Method for detecting irradiation boundary and center of light source

Abstract

The invention provides a light source irradiation boundary and center detection method, which comprises the following steps: carrying out illumination scanning on original position data, and if an illumination value on a detection position exceeds an illumination range (namely, representing that the illumination of the current position exceeds a certain illumination range or a visual field range), changing a scanning direction, reducing a scanning step length, and obtaining a detection result. Meanwhile, the number of iterations is increased to carry out illumination scanning on a detection position; and determining a detection set until the number of iterations is completed and the situation that the illumination value exceeds the view range does not exist, namely obtaining the range of the original position data for predicting the boundary. And if the predicted boundary is a polygon, performing rough classification on the site data by adopting an unsupervised clustering algorithm, determining a target classification and a central point by calculating a centroid in each classification, and finally performing fitting by adopting a ransc algorithm to obtain the boundary and finally obtaining a point location drawing. According to the embodiment of the invention, the speed of light source illumination detection and the accuracy of illumination detection are improved.

Description

Technical Field

[0001] The present application relates to the field of semiconductor manufacturing technology, and in particular to a method for detecting a boundary and a center of a light source. Background Art

[0002] The semiconductor equipment manufacturing field has high requirements for the accuracy, uniformity and reliability of ultraviolet light sources. The uniformity of the light source, the center of illumination, etc. may vary slightly due to production reasons, so after integration and debugging, it is necessary to rely on algorithms, designs and other methods to correct them.

[0003] Traditional detection methods, such as using a UV intensity sensor to traverse the illumination range, thereby obtaining the relationship between all light intensities and actual positions within the illumination range, and then obtaining the boundary and center of the illumination through fitting, use this method to calibrate the light source of the integrated device. Another example is using images to analyze data using algorithms such as Hough transform, which has defects such as data redundancy, long detection time, and can only be used for detection in special occasions.

[0004] Therefore, a new solution for detecting the boundary and center of light source illumination is needed. Summary of the invention

[0005] In view of this, an embodiment of the present specification provides a method for detecting the boundary and center of a light source.

[0006] The embodiments of this specification provide the following technical solutions:

[0007] An embodiment of this specification provides a method for detecting a boundary and a center of a light source, including:

[0008] Scan the position along a scanning direction using an initial step length, and detect changes in the illumination value at the detection position. When the illumination value at a detection position exceeds a preset illumination range, it is determined that the current detection position exceeds the detection set, wherein each scanning direction, scanning step length, and number of iterations are preset;

[0009] The scanning step length is reduced from the initial scanning step length to the second step length, the initial iteration number is increased to the second iteration number, and the scanning direction is changed to scan. When the iteration number is greater than the second iteration number and the illumination value of the detection position does not exceed the field of view, a detection set is determined, and each point in the detection set is used to predict the boundary;

[0010] If the boundary information predicted from the point data in the detection set is a polygon, an unsupervised clustering algorithm is used to roughly classify the point data, and the target classification and the center point are determined by calculating the centroid in each classification;

[0011] The ransc algorithm is used to perform straight line fitting according to each target classification and the corresponding center point. The fitting result is to determine the intersection point of the corresponding boundaries of two adjacent classifications.

[0012] According to the fitting results, after filtering out the abnormal points from the detection set, the target graphic boundary is determined from the predicted boundary and the point drawing is obtained.

[0013] The embodiments of this specification also provide a method for light intensity correction, which adopts the method for detecting the boundary and center of the light source illumination described in the above technical solution. During the debugging process of the template printing equipment, the relationship between the fitting accuracy, initial step size and number of iterations is set, and the ransc confidence threshold is set to fit the field of view of the light intensity and obtain the field of view coordinates.

[0014] The embodiments of this specification also provide a method for wafer edge finding, which adopts the detection method of the boundary and center of light source illumination described in the above technical solution, uses a spectral sensor in the wafer bonding equipment to collect data based on the set boundary and the equation of the circle to perform RANSC least squares fitting, and determines the center and radius of the wafer after removing interference points.

