Anomaly identification method and system combining hierarchical parameter analysis

Abstract

This invention discloses an anomaly identification method and system combining hierarchical parameter analysis, relating to the field of fault detection technology. The method includes: extracting fault detection information of a target semiconductor device; searching a fault record database to determine a first set of sub-events; extracting key parameters to obtain a set of key parameter value sequences for the first sub-events; performing parameter fluctuation duration integration analysis to obtain the first parameter fluctuation duration; extracting associated devices to obtain a set of K secondary faulty components; obtaining a set of K secondary operating parameters; obtaining a set of abnormal components and a set of abnormal detection items; and performing potential fault detection on the target semiconductor device. This invention solves the technical problems of existing semiconductor device fault detection methods, such as the difficulty in accurately predicting and locating potential fault sources in complex semiconductor devices and low detection efficiency, achieving the technical effect of improving the efficiency and accuracy of semiconductor device fault detection.

Description

Technical Field

[0001] This application relates to the field of fault detection technology, specifically to an anomaly identification method and system that combines hierarchical parameter analysis. Background Technology

[0002] With the rapid development of semiconductor technology, semiconductor equipment plays a crucial role in the electronics manufacturing industry. Semiconductor equipment mainly includes etching equipment, thin film deposition equipment, cleaning equipment, wet processing equipment, packaging and testing equipment, polishing equipment, silicon wafer manufacturing equipment, etc. The complexity and precision of semiconductor equipment make fault detection a challenge. With the advancement of technology, modern semiconductor equipment fault detection methods are becoming increasingly diversified, including multimeter testing, oscilloscope measurement, fault mode analysis, non-destructive testing (such as X-ray, ultrasonic scanning, etc.), electrical testing, etc. However, these methods can only measure and locate faults that have already occurred, and cannot accurately predict and locate potential faults in the equipment based on the analysis of historical fault data, so as to carry out early maintenance and prevention, and at the same time affect the efficiency of fault detection.

[0003] Therefore, current technologies related to fault detection in semiconductor equipment suffer from difficulties in accurately predicting and locating potential fault sources in complex semiconductor devices, as well as low detection efficiency. Summary of the Invention

[0004] This application provides an anomaly identification method and system that combines hierarchical parameter analysis. By employing techniques such as feature extraction and matching, and constructing a noise reduction and convergence model, it solves the technical problems of existing semiconductor equipment fault detection, such as the difficulty in accurately predicting and locating potential fault sources in complex semiconductor equipment and the low detection efficiency. This achieves the technical effect of improving the efficiency and accuracy of semiconductor equipment fault detection.

[0005] This application provides an anomaly identification method combining hierarchical parameter analysis. The method includes: extracting fault detection information of a target semiconductor device, wherein the fault detection information includes the fault occurrence time point, fault event, and K primary faulty components; retrieving from a fault record database based on the fault event to determine a first sub-event set; extracting key parameters from the first sub-event set to obtain a set of key parameter value sequences for the first sub-event; traversing the set of key parameter value sequences for the first sub-event and performing parameter fluctuation duration integration analysis to obtain a first parameter fluctuation duration; using the K primary faulty components as indexes, extracting associated devices from the target semiconductor device to obtain a set of K secondary faulty components; extracting operating parameters from the set of K secondary faulty components based on the first parameter fluctuation time and the fault occurrence time to obtain a set of K secondary operating parameters; identifying parameter anomaly trends in the set of K secondary operating parameters to obtain a set of abnormal components and a set of abnormal detection items; and detecting potential faults in the target semiconductor device based on the set of abnormal components and the set of abnormal detection items.

[0006] In a possible implementation, the parameter fluctuation duration is integrated and analyzed by traversing the set of key parameter value sequences of the first sub-event to obtain the first parameter fluctuation duration. The following processing is then performed: a set of first parameter change curves is constructed based on the set of key parameter value sequences of the first sub-event, wherein the horizontal axis of the curves in the set of first parameter change curves is time, and the vertical axis is the key parameter value; the set of first parameter change curves is divided according to a preset parameter tolerance threshold set to obtain a cluster of first parameter abnormal time periods, wherein the cluster of first parameter abnormal time periods includes multiple sets of first parameter abnormal time periods, and each set of first parameter abnormal time periods corresponds to a first parameter change curve; the parameter fluctuation duration is integrated and analyzed on the cluster of first parameter abnormal time periods to determine the first parameter fluctuation duration.

[0007] In a possible implementation, the set of key parameter value sequences of the first sub-event is traversed to perform parameter fluctuation duration integration analysis to obtain the first parameter fluctuation duration. The following processing is then performed: the duration of multiple first parameter abnormal time periods is statistically analyzed to obtain multiple first parameter abnormal time length sets; the mean of each of the multiple first parameter abnormal time length sets is calculated to obtain multiple first parameter abnormal time mean values; starting from the mode of the multiple first parameter abnormal time mean values, discrete cleaning is performed on the multiple first parameter abnormal time mean values ​​to obtain a target first parameter abnormal time mean value set; the target first parameter abnormal time mean value set is weighted according to the distance to the starting point to generate the first parameter fluctuation duration.

[0008] In a possible implementation, the anomaly identification method combined with hierarchical parameter analysis further performs the following processing: Calculate the proportion of the mode of the multiple first parameter anomaly time mean values ​​among the multiple first parameter anomaly time mean values, and determine the cleaning step size based on the calculation result; determine a first discrete cleaning interval with the starting point as the interval center and the cleaning step size as the interval radius, wherein the first discrete cleaning interval includes a first interval density; expand the first discrete cleaning interval according to the cleaning step size to obtain a second discrete cleaning interval, wherein the second discrete cleaning interval includes a second interval density; when the first interval density is less than or equal to the second interval density, continue to expand and iterate the second discrete cleaning interval until the interval density difference between two adjacent cleanings is less than a preset interval density difference threshold, stop the iteration, and take the Nth discrete cleaning interval obtained in the last iteration as the target discrete cleaning interval; take the multiple first parameter anomaly time mean values ​​within the target discrete cleaning interval as the target first parameter anomaly time mean set.

