Data feature extraction method, device and equipment of wafer, medium and product

Abstract

The application provides a wafer data feature extraction method, device, equipment, medium and product, the method comprises the following steps: obtaining sensor data of a sub-process in a process to be analyzed; wherein the sensor data of the sub-process comprises sensor data of each normal wafer and abnormal wafer under the sub-process; determining the data type of the sensor data of the sub-process; the data type comprises a linear type, a partial linear type and a nonlinear type; according to the data type, the sensor data of the sub-process is subjected to feature extraction to obtain feature data of the sub-process; according to the feature data of the sub-process, the feature data of the process to be analyzed is obtained, and the feature data of the process to be analyzed comprises the feature data of the sub-process in the process to be analyzed; the scheme improves the accuracy of wafer feature data extraction.

Description

Technical Field

[0001] This application relates to the field of semiconductor technology, and in particular to a method, apparatus, device, medium and product for extracting data features from a wafer. Background Technology

[0002] In semiconductor manufacturing, fault data collection systems are crucial monitoring and analysis tools used to collect, store, and analyze fault or defective product data during the production process. Production data in semiconductor manufacturing includes sensor data, equipment parameters, production batch information, test results, and process parameters. To accurately analyze the root causes of faults or decreased product yield, precise feature extraction from the sensor data of the process is necessary.

[0003] Currently, some feature data extraction schemes are based on the statistical values ​​of signals returned by sensors to extract features; however, using this scheme will result in the loss of signal detail and inaccurate feature data; using inaccurate feature data for fault analysis will lead to inaccurate analysis results. Summary of the Invention

[0004] The wafer data feature extraction method, apparatus, device, medium, and product provided in this application are used to improve the accuracy of wafer feature data extraction.

[0005] In a first aspect, embodiments of this application provide a wafer data feature extraction method, comprising: acquiring sensor data of a sub-process in a process to be analyzed; wherein the sensor data of the sub-process includes sensor data of each normal wafer and abnormal wafer in the sub-process; determining the data type of the sensor data of the sub-process; the data type includes linear type, partially linear type and nonlinear type; performing feature extraction on the sensor data of the sub-process according to the data type to obtain feature data of the sub-process; and obtaining feature data of the process to be analyzed according to the feature data of the sub-process, wherein the feature data of the process to be analyzed includes the feature data of the sub-process in the process to be analyzed.

[0006] In one possible implementation, determining the data type of the sensor data for the sub-process includes: performing linear fitting based on the sensor data of the sub-process to obtain the linear fit degree; the linear fit degree characterizes the deviation between the fitted data obtained by linear fitting and the sensor data of the sub-process; if the linear fit degree is not greater than a first threshold, the data type is determined to be linear; if the linear fit degree is greater than the first threshold, the data type of the sensor data for the sub-process is determined based on the longest near-zero second-order difference segment of the fitted curve obtained by linear fitting; the near-zero second-order difference segment characterizes a continuous segment where the absolute value of the second-order difference of the sensor data is less than a preset value.

[0007] In one possible implementation, the data type of the sensor data of the sub-process is determined based on the longest near-zero second-order difference segment of the fitted curve obtained by linear fitting, including: if the length of the near-zero second-order difference segment is longer than a second threshold, the data type of the sensor data is determined to be partially linear; if the length of the near-zero second-order difference segment is not longer than the second threshold, the data type of the sensor data is determined to be nonlinear.

[0008] In one possible implementation, feature extraction is performed on the sensor data of the sub-process according to the data type to obtain the feature data of the sub-process, including: if the data type is linear, linear feature extraction is performed on the sensor data of the sub-process to obtain the feature data of the sub-process; if the data type is partially linear, linear and nonlinear feature extraction is performed on the sensor data of the sub-process to obtain the feature data of the sub-process; if the data type is nonlinear, nonlinear feature extraction is performed on the sensor data of the sub-process to obtain the feature data of the sub-process.

[0009] In one possible implementation, linear fitting is performed based on the sensor data of the sub-process to obtain the linear fit degree, including: establishing a first objective function based on the sensor data of the sub-process, performing linear fitting to obtain the linear fitting parameters of the first objective function; substituting the linear fitting parameters into the first objective function to obtain the fitting data; and obtaining the linear fit degree based on the fitting data and the sensor data.

[0010] In one possible implementation, linear feature extraction is performed on the sensor data of the sub-process to obtain the feature data of the sub-process, including: using the linear fitting parameters and the linear fitting degree as the feature data of the sub-process.

[0011] In one possible implementation, nonlinear feature extraction is performed on the sensor data to obtain feature data of the sub-process, including: establishing a second objective function based on the sensor data of the sub-process, performing nonlinear fitting to obtain nonlinear fitting parameters of the second objective function; substituting the nonlinear fitting parameters into the second objective function to obtain nonlinear fitting data; obtaining the nonlinear fitting degree based on the nonlinear fitting data and the sensor data of the sub-process; the nonlinear fitting degree characterizes the deviation between the nonlinear fitting data and the sensor data; and using the nonlinear fitting parameters and the nonlinear fitting degree as feature data of the sub-process.

[0012] In one possible implementation, the method further includes: calculating the second-order difference of the sensor data of the sub-process based on the sensor data of the sub-process; determining a continuous segment in which the absolute value of the second-order difference is less than a preset value; and selecting the longest continuous segment in the continuous segment as the near-zero second-order difference segment.

[0013] In one possible implementation, acquiring sensor data of a sub-process in the process to be analyzed includes: acquiring sensor data of the process to be analyzed; the sensor data includes tags of the corresponding sub-processes; and dividing the sensor data of the process to be analyzed according to the tags to obtain sensor data of the sub-processes in the process to be analyzed.

[0014] In one possible implementation, the method further includes: performing fault analysis processing based on the characteristic data of the process to be analyzed, and obtaining fault analysis results.

[0015] Secondly, embodiments of this application provide a wafer data feature extraction apparatus, comprising: an acquisition module for acquiring sensor data of a sub-process in a process to be analyzed; wherein the sensor data of the sub-process includes sensor data of each normal wafer and abnormal wafer in the sub-process; a determination module for determining the data type of the sensor data of the sub-process; the data type includes linear type, partially linear type and nonlinear type; an extraction module for performing feature extraction on the sensor data of the sub-process according to the data type to obtain feature data of the sub-process; and a generation module for obtaining feature data of the process to be analyzed according to the feature data of the sub-process, wherein the feature data of the process to be analyzed includes the feature data of the sub-process in the process to be analyzed.

[0016] Thirdly, embodiments of this application provide an electronic device, including: a memory and a processor; the memory stores computer-executable instructions; the processor executes the computer-executable instructions stored in the memory, causing the processor to perform the first aspect and / or various possible implementations of the first aspect as described above.

[0017] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the first aspect and / or various possible implementations of the first aspect.

[0018] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the first aspect and / or various possible implementations of the first aspect.

[0019] The method, apparatus, equipment, medium, and product for wafer data feature extraction provided in this application acquire sensor data from normal and abnormal wafers in a sub-process of the process to be analyzed, and determine the data type of the acquired sensor data. The data types include linear, partially linear, and nonlinear types. Then, feature extraction is performed on the sensor data of the sub-process according to the specific data type to obtain the feature data of the sub-process. Based on the feature data of the sub-process, the feature data of the process to be analyzed is obtained. Extracting feature data from sensor data according to different data types improves the comprehensiveness and accuracy of feature extraction; it effectively captures feature data of different linear types from the sensor data; and subsequent fault or anomaly analysis based on comprehensive and accurate feature data of the process to be analyzed can improve the accuracy of fault analysis results. Attached Figure Description

[0020] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0021] Figure 1 An exemplary flowchart of a method for extracting data features from a wafer is shown.

[0022] Figure 2 An exemplary flowchart of a method for extracting data features from a wafer is shown.

[0023] Figure 3 An exemplary flowchart of a method for extracting data features from a wafer is shown.

[0024] Figure 4 An exemplary flowchart of a method for extracting data features from a wafer is shown.

[0025] Figure 5 An exemplary flowchart of a method for extracting data features from a wafer is shown.

[0026] Figure 6 An exemplary flowchart of a method for extracting data features from a wafer is shown.

[0027] Figure 7 This is a schematic diagram illustrating the segmentation of a single wafer and a single sensor according to sub-processes, as exemplified in this application.

