Intelligent semiconductor wafer spraying control method and system

Abstract

The invention provides an intelligent semiconductor wafer spraying control method and system, and the method comprises the steps: obtaining a recommended cleaning time range of a wafer from a preset spraying time-wafer attribute association database according to the material and surface roughness of the wafer; the spraying parameters, obtained in the first three steps, of the sub-areas are integrated, a wafer surface cleaning scheme is formed, the cleaning scheme comprises the nozzle layout, the nozzle number, the spraying liquid components and the temperature of all the sub-areas, and the cleaning scheme is transmitted to a spraying control unit; after cleaning is completed, the surface image of the wafer is compared with a preset standard image, the similarity between the surface image of the wafer and the preset standard image is calculated, if the similarity is lower than a preset threshold value, the serial number and cleaning parameters of the wafer are recorded in an abnormal cleaning log, meanwhile, the wafer is conveyed to an unqualified product cache region, and the unqualified product cache region is used for cleaning the wafer. By means of spraying control, the cleaning efficiency and quality are improved, resource waste is reduced, and an advanced cleaning solution is provided for semiconductor manufacturing.

Description

Technical Field

[0001] The present invention relates to the field of information technology, in particular to the field of semiconductor wafers, and specifically to an intelligent semiconductor wafer spray control method and system. Background Art

[0002] During the wafer cleaning process, there is a technical contradiction between real-time monitoring of spray pressure and flow and automatic adjustment of spray parameters according to the degree of wafer surface contamination. On the one hand, in order to ensure the cleaning effect, it is necessary to use clustering and decision tree algorithms to divide sub-areas according to the contamination distribution and graphic structure characteristics of the wafer surface, and select the best spray parameters for each sub-area. This requires the system to be able to flexibly adjust parameters such as spray pressure, flow and angle. On the other hand, in order to ensure the stability and controllability of the cleaning process, it is necessary to monitor the actual operating parameters of the spray device in real time and compare them with the preset values. When the monitored values ​​deviate, the system needs to dynamically adjust the nozzle opening to ensure that the parameters meet the requirements. However, frequent adjustment of spray parameters may affect the continuity and uniformity of cleaning, and even lead to insufficient cleaning of certain areas. How to maintain the relative stability of spray parameters while ensuring the cleaning effect has become a technical problem that needs to be solved urgently. In addition, how to organically combine real-time monitoring data with contamination analysis results to establish a dynamically optimized cleaning parameter model to adapt to different types and degrees of wafer contamination is also an important challenge facing the system. Summary of the invention

[0003] The present invention provides an intelligent semiconductor wafer spray control method, which mainly includes:

[0004] Obtaining a surface contamination distribution map of the wafer to be cleaned, dividing the contaminated area on the wafer surface by a clustering algorithm to obtain a number of sub-areas with different contamination degrees, and obtaining corresponding spray pressure and flow parameters from a preset spray parameter database for each sub-area;

[0005] Acquire the graphic structure data of the wafer surface, partition the wafer surface using a decision tree algorithm to obtain a plurality of sub-regions with different graphic structure complexities, and acquire corresponding spray angle parameters from the preset spray parameter database for each of the sub-regions;

[0006] According to the material and surface roughness of the wafer, a recommended cleaning time range for the wafer is obtained from a preset spray time-wafer attribute association database;

[0007] Integrate the sub-region spray parameters obtained in the first three steps to form a wafer surface cleaning plan, the cleaning plan including the nozzle layout, the number of nozzles, the spray liquid composition and the temperature of each sub-region, and transmit the cleaning plan to the spray control unit;

[0008] The spray control unit controls the nozzle of the spray device to move on the surface of the wafer at a preset trajectory and time interval according to the received cleaning scheme, and monitors the pressure and flow parameters of each nozzle in real time. If the monitored value deviates from the set value of the cleaning scheme, the nozzle opening is dynamically adjusted;

[0009] During the wafer cleaning process, the surface image of the wafer is collected in real time by a machine vision system, the image is processed and analyzed, and the amount of residual pollutants on the surface of the wafer is calculated. If the residual amount is higher than a preset threshold, the wafer is re-transferred to the spray device for further cleaning.

[0010] After cleaning is completed, the wafer surface image is compared with a preset standard image to calculate the similarity between the two. If the similarity is lower than a preset threshold, the number and cleaning parameters of the wafer are recorded in the abnormal cleaning log, and the wafer is transferred to the defective product buffer area.

[0011] For the wafers that have passed the cleaning, their actual cleaning parameters are uploaded to the spray parameter optimization module. The optimization module trains the historical accumulated cleaning data based on the support vector machine algorithm to obtain a dynamically updated spray parameter-cleaning effect mapping model, which is used to optimize and predict the subsequent cleaning parameters of new wafers.

[0012] The present invention provides an intelligent semiconductor wafer spray control system, which mainly includes:

[0013] The contamination analysis module is used to obtain the contamination distribution on the wafer surface and divide it into sub-areas;

[0014] Graphic structure analysis module, used to obtain the graphic structure of the wafer surface and divide it into sub-areas;

[0015] A cleaning parameter acquisition module is used to obtain a recommended cleaning time range based on wafer properties;

[0016] A cleaning plan generation module is used to integrate parameters to form a wafer surface cleaning plan;

[0017] A cleaning control module is used to control the spray device to perform cleaning and monitor and adjust in real time;

[0018] The quality detection module is used to monitor the cleaning effect in real time and perform subsequent processing.

