Micron-sized film morphology detection method

Abstract

The invention belongs to the technical field of ink-jet printing and manufacturing, and particularly relates to a micron-sized film morphology detection method, which comprises the following steps: acquiring original point cloud data of the whole surface including a substrate area and a printed film, calculating the gradient of each point of the point cloud, and removing points in a film climbing area according to a set gradient threshold value to obtain a thin film morphology detection result; obtaining a point cloud of a substrate area and a point cloud of a printing film gentle area based on distance clustering; extracting one-time measurement point cloud of the edge of the printing film corresponding to the single-stroke scanning to calculate the approximate film thickness of the printing film; downwards translating the point cloud of the flat area of the printing film by a distance of approximate thickness; the translated point cloud and the substrate point cloud are fused and down-sampled to obtain a sparse point cloud for fitting to obtain a fitting curved surface of the whole substrate; and the corrected relative height of the point cloud is obtained by subtracting the curved surface point cloud of the fitting substrate from the original point cloud. According to the invention, high-efficiency online measurement and detection can be carried out on the film manufactured by ink-jet printing.

Description

Technical Field

[0001] This invention belongs to the technical field of inkjet printing manufacturing, and more specifically, relates to a method for detecting the morphology of micron-scale thin films. Background Technology

[0002] With the increasing demand for OLED screens, high resolution, low power consumption, and flexibility have become the development direction of new display technologies. OLEDs (Organic Light Emitting Diodes) are widely used in consumer electronics such as smartphones, tablets, computers, and televisions due to their high contrast, fast response speed, and flexibility. However, OLED devices are extremely sensitive to the environment; exposure to water and oxygen can lead to performance degradation or even failure. Therefore, thin-film encapsulation (TFE) is crucial for ensuring the reliability and lifespan of OLED devices. Furthermore, the morphology of the thin film affects the light-emitting performance of the device; therefore, accurate morphology measurement is one of the core technologies for improving OLED yield.

[0003] Common methods for measuring topography include line laser, linearly scanned confocal, and white light interferometry. Line laser measurement accuracy is typically limited to the tens of micrometers, making it unsuitable for measuring micrometer-scale thin films. White light interferometry achieves sub-nanometer accuracy; however, its single field of view is typically 1.5 mm. The 1.5 mm field of view requires longitudinal scanning to complete the measurement, resulting in relatively low measurement efficiency. Linear scanning confocal scanning offers micrometer-level measurement accuracy and can reach kilohertz measurement frequencies, enabling a balance between accuracy and efficiency for measuring micrometer-scale thin films.

[0004] Linear scanning confocal sensors can measure multilayer thin films. However, in this measurement scenario—measuring a transparent thin film approximately 10 μm thick on a transparent glass substrate—it's impossible to obtain the relative height information between different locations on the overall surface of both the substrate and the thin film by measuring the actual film thickness. This is because directly measuring the actual film thickness is not possible, and therefore, the relative height information cannot be obtained by setting the reference to 0. There are two reasons why directly measuring the actual film thickness is not possible: First, the maximum measurable film thickness is 20 μm. When measuring films thinner than 20 μm, the height signal peaks of the top and bottom surfaces of the film in the spectrum are too close, making it impossible to measure the film thickness by subtracting the signals from the top and bottom surfaces. Second, the refractive indices of the thin film and the transparent substrate are too close, both around 1.5. The interface between the film and the substrate cannot generate a discernible optical height signal; only the signal from the top of the film and the bottom of the substrate can be obtained, making it impossible to obtain the film thickness by subtracting the signals from the top and bottom surfaces.

[0005] In addition, there is currently a method to obtain the relative height information between different positions on the overall surface of the substrate and the film by setting a reference plane to correct the original height information of the film collected by the sensor. Specifically, when the original height data of the film is measured by a line scan confocal sensor, the substrate will be tilted. When measuring with an instrument, a certain part (i.e. the substrate) is usually selected as the reference plane for plane fitting, and then the overall surface morphology of the substrate and the film (i.e. the original height information collected by the sensor) is flattened with the fitted plane as the reference to achieve correction. For example, (1) the original height point cloud of the reference plane part obtained by the sensor is fitted with the RANSAC fitting algorithm, and then the original height point cloud of the overall surface morphology of the substrate and the film is subtracted from the fitted point cloud of the plane to obtain the relative height information between different positions on the overall surface of the substrate and the film, thereby realizing the correction of the overall surface morphology of the substrate and the film. However, since inkjet printed films are multi-scale, their length and width are much larger than their thickness, so the entire substrate plane is microscopically curved and is not the so-called "reference plane part", so this method cannot achieve good results. (2) Fitting the shape of the surrounding parts of the substrate using the Poisson fitting algorithm also fails to achieve good results. This is because the middle part of the film is far from the surrounding substrate, making it difficult to accurately predict the point cloud of the middle substrate using the original height point cloud of the surrounding parts measured by the sensor, resulting in poor relative height correction.

[0006] Therefore, a new overall surface morphology correction scheme that includes the substrate area and the printed film is needed. Summary of the Invention

[0007] In view of the above-mentioned defects or improvement needs of the existing technology, the present invention provides a micron-level thin film morphology detection method, the purpose of which is to perform high-efficiency online measurement and detection of thin films manufactured by inkjet printing.

