Gate oxide stability monitoring method for MOSFET fabrication

Abstract

The present invention discloses a gate oxide stability monitoring method for MOSFET preparation, which relates to the field of semiconductor device manufacturing. The method comprises: using a sensor group to monitor the real-time data of the gate oxide; transmitting the real-time data set to a central database for time series data analysis and extracting key time series features; inputting to an independent abnormality identification channel and outputting a parameter abnormality identification value; establishing a conditional dependency relationship with a labeling database, initializing a joint prediction network, performing a time series joint probability prediction of gate oxide failure for the parameter abnormality identification value, establishing a time series joint probability prediction result, and completing stability monitoring according to the time series joint probability prediction result. The method solves the technical problems of poor real-time performance and difficulty in timely warning of potential problems in the existing MOSFET gate oxide stability monitoring, and achieves the technical effect of real-time monitoring of the state change of the gate oxide layer and timely discovery of potential problems.

Description

Technical Field

[0001] The present application relates to the field related to semiconductor device manufacturing, and in particular to a gate oxide layer stability monitoring method for MOSFET preparation. Background Art

[0002] In modern semiconductor technology, metal oxide semiconductor field effect transistors (MOSFETs) are core components in integrated circuits, and their performance stability and reliability are crucial to the operation of the entire circuit system. As the key interface for controlling current, the quality and stability of the gate oxide layer of MOSFET directly affect the performance and life of the device. As integrated circuit technology develops towards smaller line widths and higher integration, the thickness of the gate oxide layer continues to decrease, making it more susceptible to the environment, stress, and tiny defects in the process, which in turn causes problems such as oxide layer degradation, increased leakage, and even failure. In the existing MOSFET preparation process, the monitoring of the stability of the gate oxide layer mainly relies on offline testing methods, such as breakdown voltage testing and capacitance-voltage (CV) characteristic analysis. However, offline testing cannot reflect the subtle changes in the gate oxide layer during the preparation process in real time, and it is difficult to warn of potential problems in a timely manner.

[0003] In the current related technologies, the gate oxide layer stability monitoring used in MOSFET preparation has poor real-time performance and is difficult to promptly warn of potential problems. Summary of the invention

[0004] The present application provides a gate oxide stability monitoring method for MOSFET preparation, uses integrated multiple sensors (electrical sensor, temperature and humidity sensor, stress sensor, optical sensor) to monitor the gate oxide data in real time, uses Internet of Things technology and big data analysis methods to conduct in-depth mining and timing analysis of the monitoring data, identifies parameter anomalies through independent anomaly recognition channels, and performs timing joint probability prediction of gate oxide failure based on a joint prediction network, thereby achieving the technical effect of real-time monitoring of the state changes of the gate oxide and timely discovery of potential problems.

[0005] The present application provides a gate oxide layer stability monitoring method for MOSFET preparation, comprising:

[0006] A sensor group is used to monitor the real-time data of the gate oxide layer and establish a real-time data set, wherein the sensor group includes an electrical sensor, a temperature and humidity sensor, a stress sensor, and an optical sensor; the real-time data set is transmitted to a central database through an Internet of Things platform, and a time series data analysis is performed on each group of data mapped with the sensor in the real-time data set to extract key time series features; the key time series features are used as input data and input into an independent abnormality identification channel, wherein the independent abnormality identification channel is integrated with an identification sub-channel mapped with the sensor, and a parameter abnormality identification value with a time series identifier is output through the independent abnormality identification channel; an annotation database is obtained, a conditional dependency relationship is established with the annotation database, and a joint prediction network is initialized based on the conditional dependency relationship, and a time series joint probability prediction of gate oxide failure is performed on the parameter abnormality identification value based on the initialized joint prediction network, a time series joint probability prediction result is established, and stability monitoring is completed according to the time series joint probability prediction result.

