A method and apparatus for cleaning silicon-based epitaxial wafers for feature analysis

Abstract

The application discloses a silicon-based epitaxial wafer cleaning method and device, and relates to the technical field of silicon-based epitaxial wafer processing. The method comprises the following steps: detecting the state of a plurality of silicon-based epitaxial wafers to be cleaned, and obtaining a plurality of metal distribution information and a plurality of particle distribution information; grouping the silicon-based epitaxial wafers according to the current cleaning equipment state, the plurality of metal distribution information and the plurality of particle distribution information; performing parameter optimization on a metal-particle removal link of a plurality of silicon-based epitaxial wafer groups, and outputting a plurality of adaptive cleaning parameters; and performing batch cleaning processing on the plurality of silicon-based epitaxial wafer groups by using a cleaning equipment. The method solves the problem that the existing silicon-based epitaxial wafer cleaning method lacks pertinence, cannot effectively clean the silicon-based epitaxial wafers according to the actual state of the silicon-based epitaxial wafers, and results in uneven cleaning effect and low cleaning efficiency.

Description

Technical Field

[0001] This application relates to the field of silicon-based epitaxial wafer processing technology, specifically to a silicon-based epitaxial wafer cleaning method and apparatus based on feature analysis. Background Technology

[0002] With the development of the semiconductor industry, silicon-based epitaxial wafers are increasingly widely used in integrated circuit manufacturing and other fields. However, traditional silicon-based epitaxial wafer cleaning methods lack precise analysis of the individual characteristics of silicon-based epitaxial wafers. They typically use uniform cleaning parameters to clean all silicon-based epitaxial wafers without making reasonable grouping and parameter adjustments based on the actual working conditions of the cleaning equipment and the preset cleaning time requirements. This results in inconsistent cleaning effects and low cleaning efficiency, thereby affecting the product yield and performance stability. Summary of the Invention

[0003] This application provides a silicon-based epitaxial wafer cleaning method and apparatus based on feature analysis, which solves the technical problem that existing silicon-based epitaxial wafer cleaning methods lack specificity, cannot effectively clean according to the actual state of the silicon-based epitaxial wafer, resulting in inconsistent cleaning effects and low cleaning efficiency.

[0004] The technical solution to the above-mentioned technical problems in this application is as follows: In a first aspect, this application provides a method for cleaning silicon-based epitaxial wafers based on feature analysis, the method comprising: The state of several silicon-based epitaxial wafers to be cleaned is monitored to obtain several metal distribution information and several particle distribution information. Constrained by a preset cleaning time, the silicon-based epitaxial wafers are grouped according to the current cleaning equipment status, the distribution information of the several metals, and the distribution information of the several particles, to generate multiple silicon-based epitaxial wafer groups. With the goal of maximizing the overall cleaning quality within the group, the cleaning parameters for the metal-particle removal process are optimized for each of the multiple silicon-based epitaxial wafer groups, and multiple suitable cleaning parameters are output. According to the aforementioned multiple adaptive cleaning parameters, the multiple silicon-based epitaxial wafer groups are subjected to batch cleaning processes using cleaning equipment.

[0005] Secondly, this application provides a silicon-based epitaxial wafer cleaning apparatus for feature analysis, comprising: The information acquisition module is used to detect the status of several silicon-based epitaxial wafers to be cleaned and to acquire several metal distribution information and several particle distribution information. The sample grouping module is used to group the silicon-based epitaxial wafers according to the current cleaning equipment status, the distribution information of the metals and the distribution information of the particles, with a preset cleaning time as a constraint, to generate multiple silicon-based epitaxial wafer groups. The parameter output module is used to optimize the cleaning parameters for the metal-particle removal process of the multiple silicon-based epitaxial wafer groups with the goal of maximizing the overall cleaning quality within the group, and output multiple suitable cleaning parameters. The sample processing module is used to perform batch cleaning of the multiple silicon-based epitaxial wafer groups according to the multiple adapted cleaning parameters and through cleaning equipment.

[0006] This application provides one or more technical solutions, which have at least the following technical effects or advantages: This application provides a silicon-based epitaxial wafer cleaning method and apparatus based on feature analysis. First, the silicon-based epitaxial wafers to be cleaned are subjected to state detection to obtain their metal and particle distribution information, providing a detailed and targeted data foundation for subsequent cleaning work. Second, the silicon-based epitaxial wafers are rationally grouped according to a preset cleaning time and the status of the cleaning equipment, enabling the cleaning work to be completed efficiently within the specified time and avoiding the problem of low cleaning efficiency caused by unreasonable grouping. Then, cleaning parameters are optimized with the goal of maximizing the overall cleaning quality within the group, ensuring that each silicon-based epitaxial wafer group receives the most suitable cleaning parameters, thereby improving the consistency of cleaning results. Finally, the silicon-based epitaxial wafer groups are batch-cleaned according to the suitable cleaning parameters, ensuring the standardization and efficiency of the cleaning process.

[0007] Through the above technical solutions, this application effectively reduces the phenomena of insufficient and excessive cleaning, ensuring high product yield, and maintaining the production line rhythm through batch processing within the group. Thus, while ensuring high yield, it can also take into account high production efficiency, improving the cleaning efficiency and cleaning quality of silicon-based epitaxial wafers in the metal and particle removal process. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 This is a schematic flowchart of a silicon-based epitaxial wafer cleaning method based on feature analysis provided in an embodiment of this application; Figure 2 This is a schematic diagram of a silicon-based epitaxial wafer cleaning device for feature analysis provided in an embodiment of this application.

[0010] The components represented by each number in the attached diagram are explained below: Information acquisition module 11, sample grouping module 12, parameter output module 13, sample processing module 14. Detailed Implementation

[0011] This application provides a silicon-based epitaxial wafer cleaning method and apparatus based on feature analysis, which addresses the technical problem that existing silicon-based epitaxial wafer cleaning methods lack specificity, cannot effectively clean according to the actual condition of the silicon-based epitaxial wafer, resulting in inconsistent cleaning effects and low cleaning efficiency.

[0012] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0013] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0014] In the description of this application, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use this application. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that this application can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid unnecessarily obscuring the description of this application. Therefore, this application is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.

[0015] Example 1, as Figure 1 As shown in the figure, this application provides a silicon-based epitaxial wafer cleaning method based on feature analysis, including: S10: Perform state detection on several silicon-based epitaxial wafers to be cleaned, and obtain several metal distribution information and several particle distribution information; In this embodiment, the state of several silicon-based epitaxial wafers to be cleaned is first monitored. This can be done using monitoring equipment such as a scanning electron microscope (SEM) or an energy dispersive spectroscopy (EDS) analyzer to scan the surface of the silicon-based epitaxial wafers. Specifically, the SEM is used to observe the microstructure and particle distribution on the surface of the silicon-based epitaxial wafers, while the EDS is used to analyze the types and distribution of metal elements. Through the coordinated operation of these devices, metal distribution information and particle distribution information for each silicon-based epitaxial wafer are obtained.

[0016] Secondly, the acquired metal and particle distribution information is stored and displayed in the form of data tables or images. Meanwhile, to ensure the accuracy and reliability of the data, the testing process must be carried out according to operating procedures, and the testing equipment must be calibrated and maintained regularly.

[0017] The acquired metal and particle distribution information is further analyzed using data analysis software. Through data analysis, the distribution patterns of metals and particles are understood, such as the main areas of metal concentration, and the size and density distribution of particles.