[0019] The technical solution provided by the embodiment of the present invention may include the following beneficial effects:

[0020] The present invention discloses an intelligent semiconductor wafer spray control method. The method obtains the contamination distribution map and graphic structure data of the wafer surface, partitions the wafer surface by using clustering and decision tree algorithms, obtains corresponding spray parameters from a preset database, determines the recommended cleaning time according to the wafer material and surface roughness, and integrates to form a cleaning scheme. During the cleaning process, the spray control unit controls the nozzle movement and adjusts the parameters in real time according to the scheme, the machine vision system monitors the residual amount of pollutants, and judges the eligibility through image comparison after cleaning, and uses the cleaning data for optimization model training. The present invention realizes the intelligent and refined control of the wafer cleaning process, and the adjustable spray design ensures the spray uniformity and reliability, improves the cleaning efficiency and quality, reduces the waste of resources, and provides an advanced cleaning solution for semiconductor manufacturing. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 The present invention is a flow chart of an intelligent semiconductor wafer spray control method.

[0022] Figure 2 The figure is a schematic diagram of the framework structure of an intelligent semiconductor wafer spray control system of the present invention. DETAILED DESCRIPTION

[0023] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0024] like Figure 1-Figure 2 In this embodiment, a smart semiconductor wafer spray control method may specifically include:

[0025] Step S101, obtaining a surface contamination distribution map of the wafer to be cleaned, dividing the contaminated area on the wafer surface by a clustering algorithm to obtain several sub-areas with different contamination degrees, and obtaining corresponding spray pressure and flow parameters from a preset spray parameter database for each sub-area.

[0026] Acquire a surface contamination distribution image of a wafer to be cleaned; preprocess the contamination distribution image to obtain a preprocessed contamination distribution image; perform cluster analysis on the preprocessed contamination distribution image using a K-means clustering algorithm to obtain a number of contaminated sub-regions with different contamination degrees; acquire a preset spray parameter database, wherein the spray parameter database stores spray pressure and flow parameters corresponding to different contamination degrees; acquire the spray pressure and flow parameters corresponding to each contaminated sub-region from the spray parameter database according to the contamination degrees of the several contaminated sub-regions; place the wafer in a cleaning device, and control the spray head of the cleaning device to perform differential cleaning on each contaminated sub-region according to the acquired spray pressure and flow parameters to obtain a cleaned wafer; detect the surface of the cleaned wafer to determine whether it meets the preset cleanliness requirement; if the cleanliness requirement is not met, put the cleaned wafer back into the cleaning device for cleaning; if the cleanliness requirement is met, send the cleaned wafer to the next process.

[0027] Specifically, obtain the surface contamination distribution image of the wafer to be cleaned. This can be obtained through dark-field microscope imaging technology. The working principle of the dark-field microscope is to illuminate the light source obliquely onto the sample surface, so that only scattered light can enter the objective lens, so that the contaminants appear bright against a dark background, and the tiny contaminant particles on the surface of the wafer can be clearly displayed. For example, using a 500nm LED light source and a 500x magnification dark-field microscope to image a 12-inch silicon wafer, a high-resolution contamination distribution image can be obtained. The higher the brightness in the image, the higher the concentration of contaminants. The reason for using a dark-field microscope is that it has a better imaging effect on tiny particles and can more accurately reflect the contamination status of the wafer surface.

[0028] Preprocess the pollution distribution image to remove noise and interference information. Image smoothing and filtering techniques can be used to remove noise in the image. For example, using a Gaussian filter to smooth the acquired wafer surface pollution distribution image can effectively remove random noise in the image. Morphological operations, such as corrosion and dilation operations, can also be used to remove interference information in the image, such as dust or scratches. Assume that a small area on the wafer surface is mistakenly identified as a contaminant, but it is actually caused by light reflection. The misidentified area can be removed by corrosion operation. The purpose of image preprocessing is to improve the accuracy of subsequent clustering analysis. The K-means clustering algorithm is used to perform cluster analysis on the preprocessed pollution distribution image, and the contaminated area is divided into several sub-areas with different degrees of contamination. The K-means clustering algorithm is an unsupervised learning algorithm that divides data points into K clusters, and the center point of each cluster is called the centroid. The brightness value of each pixel in the preprocessed image is used as a feature vector, and the pixels are divided into K clusters, each cluster representing a sub-area with different degrees of contamination. For example, if the K value is set to 3, the contaminated area on the wafer surface can be divided into three sub-areas: lightly contaminated, moderately contaminated, and heavily contaminated. The K-means algorithm is chosen because it is simple and efficient and can effectively classify contaminated areas. According to the degree of contamination of each sub-area, the corresponding spray pressure and flow parameters are obtained from the preset spray parameter database.

[0029] The preset spray parameter database stores the optimal spray pressure and flow parameters corresponding to different degrees of contamination. For example, the spray pressure corresponding to the lightly contaminated area is 1MPa and the flow rate is 1L / min; the spray pressure corresponding to the moderately contaminated area is 2MPa and the flow rate is 2L / min; the spray pressure corresponding to the heavily contaminated area is 3MPa and the flow rate is 3L / min. The parameters in the database are obtained through a large amount of experimental data to ensure the best cleaning effect. The purpose of establishing this database is to achieve differentiated cleaning, improve cleaning efficiency and reduce damage to the wafer. The wafer is placed in the cleaning equipment, and the spray head is controlled to perform differentiated cleaning on different contaminated sub-areas according to the obtained spray parameters. For example, for the three contaminated sub-areas divided, the spray head will apply a pressure of 3MPa and a flow rate of 3L / min to the heavily contaminated area for cleaning according to the preset parameters, a pressure of 2MPa and a flow rate of 2L / min to the moderately contaminated area, and a pressure of 1MPa and a flow rate of 1L / min to the lightly contaminated area. This differentiated cleaning method can accurately clean areas with different levels of contamination, avoid wafer damage caused by excessive cleaning, and improve cleaning efficiency.