[0007] In a possible implementation, the conditional dependency relationship is established by using the annotation database, and the following processing is performed:

[0008] Classify the conditional dependency relationships and establish direct dependency, indirect dependency and cross dependency; establish the direct dependency, indirect dependency and cross dependency through domain knowledge to generate a basic establishment result, and the basic establishment result is set with a first stable value; use the annotation database to calculate the joint probability distribution of each pair of parameters and the marginal probability distribution of each parameter, calculate the common information value based on the joint probability distribution and the marginal probability distribution, and generate additional establishment results of direct dependency, indirect dependency and cross dependency with the common information value, and the additional establishment result is set with a second stable value; establish the conditional dependency relationship according to the basic establishment result, the additional establishment result, the first stable value and the second stable value.

[0009] In a possible implementation, the conditional dependency relationship is established according to the basic establishment result, the additional establishment result, the first stable value, and the second stable value, and the following processing is performed:

[0010] A conflict analysis is performed on the basic establishment results and the additional establishment results, and a penalty verification is performed on the conflict analysis results through a first stable value and a second stable value to establish a penalty verification result; the labeled database is used as test data, and the state-action reward reinforcement of the penalty verification result is executed, and the penalty verification result is updated according to the state-action reward reinforcement result to establish a conditional dependency relationship.

[0011] In a possible implementation, the state action reward reinforcement of the execution penalty verification result performs the following processing:

[0012] Establish the state-action value function as follows: ;

[0013] in, Indicates in status Take action When , the expected cumulative reward is, is the index of the time step, is the discount factor, , , They are the short-term, medium-term and long-term reward weight factors respectively. For the time step Short-term immediate rewards, For the time step Mid-term immediate reward, At time step Long-term immediate rewards, Characterize the system at time status, Characterize the system at time Actions, Representation state, Represent actions; use state-action value function to reinforce state-action rewards.

[0014] In a possible implementation, the following processing is performed:

[0015] Establish independent parameter thresholds for each sensor in the sensor group; determine whether the parameter abnormality identification value triggers the independent parameter threshold; if the independent parameter threshold is triggered, generate first verification information; if the time series joint probability prediction result is an abnormal result based on the parameter abnormality identification value corresponding to the triggered independent parameter threshold, eliminate the first verification information, and complete stability monitoring based on the time series joint probability prediction result.

[0016] In a possible implementation, the following processing is performed:

[0017] If the time series joint probability prediction result is not an abnormal result based on the parameter abnormality identification value corresponding to the triggering independent parameter threshold, the first verification information is set as the retained abnormality information, and the retained time series abnormality identification is performed according to the retained abnormality information.

[0018] In a possible implementation, after the real-time data set is transmitted to the central database through the Internet of Things platform, the following processing is performed:

[0019] The neighborhood data of the real-time data set is smoothed based on the central database, and the abnormal area is identified; after adding abnormal attention to the abnormal area, time series data analysis is performed.

[0020] In a possible implementation, the stability monitoring is completed according to the time series joint probability prediction results, and the following processing is performed: a stable anomaly identification library is established, anomaly matching of the stable anomaly identification library is performed through the time series joint probability prediction results, and anomaly matching results are obtained; stable anomaly reporting is performed according to the anomaly level corresponding to the anomaly matching result.

[0021] The gate oxide stability monitoring method for MOSFET preparation proposed in the present application first uses a sensor group to perform real-time data monitoring of the gate oxide layer to establish a real-time data set, the sensor group includes an electrical sensor, a temperature and humidity sensor, a stress sensor, and an optical sensor, and then transmits the real-time data set to a central database through an Internet of Things platform, and performs time series data analysis on each group of data mapped to the sensor in the real-time data set, extracts key timing features, and then uses the key timing features as input data and inputs them into an independent abnormality recognition channel, the independent abnormality recognition channel integrates an identification sub-channel mapped to the sensor, and outputs a parameter abnormality recognition value with a timing identifier through the independent abnormality recognition channel, and finally obtains a labeling database, establishes a conditional dependency relationship with the labeling database, and initializes a joint prediction network based on the conditional dependency relationship, performs a time series joint probability prediction of gate oxide failure on the parameter abnormality recognition value based on the initialized joint prediction network, establishes a time series joint probability prediction result, and completes stability monitoring based on the time series joint probability prediction result, thereby achieving the technical effect of real-time monitoring of the state change of the gate oxide layer and timely discovering potential problems. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solution of the embodiment of the present invention, the accompanying drawings of the embodiment of the present invention will be briefly introduced below. A flow chart is used in the present application to illustrate the operations performed by the method according to the embodiment of the present application. It should be understood that the preceding or following operations are not necessarily performed accurately in order. On the contrary, various steps can be processed in reverse order or simultaneously as needed. At the same time, other operations can also be added to these processes, or one or more operations can be removed from these processes.