[0008] To achieve the above objectives, according to one aspect of the present invention, a method for detecting the morphology of micron-scale thin films is provided, comprising: The original point cloud data of the entire surface, including the substrate area and the printed film, is acquired and stitched together using a line scan confocal sensor. The normal vector of each point in the stitched point cloud is calculated. The normal vector of each point is normalized, and the angle between the normalized normal vector of each point and the vertical unit vector is calculated as the tilt of that point. Points in the point cloud with a tilt exceeding a preset threshold are deleted. The remaining point cloud is clustered based on distance to obtain the point cloud of the substrate area and the point cloud of the smooth area of ​​the printed film. The original point cloud of a single-pass scan corresponding to the edge of the printed film is extracted. The film thickness is calculated based on this original point cloud, and the calculation result is used as the approximate thickness of the printed film. The edge of the printed film spans the sloping area of ​​the printed film and includes part of the substrate area and part of the flat area of ​​the printed film. The point cloud of the flat area of ​​the printed film is translated downward by a distance equal to the approximate thickness mentioned above. The translated point cloud is fused with the substrate point cloud. The fused point cloud is voxelized and downsampled to obtain a sparse point cloud. The Poisson fitting algorithm is used to fit the sparse point cloud to obtain the implicit function of the fitted surface of the entire substrate. The reconstructed mesh of the entire substrate is obtained based on the implicit function. The reconstructed mesh is linearly upsampled to obtain the fitted surface of the entire substrate. The original point cloud data of the entire surface is voxelized downsampled and denoised. Each point in the processed point cloud is traversed. The nearest bottom point in the x and y directions of the current point is found among the points of the fitted surface of the entire substrate. The z-coordinate of the current point is subtracted from the z-coordinate of the nearest bottom point to obtain the corrected relative height of the current point, thus completing the micron-level thin film morphology detection.

[0009] Furthermore, the method for acquiring the original point cloud data of the entire surface is as follows: A line scan confocal sensor is used to scan the entire surface, including the substrate area and the printed film, through multi-pass line scanning, obtaining the scan data of all single passes. Among them, there is partial overlap between the scan areas of two adjacent single passes. Based on the ICP algorithm, the scan data of all scan areas are stitched together in a multi-scale manner: (1) Between the scan data of each adjacent single-stroke overlapping area, firstly, the rigid transformation matrix in the ICP algorithm is initialized as the matrix of sensor physical translation. Based on a preset large matching point deviation threshold, the rigid transformation matrix used to achieve coarse stitching is obtained through the ICP algorithm. Then, the rigid transformation matrix in the ICP algorithm is initialized as the rigid transformation matrix used to achieve coarse stitching. Based on a preset small matching point deviation, the rigid transformation matrix used to achieve fine stitching is obtained through the ICP algorithm. The formula of the ICP algorithm is: In the formula, To obtain a rigid transformation matrix from the source point cloud to the target point cloud, by and composition, Here is the rotation transformation matrix; The translation transformation matrix is ​​used; the source point cloud is the partial scan data corresponding to the overlapping area of ​​the scanned single stroke in the latter scanned single stroke and the scanned single stroke in the former scanned single stroke in the two adjacent single strokes; the target point cloud is the partial scan data corresponding to the overlapping area of ​​the scanned single stroke in the former scanned single stroke and the scanned single stroke in the latter scanned single stroke in the two adjacent single strokes. The first point cloud in the source cloudi One point; For points in the target point cloud, nearest neighbor matching is used to find points in the source point cloud. The corresponding matching point; (2) By sequentially multiplying the rigid transformation matrix obtained when each single stroke is used as the source point cloud with the rigid transformation matrices of each single stroke before the single stroke scan as the source point cloud, the result of the cumulative multiplication is used as the rigid transformation matrix of the single stroke scan data and spliced ​​into the first stroke scan data. (3) Use a rigid transformation matrix to stitch each single-stroke scan data into the first stroke scan data, transform the single-stroke scan data, and stitch all the transformed scan data with the first stroke scan data.

[0010] Furthermore, the method for acquiring the original point cloud data of the entire surface also includes: after the scan data is stitched together, preprocessing the stitched scan data, including magnifying the z-coordinate data of each scan point by a preset factor; Then in the current point Coordinates minus the nearest bottom point When calculating coordinates, the result of the subtraction will be divided by the preset multiple, and the result of the division will be used as the corrected relative height of the current point.

[0011] Furthermore, a breadth-first search algorithm is used to cluster the retained point clouds.

[0012] Furthermore, the film thickness is calculated as follows: Points belonging to the substrate region in the first measurement point cloud are selected and fitted into a straight line using the least squares method, where the fitting formula is: ,in, , The first part of the substrate region i The x and z coordinates of each point, This represents the total number of points in the partial substrate region within the single measurement point cloud. and These are the slope and intercept of the fitted line, respectively; Based on the x-coordinate of each point in the measured point cloud and the fitted linear equation... Calculate the corresponding ,Will The z-coordinate of the point after correction is used to obtain the corrected first measurement point cloud; the z-coordinate of the points in the smooth area of ​​the printed film is selected from the corrected first measurement point cloud and the average value is taken as the approximate thickness.

[0013] Furthermore, the method for linear upsampling the reconstructed mesh is as follows: The reconstructed mesh is grouped into sets of three adjacent mesh points, and linear upsampling is performed on each set. The formula for linear upsampling is as follows: ,in, To determine the regular grid coordinates of the upsampled points in this group, These are the coordinates of the k-th grid point in the group; for The corresponding z-coordinate; The linear interpolation coefficients are derived from the equation. The solution is obtained.

[0014] Furthermore, the search is performed using a kd-tree nearest neighbor search algorithm: searching for points on the fitted surface of the entire substrate that are closest to the current point. closest bottom point And find its index, the calculation formula is as follows: .

[0015] Furthermore, after obtaining the corrected relative heights at each point, the method also includes: The relative height information of each point obtained by correction is projected along the z-direction to obtain a depth map. That is, the point is assigned a color according to the value of the corrected z-coordinate of each point in the point cloud to obtain a corrected topographic depth map. The particle detection algorithm, overflow detection algorithm and Mura detection algorithm are used on the depth map to realize the identification and localization of three types of defects.

[0016] According to another aspect of the present invention, an electronic device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the method described above.

[0017] According to another aspect of the invention, a computer-readable storage medium is provided, the computer-readable storage medium including a stored computer program, wherein, when the computer program is run by a processor, it controls the device where the storage medium is located to perform the steps of the method described above.