[0028] Figure 8 This is a schematic diagram of a data segment of linear type sensor feature data as an example of this application;

[0029] Figure 9 This is a schematic diagram of a data segment of a partial linear sensor feature data example from this application;

[0030] Figure 10 This is a schematic diagram of a data segment of nonlinear sensor characteristic data as an example of this application;

[0031] Figure 11 This is a flowchart illustrating the wafer feature data extraction method of this application;

[0032] Figure 12 An exemplary schematic diagram of a wafer data feature extraction device is shown;

[0033] Figure 13 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0034] The accompanying drawings have illustrated specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to specific embodiments. Detailed Implementation

[0035] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0036] It should be noted that the brief descriptions of terms in this application are only for the convenience of understanding the embodiments described below, and are not intended to limit the embodiments of this application. Unless otherwise stated, these terms should be understood in their ordinary and common meaning. The terms "comprising" and "having," and any variations thereof, in the specification, claims, and accompanying drawings of this application are intended to be omnipresent but not exclusive. For example, a product or device that comprises a series of components is not necessarily limited to those components that are explicitly listed, but may include other components that are not explicitly listed or that are inherent to such products or devices. The term "module" as used in this application refers to any known or subsequently developed hardware, software, firmware, artificial intelligence, fuzzy logic, or combination of hardware and / or software code capable of performing the functions associated with that element.

[0037] In semiconductor manufacturing, the FDC (Failure Data Collection) system is a crucial monitoring and analysis tool used to collect, store, and analyze data on failures or defective products during the production process. Production data in semiconductor manufacturing typically includes sensor data, equipment parameters, production batch information, test results, and process parameters. This data can be used for root cause analysis (RCA) to help engineers identify the root causes of failures or decreased product yield. To accurately analyze the root causes of failures or decreased product yield, comprehensive and accurate feature extraction of the sensor data from the process is required.

[0038] Currently, some feature data extraction schemes are based on the statistical values ​​of signals returned by various sensors in the process steps to extract feature values, such as average value, standard deviation, maximum value, minimum value, peak value, etc. However, using this scheme will lose the detailed performance of the signal, and the extracted feature data will be inaccurate. Using inaccurate feature data for fault analysis will lead to inaccurate analysis results.

[0039] The technical content provided in this application aims to solve the aforementioned technical problems in related technologies. The wafer feature data extraction method, apparatus, equipment, medium, and product provided in this application acquire sensor data from normal and abnormal wafers in a sub-process of the process to be analyzed, and determine the data type of the acquired sensor data. The data types include linear, partially linear, and nonlinear types. Then, feature extraction is performed on the sensor data of the sub-process according to the specific data type to obtain the feature data of the sub-process. Based on the feature data of the sub-process, the feature data of the process to be analyzed is obtained. Extracting feature data from sensor data according to different data types improves the comprehensiveness and accuracy of feature extraction; it effectively captures feature data of different linear types from sensor data; and subsequent fault or anomaly analysis based on comprehensive and accurate feature data of the process to be analyzed can improve the accuracy of fault analysis results.

[0040] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0041] Example 1

[0042] Figure 1 An exemplary flowchart of a wafer data feature extraction method is shown. The execution entity in this example can be a wafer data feature extraction device. Figure 1 As shown, the method includes:

[0043] Step 101: Obtain sensor data for the sub-process in the process to be analyzed; wherein, the sensor data for the sub-process includes sensor data for each normal wafer and abnormal wafer in the sub-process;

[0044] Step 102: Determine the data type of the sensor data for the sub-process; data types include linear, partially linear, and nonlinear types.

[0045] Step 103: Based on the data type, extract features from the sensor data of the sub-process to obtain the feature data of the sub-process;

[0046] Step 104: Based on the characteristic data of the sub-process, obtain the characteristic data of the process to be analyzed. The characteristic data of the process to be analyzed includes the characteristic data of the sub-processes in the process to be analyzed.

[0047] In practical applications, the execution entity of this method can be a matching device for extracting data features from wafers. There are various ways to implement it. For example, it can be implemented through a computer program, such as application software; or it can be implemented through a physical device that integrates or installs relevant computer programs, such as a chip.

[0048] Production data in the semiconductor manufacturing process is generally stored in the data center or database of the semiconductor factory in structured, unstructured, or database formats. When it is necessary to analyze the causes of faults or defects, it is necessary to first use libraries and tools in programming languages ​​to export the data of the process to be analyzed into a structured format that is easy to analyze or process.

[0049] As an example, production data is typically a dictionary with wafer numbers as keys, where the key value for each wafer is a two-dimensional array. Each two-dimensional array contains data about the wafer's production process, typically including: data index, time, product batch information (Lot ld), wafer number (Wafer ld), silicon substrate number (Substrate ld), step name, and various sensor readings (Sensor1 Value, Sensor2 Value, etc.).

[0050] Specifically, wafer manufacturing involves multiple process steps. Typically, when root cause analysis is needed to identify defective products or malfunctions, the specific process step with the problem can be determined based on the actual process parameters and sensor data from each step. For the identified problematic process step, further analysis of the specific defects and malfunctions within that process's sub-processes is then conducted. For example, wafer manufacturing processes include oxidation, photolithography, etching, and thin film deposition. The oxidation process includes multiple sub-processes such as heating, gas deposition, and cooling. Sensor data obtained from the database is categorized according to sub-process tags to obtain sub-process sensor data; each group of sensor data includes sensor data from multiple wafers within that sub-process.

[0051] Specifically, the sensor data for each sub-process within the process to be analyzed is first acquired. This sub-process sensor data includes sensor data from normal wafers and sensor data from abnormal wafers. Each sub-process's sensor data includes multiple different types of sensor data, such as temperature sensor data, humidity sensor data, and pressure sensor data. Anomaly detection is then performed on the acquired sensor data for each sub-process to eliminate abnormal data. Here, abnormal data refers to data that is significantly different from other data as detected by the dataset's algorithm. For example, anomaly detection algorithms could be the Isolation Forest algorithm, the percentile method, or the distribution fitting algorithm. This example does not limit the anomaly detection method.

[0052] Based on the sensor data acquired from each sub-process, the data type of the sensor data is determined. In this example, the data types include linear, partially linear, and nonlinear types. Specifically, the process to be analyzed includes multiple sub-process steps; each sub-process step is equipped with multiple sensors, such as temperature sensors, humidity sensors, and pressure sensors, to collect sensor data from each sub-process step, resulting in temperature sensor data, humidity sensor data, pressure sensor data, etc. The data type of each sensor data is determined, thus obtaining the data type of each sensor data point within the sub-process. Subsequently, feature data is extracted based on different data types to obtain the feature data of each sensor data point within each sub-process. For each sensor data point within a sub-process, the feature data of the sub-process extracted through feature extraction is used to obtain the feature data of the process to be analyzed. Subsequently, fault analysis and defect analysis are performed based on the feature data of the process to be analyzed. The feature data in this example includes fitting parameters obtained by fitting according to different linear types, and a numerical representation of the deviation between the fitted data and the sensor data; for example, the numerical representation of the deviation can be mean square error (MSE), root mean square error (RMSE), coefficient of determination (R2), mean absolute error (MAE), mean absolute percentage error (MAPE), etc.; these error values ​​are used as feature data in this example for subsequent defect analysis and fault analysis.

[0053] In this example, the data type of the sensor data is determined based on the sensor data of each sub-process of the process to be analyzed, and feature data is extracted from the sensor data for different data types. The feature data of the process to be analyzed is obtained based on the extracted feature data of the sub-process. The solution of this application extracts feature data of wafer sensors more comprehensively and accurately. The subsequent fault analysis based on the extracted feature data yields more accurate analysis results.

[0054] Furthermore, it is necessary to determine the data type based on the acquired sensor data; accordingly, as an example, Figure 2 An exemplary flowchart of a wafer data feature extraction method is shown; based on any example, the data type of sensor data for a sub-process is determined, including:

[0055] Step 201: Based on the sensor data of the sub-process, perform linear fitting to obtain the linear fit degree; the linear fit degree characterizes the deviation between the fitted data obtained by linear fitting and the sensor data of the sub-process.

[0056] Step 202: If the linear fit is not greater than the first threshold, then the data type is determined to be linear.

[0057] Step 203: If the linear fit is greater than the first threshold, then the data type of the sensor data of the sub-process is determined based on the longest near-zero second-order difference segment of the fitted curve obtained by linear fitting; the near-zero second-order difference segment represents a continuous segment where the absolute value of the second-order difference of the sensor data is less than a preset value.

