A method for detecting crystal cells of a semiconductor crystal growth furnace in a low-resolution scenario

Abstract

The present invention discloses a method for detecting crystal cells of a semiconductor crystal growth furnace in a low-resolution scenario, which relates to the technical field of semiconductor crystal growth furnaces and includes the following steps: acquiring an original image of crystal seeding through a global industrial camera in advance, calibrating the double-aperture part of the original image of crystal seeding as a target feature for cropping, and obtaining a target image containing the double-aperture part; synchronously performing hollow detection and crystal cell forming detection on the target image respectively; outputting the detection results of the hollow detection and the crystal cell forming detection, and determining whether crystal cells are generated in the current seeding stage. The present invention helps to more accurately judge the state of crystal cells, can automatically detect crystal cells, reduces the manual burden, and realizes the accuracy, efficiency and reliability of detection.

Description

Technical Field

[0001] The present invention relates to the technical field of semiconductor crystal growth furnaces. Specifically, it relates to a method for detecting crystal cells of a semiconductor crystal growth furnace in a low-resolution scenario. Background Art

[0002] A crystal cell is a phenomenon that occurs during the seed stage. Small crystals similar to a lattice structure grow on the finished crystal, which is related to the production process. When crystal cells appear, it is necessary to control the temperature, crucible rotation speed, and pulling speed to further control the crystal growth. Moreover, the generation of crystal cells is also related to the formation of the crystal column shape later, so it is necessary to observe the moment when crystal cells appear.

[0003] Currently, when crystal cells appear during the seed stage, the diameter of the crystal column is very small, about 10 - 20 mm. Traditionally, the manual observation method is that workers use an optical microscope to visually observe whether crystal cells appear through the filter of the flange observation port. However, there are mainly the following disadvantages:

[0004] 1. Limited resolution: The resolution of an optical microscope is limited by the wavelength of light. Generally, the resolution limit of an optical microscope is about 0.2 micrometers. For crystal cells with such microscopic structures, the sizes of many crystal cells are at the nanometer level. For example, the edge length of the crystal cell of a common metal crystal may be between several nanometers and dozens of nanometers. In this case, it is very difficult for an optical microscope to clearly distinguish the shape and structural details of a single crystal cell, which may lead to inaccurate observation results and an inability to accurately judge whether crystal cells appear. At the same time, the depth of field of an optical microscope is relatively shallow. If the sample where the crystal cells are located has a certain thickness, or the crystal cells are distributed at different depth levels, it is very difficult to simultaneously focus on observing all the crystal cells, thus easily missing some information of the crystal cells.

[0005] 2. Limited observation angle: Observing through the flange observation port, the observation angle is relatively fixed. This makes it very difficult to observe crystal cells from multiple angles. For some crystal cells with complex three-dimensional structures, a single-angle observation may not comprehensively understand the true shape and distribution of the crystal cells. Just like observing a three-dimensional sculpture, looking only from one direction may ignore some important features on the back or side of the sculpture, thus causing deviations in the judgment of whether crystal cells appear and their state.

[0006] 3. Strong subjectivity: Visual observation depends on the vision and experience of workers. Different workers may make different judgments on whether crystal cells appear due to factors such as vision differences and differences in the understanding of crystal cell morphology. For example, an inexperienced worker may misjudge some impurities or other microscopic structures as crystal cells, or conversely, ignore some newly formed tiny crystal cells. This subjectivity-strong judgment method is not conducive to accurately and consistently determining the appearance of crystal cells.

[0007] In the related art, no effective solution has been proposed for the problems yet. Summary of the Invention

[0008] In view of the problems in the related art, the present invention provides a method for detecting crystal cells of a semiconductor crystal growth furnace in a low-resolution scenario, so as to overcome the above technical problems existing in the related art.