[0009] In a possible implementation, using the K primary faulty components as indexes, the target semiconductor device is associated with other devices to obtain a set of K secondary faulty components. The following processing is also performed: obtaining multiple components of the target semiconductor device; assigning pairwise association degrees to the multiple components according to preset association indicators to obtain multiple initial association degree sets and multiple initial associated component sets; filtering the multiple initial association degree sets according to preset association degree thresholds; mapping and cleaning the multiple initial associated component sets based on the filtering results to generate multiple associated component sets; and retrieving the multiple associated component sets using the K primary faulty components as indexes to obtain a set of K secondary faulty components.

[0010] In a possible implementation, the following steps are performed: Parameter anomaly trend identification is performed on the K sets of secondary operating parameters to obtain a set of abnormal components and a set of abnormal detection items. The following processing is also performed: Obtaining a set of K compliance thresholds for the K sets of secondary faulty components; performing out-of-specification analysis on the K sets of secondary operating parameters using the K sets of compliance thresholds to determine a set of K out-of-specification operating parameters and a set of K compliant operating parameters; calculating the out-of-specification ratio based on the K sets of K out-of-specification operating parameters and the K sets of K secondary operating parameters to determine K parameter out-of-specification coefficients; identifying the K sets of compliant operating parameters according to K threshold edge bandwidths to determine a set of K compliant operating parameters at the edge; calculating the edge compliance ratio based on the K sets of K compliant operating parameters and the K sets of K compliant operating parameters to determine K parameter edge compliance coefficients; performing a weighted calculation on the K parameter out-of-specification coefficients and the K parameter edge compliance coefficients to obtain K anomaly coefficients; and identifying the K anomaly coefficients according to a preset anomaly coefficient threshold to determine the set of abnormal components.

[0011] In a possible implementation, the anomaly identification method combining hierarchical parameter analysis also performs the following processing: constructing a detection item identification network layer; using the detection item identification network layer to identify the set of abnormal components and the multiple sets of secondary operating parameters corresponding to the set of abnormal components, and determining the set of abnormal detection items.

[0012] This application also provides an anomaly identification system combining hierarchical parameter analysis, comprising: a fault detection information extraction module, which is used to extract fault detection information of a target semiconductor device, wherein the fault detection information includes the fault occurrence time point, fault event, and K first-level faulty components; a first sub-event set determination module, which searches a fault record database based on the fault event to determine a first sub-event set; a key parameter extraction module, which is used to extract key parameters from the first sub-event set to obtain a set of key parameter value sequences for the first sub-event; and a first parameter fluctuation duration acquisition module, which is used to traverse the set of key parameter value sequences for the first sub-event to perform parameter fluctuation duration integration analysis to obtain the first parameter fluctuation duration. The system includes: a secondary fault component set acquisition module, which extracts associated devices from the target semiconductor device using the K primary fault components as indices to obtain a set of K secondary fault components; a secondary operating parameter set acquisition module, which extracts operating parameters from the K secondary fault component sets based on the first parameter fluctuation time and the fault occurrence time to obtain a set of K secondary operating parameters; a parameter anomaly trend identification module, which identifies parameter anomaly trends in the K secondary operating parameter sets to obtain a set of abnormal components and a set of abnormal detection items; and a device potential fault detection module, which detects potential faults in the target semiconductor device based on the set of abnormal components and the set of abnormal detection items.

[0013] This application proposes an anomaly identification method and system that combines hierarchical parameter analysis. The method extracts fault detection information from a target semiconductor device; searches a fault record database to determine the first sub-event set; extracts key parameters to obtain a set of key parameter value sequences for the first sub-event; performs integrated analysis of parameter fluctuation duration to obtain the first parameter fluctuation duration; extracts associated devices to obtain a set of K secondary faulty components; obtains a set of K secondary operating parameters; obtains a set of abnormal components and a set of abnormal detection items; and performs potential fault detection on the target semiconductor device. This solves the technical problems of existing semiconductor device fault detection methods, such as the difficulty in accurately predicting and locating potential fault sources in complex semiconductor devices and low detection efficiency, thus achieving the technical effect of improving the efficiency and accuracy of semiconductor device fault detection. Attached Figure Description

[0014] To more clearly illustrate the technical solutions of the embodiments of this disclosure, the accompanying drawings of the embodiments of this disclosure will be briefly described below. Flowcharts are used in this application to illustrate the operations performed by the system according to the embodiments of this application. It should be understood that the preceding or following operations are not necessarily performed precisely in sequence. Instead, various steps can be processed in reverse order or simultaneously as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from these processes.

[0015] Figure 1 A schematic flowchart of an anomaly identification method combining hierarchical parameter analysis provided in an embodiment of this application;

[0016] Figure 2 This is a schematic diagram of the anomaly identification system structure combined with hierarchical parameter analysis provided in an embodiment of this application.

[0017] Explanation of reference numerals in the attached figures: Fault detection information extraction module 10, first sub-event set determination module 20, key parameter extraction module 30, first parameter fluctuation duration acquisition module 40, secondary fault component set acquisition module 50, secondary operating parameter set acquisition module 60, parameter abnormal trend identification module 70, and equipment potential fault detection module 80. Detailed Implementation

[0018] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application.

[0019] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description of this application will be provided in conjunction with the accompanying drawings. The described embodiments should not be considered as limitations on this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0020] In the following description, references to "some embodiments" describe a subset of all possible embodiments. However, it is understood that "some embodiments" may be the same or different subsets of all possible embodiments and may be combined with each other without conflict. The terms "first" and "second" are used merely to distinguish similar objects and do not represent a specific ordering of objects. The terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules not explicitly listed or inherent to these processes, methods, products, or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only.