[0018] Specifically, step S10 in the method includes: The surface of the silicon-based epitaxial wafers was scanned using a total reflection X-ray fluorescence spectrometer to obtain the content and spatial distribution of metal elements, and feature extraction was performed to generate metal distribution information, resulting in several metal distribution information, wherein the metal distribution information includes at least the total amount of metal, the proportion of key metals, and the two-dimensional distribution of key metals. A laser scanning particle detection device is used to detect the plurality of silicon-based epitaxial wafers respectively, to obtain the number of particles and the spatial coordinates of particles at different size thresholds, and to extract features to generate particle distribution information, thereby obtaining a plurality of particle distribution information, wherein the particle distribution information includes at least the total number of particles, particle size distribution and spatial uniformity.

[0019] In this embodiment, firstly, a total internal reflection X-ray fluorescence spectrometer is used to scan the surface of the silicon-based epitaxial wafer. The total internal reflection X-ray fluorescence spectrometer can detect the content and spatial distribution of metal elements on the surface of the silicon-based epitaxial wafer. During the scanning process, scanning parameters, such as scanning speed and scanning range, are set according to the size and characteristics of the silicon-based epitaxial wafer to ensure that the obtained metal element information is complete and accurate.

[0020] Secondly, after obtaining the content and spatial distribution data of metal elements, feature extraction is performed. Through data analysis and processing, the total metal content, the proportion of critical metals, and the two-dimensional distribution of critical metals are extracted to form metal distribution information. Critical metals refer to metal elements that, even when present at extremely low concentrations on or within the silicon wafer, can have a catastrophic impact on the electrical performance, reliability, and long-term stability of semiconductor devices.

[0021] Next, particle distribution information is obtained using a laser scanning particle detection device. This device can detect the number of particles and their spatial coordinates at different size thresholds. During the detection process, the device is debugged and calibrated to ensure accurate detection of particles of various sizes. Simultaneously, size thresholds are set based on the actual conditions of the silicon-based epitaxial wafer to obtain particle information that better meets practical needs.

[0022] Furthermore, after obtaining the number of particles and their spatial coordinates at different size thresholds, feature extraction is performed. Features such as the total number of particles, particle size distribution, and spatial uniformity are extracted to form particle distribution information. After acquiring both metal distribution information and particle distribution information, they are managed and stored.

[0023] For example, a database is established to classify and store the metal distribution and particle distribution information of each silicon-based epitaxial wafer, facilitating subsequent queries and use. Simultaneously, the database is regularly maintained and updated to ensure the accuracy and completeness of the data.

[0024] S20: Based on the preset cleaning time, according to the current cleaning equipment status, the distribution information of the several metals and the distribution information of the several particles, the several silicon-based epitaxial wafers are grouped to generate multiple silicon-based epitaxial wafer groups. In this embodiment, a preset cleaning time is used as a constraint, which affects the efficiency and progress of the cleaning work. Meanwhile, if the cleaning equipment is in good working order, it can handle a relatively high volume of cleaning tasks and achieve high cleaning efficiency; if the equipment malfunctions or is damaged, the grouping strategy is adjusted to avoid poor cleaning results or excessively long cleaning times due to equipment problems.

[0025] Next, silicon-based epitaxial wafers are grouped by combining metal distribution information and several particle distribution information. Factors such as total metal content, the proportion of key metals, the total number of particles, and particle size distribution can be considered comprehensively. For example, silicon-based epitaxial wafers with similar total metal content and similar particle size distribution can be grouped together, allowing for the use of relatively uniform cleaning parameters in subsequent cleaning processes, thus improving cleaning efficiency.

[0026] The resulting multiple silicon-based epitaxial wafer groups ensure that the silicon-based epitaxial wafers within each group have a certain similarity in metal and particle characteristics. Considering the actual state of the equipment, the number of groups is dynamically set, and the cleaning time limit is taken into account, ensuring that the cleaning time for each group is within the preset cleaning time range. This lays the foundation for subsequent optimization of cleaning parameters and batch cleaning processing.

[0027] Specifically, step S20 in the method includes: Read the current cleaning equipment status of the cleaning equipment, wherein the current cleaning equipment status is the number of currently working equipment and equipment attributes, and the equipment attributes include the maximum number of silicon-based epitaxial wafers that can be cleaned simultaneously; With a constraint that the cleaning time is less than the preset cleaning time, the grouping quantity analysis is performed based on the number of currently working devices and device attributes, and the optimal grouping quantity is output. Feature extraction, numerical representation, and feature fusion are performed on the aforementioned metal distribution information and particle distribution information to generate several silicon-based epitaxial wafer feature vectors. Using the optimal number of groups as the number of cluster centers, and based on the feature vectors of the silicon-based epitaxial wafers, the K-means algorithm is used to group the silicon-based epitaxial wafers to generate multiple silicon-based epitaxial wafer groups.

[0028] In this embodiment, the current status of the cleaning equipment is first read. Through data interaction with the control system of the cleaning equipment, the number of currently operational devices and their attributes are obtained, including the maximum number of silicon-based epitaxial wafers that can be cleaned simultaneously.

[0029] Secondly, constrained by a minimum cleaning time, the optimal number of groups is analyzed based on the number of currently available and functional devices. Considering the cleaning equipment's efficiency, the maximum number of wafers each device can clean, and the preset cleaning time, the optimal number of groups is determined based on the total number of silicon-based epitaxial wafers and the approximate cleaning time. For example, a mathematical model can be established, combining historical cleaning data and equipment performance data for analysis, ultimately outputting the optimal number of groups.

[0030] Then, feature extraction, numerical representation, and feature fusion are performed on several metal distribution information sets and several particle distribution information sets, respectively. For the metal distribution information, in addition to the total metal amount, key metal ratio, and two-dimensional distribution of key metals mentioned above, features such as metal concentration gradient and the correlation between metals and particles can be further extracted. Furthermore, feature extraction extracts quantitative indicators with physical and technological significance from the original metal and particle distribution information. For example, the total metal amount and key metal ratio are extracted from the metal information; the total number of particles and the proportion of particles of a specific size are extracted from the particle information.

[0031] For particle distribution information, in addition to the total number of particles, particle size distribution, and spatial uniformity, the analysis also includes particle shape characteristics and particle aggregation. These characteristics are then numerically represented, for example, using specific numerical values ​​to represent the metal concentration gradient and particle shape irregularity. Finally, the features from the metal distribution information and particle distribution information are fused to generate several silicon-based epitaxial wafer feature vectors.

[0032] Finally, the optimal number of groups is used as the number of cluster centers. Based on the feature vectors of several silicon-based epitaxial wafers, the K-means algorithm is used to group the silicon-based epitaxial wafers. The K-means algorithm is a commonly used clustering algorithm that iteratively divides data points into different clusters, resulting in high similarity among data points within clusters and low similarity among data points between clusters.

[0033] During the grouping process, cluster centers are first randomly initialized. Then, the distance from the eigenvector of each silicon-based epitaxial wafer to each cluster center is calculated, and the wafer is assigned to the nearest cluster. Next, the cluster center of each cluster is recalculated, and the division is repeated until the cluster centers no longer change significantly or the preset number of iterations is reached.

[0034] By using the K-means algorithm to group silicon-based epitaxial wafers, multiple groups of silicon-based epitaxial wafers are generated, ensuring that the silicon-based epitaxial wafers in each group have high similarity in metal and particle characteristics, which provides a good foundation for subsequent optimization of cleaning parameters and batch cleaning processing.