[0058] Specifically, to determine the data type of sensor data in a sub-process, a linear fit must first be performed on the sensor data to obtain the linear fit degree. The linear fit degree characterizes the deviation between the fitted data obtained from the linear fit and the sensor data of the sub-process. This deviation can be characterized by the aforementioned mean square error, root mean square error, coefficient of determination, mean absolute error, and mean absolute percentage error. Specifically, the linear fit degree is compared with a preset first threshold. If the linear fit degree is not greater than the first threshold, the sensor data of the sub-process is determined to be linear. If the linear fit degree is greater than the first threshold, the data type of the sensor data in the sub-process needs to be further determined based on the longest near-zero second-order difference segment of the fitted curve obtained from the linear fit, to determine whether the data type of the sensor data in the sub-process is partially linear or nonlinear. In this example, a linear fit is performed on the sensor data of the sub-process to obtain the linear fit degree; the obtained linear fit degree is compared with a preset first threshold to determine the data type of the current sub-process sensor data, improving the accuracy of determining the data type of the sensor data.

[0059] When it is determined that the data type of the sensor data in a sub-process is not linear, it is necessary to obtain the near-zero second-order difference segment of the sensor data based on the sensor data of the sub-process. Subsequently, the data type of the sensor data is determined based on the near-zero second-order difference segment. Accordingly, as an example, the method also includes:

[0060] Calculate the second-order difference of the sensor data of the sub-process based on the sensor data of the sub-process.

[0061] Identify the continuous segments where the absolute value of the second difference is less than a preset value;

[0062] The longest continuous segment among the continuous segments is selected as the near-zero second-order difference segment.

[0063] Specifically, firstly, based on the sensor data of the sub-process, the second-order difference of the sub-process sensor data is calculated. In practical applications, calculating the second-order difference of data is often used to process non-stationary time series data. The second-order difference can reveal the acceleration or deceleration of data changes, thereby helping to analyze trend changes in the data. The second-order difference can eliminate trend changes in the data, making it easier to capture the true patterns in the dataset. Next, based on the calculated second-order difference values, a continuous segment with an absolute value less than a preset value is determined, and the longest continuous segment is selected as the near-zero second-order difference segment in this example. The near-zero second-order difference segment typically indicates that the data is stationary within this segment, and the trend is linear. In this example, by calculating the second-order difference of the sensor data, the near-zero second-order difference segment of the sub-process sensor data is determined, improving the accuracy of obtaining the near-zero second-order difference segment of the sub-process sensor data.

[0064] Furthermore, after obtaining the near-zero second-order difference segment from the sensor data of the sub-process, the data type of the sensor data of the sub-process is further determined based on the near-zero second-order difference segment; correspondingly, as an example, Figure 3 An exemplary flowchart of a wafer data feature extraction method is shown; based on any example, the data type of sensor data for the sub-process is determined according to the longest near-zero second-order difference segment of the fitted curve obtained by linear fitting, including:

[0065] Step 301: If the length of the near-zero second-order difference segment is longer than the second threshold, then the data type of the sensor data is determined to be partially linear.

[0066] Step 302: If the length of the near-zero second-order difference segment is not longer than the second threshold, then the data type of the sensor data is determined to be nonlinear.

[0067] Specifically, the obtained near-zero second-order difference segment is compared with a preset second threshold. If the length of the near-zero second-order difference segment is longer than the second threshold, the data type of the sub-process is considered to be partially linear; if the length of the near-zero second-order difference segment is not longer than the second threshold, the data type of the sensor data in the sub-process is considered to be nonlinear. For example, if the length of the near-zero second-order difference segment is longer than the preset second threshold, then in the sub-process, the data type of the sensor data corresponding to the near-zero second-order difference segment is linear, and the remaining part is nonlinear. In this example, the near-zero second-order difference segment of the sensor data in the sub-process is used to determine whether the data type of the sensor data in the current sub-process is partially linear or nonlinear, thus improving the accuracy of data type determination.

[0068] Furthermore, different feature data extraction methods need to be performed based on the different data types of the sensor data; correspondingly, as an example, Figure 4An exemplary flowchart of a wafer data feature extraction method is shown. Based on any example, feature extraction is performed on the sensor data of a sub-process according to the data type to obtain the feature data of the sub-process, including:

[0069] Step 401: If the data type is linear, then perform linear feature extraction on the sensor data of the sub-process to obtain the feature data of the sub-process.

[0070] Step 402: If the data type is partially linear, then perform linear and nonlinear feature extraction on the sensor data of the sub-process to obtain the feature data of the sub-process.

[0071] Step 403: If the data type is non-linear, perform non-linear feature extraction on the sensor data of the sub-process to obtain the feature data of the sub-process.

[0072] Specifically, when the data type of the sensor data in a sub-process is determined to be linear according to the aforementioned method, linear feature extraction is performed on the sensor data to obtain the feature data of the sub-process. When the data type of the sensor data in a sub-process is determined to be partially linear, linear feature extraction is performed on the linear sensor data and nonlinear feature extraction is performed on the nonlinear sensor data to obtain the corresponding feature data of the sub-process. When the data type of the sensor data in a sub-process is determined to be nonlinear, nonlinear feature extraction is performed on the sensor data to obtain the feature data of the sub-process.

[0073] For example, when the data type of a sub-process is determined to be partially linear, the sensor data corresponding to the near-zero second-order difference segment in the sensor data is linear data. Specifically, when the near-zero difference segment is obtained, the start and end times and duration of the near-zero second-order difference segment can be determined based on the sensor data of the sub-process. The data type of the sensor data corresponding to this part is linear, and the data type of the sensor data corresponding to the remaining part is nonlinear. When performing feature extraction for sub-process steps with partially linear sensor data types, linear feature extraction is performed for linear sensor data, and nonlinear feature extraction is performed for nonlinear sensor data to obtain the feature data of the sub-process.

[0074] In this example, different feature data are extracted based on the different data types of sensor data in the sub-process, which improves the comprehensiveness and accuracy of feature data extraction from sensor data; it avoids the loss of detailed feature data due to incomplete feature data extraction from sensor data, and improves the accuracy of subsequent analysis of defects and fault causes.

[0075] Furthermore, it is necessary to accurately obtain the linear fit of the sensor data; correspondingly, as an example, Figure 5 An exemplary flowchart of a wafer data feature extraction method is shown; based on any example, linear fitting is performed according to sensor data from a sub-process to obtain the linear fitting degree, including:

[0076] Step 501: Based on the sensor data of the sub-process, establish the first objective function, perform linear fitting, and obtain the linear fitting parameters of the first objective function;

[0077] Step 502: Substitute the linear fitting parameters into the first objective function to obtain the fitted data;

[0078] Step 503: Obtain the linear fit degree based on the fitted data and sensor data.

[0079] In practical applications, to ensure process repeatability and reliability, semiconductor process steps involve many static environmental controls, such as temperature and pressure control in chemical vapor deposition (CVD) or physical vapor deposition (PVD), current control in electroplating or chemical mechanical polishing (CMP), and voltage control in oxidation or diffusion. Theoretically, the sensor data acquired during these steps should be linear data with a slope close to 0. However, abnormal fluctuations in actual sensor data often indicate equipment failure, environmental changes, material problems, or deviations in process control, which are the root causes of product defects, performance degradation, reduced yield, shortened device lifespan, or even direct failure. Therefore, it is necessary to obtain the corresponding characteristic data when the sensor data is linear for subsequent anomaly analysis. Specifically, based on the acquired sub-process sensor data, a first objective function is established, and linear fitting is performed to obtain the linear fitting parameters of the first objective function. For example, abnormal data in the sub-process sensor data is removed to obtain a set of normal data after removing anomalies. For the set of normal data, the first objective function is established as follows:

[0080] S = at1 + b

[0081] Where S represents sensor data, a and b represent linear fitting parameters, and t1 represents relative time, which is obtained by subtracting the initial time of the process to be analyzed from the timestamp of the sensor data. Based on the first function, the least squares method is used for linear fitting to obtain the fitting parameters of the first objective function; the obtained linear fitting parameters are expressed as:

[0082]

[0083] Where a and b represent linear fitting parameters, n represents the number of sensor data, and S i，nt represents the data from the i-th sensor (excluding outliers). i，n This represents the relative time corresponding to the i-th sensor data. The obtained linear fitting parameters are then substituted into the first objective function to obtain the fitted data; the linear fit degree is obtained based on the fitted data and the sensor data. For example, the root mean square error (RMSE) is used to represent the numerical characterization of the linear fit degree; the formula for the RMS error is:

[0084]

[0085] Where RMSE represents the root mean square error, n represents the number of sensor data, and S i S represents sensor data (including data from outliers). i，f represents the fitted data; i represents the data from any one sensor.