[0018] In summary, compared with the prior art, the technical solutions conceived by this invention have the following main advantages: 1. This invention provides a method for detecting the morphology of micron-level thin films. First, the point cloud of the substrate region and the point cloud of the flat area of ​​the printed film are separated from the original point cloud data of the entire surface, including the substrate region and the printed film. Then, the point cloud of the film edge, which spans the sloping area and includes parts of the substrate region and the flat area of ​​the printed film, is extracted. Based on this point cloud, the approximate thickness of the printed film is calculated. The point cloud of the flat area of ​​the printed film is shifted downwards by a distance equal to the approximate thickness. The shifted point cloud is then fused with the substrate point cloud. The fused point cloud is voxelized and downsampled to obtain a sparse point cloud. The sparse point cloud is then fitted to obtain the fitted surface of the entire substrate. Since this method is designed for printed films that meet uniformity requirements, it calculates the approximate thickness of the film using points at the edge of the printed film, shifts the point cloud of the flat area of ​​the film downwards by the approximate thickness of the film, and fuses it with the point cloud of the surrounding area of ​​the substrate, thus enabling the fitting of the substrate surface. Therefore, this invention enables the measurement of film thickness using linear scan confocal measurements at the edge, even when the precise refractive index of the film is unknown. The algorithm directly calculates the film thickness based on morphology data, and then morphology correction is performed based on the fitted substrate surface, specifically the correction of the relative height of each point. Thus, this method achieves overall film morphology correction even when the substrate is curved, enabling efficient online measurement and inspection of inkjet-printed films.

[0019] 2. This invention also proposes a method for stitching raw scan data of all scan areas in a multi-scale manner based on the ICP algorithm. For each pair of adjacent single passes, the point clouds of the overlapping scan areas of these two single passes are first used for coarse stitching and fine stitching in sequence to obtain a rigid transformation matrix. This matrix is ​​then used for the point cloud of the single pass scanned later in the adjacent pair of single passes to achieve stitching. Specifically, during coarse stitching, the initial rigid transformation matrix is ​​set according to the physical translation of the sensor, that is, the x-axis movement distance in the rigid transformation matrix is ​​set as the difference between the line width of the field of view and the overlap distance of adjacent point clouds. A relatively large distance threshold is set (the maximum allowable matching deviation of corresponding points in different passes in the overlapping area when transformed and matched by the rigid transformation matrix), and the coarse stitching rigid transformation matrix is ​​obtained through the ICP algorithm. Then, for fine stitching, the coarse stitching rigid transformation matrix is ​​used as the initial rigid transformation matrix, and a relatively small distance threshold is set (the minimum allowable matching deviation of corresponding points in different passes in the overlapping area when transformed and matched by the rigid transformation matrix), and the fine stitching rigid transformation matrix is ​​obtained through the ICP algorithm. This method selects two sets of point clouds from two adjacent single-pass overlapping scan regions for registration. This has two advantages: firstly, it avoids using all scan data from a single pass, reducing the size of the point cloud to be registered; secondly, it increases the proportion of matching points, resulting in more accurate matching. Furthermore, this method uses multi-scale stitching, which has two advantages: firstly, if a small matching distance threshold is directly set, it may be impossible to find the initial rigid transformation matrix; secondly, after coarse stitching, a very small matching distance threshold can be set for the second stitching, which helps improve stitching accuracy.

[0020] 3. This invention also proposes a method for approximate film thickness based on cloud computing of edge points of the printed film. By setting the measurement range to span the substrate area, the sloping area of ​​the printed film, and the flat area of ​​the printed film, a measurement object similar to a step meter can be obtained. First, a linear fit is performed on some points in the substrate area. Based on the fitting equation, the z-axis height of the point cloud in the substrate area, the sloping area of ​​the printed film, and the flat area is corrected. That is, the entire extracted single-measurement point cloud is corrected based on the fitting equation. Then, the z-axis coordinates of points in the flat area of ​​the printed film are selected from the corrected single-measurement point cloud and the average value is taken as the approximate thickness of the printed film. Since this method extracts the original point cloud of a single measurement, that is, without preprocessing, including without rigid transformation matrix splicing, using such data for linear fitting and correction can more realistically reflect the true information of the film under a single measurement in the edge area of ​​the printed film. Furthermore, since the detection method of this invention is aimed at printed films that meet the requirements of uniformity and consistency, the film thickness obtained by the above method can be basically used as the thickness of the entire printed film. The film thickness calculation result has high reliability, ensuring the accuracy of subsequent substrate surface fitting, and thus ensuring the accuracy of the final surface morphology correction.

[0021] 4. When preprocessing the spliced ​​point cloud data of the entire surface, this invention proposes to magnify the z-coordinate data of each scanning point by a preset factor. Since the film thickness is on the micrometer level and the length and width are on the decimeter level, which is four orders of magnitude different, magnifying the z-coordinate can make the characteristics of the film more obvious, which facilitates subsequent operations and improves the accuracy of computation and processing. Attached Figure Description