[0035] Among them, with a constraint of less than a preset cleaning time, the optimal number of groups is output based on the number of currently working devices and their attributes, and a grouping analysis is performed, including: Based on the aforementioned silicon-based epitaxial wafers, grouping and enumerating are performed to generate several initial group quantities; Based on the current number of working devices and device attributes, the cleaning time is predicted according to the number of initial groups to obtain a number of minimum cleaning times. The initial number of groups whose minimum cleaning time is less than the preset cleaning time is set as the number of qualified groups, resulting in multiple qualified group numbers. The qualified group number with the largest value is selected as the optimal group number.

[0036] In this embodiment, firstly, a grouping enumeration is performed on a number of silicon-based epitaxial wafers. Grouping enumeration involves trying different combinations of the silicon-based epitaxial wafers to generate several initial grouping numbers. For example, assuming there are 100 silicon-based epitaxial wafers, they can be tried to be divided into 2 groups, 4 groups, 5 groups, 10 groups, etc., to obtain the initial grouping number.

[0037] Secondly, based on the current number of working devices and their attributes, cleaning time is predicted for several initial group quantities. For example, if there are currently 3 working devices, and each device has a maximum cleaning capacity of 20 pieces, when there are 2 groups, each group contains 50 pieces. The devices cannot clean this group simultaneously and need to clean it in batches. The time required to clean this group is calculated based on parameters such as the device's cleaning rate. Similarly, similar cleaning time predictions are performed for other initial group quantities to ultimately obtain several minimum cleaning times.

[0038] Then, the initial number of groups whose minimum cleaning time is less than the preset cleaning time is set as the number of qualified groups. Assuming the preset cleaning time is 3 hours, among the calculated minimum cleaning times, if the minimum cleaning time is 2.5 hours when there are 4 groups, 2.8 hours when there are 5 groups, and 3.5 hours when there are 2 groups, then the initial number of groups of 4 and 5 is the number of qualified groups, resulting in multiple qualified group counts.

[0039] Finally, the largest number of qualified groups was selected as the optimal number of groups. This is because a larger number of groups increases the likelihood of similarity in metal and particle characteristics among the silicon-based epitaxial wafers within each group, resulting in higher cleaning precision and facilitating subsequent optimization of cleaning parameters and batch cleaning processes. In other words, the number of groups with the highest cleaning precision was chosen while meeting the cleaning time constraints.

[0040] Therefore, based on the above example, if the acceptable number of groups is 4 and 5, then the optimal number of groups is 5. Determining the optimal number of groups can improve cleaning efficiency and quality while meeting the preset cleaning time.

[0041] Specifically, based on the current number of working devices and device attributes, cleaning time is predicted according to the several initial grouping quantities to obtain several minimum cleaning times, including: Randomly select any initial group number from the plurality of initial group numbers as the first initial group number; Based on the current number of working devices and device attributes, the number of repeated cleaning cycles for each device is randomly divided according to the first initial grouping number, generating multiple first device working schemes. The cleaning time is predicted for each of the multiple first equipment working schemes to obtain multiple first predicted cleaning times. The first predicted cleaning time with the shortest time is selected as the first minimum cleaning time and added to the multiple minimum cleaning times.

[0042] In this embodiment, firstly, one initial group number is randomly selected from several initial group numbers as the first initial group number. For example, if there are 3, 5, or 7 initial group numbers, 5 groups are randomly selected as the first initial group number. Then, based on the current number of working devices and device attributes, each device may clean once or multiple times. Using the maximum number of silicon-based epitaxial wafers that can be cleaned simultaneously as a constraint, the number of repeated cleaning cycles for each device is randomly divided according to the first initial group number, generating multiple first device working schemes.

[0043] For example, there are currently 4 working devices, each device can clean a maximum of 15 wafers, and the total number of silicon-based epitaxial wafers is 100. When the first initial grouping is 5 groups, each group has 20 wafers.

[0044] The first equipment's working plan is as follows: Equipment 1 cleans the first 15 pieces of the first group, then cleans the remaining 5 pieces; Equipment 2 cleans the first 15 pieces of the second group, then cleans the remaining 5 pieces; Equipment 3 cleans the third group; Equipment 4 cleans the fourth group; for the remaining fifth group, after Equipment 1 cleans the remaining 5 pieces of the first group, it cleans the first 10 pieces of the fifth group, and after Equipment 2 cleans the remaining 5 pieces of the second group, it cleans the remaining 10 pieces of the fifth group. Other different methods of dividing the cleaning cycle can also be used, thus generating multiple working plans for the first equipment.

[0045] Subsequently, cleaning duration predictions were made for multiple working schemes of the first equipment. The predictions considered factors such as the equipment's cleaning rate, start-up and stop times, and the transition time between equipment. For example, the cleaning rate was 2 minutes per piece, the start-up and stop times were each 1 minute, and the transition time between equipment was 0.5 minutes.

[0046] Based on one of the above-described first equipment operating schemes, the cleaning time for that scheme is calculated. Calculations are performed for all first equipment operating schemes to obtain multiple first predicted cleaning times.

[0047] Finally, the shortest cleaning time from the first predicted cleaning times is selected as the first minimum cleaning time and added to several minimum cleaning times. This process of randomly selecting the initial number of groups, generating the equipment operating plan, predicting cleaning times, and selecting the minimum cleaning time is repeated continuously until the corresponding operations have been completed for all initial group numbers, thus obtaining a complete set of minimum cleaning times, providing data support for subsequently determining the optimal number of groups.

[0048] S30: With the goal of maximizing the overall cleaning quality within the group, optimize the cleaning parameters for the metal-particle removal process of the multiple silicon-based epitaxial wafer groups respectively, and output multiple suitable cleaning parameters; In the embodiments of this application, Specifically, step S30 in the method includes: Randomly select a first silicon-based epitaxial wafer group from the plurality of silicon-based epitaxial wafer groups; The cleaning parameter adjustment space of the cleaning equipment in the metal-particle removal process is obtained. Multiple initial cleaning parameters are randomly selected within the cleaning parameter adjustment space. The cleaning equipment is a fully automatic single-plate cleaning machine. The cleaning parameters include chemical agent concentration, reaction temperature, reaction time, and mega-sound energy. A cleaning quality prediction plugin for silicon-based epitaxial wafers is constructed, wherein a deep learning model is trained to convergence using a sample metal distribution information set, a sample particle distribution information set, a sample cleaning parameter set, and a sample cleaning quality coefficient set to generate the cleaning quality prediction plugin. With the goal of maximizing the overall cleaning quality within the group, the cleaning quality prediction plugin and multiple initial cleaning parameters are used to optimize the cleaning parameters of the first silicon-based epitaxial wafer group, output the first adaptive cleaning parameter, and add it to the multiple adaptive cleaning parameters.

[0049] In this embodiment, one silicon-based epitaxial wafer group is randomly selected from multiple groups as the first silicon-based epitaxial wafer group. Random selection ensures that the subsequent optimization process can cover silicon-based epitaxial wafer groups with different characteristics, making the final adaptive cleaning parameters more universal.