[0009] The technical solution of the present invention is realized as follows:

[0010] A method for detecting crystal cells of a semiconductor crystal growth furnace in a low-resolution scenario includes the following steps:

[0011] Pre-collect the original image of crystal seeding through a global industrial camera, and calibrate the double-aperture part of the original image of crystal seeding as the target feature for cropping to obtain a target image including the double-aperture part;

[0012] Perform hollow detection and crystal cell forming detection on the target image synchronously, where: the hollow detection is used to detect whether the current target image is a hollow polygon, and the crystal cell forming detection is used to detect whether the current target image is multiple lines;

[0013] Output the detection results of the hollow detection and the crystal cell forming detection, and determine whether a crystal cell is generated in the current seeding stage, where: if the output of the hollow detection is a hollow polygon and the output of the crystal cell forming detection is multiple lines, it indicates that a crystal cell is generated in the current seeding stage.

[0014] Further, the obtaining of the target image including the double-aperture part includes the following steps:

[0015] Convert the obtained target image including the double-aperture part into a binary image.

[0016] Further, the hollow detection of the target image includes the following steps:

[0017] Read and process the binary image: load and process the input binary image to ensure that the image only contains two gray values of black and white;

[0018] Find the contours in the image: obtain all external contours through the findContours function, including: use the cv::findContours function to find all the contours in the image;

[0019] Traverse each found contour, where: for each contour, check whether the size of its point set is greater than the point set threshold, and filter out the contours with the size of the point set less than the point set threshold to filter out too small contours;

[0020] Check the number of corner points of the filtered contour, determine whether the number of points of the contour is greater than the point number threshold, and eliminate the contours with the number of points less than the point number threshold;

[0021] Perform convex hull detection on the eliminated contours, calculate the convex hull through the convexHull function, and obtain the smallest convex polygon that can enclose the contour point set;

[0022] Obtain the hollow detection result, including: comparing the number of points of the obtained external contour with the number of points of the convex hull, including the following steps:

[0023] If the number of points of the external contour is equal to the number of points of the convex hull, it means that the current contour is not a hollow polygon, and output the hollow detection result as no;

[0024] If the number of points of the external contour is not equal to the number of points of the convex hull, it means that the current contour is a hollow polygon, and draw the contour containing the hollow polygon as the hollow detection output, that is, it means that the target image has unit cells.

[0025] Furthermore, perform unit cell forming detection on the target image, including the following steps:

[0026] Read and process the binary image: load and process the input binary image to ensure that the image only contains two gray values, black and white;

[0027] Find the contours in the image: obtain all external contours through the findContours function, including: using the cv::findContours function to find all the contours in the image;

[0028] Traverse each found contour, including: for each contour, check whether the size of its point set is greater than the point set threshold, and filter out the contours with the size of the point set less than the point set threshold to filter out too small contours;

[0029] Calculate the perimeter of each contour, including: obtain the complete length of the contour through the arcLength function,

[0030] Judge whether the contour meets the continuous line according to the set line threshold, including: set the number of image segments to segment the contour, and regard the perimeter of the segmented contour greater than the set line threshold as meeting the line, and obtain the number of lines that meet the line;

[0031] Compare the number of lines that meet the line with the preset line threshold to judge whether the current unsegmented contour conforms to multiple lines, including: if the current number of lines that meet the line is greater than the line threshold, the current unsegmented contour conforms to multiple lines.

[0032] The beneficial effects of the present invention:

[0033] The present invention has high detection accuracy and fast detection speed. It collects the original image of crystal seeding through a global industrial camera, and simultaneously performs hollow detection and cell forming detection on the target image, outputs the detection results of hollow detection and cell forming detection, determines whether cells are generated in the current seeding stage, helps to more accurately judge the state of cells, can automatically detect cells, reduces the manual burden, and realizes the accuracy, efficiency and reliability of detection. At the same time, the present invention has a wide range of applicable scenarios. It simultaneously performs hollow detection and cell forming detection on the target image to identify the state of cells, can adapt to a low-resolution global industrial camera, further reduces the production cost of enterprises, and meets the low-cost production configuration requirements of enterprises. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required to be used in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0035] Figure 1 is a flowchart of a method for detecting cells in a semiconductor crystal growth furnace in a low-resolution scenario according to an embodiment of the present invention;