[0022] Figure 1 This is a flowchart of a method for detecting the morphology of a micron-scale thin film provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of morphology correction provided in an embodiment of the present invention; Figure 3 The measurement method of the linear scan confocal sensor provided in the embodiments of the present invention; Figure 4 The scanning method of the linear confocal sensor provided in the embodiments of the present invention; Figure 5 This is a schematic diagram of the point cloud stitching method provided in an embodiment of the present invention; Figure 6 A schematic diagram illustrating the partitioning of the thin film flat region, the thin film ramp region, and the substrate region provided in an embodiment of the present invention; Figure 7 A flowchart of a point cloud segmentation algorithm provided in an embodiment of the present invention; Figure 8 A flowchart of the breadth-first search algorithm provided in an embodiment of the present invention; Figure 9 This is a schematic diagram of two-dimensional thin film morphology correction provided in an embodiment of the present invention; Figure 10 This is a schematic diagram of thin film detection parameters provided in an embodiment of the present invention; Figure 11 A schematic diagram of a thin film Mulla defect provided in an embodiment of the present invention; Figure 12 A schematic diagram of the depth map of a thin film with defects provided in an embodiment of the present invention; Figure 13 This is a schematic diagram of the detection device corresponding to the detection method of the present invention provided in an embodiment of the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0024] Example 1 A method for detecting the morphology of micron-scale thin films, such as Figure 1 As shown, it includes: The original point cloud data of the entire surface, including the substrate area and the printed film, is acquired and stitched together using a line scan confocal sensor. The normal vector of each point in the stitched point cloud is calculated. The normal vector of each point is normalized, and the angle between the normalized normal vector of each point and the vertical unit vector is calculated as the tilt of that point. Points in the point cloud with a tilt exceeding a preset threshold are deleted. The remaining point cloud is clustered based on distance to obtain the point cloud of the substrate area and the point cloud of the smooth area of ​​the printed film. The original point cloud of a single-pass scan corresponding to the edge of the printed film is extracted. The film thickness is calculated based on this original point cloud, and the calculation result is used as the approximate thickness of the printed film. The edge of the printed film spans the sloping area of ​​the printed film and includes part of the substrate area and part of the flat area of ​​the printed film. The point cloud of the flat area of ​​the printed film is translated downward by a distance equal to the approximate thickness. The translated point cloud is fused with the substrate point cloud. The fused point cloud is voxelized and downsampled to obtain a sparse point cloud. The Poisson fitting algorithm is used to fit the sparse point cloud to obtain the implicit function of the fitted surface of the entire substrate. The reconstructed mesh of the entire substrate is obtained based on the implicit function. The reconstructed mesh is linearly upsampled to obtain the fitted surface of the entire substrate. The original point cloud data of the entire surface is voxelized downsampled and denoised. Each point in the processed point cloud is traversed. The nearest bottom point in the x and y directions of the current point is found among the points of the fitted surface of the entire substrate. The z-coordinate of the current point is subtracted from the z-coordinate of the nearest bottom point to obtain the corrected relative height of the current point, thus completing the micron-level thin film morphology detection.

[0025] The method in this embodiment is applied to the detection of printed films that meet the requirements of uniformity and consistency. It only corrects the microscopically curved morphology of the film, such as... Figure 2 As shown.

[0026] A linear scan confocal sensor is used to measure various regions of the entire surface. The overall film morphology is obtained by stitching the data together, and point cloud segmentation is achieved through normal vector filtering and clustering algorithms. The approximate film thickness is calculated from the edge morphology. The point cloud of the smooth area of ​​the printed film is shifted downwards by a distance equal to the approximate thickness. The shifted point cloud is then fused with the substrate point cloud, and Poisson fitting is used to obtain the overall substrate morphology. This is then used to reconstruct the corrected film morphology. In other words, this embodiment addresses the situation where the precise value of the film refractive index is unknown during film measurement, and where the substrate may exhibit a curved surface due to warping. This embodiment proposes a film morphology correction algorithm. The corrected morphology can be used for film thickness measurement; film edge morphology measurement (including ramp distance and protrusion height); and film defect detection (including particles, overflow, and Moura).

[0027] In practice, a linear scan confocal sensor is first used to acquire thin film point cloud data. Then, an off-axis linear scan confocal sensor is used to measure the micron-scale thin film on a transparent glass substrate, such as... Figure 3 , Figure 4 As shown, the sensor moves along the Y-axis until a single scan is completed. Then, the sensor translates along the X-axis, scanning neighboring areas. A certain overlap distance is set in the X-axis when scanning two adjacent areas to stitch together adjacent point clouds. The entire thin film is scanned sequentially using a line-scan confocal sensor to obtain point cloud data.

[0028] Using the ICP algorithm to stitch adjacent point clouds is a preferred implementation method. Based on the ICP algorithm, the scan data of all scanned areas are stitched together in a multi-scale manner. (1) Between the scan data of each adjacent single-stroke overlapping area, firstly, the rigid transformation matrix in the ICP algorithm is initialized as the matrix of sensor physical translation. Based on a preset large matching point deviation threshold, the rigid transformation matrix used to achieve coarse stitching is obtained through the ICP algorithm. Then, the rigid transformation matrix in the ICP algorithm is initialized as the rigid transformation matrix used to achieve coarse stitching. Based on a preset small matching point deviation, the rigid transformation matrix used to achieve fine stitching is obtained through the ICP algorithm. The formula of the ICP algorithm is: In the formula, To obtain a rigid transformation matrix from the source point cloud to the target point cloud, by and composition, Here is the rotation transformation matrix; The translation transformation matrix is ​​used; the source point cloud is the partial scan data corresponding to the overlapping area of ​​the scanned single stroke in the latter scanned single stroke and the scanned single stroke in the former scanned single stroke in the two adjacent single strokes; the target point cloud is the partial scan data corresponding to the overlapping area of ​​the scanned single stroke in the former scanned single stroke and the scanned single stroke in the latter scanned single stroke in the two adjacent single strokes. The first point cloud in the source cloud i One point; For points in the target point cloud, nearest neighbor matching is used to find points in the source point cloud. The corresponding matching point; (2) By sequentially multiplying the rigid transformation matrix obtained when each single stroke is used as the source point cloud with the rigid transformation matrices of each single stroke before the single stroke scan as the source point cloud, the result of the cumulative multiplication is used as the rigid transformation matrix of the single stroke scan data and spliced ​​into the first stroke scan data. (3) Use a rigid transformation matrix to stitch each single-stroke scan data into the first stroke scan data, transform the single-stroke scan data, and stitch all the transformed scan data with the first stroke scan data.