[0050] Secondly, the adjustment range of cleaning parameters for the cleaning equipment in the metal-particle removal process was determined. Since the cleaning equipment is a fully automatic single-plate cleaner, its cleaning parameters include chemical concentration, reaction temperature, reaction time, and megaphonic energy. Different chemical concentrations affect the removal efficiency of metals and particles, while reaction temperature and time play a crucial role in the chemical reaction. Megaphonic energy enhances the physical action during the cleaning process, improving cleaning efficiency. Several initial cleaning parameters were randomly selected within the adjustment range to provide a starting point for the subsequent optimization process.

[0051] Then, a plugin for predicting the cleaning quality of silicon-based epitaxial wafers was constructed. A deep learning model was trained until convergence using sample metal distribution information sets, sample particle distribution information sets, sample cleaning parameter sets, and sample cleaning quality coefficient sets. The sample metal distribution information set contains the distribution of metals on different silicon-based epitaxial wafers; the sample particle distribution information set reflects the relevant characteristics of the particles; the sample cleaning parameter set records the combinations of various cleaning parameters; and the sample cleaning quality coefficient set, based on historical cleaning results, is a quantitative representation of the cleaning effect. By training the deep learning model with sample data, it can accurately predict the cleaning quality of silicon-based epitaxial wafers under different cleaning parameters.

[0052] Finally, with the goal of maximizing the overall cleaning quality within the group, the cleaning parameters for the first silicon-based epitaxial wafer group were optimized using a cleaning quality prediction plugin and multiple initial cleaning parameters. During the optimization process, the initial cleaning parameters were continuously adjusted, and the cleaning quality prediction plugin was used to predict the cleaning quality under different parameter combinations, gradually finding the parameter combination that yielded the highest overall cleaning quality for the group.

[0053] The parameter combination is used as the first adaptive cleaning parameter and added to multiple adaptive cleaning parameters. The process of randomly selecting silicon-based epitaxial wafer groups, obtaining initial cleaning parameters, building prediction plugins, and optimizing parameters is repeated until the corresponding operations are completed for all silicon-based epitaxial wafer groups, thereby obtaining a complete set of multiple adaptive cleaning parameters, providing accurate parameter basis for subsequent batch cleaning processes.

[0054] With the goal of maximizing the overall cleaning quality within the group, the cleaning quality prediction plugin and multiple initial cleaning parameters are used to optimize the cleaning parameters of the first silicon-based epitaxial wafer group, and the first adaptive cleaning parameters are output, including: Using the cleaning quality prediction plugin, the cleaning quality is predicted for the multiple initial cleaning parameters based on the first metal distribution information set and the first particle distribution information set of the first silicon-based epitaxial wafer group, and multiple sets of predicted cleaning quality coefficients are output. The predicted cleaning quality coefficients included in the multiple predicted cleaning quality coefficient sets are summed to obtain multiple predicted overall cleaning quality coefficients; The initial cleaning parameters are set as the initial solution, and multiple initial solutions are arranged in descending order according to the predicted overall cleaning quality coefficient to generate an initial solution sequence. The first solution of the initial solution sequence is selected as the optimal solution, the last solution is selected as the poor solution, and the remaining solutions are selected as the inferior solutions, thus obtaining the optimal solution, the poor solution, and multiple inferior solutions; The cleaning quality deviations between the multiple inferior solutions and the superior and inferior solutions are evaluated respectively, and a first optimization strategy is formulated based on the cleaning quality deviations. According to the first optimization strategy, the multiple inferior solutions are adjusted according to the preset optimization adjustment step size to obtain multiple updated inferior solutions. The superior solutions, inferior solutions and multiple updated inferior solutions are reordered according to the predicted overall cleaning quality coefficient from large to small to generate the first updated solution sequence. Following an iterative optimization mechanism of selecting superior, inferior, and poor solutions, updating the optimization strategy, optimizing inferior solutions, and updating the initial solution sequence, the cleaning parameters are optimized based on the first updated solution sequence until a preset number of convergences is reached. The superior solutions in the current updated solution sequence are then output as the first matching cleaning parameters.

[0055] In this embodiment, firstly, a cleaning quality prediction plugin is used to predict the cleaning quality of multiple initial cleaning parameters by combining the first metal distribution information set and the first particle distribution information set of the first silicon-based epitaxial wafer group. Based on a previously trained deep learning model, the cleaning quality prediction plugin outputs multiple sets of predicted cleaning quality coefficients according to the metal and particle distribution information of the silicon-based epitaxial wafers and different cleaning parameters. These predicted cleaning quality coefficient sets reflect the cleaning quality of the first silicon-based epitaxial wafer group under different initial cleaning parameters.

[0056] Secondly, the predicted cleaning quality coefficients contained in multiple sets of predicted cleaning quality coefficients are summed. Since each set of predicted cleaning quality coefficients may contain coefficients of multiple different dimensions, summing them together yields a comprehensive overall predicted cleaning quality coefficient, which facilitates subsequent comparison of cleaning effects with different initial cleaning parameters.

[0057] Next, the initial cleaning parameters are set as the initial solutions, and multiple initial solutions are arranged in descending order of the predicted overall cleaning quality coefficient to generate an initial solution sequence. Initial solutions with better cleaning performance are placed first, and initial solutions with poorer cleaning performance are placed last, which facilitates the subsequent selection of superior, inferior, and poor solutions.

[0058] After the initial solution sequence is generated, the first solution is designated as the optimal solution, the last solution as the suboptimal solution, and the remaining solutions as the inferior solutions. The optimal solution represents the parameter combination with the best cleaning effect among the current initial solutions, the suboptimal solution is the parameter combination with the worst cleaning effect, and the inferior solution is the parameter combination in an intermediate state.

[0059] Then, the cleaning quality deviations of multiple suboptimal solutions compared to optimal and poor solutions are evaluated. By comparing the predicted overall cleaning quality coefficients of suboptimal, optimal, and poor solutions, the difference is calculated to determine the gap in cleaning quality between suboptimal, optimal, and poor solutions. A first optimization strategy is formulated based on the cleaning quality deviations. For example, if the difference between a suboptimal solution and a optimal solution is large, a larger adjustment step size is needed to move it closer to the optimal solution; if the difference is small, the adjustment step size can be appropriately reduced to avoid over-adjustment.

[0060] Following the first optimization strategy, multiple suboptimal solutions are adjusted according to a preset optimization adjustment step size, resulting in multiple updated suboptimal solutions. The preset optimization adjustment step size is a fixed value pre-set during the optimization process to control the adjustment magnitude of suboptimal solutions. The adjusted updated suboptimal solutions may be closer to the optimal solutions, thereby improving the cleaning quality. Then, the optimal solutions, suboptimal solutions, and multiple updated suboptimal solutions are reordered from largest to smallest according to the predicted overall cleaning quality coefficient, generating the first updated solution sequence.

[0061] Finally, following the iterative optimization mechanism of selecting superior, inferior, and poor solutions, updating the optimization strategy, optimizing inferior solutions, and updating the initial solution sequence, parameter cleaning and optimization continue based on the first updated solution sequence. In each iteration, the steps of solution selection, strategy formulation, inferior solution adjustment, and sequence updating are repeated until the preset number of convergence iterations is reached.

[0062] Specifically, the preset number of convergences is a pre-set upper limit for the number of iterations. For example, the preset number of convergences can be set to 100. When the preset number of iterations is reached, the optimization process is considered to have converged. At this time, the optimal solution in the current updated solution sequence is output as the first fitting cleaning parameter.