[0086] It should be noted that the specific numerical type of the linear fit degree represented in this example is not limited; it can be any one or more of the aforementioned mean squared error, coefficient of determination, mean absolute error, and mean absolute percentage error. In this example, by establishing an objective function and performing linear fitting to obtain the linear fit degree, the accuracy of obtaining the linear fit degree is improved. Subsequently, based on the linear fit degree, the data type of the sensor data is determined, improving the accuracy of the sensor data type determination.

[0087] Furthermore, linear feature extraction is performed on the sensor data of the sub-process to obtain the feature data of the sub-process, including:

[0088] The linear fitting parameters and the degree of linear fitting are used as feature data for the sub-process.

[0089] Specifically, when the data type of the sensor in the sub-process is determined to be linear according to the aforementioned method, the fitting parameters and fitting data obtained during linear fitting are the feature data when the data type of the sensor data in the sub-process is linear. For example, the feature data includes linear fitting parameters a and b, the root mean square error (RMSE), and the initial time and duration of the sub-process, etc.; this example does not limit the specific type of feature data. In this example, using the linear fitting parameters and linear fit degree obtained when the data type of the sub-process sensor is linear as the feature data of the sub-process improves the accuracy of obtaining the sub-process feature data.

[0090] Furthermore, for nonlinear sensor data, corresponding nonlinear feature extraction is required; accordingly, as an example, Figure 6 An exemplary schematic diagram of a wafer data feature extraction process is shown; based on any example, nonlinear feature extraction is performed on sensor data to obtain sub-process feature data, including:

[0091] Step 601: Based on the sensor data of the sub-process, establish a second objective function, perform nonlinear fitting, and obtain the nonlinear fitting parameters of the second objective function;

[0092] Step 602: Substitute the nonlinear fitting parameters into the second objective function to obtain the nonlinear fitting data;

[0093] Step 603: Obtain the nonlinear fitting degree based on the nonlinear fitting data and the sensor data of the sub-process; the nonlinear fitting degree characterizes the deviation between the nonlinear fitting data and the sensor data.

[0094] Step 604: Use the nonlinear fitting parameters and the degree of nonlinear fitting as feature data of the sub-process.

[0095] In practical applications, the sub-process steps corresponding to non-linear sensor data types are mostly transitional steps. Whether the changes in parameters such as temperature, pressure, and current are gradual during these steps may affect the dimensional accuracy, electrical performance, and reliability of the device. Considering that transitional steps often occur between two stable steps where the sensor data type is linear, non-linear fitting of the sensor data can be performed using hyperbolic tangent functions, logistic functions, Gaussian functions, exponential decay functions, sine functions, etc. This example does not restrict the specific type of objective function used for non-linear fitting; it can be any of the aforementioned examples or other functions. Specifically, a second objective function is established based on the sensor data of the sub-process. Based on this second objective function, least squares is used for nonlinear fitting to obtain the nonlinear fitting parameters. This example uses a variant of the hyperbolic tangent function for nonlinear fitting. First, based on the linear type of each sub-step in the process to be analyzed, the data segments corresponding to all linear types in the process to be analyzed are determined. Then, for the data segments corresponding to nonlinear data types, the last sensor data point of the preceding linear type segment is used as the starting coordinate of the nonlinear fitting objective function, and the first sensor data point of the following linear type segment is used as the ending coordinate of the nonlinear fitting objective function. These are respectively denoted as the starting coordinates (S...). s , t s ), endpoint coordinates (S) e , t e Among them, S s t represents the last sensor data of the preceding linear segment. s S represents the time corresponding to the sensor data at that point. e t represents the first sensor data of the subsequent linear segment. e This represents the time corresponding to the sensor data at that point; the second objective function can be expressed as:

[0096]

[0097] Where S represents sensor data, S e S represents the first sensor data of the next linear type segment. s t represents the last sensor data of the preceding linear segment. c Here, t2 represents the nonlinear fitting parameter, i.e., the timestamp correction coefficient; k represents the nonlinear fitting parameter, i.e., the nonlinear transition smoothness parameter; t2 represents the relative time corresponding to any sensor data, which is obtained by subtracting the initial time of the process to be analyzed from the timestamp of the sensor data, as mentioned above. The obtained nonlinear fitting parameter is substituted into the second objective function to obtain the corresponding nonlinear fitting data. The nonlinear fitting degree is obtained based on the nonlinear fitting data and the sensor data of the sub-process; where the nonlinear fitting degree characterizes the deviation between the nonlinear fitting data and the sensor data. Then, the nonlinear fitting parameter and the nonlinear fitting degree are used as feature data of the sub-process. For example, the feature data includes: timestamp correction coefficient t2. c The parameters include the nonlinear transition smoothness parameter k, the root mean square error (RMSE), the initial time, and the duration corresponding to the nonlinear type. In this example, the nonlinear fitting parameters and the degree of nonlinear fit are obtained by performing nonlinear fitting on the sensor data of the nonlinear type. These parameters and the degree of nonlinear fit are then used as feature data for extraction, improving the accuracy of feature data extraction.

[0098] As an example, when the sensor data type corresponding to the sub-process is partially linear, the aforementioned linear feature extraction method is used to extract the feature data of the sub-process for the linear part; and the aforementioned nonlinear feature extraction method is used to extract the feature data of the sub-process for the nonlinear part.

[0099] For example, after determining the data type based on the sensor data of the sub-process through linear and nonlinear fitting, this application marks the process segments corresponding to different data types; linear sensor data is marked as a stable segment, partially linear sensor data is marked as a quasi-stable segment, and nonlinear sensor data is marked as a transition segment. Figure 7 This is a schematic diagram illustrating the segmentation of a single wafer and a single sensor according to sub-processes, as an example of this application; Figure 7As shown, the horizontal axis corresponds to the process time of the entire process. The entire process is divided into multiple steps according to sub-process steps, such as step 1, step 2, step 3, etc.; the vertical axis corresponds to the sensor data under each sub-process step. Each sub-process step corresponds to a different time period, and each sub-process corresponds to the sensor data in that step. A smooth horizontal line in the graph represents a stable segment, where the sensor data is of linear type. A fluctuating line represents a transition segment, where the sensor data is of non-linear type. A segment with both smooth and fluctuating lines represents a quasi-stable segment, where the sensor data is of partially linear type. For example, steps 1 and 2 are stable segments, step 3 is a transition segment, step 4 is a quasi-stable segment, and step 5 is a stable segment.

[0100] Figure 8 This is a schematic diagram of a data segment of linear type sensor feature data as an example of this application; such as Figure 8 As shown, the data segments of sensor feature data of linear data type basically present a stable horizontal line state, and the red dots in the figure are the abnormal point data identified during anomaly detection. Figure 9 This is a schematic diagram of a data segment of a partial linear sensor feature data example from this application; such as... Figure 9 As shown, some segments of sensor feature data of linear data types exhibit a stable horizontal line state, as shown by the segments corresponding to the light green background in the figure; another part exhibits a fluctuating state, as shown by the segments corresponding to the light red background in the figure. Among them, the fluctuating segments have a relatively short duration, while the stable segments have a longer duration. The blue solid line in the figure represents the fitting result based on the Adaptive Piecewise Constant Approximation Algorithm (APCA), and the orange dotted dashed line represents the fitting result based on the hyperbolic tangent function variant, which means that the fitting results obtained by the two algorithms are represented. Figure 10 This is a schematic diagram of a data segment of a nonlinear sensor characteristic data example from this application; as shown below. Figure 10 As shown, the data segments of sensor feature data of nonlinear data types exhibit a fluctuating state; the blue solid line in the figure represents the fitting result based on the Adaptive Piecewise Constant Approximation Algorithm (APCA), and the orange dashed line represents the fitting result based on the hyperbolic tangent function variant, which means that the fitting results obtained by the two algorithms are represented; the data segments of nonlinear data types may also have stable phases, but the duration of the stable phase in the sub-process is usually very short.