[0015] Compared with the prior art, the at least one technical solution adopted in the embodiments of this specification can achieve the following beneficial effects:

[0016] The embodiments of this specification use variable step length scanning to obtain valid data points. The field of view edge information corresponding to the entire boundary range is divided by always detecting the data corresponding to the boundary position. Prior information is obtained through prediction, and the prior conditions are added to subsequent calculations to improve the fitting results. The fitting process can not only fit geometric figures with general trajectory equations, but also fit polygons to finally obtain accurate results. While improving the fitting accuracy, the detection efficiency is also improved; it can also adapt to different scenarios and fit various different graphics, and has universality. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0018] Figure 1 is a flow chart of a detection method for obtaining boundaries and centers provided in an embodiment of this specification;

[0019] Figure 2 It is a flow chart of determining the boundary and center according to the geometric boundary of the illumination range provided in the embodiment of this specification;

[0020] Figure 3is a flowchart of obtaining the most central position as the center provided in an embodiment of this specification;

[0021] Figure 4 It is a schematic diagram of the field of view coordinates in the light intensity correction process provided in the embodiment of this specification. DETAILED DESCRIPTION

[0022] The embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0023] The following describes the implementation methods of the present application through specific examples, and those skilled in the art can easily understand other advantages and effects of the present application from the contents disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. The present application can also be implemented or applied through other different specific implementation methods, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present application. It should be noted that, in the absence of conflict, the following embodiments and the features in the embodiments can be combined with each other. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without making creative work belong to the scope of protection of the present application.

[0024] It should be noted that various aspects of the embodiments within the scope of the appended claims are described below. It should be apparent that the aspects described herein can be embodied in a wide variety of forms, and any specific structure and / or function described herein is merely illustrative. Based on the present application, it should be understood by those skilled in the art that an aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number and aspect described herein can be used to implement the device and / or practice the method. In addition, other structures and / or functionalities other than one or more of the aspects described herein can be used to implement this device and / or practice this method.

[0025] It should also be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present application. The drawings only show components related to the present application rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component may be changed arbitrarily, and the component layout may also be more complicated.

[0026] It should be noted that the terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein.

[0027] Additionally, in the following description, specific details are provided to facilitate a thorough understanding of the examples. However, it will be understood by those skilled in the art that the examples can be practiced without these specific details.

[0028] Traditional detection methods require traversing the illumination range, such as traversing the illumination range through a UV intensity sensor to obtain the relationship distribution between all light intensities within the illumination range and the actual position, and then obtaining the boundary and center of the illumination through fitting. Another example is the use of image data analysis such as Hough transform for detection. That is, the entire field of view needs to be traversed, and the range outside the boundary and the boundary needs to be scanned, resulting in redundant data processing and a long time consumption. Moreover, it can only be applied to specific scenes for detection, and cannot be applied to all detection scenes.

[0029] In addition, traditional detection methods lack the ability to distinguish the real-time and authenticity of data, and sometimes require manual deletion of interfering information, which reduces the credibility of detection.

[0030] Based on this, the embodiment of this specification proposes a new detection scheme for the boundary and center of light source illumination: mainly by fitting the geometric figures corresponding to the illumination range to obtain parameters such as the center and boundary. This scheme uses a variable step size to scan the illumination at the detection position to obtain valid data points, reduce data redundancy processing, use an unsupervised clustering algorithm to predict the data and obtain prior information, pre-process the data to remove interference, and can be applied to fitting targets with multiple deformations, roughly classify the original data and obtain the center point, and finally use ransc for boundary fitting to obtain the final accurate result.

[0031] The technical solutions provided by various embodiments of the present application are described below in conjunction with the accompanying drawings.

[0032] like Figure 1 and Figure 2 As shown, the embodiment of this specification provides a method for detecting the boundary and center of light source illumination.

[0033] The first aspect is to solve the problem of obtaining effective data. In actual operation, a large amount of data is within the geometric boundary range of the illumination range. Using all data processing will cause redundancy, which is not only time-consuming but also inflexible. Therefore, in the embodiment of this specification, variable step length scanning is used to obtain effective data points. That is, the field edge information corresponding to the entire boundary range is divided by always detecting the data corresponding to the boundary position.

[0034] Secondly, after obtaining the edge information of the field of view, the data can be preprocessed to remove interference, that is, prior information can be obtained through prediction, and the prior conditions can be added to subsequent calculations to improve the fitting results.

[0035] Thirdly, in the fitting process, the target is increased as a polygon, that is, the fitting of a mathematical model without a general trajectory equation is added. Specifically, the original data is roughly classified and then fitted using ransc to obtain the final accurate result.