[0036] Figure 2 is a schematic diagram of cropping a target image of a method for detecting cells in a semiconductor crystal growth furnace in a low-resolution scenario according to an embodiment of the present invention;

[0037] Figure 3 is a schematic diagram of a binarized image of a method for detecting cells in a semiconductor crystal growth furnace in a low-resolution scenario according to an embodiment of the present invention;

[0038] Figure 4 is a schematic diagram of a hollow polygon scenario of a method for detecting cells in a semiconductor crystal growth furnace in a low-resolution scenario according to an embodiment of the present invention;

[0039] Figure 5 is a schematic diagram of a multi-line scenario of a method for detecting cells in a semiconductor crystal growth furnace in a low-resolution scenario according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0040] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art belong to the scope of protection of the present invention.

[0041] According to an embodiment of the present invention, a method for detecting crystal cells of a semiconductor crystal growth furnace in a low-resolution scenario is provided.

[0042] As Figure 1 shown, the method for detecting crystal cells of a semiconductor crystal growth furnace in a low-resolution scenario according to an embodiment of the present invention includes the following steps:

[0043] Step S1, collect the original image of crystal seeding through a global industrial camera in advance, and calibrate the double-aperture part of the original image of crystal seeding as the target feature for cropping to obtain a target image including the double-aperture part, as Figure 2 shown.

[0044] In this technical solution, as Figure 3 shown, convert the obtained target image including the double-aperture part into a binary image.

[0045] Step S2, perform hollow detection and crystal cell forming detection on the target image synchronously, wherein, including: the hollow detection is used to detect whether the current target image is a hollow polygon, and the crystal cell forming detection is used to detect whether the current target image is multiple lines;

[0046] Among them, performing hollow detection on the target image includes the following steps:

[0047] Read and process the binary image: Load and process the input binary image to ensure that the image only contains two gray values, black (0) and white (255).

[0048] Find the contours in the image: Obtain all external contours through the findContours function, including: Use the cv::findContours function to find all the contours in the image.

[0049] Specifically, in application, the findContours function is a function in OpenCV (a library for computer vision tasks) used to find contours in a binary image. The parameter RETR_EXTERNAL specifies to extract only external contours, and CHAIN_APPROX_SIMPLE means to simplify the contours, merging redundant points into a single line segment.

[0050] Traverse each found contour, wherein, including: For each contour, check whether the size of its point set is greater than the point set threshold, with a value of 4, and filter out the contours with a point set size less than 4 to filter out too small contours;

[0051] In the application of this technical solution, in OpenCV, for a contour, the size of its point set refers to the number of points contained in the contour, and each point is represented by (x, y) coordinates. For a contour, it is composed of a series of points. By checking whether the number of points contained in the contour is greater than 4, too small contours are filtered out to ensure that the contour is large enough.

[0052] Check the number of corner points of the filtered contour, determine whether the number of points of the contour is greater than the point number threshold, which is 4, and remove the contours with the number of points less than the point number threshold to ensure that there are enough vertices for subsequent analysis.

[0053] Perform convex hull detection on the removed contour, calculate the convex hull through the convexHull function, and obtain the smallest convex polygon that can enclose the contour point set.

[0054] Specifically, the convex hull is a polygon whose vertices are composed of points in the contour, enclosing the entire contour, and for any two points, they are inside the convex hull. And the cv::convexHull function in OpenCV is used to calculate the convex hull of a given point set.