[0021] This application provides an anomaly identification method that combines hierarchical parameter analysis, such as... Figure 1 As shown, the method includes:

[0022] Step S100: Extract fault detection information of the target semiconductor equipment. This fault detection information includes the fault occurrence time, fault events, and K primary faulty components. Extracting fault detection information of the target semiconductor equipment mainly refers to collecting and analyzing various detection information and data related to the semiconductor equipment fault from operation records or equipment operation logs before and after the fault. This includes the fault occurrence time, fault events, and K primary faulty components. Specifically, the fault occurrence time refers to the specific time the fault occurred, such as a specific date, time, minute, and second. A precise time helps analyze the equipment status, operating conditions, and other possible external factors at the time of the fault, thereby more accurately locating the cause of the fault. Fault events refer to the specific manifestations or phenomena of the fault, such as inability to operate normally (failure of a certain function of the equipment), performance degradation (product quality not meeting standards, reduced output), abnormal sounds, abnormal temperatures, etc. Primary faulty components refer to equipment components initially judged to be directly related to the fault events. k is a positive integer greater than or equal to 1, representing the specific number of faulty equipment components. Analyzing the primary faulty components can narrow down the fault range and further locate the specific position of the fault, helping to reduce fault investigation time and improve fault handling efficiency.

[0023] Step S200: Based on the fault event, a search is performed in the fault record database to determine the first sub-event set. When performing semiconductor equipment fault detection, an existing fault record database is used to assist in the diagnosis and location of the current fault, determining the first sub-event set. The fault record database is typically a database containing historical fault events and related information. Specifically, the extracted fault event information is used as query conditions to search the fault record database, finding historical fault records similar to or related to the current fault event. These historical fault records constitute the first sub-event set. Each element in the set is a historical fault event similar to or related to the current fault event, containing information such as fault descriptions, causes, and solutions that are valuable for reference regarding the current fault. For example, power supply fault events include voltage fluctuations, power outages, and loose wiring. After determining the first sub-event set, these historical fault records can be further analyzed to extract useful information to assist in the diagnosis and location of the current fault. For example, comparing the similarities and differences between the current fault and historical faults in terms of performance, parameters, etc., thereby narrowing the scope of fault investigation.

[0024] Step S300: Extract key parameters from the first sub-event set to obtain a set of key parameter value sequences for the first sub-event. A thorough analysis of the first sub-event set is conducted to extract key parameters related to the fault and their corresponding values, resulting in a set of key parameter sequences for the first sub-event. Specifically, by comparing changes in key parameters across different fault events and analyzing historical fault data, key parameters closely related to fault occurrence and diagnosis are identified and extracted from the first sub-event set. These may include equipment operating parameters (such as voltage, current, and temperature), environmental parameters (such as temperature, humidity, and dust concentration), operational parameters (such as operation time, operation sequence, and operator), and other fault-related parameters. For each extracted key parameter, its specific values ​​in historical fault events are collected to form a key parameter value sequence, reflecting the changes in the key parameter before and after the fault occurs. This provides crucial data support for fault diagnosis and analysis. The set of key parameter value sequences for the first sub-event is a collection of multiple key parameter value sequences, each corresponding to a key parameter, containing rich fault information.

[0025] Step S400: Traverse the set of key parameter value sequences of the first sub-event to perform parameter fluctuation duration integration analysis and obtain the first parameter fluctuation duration. A detailed time-domain analysis is performed on the key parameter value sequences extracted from the first sub-event set to determine the duration of fluctuations of these key parameters before and after the fault occurs. This quantifies the length of parameter fluctuations, thereby revealing the temporal characteristics of fault development. Specifically, each sequence in the first sub-event key parameter value sequence set is examined one by one, i.e., the value change sequence of each key parameter in the fault event. In each parameter value sequence, by setting a threshold or comparing adjacent data points, the intervals in which parameter values ​​change significantly (i.e., fluctuations) are identified. For example, when a parameter value exceeds or falls below a preset threshold, or when the change between adjacent data points exceeds a certain threshold, parameter fluctuation is considered to have occurred. For each identified parameter fluctuation interval, its duration is calculated, i.e., the time difference from the start to the end of the fluctuation, reflecting the length or duration of the parameter fluctuation. The fluctuation durations of all key parameters are summarized and integrated to analyze the relationships and patterns between these fluctuation durations. This may include calculating statistical indicators such as average fluctuation duration, longest fluctuation duration, and shortest fluctuation duration, as well as analyzing the correlation or differences between the fluctuation durations of different parameters. Through integrated analysis, the fluctuation duration of the first parameter (e.g., average fluctuation duration) is determined.

[0026] In one possible implementation, step S400 further includes step S410, constructing a first parameter change curve set based on the first sub-event key parameter value sequence set, wherein the horizontal axis of the curves in the first parameter change curve set represents time, and the vertical axis represents the key parameter value. Each element (i.e., curve) in the first parameter change curve set is constructed based on the aforementioned key parameter value sequence set, with the horizontal axis representing time and the vertical axis representing the key parameter value. By viewing these curves, one can intuitively understand how the key parameter value changes over time. The implementation also includes step S420, dividing the first parameter change curve set according to a preset parameter tolerance threshold set to obtain first parameter abnormal time period clusters, wherein the first parameter abnormal time period cluster includes multiple first parameter abnormal time period sets, and each first parameter abnormal time period set corresponds to a first parameter change curve. The preset parameter tolerance threshold set is a set of thresholds based on historical data used to determine when a parameter value is considered abnormal. For each curve in the first parameter change curve set, the preset parameter tolerance threshold set is used to determine which time periods are abnormal. Specifically, each point on the curve (i.e., the parameter value at each time point) is checked. If the value at a certain point exceeds the corresponding threshold range, the point and its surrounding time period are marked as abnormal. Each curve is divided into multiple normal and abnormal time periods, obtaining a first parameter abnormal time period cluster, including multiple first parameter abnormal time period sets. The method also includes step S430, which performs parameter fluctuation duration integration analysis on the first parameter abnormal time period cluster to determine the first parameter fluctuation duration. All abnormal time periods in the first parameter abnormal time period cluster are integrated and analyzed to determine the total parameter fluctuation duration. Specifically, for each first parameter abnormal time period set (i.e., the abnormal time period on each curve), its duration (i.e., the time range spanned by that period) is calculated. The durations of all abnormal time periods are summed to obtain the total parameter fluctuation duration, i.e., the first parameter fluctuation duration, representing the total duration of parameter value abnormality throughout the entire observation period.