[0036] Specifically, it includes steps S101 to S105. In step S101, position scanning is performed along a scanning direction using an initial step length, and changes in illumination values ​​at the detection positions are detected. When the illumination value of a detection position exceeds a preset illumination range, it is determined that the current detection position exceeds the detection set.

[0037] In actual operation, there are various situations in the illumination field range, such as uniform illumination, illumination attenuation, and flickering illumination. The embodiment of this specification adopts variable step size to perform preliminary division of illumination effective data, mainly by detecting data points at the boundary of illumination change to obtain the edge information of the field of view of the entire illumination area. Each scanning direction, scanning step size, number of iterations, and search range can be preset.

[0038] An initial step length is set to scan along a scanning direction and detect the illumination value in real time. When the illumination value of a detection position exceeds a preset illumination range, it is determined that the current detection position exceeds the detection set.

[0039] The preset illumination range can be specifically limited according to the actual situation. For example, the preset illumination range includes exceeding the field of view, that is, the preset illumination range is less than the preset illumination value when the illumination value cannot be detected. For another example, the preset illumination range corresponds to the light changing from bright to dark, that is, the preset illumination range is the preset illumination range corresponding to strong light, etc.

[0040] When a large change in the illuminance value is detected, it means that the detection position has reached the illumination boundary; or, when the illuminance value is less than the preset illuminance value, that is, exceeds the field of view, it is necessary to further determine the field of view edge information by changing the scanning step size, etc.

[0041] Step S102, reduce the scanning step size from the initial scanning step size to the second step size, increase the initial iteration number to the second iteration number, and change the scanning direction. When the overlapping number is greater than the second iteration number and the illumination value of the detection position does not exceed the field of view, the detection set is determined, and the points in each detection set are used to predict the boundary.

[0042] In combination with the above embodiments, variable step length scanning is performed by changing the scanning direction, reducing the step length, and increasing the number of iterations. For example, the scanning step length is reduced from the initial scanning step length to the second step length, the initial number of iterations is increased to the second number of iterations, and the scanning direction is changed to perform scanning. When the number of iterations is greater than the second number of iterations and the illumination value of the detection position does not exceed the field of view, a detection set is determined, and each point in the detection set is used to predict the boundary. The points in the detection set correspond to the edge points of the field of view.

[0043] In some embodiments, if the illumination value changes greatly in the initial scanning direction, it means that the detected position has reached the illumination boundary, the scanning direction is changed, the step length is reduced, and the number of iterations is increased. When the illumination value of the detected position changes again, such as the illumination value is less than a preset illumination value, it means that the detected position has exceeded the illumination range, the scanning direction is changed again, the step length is reduced, and the number of iterations is increased. When the number of iterations is greater than the set value, it means that the required accuracy has been met. In this way, the boundary information is always detected, and the field of view edge information within the entire boundary range is finally obtained.

[0044] It should be noted that the variable step length scanning used in this manual can not only make the step length very small to improve the detection accuracy. In addition, the later data processing uses the RANSC algorithm for fitting. When applied to the semiconductor detection field, it can automatically delete interference data according to the set confidence threshold, thereby obtaining a good fitting result. The RANSC algorithm has been widely used in image-related fields, such as splicing, feature matching, etc.

[0045] It should also be noted that the predicted boundary of the light source change is obtained, the data is stored and subsequent calculations are performed. The scanning algorithm can control the accuracy according to the set number of iterations. Here, a binary search or gradient descent algorithm is used.

[0046] In some embodiments, the step further includes scanning according to the unreduced scanning step length and the current number of iterations when the illumination value at the detection position does not exceed a preset illumination range, until the number of iterations is greater than a set value to complete the boundary scan.

[0047] Specifically, when the illumination value is in a corresponding field of view, such as a light-dark area, there is no need to perform variable step scanning, and the current scan continues, that is, scanning is performed according to the unreduced scanning step and the current number of iterations until the number of iterations is greater than the set value to complete the boundary scan.

[0048] Step S103: If the boundary information predicted from the point data in the detection set is a polygon, an unsupervised clustering algorithm is used to minimize the influence between the effective boundary data, and the point data is roughly classified, and the target classification and the center point are determined by calculating the centroid in each classification. The unsupervised clustering algorithm includes but is not limited to k-means, SVM, k-medoids algorithm, etc.

