Intelligent calibration method and system for semiconductor detection equipment

Abstract

The invention discloses an intelligent calibration method and system for semiconductor detection equipment, particularly relates to the field of semiconductor detection equipment, is used for solving the problem of calibration error traceability, and aims at solving the problem of calibration error traceability by performing time serialization splitting on measurement data and constructing a corresponding relation between a measurement time grid and a measurement area. According to the method, the source of the error can be accurately positioned in time and space, meanwhile, whether the error is short-term fluctuation or long-term abnormity is identified by combining deep analysis on the trend of the measurement error, so that accidental errors and hardware faults are distinguished, and when the error belongs to the short-term fluctuation, the previous measurement data is backtracked according to the measurement time lattice, and the error is accurately determined. The method comprises the following steps of: selecting an interval with a stable environment, determining an earliest error occurrence position and a corresponding measurement area, and executing local calibration on the area, so that redundant operation repeatedly executed in a whole process in a traditional method is avoided, and semiconductor detection equipment can still maintain relatively high measurement stability and data reliability under a high-precision requirement.

Description

Technical Field

[0001] The present invention relates to the field of semiconductor detection equipment. More specifically, the present invention relates to an intelligent calibration method and system for a semiconductor detection equipment. Background Art

[0002] In semiconductor detection, as a high-precision measurement device, the calibration process of an optical profiler is directly related to the accuracy of measurement data and production efficiency. Traditional calibration methods usually adopt a full-process retry method to deal with any out-of-tolerance situation. Once the calibration result exceeds the preset tolerance, the system will re-execute the entire process without carefully splitting and analyzing the collected measurement data and environmental data. This approach fails to utilize the distribution characteristics of data in the time grid and space, and also ignores the influence of environmental conditions on errors, making it impossible to accurately locate the specific time period and area where errors occur.

[0003] When the current system encounters an out-of-tolerance calibration result, it directly adopts an operation strategy of full-process retry, and fails to comprehensively evaluate the deviation degree of the error from the historical data based on the collected preprocessed data, combined with environmental impact and time fluctuation. As a result, it is impossible to distinguish accidental errors from hardware failures, and it is also impossible to determine the starting point of the error and the local abnormal area through backtracking, leading to repeated operations, waste of detection time, and difficulty in targeted maintenance.

[0004] To solve the above problems, a technical solution is provided now. Summary of the Invention

[0005] To overcome the above-mentioned defects of the prior art, embodiments of the present invention provide an intelligent calibration method and system for a semiconductor detection equipment. By performing time serialization splitting on measurement data and constructing the corresponding relationship between the measurement time grid and the measurement area, the source of the error can be accurately located in terms of time and space. At the same time, combined with in-depth analysis of the measurement error trend, it is possible to identify whether the error is short-term fluctuation or long-term anomaly, so as to distinguish accidental errors from hardware failures. In the case where the error belongs to short-term fluctuation, the previous measurement data is traced back according to the measurement time grid, and the interval with stable environment is screened to determine the earliest occurrence position of the error and its corresponding measurement area, and local calibration is performed on this area, avoiding the redundant operation of full-process repeated execution in the traditional method, improving the measurement efficiency, reducing the equipment downtime, and enhancing the accuracy of measurement error correction, so that the semiconductor detection equipment can still maintain high measurement stability and data reliability under high-precision requirements, in order to solve the problems raised in the above background art.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] An intelligent calibration method for a semiconductor detection equipment, comprising the steps:

[0008] S1: At the beginning of calibration, preprocess the measurement data and environmental data within fixed sampling points and time intervals, and split and establish samples according to the measurement time grid;

[0009] S2: Extract the deviation degree between the current measurement error and historical measurement data, quantify its stability in combination with the influence of environmental conditions, calculate the error fluctuation range within the measurement time grid at the same time, and adjust the calculated deviation amplitude in combination with time factors;

[0010] S3: Extract the overall error trend, compare it with the historical stable state, form a comprehensive deviation degree, match it with the set reference limit, and judge whether it belongs to a hardware fault;

[0011] S4: If it is determined that the error belongs to a non-hardware fault, trace back to the previous measurement data, align the historical error data according to the measurement time grid, check the degree of environmental change within each time grid, screen out the stable interval, and determine the measurement time grid and corresponding measurement area where the error first appears;

[0012] S5: According to the determined measurement time grid and corresponding measurement area, re-execute the local calibration of the corresponding measurement area according to the preset rules.

[0013] In a preferred embodiment, step S1 includes the following content:

[0014] During calibration, the data of all sensors need to be stored at the set time interval and corresponding to specific measurement time grids to form a time series data set; after the data acquisition is completed, perform denoising processing and normalization processing on the collected data. The processed data is split according to the time grid, and each measurement time grid contains the measurement data and environmental data within the corresponding time period, and a complete time series data set is constructed according to the measurement order to form a standardized data structure.