[0027] In one possible implementation, step S430 further includes step S431, which involves statistically analyzing the duration of multiple sets of abnormal first-parameter time periods to obtain multiple sets of abnormal first-parameter time lengths. For each set of abnormal first-parameter time periods, the duration of each abnormal time period (i.e., the difference between the start and end times of the abnormal time period) is calculated. The durations of each abnormal first-parameter time period are combined into a set, called the set of abnormal first-parameter time lengths, ultimately obtaining multiple sets of abnormal first-parameter time lengths. Step S432 further includes calculating the mean of each set of abnormal first-parameter time lengths to obtain multiple average values ​​for abnormal first-parameter times. The mean of all time lengths in each set of abnormal first-parameter time lengths is calculated, representing the average duration of abnormal time periods in that set, thus obtaining multiple average values ​​for abnormal first-parameter times.

[0028] Step S430 further includes step S433, which involves using the mode of the multiple abnormal time mean values ​​of the first parameter as a starting point, performing discrete cleaning on the multiple abnormal time mean values ​​of the first parameter to obtain a target set of abnormal time mean values ​​of the first parameter. The mode of the multiple abnormal time mean values ​​of the first parameter is calculated, i.e., the mean value that appears most frequently, and discrete cleaning is performed (i.e., removing or adjusting abnormal time mean values ​​that differ significantly from the mode) to obtain the target set of abnormal time mean values ​​of the first parameter. Step S434 further includes step S434, which involves weighting the target set of abnormal time mean values ​​of the first parameter according to their distance from the starting point to generate the first parameter fluctuation duration. In the target set of abnormal time mean values ​​of the first parameter, the distance between each mean and the mode (i.e., the starting point) is used as the basis for weighting. Specifically, mean values ​​closer to the starting point are given a larger weight, while mean values ​​farther away are given a smaller weight. Then, the weighted mean values ​​are summarized or calculated to obtain a comprehensive value, which is the first parameter fluctuation duration, representing the average level of all abnormal time periods and considering the relative distance between different abnormal durations and the mode.

[0029] In one possible implementation, step S433 further includes step S4331, calculating the proportion of the mode of the multiple first parameter abnormal time means among the multiple first parameter abnormal time means, and determining the cleaning step size based on the calculation result. Among the multiple first parameter abnormal time means, the proportion of the mode (i.e., the mean that occurs most frequently) is calculated, and an initial cleaning step size is determined based on the proportion of the mode for subsequent interval cleaning. It also includes step S4332, determining a first discrete cleaning interval with the starting point as the interval center and the cleaning step size as the interval radius, wherein the first discrete cleaning interval includes a first interval density. Using the mode as the center (i.e., the starting point) of the interval and a determined cleaning step size as the interval radius, an interval range is defined, namely the first discrete cleaning interval. This interval includes the numerical ranges extended to the left and right by the cleaning step size starting from the mode. A set of anomalous time mean values ​​of the first parameter that are close to or similar to the mode is selected, thereby eliminating outliers or noise data that are far from the mode. The ratio of the number of anomalous time mean values ​​of the first parameter in the first discrete cleaning interval to the total length of the interval is calculated to obtain the first interval density, which reflects the density of data points in the interval. If the density is high, it indicates that the data points in the interval are relatively concentrated; otherwise, it indicates that the data points are relatively scattered.

[0030] Step S433 further includes step S4333, expanding the first discrete cleaning interval according to the cleaning step size to obtain a second discrete cleaning interval, wherein the second discrete cleaning interval includes a second interval density. After determining the first discrete cleaning interval (i.e., the interval centered on the mode with the cleaning step size as its radius), this interval is expanded according to the cleaning step size. Typically, the length of one cleaning step size is added to both sides of the first discrete cleaning interval to obtain a larger interval, namely the second discrete cleaning interval. This helps to more comprehensively assess the distribution of the data and may discover more data points similar to the mode. The ratio of the number of first parameter outlier time mean values ​​falling into this interval to the total length of the interval is calculated to obtain the second interval density, which reflects the density of data points within the second discrete cleaning interval. Compared with the first interval density, the second interval density may vary because it contains more data points. If the second interval density is high, it indicates that the data points are still relatively concentrated over a larger area; if the second interval density is low, it may mean that the data points are more dispersed within that area.

[0031] Step S433 further includes step S4334: when the density of the first interval is less than or equal to the density of the second interval, the second discrete cleaning interval is expanded iteratively until the difference in interval density between two adjacent cleanings is less than a preset interval density difference threshold, the iteration stops, and the Nth discrete cleaning interval obtained in the last iteration is taken as the target discrete cleaning interval. If the density of the first interval is less than or equal to the density of the second interval, the second discrete cleaning interval is expanded to obtain a new, larger discrete cleaning interval (which can be called the third discrete cleaning interval), and the density of this new interval is calculated. This process is repeated, and each iteration will obtain a larger discrete cleaning interval and a corresponding interval density. The iteration ends when the stopping condition is met. The stopping condition is usually that the difference in interval density between two adjacent cleanings is less than a preset threshold (i.e., the preset interval density difference threshold). The preset interval density difference threshold is used to control the accuracy and efficiency of the iteration, ensuring that an interval that contains enough data points while maintaining relatively concentrated data is found. When the iteration stops, the discrete cleaning interval obtained in the last iteration (i.e., the Nth discrete cleaning interval) will be regarded as the target discrete cleaning interval. It also includes step S4335, which takes the average of multiple first parameter abnormal times within the target discrete cleaning interval as the target first parameter abnormal time average set.

