characteristic trend of data changes in the first type of curve, and one of the first typical fault types corresponds to at least one of the data change types; S105, performing a second classification operation on the plurality of the first-class curves according to the plurality of data change types, so as to classify the first-class curves into a plurality of curve data sets, wherein one curve data set includes: a plurality of the first-class curves, and the plurality of the first-class curves have the same or similar data change trends (or data change types). For example, in some embodiments, the second numerical curves conforming to the same data change type are classified into one curve data set.

[0053] Correspondingly, based on the two classifications, on the one hand, most of the invalid data can be filtered out, and on the other hand, a class of curves in the same curve data set can have similar characteristics (or, in other words, they are likely to originate from the same or a class of faults), so it is also convenient to assign them to different wafer engineers for subsequent analysis.

[0054] In some embodiments, the first typical fault type may be a fault that occurs frequently in the wafer production line and has a certain degree of universality. In addition, corresponding to these known faults, the fluctuation of the working indicators caused by them also has certain characteristics, and these characteristics are recorded as different data change types to facilitate secondary classification of a type of curve with abnormal fluctuations.

[0055] Specifically, the present invention divides the second numerical curve into different curve data sets according to possible fault types through the first classification operation and the second classification operation, thereby facilitating more reasonable differentiation and classification of massive data points through the coordinated application of curve types and fault types.

[0056] In other words, through the collaborative classification of curve types and typical fault types, the classification degree can be reasonably set. Therefore, the reasonable setting of the classification degree can avoid over-classification, that is, avoid prematurely ignoring the possible associations between different curve data sets, which increases the difficulty for later wafer engineers to accurately locate faults; on the other hand, it can also avoid too low a classification degree (equivalent to invalid classification), which results in wafer engineers still needing a large degree of manual intervention to judge the correlation between data and perform manual classification, which will also put higher requirements on the engineering experience of wafer engineers.

[0057] In particular, the present invention uses typical fault types in conjunction with the second numerical curve to complete classification analysis of data points, which can ensure that the classification degree and classification results have better versatility and reliability in multiple application scenarios (such as different wafer production lines).

[0058] Preferably, in some embodiments, the first typical fault type may include one or more of the following types: cleaning agent residue fault, voltage regulation fault, probe fault, etching byproduct fault, etc. The applicant has noticed that these types of faults are somewhat common in wafer processing plants, and therefore, they are preferably set as initial classification conditions for curve classification.

[0059] In some embodiments, the probe failure may be usually caused by voltage instability or aging of the power supply line. Specifically, there may be a specific fluctuation trend (or characteristic trend) on the second value curve of the voltage or current.

[0060] In some embodiments, when there are certain errors in the cleaning process (such as temperature, pressure, cleaning valve flow, time and other cleaning parameters), the cleaning effect may be poor, which may cause cleaning agent residue on the wafer and cause cleaning agent residue failure. Correspondingly, the numerical curves of temperature, pressure, cleaning valve flow, residual water pH value and so on may also form specific fluctuation trends.

[0061] For example, when the nozzle of the cleaning valve is clogged or filled, it will form a specific trend on the flow curve (such as the flow gradually decreases to a certain flow value and maintains the flow value unchanged).

[0062] Preferably, these selected typical faults usually cause fluctuations with a certain degree of recognition on the first numerical curve or the second numerical curve. For example, the fluctuation may be reflected in that the value is far above or below the reference value for a long time. For another example, the fluctuation may be reflected in that the value fluctuates up and down periodically over a period of time. For another example, the fluctuation may be reflected in that the trend of change of the value changes periodically over a period of time, such as the slope of the numerical curve formed by it fluctuates up and down over a period of time. In this embodiment, the data fluctuations are distinguished and classified by the graphical change characteristics reflected on the numerical curve.

[0063] Preferably, in some embodiments, the first typical fault type includes multiple first-level faults, such as a first-level fault that can be one or more of a cleaning agent residue fault, a voltage stabilization fault, a probe fault, etc., and this type of fault can be further divided into multiple second-level faults. For example, taking a probe fault as an example, the probe fault can be specifically divided into one or more of the following faults: voltage instability, power line aging (equivalent to multiple second-level faults). For another example, taking a cleaning agent residue fault, it can be specifically divided into one or more of the following faults: temperature control fault, pressure control fault, cleaning valve blockage fault (equivalent to multiple second-level faults), etc.