[0015] In a preferred embodiment, step S2 includes the following content:

[0016] Extract the deviation degree between the current measurement error and historical measurement data, quantify its stability in combination with the influence of environmental conditions, and obtain the time-varying entropy dispersion index through measurement. The processing steps are as follows:

[0017] Set the current measurement error within the measurement time grid as , the historical measurement error as , where represents the time grid index, represents the sampling point index within the time grid, represents the measurement error after normalization processing; the error deviation is calculated using the following formula: ; where, is the time grid Environment impact factor after internal normalization; time-varying entropy dispersion index It is defined as follows: ; where is the number of data points within the time grid , is the central moment of the time grid , is the measurement reference moment is the time normalization parameter

[0018] In a preferred embodiment, step S2 further includes the following:

[0019] Calculate the error fluctuation range within the measurement time grid, adjust the calculated deviation amplitude in combination with time factors, and finally obtain the time-domain ripple jump index. The processing steps are as follows:

[0020] Within each measurement time grid , perform an exponential scale transformation on the error change between adjacent sampling points, and set the normalized measurement error to , then the error jump value is defined as: ; where is the sensitivity adjustment parameter, and the time-domain ripple jump index is calculated as follows: ; where is the duration of the time grid , is the time balance constant

[0021] In a preferred embodiment, step S3 includes the following:

[0022] Set the measurement time grid index to , and perform a double-difference transformation on the time-varying entropy dispersion index and the time-domain ripple jump index within the time grid; then, perform a time-series integration on the transformation results of the time-varying entropy dispersion index and the time-domain ripple jump index to obtain the overall error trend; obtain the overall error trend After that, compare it with the historical stable state to determine the overall deviation degree of the measurement error; the historical stable state is fitted using the means of the time-varying entropy dispersion index and the time-domain ripple jump index of the measurement device under stable operating conditions, and the historical stable error feature is defined as: ; where is the total number of measurement time grids of historical stable data represents the measurement time grid index used to calculate the historical stable error feature, and its value range is from 1 to ; Subsequently, calculate the comprehensive deviation degree of the current overall error trend compared with the historical stable error: ; where To adjust the parameters of the deviation mapping.

[0023] In a preferred embodiment, step S3 further includes the following:

[0024] After calculating the comprehensive deviation it is matched with a set reference limit to determine whether the current error belongs to short-term fluctuations or long-term anomalies, and to determine whether it belongs to a hardware fault. The set reference limit includes the upper limit of short-term fluctuations and the long-term anomaly threshold and the judgment is made according to the following rules:

[0025] If then it is determined that the error is within the normal range and no additional adjustment is required;

[0026] If then it is determined that the error belongs to short-term fluctuations and is not a hardware fault;

[0027] If then it is determined that the measurement error has deviated from the historical stable state, exceeded the normal measurement working condition range, and there is a hardware fault.

[0028] In a preferred embodiment, step S4 includes the following:

[0029] If it is determined that there is no hardware fault, first call the data records stored during the previous complete measurement process, including the normalized measurement error data and environmental data, and align them one by one according to the current measurement time grid structure; set the current measurement time grid index as and the historical measurement time grid index as then perform time grid matching on the time axis. After the time grid data alignment is completed, analyze the degree of change of the environmental data within the measurement time grid; use weighted calculation to calculate the normalized offset of the current measurement time grid from the corresponding historical measurement time grid in terms of environmental parameters; if the normalized offset value is less than the environmental change threshold, it is determined that the corresponding time grid is in the environmental stable interval, otherwise the corresponding time grid is excluded.

[0030] In a preferred embodiment, step S4 further includes the following:

[0031] After screening out the environmental stable intervals, based on the difference between the current measurement error and the historical error data, determine the time grid where the error first appears and the corresponding measurement area;

[0032] Then, search all the time grids that meet the environmental stability conditions in chronological order to find the first time grid where the change rate of the measurement error exceeds the error mutation threshold Determine the corresponding measurement area as the error starting time cell;

[0033] To locate the error starting time cell For the corresponding measurement area, determine the physical area covered by the measuring device within the corresponding time cell according to the correspondence between the time cell and the spatial scanning path; set the measurement area index as , then the measurement area corresponding to the error starting time cell is: ; where refers to the mapping relationship between the measurement time cell and the measurement area;

[0034] Finally, locate the measurement time cell where the error first appears and its corresponding measurement area .