[0030] After cleaning, the wafer surface is inspected to determine whether it meets the preset cleanliness requirements. A dark field microscope or other surface inspection equipment can be used to inspect the cleaned wafers, for example, to detect the number and size of residual particles on the wafer surface and compare them with the preset cleanliness standards. The cleanliness standard can be set to less than 10 particles / square centimeter and the maximum particle size is less than 1 micron. The reason for the inspection is to ensure that the cleaning effect meets the requirements and prevent unqualified wafers from entering the next process. If the cleanliness requirements are not met, return and re-clean until the requirements are met. When the test results show that the cleanliness of the wafer surface does not meet the requirements, for example, there are still residual particles in the heavily contaminated area, it is necessary to readjust the spray parameters, such as increasing the spray pressure or extending the cleaning time, and then clean again. This cycle will continue until the wafer surface meets the preset cleanliness requirements. If the requirements are met, the cleaned wafer is sent to the next process. When the test results show that the cleanliness of the wafer surface has met the requirements, the cleaned wafer can be sent to the next process, such as photolithography or etching. This ensures the smooth progress of subsequent processes and ultimately produces high-quality chips.

[0031] Step S102, obtaining the graphic structure data of the wafer surface, partitioning the wafer surface using a decision tree algorithm to obtain a number of sub-regions with different graphic structure complexities, and obtaining corresponding spray angle parameters from the preset spray parameter database for each sub-region.

[0032] The graphic structure data of the wafer surface is obtained, and the graphic structure data is preprocessed to extract the graphic structure features. According to the extracted graphic structure features, the decision tree algorithm is used to partition the wafer surface to obtain several sub-regions with different graphic structure complexity. For each sub-region, according to the complexity of its graphic structure, the corresponding spray angle parameter is obtained from the preset spray parameter database. According to the obtained spray angle parameter, the spray angle of each sub-region is determined to obtain the spray angle data of each sub-region on the wafer surface. The spray angle data of each sub-region on the wafer surface is input into the spray equipment, and the spray equipment is controlled to perform differentiated spray treatment on each sub-region on the wafer surface. During the spray treatment process, the image data of the wafer surface is collected in real time, and the image data is analyzed to determine whether the spray effect meets the expectations. If the spray effect does not meet the expectations, the spray angle parameter is adjusted according to the image analysis results, and the adjusted parameter is updated to the spray parameter database to optimize the subsequent wafer spray treatment process.

[0033] Specifically, the graphic structure data of the wafer surface can be obtained through optical microscopes, scanning electron microscopes and other equipment. For example, by scanning the wafer surface with a scanning electron microscope, a high-resolution surface morphology image can be obtained, which contains information on various graphic structures on the wafer surface, such as lines, grooves, holes, etc. The reason for obtaining graphic structure data is that different graphic structures have different effects on the flow and coverage of the cleaning liquid, thereby affecting the cleaning effect. The acquired graphic structure data usually contains a lot of noise and redundant information and needs to be preprocessed. For example, image smoothing and filtering techniques can be used to remove noise in the image, such as Gaussian filtering or median filtering. Edge detection algorithms can also be used to extract edge information of the graphic structure, such as Canny operator or Sobel operator. The purpose of preprocessing is to improve the accuracy and efficiency of subsequent feature extraction. Assume that there is a long and thin scratch on the surface of the wafer, which is noise information and needs to be removed by image processing technology. Graphic structure features are extracted from the preprocessed graphic structure data. For example, features such as density, complexity, and directionality of the graphic structure can be extracted. The graphic structure density can be defined as the proportion of the graphic structure per unit area. Complexity can be defined as the total length of the edges of the graph structure. Directionality can be defined as the main direction of the graph structure. These features can be used to describe the complexity of the graph structure. For example, an area with dense lines and complex directions has high complexity. The purpose of extracting these features is to provide a basis for subsequent partitioning.

[0034] The decision tree algorithm is used to partition the wafer surface. The decision tree algorithm is a classification algorithm based on a tree structure. It can divide the wafer surface into several sub-regions with different graphic structure complexities according to the characteristics of the graphic structure. For example, the wafer surface can be divided into low-complexity regions, medium-complexity regions, and high-complexity regions according to the density and complexity of the graphic structure. The decision tree algorithm is chosen because it can effectively process multidimensional feature data and has good interpretability. Assuming that one part of the wafer surface is dense circuit graphics and the other part is relatively sparse connection lines, the decision tree algorithm can divide these two parts into different regions according to the graphic density. For each sub-region, according to the complexity of its graphic structure, the corresponding spray angle parameter is obtained from the preset spray parameter database. The preset spray parameter database stores the optimal spray angle parameters corresponding to different graphic structure complexities. For example, the spray angle corresponding to the low-complexity region is vertically downward, the spray angle corresponding to the medium-complexity region is 45 degrees, and the spray angle corresponding to the high-complexity region is 30 degrees. The reason for setting different spray angles is that different angles can change the flow and coverage of the cleaning liquid on the wafer surface, thereby improving the cleaning efficiency. The parameters in the database are obtained through a large amount of experimental data to ensure the best cleaning effect. For highly complex areas, using a smaller spray angle can allow the cleaning liquid to better penetrate into small grooves and holes.