[0063] The first set of adaptive cleaning parameters maximized the overall cleaning quality of the first silicon-based epitaxial wafer group, providing accurate parameter basis for subsequent batch cleaning processes.

[0064] Further, the cleaning quality deviations between the plurality of inferior solutions and the superior and inferior solutions are evaluated respectively, and a first optimization strategy is formulated based on the cleaning quality deviations, including: The average of the multiple predicted overall cleaning quality coefficients of the multiple suboptimal solutions is calculated as the comprehensive cleaning quality coefficient of the suboptimal solutions. The predicted overall cleaning quality coefficient of the optimal solution is used as the cleaning quality coefficient of the optimal solution, and the predicted overall cleaning quality coefficient of the poor solution is used as the cleaning quality coefficient of the poor solution. Calculate the absolute values ​​of the deviations between the comprehensive cleaning quality coefficient of the inferior solution and the cleaning quality coefficients of the superior and inferior solutions, respectively, to obtain the cleaning quality deviations of the superior and inferior solutions. If the cleaning quality deviation of the optimal solution is greater than or equal to the cleaning quality deviation of the poor solution, the direction that approaches the optimal solution will be used as the first optimization strategy. If the cleaning quality deviation of the optimal solution is less than the cleaning quality deviation of the poor solution, the direction away from the poor solution will be used as the first optimization strategy.

[0065] In this embodiment, firstly, the average of multiple predicted overall cleaning quality coefficients for multiple suboptimal solutions is calculated to obtain the comprehensive cleaning quality coefficient of the suboptimal solutions. The average value reflects the overall level of cleaning quality of multiple suboptimal solutions, facilitating subsequent comparison with optimal and poor solutions.

[0066] Secondly, the overall cleaning quality coefficient predicted by the optimal solution is defined as the optimal solution cleaning quality coefficient, and the overall cleaning quality coefficient predicted by the poor solution is defined as the poor solution cleaning quality coefficient.

[0067] Next, the absolute values ​​of the deviations between the comprehensive cleaning quality coefficient of the inferior solution and the cleaning quality coefficients of the superior and inferior solutions were calculated. By taking the absolute values, the positive and negative effects of the deviations were eliminated, and only the magnitude of the deviations was considered. The obtained cleaning quality deviations of the superior and inferior solutions visually demonstrate the gap between the comprehensive cleaning quality of the inferior solution and the cleaning quality of the superior and inferior solutions.

[0068] The first optimization strategy is formulated based on the relationship between the cleaning quality deviations of the optimal and suboptimal solutions. If the cleaning quality deviation of the optimal solution is greater than or equal to that of the suboptimal solution, it indicates that the overall cleaning quality of the suboptimal solution is significantly different from that of the optimal solution. In this case, the direction that approaches the optimal solution is adopted as the first optimization strategy. This means that in subsequent optimization processes, cleaning parameters should be adjusted in a direction that brings the suboptimal solution closer to the optimal solution to improve cleaning quality.

[0069] If the cleaning quality deviation of the optimal solution is less than that of the poor solution, it indicates that the overall cleaning quality of the inferior solution is significantly different from that of the poor solution. In this case, the first optimization strategy should be to move away from the poor solution. That is, in subsequent adjustments, we should prevent the inferior solution from developing into a poor solution, thereby improving the overall cleaning quality.

[0070] For example, the overall cleaning quality coefficient of the inferior solution is 80, the cleaning quality coefficient of the superior solution is 95, and the cleaning quality coefficient of the poor solution is 60. Then the cleaning quality deviation of the superior solution is |80-95|=15, and the cleaning quality deviation of the poor solution is |80-60|=20. Since the cleaning quality deviation of the superior solution is less than that of the poor solution, the direction away from the poor solution is taken as the first optimization strategy.

[0071] Developing optimization strategies based on the magnitude of deviation allows for more targeted adjustments to suboptimal solutions, improving optimization efficiency and helping to find the first suitable cleaning parameters that maximize the overall cleaning quality within the first silicon-based epitaxial wafer group more quickly. This provides more accurate parameter data for subsequent batch cleaning processes.

[0072] S40: According to the multiple adaptive cleaning parameters, the multiple silicon-based epitaxial wafer groups are subjected to batch cleaning treatment by the cleaning equipment.

[0073] In this embodiment, multiple silicon-based epitaxial wafer groups are batch-cleaned using a fully automated single-wafer cleaning machine according to several adapted cleaning parameters. During the cleaning process, the concentration of chemical agents is controlled to ensure that it precisely matches the values ​​set by the adapted cleaning parameters, thereby guaranteeing the effective removal of metals and particles. Simultaneously, the reaction temperature is adjusted and stabilized within the adapted temperature range to create favorable conditions for the smooth progress of the chemical reaction.

[0074] For the reaction time, follow the appropriate parameters to avoid affecting the cleaning effect due to excessively long or short times. Excessively long reaction times may cause unnecessary damage to the silicon epitaxial wafer, while excessively short times may fail to adequately remove metals and particles. The megaphonic energy also needs to be adjusted according to the appropriate parameters. Appropriate megaphonic energy can enhance the physical action during the cleaning process and improve cleaning efficiency, but excessively high energy may have adverse effects on the surface of the silicon epitaxial wafer.

[0075] During the batch cleaning process, the operating status of the cleaning equipment is monitored in real time to ensure the stability of all parameters. Simultaneously, the cleaned silicon-based epitaxial wafers are sampled and tested to check the removal of metals and particles, and to assess whether the cleaning effect meets expectations.

[0076] If the sampling test results show that the cleaning effect does not meet the requirements, analyze the reasons and adjust the appropriate cleaning parameters accordingly. This may be due to subtle differences in the characteristics of the silicon-based epitaxial wafer group, causing the original appropriate parameters to be inapplicable. In this case, re-optimize the parameters based on the test results to obtain more suitable cleaning parameters.

[0077] After completing the batch cleaning process, the cleaning equipment is cleaned and maintained to prepare for the next cleaning operation. Simultaneously, all data and results from this cleaning process are recorded, including cleaning parameters, cleaning time, and cleaning effects. This provides a reference and accumulates experience for subsequent cleaning work, continuously optimizing the cleaning method for silicon-based epitaxial wafers and improving cleaning quality and efficiency.

[0078] Specifically, step S40 in the method includes: The equipment operation scheme corresponding to the minimum cleaning time of the optimal group number is used as the adapted equipment operation scheme. According to the working scheme of the adapter equipment and the multiple adapter cleaning parameters, the multiple silicon-based epitaxial wafer groups are subjected to batch cleaning treatment by the cleaning equipment.

[0079] In this embodiment, the optimal number of groups is first determined. Factors such as the number of silicon-based epitaxial wafer groups, the processing capacity of the cleaning equipment, and the cleaning efficiency are considered. By simulating and analyzing the cleaning time under different group numbers, the group number that minimizes the total cleaning time is identified.

[0080] Secondly, given the determined optimal number of groups, calculate the equipment operation plan and its cleaning duration for each group. The equipment operation plan may involve the operating mode of the cleaning equipment, the sequence of operations for each step, and the start-up and shutdown times of the equipment. By recording and analyzing the cleaning duration under different operation plans, identify the equipment operation plan corresponding to the minimum cleaning duration, and use this as the suitable equipment operation plan.