[0023] Figure 1 A schematic flow chart of a gate oxide layer stability monitoring method for MOSFET fabrication provided in an embodiment of the present application.

[0024] Figure 2 A schematic diagram of a flow chart of establishing conditional dependencies using a labeled database in a gate oxide layer stability monitoring method for MOSFET fabrication provided in an embodiment of the present application. DETAILED DESCRIPTION

[0025] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below.

[0026] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings. The described embodiments should not be regarded as limiting the present application. All other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of this application.

[0027] In the following description, reference is made to "some embodiments", which describe a subset of all possible embodiments, but it is understood that "some embodiments" may be the same subset or different subsets of all possible embodiments, and may be combined with each other without conflict, and the terms "first\second" involved are merely to distinguish similar objects and do not represent a specific ordering of objects. The terms "including" and "having" and any variations are intended to cover non-exclusive inclusions, 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 clearly listed, but may include other steps or modules that are not clearly 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 those generally understood by technicians in the technical field of this application. The terms used herein are for the purpose of describing the embodiments of the present application only.

[0028] The present application provides a gate oxide layer stability monitoring method for MOSFET preparation, such as Figure 1 As shown, the method includes:

[0029] Step S100, using a sensor group to monitor the real-time data of the gate oxide layer and establish a real-time data set, the sensor group includes an electrical sensor, a temperature and humidity sensor, a stress sensor, and an optical sensor. Specifically, during the preparation of the MOSFET, the electrical sensor, the temperature and humidity sensor, the stress sensor, and the optical sensor are deployed near the gate oxide layer or in direct contact with it. These sensors are used to sense and collect various physical quantities related to the state of the gate oxide layer in real time. Each sensor continuously collects data according to the set sampling frequency and transmits the data to the data collection device, including current, voltage (electrical sensor), temperature, humidity (temperature and humidity sensor), stress level (stress sensor) and optical properties (such as reflectivity, transmittance, etc., optical sensor). The collected raw data is sorted, cleaned and preprocessed to remove noise, fill missing values, etc. to ensure the quality and consistency of the data. The sorted data forms a real-time data set for subsequent data analysis and monitoring.

[0030] Step S200, transmit the real-time data set to the central database through the Internet of Things platform, and perform time series data analysis on each set of data mapped to the sensor in the real-time data set to extract key time series features. Specifically, the real-time data set is transmitted from the data collection device to the central database using the Internet of Things platform (such as MQTT, CoAP and other protocols). Among them, the Internet of Things platform is used to provide a stable and efficient data transmission channel to ensure the real-time and integrity of the data. Perform time series analysis on each set of data mapped to the sensor in the central database, including data trend analysis, periodicity detection, outlier identification, etc., to reveal the laws and characteristics of data changes over time. Key features are extracted from the time series data by means of signal processing techniques (such as Fourier transform, wavelet transform), statistical methods (such as mean, variance, standard deviation), etc., which can reflect important information about the state of the gate oxide layer.