[0049] Combined with the above variable step length scanning to obtain valid data point position process, the detection set is obtained, and then the geometric figure is obtained according to the target requirements. If the detection set, that is, the mathematical model with a general trajectory equation when fitting the edge information of the field of view, the subsequent processing can directly apply the ransc algorithm to perform anti-noise and interference fitting to obtain accurate results.

[0050] If the fitting target is a polygon, a rough classification is required, and then ransc is used for fitting to obtain the final accurate result.

[0051] In some embodiments, in order to improve the accuracy of the edge fitting of the field of view, the raw data needs to be preprocessed to obtain prior information, and then the boundary and range of the polygon are determined. In particular, when the determination of the center point is extremely inaccurate, the distribution of the data will be distorted, resulting in inaccurate geometric figures.

[0052] If k-means is used to select the average value of the cluster in the detection set as the new center, it will be iterated until the distribution of objects in the cluster no longer changes. However, there will be a situation where the center point is extremely inaccurate. The k-means algorithm cannot achieve a higher accuracy because large interference will be detected during the illumination acquisition process. Therefore, based on the calculation of clusters in k-means, the centroid algorithm is used to eliminate the above interference, that is, the most central object at the center of the cluster is taken as the centroid, which is similar to the way K-medoids finds the centroid to improve accuracy, and the detection is faster, and the number of convergences is fixed.

[0053] In other embodiments, the raw data is subjected to rough classification of point data, and the process of directly calculating the centroid is similar to using a priori conditions to obtain the target classification and the center point after removing interference.

[0054] Step S104: perform straight line fitting using the ransc algorithm according to each target category and the corresponding center point. The fitting result is to determine the intersection point of the corresponding boundaries of two adjacent categories, thus completing the fitting process.

[0055] In some embodiments, if the boundary information predicted according to the point data in the detection set is a geometric figure with a curve equation, the ransc algorithm is directly applied to perform anti-noise and interference fitting to obtain a point drawing.

[0056] Specifically, if the fitting target in the process of fitting the edge information of the field of view is a mathematical model with a general trajectory equation, the subsequent processing can directly apply the ransc algorithm to use the geometric figures represented by the curve equation to achieve anti-noise and interference fitting to obtain accurate results.

[0057] In other embodiments, if the fitting target is a polygon, the target classification and center point are determined by combining the above calculation of the centroid in each classification, and the ransc algorithm is used for straight line fitting. The fitting result is to determine the intersection of the corresponding boundaries of two adjacent classifications.

[0058] For example, to eliminate interference, the original data is fitted to remove abnormal points according to the prior conditions of the cluster average value, that is, the prior is obtained by combining the above clustering algorithm, and the range of the prior information belonging to one edge is classified into one category. After a certain distance away from the prior, it is considered that the position belongs to the range of a certain edge of the polygon, and these positions are classified into another category. These classes are fitted with straight lines using the ransc algorithm, and after multiple iterations, a result with high credibility is obtained, that is, the intersection of the corresponding boundaries of two adjacent categories is determined. The intersection of these boundaries is the intersection of the polygon and is near the prior.

[0059] Another example is to directly use the centroid to determine the target classification and center point, which is similar to using prior conditions to remove interference. The ransc algorithm is used for straight line fitting to achieve anti-noise and interference fitting, thereby determining the intersection of the corresponding boundaries of two adjacent target classifications.

[0060] In the embodiments of this specification, unsupervised clustering and other machine learning algorithms preprocess the data to obtain prior conditions, then set the range according to the prior conditions, and adaptively remove the interference information to obtain the result, which greatly improves the efficiency. Alternatively, in the process of calculating the center point, the centroid calculation method is used to directly process the process similar to the process of obtaining the prior conditions including preprocessing, and finally adaptively remove the interference information to obtain the result, which greatly improves the efficiency.

[0061] Step S105: According to the fitting result, after filtering out the abnormal points from the detection set, the target graphic boundary is determined from the predicted boundary and the point drawing is obtained.

[0062] Specifically, in the data drawing process, the fitting result is drawn using the remaining points after the points filtered out from the original points, so as to determine the target graphic boundary in the predicted boundary and obtain the point drawing. It is verified by observation that the point drawing obtained by the embodiment of this specification is not only for a specific scene, but also for a target polygon, and can obtain accurate point drawing, such as high-precision calibration for semiconductor equipment. After the module is designed, it can adapt to different scenes and different target graphics, and has good effects on the calculation of illumination uniformity, high-precision fitting of center points and edge points, and has universal applicability.