[0035] According to the acquired spray angle parameters, the spray angle of each sub-area is determined, and the spray angle data of each sub-area on the wafer surface is obtained. For example, for the three divided sub-areas, the corresponding spray angles are vertical downward, 45 degrees and 30 degrees respectively. These angle data are stored in a data structure for subsequent use by the spray equipment. The spray angle data of each sub-area on the wafer surface is input into the spray equipment, and the spray equipment is controlled to perform differential spray treatment on each sub-area on the wafer surface. The spray equipment adjusts the position and angle of the spray head according to the input spray angle data, and performs differential spraying on different sub-areas. For example, for a highly complex area, the spray head will spray at an angle of 30 degrees. This differentiated spraying method can accurately clean areas with different graphic structure complexity, improve cleaning efficiency and reduce damage to the wafer. During the spraying process, the image data of the wafer surface is collected in real time. A high-speed camera or other imaging device can be used to collect image data of the wafer surface during the spraying process in real time. These image data can be used to monitor the cleaning process and evaluate the spraying effect. The image data is analyzed to determine whether the spray effect meets expectations. For example, the spray effect can be determined by analyzing the number and size of residual contaminants in the image. When the number of residual contaminants exceeds the preset threshold, it is considered that the spray effect does not meet expectations. If the spray effect does not meet expectations, the spray angle parameters are adjusted according to the image analysis results. For example, when it is found that the cleaning effect of a high-complexity area is not ideal, you can try to reduce the spray angle, such as from 30 degrees to 20 degrees, or increase the spray time. The adjusted parameters are updated to the spray parameter database to optimize the subsequent wafer spray process. Updating the adjusted spray angle parameters to the database can enable subsequent wafers with the same graphic structure to obtain better cleaning effects. This is a data-driven optimization method that can continuously improve the cleaning process.

[0036] Step S103 , obtaining a recommended cleaning time range for the wafer from a preset spray time-wafer attribute association database according to the material and surface roughness of the wafer.

[0037] Obtain the material information and surface roughness value of the wafer and use them as query conditions. Access the preset spray time-wafer attribute association database and perform matching queries based on the material and roughness value. If a matching record is found in the association database, extract the corresponding recommended cleaning time range. Determine the target cleaning time for the wafer based on the obtained recommended cleaning time range. Pass the target cleaning time to the spray control module and perform a cleaning process on the wafer for the corresponding length of time. After cleaning, perform a roughness test on the wafer surface and obtain the roughness value after cleaning. Compare the roughness value after cleaning with the preset threshold to determine whether the cleaning effect meets the standard, so as to determine whether secondary cleaning or adjustment of cleaning parameters is required.

[0038] Step S104, integrating the sub-area spray parameters obtained in the first three steps to form a wafer surface cleaning plan, the cleaning plan including the nozzle layout, number of nozzles, spray liquid composition and temperature of each sub-area, and transmitting the cleaning plan to the spray control unit.

[0039] The contamination degree data and material property data of the wafer surface are obtained, and the spray parameters of each sub-area of ​​the wafer surface are determined according to the contamination degree data and material property data, wherein the spray parameters include nozzle layout, number of nozzles, spray liquid composition and temperature; the spray parameters of each sub-area are integrated to obtain a wafer surface cleaning plan; a machine learning algorithm is used to optimize the parameters in the cleaning plan according to preset historical cleaning data and current wafer surface condition data to obtain an optimized cleaning plan; the optimized cleaning plan is converted into a control instruction, and the control instruction is transmitted to a spray control unit; during the cleaning process, the real-time spray parameters and wafer surface condition data of each sub-area are obtained; if the real-time spray parameters and wafer surface condition data do not conform to the preset range, the cleaning plan is dynamically adjusted; after the cleaning is completed, the cleaning effect data of the wafer surface is obtained, and the cleaning effect data is input into the machine learning algorithm, the machine learning algorithm is trained, and the optimized machine learning algorithm is obtained for the optimization of subsequent cleaning plans.

[0040] In step S105, the spray control unit controls the nozzle of the spray device to move on the surface of the wafer at a preset trajectory and time interval according to the received cleaning scheme, and monitors the pressure and flow parameters of each nozzle in real time. If the monitored value deviates from the set value of the cleaning scheme, the nozzle opening is dynamically adjusted.

[0041] S1051, the spray control unit receives the cleaning scheme, obtains the preset nozzle trajectory and time interval parameters, and controls the nozzle to move on the wafer surface according to the parameters. S1052, during the movement of the nozzle, the pressure and flow data of the nozzle are collected in real time, and the collected real-time data are compared with the pressure and flow thresholds set in the cleaning scheme. S1053, if the comparison result shows that the real-time pressure or flow data deviates from the set threshold, it is determined to be an abnormal situation and the nozzle opening adjustment mechanism is triggered. S1054, according to the direction and amount of pressure deviation, the nozzle opening adjustment value is dynamically calculated, and the nozzle opening is adjusted through the control signal to restore the pressure to the set range. S1055, according to the direction and amount of flow deviation, the nozzle opening adjustment value is dynamically calculated, and the nozzle opening is adjusted through the control signal to restore the flow to the set range. In a cleaning cycle, steps S1052-S1055 are continuously executed in a loop until the cleaning cycle ends, and the cleaning operation on the wafer surface is completed. The pressure and flow monitoring data in a cleaning cycle are recorded and counted to form a cleaning quality report for evaluating the cleaning effect and optimizing the cleaning scheme.