[0035] An intelligent calibration system for a semiconductor detection device includes: a data sequencing module, a deviation analysis module, a trend induction module, an error tracing module, and a local calibration module;

[0036] Data sequencing module: At the start of calibration, preprocess the measurement data and environmental data within a fixed sampling point and time interval, split and establish samples according to the measurement time cell, and transfer the preprocessed data to the deviation analysis module;

[0037] Deviation analysis module: Extract the deviation degree between the current measurement error and the historical measurement data, quantify its stability in combination with the influence of environmental conditions, calculate the error fluctuation range within the measurement time cell, and adjust the calculation according to the time factor to obtain the deviation amplitude, and transfer the calculation result to the trend induction module;

[0038] Trend induction module: Extract the overall error trend, compare it with the historical stable state to form a comprehensive deviation degree, match it with the set reference limit, and judge whether it belongs to a hardware fault, and transfer the judgment result to the error tracing module;

[0039] Error tracing module: If it is determined that the error belongs to a non-hardware fault, trace back to the previous measurement data, align the historical error data according to the measurement time cell, check the degree of environmental change within each time cell, screen out the stable interval, determine the measurement time cell where the error first appears and the corresponding measurement area, and transfer the determined result to the local calibration module;

[0040] Local calibration module: According to the determined measurement time cell and the corresponding measurement area, re-execute the local calibration of the corresponding measurement area according to the preset rules.

[0041] The technical effects and advantages of the intelligent calibration method and system for a semiconductor detection device of the present invention:

[0042] The present invention splits the measurement data into time series, and constructs the correspondence between the measurement time grid and the measurement area, so that the sources of errors can be accurately located in terms of time and space. At the same time, by combining in-depth analysis of the measurement error trend, it is identified whether the error is short-term fluctuation or long-term anomaly, so as to distinguish accidental errors from hardware failures. In the case where the error is a short-term fluctuation, the previous measurement data is traced back according to the measurement time grid, and the interval with stable environment is screened to determine the earliest occurrence position of the error and its corresponding measurement area. Local calibration is performed for this area, avoiding the redundant operations of repeated execution in the whole process in the traditional method, improving the measurement efficiency, reducing the equipment downtime, and enhancing the accuracy of measurement error correction, so that the semiconductor detection equipment can still maintain high measurement stability and data reliability under high-precision requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 It is a schematic flow chart of an intelligent calibration method for a semiconductor detection device of the present invention;

[0044] Figure 2 It is a schematic structural diagram of an intelligent calibration system for a semiconductor detection device of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0045] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0046] Embodiment 1: Figure 1 An intelligent calibration method for a semiconductor detection device of the present invention is given, including:

[0047] S1: At the start of calibration, the measurement data and environmental data at fixed sampling points and within a time interval are preprocessed, and samples are established by splitting according to the measurement time grid.

[0048] S2: Extract the deviation degree between the current measurement error and the historical measurement data, and quantify its stability in combination with the influence of environmental conditions. At the same time, calculate the error fluctuation range within the measurement time grid, and adjust the calculated deviation amplitude in combination with time factors.

[0049] S3: Extract the overall error trend, compare it with the historical stable state to form a comprehensive deviation degree, and match it with the set reference limit to determine whether it belongs to a hardware failure.

[0050] S4: If it is determined that the error is not a hardware fault, trace back the last measurement data, align the historical error data according to the measurement time grid, check the degree of environmental change in each time grid, screen out the stable interval, and determine the measurement time grid where the error first appeared and the corresponding measurement area.

[0051] S5: According to the determined measurement time grid and the corresponding measurement area, re-perform local calibration of the corresponding measurement area according to a preset rule.

[0052] Step S1 includes the following contents:

[0053] When starting calibration, first determine the fixed measurement sampling points and their corresponding time intervals, and establish a data collection mechanism to ensure that the data in each measurement time grid can be fully recorded. The measurement data includes core measurement information such as height data, reflectivity, and interference image signals obtained by the optical profiler. At the same time, environmental data is collected synchronously, including temperature, humidity, vibration status, light source power, and other external factors that affect measurement stability. The data collection process must ensure time synchronization, that is, the data of all sensors must be stored at the set time interval and correspond to the specific measurement time grid to form a time series data set.

[0054] After the data acquisition is completed, the collected data is denoised. First, for the measured data, the neighborhood difference is used to screen outliers and remove abnormal signals that significantly deviate from their adjacent measured values ​​to reduce the impact of transient fluctuations in the measuring equipment. Secondly, the environmental data is smoothed by using sliding window averaging or median filtering to remove single sampling errors so that it can better reflect the overall trend rather than being affected by short-term interference. For signals with random noise interference, such as interferometric image signals or reflectivity data, high-frequency noise components are filtered out through signal frequency domain analysis to retain key features.