[0031] In a possible implementation, after the real-time data set is transmitted to the central database through the Internet of Things platform, step S200 further includes step S210, performing neighborhood data smoothing of the real-time data set based on the central database and identifying abnormal areas. Specifically, the real-time data set is retrieved from the central database, and these data sets contain various types of data about the state of the gate oxide layer collected by different sensors. For each data point (or data sequence) in the data set, the data point in its neighborhood is selected, wherein the selection of the neighborhood can be determined according to the time window (such as data of a few minutes or hours before and after) or the spatial position (data has spatial attributes). Smoothing algorithms (such as moving average, exponential smoothing, median filtering, etc.) are applied to the selected neighborhood data to reduce the impact of noise and random fluctuations on the data. Using statistical methods (such as threshold detection, distance-based anomaly detection), etc., anomaly detection is performed on the smoothed data to find data points or data areas that are significantly different from the normal data pattern. The detected abnormal data points or data areas are identified in the real-time data set so that these areas can be focused on in subsequent analysis. Step S220, after adding abnormal attention to the abnormal area, time series data analysis is performed. Specifically, before performing the time series data analysis, attention is added to the data in these areas according to the abnormal areas identified in step S210. On the basis of adding attention to the abnormal areas, time series data analysis is performed on the entire real-time data set (including but not limited to the abnormal areas) to extract the key time series features that reflect the changes in the state of the gate oxide layer. The time series analysis results of the abnormal areas and non-abnormal areas are integrated to form a more comprehensive data set feature representation. Since the original data contains a large amount of noise and random fluctuations, these factors will interfere with the judgment of the true state of the gate oxide layer. Therefore, this implementation method first performs neighborhood data smoothing and anomaly detection after transferring the data to the central database, which significantly reduces noise interference and improves data quality. At the same time, after adding abnormal attention to the abnormal area, the time series data analysis is performed again, which more accurately captures the changing trend and abnormal characteristics of the gate oxide state, which helps to more accurately identify parameter anomalies in the future.

[0032] Step S300, the timing key features are used as input data and input into an independent abnormality identification channel, in which an identification sub-channel mapped with a sensor is integrated, and a parameter abnormality identification value with a timing mark is output through the independent abnormality identification channel. Specifically, the independent abnormality identification channel is a module dedicated to abnormality identification, which contains multiple identification sub-channels mapped with sensors. The extracted timing key features are used as input data and input into the independent abnormality identification channel. Each sub-channel in the independent abnormality identification channel is responsible for monitoring whether the data of the corresponding sensor is abnormal. Among them, the recognition algorithm includes threshold-based detection, statistical tests (such as t-test, chi-square test), machine learning models (such as classifiers, abnormality detection algorithms), etc. When an abnormality is detected, the identification sub-channel outputs a parameter abnormality identification value with a timing mark, wherein the timing mark is used to identify the time point and duration of the abnormality, which is used for subsequent analysis and tracking.

[0033] Step S400, obtaining a labeling database, establishing a conditional dependency relationship with the labeling database, and initializing a joint prediction network based on the conditional dependency relationship, performing a time-series joint probability prediction of gate oxide failure on the parameter abnormality identification value based on the initialized joint prediction network, establishing a time-series joint probability prediction result, and completing stability monitoring according to the time-series joint probability prediction result. Specifically, the labeling database is a data set containing known gate oxide states (normal or failed) and their corresponding parameter values. These data samples have been manually labeled and experimentally verified, have high accuracy and reliability, and are used to train and verify the prediction model. Using the data in the labeling database, analyze the correlation between different parameters, establish conditional dependencies, including direct dependencies, indirect dependencies, and cross dependencies, to reflect the complex interactions and influences between parameters. Based on the conditional dependency relationship, initialize the joint prediction network, which is a prediction model that can comprehensively consider multiple parameter information and can output a time-series joint probability prediction result of gate oxide failure. The parameter abnormality identification value is used as an input and input into the joint prediction network to perform a time-series joint probability prediction of gate oxide failure. The prediction result gives the probability distribution of gate oxide failure in the future, providing a basis for stability monitoring. The embodiment of the present application integrates multiple sensors (electrical sensor, temperature and humidity sensor, stress sensor, optical sensor) to monitor the gate oxide layer in real time, uses the Internet of Things technology and big data analysis methods to perform deep mining and timing analysis on the monitoring data, identifies parameter anomalies through independent abnormality identification channels, and performs timing joint probability prediction of gate oxide failure based on a joint prediction network. The technical effect of real-time monitoring of the state changes of the gate oxide layer and timely discovery of potential problems is achieved.