[0032] Step S500: Using the K primary faulty components as indexes, perform associated device extraction on the target semiconductor device to obtain a set of K secondary faulty components. Based on these K primary faulty components, associated devices are extracted using them as indexes. Associated devices refer to other devices or components that are closely related to the primary faulty components in terms of function, structure, or operation. These devices or components may be affected by the primary faulty components, or their failures may also cause problems with the primary faulty components. For each primary faulty component, the associated secondary faulty components are extracted to form a set of secondary faulty components, ultimately resulting in K sets of secondary faulty components, each set corresponding to one primary faulty component.

[0033] In one possible implementation, step S500 further includes step S510, acquiring multiple components of the target semiconductor device. This involves identifying, listing, and collecting all relevant constituent components of the target semiconductor device, i.e., the various parts constituting the semiconductor device. It also includes step S520, assigning pairwise correlation degrees to the multiple components according to preset correlation indicators to obtain multiple initial correlation degree sets and multiple initial associated component sets. Preset correlation indicators may be determined based on the functional relevance, physical connectivity, fault propagation, or any other attributes related to the operation of semiconductor devices. For example, functional relevance considers whether two components work together to complete a specific task; physical connectivity focuses on whether there is a direct physical connection (such as wires, pipes, etc.) between two components. For multiple components obtained, correlation analysis is performed on each pair of components according to the preset correlation indicators. The correlation between each component and multiple other components is collected to form an initial correlation set. After determining that there is a certain correlation between two components, these two components are combined into a correlated component pair, and all such correlated component pairs are collected to form an initial correlated component set. Each initial correlated component set contains multiple correlated component pairs, and the correlation values ​​between these component pairs may be different.

[0034] Step S500 further includes step S530, which involves filtering the multiple initial correlation sets according to a preset correlation threshold, and mapping and cleaning the multiple initial correlation component sets based on the filtering results to generate multiple correlation component sets. The preset correlation threshold is set based on historical data, actual situation, and needs, and is used to determine which correlations are significant or important. Using the preset correlation threshold, each correlation value in the multiple initial correlation sets is compared and judged. If the correlation value of a component pair exceeds or equals the correlation threshold, it is considered that the pair of components has a significant correlation, and this correlation value and the corresponding component pair information are retained. If the correlation value is lower than the threshold, the initial correlation component sets are mapped and cleaned, that is, those component pairs in the initial correlation component sets with correlations lower than the threshold are removed, and finally multiple correlation component sets are obtained, each correlation component set containing a set of component pairs with significant correlations.

[0035] Step S500 further includes step S540, which uses the K primary faulty components as indexes to retrieve the multiple sets of associated components to obtain K sets of secondary faulty components. Using these K primary faulty components as indexes or query conditions, the previously generated multiple sets of associated components are retrieved. Other components directly or indirectly related to the primary faulty components are found within these sets. For each primary faulty component, its associated secondary components are collected to form a set of secondary faulty components, ultimately yielding K sets of secondary faulty components.

[0036] Step S600: Based on the first parameter fluctuation time and the fault occurrence time, extract operating parameters from the K sets of secondary faulty components to obtain K sets of secondary operating parameters. A time range is determined based on the first parameter fluctuation time and the fault occurrence time, typically including a period before and after the fault occurrence time, to ensure that complete operating parameter data related to the fault can be captured. Specifically, within the determined time range, operating parameters related to the secondary faulty components are extracted from the equipment's operating data, which may include voltage, current, temperature, pressure, and speed. The extracted operating parameters are screened and organized, removing parameters irrelevant to the fault or redundant parameters, retaining key parameters closely related to the occurrence and evolution of the fault, and categorized according to the corresponding set of secondary faulty components to form K sets of secondary operating parameters, each set containing key operating parameter data related to the corresponding secondary faulty component.

[0037] Step S700 involves identifying abnormal trends in the K sets of secondary operating parameters to obtain a set of abnormal components and a set of abnormal detection items. Specifically, identifying abnormal trends in the K sets of secondary operating parameters involves using Z-scores, anomaly detection algorithms, etc., to analyze the parameter values ​​in each set, identifying parameter values ​​that significantly differ from the normal parameter distribution pattern (i.e., abnormal values), indicating an abnormal operating state of the equipment or component. The identified abnormal parameter values ​​are then categorized according to their corresponding secondary faulty components. If multiple key operating parameters of a secondary faulty component exhibit abnormal trends, the component is identified as an abnormal component and added to the abnormal component set. From the operating parameters of each abnormal component, specific abnormal parameter items or parameter combinations are extracted to form a set of abnormal detection items. Here, abnormal detection items refer to specific parameter items or parameter combinations that are identified as abnormal in the abnormal trend identification.

[0038] In one possible implementation, step S700 further includes step S710, obtaining a set of K compliance thresholds for operating parameters for K sets of secondary faulty components. For each set of secondary faulty components, there is a corresponding set of operating parameters (such as temperature, pressure, current, etc.), and each parameter has a compliance threshold, representing the range of the parameter under normal operation. Multiple compliance thresholds for operating parameters of each component in the set of secondary faulty components are obtained to form a set of K compliance thresholds for operating parameters. Step S720 further includes using the set of K compliance thresholds for operating parameters to perform out-of-compliance analysis on the K sets of secondary operating parameters, determining a set of K out-of-compliance operating parameters and a set of K compliant operating parameters. For each operating parameter in each set of secondary faulty components, it is compared with its corresponding compliance threshold. If the value of a parameter exceeds its compliance threshold, it is considered out of compliance, ultimately determining a set of K out-of-compliance operating parameters and a set of K compliant operating parameters. Step S730 further includes calculating the out-of-compliance ratio based on the set of K out-of-compliance operating parameters and the K sets of K secondary operating parameters, determining the out-of-compliance coefficients for the K parameters. For each set of secondary faulty components, the ratio of the number of parameters in its out-of-range operating parameter set to the number of parameters in the entire secondary operating parameter set is calculated to obtain K parameter out-of-range coefficients, which reflect the severity of the component's out-of-range operating parameters.