[0055] After denoising, the measured data and environmental data are normalized so that the values ​​of different dimensions are within the same scale range for subsequent analysis. Normalization uses interval mapping to map the measured data and environmental data to the preset standard range respectively, while retaining the relative change trend of the original data to ensure that the normalization process does not affect the essential information of the data. The normalized data needs to be split according to the time grid. Each measurement time grid contains the measured data and environmental data in the corresponding time period, and a complete time series data set is constructed according to the measurement order to form a standardized data structure that can be used for subsequent error analysis.

[0056] A measurement time grid refers to the measurement data unit divided at fixed time intervals during the calibration or measurement process. Since the measurement process of an optical profiler is continuous, in order to facilitate the analysis of the stability and change trend of measurement data, it is necessary to divide the measurement data in the time dimension. The length of the measurement time grid is jointly determined by the sampling frequency, data processing capacity, and measurement requirements, and is usually set to range from several milliseconds to several seconds. Within each measurement time grid, all measurement data and environmental data during that time period are included, including information such as height, reflectivity, interference images, temperature and humidity, vibration, etc. The role of the measurement time grid is to provide a data structure in the time dimension, enabling data analysis to not only focus on the state of the measurement points themselves, but also observe the change trend of errors by combining historical data.

[0057] The measurement area refers to the physical space area covered by the optical profiler during the measurement process when scanning the surface of the workpiece. The size and shape of the measurement area depend on the optical system of the measurement device, the scanning mode, and the geometric characteristics of the measurement object. During actual measurement, the device will collect data at different physical positions according to the preset scanning path, so as to obtain the complete contour information of the workpiece surface. The measurement area can be further divided into multiple local measurement units, each unit corresponding to a certain spatial coordinate range and recording the measurement data within that range. The role of the measurement area is to provide a data structure in the spatial dimension, enabling the system to track the distribution of errors in the physical position.

[0058] The measurement time grid and the measurement area are two interrelated dimensions. The former describes the time distribution characteristics of the data, and the latter describes the spatial distribution characteristics of the data. Since during the scanning process of the optical profiler, the measurement head will sequentially collect data from different areas according to the established path, the measurement data corresponding to each measurement time grid is not limited to a single measurement area, but is related to the scanning speed and path of the device. During the data analysis process, the relationship between the measurement time grid and the measurement area can be established through time - space mapping, that is: within a certain measurement time grid, the system can determine the physical area where the measurement head of the device is located, and combine the measurement data within that area for error analysis. This time - space correlation method enables errors to be traced not only in the time dimension, but also accurately located to a specific area in the physical space, so that when the error exceeds the tolerance, it can be traced back to the corresponding measurement area, and local calibration can be performed for that area without having to perform a full - process calibration.

[0059] Through the above processing flow, a time - series sample containing complete measurement data and environmental data is finally obtained, and the data quality is guaranteed, enabling subsequent error analysis to be carried out on the basis of noise - free and normalized data, and being able to trace the error source and change trend according to the measurement time grid structure.

[0060] Step S2 includes the following content:

[0061] Extract the deviation degree between the current measurement error and historical measurement data, and quantify its stability in combination with the influence of environmental conditions. The time-varying entropy dispersion index is obtained through measurement, which is used to quantify the deviation degree of the current measurement error compared with historical measurement data, and correct the error stability in combination with environmental conditions. The calculation is based on the normalized data split according to the measurement time grid. Each time grid contains measurement data and environmental data, and a complete time series data set is constructed. The processing steps are as follows:

[0062] Since the change trend of the error may be affected by environmental parameters, before calculating the error deviation degree, it is necessary to perform normalization adjustment on the current measurement error and historical error. Set the current measurement error within the measurement time grid as , and the historical measurement error as , where represents the time grid index, represents the sampling point index within the time grid, represents the measurement error after normalization processing. The error deviation is calculated using the following formula: ; where is the normalized environmental impact factor within the time grid . This factor is calculated by normalizing environmental parameters such as temperature, humidity, and vibration, aiming to compensate for the influence of the environment on the measurement error in the error calculation.

[0063] After calculating the error deviation , it is necessary to introduce a time influence factor to reflect the dispersion characteristics of the error in the time dimension. The time-varying entropy dispersion index is defined as follows: ; where is the number of data points within the time grid , is the central moment of the time grid , is the measurement reference moment, is the time normalization parameter. This index introduces the time factor into the error deviation calculation through non-linear logarithmic modulation to capture the long-term trend of error changes.