[0063] The detection method of the embodiment of this specification has the following beneficial effects:

[0064] 1. Improve detection speed: Existing methods traverse the entire field of view, such as the range outside the boundary of the target graphic and the range inside the boundary, there is redundant data. This application uses a smaller step size to always detect boundary information, and finally obtains the edge information of the field of view within the entire boundary range. The use of a reduced step size not only improves processing accuracy, but also compares to the existing method with an algorithm with a time complexity of O(n^2), that is, processing n data requires n^2 or n*n operations, while this application uses a variable step size method, processing log(n) data requires log2(n) operations, and the time complexity is O(log2(n)), and does not scan the range outside and inside the boundary, which shortens the time spent on scanning under high precision and shortens the detection time.

[0065] 2. It has high decoupling and is suitable for high-precision calibration of semiconductor equipment. The module design can adapt to different scenes and different target graphics, and has good effects on the calculation of illumination uniformity and high-precision fitting of center points and edge points.

[0066] 3. Improve accuracy. Variable step size can make the step size very small, which makes the detection accuracy very high. The later data processing uses the ransc algorithm for fitting. When applied to the field of semiconductor detection, it can automatically delete interference data according to the set confidence threshold, thereby obtaining a good fitting result.

[0067] 4. No human intervention. Compared with the existing methods, which require designing prior conditions and removing interference points after collecting data, which consumes manpower, this application uses machine learning algorithms such as unsupervised clustering to pre-process the data to obtain prior conditions, and then sets the range according to the prior conditions, adaptively removes interference information to obtain results, greatly improving efficiency.

[0068] In some embodiments, an unsupervised clustering algorithm is used to perform a preliminary classification of site data, and the target classification and center point are determined by calculating the centroid point in each classification, including: presetting the preliminary classification and the initial cluster center corresponding to each classification from the detection set; assigning sample points to each classification by calculating the distance from each site to each of the initial cluster centers in the detection set, wherein the sample points in each classification are determined based on whether the distance between the sample point and the initial cluster center meets a first threshold; within each classification sample range, the initial cluster center is moved and updated to a new cluster center according to the average value of all samples, and each classification sample range is updated each time the cluster center is updated; when the moving distance when the initial cluster center is updated to the new cluster center is less than a second threshold, the first center point corresponding to each classification is determined; within each classification sample range, the point closest to the first center point is updated as the center of mass; when the distance between the previous center of mass and the current center of mass is less than a third threshold, the target center of mass of the target classification is obtained, that is, the target classification is obtained.

[0069] Specifically, taking the K-means algorithm as an example, the cluster centers are initialized first. This step requires randomly selecting k samples as the initial cluster centers, where k is required. In some embodiments, the points of the detection set are respectively used as samples, and the initial cluster centers are initialized from each detection set.

[0070] Then, samples are assigned to k cluster centers, the distance between each sample and each cluster center is calculated, and each sample is assigned to a cluster center that meets the preset conditions, such as the distance between the sample and the cluster center meets the first threshold. Then, the cluster center is moved to the average value of all samples in this cluster. At the same time, the sample range of each classification is updated, that is, the process of assigning samples is the same as above, which will not be repeated here.

[0071] Finally, until the distance moved by the cluster center is less than a certain set threshold or remains unchanged, the movement stops, indicating that the data has reached convergence. According to the current scenario, the k we set is known.

[0072] That is, the K-means algorithm is used to obtain prior information by taking the cluster average as the new center, and then the centroid algorithm is used to obtain the most central position in the cluster as the center point.

[0073] Combined with the process of obtaining prior information, within each classification sample range, the point closest to the first center point is updated as the centroid, and each classification sample range is updated to obtain the target classification; when the distance between the previous centroid and the current centroid is less than the third threshold, the target centroid and target classification of the target classification are obtained.

[0074] In some embodiments, an unsupervised clustering algorithm is used to perform a preliminary classification of site data, and the target classification and center point are determined by calculating the centroid in each classification, which also includes: presetting the preliminary classification and the initial cluster center corresponding to each classification from the detection set; assigning sample points to each classification by calculating the distance from each site to each of the initial cluster centers in the detection set, wherein the sample points in each classification are determined based on whether the distance between the sample point and the initial cluster center meets a fourth threshold; within each classification sample range, the point closest to the initial cluster center is updated as the centroid, and each classification sample range is updated to obtain the target classification; when the distance between the previous centroid and the current centroid is less than the fifth threshold, the target centroid of each class is obtained and the target classification is obtained.