[0042] Specifically, the spray control unit plays a vital role in the wafer surface cleaning process. It receives instructions from the cleaning plan and accurately controls the movement trajectory and time interval of the nozzle to ensure that the cleaning liquid is evenly sprayed on the wafer surface. For example, the cleaning plan stipulates that the nozzle moves in a spiral trajectory with a time interval of 5 seconds between each circle. The spray control unit will accurately control the nozzle to execute according to this plan. The setting of the nozzle trajectory and time interval parameters needs to take into account factors such as the size of the wafer, the distribution of contaminants, and the characteristics of the cleaning liquid. For example, for a wafer with a diameter of 300 mm, there are usually more contaminants in the edge area, so the nozzle can be set to move slower and stay longer in the edge area to ensure the cleaning effect of the edge area.

[0043] During the movement of the nozzle, the pressure and flow data of the nozzle need to be collected in real time. These data reflect the operating status of the spray system and are crucial to ensuring the cleaning quality. For example, a pressure sensor and a flow meter can be used to measure the pressure and flow of the nozzle respectively, and the data can be transmitted to the spray control unit. The collected real-time data needs to be compared with the pressure and flow thresholds set in the cleaning scheme. The setting of the threshold needs to take into account factors such as the characteristics of the cleaning fluid, the model of the nozzle, and the material of the wafer surface. For example, for a highly corrosive cleaning fluid, a lower pressure threshold needs to be set to prevent the wafer surface from being corroded. When the comparison result shows that the real-time pressure or flow data exceeds the set threshold, it indicates that an abnormality has occurred in the spray system and immediate measures need to be taken to adjust it. For example, when the real-time pressure exceeds the threshold of 2MPa, the nozzle opening needs to be reduced to reduce the pressure. The adjustment of the nozzle opening needs to be dynamically calculated according to the direction and amount of the pressure or flow deviation. For example, when the real-time pressure is 1MPa lower than the set value, the nozzle opening needs to be increased to increase the pressure. The specific adjustment value can be calculated by the PID control algorithm to ensure that the pressure or flow can be quickly and stably restored to the set range.

[0044] In a cleaning cycle, data collection, comparison, adjustment and other steps need to be performed in a continuous cycle until the cleaning cycle ends. The setting of the cleaning cycle needs to take into account factors such as the degree of wafer contamination and the efficiency of the cleaning solution. For example, for lightly contaminated wafers, the cleaning cycle can be set to 1 minute, while for heavily contaminated wafers, the cleaning cycle needs to be extended to 3 minutes or longer. After the cleaning cycle is over, the pressure and flow monitoring data within a cycle need to be recorded and counted to form a cleaning quality report. This report can be used to evaluate the cleaning effect and provide a reference for optimizing the cleaning solution. For example, when the cleaning quality report shows that the cleaning pressure in a certain area is consistently lower than the set value, the nozzle layout can be adjusted to increase the number of nozzles in the area to improve the cleaning effect. Through continuous monitoring, adjustment and optimization, a set of intelligent and adaptive wafer surface cleaning solutions can be established to improve production efficiency and product yield.

[0045] Step S106, during the wafer cleaning process, the wafer surface image is collected in real time by a machine vision system, the image is processed and analyzed, and the current residual pollutant amount on the wafer surface is calculated. If the residual amount is higher than a preset threshold, the wafer is re-transferred to the spray device for further cleaning.

[0046] The real-time image of the wafer surface is acquired through the machine vision system and transmitted to the image processing module. In the image processing module, the acquired wafer surface image is preprocessed, including image denoising, enhancement and correction operations to improve the image quality. According to the preprocessed wafer surface image, the image segmentation algorithm is used to segment the image and extract the pollutant area on the wafer surface. For the segmented pollutant area, its area ratio is calculated to obtain the residual pollutant amount on the current wafer surface. The calculated residual pollutant amount is compared with the preset threshold to determine whether the wafer surface meets the cleaning requirements. If the residual pollutant amount is higher than the preset threshold, the conveying device is controlled to re-convey the wafer to the spray device for re-cleaning the wafer. If the residual pollutant amount is lower than the preset threshold, the wafer surface is judged to be qualified for cleaning and the wafer is conveyed to the next process for subsequent processing.

[0047] Specifically, in the wafer manufacturing process, the application of machine vision systems is to monitor the cleanliness of the wafer surface in real time. For example, a high-resolution camera is used to capture images of the wafer surface, which are then transmitted to the image processing module. In the image processing module, the first step is image denoising, which can reduce random noise in the image through software algorithms and improve the usability of the image. Next, the image enhancement operation can adjust the contrast and brightness to make the details in the image clearer for subsequent processing. Image correction is to correct the distortion caused by the camera shooting angle or lighting changes to ensure that the obtained image can truly reflect the state of the wafer. The processed image will be processed using an image segmentation algorithm, which can identify the contaminant area in the image, which usually involves complex edge detection technology to accurately separate the contaminant and the image area of ​​the wafer itself. After identifying the contaminants, the system will calculate the area ratio of these areas. This data is achieved by analyzing the ratio of the contaminated area to the entire wafer surface in the segmented image. When this ratio is higher than the threshold set by the factory, it means that the cleanliness of the wafer surface does not meet the production standards and needs to be re-cleaned. At this time, the control system will instruct the conveyor device to send the wafer back to the spray device for further cleaning to ensure that each wafer meets the cleaning standards required for production.

[0048] After the wafer is cleaned and qualified, that is, the residual contaminants are lower than the preset threshold, the wafer is transferred to the next process. This series of operations not only ensures product quality, but also improves production efficiency, because by accurately controlling the cleaning process, the number of wafers that need to be repeatedly cleaned can be reduced, thus saving costs and time. This automated inspection and cleaning system based on image processing demonstrates the application of high technology in modern semiconductor manufacturing, ensuring high product quality and high production efficiency.