[0064] Calculate the error fluctuation range within the measurement time grid, adjust the calculated deviation amplitude in combination with the time factor, and finally obtain the time-domain ripple jump index, which is used to characterize the fluctuation amplitude of the error within the measurement time grid and adjust the fluctuation cumulative effect in combination with the time scale. The calculation is based on the processing results of the first step, that is, the measurement data and environmental data after time grid division, and is processed twice relying on the calculation results of the time-varying entropy dispersion index. The processing steps are as follows:

[0065] In each measurement time grid Within, perform an exponential scale transformation on the error change between adjacent sampling points to amplify the influence of minor fluctuations. Set the normalized measurement error as , then the error jump value is defined as: ; where is the sensitivity adjustment parameter, controlling the response degree of the error jump calculation to small fluctuations. This calculation method can ensure a low response when the error fluctuates within a small range, while exponentially amplifying its influence when the error mutates.

[0066] After calculating the error jump, it is necessary to perform a time scale adjustment on the overall fluctuation situation of the measurement time grid to consider the influence of different time grid lengths on error accumulation. The time domain ripple jump index is calculated as follows: ; where is the duration of the time grid , is the time balance constant. Through the square root time modulation function, this index can automatically adjust the error accumulation effect of different time grids, ensuring that the error of a long time grid will not be overly amplified due to the increase in the number of samples, and at the same time ensuring the stability of error fluctuations within a short time grid.

[0067] Step S3 includes the following content:

[0068] Within the measurement time grid, first obtain the temporal variation characteristics of the error based on the time-varying entropy dispersion index and the time domain ripple jump index, and use the time series transformation method to extract the overall error trend of the measurement time grid. Set the measurement time grid index as , and the values of the time-varying entropy dispersion index and the time domain ripple jump index calculated within the time grid are respectively and , then within consecutive measurement time grids, construct an error trend function according to the non-linear transformation method, and extract the overall error trend based on this.

[0069] First, perform a double difference calculation on the time-varying entropy dispersion index and the time domain ripple jump index in the time series to eliminate the linear growth term of the measurement error and highlight the non-linear change trend within a short time:

[0070] ;

[0071] ;

[0072] The purpose of this double difference calculation is to extract the acceleration characteristics of the error change in the time grid sequence, ensuring a low trend amplitude when the error changes slowly, while being able to amplify the change signal when the error mutates. After that, perform a temporal integration on the transformation results of the time-varying entropy dispersion index and the time domain ripple jump index to obtain the overall error trend: ; where, represents the base of the natural logarithm, is the time decay factor, which weakens the influence of errors in more distant time grids, while the influence of recent error fluctuations is stronger. This calculation method can ensure that the system gives priority to the error changes in the most recent time grid, while taking into account the error trend evolution in a longer time range.

[0073] After obtaining the overall error trend , it is necessary to compare it with the historical stable state to determine the overall deviation degree of the measurement error. The historical stable state is fitted by the mean of the time-varying entropy dispersion index and the time-domain ripple jump index of the measuring device under stable working conditions, and the historical stable error feature is defined as: ; where, is the total number of measurement time grids of the historical stable data, represents the measurement time grid index used to calculate the historical stable error feature, and its value range is from 1 to . Subsequently, calculate the comprehensive deviation degree of the current overall error trend compared with the historical stable error: ; where, is the parameter for adjusting the deviation mapping, which is used to control the non-linear amplification effect of the error deviation. This calculation method ensures that when the overall error trend approaches the historical stable state, the comprehensive deviation degree is small, while when the error trend is far from the historical stable state, the deviation degree increases rapidly with the non-linear mapping function, enhancing the sensitivity to abnormal changes.

[0074] After calculating the comprehensive deviation degree , it is necessary to match it with the set reference limit to judge whether the current error belongs to short-term fluctuation or long-term anomaly, and further judge whether it belongs to a hardware fault. The set reference limit includes the short-term fluctuation upper limit and the long-term anomaly threshold , and the judgment is made according to the following rules:

[0075] If , it is determined that the error belongs to the normal range and no additional adjustment is required.

[0076] If , it is determined that the error belongs to short-term fluctuation, not a hardware fault, the measurement error is within the fluctuation range, and the error change does not form a continuous trend, usually caused by environmental fluctuations, instantaneous interference or single measurement deviation. The system will trace back to the previous measurement data, screen the stable interval according to the environmental data within the measurement time grid, and further analyze the measurement time grid where the error first appears and its corresponding measurement area to provide a precise adjustment range for subsequent local calibration.

[0077] If , it is determined that the measurement error has deviated far from the historical stable state, exceeding the normal measurement operating range, indicating a hardware failure. The system triggers a fault warning and executes a complete calibration process.