[0075] Specific as Figure 3 As shown in the figure, the k-medoids algorithm directly calculates the centroid of each classification to determine the target classification and center point. The main process is as follows:

[0076] First, input the data data and the number of classes k, where data is a set of points of x and y, such as the detection set. k is the number of detection sets.

[0077] k classes need to be initialized with k initial points, usually random points. The positions of the points will be continuously updated in subsequent iterations until the updated position offset is less than a certain threshold, which means the algorithm converges.

[0078] Then enter the loop, traverse all points, calculate the distance of each point to k centroid points, select the centroid whose distance to this point meets the fourth threshold, and determine that this point belongs to this class. This is the process of assigning sample points to each classification.

[0079] Calculate the centroids of the points in all classes separately, select the points closest to the centroids as the new centroids of these classes, and update the samples in each class in the same way as the process of allocating sample points mentioned above.

[0080] Finally, the cycle is repeated until the distance between the new centroid and the previous centroid is less than a certain threshold, indicating that the algorithm has reached convergence. That is, the target classification and the corresponding target centroid are obtained.

[0081] In some embodiments, each sample corresponds to a silhouette coefficient, and the range of each classification sample is determined by the silhouette coefficient; the silhouette coefficient of each sample in the target classification is greater than or equal to -1 and less than or equal to 1; wherein the silhouette coefficient includes cohesion and separation.

[0082] Taking k-means as an example, the contour system is used to judge the convergence of the scan during the continuous iteration process. Each sample has a corresponding contour coefficient, which consists of two parts:

[0083] a: The average distance between the sample and other sample points in the same cluster (quantified cohesion)

[0084] b: The average distance between the sample and all sample points in the nearest cluster (quantitative separation)

[0085] max(a,b) means taking the maximum value of a and b.

[0086] The silhouette coefficient for each sample is defined as:

[0087]

[0088] In the process of obtaining the target classification and the center point, the silhouette coefficient of each sample in the target classification is greater than or equal to -1 and less than or equal to 1.

[0089] In some embodiments, the average distance between all points in the target classification and the target centroid meets a preset threshold.

[0090] Taking the k-means example again, during the continuous iteration process, SSE is used to determine the convergence of the scan.

[0091] Among them, SSE:

[0092]

[0093] Given a data set D containing n data objects, D = {x1, x2, x3, ..., x n}, define the category set generated by clustering through the K-means algorithm as set C = {C1, C2, C3, …, C n}, where m k It is cluster C k The center point is calculated as follows:

[0094]

[0095] SSE is the sum of squares error, which mainly measures the average distance between all data in a certain category and the center point of the data cluster. The smaller the SSE, the denser the data cluster, that is, the better the clustering effect. That is, the average distance between all points in the target classification and the target centroid meets the preset threshold.

[0096] In some embodiments, it also includes: in each classification, taking the target centroid as the center, drawing a circle with a gradually increasing radius until a target circle is formed with the initial step length as the radius, and determining that the target circle includes the initial cluster center, then determining the target centroid.

[0097] After obtaining the target centroid, the accuracy of the target centroid is verified by the following method. Taking k-means as an example, a circle with a gradually increasing radius is made with the center obtained by k-means as the center, and the initial step size S is set to increase gradually until the circle contains the real sampling point, such as Figure 3 As shown, by obtaining sampling points such as the initial cluster center, the cluster center is obtained through the k-medoids algorithm, such as the target centroid to obtain the object at the center of the cluster. Among them, the initial cluster center can be similar to the first point in the k-medoids algorithm. The first point is assumed to be a theoretical point in the space. The theoretical point is calculated by the distance and weight of the surrounding points and is set as a virtual point. It does not really exist in the feature space. With this virtual point as the center of the circle, the radius is expanded until it includes the first real point, and this real point is used as the real centroid to obtain the real centroid. The initial cluster center can be a real point.

[0098] In combination with the above embodiments, the embodiments of this specification provide a method for light intensity correction. During the debugging process of the template printing equipment, the relationship between the fitting accuracy, initial step size and number of iterations is set, and the ransc confidence threshold is set to fit the field of view of the light intensity and obtain the field of view coordinates.