[0064] In some embodiments, based on the results of these classifications, namely, multiple curve data sets, wafer engineers can perform analysis in combination with engineering experience to predict or evaluate the actual cause of the failure, and then perform timely inspection and maintenance on the machine.

[0065] Alternatively, in other embodiments, wafer engineers may also choose to separate these classified multiple curve data sets into new bin codes and input them into the YMS (Yield Management System) so that the YMS can adaptively update the data processing mechanism based on the new classification results.

[0066] YMS is a set of professional data analysis tools that integrates data management, data analysis, visualization, and standardization functions. In the process of semiconductor design, wafer manufacturing, packaging and testing (especially in the mass production stage), it can help customer engineers greatly improve data analysis efficiency, quickly analyze one or more types of data, and find the key points to improve yield, thereby further promoting the stability and controllability of the entire production process, improving product quality, and reducing corporate costs. In this embodiment, by reasonably classifying massive data points, it is also beneficial for YMS to use its adaptive learning ability to further analyze and judge different data sets.

[0067] In some embodiments, steps S100 - S103 and S104 may be executed simultaneously or sequentially.

[0068] In some embodiments, one of the first typical fault types corresponds to at least one of the test results, see Figure 2 As shown, the method also includes: S106, calculating multiple correlation number sets between at least one of the curve data sets and at least one test result; wherein, one of the curve data sets forms a correlation number set corresponding to one of the test results, and the correlation number set includes: multiple correlations between multiple of the one type of curves and the test results respectively calculated by the correlation calculation method.

[0069] For example, in some embodiments, when a curve data set is preliminarily evaluated to be related to fault I, a correlation calculation is performed between the curve data set and the test result corresponding to fault I. In a specific embodiment, a curve data set includes: n first-class curves, and n correlations between the n first-class curves and the test results are calculated accordingly.

[0070] For example, in some embodiments, the correlation refers to the correlation coefficient r between two curves.

[0071] For example, in some embodiments, the correlation calculation method may adopt a statistical method or a machine learning algorithm, which is not limited in the present invention.

[0072] S107, calculating the difference between at least two of the correlations in the correlation data set.

[0073] S108: When the difference is greater than a first set difference, a first prompt signal is output to prompt the user to perform a reclassification operation on the curve data set.

[0074] For example, in some embodiments, when there are two or more correlations with large deviations in the difference among the values ​​of multiple correlations formed by a curve data set, it is considered that the primary fault corresponding to the curve data set may be generated by two or more secondary faults, so the classification degree of the curve data set can be further refined. For example, the current curve data set is further divided into two or more sub-curve data sets, and among the multiple correlations corresponding to a sub-curve data set, the difference between any two correlations is less than or equal to the second set difference. In some embodiments, the first set difference and the second set difference may be equal.

[0075] For example, in this embodiment, the test results may include one or more of the following: yield parameters, defect parameters, measurement parameters, device electrical parameters, and the like.

[0076] Taking defect parameters as an example, this embodiment is explained exemplarily: when a preliminary analysis shows that the fault I reflected by one of the curve data sets may be the main cause of defect A, n correlations between n first-class curves in the curve data set and the defect parameters of defect A are calculated. If the results of the n correlations are close to the same, it can be preliminarily considered that the classification result of the curve data set is relatively reliable, and the classification result can be directly output to the user (for example, a wafer engineer) for subsequent analysis. If the results of the n correlations show significant differences, it is preferred to further subdivide the curve data set so that the user can analyze that the fault I may be a primary fault formed by multiple secondary faults, and the user may be advised to analyze these secondary faults one by one.

[0077] In some embodiments, one of the first typical fault types corresponds to at least one test result, and correspondingly, the method comprises the steps of: Calculating the correlation between the type of curve and the test result; wherein, when the correlation is lower than a set first correlation value, it is considered that the type of curve is not correlated with the test result, otherwise, it is considered that the type of curve is correlated with the test result; When the number of a type of curves related to the test result is less than a set first correlation amount, the corresponding test result is marked as an unknown result, and a second prompt signal is issued accordingly to prompt the user to correct the first numerical curve.

[0078] In this embodiment, when a fault is displayed at the test end, that is, in the test result, but no meaningful fault data (that is, a type I curve) can be matched from the classification result, it is preferred to try to re-process the type II curve that was screened out in S103 to avoid missing abnormal data due to improper previous data processing.