[0078] Step S4 includes the following:

[0079] If it is determined that there is no hardware failure, first call the data records stored during the previous complete measurement process, including the normalized measurement error data and environmental data, and align them one by one according to the current measurement time grid structure. Set the current measurement time grid index as , and the historical measurement time grid index as . Then perform time grid matching on the time axis so that for any measurement data, its corresponding time grid in the previous measurement data can be found . During the data alignment process, ensure that the measurement error data and environmental data are synchronized at the same time interval, so that the measurement error trend and environmental change situation can be accurately corresponding on the time axis to avoid data deviation caused by time drift.

[0080] After the time grid data alignment is completed, analyze the degree of change of the environmental data within the measurement time grid to screen out the stable interval where the error can be traced. Set the environmental data to include temperature , humidity , vibration and other factors. Use weighted calculation to calculate the normalized offset of the current measurement time grid and the corresponding historical measurement time grid in environmental parameters: ; where , , are the standard normalization factors of their respective environmental parameters, enabling different physical quantities to be compared on the same scale, , , are the preset proportionality coefficients of their respective environmental parameters. If the normalized offset value is less than the environmental change threshold, it is determined that the corresponding time grid is in the environmental stable interval; otherwise, the corresponding time grid is excluded and not involved in the error backtracking analysis to ensure that the subsequent analysis is based on the time grid with environmental stability for calculation.

[0081] After screening out the environmental stable interval, based on the difference between the current measurement error and the historical error data, determine the time grid where the error first appears and the corresponding measurement area. Set the measurement error data as and the historical error data as , and calculate the change rate of the measurement error: ; Then, search all time cells that meet the environmental stability condition in chronological order to find the first time cell where the change rate of the measurement error exceeds the error mutation threshold. . This time cell is used as the error starting time cell to further determine the corresponding measurement area.

[0082] To locate the measurement area corresponding to the error starting time cell , based on the correspondence between the time cell and the spatial scanning path, determine the physical area covered by the measuring device within this time cell. Set the measurement area index as , then the measurement area corresponding to the error starting time cell is: ; where refers to the mapping relationship between the measurement time cell and the measurement area, that is, within a certain time cell, the physical area covered by the scanning path of the measuring device. Since the optical profiler moves along a predefined trajectory during scanning, the scanning position corresponding to each measurement time cell can be deduced from the device's motion trajectory, scanning speed, and the duration of the time cell. For example, if the optical profiler scans the surface of a wafer at a fixed rate and each measurement time cell lasts for 50 milliseconds, then during the time cell , the measurement head moves from position to position , then the measurement area of this time cell is the spatial area from to . If the error first appears in the time cell , then can indicate the specific physical location where the error appears, such as a local area on the wafer or the edge position of the chip package. Through this mapping, the measurement system can accurately trace back the source of the error and perform local calibration in this area without affecting the entire measurement process.

[0083] Finally, through the above data alignment, environmental stability screening, and error starting point identification, accurately locate the measurement time cell where the error first appears and its corresponding measurement area , providing a basis for subsequent local calibration and avoiding unnecessary full-process recalibration.

[0084] Step S5 includes the following content:

[0085] After determining the measurement time cell where the error starts and its corresponding measurement area, first enter the local calibration mode. At this time, all data that exactly corresponds to this measurement time cell will be extracted from the complete measurement dataset, and based on Determine the physical area scanned by the device during this time period. Ensure that during the data extraction process, the time grid is consistent with the information of the measurement area to avoid subsequent processing errors caused by data confusion.

[0086] Subsequently, according to the preset local calibration rules, perform local calibration operations on the extracted measurement area. The specific steps include: First, analyze the differences between the historical measurement data and the current data in this area to determine the calibration parameters that need to be adjusted; then call the pre-configured calibration program to finely tune and correct the optical elements, scanning speed, exposure parameters, etc. in this area; during the adjustment process, continuously monitor the measurement data and environmental data to ensure that the changes in various parameters during the local calibration process are within the expected range.

[0087] After completing the local calibration operation, verify the re-calibrated measurement area, that is, compare the new data after calibration with the preset reference standard to confirm whether the error has returned to the normal fluctuation range. If the verification result meets the preset standard, update the calibration parameters of this area; otherwise, the system will perform the local calibration operation again according to the preset rules until it reaches a qualified state. This whole set of processes ensures fine and targeted local calibration without affecting the overall detection work.