[0034] like Figure 2As shown, in a possible implementation, the conditional dependency relationship is established with the annotation database, and step S400 further includes step S410, classifying the conditional dependency relationship, and establishing direct dependency, indirect dependency and cross dependency. Specifically, historical data and known association information between parameters are extracted from the annotation database, and these parameters include data from different sensors, such as electrical parameters, temperature and humidity, stress level and optical properties. According to the physical meaning of the parameters and the nature of the interaction, the conditional dependency relationship is divided into three categories: direct dependency (two or more parameters directly affect each other, and the change of one parameter directly leads to the change of another parameter), indirect dependency (parameters affect each other through intermediate variables or processes, that is, the change of one parameter indirectly leads to the change of another parameter) and cross dependency (a complex interaction network is formed between multiple parameters, in which the change of any parameter may have an impact on multiple other parameters). According to the above standards, the parameter relationships in the annotation database are classified to determine which are direct dependencies, indirect dependencies and cross dependencies. Step S420, the direct dependency, indirect dependency and cross dependency are established through domain knowledge to generate a basic establishment result, and the basic establishment result is set with a first stable value. Specifically, combined with professional knowledge in the field of MOSFET preparation, such as materials science, semiconductor physics, etc., the dependencies classified in step S410 are further analyzed and confirmed. According to domain knowledge, a mathematical model or logical framework is established for each dependency to describe the interaction mechanism between parameters. An initial stable value (first stable value) is set for each dependency, which reflects the stability and reliability of the dependency under current conditions. Step S430, the joint probability distribution of each pair of parameters and the marginal probability distribution of each parameter are calculated using the annotation database, and the common information value is calculated based on the joint probability distribution and the marginal probability distribution. The additional establishment results of direct dependence, indirect dependence and cross-dependence are generated with the common information value, and the additional establishment results are set with a second stable value. Specifically, a statistical method is used to calculate the joint probability distribution of each pair of parameters (i.e., the probability of two parameters appearing at the same time) and the marginal probability distribution of each parameter (i.e., the probability of a single parameter appearing) from the annotation database. Based on the joint probability distribution and the marginal probability distribution, the common information or related concepts in information theory are used to calculate the amount of information or the degree of dependence between the parameters. The common information value reflects the amount of shared information between the parameters, which is used to quantify the strength of the dependency. According to the common information value, an additional establishment result is generated for each dependency relationship, and the additional establishment result is provided with a second stability value, which further refines the stability and reliability of the dependency relationship.

[0035] Step S440, establish a conditional dependency relationship based on the basic establishment results, the additional establishment results, the first stable value, and the second stable value. Specifically, the basic establishment results (including the first stable value) and the additional establishment results (including the second stable value) are comprehensively evaluated to fully consider various aspects of the dependency relationship (such as physical mechanisms, historical data, statistical characteristics, etc.). Based on the results of the comprehensive evaluation, the conditional dependency relationship is finally established. These relationships not only reflect the direct, indirect and cross-dependencies between parameters, but also contain the stability and reliability information of these dependencies. This implementation method establishes a more accurate and comprehensive conditional dependency relationship by comprehensively considering multiple factors (including physical mechanisms, historical data, statistical characteristics, etc.), combining the basic establishment results and the additional establishment results, and provides a more reliable basis for gate oxide failure prediction.