[0029] As shown in the splicing diagram Figure 5 As shown. During stitching, the overlapping region on the left side of the second point cloud is selected as the source point cloud, and the overlapping region on the right side of the first point cloud is selected as the target point cloud. The transformation matrix is ​​solved using the ICP algorithm, and then the transformation matrix is ​​applied to the entire second point cloud to achieve stitching. This operation is repeated for subsequent point clouds to achieve stitching of the entire film morphology.

[0030] More specifically, a multi-scale stitching method is used to stitch adjacent point clouds together. During coarse stitching, an initial transformation matrix is ​​set based on the physical translation of the sensor; that is, the distance moved in the x-direction of the transformation matrix is ​​set to the difference between the linewidth of the field of view and the overlap distance of adjacent point clouds. A relatively large distance threshold is set (limiting the maximum allowed inter-point distance during transformation matching), and the coarse stitching transformation matrix is ​​obtained through the ICP algorithm. The stitching effect is mainly evaluated based on the matching ratio (fitness) and the root mean square error (RMSE) of the matching points. The formula for calculating the matching ratio is:

[0031] Where N inlier N is the number of source point cloud points that match the target point cloud points. source The number of source point cloud points.

[0032] The root mean square error of the matching points is the root mean square value of the distance between successfully registered corresponding points, and its calculation formula is as follows:

[0033] Then, fine stitching is performed. The coarse stitching transformation matrix is ​​used as the initial transformation matrix, a small distance threshold is set, and the fine stitching transformation matrix is ​​obtained through the ICP algorithm.

[0034] Preprocessing the stitched point cloud, as a preferred implementation method, includes: voxel downsampling to reduce the size of the point cloud for easier processing; magnifying the z-coordinate of the points by 100 times (exemplary), since the film thickness is on the micrometer scale while the length and width are on the decimeter scale, differing by four orders of magnitude, magnifying the z-coordinate can make the film's features more apparent, facilitating subsequent operations; and point cloud denoising, removing noise points far from the main point cloud through statistical methods. The denoising method is as follows: (1) Calculate the average distance from each point in the point cloud to its nearest neighbor. The calculation formula is as follows:

[0035] Where p i Let i be the i-th point in the point cloud. Let d be the j-th nearest neighbor of this point. i is the average distance to that point, and k is the number of nearest neighbors; (2) Calculate the mean of the average distances of all points in the point cloud. and variance The calculation formula is:

[0036]

[0037] Where N is the number of points in the point cloud; (3) Identify outliers. If the average distance of a point is less than a threshold, retain it; otherwise, remove it, as shown in the following formula: ,in is a coefficient. .

[0038] Next, the verticality of each point in the point cloud is calculated, thus obtaining the tilt of each point, and a threshold is set. The point cloud is then filtered by tilt, removing points at the slopes of the thin film, resulting in point clouds of the flat portion of the thin film and the substrate portion. At this point, there is a gap between the two point clouds, which can be separated using a clustering algorithm.

[0039] Regarding the point cloud clustering and segmentation method in this embodiment: Since the thin film on the glass substrate has certain geometric features, it can be segmented based on these features. The thin film point cloud can be divided into three parts: the substrate region, the printing film ramp region, and the printing film flat region. The normal vector of each point in the point cloud is calculated. The covariance of each point is calculated using the following formula: Where C is the covariance matrix of the point, and k is the number of nearest neighbors of the point determined under the constraints of a preset radius and a preset maximum number of nearest neighbors. It is the j-th nearest neighbor of this point. Let C be the centroid of its k nearest neighbors. The formula for calculating the eigenvalues ​​of C is: ,in Let E be the eigenvalues, E be the identity matrix, and C be the covariance matrix. Given eigenvectors, find the eigenvalues ​​of the covariance matrix. (in and the corresponding feature vectors , That is, the normal vector of that point ( Convert the normal vector to a unit vector using the following formula: ,in Let be the normal vector of the i-th point. They are respectively The x, y, and z components, Let be the unit normal vector of the i-th point. Then, calculate the slope of each point using the following formula: .in Let be the slope of that point, and z be a unit vector along the positive z-axis, i.e., (0,0,1). The unit normal vector at that point The magnitude of the z-component. For the smooth regions of the substrate and thin film, the normal vector is almost parallel to the z-axis, so its tilt is small. For the steep regions of the thin film, the normal vector makes a larger angle with the z-axis, so its tilt is larger. By setting a tilt threshold, points with large tilt are filtered out, and points with tilt less than the threshold are the point clouds of the smooth regions of the substrate and thin film. Since the points in the steep regions of the thin film are deleted, the distance between the substrate point cloud and the thin film point cloud is increased, which is equivalent to being scattered. Then, using a breadth-first search algorithm (as a preferred implementation), a distance threshold is set, for example, π / 3, and the point cloud is divided into two parts. i Points and the j Points Distance between The calculation formula is: ,in, , , and , , They are respectively and The coordinates are used for clustering; points with a distance less than a threshold are grouped into one class, and points with a distance greater than the threshold are grouped into two different classes. Specifically, points with a slope less than the threshold indicate a relatively flat area, while points with a slope greater than the threshold indicate a relatively steep area. This allows for the separation of flat and steep portions of the point cloud. The steep portions represent the sloping areas at the film edges, thus the point cloud can be divided into two parts: the flat areas on the substrate and film, and the sloping areas of the film. Points in the sloping areas of the film are then removed, such as... Figure 6 As shown.

[0040] Based on the distances between points in the point cloud, a breadth-first search algorithm is used to cluster the point cloud in the flat areas. The distance calculation formula is as follows: Set a distance threshold. Starting from the initial point, search for nearby points within the threshold. If a point is found, add it to the category. Repeat this process until all unclassified points have been traversed. The flowchart of the clustering algorithm is as follows: Figure 7 As shown, the flowchart of the breadth-first search algorithm is as follows: Figure 8 As shown. The point clouds are sorted according to the data size of each point cloud to obtain the point clouds of the surrounding substrate and the point clouds of the flat top surface of the thin film.