[0088] Embodiment 2: Figure 2 An intelligent calibration system for a semiconductor detection device according to the present invention is given, including:

[0089] A data sequencing module, a deviation analysis module, a trend induction module, an error tracing module, and a local calibration module;

[0090] Data sequencing module: At the beginning of calibration, preprocess the measurement data and environmental data at fixed sampling points and time intervals, and split and establish samples according to the measurement time grid, and transfer the preprocessed data to the deviation analysis module;

[0091] Deviation analysis module: Extract the deviation degree between the current measurement error and the historical measurement data, and quantify its stability in combination with the influence of environmental conditions. At the same time, calculate the error fluctuation range within the measurement time grid, and adjust the calculated deviation amplitude in combination with time factors, and transfer the calculation result to the trend induction module;

[0092] Trend induction module: Extract the overall error trend, compare it with the historical stable state to form a comprehensive deviation degree, and match it with the set reference limit to determine whether it belongs to a hardware failure, and transfer the judgment result to the error tracing module;

[0093] Error tracing module: If it is determined that the error is not a hardware failure, trace back the previous measurement data, align the historical error data according to the measurement time grid, check the degree of environmental change within each time grid, filter out the stable intervals, determine the measurement time grid where the error first appears and the corresponding measurement area, and transfer the determined results to the local calibration module;

[0094] Local calibration module: According to the determined measurement time grid and the corresponding measurement area, re - execute the local calibration of the corresponding measurement area according to the preset rules.

[0095] The above formulas are all dimensionless and take their numerical values for calculation. The formulas are obtained by collecting a large amount of data for software simulation to get a formula closest to the actual situation. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0096] Only some exemplary embodiments of the present invention have been described by way of illustration above. Undoubtedly, for those of ordinary skill in the art, without departing from the spirit and scope of the present invention, the described embodiments can be modified in various different ways. Therefore, the above - mentioned drawings and description are illustrative in nature and should not be construed as limiting the protection scope of the claims of the present invention.

[0097] It should be noted that in this text, if there are relational terms such as first and second, they are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non - exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not explicitly listed, or further includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the existence of another identical element in the process, method, article or device comprising the element.

[0098] The above - mentioned is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application can easily think of changes or substitutions, which should all be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. An intelligent calibration method for semiconductor testing equipment, characterized in that: Includes steps: S1: When starting calibration, pre-process the measurement data and environmental data within fixed sampling points and time intervals, and split them into samples according to the measurement time grid; S2: Extract the degree of deviation between the current measurement error and the historical measurement data, and quantify its stability in combination with the influence of environmental conditions. At the same time, calculate the error fluctuation range within the measurement time grid, and adjust the calculation in combination with the time factor to obtain the deviation amplitude; S3: Extract the overall error trend and compare it with the historical stable state to form a comprehensive deviation, and match it with the set reference limit to determine whether it is a hardware failure; S4: If it is determined that the error is not a hardware fault, trace back the last measurement data, align the historical error data according to the measurement time grid, check the degree of environmental change in each time grid, screen out the stable interval, and determine the measurement time grid where the error first appeared and the corresponding measurement area; S5: According to the determined measurement time grid and the corresponding measurement area, re-perform local calibration of the corresponding measurement area according to a preset rule.

2. The intelligent calibration method for semiconductor testing equipment according to claim 1, characterized in that: Step S1 includes the following contents: During calibration, the data of all sensors must be stored at set time intervals and correspond to specific measurement time grids to form a time series data set. After data acquisition is completed, the collected data is denoised and normalized, and the processed data is split according to the time grid. Each measurement time grid contains the measurement data and environmental data within the corresponding time period, and a complete time series data set is constructed according to the measurement order to form a standardized data structure.

3. The intelligent calibration method for semiconductor testing equipment according to claim 2, characterized in that: Step S2 includes the following contents: The degree of deviation between the current measurement error and the historical measurement data is extracted, and its stability is quantified in combination with the influence of environmental conditions. The time-varying entropy diffusion index is obtained through measurement. The processing steps are as follows: Set the current measurement error within the measurement time grid to , the historical measurement error is ,in represents the time grid index, Represents the sampling point index within the time grid, represents the normalized measurement error; The error deviation is calculated using the following formula: ;in, For time grid Internally normalized environmental impact factor; time-varying entropy diffusion index The definition is as follows: ;in, For time grid The number of data points within For time grid The central moment of To measure the base time, is the time normalization parameter.

4. The intelligent calibration method for semiconductor testing equipment according to claim 3, characterized in that: Step S2 also includes the following contents: Calculate the error fluctuation range within the measurement time grid, adjust the calculation based on the time factor to get the deviation amplitude, and finally get the time domain ripple jump index. The processing steps are as follows: At each measurement time Within, the error change between adjacent sampling points is transformed into an exponential scale, and the normalized measurement error is set to , then the error jump value is defined as: ;in, For sensitivity adjustment parameters, the time domain ripple jump index is calculated as follows: ;in, For time grid duration, is the time equilibrium constant.