[0036] In a possible implementation, the conditional dependency is established according to the basic establishment result, the additional establishment result, the first stable value, and the second stable value. Step S440 further includes step S441, performing conflict analysis on the basic establishment result and the additional establishment result, and performing penalty verification on the conflict analysis result through the first stable value and the second stable value to establish a penalty verification result. Specifically, a comparison is made between the basic establishment result (based on domain knowledge) and the additional establishment result (based on statistics and common information value), and possible conflict points are identified, which appear in the strength, stability or impact direction of different dependencies. The identified conflict points are analyzed in depth to evaluate the causes and possible consequences thereof, including considering factors such as the physical mechanisms of different dependencies, the reliability of historical data, and the applicability of statistical methods. The first stable value (the basic stable value based on domain knowledge) and the second stable value (the additional stable value based on statistics and common information value) are used as references to perform penalty verification on the conflict analysis result. That is, if the result of a certain conflict point is significantly inconsistent with the first stable value or the second stable value, a penalty adjustment is made to the conflict point to reduce its weight or influence in the final conditional dependency. According to the result of the penalty verification, the processing method and weight of the conflict point are updated to generate the final penalty verification result. Step S442, the annotated database is used as test data, the state action reward reinforcement of the penalty verification result is performed, and the penalty verification result is updated according to the state action reward reinforcement result to establish a conditional dependency relationship. Specifically, part or all of the data in the annotated database is used as test data, and these data cover various possible scenarios and conditions. The penalty verification result is applied to the test data to simulate the dependency changes under different states (i.e., different parameter combinations and conditions). For each state, the corresponding action (i.e., adjustment or prediction of the dependency) is performed, and rewards or penalties are given according to the difference between the predicted result and the real result. According to the result of the state action reward reinforcement, the penalty verification result is updated. That is, if a dependency performs well in the test data (i.e., the prediction accuracy is high), its weight or stability is increased; otherwise, its weight or stability is reduced. Based on the updated penalty verification result and the state action reward reinforcement result, a conditional dependency relationship is finally established. This relationship comprehensively considers the domain knowledge, statistical information, common information value and the verification result of the test data, and provides a reliable basis for the prediction of gate oxide failure. This implementation method effectively solves the conflicts and inconsistencies that may exist between different data sources through conflict analysis and penalty verification mechanisms. It uses the labeled database as test data and verifies and updates it through state-action reward reinforcement, ensuring that the final dependency relationship is highly accurate and adaptable in practical applications.

[0037] In a possible implementation, the state-action reward reinforcement of the execution penalty verification result, step S442 further includes step S4421, establishing a state-action value function, as follows: ;

[0038] in, Indicates in status Take action When , the expected cumulative reward is, is the index of the time step, is the discount factor, , , They are the short-term, medium-term and long-term reward weight factors respectively. For the time step Short-term immediate rewards, For the time step Mid-term immediate reward, At time step Long-term immediate rewards, Characterize the system at time status, Characterize the system at time Actions, Representation state, Representing the action; Step S4422, using the state-action value function to perform state-action reward reinforcement.

[0039] In a possible implementation, the stability monitoring is completed according to the timing joint probability prediction result, and step S400 further includes step S450, establishing a stable abnormal identification library, performing abnormal matching of the stable abnormal identification library through the timing joint probability prediction result, and obtaining the abnormal matching result. Specifically, cases that lead to gate oxide failure are collected and analyzed from past monitoring data, including abnormal parameters, abnormal types, time series of abnormal occurrence, and final failure modes in these cases. Different abnormal levels are defined according to the potential impact of the abnormality on the stability of the gate oxide layer, such as slight abnormality, medium abnormality, severe abnormality, etc. A mapping relationship is established between each abnormal type, abnormal parameter and its timing characteristics and the corresponding abnormal level, and stored in the stable abnormal identification library. Key features are extracted from the timing joint probability prediction result, and the prediction result is compared with the entry in the stable abnormal identification library using a matching algorithm, and the matched abnormal type, abnormal level, and corresponding timestamp information are recorded. Step S460, stable abnormality is reported according to the abnormal level corresponding to the abnormal matching result. Specifically, according to the abnormal matching result, the currently monitored abnormal level is determined. According to the abnormal level, the corresponding notification mechanism is triggered. For example, for serious anomalies, relevant personnel need to be notified immediately through the emergency alarm system; for minor anomalies, reminders are given via email or message push. The report content includes the type of anomaly, the level of anomaly, the time of occurrence, the possible scope of impact, and the recommended response measures. This implementation method achieves timely warning of potential failure risks by establishing a stable anomaly identification library and performing anomaly matching, and responds in a graded manner according to the level of anomalies, reasonably allocates resources, and avoids overreaction or neglect of important issues.