[0099] Specifically, during the debugging process of the template printing device, the center of the field of view of the calibrated light intensity is applicable to the detection algorithm of the light source illumination boundary and center of the above embodiment. According to the requirements, if the fitting accuracy is set to be less than a, the initial step size is s, and the number of iterations is n, the following formula relationship is obtained:

[0100]

[0101] Right now

[0102]

[0103] Set parameters according to accuracy requirements, communicate with the industrial computer to control the coarse motion stage and each sensor to collect data in real time. When the collection is completed, determine whether a priori is needed based on the mathematical model of the scan target, and then set the ransc confidence threshold and the number of iterations to obtain the fitting result, as shown below Figure 4 As shown in the figure: the straight line is the final accurate fitting result, the large X mark is the intersection of the accurate fitting, the small x mark is the prior condition, the reliable points considered by the ransc algorithm are dark dots, and the removed points are light dots. Obtaining accurate field of view coordinates can improve the accuracy of the device.

[0104] In combination with the above embodiments, the embodiments of this specification provide a method for wafer edge finding, which uses a spectral sensor in a wafer bonding device to collect data based on the set boundary and the equation of the circle to perform RANSC least squares fitting, and determines the center and radius of the wafer after removing interference points.

[0105] Specifically, in wafer bonding equipment, a spectral confocal sensor is used to find the edge of the wafer. Due to problems such as unstable cutting angle and position deviation at the edge of the wafer, the sensor collects unstable data. In this scenario, the above-mentioned light source illumination boundary and center detection algorithm is applicable to fit the edge and calculate the center.

[0106] Set the boundary acquisition data as required. When the spectral confocal sensor data mutates, it means that the current position is near the wafer boundary. Collect multiple boundary information and perform RANSC least squares fitting based on the following formula, using the general equation of a three-dimensional circle as the mathematical model.

[0107] x 2 +y 2 +Dx+Ey+F=0(Formula 3)

[0108] Collect data at multiple data points to obtain a linear equation system:

[0109]

[0110] Set the number of iterations to perform ransc iterations at the confidence level to obtain D, E, and F after removing interference. The center and radius of the circle are obtained according to the formula:

[0111]

[0112] Among them, D, E, and F are constants.

[0113] The embodiments of this specification are applicable to semiconductor equipment precision calibration, equipment stability detection, system error correction, non-contact anti-interference high-precision fitting, and are particularly applicable to high-precision calibration of semiconductor detection equipment, imprinting equipment, lithography equipment, etc.

[0114] In combination with the above embodiments, the traditional detection method has the following shortcomings:

[0115] 1. Obtain data redundancy. According to the test performance, we only need the boundary information of the geometric figure within the illumination range to obtain its center, boundary and other parameters through fitting. However, in actual operation, a large amount of data is within the geometric boundary range of this illumination. These data are redundant, which is not only time-consuming but also less flexible. If you want to improve the detection accuracy, you need a smaller detection step size.

[0116] 2. Low flexibility. After obtaining the data, the traditional method needs to perform post-processing to reduce interference, and then perform least squares fitting to obtain parameters. This method has very high requirements on the real-time and authenticity of the data. If necessary, it is also necessary to manually delete the interference information to obtain more reliable results.

[0117] 3. The scope of application is narrow. Traditional methods can only be applied to specific scenarios.

[0118] 4. Low anti-interference ability. If the traditional method is directly fitted, noise, anomalies and other information will be added to the result. Therefore, it is necessary to manually process the data, which is time-consuming and labor-intensive.

[0119] The same or similar parts between the various embodiments in this specification can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the product embodiment described later, since it corresponds to the method, the description is relatively simple, and the relevant parts can be referred to the partial description of the system embodiment.