[0081] Next, following the adapted equipment's operating plan and multiple adapted cleaning parameters, batch cleaning of multiple silicon-based epitaxial wafer groups was performed using the cleaning equipment. During the cleaning process, the cleaning equipment was operated according to the procedures and steps outlined in the adapted equipment's operating plan. For example, different cleaning function modules were activated in a predetermined sequence, and the operating time and intensity of each module were controlled. Simultaneously, adapted cleaning parameters such as chemical concentration, reaction temperature, reaction time, and megahertz energy were adjusted to ensure that each silicon-based epitaxial wafer group was cleaned under optimal conditions.

[0082] During batch cleaning, a comprehensive monitoring mechanism is established. Various operating parameters of the cleaning equipment, such as pressure, flow rate, and temperature, are monitored in real time to ensure stable operation. If any abnormalities are detected, timely adjustments are made to prevent impact on cleaning quality. Simultaneously, various adaptive parameters during the cleaning process are monitored in real time to ensure that chemical concentrations, reaction temperatures, reaction times, and megaphonic energy remain within the set ranges.

[0083] After cleaning, a comprehensive quality inspection is conducted on each silicon-based epitaxial wafer group. Spectroscopic analysis and microscopic observation are used to detect residual metals and particles on the silicon-based epitaxial wafers and to assess whether the cleaning effect meets the expected standards. If the cleaning effect of some silicon-based epitaxial wafer groups is found to be unsatisfactory, the reasons are analyzed, and the working scheme and cleaning parameters of the adapted equipment are optimized and adjusted. This may be due to a problem in a certain part of the equipment's working scheme, or the adapted cleaning parameters may require further fine-tuning. Through continuous optimization and improvement, the cleaning quality and efficiency of silicon-based epitaxial wafers are improved.

[0084] Finally, the process and results of this batch of cleaning treatment were summarized and recorded. The records included the optimal number of groups, the appropriate equipment operating plan, the appropriate cleaning parameters, the equipment operation status during the cleaning process, and the cleaning effect test results.

[0085] In summary, compared with the prior art, this application, by considering the actual state of the equipment, dynamically setting the number of groups, and then iteratively optimizing the cleaning parameters through an optimization mechanism, finally obtains the first suitable cleaning parameters. This can maximize the overall cleaning quality within the first silicon-based epitaxial wafer group, fully consider the differences in cleaning quality between different initial solutions, and gradually approach the optimal solution through reasonable solution selection, strategy formulation, and adjustment of inferior solutions.

[0086] In summary, the embodiments of this application have at least the following technical effects: This application provides a silicon-based epitaxial wafer cleaning method based on feature analysis. First, the silicon-based epitaxial wafers to be cleaned are subjected to state detection to obtain their metal and particle distribution information, providing a detailed and targeted data foundation for subsequent cleaning work. Second, the silicon-based epitaxial wafers are rationally grouped according to the preset cleaning time and the status of the cleaning equipment, enabling the cleaning work to be completed efficiently within the specified time and avoiding the problem of low cleaning efficiency caused by unreasonable grouping. Then, the cleaning parameters are optimized with the goal of maximizing the overall cleaning quality within the group, ensuring that each silicon-based epitaxial wafer group receives the most suitable cleaning parameters, thereby improving the consistency of cleaning results. Finally, the silicon-based epitaxial wafer groups are batch-cleaned according to the suitable cleaning parameters, ensuring the standardization and efficiency of the cleaning process. Through the above technical solution, this application effectively reduces under-cleaning and over-cleaning phenomena, ensuring high product yield, and maintaining production line rhythm through batch processing within groups. Therefore, it can ensure high yield while also achieving high production efficiency, improving the cleaning efficiency and quality of silicon-based epitaxial wafers in the metal and particle removal stage.

[0087] Example 2, as Figure 2 As shown, based on the same inventive concept as the silicon-based epitaxial wafer cleaning method for feature analysis provided in Embodiment 1, this application also provides a silicon-based epitaxial wafer cleaning apparatus for feature analysis, comprising: The information acquisition module 11 is used to perform state detection on several silicon-based epitaxial wafers to be cleaned, and to acquire several metal distribution information and several particle distribution information. The sample grouping module 12 is used to group the several silicon-based epitaxial wafers according to the current cleaning equipment status, the several metal distribution information and the several particle distribution information, with a preset cleaning time as a constraint, to generate multiple silicon-based epitaxial wafer groups. The parameter output module 13 is used to optimize the cleaning parameters for the metal-particle removal process of the multiple silicon-based epitaxial wafer groups with the goal of maximizing the overall cleaning quality within the group, and output multiple adaptive cleaning parameters. The sample processing module 14 is used to perform batch cleaning processing on the multiple silicon-based epitaxial wafer groups according to the multiple adaptive cleaning parameters through the cleaning equipment.

[0088] In one embodiment, the information acquisition module 11 is specifically used for: The surface of the silicon-based epitaxial wafers was scanned using a total reflection X-ray fluorescence spectrometer to obtain the content and spatial distribution of metal elements, and feature extraction was performed to generate metal distribution information, resulting in several metal distribution information, wherein the metal distribution information includes at least the total amount of metal, the proportion of key metals, and the two-dimensional distribution of key metals. A laser scanning particle detection device is used to detect the plurality of silicon-based epitaxial wafers respectively, to obtain the number of particles and the spatial coordinates of particles at different size thresholds, and to extract features to generate particle distribution information, thereby obtaining a plurality of particle distribution information, wherein the particle distribution information includes at least the total number of particles, particle size distribution and spatial uniformity.

[0089] In one embodiment, the sample grouping module 12 is specifically used for: Read the current cleaning equipment status of the cleaning equipment, wherein the current cleaning equipment status is the number of currently working equipment and equipment attributes, and the equipment attributes include the maximum number of silicon-based epitaxial wafers that can be cleaned simultaneously; With a constraint that the cleaning time is less than the preset cleaning time, the grouping quantity analysis is performed based on the number of currently working devices and device attributes, and the optimal grouping quantity is output. Feature extraction, numerical representation, and feature fusion are performed on the aforementioned metal distribution information and particle distribution information to generate several silicon-based epitaxial wafer feature vectors. Using the optimal number of groups as the number of cluster centers, and based on the feature vectors of the silicon-based epitaxial wafers, the K-means algorithm is used to group the silicon-based epitaxial wafers to generate multiple silicon-based epitaxial wafer groups.

[0090] Furthermore, in one embodiment, constrained by a minimum cleaning time, a grouping quantity analysis is performed based on the current number of working devices and device attributes to output the optimal grouping quantity, including: Based on the aforementioned silicon-based epitaxial wafers, grouping and enumerating are performed to generate several initial group quantities; Based on the current number of working devices and device attributes, the cleaning time is predicted according to the number of initial groups to obtain a number of minimum cleaning times. The initial number of groups whose minimum cleaning time is less than the preset cleaning time is set as the number of qualified groups, resulting in multiple qualified group numbers. The qualified group number with the largest value is selected as the optimal group number.

[0091] Furthermore, in one embodiment, based on the current number of working devices and device attributes, the cleaning time is predicted according to the several initial grouping quantities to obtain several minimum cleaning times, including: Randomly select any initial group number from the plurality of initial group numbers as the first initial group number; Based on the current number of working devices and device attributes, the number of repeated cleaning cycles for each device is randomly divided according to the first initial grouping number, generating multiple first device working schemes. The cleaning time is predicted for each of the multiple first equipment working schemes to obtain multiple first predicted cleaning times. The first predicted cleaning time with the shortest time is selected as the first minimum cleaning time and added to the multiple minimum cleaning times.