[0101] Based on the aforementioned example, Table 1 presents the feature data extraction values ​​for different sub-process steps of one example. As shown in Table 1, for... Figure 7 Step 3 in the process corresponds to the nonlinear type of sensor feature data, and the timestamp correction coefficient t of the nonlinear fitting parameters is extracted. cThe nonlinear transition smoothness parameter k, root mean square error (RMSE), initial time, and duration are used as feature data for this step. For the partial linear sensor feature data corresponding to step 4, the timestamp correction coefficient t of the nonlinear fitting parameters corresponding to the nonlinear type is extracted. c The nonlinear transition smoothness parameter k, root mean square error (RMSE), initial time, duration, and linear fitting parameters a and b corresponding to the linear type are used as the feature data for this step. For the sensor feature data of the linear type corresponding to step 5, the linear fitting parameters a and b, root mean square error (RMSE), initial time, and duration are extracted as the feature data for this step.

[0102] Table 1

[0103]

[0104] Furthermore, sensor data for each sub-process is obtained based on the acquired sensor data of the process to be analyzed; correspondingly, as an example, the sensor data of the sub-processes in the process to be analyzed are obtained, including:

[0105] Acquire sensor data for the process to be analyzed; the sensor data includes the labels of the corresponding sub-processes;

[0106] Based on the labels, the sensor data of the process to be analyzed is divided into sub-processes to obtain the sensor data of the sub-processes within the process to be analyzed.

[0107] Specifically, the sensor data for the process step to be analyzed is first obtained. This sensor data includes labels for each sub-process within the analyzed process. Based on these sub-process labels, the sensor data for the analyzed process is segmented, resulting in sensor data for each sub-process within the analyzed process. In this example, the sensor data is segmented based on time sequence according to the sub-process labels. Subsequently, feature data is extracted from the segmented sub-process sensor data, avoiding the omission of information from key steps and improving the comprehensiveness and accuracy of feature data extraction.

[0108] After extracting the characteristic data of the process to be analyzed, it is necessary to analyze the faults and causes of defects based on the characteristic data; accordingly, as an example, the method also includes:

[0109] Based on the characteristic data of the process to be analyzed, fault analysis processing is performed to obtain the fault analysis results.

[0110] Specifically, after obtaining the feature data of each sub-process using the aforementioned method, the feature data of all sub-processes are summarized to obtain the complete feature data of the process to be analyzed. For example, Table 2 is a sample feature data statistics table for the process to be analyzed. As shown in Table 2, this table includes feature data corresponding to different samples (wafers); the feature data corresponding to each sensor data is divided into feature data corresponding to linear types and feature data corresponding to nonlinear types according to different data types.

[0111] Table 2

[0112]

[0113] Next, the causes of defects and failures are analyzed based on the feature data table; this example does not limit the method for analyzing the causes of defects and failures. For example, correlation analysis is performed based on the feature data of the process to be analyzed; first, the target correlation object is determined, such as yield or defect rate. Based on the determined target correlation object, regression analysis or classification analysis is performed based on the obtained feature data to obtain the cause of the target correlation object. For example, if regression analysis shows that under a certain feature data, the feature data corresponding to normal wafers are all higher than that feature data, and the feature data corresponding to abnormal wafers are all lower than that feature data, then the cause of the defect is considered to be that feature data. Next, further analysis is performed to determine the process conditions and / or process parameters under the actual process steps corresponding to that feature data to obtain the analysis results. In this example, by analyzing the linear, partially linear, and nonlinear feature data obtained by the aforementioned method, the causes of defects and failures are analyzed, avoiding the inability to extract detailed features that could lead to the failure or failure, thus improving the accuracy of the failure analysis results.

[0114] As an example, Figure 11 This is a flowchart illustrating the wafer feature data extraction method of this application; as shown. Figure 11As shown, the sensor data corresponding to the process step to be analyzed is first obtained; the sensor data is divided according to the labels of each sub-process in the process to be analyzed to obtain the sensor data of the sub-process; firstly, it is determined whether the sensor data type under k=1 in each sub-process step in the process to be analyzed is linear. If it is linear, it is marked as a stable segment and the feature data corresponding to the linear sensor data is extracted. The non-linear sensor data is temporarily recorded in the data set to be analyzed in the second step; the feature data of the linear sensor data of each sub-process is extracted to obtain the feature data of the sub-process, until the linear feature data of all sub-processes in the process to be analyzed are obtained. After identifying and extracting feature data from the linear components of all sub-processes, the sensor data in the dataset to be analyzed a second time is again labeled according to the sub-process, with k'=1 selected to confirm whether the data type is partially linear. If so, it is marked as a quasi-stable segment of the sub-process, and linear and nonlinear feature extraction is performed on the sensor data of the sub-process. Feature data corresponding to partially linear sensor data is obtained. Sensor data that is not partially linear is temporarily added to the dataset to be analyzed a third time. Feature extraction is performed on the partially linear sensor data of each sub-process to obtain the sub-process's feature data, until the feature data of all sub-processes in the process to be analyzed is obtained. Further, the sensor data in the dataset to be analyzed a third time is again labeled according to the sub-process, with k”=1 selected to extract nonlinear features from the nonlinear sensor data. Feature extraction is performed on the nonlinear sensor data of each sub-process to obtain the sub-process's feature data, until the feature data of all sub-processes in the process to be analyzed is obtained. After extraction, the feature data of the process to be analyzed is obtained, including both linear and nonlinear feature data.

[0115] The wafer data feature extraction method provided in this embodiment obtains sensor data from normal and abnormal wafers in a sub-process of the process to be analyzed, and determines the data type of the obtained sensor data. The data types include linear, partially linear, and nonlinear types. Then, feature extraction is performed on the sensor data of the sub-process according to the specific data type to obtain the feature data of the sub-process. Based on the feature data of the sub-process, the feature data of the process to be analyzed is obtained. Extracting feature data from sensor data according to different data types improves the comprehensiveness and accuracy of feature extraction; it effectively captures feature data of different linear types from the sensor data; subsequently, based on comprehensive and accurate feature data of the process to be analyzed, fault or anomaly analysis can be performed, improving the accuracy of fault analysis results.

[0116] Example 2

[0117] Figure 12 An exemplary schematic diagram of a wafer data feature extraction device is shown, such as... Figure 12 As shown, the wafer data analysis apparatus provided in this embodiment includes:

[0118] The acquisition module 21 is used to acquire sensor data of the sub-process in the process to be analyzed; wherein, the sensor data of the sub-process includes sensor data of each normal wafer and abnormal wafer in the sub-process;

[0119] The determination module 22 is used to determine the data type of the sensor data of the sub-process; the data types include linear type, partially linear type and nonlinear type;

[0120] The extraction module 33 is used to extract features from the sensor data of the sub-process according to the data type to obtain the feature data of the sub-process;

[0121] The generation module 33 is used to obtain the feature data of the process to be analyzed based on the feature data of the sub-processes. The feature data of the process to be analyzed includes the feature data of the sub-processes in the process to be analyzed.

[0122] Production data in the semiconductor manufacturing process is generally stored in the data center or database of the semiconductor factory in structured, unstructured, or database formats. When it is necessary to analyze the causes of faults or defects, it is necessary to first use libraries and tools in programming languages ​​to export the data of the process to be analyzed into a structured format that is easy to analyze or process.

[0123] As an example, production data is typically a dictionary with wafer numbers as keys, where the key value for each wafer is a two-dimensional array. Each two-dimensional array contains data about the wafer's production process, typically including: data index, time, product batch information (Lot ld), wafer number (Wafer ld), silicon substrate number (Substrate ld), step name, and various sensor readings (Sensor1 Value, Sensor2 Value, etc.).

[0124] Specifically, wafer manufacturing involves multiple process steps. Typically, when root cause analysis is needed to identify defective products or malfunctions, the specific process step with the problem can be determined based on the actual process parameters and sensor data from each step. For the identified problematic process step, further analysis of the specific defects and malfunctions within that process's sub-processes is then conducted. For example, wafer manufacturing processes include oxidation, photolithography, etching, and thin film deposition. The oxidation process includes multiple sub-processes such as heating, gas deposition, and cooling. Sensor data obtained from the database is categorized according to sub-process tags to obtain sub-process sensor data; each group of sensor data includes sensor data from multiple wafers within that sub-process.

[0125] Specifically, the sensor data for each sub-process within the process to be analyzed is first acquired. This sub-process sensor data includes sensor data from normal wafers and sensor data from abnormal wafers. Each sub-process's sensor data includes multiple different types of sensor data, such as temperature sensor data, humidity sensor data, and pressure sensor data. Anomaly detection is then performed on the acquired sensor data for each sub-process to eliminate abnormal data. Here, abnormal data refers to data that is significantly different from other data as detected by the dataset's algorithm. For example, anomaly detection algorithms could be the Isolation Forest algorithm, the percentile method, or the distribution fitting algorithm. This example does not limit the anomaly detection method.