[0049] Step S107, after cleaning is completed, the wafer surface image is compared with a preset standard image to calculate the similarity between the two. If the similarity is lower than a preset threshold, the wafer number and cleaning parameters are recorded in the abnormal cleaning log, and the wafer is transferred to the defective product buffer area.

[0050] The wafer surface image after the wafer cleaning is completed is obtained, converted into a grayscale image, and image enhancement processing is performed to obtain the processed wafer surface image. A standard image matching the current wafer type is obtained from a preset standard image library, and the standard image is subjected to the same image enhancement processing. A feature extraction algorithm, such as SIFT or SURF algorithm, is used to extract key feature points from the processed wafer surface image and the standard image. The wafer surface image is matched with the standard image by a feature point matching algorithm, such as RANSAC algorithm, and the ratio of the number of matching point pairs to the total number of feature points is calculated to obtain a similarity value. The calculated similarity value is compared with the preset similarity threshold. If the similarity value is lower than the threshold, the wafer cleaning quality is judged to be unqualified. If the wafer cleaning quality is unqualified, the wafer number and the corresponding cleaning parameters are recorded in the abnormal cleaning log database, and the wafer is transferred to the unqualified product buffer area to wait for further processing. If the wafer cleaning quality is qualified, the wafer is transferred to the next process for subsequent processing, and the wafer processing status information is updated to track the wafer production progress in real time.

[0051] Specifically, after the wafer is cleaned, its cleaning quality needs to be evaluated. This solution uses an image comparison method based on machine vision to compare the image of the wafer to be tested with the standard image to determine whether the cleaning quality is qualified. First, obtain the image of the wafer surface after cleaning. For example, use a high-resolution industrial camera to shoot the wafer surface to obtain a 1920x1080 pixel color image. Then, convert the color image to a grayscale image for subsequent processing. This can reduce the amount of calculation, and the grayscale image can better reflect the characteristics of the contaminants on the wafer surface. Image enhancement processing is to highlight the characteristics of the wafer surface. For example, histogram equalization is used to improve the image contrast and make the contaminant area more visible. Next, it is necessary to select a standard image that matches the current wafer type from the standard image library. The standard image library pre-stores the ideal cleaning state images of various types of wafers. For example, for a 12-inch silicon wafer, select the corresponding standard image. The standard image should also undergo the same image enhancement processing to ensure that the processing method is consistent with the wafer image to be tested, so as to avoid comparison errors caused by different processing methods. A feature extraction algorithm, such as the SIFT (Scale Invariant Feature Transform) algorithm, is used to extract key feature points from the processed wafer surface image and the standard image. The SIFT algorithm can extract stable feature points in the image that are not affected by image rotation, scaling, and illumination changes. For example, 500 feature points were extracted from the wafer surface image and 600 feature points were extracted from the standard image.

[0052] Then, a feature point matching algorithm, such as the RANSAC (random sampling consensus) algorithm, is used to match the feature points of the wafer surface image with the feature points of the standard image. The RANSAC algorithm can effectively eliminate mismatched points and improve matching accuracy. Assume that 400 feature points are matched. Calculate the ratio of the number of matching point pairs to the total number of feature points to obtain a similarity value. In this example, the similarity value is 400 / (500+600)≈34%. Compare the calculated similarity value with the preset similarity threshold. Assume that the preset similarity threshold is 80%. Since 34% is less than 80%, it is judged that the wafer cleaning quality is unqualified. When the wafer cleaning quality is unqualified, the wafer number (for example, SN123456789) and the corresponding cleaning parameters (for example, cleaning time 5 minutes, cleaning solution concentration 10%) are recorded in the abnormal cleaning log database, so as to analyze the causes of problems in the cleaning process and make improvements.

[0053] At the same time, the wafer is transferred to the defective product buffer area to wait for further processing, such as re-cleaning or scrapping. When the wafer cleaning quality is qualified, the wafer is transferred to the next process (for example, the photolithography process) for subsequent processing. At the same time, the wafer processing status information is updated (for example, cleaning completed, waiting for photolithography) to track the wafer production progress in real time and improve production efficiency.

[0054] Step S108, for the wafers that have passed the cleaning, their actual cleaning parameters are uploaded to the spray parameter optimization module. The optimization module trains the historical accumulated cleaning data based on the support vector machine algorithm to obtain a dynamically updated spray parameter-cleaning effect mapping model, which is used to optimize and predict the subsequent cleaning parameters of new wafers.

[0055] The actual cleaning parameters of the wafers that have passed the cleaning are obtained and uploaded to the spray parameter optimization module. The spray parameter optimization module obtains the historical accumulated cleaning data and uses the support vector machine algorithm to train the cleaning data. Through the training of the support vector machine algorithm, a dynamically updated spray parameter-cleaning effect mapping model is obtained. When a new wafer needs to be cleaned, the initial cleaning parameters of the new wafer are input into the spray parameter-cleaning effect mapping model. The spray parameter-cleaning effect mapping model predicts the optimized cleaning parameters based on the input initial cleaning parameters. The optimized cleaning parameters are applied to the cleaning process of the new wafer and the new wafer is cleaned. After the cleaning is completed, the actual cleaning parameters and cleaning effects of the new wafer are obtained and added to the historical cleaning data for subsequent model updates and optimizations.