[0041] Regarding the substrate morphology fitting in this embodiment: For most printed films, the film thickness is approximately uniform because the liquid will level out in the absence of external interference (that is, the film thickness is theoretically uniform because the droplets will level out after printing). Point clouds are extracted from the film edges (across the substrate area, the ramp area, and the flat area), and the film thickness is calculated. The calculated result is considered as the approximate thickness of the printed film. The point cloud in the flat area of ​​the film is shifted downwards by a distance equal to the approximate film thickness and then fused with the substrate point cloud. The fused point cloud is voxelized and downsampled to obtain a sparse point cloud. A Poisson fitting algorithm is then used to fit the sparse point cloud: the Poisson equation is: ,in The implicit function of the fitted surface for the required solution is the isosurface of the fitted surface to be reconstructed. is the Laplace operator; V is a continuous normal vector field constructed from the sparse point cloud; It is the divergence of V.

[0042] Then, based on the implicit function, a reconstructed mesh of the entire substrate is obtained, and linear upsampling is performed on the reconstructed mesh to obtain the fitted surface of the entire substrate. As a preferred embodiment, every three adjacent mesh points of the reconstructed mesh are grouped together, and linear upsampling is performed on each group. The linear upsampling calculation formula is as follows: ,in, To determine the regular grid coordinates of the upsampled points in this group, These are the coordinates of the k-th grid point in the group; for The corresponding z-coordinate; The linear interpolation coefficients are derived from the equation. The solution is obtained.

[0043] Regarding the calculation of the film thickness at the edge of the printed film, since a reference surface is required for measuring the printed film thickness, only the point cloud at the edge of the printed film can be used to calculate the film thickness. Specifically, in the linear confocal scanning process, the sensor's line profile needs to span both the substrate and the film edge for the highest measurement accuracy. The calculation process, which can be considered a preferred implementation, involves selecting points belonging to the aforementioned portion of the substrate region from the point cloud of the first measurement and fitting them to a straight line using the least squares method, where the fitting formula is... ,in, , The first part of the substrate region i The x and z coordinates of each point, This represents the total number of points in the partial substrate region within the single measurement point cloud. and These are the slope and intercept of the fitted line, respectively; based on the x-coordinate of each point in the measured point cloud and the equation of the fitted line. Calculate the corresponding ,Will The z-coordinate of the point after correction is used to obtain the corrected first measurement point cloud; the z-coordinate of the points in the smooth area of ​​the printed film is selected from the corrected first measurement point cloud and the average value is taken as the approximate thickness.

[0044] For the correction of points on a single line at the edge (i.e., the point cloud of a single measurement), the specific algorithm is as follows: Select points on the substrate and perform linear fitting on them, denoted as l. Use the least squares method to minimize the fitting error; the calculation formula is:

[0045]

[0046]

[0047] Where a and b are the slope and intercept of the fitted line, respectively.

[0048] Keeping the x-coordinate of each point on the line constant, the corresponding (x, z0) is calculated according to the equation of line l, and (x, z-z0) is used as the contour of the corrected edge. The film thickness is obtained based on the corrected morphology. The film morphologies before and after correction are shown below. Figure 9 As shown. Through correction, the film thickness, protrusion height, and ramp distance can be obtained, such as... Figure 10 As shown.

[0049] Furthermore, since the droplets level out after printing, the film thickness is theoretically uniform. The flat portion of the top surface of the film is translated downwards; the point cloud of this portion should represent the topography of the substrate at the bottom of the film. The translated point cloud is then merged with the point clouds from the segmented perimeter of the substrate to obtain a new point cloud. This new point cloud is downsampled and then subjected to Poisson fitting. It is crucial to set the Poisson fitting depth appropriately; excessive depth will result in overfitting of the fitted surface. In this embodiment, the Poisson fitting depth is set to 5, resulting in a mesh point cloud.

[0050] Regarding the method or thin film morphology correction in this embodiment, for the original point cloud encompassing the entire surface of the substrate and the thin film, voxelization downsampling and noise reduction are performed, traversing each point P in the processed point cloud. i (x i , y i , z i Using the kd-tree nearest neighbor search algorithm, a match is found among the points on the fitted surface of the entire substrate at the current point. closest bottom point And find its index, the calculation formula is as follows: Find the point at the bottom that corresponds to the top point based on the index. , top point Coordinates minus the corresponding bottom point The coordinates are then divided by the aforementioned magnification factor 'a' (to offset the previous magnification) to obtain the corrected relative height. The calculation formula is as follows: .

[0051] By drawing lines along the sensor's motion direction after correction, relatively obvious Mura features can be obtained, such as... Figure 11 As shown.

[0052] As a preferred embodiment, after obtaining the corrected relative heights at each point, the method further includes: The relative height information of each point obtained by correction is projected along the z-direction to obtain a depth map. That is, the point is assigned a color according to the value of the corrected z-coordinate of each point in the point cloud to obtain a corrected topographic depth map. The particle detection algorithm, overflow detection algorithm and Mura detection algorithm are used on the depth map to realize the identification and localization of three types of defects.

[0053] Here, the point cloud is assigned a color based on the numerical value of the z-coordinate of each point, as shown below:

[0054] in For the z-coordinate of each point, , These are the maximum and minimum z-coordinates in the point cloud, respectively. The depth map is obtained by color mapping based on the normalized z-coordinate. For example... Figure 12 As shown, particle detection algorithm, overflow detection algorithm and MLA detection algorithm are used on the depth map to identify and locate three types of defects.

[0055] Specifically, the particle detection algorithm and overflow detection algorithm involve: First, color channel decomposition of the image to obtain the r-channel portion. Then, an opening operation is performed on the image to remove burrs, followed by a closing operation to fill gaps, revealing the region of the thin film in the original image. Next, this region is cropped from the original image to obtain the thin film portion. The image after subtracting the mean is then segmented using dynamic thresholding to obtain particle defects. Next, color channel decomposition of the image yields the b-channel portion, thus revealing the substrate portion. A closing operation is performed on the image to fill gaps, and then subtracted from the image before the closing operation to obtain the gap features. Finally, an opening operation is performed on the image to separate each overflow defect.