[0126] Based on the acquired sensor data from each sub-process, the data type of the sensor data is determined. In this example, the data types include linear, partially linear, and nonlinear types. Specifically, the process to be analyzed includes multiple sub-process steps; each sub-process step is equipped with multiple sensors, such as temperature sensors, humidity sensors, and pressure sensors, to collect sensor data from each sub-process step, resulting in temperature sensor data, humidity sensor data, pressure sensor data, etc. The data type of each sensor data is determined, thus obtaining the data type of each sensor data point within the sub-process. Subsequently, feature data is extracted based on different data types to obtain the feature data of each sensor data point within each sub-process. For each sensor data point within a sub-process, the extracted feature data of the sub-process yields the feature data of the process to be analyzed. Subsequently, fault analysis and defect analysis are performed based on the extracted feature data of the process to be analyzed. The feature data in this example includes fitting parameters obtained from fitting different linear types, and a numerical representation of the deviation between the fitted data and the sensor data. For example, the numerical representation of the deviation can be mean square error (MSE), root mean square error (RMSE), coefficient of determination (R²), mean absolute error (MAE), mean absolute percentage error (MAPE), etc. These error values ​​serve as the feature data in this example and are used for subsequent defect and fault analysis. In this example, the data type of the sensor data is determined based on the obtained sensor data of each sub-process of the process to be analyzed, and feature data is extracted from the sensor data for different data types. The feature data of the process to be analyzed is obtained based on the extracted feature data of the sub-process. The solution of this application extracts feature data of wafer sensors more comprehensively and accurately; subsequent fault analysis based on the extracted feature data yields more accurate analysis results.

[0127] Furthermore, it is necessary to determine the data type based on the acquired sensor data; correspondingly, as an example, module 22 is used for:

[0128] Based on the sensor data of the sub-process, a linear fit is performed to obtain the linear fit degree; the linear fit degree characterizes the deviation between the fitted data obtained by the linear fit and the sensor data of the sub-process.

[0129] If the linear fit is not greater than the first threshold, the data type is determined to be linear.

[0130] If the linear fit is greater than the first threshold, the data type of the sensor data in the sub-process is determined based on the longest near-zero second-order difference segment of the fitted curve obtained by linear fitting; the near-zero second-order difference segment represents a continuous segment where the absolute value of the second-order difference of the sensor data is less than a preset value.

[0131] Specifically, to determine the data type of sensor data in a sub-process, a linear fit must first be performed on the sensor data to obtain the linear fit degree. The linear fit degree characterizes the deviation between the fitted data obtained from the linear fit and the sensor data of the sub-process. This deviation can be characterized by the aforementioned mean square error, root mean square error, coefficient of determination, mean absolute error, and mean absolute percentage error. Specifically, the linear fit degree is compared with a preset first threshold. If the linear fit degree is not greater than the first threshold, the sensor data of the sub-process is determined to be linear. If the linear fit degree is greater than the first threshold, the data type of the sensor data in the sub-process needs to be further determined based on the longest near-zero second-order difference segment of the fitted curve obtained from the linear fit, to determine whether the data type of the sensor data in the sub-process is partially linear or nonlinear. In this example, a linear fit is performed on the sensor data of the sub-process to obtain the linear fit degree; the obtained linear fit degree is compared with a preset first threshold to determine the data type of the current sub-process sensor data, improving the accuracy of determining the data type of the sensor data.

[0132] When it is determined that the data type of the sensor data in the sub-process is not linear, it is necessary to further obtain the near-zero second-order difference segment of the sensor data based on the sensor data of the sub-process. Subsequently, the data type of the sensor data is determined based on the near-zero second-order difference segment. Accordingly, as an example, the device also includes: a calculation module 25, which is used for:

[0133] Calculate the second-order difference of the sensor data of the sub-process based on the sensor data of the sub-process.

[0134] Identify the continuous segments where the absolute value of the second difference is less than a preset value;

[0135] The longest continuous segment among the continuous segments is selected as the near-zero second-order difference segment.

[0136] Specifically, firstly, based on the sensor data of the sub-process, the second-order difference of the sub-process sensor data is calculated. In practical applications, calculating the second-order difference of data is often used to process non-stationary time series data. The second-order difference can reveal the acceleration or deceleration of data changes, thereby helping to analyze trend changes in the data. The second-order difference can eliminate trend changes in the data, making it easier to capture the true patterns in the dataset. Next, based on the calculated second-order difference values, a continuous segment with an absolute value less than a preset value is determined, and the longest continuous segment is selected as the near-zero second-order difference segment in this example. The near-zero second-order difference segment typically indicates that the data is stationary within this segment, and the trend is linear. In this example, by calculating the second-order difference of the sensor data, the near-zero second-order difference segment of the sub-process sensor data is determined, improving the accuracy of obtaining the near-zero second-order difference segment of the sub-process sensor data.

[0137] Furthermore, after obtaining the near-zero second-order difference segment of the sensor data from the sub-process, the data type of the sensor data from the sub-process is further determined based on the near-zero second-order difference segment; correspondingly, as an example, the determination module 22 is used for:

[0138] If the length of the near-zero second-order difference segment is longer than the second threshold, then the data type of the sensor data is determined to be partially linear.

[0139] If the length of the near-zero second-order difference segment is not longer than the second threshold, then the data type of the sensor data is determined to be nonlinear.

[0140] Specifically, the obtained near-zero second-order difference segment is compared with a preset second threshold. If the length of the near-zero second-order difference segment is longer than the second threshold, the data type of the sub-process is considered to be partially linear; if the length of the near-zero second-order difference segment is not longer than the second threshold, the data type of the sensor data in the sub-process is considered to be nonlinear. For example, if the length of the near-zero second-order difference segment is longer than the preset second threshold, then in the sub-process, the data type of the sensor data corresponding to the near-zero second-order difference segment is linear, and the remaining part is nonlinear. In this example, the near-zero second-order difference segment of the sensor data in the sub-process is used to determine whether the data type of the sensor data in the current sub-process is partially linear or nonlinear, thus improving the accuracy of data type determination.

[0141] Furthermore, different feature data extractions need to be performed based on the different data types of the sensor data; correspondingly, as an example, extraction module 23 is used for:

[0142] If the data type is linear, then linear feature extraction is performed on the sensor data of the sub-process to obtain the feature data of the sub-process;

[0143] If the data type is partially linear, then linear and nonlinear features are extracted from the sensor data of the sub-process to obtain the feature data of the sub-process.

[0144] If the data type is non-linear, then non-linear feature extraction is performed on the sensor data of the sub-process to obtain the feature data of the sub-process.

[0145] Specifically, when the data type of the sensor data in a sub-process is determined to be linear according to the aforementioned method, linear feature extraction is performed on the sensor data to obtain the feature data of the sub-process. When the data type of the sensor data in a sub-process is determined to be partially linear, linear feature extraction is performed on the linear sensor data and nonlinear feature extraction is performed on the nonlinear sensor data to obtain the corresponding feature data of the sub-process. When the data type of the sensor data in a sub-process is determined to be nonlinear, nonlinear feature extraction is performed on the sensor data to obtain the feature data of the sub-process.

[0146] For example, when the data type of a sub-process is determined to be partially linear, the sensor data corresponding to the near-zero second-order difference segment in the sensor data is linear data. Specifically, when the near-zero difference segment is obtained, the start and end times and duration of the near-zero second-order difference segment can be determined based on the sensor data of the sub-process. The data type of the sensor data corresponding to this part is linear, and the data type of the sensor data corresponding to the remaining part is nonlinear. When performing feature extraction for sub-process steps with partially linear sensor data types, linear feature extraction is performed for linear sensor data, and nonlinear feature extraction is performed for nonlinear sensor data to obtain the feature data of the sub-process.

[0147] In this example, different feature data are extracted based on the different data types of sensor data in the sub-process, which improves the comprehensiveness and accuracy of feature data extraction from sensor data; it avoids the loss of detailed feature data due to incomplete feature data extraction from sensor data, and improves the accuracy of subsequent analysis of defects and fault causes.