[0079] In some embodiments, see Figure 3 As shown, the step of correcting the first numerical curve includes: (1) obtaining the first numerical curve processed by the first type of processing method, and processing the corresponding first numerical curve by the second type of processing method to obtain a new second numerical curve; (2) determining whether the similarity between the new second numerical curve and the pre-stored second reference curve is less than or equal to a set second similarity threshold; If not, proceed to (3); (3) calculating the correlation between the second numerical curve and the test result marked as an unknown result; (4) If the correlation is greater than a set second correlation value, the second numerical curve is classified as a new curve data set.

[0080] That is to say, in this embodiment, when there are test results (or fault results) that are not matched to a related type of curve, the trend amplification processing can be performed again on the filtered first numerical curve to avoid some fluctuations being ignored due to being too small.

[0081] Similar to the first reference curve, the second reference curve may also be preset by a wafer engineer based on experience.

[0082] Specifically, in this embodiment, the first numerical curve processed by a type of processing method is preferably reprocessed to efficiently screen out data results that may be omitted.

[0083] In some embodiments, the method further includes the step of obtaining the fault type of the new curve data set and marking the fault type as a second typical fault type.

[0084] In some embodiments, the method further includes recording the data change type corresponding to the second typical fault type to update the fault type and the data change type in S104. For example, in a new data classification process, the second numerical curve is classified as a whole according to the first typical fault type and the second typical fault type.

[0085] It is understandable that this embodiment also provides a classification mechanism with autonomous updating of classification degree, that is, by analyzing the accuracy and applicability of the real-time classification degree, obtaining the latest typical fault type, and being able to timely adjust the classification degree (that is, the number of curve data sets divided in S104) in the next classification process, so as to ensure that the classification mechanism has a high adaptability to the actual operation of the wafer production line. In other words, this classification mechanism with autonomous updating of classification degree also makes the method flexible and versatile when facing different production lines, or different production stages of the same production line.

[0086] For example, in some embodiments, before the method is actually applied, historical data for a period of time in the current wafer production line (including data points collected by the operation end and test results collected by the test end) can be selected to apply the historical data to execute S100-S105 (or the method or steps in any embodiment of the present invention can be executed) to determine the typical fault type that is compatible with the current wafer production line (i.e., determine the degree of classification of the preliminary classification), and then balance the problems of over-classification and invalid classification.

[0087] In some embodiments, the method further comprises the step of: selecting a first typical fault type from a historical database of the machine.

[0088] In some embodiments, the type of processing methods includes one or more of the following: calculus processing.

[0089] For example, in some embodiments, the calculus processing refers to deriving the first numerical curve. In other words, the second numerical curve is the result of deriving the first numerical curve.

[0090] In some embodiments, the second type of processing method includes one or more of the following: exponential processing. Exponential processing refers to a data representation method that converts raw data into an exponential form through a mathematical formula.

[0091] For example, in some embodiments, the first numerical curve obtained is: Y1=A1x+B1; the second numerical curve obtained after exponential processing is: Y2=A2e kx +B2e l（x-1） +……+N2e 0 .

[0092] It should be noted that the second type of processing method in this embodiment can also adopt other processing methods, which mainly focus on amplifying the degree of change of the data. Therefore, the selected method only needs to be able to amplify the degree of change of the data for easy capture. In some embodiments, a corresponding data processing method is selected based on a data label of a type of curve; wherein the data label includes: a machine type, or a typical failure type of a machine; and the machine type or the typical failure type is correspondingly marked with at least one preferred data processing method.

[0093] For example, for some specific fault types, the fluctuation caused by them on the first value curve may be very small. For this, it is preferred to use the second type of processing method to amplify the fluctuation trend to a certain extent to capture abnormal situations more sensitively. Correspondingly, for these specific fault types, the recommended data processing methods will be marked to further improve the reliability and accuracy of the classification degree.