[0056] The above embodiment is only one of the preferred implementation modes of the present invention and should not be used to limit the protection scope of the present invention. Any changes or modifications that are made to the main design concept and spirit of the present invention and have no substantive significance, and the technical problems they solve are still consistent with the present invention, should be included in the protection scope of the present invention.

Claims

1. An intelligent semiconductor wafer spray control method, characterized in that: The method comprises: S101: Obtaining a surface contamination distribution map of a wafer to be cleaned, dividing the contaminated area on the wafer surface by using a clustering algorithm to obtain a number of sub-areas with different contamination degrees, and obtaining corresponding spray pressure and flow parameters from a preset spray parameter database for each sub-area; S102: Acquire the graphic structure data of the wafer surface, partition the wafer surface using a decision tree algorithm to obtain a plurality of sub-regions with different graphic structure complexities, and acquire a corresponding spray angle parameter from the preset spray parameter database for each sub-region; S103 obtains a recommended cleaning time range for the wafer from a preset spray time-wafer attribute association database according to the material and surface roughness of the wafer; S104 integrates the sub-region spray parameters obtained in the first three steps to form a wafer surface cleaning plan, the cleaning plan including the nozzle layout, the number of nozzles, the spray liquid composition and the temperature of each sub-region, and transmits the cleaning plan to the spray control unit; S105: the spray control unit controls the nozzle of the spray device to move on the surface of the wafer at a preset trajectory and time interval according to the received cleaning scheme, and monitors the pressure and flow parameters of each nozzle in real time. If the monitored value deviates from the set value of the cleaning scheme, the nozzle opening is dynamically adjusted; S106: During the wafer cleaning process, the machine vision system is used to collect the wafer surface image in real time, the image is processed and analyzed, and the amount of pollutant residue on the current wafer surface is calculated. If the residue is higher than a preset threshold, the wafer is re-transferred to the spray device for further cleaning. After the cleaning is completed, the wafer surface image is compared with a preset standard image to calculate the similarity between the two. If the similarity is lower than a preset threshold, the wafer number and cleaning parameters are recorded in the abnormal cleaning log, and the wafer is transferred to the unqualified product buffer area. S108 For the wafers that have passed the cleaning, their actual cleaning parameters are uploaded to the spray parameter optimization module. The optimization module trains the historical accumulated cleaning data based on the support vector machine algorithm to obtain a dynamically updated spray parameter-cleaning effect mapping model for optimizing and predicting the subsequent cleaning parameters of new wafers.

2. The method according to claim 1, characterized in that The step S101 includes: Acquire a surface contamination distribution image of the wafer to be cleaned; Preprocessing the pollution distribution image to obtain a preprocessed pollution distribution image; Using K-means clustering algorithm to perform cluster analysis on the preprocessed pollution distribution image to obtain a number of pollution sub-areas with different pollution degrees; Acquire a preset spray parameter database, wherein the spray parameter database stores spray pressure and flow parameters corresponding to different pollution degrees; According to the pollution degree of the plurality of pollution sub-areas, the spray pressure and flow rate parameters corresponding to each pollution sub-area are obtained from the spray parameter database; Placing the wafer in a cleaning device, and controlling the spray head of the cleaning device to perform differential cleaning on each contaminated sub-area according to the acquired spray pressure and flow rate parameters, to obtain a cleaned wafer; Inspecting the surface of the cleaned wafer to determine whether it meets the preset cleanliness requirement; If the cleanliness requirement is not met, the cleaned wafer is put back into the cleaning device for cleaning; If the cleanliness requirement is met, the cleaned wafer is sent to the next process.

3. The method according to claim 1, characterized in that The step S102 includes: Acquire graphic structure data on the wafer surface, pre-process the graphic structure data, and extract graphic structure features; According to the extracted graphic structure features, the decision tree algorithm is used to partition the wafer surface to obtain several sub-areas with different graphic structure complexity; For each sub-area, according to the complexity of its graphic structure, the corresponding spray angle parameters are obtained from the preset spray parameter database; Determine the spray angle of each sub-area according to the acquired spray angle parameters, and obtain the spray angle data of each sub-area on the wafer surface; Inputting the spray angle data of each sub-area on the wafer surface into the spray equipment, and controlling the spray equipment to perform differentiated spray treatment on each sub-area on the wafer surface; During the spray treatment process, the image data of the wafer surface is collected in real time, and the image data is analyzed to determine whether the spray effect meets the expectations; If the spraying effect does not meet expectations, the spray angle parameters are adjusted according to the image analysis results, and the adjusted parameters are updated to the spray parameter database to optimize the subsequent wafer spraying process.

4. The method according to claim 1, characterized in that The step S103 includes: Obtain the material information and surface roughness value of the wafer and use them as query conditions; Access the preset spray time-wafer attribute association database and perform matching queries based on material and roughness values; If a matching record is found in the associated database, the corresponding recommended cleaning time range is extracted; Determine the target cleaning time of the wafer based on the obtained recommended cleaning time range; The target cleaning time is transmitted to the spray control module, and the wafer is cleaned for a corresponding period of time; After cleaning is completed, the roughness of the wafer surface is tested to obtain the roughness value after cleaning; The roughness value after cleaning is compared with the preset threshold to determine whether the cleaning effect meets the standard, so as to determine whether secondary cleaning or adjustment of cleaning parameters is required.