[0148] Furthermore, it is necessary to accurately obtain the linear fit of the sensor data; accordingly, as an example, extraction module 23 is used for:

[0149] Based on the sensor data of the sub-process, a first objective function is established, and linear fitting is performed to obtain the linear fitting parameters of the first objective function;

[0150] Substitute the linear fitting parameters into the first objective function to obtain the fitted data;

[0151] The linear fit is obtained based on the fitted data and sensor data.

[0152] In practical applications, to ensure process repeatability and reliability, semiconductor process steps involve many static environmental controls, such as temperature and pressure control in chemical vapor deposition (CVD) or physical vapor deposition (PVD), current control in electroplating or chemical mechanical polishing (CMP), and voltage control in oxidation or diffusion. Theoretically, the sensor data acquired during these steps should be linear data with a slope close to 0. However, abnormal fluctuations in actual sensor data often indicate equipment failure, environmental changes, material problems, or deviations in process control, which are the root causes of product defects, performance degradation, reduced yield, shortened device lifespan, or even direct failure. Therefore, it is necessary to obtain the corresponding characteristic data when the sensor data is linear for subsequent anomaly analysis. Specifically, based on the acquired sub-process sensor data, a first objective function is established, and linear fitting is performed to obtain the linear fitting parameters of the first objective function. For example, abnormal data in the sub-process sensor data is removed to obtain a set of normal data after removing anomalies. For the set of normal data, the first objective function is established as follows:

[0153] S = at1 + b

[0154] Where S represents sensor data, a and b represent linear fitting parameters, and t1 represents relative time, which is obtained by subtracting the initial time of the process to be analyzed from the timestamp of the sensor data. Based on the first function, the least squares method is used for linear fitting to obtain the fitting parameters of the first objective function; the obtained linear fitting parameters are expressed as:

[0155]

[0156] Where a and b represent linear fitting parameters, n represents the number of sensor data, and S i，n Indicates the first i Data from one sensor (excluding data from outliers), t i，n This represents the relative time corresponding to the i-th sensor data. The obtained linear fitting parameters are then substituted into the first objective function to obtain the fitted data; the linear fit degree is obtained based on the fitted data and the sensor data. For example, the root mean square error (RMSE) is used to represent the numerical characterization of the linear fit degree; the formula for the RMS error is:

[0157]

[0158] Where RMSE represents the root mean square error, n represents the number of sensor data, and S i S represents sensor data (including data from outliers).i，f represents the fitted data; i represents the data from any one sensor.

[0159] It should be noted that the specific numerical type of the linear fit degree represented in this example is not limited; it can be any one or more of the aforementioned mean squared error, coefficient of determination, mean absolute error, and mean absolute percentage error. In this example, by establishing an objective function and performing linear fitting to obtain the linear fit degree, the accuracy of obtaining the linear fit degree is improved. Subsequently, based on the linear fit degree, the data type of the sensor data is determined, improving the accuracy of the sensor data type determination.

[0160] Furthermore, linear feature extraction is performed on the sensor data of the sub-process to obtain the feature data of the sub-process, including:

[0161] The linear fitting parameters and the degree of linear fitting are used as feature data for the sub-process.

[0162] Specifically, when the data type of the sensor in the sub-process is determined to be linear according to the aforementioned method, the fitting parameters and fitting data obtained during linear fitting are the feature data when the data type of the sensor data in the sub-process is linear. For example, the feature data includes linear fitting parameters a and b, the root mean square error (RMSE), and the initial time and duration of the sub-process, etc.; this example does not limit the specific type of feature data. In this example, using the linear fitting parameters and linear fit degree obtained when the data type of the sub-process sensor is linear as the feature data of the sub-process improves the accuracy of obtaining the sub-process feature data.

[0163] Furthermore, for nonlinear sensor data, corresponding nonlinear feature extraction is required; accordingly, as an example, extraction module 23 is used for:

[0164] Based on the sensor data from the sub-process, a second objective function is established, and nonlinear fitting is performed to obtain the nonlinear fitting parameters of the second objective function.

[0165] Substituting the nonlinear fitting parameters into the second objective function yields the nonlinear fitting data;

[0166] The nonlinear fitting degree is obtained by comparing the nonlinear fitting data with the sensor data of the sub-process; the nonlinear fitting degree characterizes the deviation between the nonlinear fitting data and the sensor data.

[0167] The nonlinear fitting parameters and the degree of nonlinear fitting are used as feature data of the sub-process.

[0168] In practical applications, the sub-process steps corresponding to non-linear sensor data types are mostly transitional steps. Whether the changes in parameters such as temperature, pressure, and current are gradual during these steps may affect the dimensional accuracy, electrical performance, and reliability of the device. Considering that transitional steps often occur between two stable steps where the sensor data type is linear, non-linear fitting of the sensor data can be performed using hyperbolic tangent functions, logistic functions, Gaussian functions, exponential decay functions, sine functions, etc. This example does not restrict the specific type of objective function used for non-linear fitting; it can be any of the aforementioned examples or other functions. Specifically, a second objective function is established based on the sensor data of the sub-process. Based on this second objective function, least squares is used for nonlinear fitting to obtain the nonlinear fitting parameters. This example uses a variant of the hyperbolic tangent function for nonlinear fitting. First, based on the linear type of each sub-step in the process to be analyzed, the data segments corresponding to all linear types in the process to be analyzed are determined. Then, for the data segments corresponding to nonlinear data types, the last sensor data point of the preceding linear type segment is used as the starting coordinate of the nonlinear fitting objective function, and the first sensor data point of the following linear type segment is used as the ending coordinate of the nonlinear fitting objective function. These are respectively denoted as the starting coordinates (S...). s , t s ), endpoint coordinates (S) e , t e Among them, S s t represents the last sensor data of the preceding linear segment. s S represents the time corresponding to the sensor data at that point. e t represents the first sensor data of the subsequent linear segment. e This represents the time corresponding to the sensor data at that point; the second objective function can be expressed as:

[0169]

[0170] Where S represents sensor data, S e S represents the first sensor data of the next linear type segment. s t represents the last sensor data of the preceding linear segment. cHere, t2 represents the nonlinear fitting parameter, i.e., the timestamp correction coefficient; k represents the nonlinear fitting parameter, i.e., the nonlinear transition smoothness parameter; t2 represents the relative time corresponding to any sensor data, which is obtained by subtracting the initial time of the process to be analyzed from the timestamp of the sensor data, as mentioned above. The obtained nonlinear fitting parameter is substituted into the second objective function to obtain the corresponding nonlinear fitting data. The nonlinear fitting degree is obtained based on the nonlinear fitting data and the sensor data of the sub-process; where the nonlinear fitting degree characterizes the deviation between the nonlinear fitting data and the sensor data. Then, the nonlinear fitting parameter and the nonlinear fitting degree are used as feature data of the sub-process. For example, the feature data includes: timestamp correction coefficient t2. c The parameters include the nonlinear transition smoothness parameter k, the root mean square error (RMSE), the initial time, and the duration corresponding to the nonlinear type. In this example, the nonlinear fitting parameters and the degree of nonlinear fit are obtained by performing nonlinear fitting on the sensor data of the nonlinear type. These parameters and the degree of nonlinear fit are then used as feature data for extraction, improving the accuracy of feature data extraction.

[0171] Furthermore, sensor data for each sub-process is obtained based on the acquired sensor data of the process to be analyzed; correspondingly, as an example, the acquisition module 21 is used for:

[0172] Acquire sensor data for the process to be analyzed; the sensor data includes the labels of the corresponding sub-processes;

[0173] Based on the labels, the sensor data of the process to be analyzed is divided into sub-processes to obtain the sensor data of the sub-processes within the process to be analyzed.

[0174] Specifically, the sensor data for the process step to be analyzed is first obtained. This sensor data includes labels for each sub-process within the analyzed process. Based on these sub-process labels, the sensor data for the analyzed process is segmented, resulting in sensor data for each sub-process within the analyzed process. In this example, the sensor data is segmented based on time sequence according to the sub-process labels. Subsequently, feature data is extracted from the segmented sub-process sensor data, avoiding the omission of information from key steps and improving the comprehensiveness and accuracy of feature data extraction.

[0175] After extracting the characteristic data of the process to be analyzed, it is necessary to analyze the faults and causes of defects based on the characteristic data; accordingly, as an example, the device also includes: analysis module 26, which is used for:

[0176] Based on the characteristic data of the process to be analyzed, fault analysis processing is performed to obtain the fault analysis results.