[0094] Embodiment 2 Correspondingly, the present invention also provides a data classification system for a machine, see Figure 4 As shown, the system includes: A data acquisition module 10, used to acquire a plurality of data points from at least one machine, wherein the plurality of data points are used to reflect at least one working indicator of the machine; A data fitting module 11 is used to fit a plurality of first numerical curves corresponding to the machine according to the plurality of data points; A data processing module 12 is used to perform data processing on the plurality of first numerical curves to obtain a plurality of second numerical curves correspondingly, wherein the data processing method includes: a first processing method, the first processing method is used to calculate the change trend of the first numerical curve, and / or a second processing method, the second processing method is used to amplify the change trend of the first numerical curve; A first classification module 13 is used to perform a first classification operation on the plurality of second value curves to select a type of curves from the plurality of second value curves, wherein the similarity between the type of curves and the pre-stored first reference curve is less than or equal to a set first similarity threshold; The type determination module 14 is used to obtain multiple fault types and determine corresponding multiple data change types according to the fault types, wherein the fault types include: a first typical fault type, the data change type is used to describe a first characteristic trend of data changes in the first type of curve, and one of the first typical fault types corresponds to at least one of the data change types; The second classification module 15 is used to perform a second classification operation on the multiple first-class curves according to the multiple data change types to divide the first-class curves into multiple curve data sets, and one curve data set includes: multiple first-class curves, and the multiple first-class curves have the same or similar data change types.

[0095] In some embodiments, one of the first typical fault types corresponds to at least one test result, and correspondingly, the system further includes: A first calculation module 16 is used to calculate a plurality of correlation number sets between at least one of the curve data sets and at least one of the test results; wherein one of the curve data sets corresponds to one of the test results and forms one of the correlation number sets, and the correlation number set includes: a plurality of correlations between a plurality of the one type of curves and the test results respectively calculated by a correlation calculation method; A second calculation module 17, configured to calculate a difference between at least two of the correlations in the correlation data set; The prompt module 18 is used to output a first prompt signal when the difference is greater than a first set difference, so as to prompt the user to perform a re-classification operation on the curve data set.

[0096] In some embodiments, one of the first typical fault types corresponds to at least one test result, and correspondingly, the system further includes: A correlation calculation module, used for calculating the correlation between the one type of curve and the test result; wherein, when the correlation is lower than a set first correlation value, it is considered that the one type of curve is not correlated with the test result, otherwise, it is considered that the one type of curve is correlated with the test result; The correction module is used to mark the corresponding test result as an unknown result when the number of a type of curves related to the test result is less than a set first correlation amount, and to send a second prompt signal accordingly to prompt the user to correct the first numerical curve.

[0097] In some embodiments, the correction module further comprises: a reprocessing unit, configured to obtain the first numerical curve processed by the first type of processing method, and process the first numerical curve by using the second type of processing method to obtain a new second numerical curve; A judging unit, used to judge whether the similarity between the new second value curve and the second reference curve is less than or equal to a set second similarity threshold; If not, then enter the correlation calculation unit; a correlation calculation unit, configured to calculate the correlation between the second value curve and the test result marked as an unknown result; An updating unit is used for classifying the second numerical curve as a new curve data set if the correlation is greater than a set second correlation value.

[0098] In some embodiments, the method further includes: a first fault updating module, configured to obtain the fault type of the new curve data set and mark the fault type as a second typical fault type.

[0099] In some embodiments, it also includes: a second fault updating module, which is used to record the data change type corresponding to the second typical fault type, so as to update the fault type and the data change type in the type determination module 14.

[0100] In some embodiments, it also includes: a selection module, which is used to select a first typical fault type from the historical database of the machine.

[0101] It can be understood that the system in the embodiment of the present invention can be used to implement the method or steps in any of the above embodiments, which will not be described in detail here.

[0102] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0103] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, a magnetic disk, or an optical disk), and includes a number of instructions for a computer terminal (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present invention.

[0104] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation modes, which are merely illustrative rather than restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the present invention and the claims, all of which are within the protection of the present invention.

Claims

1. A data classification method for a machine, characterized in that: Includes steps: S100, acquiring a plurality of data points from at least one machine, wherein the plurality of data points are used to reflect at least one working indicator of the machine; S101, fitting a plurality of first numerical curves corresponding to the machine according to the plurality of data points; S102, performing data processing on the plurality of first numerical curves to obtain a plurality of second numerical curves correspondingly, wherein the data processing method comprises: a first processing method, the first processing method is used to calculate the change trend of the first numerical curve, and / or a second processing method, the second processing method is used to amplify the change trend of the first numerical curve; S103, performing a first classification operation on the plurality of second value curves to select a type of curves from the plurality of second value curves, wherein the similarity between the type of curves and the pre-stored first reference curve is less than or equal to a set first similarity threshold; S104, acquiring multiple fault types, and determining corresponding multiple data change types according to the fault types, respectively, wherein the data change type is used to describe a first characteristic trend of data changes in the type of curve, and the fault type includes: a first typical fault type, and one of the first typical fault types corresponds to at least one of the data change types; S105, performing a second classification operation on the plurality of the first-class curves according to the plurality of data change types, so as to classify the first-class curves into a plurality of curve data sets, wherein one curve data set includes: a plurality of the first-class curves, and the plurality of the first-class curves have the same or similar data change types.