[0055] As Figure 4 shown, obtain the hollow detection result, which includes: comparing the number of points of the obtained external contour with the number of points of the convex hull, including the following steps:

[0056] If the number of points of the external contour is equal to the number of points of the convex hull, it means that the current contour is not a hollow polygon, and the hollow detection result is output as no;

[0057] If the number of points of the external contour is not equal to the number of points of the convex hull, it means that the current contour is a hollow polygon, and draw the contour containing the hollow polygon as the hollow detection output, that is, it means that there are unit cells in the target image;

[0058] Among them, the unit cell forming detection of the target image includes the following steps:

[0059] Read and process the binary image: Load and process the input binary image to ensure that the image only contains two gray values, black (0) and white (255).

[0060] Find the contours in the image: Obtain all external contours through the findContours function, including: Use the cv::findContours function to find all the contours in the image.

[0061] Traverse each found contour, which includes: For each contour, check whether the size of its point set is greater than the point set threshold, which is 4, and filter out the contours with the point set size less than 4 to filter out too small contours;

[0062] Calculate the perimeter of each contour, including: obtaining the complete length of the contour through the arcLength function,

[0063] In this technical solution, cv::arclength is a function in the OpenCV library used to calculate the arc length of a curve. For a closed contour, the arcLength function returns the complete length of the contour, that is, the perimeter of the region enclosed by the contour.

[0064] Judge whether the contour meets the continuous line according to the set line threshold, including: setting the number of image segments to segment the contour, and taking the perimeter of the segmented contour greater than the set line threshold as a satisfied line, and obtaining the number of lines that meet the line;

[0065] Specifically, in application, according to production requirements, the value of the number of image segments is set to 9, and the calibrated image is 1000x1000 pixels in size, that is, each segmented image is 333x333. According to experiments in production, when there is a halo, the lines in the binary segmented image generally exceed 100 in length. Therefore, the line threshold is set to 100 here. When the perimeter is greater than the line threshold of 100, it is considered a line.

[0066] Compare the number of lines that meet the line with the preset line threshold to judge whether the current unsegmented contour conforms to multiple white continuous lines, including:

[0067] If the current number of lines that meet the line is greater than the line threshold, the current unsegmented contour conforms to multiple white continuous lines, indicating that there are unit cells in the target image, that is, there are double apertures in the target image, as Figure 5 shown.

[0068] Specifically, according to actual production, the line threshold can take the value of 14. When the number of lines that meet the line is 14, that is, at least 7 of the 9 segmented images have two lines, it means that multiple white continuous lines (greater than or equal to 2) are detected in the complete contour (unsegmented contour), indicating that there are double apertures, indicating that there are unit cells in the target image.

[0069] In addition, for the above-mentioned obtaining the number of lines that meet the line, in application, if the contour meets the conditions, increase the line count, and at the same time judge whether there are still unexamined contours.

[0070] Step S3, output the detection results of hollow detection and unit cell formation detection, and determine whether unit cells are generated in the current crystal seeding stage, including: if the hollow detection output is a hollow polygon and the unit cell formation detection output is multiple lines, it means that unit cells are generated in the current crystal seeding stage.

[0071] In this technical solution, the existence of the crystal cell is determined only when the hollow detection output is a hollow polygon and the crystal cell forming detection output is multiple lines; otherwise, the existence of the crystal cell cannot be determined.

[0072] In summary, by means of the above technical solution of the present invention, the following benefits can be achieved:

[0073] The present invention has high detection accuracy and fast detection speed. The original image of crystal seeding is collected by a global industrial camera, and the hollow detection and the crystal cell forming detection are respectively and synchronously performed on the target image, and the detection results of the hollow detection and the crystal cell forming detection are output to determine whether a crystal cell is generated in the current seeding stage, which helps to more accurately judge the state of the crystal cell, can automatically detect the crystal cell, reduce the manual burden, and achieve the accuracy, efficiency and reliability of the detection; at the same time, the present invention has a wide range of applicable scenarios. The hollow detection and the crystal cell forming detection are respectively and synchronously performed on the target image to identify the state of the crystal cell, and it can adapt to a low-resolution global industrial camera, further reducing the production cost of the enterprise and meeting the low-cost production configuration requirements of the enterprise.