[0039] Step S700 further includes step S740, identifying the K sets of compliant operating parameters according to K threshold edge bandwidths to determine K sets of edge compliant operating parameters. Parameter values ​​are within the compliance threshold range, but if they are close to the threshold edge (i.e., close to the minimum or maximum threshold), they may also indicate potential problems or faults. The threshold edge bandwidth is a range that includes parameter values ​​close to the edge of the compliance threshold. Parameters within the compliance range but close to the threshold edge are identified to form K sets of edge compliant operating parameters. Step S750 further includes calculating the edge compliance ratio based on the K sets of edge compliant operating parameters and the K sets of compliant operating parameters to determine the K parameter edge compliance coefficients. For each set of secondary fault components, the ratio of the number of parameters in its edge compliant operating parameter set to the number of parameters in the entire compliant operating parameter set is calculated, reflecting the proportion of component operating parameters within the compliance range but close to the edge, i.e., the degree of potential risk, thereby determining the K parameter edge compliance coefficients.

[0040] Step S700 further includes step S760, which involves weighting the K out-of-specification coefficients and the K edge compliance coefficients to obtain K anomaly coefficients. Based on the importance or impact of the K out-of-specification coefficients and the K edge compliance coefficients, weighted calculations are performed to obtain K anomaly coefficients. Each anomaly coefficient corresponds to a set of secondary faulty components, reflecting the potential anomaly level of the components in that set. Step S770 further includes step S770, which identifies the K anomaly coefficients according to a preset anomaly coefficient threshold to determine the set of abnormal components. The preset anomaly coefficient threshold is an threshold set based on actual conditions and needs, used to determine which anomaly coefficients represent true anomalies or potential faults. The calculated K anomaly coefficients are compared with the preset anomaly coefficient threshold. If the anomaly coefficient of a certain set of secondary faulty components exceeds the preset anomaly coefficient threshold, the components in that set are considered to have anomalies or potential faults, and the corresponding components are identified as abnormal components. Components in all sets of secondary faulty components with anomaly coefficients exceeding the threshold are combined to form an abnormal component set.

[0041] In one possible implementation, step S770 further includes step S771, constructing a detection item recognition network layer. A recognition model is built based on convolutional neural networks (CNN), recurrent neural networks (RNN), etc., and the network is trained using a labeled dataset to obtain the detection item recognition network layer, which is used to identify abnormal detection items in the abnormal component set and the secondary operating parameter set. Step S772 further includes using the detection item recognition network layer to identify the abnormal component set and the multiple secondary operating parameter sets corresponding to the abnormal component set, determining the abnormal detection item set. The abnormal component set and the multiple secondary operating parameter sets corresponding to these components are used as input data and input into the detection item recognition network layer. Useful features are extracted through components such as convolutional layers, pooling layers, and activation functions in the network layer. Specifically, patterns or information that can represent the operating state and abnormal characteristics of components are extracted from the original data, i.e., determining which secondary operating parameters or component states belong to abnormal detection items, such as parameter values ​​exceeding the normal range, sudden changes in parameter trends, or component behaviors significantly different from normal patterns. All identified abnormal detection items are combined into a set, i.e., the abnormal detection item set.

[0042] Step S800: Potential fault detection of the target semiconductor device is performed based on the set of abnormal components and the set of abnormal detection items. In-depth analysis is conducted on the operating data of each component in the set of abnormal components, particularly data related to parameters or parameter combinations listed in the set of abnormal detection items. By comparing historical data, normal operating data, and current data, abnormal patterns are identified. An attempt is made to match the identified abnormal patterns with known fault patterns. If a matching fault pattern is found, the type, cause, and location of the potential fault can be preliminarily determined (locating the potential faulty component). Then, the impact of the potential fault on the overall performance and reliability of the equipment is assessed, including analyzing the possible consequences of the fault, such as production interruption, product quality degradation, or equipment damage, and finally predicting the probability of the potential fault occurring.

[0043] In the above text, refer to Figure 1 An anomaly identification method incorporating hierarchical parameter analysis according to embodiments of the present invention has been described in detail. Next, reference will be made to... Figure 2 An anomaly identification system incorporating hierarchical parameter analysis according to an embodiment of the present invention is described.

[0044] The anomaly identification system combining hierarchical parameter analysis according to embodiments of the present invention addresses the technical problems of existing semiconductor equipment fault detection, namely, the difficulty in accurately predicting and locating potential fault sources in complex semiconductor equipment and the low detection efficiency, thereby achieving the technical effect of improving the efficiency and accuracy of semiconductor equipment fault detection. The anomaly identification system combining hierarchical parameter analysis includes: a fault detection information extraction module 10, a first sub-event set determination module 20, a key parameter extraction module 30, a first parameter fluctuation duration acquisition module 40, a second-level fault component set acquisition module 50, a second-level operating parameter set acquisition module 60, a parameter anomaly trend identification module 70, and a potential equipment fault detection module 80.

[0045] The fault detection information extraction module 10 is used to extract fault detection information of the target semiconductor device, wherein the fault detection information includes the fault occurrence time point, fault event, and K first-level fault components.

[0046] First sub-event set determination module 20, the first sub-event set determination module 20 searches the fault record database based on the fault event to determine the first sub-event set;

[0047] The key parameter extraction module 30 is used to extract key parameters from the first sub-event set respectively to obtain a set of key parameter value sequences for the first sub-events.

[0048] The first parameter fluctuation duration acquisition module 40 is used to traverse the set of key parameter value sequences of the first sub-event to perform parameter fluctuation duration integration analysis and obtain the first parameter fluctuation duration.

[0049] The secondary fault component set acquisition module 50 is used to extract associated devices from the target semiconductor device using the K primary fault components as indexes, and obtain a set of K secondary fault components.