[0120] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

Claims

1. A method for detecting the boundary and center of a light source, characterized in that: include: Scan the position along a scanning direction using an initial step length, and detect changes in the illumination value at the detection position. When the illumination value at a detection position exceeds a preset illumination range, it is determined that the current detection position exceeds the detection set, wherein each scanning direction, scanning step length, and number of iterations are preset; The scanning step length is reduced from the initial scanning step length to the second step length, the initial iteration number is increased to the second iteration number, and the scanning direction is changed to scan. When the iteration number is greater than the second iteration number and the illumination value of the detection position does not exceed the field of view, a detection set is determined, and each point in the detection set is used to predict the boundary; If the boundary information predicted from the point data in the detection set is a polygon, an unsupervised clustering algorithm is used to roughly classify the point data, and the target classification and the center point are determined by calculating the centroid in each classification; The ransc algorithm is used to perform straight line fitting according to each target classification and the corresponding center point. The fitting result is to determine the intersection point of the corresponding boundaries of two adjacent classifications. According to the fitting results, after filtering out the abnormal points from the detection set, the target graphic boundary is determined from the predicted boundary and the point drawing is obtained.

2. The method for detecting the light source illumination boundary and center according to claim 1, characterized in that: An unsupervised clustering algorithm is used to roughly classify the site data, and the centroid points in each classification are calculated to determine the target classification and center point, including: Preset preliminary classification and initial cluster center corresponding to each classification from the detection set; In the detection set, a sample point is assigned to each category by calculating the distance from each site to each of the initial cluster centers, wherein the sample point in each category is determined according to whether the distance from the sample point to the initial cluster center satisfies a first threshold; In each classification sample range, the initial cluster center is updated to a new cluster center according to the average value of all samples, and each classification sample range is updated each time the cluster center is updated; When the initial cluster center is updated to the new cluster center and the moving distance between the two is less than a second threshold, determining a first center point corresponding to each classification; In each classification sample range, the point closest to the first center point is updated as the centroid; When the distance between the last centroid and the current centroid is less than a third threshold, the target centroid of the target classification is obtained and the target classification is obtained.

3. The method for detecting the light source illumination boundary and center according to claim 1, characterized in that: The site data is roughly classified using an unsupervised clustering algorithm, and the target classification and center point are determined by calculating the centroid of each classification, including: Preset preliminary classification and initial cluster center corresponding to each classification from the detection set; In the detection set, a sample point is allocated to each category by calculating the distance from each point to each of the initial cluster centers, wherein the sample point in each category is determined according to whether the distance from the sample point to the initial cluster center satisfies a fourth threshold; In each classification sample range, the point closest to the initial cluster center is updated as the centroid, and each classification sample range is updated to obtain the target classification; When the distance between the previous centroid and the current centroid is less than the fifth threshold, the target centroid of each category is obtained and the target classification is obtained.

4. The method for detecting the light source illumination boundary and center according to claim 2 or 3, characterized in that: Each sample has a corresponding silhouette coefficient, and the range of each classification sample is determined by the silhouette coefficient; The silhouette coefficient of each sample in the target classification is greater than or equal to -1 and less than or equal to 1; The silhouette coefficient includes cohesion and separation.

5. The method for detecting the light source illumination boundary and center according to claim 2 or 3, characterized in that: The average distance between all points in the target classification and the target centroid meets a preset threshold.

6. The method for detecting the light source illumination boundary and center according to claim 2 or 3, characterized in that: Also includes: In each classification, a circle with a gradually increasing radius is drawn with the target centroid as the center until a target circle is formed with the initial step length as the radius, and it is determined that the target circle includes the initial cluster center, and then the target centroid is determined.

7. The method for detecting the light source illumination boundary and center according to claim 1, characterized in that: Also includes: When the illumination value at the detection position does not exceed the preset illumination range, scanning is performed according to the unreduced scanning step length and the current number of iterations until the number of iterations is greater than the set value to complete the boundary scan.

8. The method for detecting the light source illumination boundary and center according to claim 1, characterized in that: Also includes: If the boundary information predicted according to the point data in the detection set is a geometric figure with a curve equation, the ransc algorithm is directly applied to perform anti-noise and interference fitting to obtain a point drawing.

9. A method for light intensity calibration, characterized in that: The method for detecting the boundary and center of light source illumination as described in any one of claims 1 to 8 is adopted. During the debugging process of the template printing equipment, the relationship among the fitting accuracy, the initial step size and the number of iterations is set, and the ransc confidence threshold is set to fit to obtain the field of view of the light intensity and the field of view coordinates.

10. A wafer edge finding method, characterized in that: The method for detecting the boundary and center of light source illumination as described in any one of claims 1 to 8 is adopted, and a spectral sensor is used in a wafer bonding device to collect data according to the set boundary and the equation of the circle to perform RANSC least squares fitting, and the center and radius of the wafer are determined after removing the interference points.

Images