[0056] Specifically, the Mura detection algorithm involves preprocessing the image, performing a forward Fourier transform, and then convolving it with a bandpass filter in the y-direction to identify horizontal stripe features in the image. The filter calculation formula is as follows: , ,in It is a two-dimensional Gaussian filter; The standard deviation is Gaussian (to control the degree of smoothing). A differential Gaussian filter is used. Then, an inverse Fourier transform is performed, and finally, threshold segmentation and aspect ratio filtering are used to obtain the Mura defects.

[0057] The method in this embodiment can be implemented by a detection system, such as... Figure 13 As shown, the system may include an off-axis scanning confocal sensor, a marble platform, a gantry, a high-precision linear motor, a sample stage, and an industrial control computer. The off-axis scanning confocal sensor is used to measure the thin film morphology; the marble platform ensures the system is less affected by vibration; the sample stage is driven by a linear motor for high-precision sample movement; and the industrial control computer runs the detection algorithm proposed in this embodiment.

[0058] During system measurement, the linear confocal sensor remains stationary while the stage moves along the X-axis until a single scan is completed. The linear confocal sensor then moves along the Y-axis for the next measurement. Adjacent measurements must have a certain overlap in the Y-axis for point cloud stitching. This process is repeated until the topography of the entire surface is measured, obtaining point cloud data for each region. To facilitate subsequent point cloud stitching, a certain overlap area must be set between adjacent scan areas; typically, the overlap distance in the X-axis is set to 5-10% of the linewidth.

[0059] The core of this method is to move the point cloud of the flat area of ​​the thin film downwards by approximately the thickness of the film, and then fuse it with the point cloud of the surrounding area of ​​the substrate to fit the curved surface of the substrate. If the thickness uniformity of the thin film is poor, the premise of this algorithm cannot be achieved. Therefore, the purpose of the algorithm proposed in this embodiment is to further detect the quality of the thin film while meeting certain thickness uniformity requirements. If the thickness uniformity of the thin film is poor, there is no need for further detection. The main effect is that, since the length and width of the thin film are much greater than its thickness, and the substrate is difficult to guarantee as a completely flat surface, the original morphology data collected by the line spectrum confocal sensor usually presents a curved surface, and the depth map obtained by projection cannot provide effective information about the quality of the thin film. After correction using this algorithm, the relative height information of the thin film can be obtained, making the characteristics of thin film defects, such as particles, overflows, and Mula, more obvious, which is beneficial for subsequent image processing to detect thin film defects. Therefore, the method of this embodiment can still achieve the correction of the overall morphology of the thin film when the substrate is curved. When the substrate is under certain conditions (such as four-corner support), it exhibits a curved surface on a microscopic scale. Traditional planar fitting algorithms (such as RANSAC) cannot be used to fit the substrate during large-scale measurements, thus failing to correct the film morphology and obtain its relative height. Using the morphology correction algorithm of this embodiment allows for a more intuitive display of the film morphology. Furthermore, by processing the depth map of the corrected morphology, film defect detection can be achieved. Directly projecting the original point cloud to obtain a depth map results in unclear film features, making it difficult to detect various defects. Moreover, some defects in the depth map of the original point cloud may be film defects, while others may be optical noise, leading to misjudgments. Using this morphology correction algorithm and then processing to obtain a depth map makes film defect detection much easier.

[0060] Example 2 This application also relates to an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method described above.

[0061] The electronic device can be a desktop computer, laptop, handheld computer, or cloud server, etc. The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The memory can be used to store computer programs and / or modules. The processor performs various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory, and by accessing data stored in the memory.

[0062] The relevant technical solutions are the same as above, and will not be repeated here.

[0063] Example 3 This application also relates to a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described above.

[0064] Specifically, the memory may include high-speed random access memory, as well as non-volatile memory, such as hard disks, RAM, plug-in hard disks, smart media cards (SMC), secure digital (SD) cards, flash cards, at least one disk storage device, flash memory device, or other volatile solid-state storage devices.

[0065] The relevant technical solutions are the same as above, and will not be repeated here.

[0066] In summary, this embodiment provides a method for measuring and detecting thin film morphology over a wide range. It enables the measurement of thin film thickness using edge-scan confocal measurements even when the precise refractive index of the film is unknown. It also allows for the correction of the overall thin film morphology when the substrate is curved. When the substrate is under certain conditions (such as corner support), it exhibits a microscopic curved surface. Traditional planar fitting algorithms (such as RANSAC) cannot be used to fit the substrate during large-scale measurements, thus failing to correct the thin film morphology and obtain its relative height. Using the morphology correction algorithm of this embodiment allows for a more intuitive display of the thin film morphology. Furthermore, by performing image processing on the depth map of the corrected morphology, thin film defect detection can also be achieved.