2. The data classification method according to claim 1, characterized in that: One of the first typical fault types corresponds to at least one test result, and the method further includes: S106, calculating a plurality of correlation number sets between at least one of the curve data sets and at least one of the test results; wherein one of the curve data sets forms one of the correlation number sets corresponding to one of the test results, and the correlation number set includes: a plurality of correlations between a plurality of the one type of curves and the test results respectively calculated by a correlation calculation method; S107, calculating the difference between at least two of the correlations in the correlation data set; S108: When the difference is greater than a first set difference, a first prompt signal is output to prompt the user to perform a reclassification operation on the curve data set.

3. The data classification method according to claim 2, characterized in that: One of the first typical fault types corresponds to at least one test result, comprising the steps of: Calculating the correlation between the one type of curve and the test result; wherein, when the correlation is lower than a set first correlation value, it is considered that the one type of curve is not correlated with the test result, otherwise, it is considered that the one type of curve is correlated with the test result; When the number of the type of curves related to the test result is less than a set first correlation amount, the corresponding test result is marked as an unknown result, and a second prompt signal is issued accordingly to prompt the user to correct the corresponding first numerical curve.

4. The data classification method according to claim 3, characterized in that: The step of correcting the first numerical curve includes: (1) obtaining the first numerical curve processed by the first type of processing method, and processing the corresponding first numerical curve by the second type of processing method to obtain a new second numerical curve; (2) determining whether the similarity between the new second numerical curve and the pre-stored second reference curve is less than or equal to a set second similarity threshold; If not, proceed to (3); (3) calculating the correlation between the second numerical curve and the test result marked as an unknown result; (4) If the correlation is greater than a set second correlation value, the second numerical curve is classified as a new curve data set.

5. The data classification method according to claim 4, characterized in that: Also includes the steps: The fault type of the new curve data set is obtained, and the fault type is marked as a second typical fault type.

6. The data classification method according to claim 5, characterized in that: Also includes the steps: The data change type corresponding to the second typical fault type is recorded to update the fault type and the data change type in S104.

7. The data classification method according to claim 1, characterized in that: The method comprises the steps of: selecting a first typical fault type from a historical database of the machine.

8. The data classification method according to claim 1, characterized in that: The first type of processing method includes one or more of the following: calculus processing; and / or, the second type of processing method includes one or more of the following: exponential processing.

9. The data classification method according to claim 1, characterized in that: Also includes the steps: A corresponding data processing method is selected according to a data label of a type of curve; wherein the data label includes: a machine type, or a typical failure type of a machine; and the machine type or the typical failure type is correspondingly marked with at least one preferred data processing method.

10. A data classification system for a machine, characterized in that: include: A data acquisition module, used to acquire a plurality of data points from at least one machine, wherein the plurality of data points are used to reflect at least one working indicator of the machine; A data fitting module, used for fitting a plurality of first numerical curves corresponding to the machine according to the plurality of data points; a data processing module, configured to perform data processing on the plurality of first numerical curves to obtain a plurality of second numerical curves correspondingly, wherein the data processing method comprises: a first type of processing method, the first type of processing method is used to calculate the change trend of the first numerical curve, and / or a second type of processing method, the second type of processing method is used to amplify the change trend of the first numerical curve; a first classification module, configured to perform a first classification operation on the plurality of the second value curves, so as to select a type of curves from the plurality of the second value curves, wherein the similarity between the type of curves and the pre-stored first reference curve is less than or equal to a set first similarity threshold; A type determination module, used to obtain multiple fault types, and determine corresponding multiple data change types according to the fault types, wherein the data change type is used to describe a first characteristic trend of data changes in the first type of curve, and the fault type includes: a first typical fault type, and one of the first typical fault types corresponds to at least one of the data change types; The second classification module is used to perform a second classification operation on the multiple first-class curves according to the multiple data change types to divide the first-class curves into multiple curve data sets, and one curve data set includes: multiple first-class curves, and the multiple first-class curves have the same or similar data change types.

Images