[0050] The secondary operating parameter set acquisition module 60 extracts operating parameters from the K secondary fault component sets based on the first parameter fluctuation time and the fault occurrence time point to obtain the K secondary operating parameter sets;

[0051] The parameter anomaly trend identification module 70 is used to identify parameter anomaly trends in the K sets of secondary operating parameters to obtain a set of abnormal components and a set of abnormal detection items.

[0052] The device potential fault detection module 80 is used to perform potential fault detection of the target semiconductor device based on the abnormal component set and the abnormal detection item set.

[0053] The specific configuration of the first parameter fluctuation duration acquisition module 40 will be described in detail below. The first parameter fluctuation duration acquisition module 40 further includes: constructing a first parameter change curve set based on the first sub-event key parameter value sequence set, wherein the horizontal axis of the curves in the first parameter change curve set represents time, and the vertical axis represents the key parameter value; dividing the first parameter change curve set according to a preset parameter tolerance threshold set to obtain first parameter abnormal time period clusters, wherein the first parameter abnormal time period cluster includes multiple first parameter abnormal time period sets, each first parameter abnormal time period set corresponding to a first parameter change curve; and performing parameter fluctuation duration integration analysis on the first parameter abnormal time period clusters to determine the first parameter fluctuation duration.

[0054] The specific configuration of the first parameter fluctuation duration acquisition module 40 will be described in detail below. The first parameter fluctuation duration acquisition module 40 may further include: performing period size statistics on multiple sets of first parameter abnormal time periods to obtain multiple sets of first parameter abnormal time lengths; calculating the mean of each of the multiple sets of first parameter abnormal time lengths to obtain multiple first parameter abnormal time averages; using the mode of the multiple first parameter abnormal time averages as a starting point, performing discrete cleaning on the multiple first parameter abnormal time averages to obtain a target set of first parameter abnormal time averages; and weighting the target set of first parameter abnormal time averages according to their distance from the starting point to generate the first parameter fluctuation duration.

[0055] The specific configuration of the first parameter fluctuation duration acquisition module 40 will be described in detail below. The first parameter fluctuation duration acquisition module 40 may further include: calculating the proportion of the mode of the multiple first parameter abnormal time mean values ​​among the multiple first parameter abnormal time mean values, and determining the cleaning step size based on the calculation result; determining a first discrete cleaning interval with the starting point as the interval center and the cleaning step size as the interval radius, wherein the first discrete cleaning interval includes a first interval density; expanding the first discrete cleaning interval according to the cleaning step size to obtain a second discrete cleaning interval, wherein the second discrete cleaning interval includes a second interval density; when the first interval density is less than or equal to the second interval density, continuing to expand and iterate the second discrete cleaning interval until the interval density difference between two adjacent cleanings is less than a preset interval density difference threshold, stopping the iteration, and taking the Nth discrete cleaning interval obtained in the last iteration as the target discrete cleaning interval; and taking the multiple first parameter abnormal time mean values ​​within the target discrete cleaning interval as the target first parameter abnormal time mean set.

[0056] The specific configuration of the secondary fault component set acquisition module 50 will be described in detail below. The secondary fault component set acquisition module 50 further includes: acquiring multiple components of the target semiconductor device; identifying pairwise correlation degrees of the multiple components according to preset correlation indicators to obtain multiple initial correlation degree sets and multiple initial associated component sets; filtering the multiple initial correlation degree sets according to preset correlation degree thresholds; mapping and cleaning the multiple initial associated component sets based on the filtering results to generate multiple associated component sets; and retrieving the multiple associated component sets using the K primary fault components as indexes to obtain K secondary fault component sets.

[0057] The specific configuration of the parameter anomaly trend identification module 70 will be described in detail below. The parameter anomaly trend identification module 70 further includes: obtaining a set of K compliance thresholds for K sets of secondary faulty components; performing out-of-specification analysis on the K sets of secondary operating parameters using the K compliance thresholds to determine K sets of out-of-specification operating parameters and K sets of compliant operating parameters; calculating the out-of-specification ratio based on the K sets of out-of-specification operating parameters and the K sets of secondary operating parameters to determine K parameter out-of-specification coefficients; identifying the K sets of compliant operating parameters according to K threshold edge bandwidths to determine K sets of edge compliant operating parameters; calculating the edge compliance ratio based on the K sets of edge compliant operating parameters and the K sets of compliant operating parameters to determine K parameter edge compliance coefficients; performing a weighted calculation on the K parameter out-of-specification coefficients and the K parameter edge compliance coefficients to obtain K anomaly coefficients; and identifying the K anomaly coefficients according to a preset anomaly coefficient threshold to determine the set of abnormal components.

[0058] The specific configuration of the parameter anomaly trend identification module 70 will be described in detail below. The parameter anomaly trend identification module 70 may further include: constructing a detection item identification network layer; using the detection item identification network layer to identify the set of abnormal components and the multiple sets of secondary operating parameters corresponding to the set of abnormal components, and determining the set of abnormal detection items.

[0059] The anomaly identification system combining hierarchical parameter analysis provided in this embodiment of the invention can execute the anomaly identification method combining hierarchical parameter analysis provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0060] Although this application makes various references to certain modules in the system according to the embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy distinction between each other and are not used to limit the scope of protection of this invention.