[0040] In a possible implementation, step S400 further includes step S470, establishing independent parameter thresholds for each sensor in the sensor group. Specifically, based on historical data, experimental data and expert experience, an independent parameter threshold is set for each sensor (electrical sensor, temperature and humidity sensor, stress sensor, optical sensor). These thresholds are used to determine whether the data monitored by the sensor is within the normal range, thereby preliminarily identifying possible anomalies. For example, the electrical sensor monitors the resistance change of the gate oxide layer and sets a threshold range for the resistance change; the temperature and humidity sensor sets the upper and lower thresholds of temperature and humidity. Step S480, determine whether the parameter abnormality identification value triggers the independent parameter threshold. Specifically, the obtained parameter abnormality identification value is compared with the set independent parameter threshold. If a parameter abnormality identification value exceeds its corresponding threshold range, it is considered that the parameter triggers an abnormal condition and needs further processing. Step S490, if the independent parameter threshold is triggered, the first verification information is generated. Specifically, when a parameter triggers an abnormal threshold, a first verification information is generated as a preliminary confirmation of the abnormal event. This information contains information such as the type of sensor that triggers the abnormality, the abnormal parameter value, and the timestamp, which is used for subsequent analysis and verification. Step S4100: If the time series joint probability prediction result is an abnormal result based on the parameter abnormality identification value corresponding to the triggering independent parameter threshold, the first verification information is eliminated, and stability monitoring is completed according to the time series joint probability prediction result. Specifically, if the time series joint probability prediction result confirms the abnormal result based on the parameter abnormality identification value corresponding to the triggering independent parameter threshold, then the abnormality is considered to be true and important, so the first verification information generated previously is eliminated (because the joint prediction result has provided a more comprehensive verification), and stability monitoring is completed directly according to the time series joint probability prediction result. This implementation method performs preliminary screening through independent parameter thresholds and then verifies through joint predictions, which not only ensures the timeliness of abnormal identification, but also improves accuracy. This double verification mechanism helps to reduce the misjudgment rate and improve the reliability of the system.

[0041] In a possible implementation, step S400 further includes step S4110, if the timing joint probability prediction result is not an abnormal result based on the parameter abnormality identification value corresponding to the triggering independent parameter threshold, the first verification information is set as the reserved abnormal information, and the reserved timing abnormality identification is performed according to the reserved abnormal information. Specifically, when the timing joint probability prediction result output by the joint prediction network is inconsistent with the abnormality triggered by the independent parameter threshold, the previously generated first verification information is set as the reserved abnormal information. That is, although a certain sensor data triggers its independent parameter threshold, the joint prediction network does not consider this to be a significant abnormality sufficient to affect the stability of the gate oxide layer. Continue to track this reserved abnormal information and associate it with subsequent timing data for analysis, including further analyzing the time series pattern of the sensor data that triggers the threshold to see if other sensor data also show similar abnormal trends or evaluate whether this abnormality may be a temporary, unimportant fluctuation. According to the result of the reserved timing abnormality identification, decide whether further action is needed, such as adjusting the monitoring parameters, increasing the monitoring frequency, starting additional diagnostic procedures, etc., and the reserved abnormal information and its analysis results can also be fed back to the operator or maintenance team so that they can make decisions based on the actual situation. The joint prediction network can comprehensively consider the data from multiple sensors and the time dependency between the data to accurately evaluate the overall state of the gate oxide layer. This implementation improves the accuracy of anomaly identification by retaining abnormal information and performing further analysis when the independent parameter threshold is inconsistent with the joint prediction result.