[0074] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Those skilled in the art will easily think of other implementation schemes of the present disclosure after considering the disclosure in the specification and the embodiments. This application aims to cover any variations, uses or adaptations of the present disclosure, which follow the general principles of the present disclosure and include the common general knowledge or conventional technical means in the technical field not disclosed in the present disclosure. The specification and the embodiments are only regarded as exemplary, and the true scope and spirit of the present disclosure are pointed out by the claims.

[0075] It should be understood that the present disclosure is not limited to the exact structures already described and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the present disclosure is only limited by the appended claims.

Claims

1. A unit cell detection method for a semiconductor crystal growth furnace in a low-resolution scenario, characterized in that: The following steps are involved: The crystal seeding original image is collected in advance by a global industrial camera, and the double aperture part of the crystal seeding original image is calibrated as a target feature for cropping to obtain a target image including the double aperture part; The target image is subjected to a hollow detection and a cell forming detection respectively and synchronously, wherein: the hollow detection is used to detect whether the current target image is a hollow polygon, and the cell forming detection is used to detect whether the current target image is a plurality of lines; Output the detection results of hollow detection and cell forming detection to determine whether a cell is generated in the current seeding stage, including: if the hollow detection output is a hollow polygon and the cell forming detection output is a plurality of lines, it indicates that a cell is generated in the current seeding stage.

2. The unit cell detection method for a semiconductor crystal growth furnace in a low-resolution scenario according to claim 1, characterized in that: The step of acquiring the target image including the dual aperture portion comprises the following steps: The target image containing the double aperture part is converted into a binary image.

3. The unit cell detection method for a semiconductor crystal growth furnace in a low-resolution scenario according to claim 2, characterized in that: Hollow detection of the target image includes the following steps: Read and process binary images: Load and process the input binary image to ensure that the image contains only two grayscale values: black and white; Find contours in the image: Get all external contours through the findContours function, including: Use the cv::findContours function to find all contours in the image; Traversing each found contour, including: for each contour, checking whether its point set size is greater than the point set threshold, filtering contours whose point set size is less than the point set threshold, so as to filter out contours that are too small; Check the number of corner points of the filtered contour to determine whether the number of points of the contour is greater than the point threshold, and remove the contours with points less than the point threshold; Perform convex hull detection on the contour after elimination, calculate the convex hull through the convexHull function, and obtain the minimum convex polygon that can enclose the contour point set; Obtaining hollow detection results, including: comparing the number of points of the external contour obtained with the number of points of the convex hull, including the following steps: If the number of points of the outer contour is equal to the number of points of the convex hull, it means that the current contour is not a hollow polygon, and the output hollow detection result is no; If the number of points of the outer contour is not equal to the number of points of the convex hull, it means that the current contour is a hollow polygon, and the contour containing the hollow polygon is drawn as the hollow detection output, which means that the target image has unit cells.

4. The unit cell detection method for a semiconductor crystal growth furnace in a low-resolution scenario according to claim 2, characterized in that: Performing cell forming detection on the target image includes the following steps: Read and process binary images: Load and process the input binary image to ensure that the image contains only two grayscale values: black and white; Find contours in the image: Get all external contours through the findContours function, including: Use the cv::findContours function to find all contours in the image; Traversing each found contour, including: for each contour, checking whether its point set size is greater than the point set threshold, filtering contours whose point set size is less than the point set threshold, so as to filter out contours that are too small; Calculate the perimeter of each contour, including: obtaining the complete length of the contour through the arcLength function, Judging whether the contour satisfies the continuous line condition according to the set line threshold, which includes: setting the image segmentation number to segment the contour, and taking the contour perimeter after segmentation greater than the set line threshold as a satisfying line, and obtaining the number of satisfying lines; The number of lines that meet the line condition is compared with a preset line threshold to determine whether the current unsegmented contour meets multiple lines, including: if the number of lines that meet the line condition is greater than the line threshold, the current unsegmented contour meets multiple lines.

Images