[0061] The specific embodiments described above do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. An anomaly detection method in combination with hierarchical parameter analysis, characterized by, The method comprises: extracting fault detection information of a target semiconductor device, wherein the fault detection information comprises a fault occurrence time point, a fault event and K primary fault components; searching in a fault record database based on the fault event to determine a first sub-event set; extracting key parameters from the first sub-event set respectively to obtain a first sub-event key parameter value sequence set; performing parameter fluctuation duration integration analysis on the first sub-event key parameter value sequence set to obtain a first parameter fluctuation duration; taking the K primary fault components as indexes to extract associated devices of the target semiconductor device to obtain a K secondary fault component set; performing running parameter extraction on the K secondary fault component set based on the first parameter fluctuation duration and the fault occurrence time point to obtain a K secondary running parameter set; performing parameter abnormal trend identification on the K secondary running parameter set to obtain an abnormal component set and an abnormal detection item set; performing potential fault detection of the target semiconductor device according to the abnormal component set and the abnormal detection item set; The abnormal identification method combined with hierarchical parameter analysis comprises: constructing a first parameter change curve set based on the first sub-event key parameter value sequence set, wherein the horizontal coordinate axis of a curve in the first parameter change curve set is time and the vertical coordinate axis is a key parameter value; dividing the first parameter change curve set according to a preset parameter tolerance threshold set respectively to obtain a first parameter abnormal period cluster, wherein the first parameter abnormal period cluster comprises a plurality of first parameter abnormal period sets, and each first parameter abnormal period set corresponds to a first parameter change curve; performing parameter fluctuation duration integration analysis on the first parameter abnormal period cluster to determine the first parameter fluctuation duration; The abnormal identification method combined with hierarchical parameter analysis comprises: performing period size statistics on a plurality of first parameter abnormal period sets to obtain a plurality of first parameter abnormal time length sets; performing mean value calculation on the plurality of first parameter abnormal time length sets respectively to obtain a plurality of first parameter abnormal time means; taking the mode of the plurality of first parameter abnormal time means as a starting point, performing discrete cleaning in the plurality of first parameter abnormal time means to obtain a target first parameter abnormal time mean set; performing weighted calculation on the target first parameter abnormal time mean set according to the distance to the starting point to generate the first parameter fluctuation duration; The parameter abnormal trend identification on the K secondary running parameter set to obtain an abnormal component set and an abnormal detection item set comprises: obtaining a K running parameter compliance threshold set of the K secondary fault component set; performing out-of-limit analysis on the K secondary running parameter set by using the K running parameter compliance threshold set to determine a K running parameter out-of-limit set and a K running parameter compliance set; performing out-of-limit proportion calculation based on the K running parameter out-of-limit set and the K secondary running parameter set to determine a K parameter out-of-limit coefficient; According to K threshold edge bandwidth, the K running parameter compliance sets are identified, K running parameter edge compliance sets are determined, the threshold edge bandwidth is a range, and the parameter values of the compliance threshold edge are included. The parameters of the threshold edge within the compliance range are identified to form the K running parameter edge compliance sets; Based on the K running parameter edge compliance sets and the K running parameter compliance sets, edge compliance proportion calculation is performed to determine K parameter edge compliance coefficients; The K parameter ultra-rule coefficients and the K parameter edge compliance coefficients are weighted calculated to obtain K abnormal coefficients; According to the preset abnormal coefficient threshold, the K abnormal coefficients are identified to determine the abnormal component set; The abnormal identification method combined with hierarchical parameter analysis includes: A detection item identification network layer is constructed; The abnormal component set and the plurality of secondary running parameter sets corresponding to the abnormal component set are identified by using the detection item identification network layer to determine the abnormal detection item set.

2. The method of claim 1, wherein the abnormality identification method is combined with a hierarchical parameter analysis. It includes: The proportion of the mode of the plurality of first parameter abnormal time averages in the plurality of first parameter abnormal time averages is calculated, and the cleaning step is determined according to the calculation result; A first discrete cleaning interval is determined with the start point as the interval center and the cleaning step as the interval radius, wherein the first discrete cleaning interval includes a first interval density; The first discrete cleaning interval is expanded according to the cleaning step to obtain a second discrete cleaning interval, wherein the second discrete cleaning interval includes a second interval density; When the first interval density is less than or equal to the second interval density, the second discrete cleaning interval is continuously expanded and iterated until the interval density difference of adjacent two times of cleaning is less than a preset interval density difference threshold, and the iteration is stopped. The Nth discrete cleaning interval obtained by the last iteration is taken as a target discrete cleaning interval; The plurality of first parameter abnormal time averages in the target discrete cleaning interval are taken as the target first parameter abnormal time average set.

3. The method of claim 1, wherein the abnormality identification method is combined with a hierarchical parameter analysis. The target semiconductor equipment is associated with equipment extraction with the K primary fault components as indexes to obtain K secondary fault component sets, including: A plurality of components of the target semiconductor equipment are obtained; The plurality of components are associated with each other according to a preset association index to obtain a plurality of initial association degree sets and a plurality of initial associated component sets; The plurality of initial association degree sets are screened according to a preset association degree threshold, and the plurality of initial associated component sets are mapped and cleaned according to the screening result to generate a plurality of associated component sets; The plurality of associated component sets are searched with the K primary fault components as indexes to obtain K secondary fault component sets.

4. An anomaly detection system incorporating hierarchical parameter analysis, characterized by, The system is used to implement the abnormal identification method combined with hierarchical parameter analysis according to any one of claims 1-3, and the system includes: A fault detection information extraction module is used to extract fault detection information of a target semiconductor equipment, wherein the fault detection information includes a fault occurrence time point, a fault event, and K primary fault components; The first sub-event set determination module determines a first sub-event set based on searching the fault event in a fault record database; The key parameter extraction module is configured to extract key parameters from the first sub-event set respectively, and obtain a first sub-event key parameter value sequence set; The first parameter fluctuation duration obtaining module is configured to analyze the first sub-event key parameter value sequence set to obtain a first parameter fluctuation duration; The secondary fault component set obtaining module is configured to take the K primary fault components as indexes to extract associated equipment of the target semiconductor device, and obtain K secondary fault component sets; The secondary running parameter set obtaining module is configured to extract running parameters from the K secondary fault component sets based on the first parameter fluctuation duration and the fault occurrence time point, and obtain K secondary running parameter sets; The parameter abnormal trend identification module is configured to identify parameter abnormal trends of the K secondary running parameter sets, and obtain an abnormal component set and an abnormal detection item set; The device potential fault detection module is configured to detect potential faults of the target semiconductor device according to the abnormal component set and the abnormal detection item set.

Images