5. The method according to claim 1, characterized in that The step S104 includes: Acquire contamination degree data and material property data of the wafer surface, and determine the spray parameters of each sub-area of ​​the wafer surface according to the contamination degree data and material property data, wherein the spray parameters include nozzle layout, number of nozzles, spray liquid composition and temperature; The spray parameters of each sub-area are integrated to obtain a wafer surface cleaning solution; Using a machine learning algorithm, according to preset historical cleaning data and current wafer surface condition data, various parameters in the cleaning scheme are optimized to obtain an optimized cleaning scheme; Converting the optimized cleaning scheme into a control instruction, and transmitting the control instruction to a spray control unit; During the cleaning process, real-time spray parameters and wafer surface condition data of each sub-area are obtained; If the real-time spray parameters and wafer surface condition data do not conform to the preset range, dynamically adjusting the cleaning scheme; After cleaning is completed, the cleaning effect data of the wafer surface is obtained, and the cleaning effect data is input into the machine learning algorithm, the machine learning algorithm is trained to obtain an optimized machine learning algorithm for subsequent optimization of the cleaning scheme.

6. The method according to claim 1, characterized in that The step S105 includes: The spray control unit receives the cleaning plan, obtains the preset nozzle trajectory and time interval parameters, and controls the movement of the nozzle on the wafer surface according to the parameters; During the movement of the nozzle, the pressure and flow data of the nozzle are collected in real time, and the collected real-time data are compared with the pressure and flow thresholds set in the cleaning plan; If the comparison results show that the real-time pressure or flow data deviates from the set threshold, it is judged as an abnormal situation and the nozzle opening adjustment mechanism is triggered; According to the direction and amount of pressure deviation, the nozzle opening adjustment value is dynamically calculated, and the nozzle opening is adjusted through the control signal to restore the pressure to the set range; According to the flow deviation direction and deviation amount, the nozzle opening adjustment value is dynamically calculated, and the nozzle opening is adjusted through the control signal to restore the flow to the set range; The pressure and flow monitoring data within a cleaning cycle are recorded and counted to form a cleaning quality report for evaluating the cleaning effect and optimizing the cleaning plan.

7. The method according to claim 1, characterized in that The step S106 includes: Acquire real-time images of the wafer surface through a machine vision system and transmit the images to an image processing module; In the image processing module, the acquired wafer surface image is preprocessed, including image denoising, enhancement and correction operations, to improve the image quality; According to the pre-processed wafer surface image, an image segmentation algorithm is used to segment the image and extract the pollutant area on the wafer surface; For the segmented pollutant area, calculate its area ratio to obtain the residual pollutant amount on the current wafer surface; Compare the calculated residual contaminant amount with the preset threshold value to determine whether the wafer surface meets the cleaning requirements; If the residual amount of contaminants is higher than a preset threshold, the conveyor device is controlled to re-convey the wafer to the spray device to clean the wafer again; If the residual contaminant amount is lower than the preset threshold, the wafer surface is judged to be qualified for cleaning and the wafer is transferred to the next process for subsequent processing.

8. The method according to claim 1, characterized in that The step S107 includes: Obtaining a wafer surface image after wafer cleaning is completed, converting it into a grayscale image, and performing image enhancement processing to obtain a processed wafer surface image; Obtain a standard image matching the current wafer type from a preset standard image library, and perform the same image enhancement processing on the standard image; Using feature extraction algorithms, SIFT or SURF algorithms are used to extract key feature points from the processed wafer surface image and the standard image; Through the feature point matching algorithm, the RANSAC algorithm is used to match the feature points of the wafer surface image with the standard image, and the ratio of the number of matching point pairs to the total number of feature points is calculated to obtain the similarity value; The calculated similarity value is compared with a preset similarity threshold value, and if the similarity value is lower than the threshold value, it is judged that the wafer cleaning quality is unqualified; If the wafer cleaning quality is unqualified, the wafer number and the corresponding cleaning parameters are recorded in the abnormal cleaning log database, and the wafer is sent to the unqualified product buffer area; If the wafer cleaning quality is qualified, the wafer will be transferred to the next step for subsequent processing, and the wafer processing status information will be updated to track the wafer production progress in real time.

9. The method according to claim 1, characterized in that: The step S108 includes: Obtain the actual cleaning parameters of the wafers that have passed the cleaning process and upload them to the spray parameter optimization module; The spray parameter optimization module obtains historical accumulated cleaning data and uses the support vector machine algorithm to train the cleaning data; Through the training of support vector machine algorithm, a dynamically updated spray parameter-cleaning effect mapping model is obtained; When a new wafer needs to be cleaned, the initial cleaning parameters of the new wafer are input into the spray parameter-cleaning effect mapping model; The spray parameter-cleaning effect mapping model predicts the optimized cleaning parameters based on the input initial cleaning parameters; Applying the optimized cleaning parameters to a cleaning process of a new wafer to clean the new wafer; After cleaning is completed, the actual cleaning parameters and cleaning effects of the new wafer are obtained and added to the historical cleaning data for subsequent model updates and optimizations.

10. An intelligent semiconductor wafer spray control system, characterized in that: The system is used to execute an intelligent semiconductor wafer spray control method according to any one of claims 1 to 9, and the spray control system includes: The contamination analysis module is used to obtain the contamination distribution on the wafer surface and divide it into sub-areas; Graphic structure analysis module, used to obtain the graphic structure of the wafer surface and divide it into sub-areas; A cleaning parameter acquisition module is used to obtain a recommended cleaning time range based on wafer properties; A cleaning plan generation module is used to integrate parameters to form a wafer surface cleaning plan; A cleaning control module is used to control the spray device to perform cleaning and monitor and adjust in real time; The quality detection module is used to monitor the cleaning effect in real time and perform subsequent processing.

Images