[0092] In one embodiment, parameter output module 13 is specifically used for: Randomly select a first silicon-based epitaxial wafer group from the plurality of silicon-based epitaxial wafer groups; The cleaning parameter adjustment space of the cleaning equipment in the metal-particle removal process is obtained. Multiple initial cleaning parameters are randomly selected within the cleaning parameter adjustment space. The cleaning equipment is a fully automatic single-plate cleaning machine. The cleaning parameters include chemical agent concentration, reaction temperature, reaction time, and mega-sound energy. A cleaning quality prediction plugin for silicon-based epitaxial wafers is constructed, wherein a deep learning model is trained to convergence using a sample metal distribution information set, a sample particle distribution information set, a sample cleaning parameter set, and a sample cleaning quality coefficient set to generate the cleaning quality prediction plugin. With the goal of maximizing the overall cleaning quality within the group, the cleaning quality prediction plugin and multiple initial cleaning parameters are used to optimize the cleaning parameters of the first silicon-based epitaxial wafer group, output the first adaptive cleaning parameter, and add it to the multiple adaptive cleaning parameters.

[0093] Specifically, with the goal of maximizing the overall cleaning quality within the group, the cleaning quality prediction plugin and multiple initial cleaning parameters are used to optimize the cleaning parameters of the first silicon-based epitaxial wafer group, outputting the first adaptive cleaning parameters, including: Using the cleaning quality prediction plugin, the cleaning quality is predicted for the multiple initial cleaning parameters based on the first metal distribution information set and the first particle distribution information set of the first silicon-based epitaxial wafer group, and multiple sets of predicted cleaning quality coefficients are output. The predicted cleaning quality coefficients included in the multiple predicted cleaning quality coefficient sets are summed to obtain multiple predicted overall cleaning quality coefficients; The initial cleaning parameters are set as the initial solution, and multiple initial solutions are arranged in descending order according to the predicted overall cleaning quality coefficient to generate an initial solution sequence. The first solution of the initial solution sequence is selected as the optimal solution, the last solution is selected as the poor solution, and the remaining solutions are selected as the inferior solutions, thus obtaining the optimal solution, the poor solution, and multiple inferior solutions; The cleaning quality deviations between the multiple inferior solutions and the superior and inferior solutions are evaluated respectively, and a first optimization strategy is formulated based on the cleaning quality deviations. According to the first optimization strategy, the multiple inferior solutions are adjusted according to the preset optimization adjustment step size to obtain multiple updated inferior solutions. The superior solutions, inferior solutions and multiple updated inferior solutions are reordered according to the predicted overall cleaning quality coefficient from large to small to generate the first updated solution sequence. Following an iterative optimization mechanism of selecting superior, inferior, and poor solutions, updating the optimization strategy, optimizing inferior solutions, and updating the initial solution sequence, the cleaning parameters are optimized based on the first updated solution sequence until a preset number of convergences is reached. The superior solutions in the current updated solution sequence are then output as the first matching cleaning parameters.

[0094] Further, the cleaning quality deviations between the plurality of inferior solutions and the superior and inferior solutions are evaluated respectively, and a first optimization strategy is formulated based on the cleaning quality deviations, including: The average of the multiple predicted overall cleaning quality coefficients of the multiple suboptimal solutions is calculated as the comprehensive cleaning quality coefficient of the suboptimal solutions. The predicted overall cleaning quality coefficient of the optimal solution is used as the cleaning quality coefficient of the optimal solution, and the predicted overall cleaning quality coefficient of the poor solution is used as the cleaning quality coefficient of the poor solution. Calculate the absolute values ​​of the deviations between the comprehensive cleaning quality coefficient of the inferior solution and the cleaning quality coefficients of the superior and inferior solutions, respectively, to obtain the cleaning quality deviations of the superior and inferior solutions. If the cleaning quality deviation of the optimal solution is greater than or equal to the cleaning quality deviation of the poor solution, the direction that approaches the optimal solution will be used as the first optimization strategy. If the cleaning quality deviation of the optimal solution is less than the cleaning quality deviation of the poor solution, the direction away from the poor solution will be used as the first optimization strategy.

[0095] In one embodiment, the sample processing module 14 is specifically used for: The equipment operation scheme corresponding to the minimum cleaning time of the optimal group number is used as the adapted equipment operation scheme. According to the working scheme of the adapter equipment and the multiple adapter cleaning parameters, the multiple silicon-based epitaxial wafer groups are subjected to batch cleaning treatment by the cleaning equipment.

[0096] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0097] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0098] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. A method for cleaning silicon-based epitaxial wafers using feature analysis, characterized in that, The methods include: The state of several silicon-based epitaxial wafers to be cleaned is monitored to obtain several metal distribution information and several particle distribution information. Constrained by a preset cleaning time, the silicon-based epitaxial wafers are grouped according to the current cleaning equipment status, the distribution information of the several metals, and the distribution information of the several particles, to generate multiple silicon-based epitaxial wafer groups. With the goal of maximizing the overall cleaning quality within the group, the cleaning parameters for the metal-particle removal process are optimized for each of the multiple silicon-based epitaxial wafer groups, and multiple suitable cleaning parameters are output. According to the aforementioned multiple adaptive cleaning parameters, the multiple silicon-based epitaxial wafer groups are subjected to batch cleaning processes using cleaning equipment.

2. The silicon-based epitaxial wafer cleaning method for feature analysis according to claim 1, characterized in that, The state of several silicon-based epitaxial wafers to be cleaned is monitored to obtain several metal distribution information and several particle distribution information, including: The surface of the silicon-based epitaxial wafers was scanned using a total reflection X-ray fluorescence spectrometer to obtain the content and spatial distribution of metal elements, and feature extraction was performed to generate metal distribution information, resulting in several metal distribution information, wherein the metal distribution information includes at least the total amount of metal, the proportion of key metals, and the two-dimensional distribution of key metals. A laser scanning particle detection device is used to detect the plurality of silicon-based epitaxial wafers respectively, to obtain the number of particles and the spatial coordinates of particles at different size thresholds, and to extract features to generate particle distribution information, thereby obtaining a plurality of particle distribution information, wherein the particle distribution information includes at least the total number of particles, particle size distribution and spatial uniformity.

3. The silicon-based epitaxial wafer cleaning method for feature analysis according to claim 1, characterized in that, Constrained by a preset cleaning time, and based on the current cleaning equipment status, the distribution information of several metals, and the distribution information of several particles, the several silicon-based epitaxial wafers are grouped to generate multiple silicon-based epitaxial wafer groups, including: Read the current cleaning equipment status of the cleaning equipment, wherein the current cleaning equipment status is the number of currently working equipment and equipment attributes, and the equipment attributes include the maximum number of silicon-based epitaxial wafers that can be cleaned simultaneously; With a constraint that the cleaning time is less than the preset cleaning time, the grouping quantity analysis is performed based on the number of currently working devices and device attributes, and the optimal grouping quantity is output. Feature extraction, numerical representation, and feature fusion are performed on the aforementioned metal distribution information and particle distribution information to generate several silicon-based epitaxial wafer feature vectors. Using the optimal number of groups as the number of cluster centers, and based on the feature vectors of the silicon-based epitaxial wafers, the K-means algorithm is used to group the silicon-based epitaxial wafers to generate multiple silicon-based epitaxial wafer groups.