[0067] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for detecting the morphology of micron-scale thin films, characterized in that, include: The original point cloud data of the entire surface, including the substrate area and the printed film, is acquired and stitched together using a line scan confocal sensor, and the normal vector of each point in the stitched point cloud is calculated. Normalize the normal vector of each point, and calculate the angle between the normalized normal vector of each point and the vertical unit vector, which is taken as the inclination of that point; Delete points in the point cloud whose tilt exceeds a preset threshold; cluster the remaining point cloud based on distance to obtain the point cloud of the substrate area and the point cloud of the smooth area of ​​the printed film. The original point cloud of a single-pass scan corresponding to the edge of the printed film is extracted, and the film thickness is calculated based on this original point cloud. The calculation result is used as the approximate thickness of the printed film. The edge of the printed film spans the sloping area of ​​the printed film and includes part of the substrate area and part of the flat area of ​​the printed film. The point cloud of the flat area of ​​the printed film is translated downwards by a distance equal to the approximate thickness. The translated point cloud is fused with the substrate point cloud. The fused point cloud is voxelized and downsampled to obtain a sparse point cloud. The Poisson fitting algorithm is used to fit the sparse point cloud to obtain the implicit function of the fitted surface of the entire substrate. The reconstructed mesh of the entire substrate is obtained based on the implicit function, and the reconstructed mesh is linearly upsampled to obtain the fitted surface of the entire substrate. The original point cloud data of the entire surface is voxelized downsampled and denoised. Each point in the processed point cloud is traversed. The nearest bottom point in the x and y directions of the current point is found among the points of the fitted surface of the entire substrate. The z-coordinate of the current point is subtracted from the z-coordinate of the nearest bottom point to obtain the corrected relative height of the current point, thus completing the micron-level thin film morphology detection.

2. The method for detecting the morphology of micron-scale thin films as described in claim 1, characterized in that, The method for acquiring the original point cloud data of the entire surface is as follows: a line scan confocal sensor is used to scan the entire surface, including the substrate area and the printed film, through multi-pass line scan, and the scan data of all single passes are obtained. Among them, the scan areas of two adjacent single passes have some overlap. Based on the ICP algorithm, the scan data of all scan areas are stitched together in a multi-scale manner: (1) Between the scan data of each adjacent single-stroke overlapping area, firstly, the rigid transformation matrix in the ICP algorithm is initialized as the physical translation matrix of the sensor, with a large preset matching point deviation threshold. The rigid transformation matrix used to achieve coarse stitching is obtained through the ICP algorithm. Then, the rigid transformation matrix in the ICP algorithm is initialized as the rigid transformation matrix used to achieve coarse stitching, with a small preset matching point deviation. The rigid transformation matrix used to achieve fine stitching is obtained through the ICP algorithm. The formula of the ICP algorithm is: In the formula, To obtain a rigid transformation matrix from the source point cloud to the target point cloud, by and composition, Here is the rotation transformation matrix; The translation transformation matrix is ​​used; the source point cloud is the partial scan data corresponding to the overlapping area of ​​the scanned single stroke in the latter scanned single stroke and the scanned single stroke in the former scanned single stroke in the two adjacent single strokes; the target point cloud is the partial scan data corresponding to the overlapping area of ​​the scanned single stroke in the former scanned single stroke and the scanned single stroke in the latter scanned single stroke in the two adjacent single strokes. The first point cloud in the source cloud i One point; For points in the target point cloud, nearest neighbor matching is used to find points in the source point cloud. The corresponding matching point; (2) By sequentially multiplying the rigid transformation matrix obtained when each single stroke is used as the source point cloud with the rigid transformation matrices of each single stroke before the single stroke scan as the source point cloud, the result of the cumulative multiplication is used as the rigid transformation matrix of the single stroke scan data and spliced ​​into the first stroke scan data. (3) Use a rigid transformation matrix that splices each single-stroke scan data to the first stroke scan data to transform the single-stroke scan data, and then splice all the transformed scan data with the first stroke scan data.

3. The method for detecting the morphology of micron-scale thin films as described in claim 2, characterized in that, The method for acquiring the original point cloud data of the entire surface also includes: after the scan data is stitched together, preprocessing the stitched scan data, including enlarging the z coordinate data of each scan point by a preset factor; Then in the current point Coordinates minus the nearest bottom point When calculating coordinates, the result of the subtraction will be divided by the preset multiple, and the result of the division will be used as the corrected relative height of the current point.

4. The method for detecting the morphology of micron-scale thin films as described in claim 1, characterized in that, A breadth-first search algorithm is used to cluster the retained point clouds.

5. The method for detecting the morphology of micron-scale thin films as described in claim 1, characterized in that, The film thickness is calculated as follows: Points belonging to the substrate region in the first measurement point cloud are selected and fitted into a straight line using the least squares method, where the fitting formula is: ,in, , The first part of the substrate region i The x and z coordinates of each point, This represents the total number of points in the partial substrate region within the single measurement point cloud. and These are the slope and intercept of the fitted line, respectively; Based on the x-coordinate of each point in the measured point cloud and the fitted linear equation... Calculate the corresponding ,Will The z-coordinate of the point after correction is used to obtain the corrected first measurement point cloud; the z-coordinate of the points in the smooth area of ​​the printed film is selected from the corrected first measurement point cloud and the average value is taken as the approximate thickness.

6. The method for detecting the morphology of micron-scale thin films as described in claim 1, characterized in that, The method for linear upsampling the reconstructed mesh is as follows: The reconstructed mesh is grouped into sets of three adjacent mesh points, and linear upsampling is performed on each set. The formula for linear upsampling is as follows: ,in, To determine the regular grid coordinates of the upsampled points in this group, These are the coordinates of the k-th grid point in the group; for The corresponding z-coordinate; The linear interpolation coefficients are derived from the equation. The solution is obtained.

7. The method for detecting the morphology of micron-scale thin films as described in claim 1, characterized in that, The search is performed using a kd-tree nearest neighbor search algorithm: finding points on the fitted surface of the entire substrate that are closest to the current point. closest bottom point And find its index, the calculation formula is as follows: .

8. The method for detecting the morphology of micron-scale thin films as described in claim 1, characterized in that, After obtaining the corrected relative heights at each point, the method also includes: The relative height information of each point obtained by correction is projected along the z-direction to obtain a depth map. That is, the point is assigned a color according to the value of the corrected z-coordinate of each point in the point cloud to obtain a corrected topographic depth map. The particle detection algorithm, overflow detection algorithm and Mura detection algorithm are used on the depth map to realize the identification and localization of three types of defects.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein the computer program, when executed by a processor, controls the device on which the storage medium resides to perform the steps of the method as described in any one of claims 1 to 8.

Images