[0177] Specifically, after obtaining the feature data of each sub-process using the aforementioned method, the feature data of all sub-processes are summarized to obtain the complete feature data of the process to be analyzed. For example, Table 2 is a sample feature data statistics table for the process to be analyzed. As shown in Table 2, this table includes feature data corresponding to different samples (wafers); the feature data corresponding to each sensor data is divided into feature data corresponding to linear types and feature data corresponding to nonlinear types according to different data types.

[0178] Next, the causes of defects and failures are analyzed based on the feature data table; this example does not limit the method for analyzing the causes of defects and failures. For example, correlation analysis is performed based on the feature data of the process to be analyzed; first, the target correlation object is determined, such as yield or defect rate. Based on the determined target correlation object, regression analysis or classification analysis is performed based on the obtained feature data to obtain the cause of the target correlation object. For example, if regression analysis shows that under a certain feature data, the feature data corresponding to normal wafers are all higher than that feature data, and the value of the feature data corresponding to abnormal wafers is all lower than that feature data, then it is considered that the cause of the defect is that feature data. Next, further analysis is performed to determine the process conditions under the actual process steps corresponding to that feature data and the cause of the defect, obtaining the analysis results. In this example, by analyzing the linear, partially linear, and nonlinear feature data obtained by the aforementioned method, the causes of defects and failures are analyzed, avoiding the inability to extract detailed features that could lead to the failure or failure cause; thus improving the accuracy of the failure analysis results.

[0179] In the wafer data feature extraction apparatus provided in this embodiment, sensor data from normal and abnormal wafers in a sub-process of the process to be analyzed are acquired, and the data type of the acquired sensor data is determined. The data types include linear, partially linear, and nonlinear types. Then, feature extraction is performed on the sensor data of the sub-process according to the specific data type to obtain the feature data of the sub-process. Based on the feature data of the sub-process, the feature data of the process to be analyzed is obtained. Extracting feature data from sensor data according to different data types improves the comprehensiveness and accuracy of feature extraction; it effectively captures feature data of different linear types from the sensor data; subsequently, fault or anomaly analysis based on comprehensive and accurate feature data of the process to be analyzed can improve the accuracy of fault analysis results.

[0180] Example 3

[0181] Figure 13 A schematic diagram of the structure of an electronic device is provided for an embodiment of this application. For example... Figure 13As shown, the electronic device provided in this embodiment includes a processor 291 and a memory 292; it may also include a communication interface 293 and a bus 294. The processor 291, memory 292, and communication interface 293 can communicate with each other via the bus 294. The communication interface 293 can be used for information transmission. The processor 291 can call logical instructions in the memory 292 to execute the method described above.

[0182] Furthermore, the logic instructions in the aforementioned memory 292 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium.

[0183] The memory 292, as a computer-readable storage medium, can be used to store software programs and computer-executable programs, such as program instructions / modules corresponding to the methods in the embodiments of this application. The processor 291 executes functional applications and data processing by running the software programs, instructions, and modules stored in the memory 292, that is, it implements the methods in the above method examples.

[0184] The memory 292 may include a program storage area and a data storage area. The program storage area may store the operating system and application programs required for at least one function; the data storage area may store data created based on the use of the terminal device. Furthermore, the memory 292 may include high-speed random access memory and may also include non-volatile memory.

[0185] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0186] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the above-described method.

[0187] Finally, it should be noted that other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A method of extracting data features of a wafer, characterized by, include: Acquire sensor data for sub-processes within the process to be analyzed; wherein, the sensor data for a sub-process includes sensor data for each normal wafer and abnormal wafer within that sub-process; Determine the data type of the sensor data for the sub-process; the data type includes linear, partially linear, and nonlinear types; Based on the data type, feature extraction is performed on the sensor data of the sub-process to obtain the feature data of the sub-process; Based on the characteristic data of the sub-process, the characteristic data of the process to be analyzed is obtained, and the characteristic data of the process to be analyzed includes the characteristic data of the sub-processes in the process to be analyzed.

2. The method of claim 1, wherein, The determination of the data type of the sensor data for the sub-process includes: Based on the sensor data of the sub-process, a linear fit is performed to obtain the linear fit degree; the linear fit degree characterizes the deviation between the fitted data obtained by the linear fit and the sensor data of the sub-process. If the linear fit is not greater than the first threshold, the data type is determined to be linear. If the linear fit is greater than the first threshold, the data type of the sensor data of the sub-process is determined based on the longest near-zero second-order difference segment of the fitted curve obtained by linear fitting; the near-zero second-order difference segment represents a continuous segment in which the absolute value of the second-order difference of the sensor data is less than a preset value.

3. The method of claim 2, wherein, The step of determining the data type of the sensor data for the sub-process based on the longest near-zero second-order difference segment of the fitted curve obtained by linear fitting includes: If the length of the near-zero second-order difference segment is longer than the second threshold, then the data type of the sensor data is determined to be partially linear. If the length of the near-zero second-order difference segment is not longer than the second threshold, then the data type of the sensor data is determined to be nonlinear.

4. The method of claim 3, wherein, The step of extracting features from the sensor data of the sub-process according to the data type to obtain the feature data of the sub-process includes: If the data type is linear, then linear feature extraction is performed on the sensor data of the sub-process to obtain the feature data of the sub-process; If the data type is partially linear, then linear and nonlinear features are extracted from the sensor data of the sub-process to obtain the feature data of the sub-process. If the data type is non-linear, then non-linear feature extraction is performed on the sensor data of the sub-process to obtain the feature data of the sub-process.

5. The method of claim 4, wherein, The step of performing linear fitting based on the sensor data from the sub-process to obtain the linear fitting degree includes: Based on the sensor data of the sub-process, a first objective function is established, and linear fitting is performed to obtain the linear fitting parameters of the first objective function; Substitute the linear fitting parameters into the first objective function to obtain the fitted data; The linear fit degree is obtained based on the fitted data and the sensor data.

6. The method of claim 5, wherein, The linear feature extraction of the sensor data of the sub-process to obtain the feature data of the sub-process includes: The linear fitting parameters and the linear fitting degree are used as feature data of the sub-process.

7. The method according to claim 4, characterized in that, The step of performing nonlinear feature extraction on the sensor data to obtain the feature data of the sub-process includes: Based on the sensor data of the sub-process, a second objective function is established, and nonlinear fitting is performed to obtain the nonlinear fitting parameters of the second objective function; Substituting the nonlinear fitting parameters into the second objective function yields the nonlinear fitting data; The nonlinear fitting degree is obtained based on the nonlinear fitting data and the sensor data of the sub-process; the nonlinear fitting degree characterizes the deviation between the nonlinear fitting data and the sensor data. The nonlinear fitting parameters and the nonlinear fitting degree are used as feature data of the sub-process.

8. The method of claim 2, wherein, The method further includes: Calculate the second-order difference of the sensor data of the sub-process based on the sensor data of the sub-process; Determine a continuous segment in which the absolute value of the second-order difference is less than the preset value; The longest continuous segment among the continuous segments is selected as the near-zero second-order difference segment.

9. The method of claim 1, wherein, The acquisition of sensor data for the sub-processes in the process to be analyzed includes: Acquire sensor data for the process to be analyzed; the sensor data includes the labels of the corresponding sub-processes; Based on the labels, the sensor data of the process to be analyzed is divided to obtain the sensor data of the sub-processes within the process to be analyzed.

10. The method according to any one of claims 1 to 9, characterized in that, The method further includes: Based on the characteristic data of the process to be analyzed, fault analysis processing is performed to obtain the fault analysis results.

11. A data feature extraction apparatus for a wafer, characterized by comprising: include: The acquisition module is used to acquire sensor data of sub-processes in the process to be analyzed; wherein, the sensor data of sub-processes includes sensor data of each normal wafer and abnormal wafer in the sub-process; A determination module is used to determine the data type of the sensor data of the sub-process; the data type includes linear type, partially linear type and nonlinear type; The extraction module is used to extract features from the sensor data of the sub-process according to the data type, so as to obtain the feature data of the sub-process; The generation module is used to obtain the feature data of the process to be analyzed based on the feature data of the sub-process, wherein the feature data of the process to be analyzed includes the feature data of the sub-processes in the process to be analyzed.

12. An electronic device, comprising: include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executable instructions; the processor executes the computer-executable instructions stored in the memory to implement the method as described in any one of claims 1 to 10.

13. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any one of claims 1 to 10.

14. A computer program product, characterised in that, It includes a computer program that, when executed by a processor, implements the method as described in any one of claims 1 to 10.

Images