4. The silicon-based epitaxial wafer cleaning method for feature analysis according to claim 3, characterized in that, Constrained by a minimum cleaning time, the optimal number of groups is analyzed based on the current number of working devices and their attributes, and the number of groups is output, including: Based on the aforementioned silicon-based epitaxial wafers, grouping and enumerating are performed to generate several initial group quantities; Based on the current number of working devices and device attributes, the cleaning time is predicted according to the number of initial groups to obtain a number of minimum cleaning times. The initial number of groups whose minimum cleaning time is less than the preset cleaning time is set as the number of qualified groups, resulting in multiple qualified group numbers. The qualified group number with the largest value is selected as the optimal group number.

5. The silicon-based epitaxial wafer cleaning method for feature analysis according to claim 4, characterized in that, Based on the current number of working devices and their attributes, the cleaning time is predicted according to the several initial grouping quantities to obtain several minimum cleaning times, including: Randomly select any initial group number from the plurality of initial group numbers as the first initial group number; Based on the current number of working devices and device attributes, the number of repeated cleaning cycles for each device is randomly divided according to the first initial grouping number, generating multiple first device working schemes. The cleaning time is predicted for each of the multiple first equipment working schemes to obtain multiple first predicted cleaning times. The first predicted cleaning time with the shortest time is selected as the first minimum cleaning time and added to the multiple minimum cleaning times.

6. The silicon-based epitaxial wafer cleaning method for feature analysis according to claim 5, characterized in that, With the goal of maximizing the overall cleaning quality within the group, the cleaning parameters for the metal-particle removal process are optimized for each of the multiple silicon-based epitaxial wafer groups, and multiple suitable cleaning parameters are output, including: Randomly select a first silicon-based epitaxial wafer group from the plurality of silicon-based epitaxial wafer groups; The cleaning parameter adjustment space of the cleaning equipment in the metal-particle removal process is obtained. Multiple initial cleaning parameters are randomly selected within the cleaning parameter adjustment space. The cleaning equipment is a fully automatic single-plate cleaning machine. The cleaning parameters include chemical agent concentration, reaction temperature, reaction time, and mega-sound energy. A cleaning quality prediction plugin for silicon-based epitaxial wafers is constructed, wherein a deep learning model is trained to convergence using a sample metal distribution information set, a sample particle distribution information set, a sample cleaning parameter set, and a sample cleaning quality coefficient set to generate the cleaning quality prediction plugin. With the goal of maximizing the overall cleaning quality within the group, the cleaning quality prediction plugin and multiple initial cleaning parameters are used to optimize the cleaning parameters of the first silicon-based epitaxial wafer group, output the first adaptive cleaning parameter, and add it to the multiple adaptive cleaning parameters.

7. The silicon-based epitaxial wafer cleaning method for feature analysis according to claim 6, characterized in that, With the goal of maximizing the overall cleaning quality within the group, the cleaning quality prediction plugin and multiple initial cleaning parameters are used to optimize the cleaning parameters of the first silicon-based epitaxial wafer group, and the first adaptive cleaning parameters are output, including: Using the cleaning quality prediction plugin, the cleaning quality is predicted for the multiple initial cleaning parameters based on the first metal distribution information set and the first particle distribution information set of the first silicon-based epitaxial wafer group, and multiple sets of predicted cleaning quality coefficients are output. The predicted cleaning quality coefficients included in the multiple predicted cleaning quality coefficient sets are summed to obtain multiple predicted overall cleaning quality coefficients; The initial cleaning parameters are set as the initial solution, and multiple initial solutions are arranged in descending order according to the predicted overall cleaning quality coefficient to generate an initial solution sequence. The first solution of the initial solution sequence is selected as the optimal solution, the last solution is selected as the poor solution, and the remaining solutions are selected as the inferior solutions, thus obtaining the optimal solution, the poor solution, and multiple inferior solutions; The cleaning quality deviations between the multiple inferior solutions and the superior and inferior solutions are evaluated respectively, and a first optimization strategy is formulated based on the cleaning quality deviations. According to the first optimization strategy, the multiple inferior solutions are adjusted according to the preset optimization adjustment step size to obtain multiple updated inferior solutions. The superior solutions, inferior solutions and multiple updated inferior solutions are reordered according to the predicted overall cleaning quality coefficient from large to small to generate the first updated solution sequence. Following an iterative optimization mechanism of selecting superior, inferior, and poor solutions, updating the optimization strategy, optimizing inferior solutions, and updating the initial solution sequence, the cleaning parameters are optimized based on the first updated solution sequence until a preset number of convergences is reached. The superior solutions in the current updated solution sequence are then output as the first matching cleaning parameters.

8. The silicon-based epitaxial wafer cleaning method for feature analysis according to claim 7, characterized in that, The cleaning quality deviations between the multiple suboptimal solutions and the optimal and poor solutions are evaluated respectively, and a first optimization strategy is formulated based on the cleaning quality deviations, including: The average of the multiple predicted overall cleaning quality coefficients of the multiple suboptimal solutions is calculated as the comprehensive cleaning quality coefficient of the suboptimal solutions. The predicted overall cleaning quality coefficient of the optimal solution is used as the cleaning quality coefficient of the optimal solution, and the predicted overall cleaning quality coefficient of the poor solution is used as the cleaning quality coefficient of the poor solution. Calculate the absolute values ​​of the deviations between the comprehensive cleaning quality coefficient of the inferior solution and the cleaning quality coefficients of the superior and inferior solutions, respectively, to obtain the cleaning quality deviations of the superior and inferior solutions. If the cleaning quality deviation of the optimal solution is greater than or equal to the cleaning quality deviation of the poor solution, the direction that approaches the optimal solution will be used as the first optimization strategy. If the cleaning quality deviation of the optimal solution is less than the cleaning quality deviation of the poor solution, the direction away from the poor solution will be used as the first optimization strategy.

9. The silicon-based epitaxial wafer cleaning method for feature analysis according to claim 5, characterized in that, According to the aforementioned multiple adaptive cleaning parameters, the multiple silicon-based epitaxial wafer groups are subjected to batch cleaning processes using cleaning equipment, including: The equipment operation scheme corresponding to the minimum cleaning time of the optimal group number is used as the adapted equipment operation scheme. According to the working scheme of the adapter equipment and the multiple adapter cleaning parameters, the multiple silicon-based epitaxial wafer groups are subjected to batch cleaning treatment by the cleaning equipment.

10. A silicon-based epitaxial wafer cleaning apparatus for feature analysis, characterized in that, A silicon-based epitaxial wafer cleaning method for performing a feature analysis according to any one of claims 1-9 includes: The information acquisition module is used to detect the status of several silicon-based epitaxial wafers to be cleaned and to acquire several metal distribution information and several particle distribution information. The sample grouping module is used to group the silicon-based epitaxial wafers according to the current cleaning equipment status, the distribution information of the metals and the distribution information of the particles, with a preset cleaning time as a constraint, to generate multiple silicon-based epitaxial wafer groups. The parameter output module is used to optimize the cleaning parameters for the metal-particle removal process of the multiple silicon-based epitaxial wafer groups with the goal of maximizing the overall cleaning quality within the group, and output multiple suitable cleaning parameters. The sample processing module is used to perform batch cleaning of the multiple silicon-based epitaxial wafer groups according to the multiple adapted cleaning parameters and through cleaning equipment.

Images