[0042] The above specific implementation manner does not constitute a limitation to the protection scope of the present application. It should be understood by those skilled in the art 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 the present application should be included in the protection scope of the present application. In some cases, the actions or steps recorded in the present application can be performed in an order different from that in the embodiment and can still achieve the desired results. In addition, the process depicted in the accompanying drawings does not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

Claims

1. A gate oxide layer stability monitoring method for MOSFET preparation, characterized in that: The method comprises: Using a sensor group to monitor the real-time data of the gate oxide layer and establish a real-time data set, the sensor group includes an electrical sensor, a temperature and humidity sensor, a stress sensor, and an optical sensor; The real-time data set is transmitted to the central database through the IoT platform, and time series data analysis is performed on each set of data mapped to the sensor in the real-time data set to extract key time series features; The timing key feature is used as input data and input into an independent abnormality identification channel, wherein the independent abnormality identification channel integrates an identification sub-channel mapped with a sensor, and outputs a parameter abnormality identification value with a timing mark through the independent abnormality identification channel; Acquire a labeling database, establish a conditional dependency relationship with the labeling database, initialize a joint prediction network based on the conditional dependency relationship, perform a timing joint probability prediction of gate oxide failure on the parameter abnormality identification value based on the initialized joint prediction network, establish a timing joint probability prediction result, and complete stability monitoring according to the timing joint probability prediction result; The step of establishing the conditional dependency relationship by using the annotation database further includes: Classify the conditional dependencies and establish direct dependencies, indirect dependencies and cross dependencies; Establishing dependencies for direct dependencies, indirect dependencies, and cross dependencies through domain knowledge to generate a basic establishment result, wherein the basic establishment result is set with a first stable value; Calculate the joint probability distribution of each pair of parameters and the marginal probability distribution of each parameter using the annotation database, calculate the common information value based on the joint probability distribution and the marginal probability distribution, and generate additional establishment results of direct dependency, indirect dependency and cross dependency with the common information value, wherein the additional establishment results are provided with a second stable value; Establish a conditional dependency relationship according to the basic establishment result, the additional establishment result, the first stable value, and the second stable value; The establishing of the conditional dependency relationship according to the basic establishment result, the additional establishment result, the first stable value, and the second stable value also includes: Performing conflict analysis on the basic establishment result and the additional establishment result, and performing penalty verification on the conflict analysis result through the first stable value and the second stable value to establish a penalty verification result; Use the labeled database as test data, perform state-action reward reinforcement of the penalty verification result, and update the penalty verification result according to the state-action reward reinforcement result to establish a conditional dependency relationship; The method further comprises: Establishing independent parameter thresholds for each sensor in the sensor group; Determining whether the parameter abnormality identification value triggers the independent parameter threshold; If the independent parameter threshold is triggered, generating first verification information; If the time series joint probability prediction result is an abnormal result based on the parameter abnormality identification value corresponding to the trigger independent parameter threshold, the first verification information is eliminated, and stability monitoring is completed according to the time series joint probability prediction result; If the time series joint probability prediction result is not an abnormal result based on the parameter abnormality identification value corresponding to the triggering independent parameter threshold, the first verification information is set as the retained abnormality information, and the retained time series abnormality identification is performed according to the retained abnormality information.

2. The gate oxide layer stability monitoring method for MOSFET preparation according to claim 1, characterized in that: The state action reward reinforcement of the execution penalty verification result also includes: Establish the state-action value function as follows: ;in, Indicates in status Take action When , the expected cumulative reward is, is the index of the time step, is the discount factor, , , They are the short-term, medium-term and long-term reward weight factors respectively. For the time step Short-term immediate rewards, For the time step Mid-term immediate reward, At time step Long-term immediate rewards, Characterize the system at time status, Characterize the system at time Actions, Representation state, Represent actions; use state-action value function to reinforce state-action rewards.

3. The gate oxide layer stability monitoring method for MOSFET preparation according to claim 1, characterized in that: After transmitting the real-time data set to the central database through the Internet of Things platform, it also includes: Perform neighborhood data smoothing of real-time data sets based on the central database and identify abnormal areas; After adding abnormal attention to the abnormal area, time series data analysis is performed.

4. The gate oxide layer stability monitoring method for MOSFET preparation according to claim 1, characterized in that: The stability monitoring is completed according to the time series joint probability prediction result, and further includes: Establish a stable anomaly identification library, perform anomaly matching of the stable anomaly identification library through the time series joint probability prediction results, and obtain anomaly matching results; Stable exception reporting is performed based on the exception level corresponding to the exception matching result.

Images