5. The intelligent calibration method for semiconductor testing equipment according to claim 4, characterized in that: Step S3 includes the following contents: Set the measurement time grid index to ,exist Double difference transformation is performed on the time-varying entropy diffusion index and the time-domain ripple jump index in the time grid; Afterwards, the transformation results of the time-varying entropy diffusion index and the time-domain ripple jump index are time-series integrated to obtain the overall error trend; Get the overall error trend After that, it is compared with the historical stable state to determine the overall deviation of the measurement error; the historical stable state is fitted by the time-varying entropy diffusion index and the mean of the time-domain ripple jump index of the measuring equipment under stable working conditions, and the historical stable error characteristics are defined as: ;in, is the total number of measurement time grids of historical stable data, Indicates the measurement time grid index used to calculate the historical stable error characteristics, ranging from 1 to ; Then calculate the comprehensive deviation of the current overall error trend compared to the historical stable error: ;in, Parameters for adjusting the deviation mapping.

6. The intelligent calibration method for semiconductor testing equipment according to claim 5, characterized in that: Step S3 also includes the following: In calculating the comprehensive deviation After that, it is matched with the set reference limit to determine whether the current error is a short-term fluctuation or a long-term abnormality, and whether it is a hardware failure. The reference limit is set to include the upper limit of short-term fluctuation. and long-term abnormal threshold , judged according to the following rules: like , then the error is judged to be within the normal range and no additional adjustment is required; like , then the error is judged to be a short-term fluctuation, not a hardware failure; like , it is determined that the measurement error has deviated from the historical stable state and exceeded the normal measurement condition range, indicating that there is a hardware failure.

7. The intelligent calibration method for semiconductor testing equipment according to claim 6, characterized in that: Step S4 includes the following contents: If it is determined to be a non-hardware fault, first call the data records stored in the last complete measurement process, including the normalized measurement error data and environmental data, and align them one-to-one according to the current measurement time grid structure; set the current measurement time grid index to , the historical measurement time grid index is , then the time grid is matched on the time axis. After the time grid data is aligned, the degree of change of the environmental data in the measured time grid is analyzed; the weighted calculation of the current measured time grid is used The corresponding historical measurement time grid Normalized offset on environmental parameters; if the normalized offset value is less than the environmental change threshold, the corresponding time grid is determined to be in the environmental stability interval, otherwise the corresponding time grid is removed.

8. The intelligent calibration method for semiconductor testing equipment according to claim 7, characterized in that: Step S4 also includes the following contents: After selecting the environmental stability interval, the time grid and corresponding measurement area where the error first occurred are determined based on the difference between the current measurement error and the historical error data; Then, all time grids that meet the environmental stability condition are searched in chronological order to find the first time grid that meets the measurement error change rate exceeding the error mutation threshold. , this time frame As the error starting time grid, determine the corresponding measurement area; In order to locate the error start time grid The corresponding measurement area determines the physical area covered by the measurement device in the corresponding time grid according to the correspondence between the time grid and the spatial scanning path; the measurement area index is set to , then the measurement area corresponding to the error start time grid is: ;in, It refers to the mapping relationship between the measurement time grid and the measurement area; Finally, the measurement time grid where the error first occurred is located. and its corresponding measurement area .

9. An intelligent calibration system for semiconductor testing equipment, used to implement an intelligent calibration method for semiconductor testing equipment according to any one of claims 1 to 8, characterized in that: include: Data sorting module, deviation analysis module, trend induction module, error tracing module and local calibration module; Data sequencing module: When calibration starts, it pre-processes the measurement data and environmental data within fixed sampling points and time intervals, splits them into samples according to the measurement time grid, and passes the pre-processed data to the deviation analysis module; Deviation analysis module: extracts the degree of deviation between the current measurement error and the historical measurement data, and quantifies its stability in combination with the influence of environmental conditions. It also calculates the error fluctuation range within the measurement time grid, adjusts the calculation in combination with the time factor to obtain the deviation amplitude, and passes the calculation result to the trend induction module; Trend Summarization Module: Extracts the overall error trend and compares it with the historical stable state to form a comprehensive deviation, which is matched with the set reference limit to determine whether it is a hardware failure and pass the judgment result to the error tracing module; Error tracing module: If the error is determined to be non-hardware fault, trace back the last measurement data, align the historical error data according to the measurement time grid, check the degree of environmental change in each time grid, screen out the stable interval, determine the measurement time grid where the error first appeared and the corresponding measurement area, and pass the determined result to the local calibration module; Local calibration module: according to the determined measurement time grid and the corresponding measurement area, re-execute the local calibration of the corresponding measurement area according to the preset rules.

Images