Silicon wafer subfissure detection method and terminal

Abstract

According to the silicon wafer subfissure detection method and the terminal, the finished silicon wafer is scanned through the laser with the preset wavelength in a penetrating mode, the scanning image on the surface of the finished silicon wafer is captured through the camera, and the target image is obtained; inputting the target image into a pre-trained subfissure detection model to obtain a subfissure detection result; the hidden crack detection model adopts an improved yolk series detection network which takes a multi-layer Swin Transform module as a backbone network; according to the method, the camera is used for capturing the picture of the laser scanning silicon wafer with the specific wavelength, the hidden crack position on the surface of the silicon wafer can be clearly displayed through the brightness difference based on the penetrability of the laser and the absorption characteristic of the silicon wafer to the laser, and the hidden crack detection is carried out by adopting an improved yolk series detection network with a multi-layer Swin Transformer module as a backbone network. The self-attention mechanism of the multi-layer Swin Transformer is utilized to well screen the image features, so that the interferences of grid lines, scratches, smudginess and the like can be eliminated, and real subfissures are highlighted.

Description

Technical Field

[0001] The present invention relates to the field of silicon wafer hidden crack detection, and in particular to a silicon wafer hidden crack detection method and terminal. Background Art

[0002] Existing silicon wafer crack detection systems, such as "Photovoltaic module defect detection based on improved YOLOv5", "use the Ghost module to replace the ordinary convolution module in the YOLOv5 backbone extraction network to reduce the number of parameters in the network model: in order to ensure good detection performance, add a Squeeze-and-Excitation (SE) attention module at the end of the backbone network to improve the target detection ability of the algorithm; introduce a bidirectional feature pyramid network (BIFPN) structure in the feature fusion network to further enhance the feature fusion ability of the network." By improving the open source network, the network's expression ability is increased and the detection ability of the model is improved.

[0003] There are also works such as "Multi-loss fusion of small sample photovoltaic module hidden crack detection algorithm" that introduces the Transformer's multi-head attention mechanism to reduce the impact of differences in the distribution of each batch of products on crack detection and promote the model to focus on crack information from diversified products. Secondly, a strategy of combining multiple loss functions to constrain model training is adopted to optimize feature extraction. On the basis of direct classification loss, triplet loss is used to shorten the feature distance between crack samples. In addition, the implicit classification loss is designed to adapt to the type difference characteristics between two cells with or without cracks, and to fully learn the diversity of historical component data.

[0004] The existing technology (1) does not really improve and optimize the algorithm based on the characteristics of hidden cracks, but improves the model by stacking computing power to increase model expression and training skills; (2) the silicon wafer is not adapted to the image of the finished product. The silicon wafer image at the finished product stage contains more gate lines, which makes the prediction of hidden cracks more difficult. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a method and terminal for detecting hidden cracks in silicon wafers, so as to achieve more accurate hidden crack detection in finished silicon wafers.

[0006] In order to solve the above technical problems, the technical solution adopted by the present invention is: A method for detecting hidden cracks in silicon wafers, comprising the steps of: S1, scanning the finished silicon wafer through laser with a preset wavelength, and capturing the scanned image of the surface of the finished silicon wafer with a camera to obtain a target image; S2, inputting the target image into a pre-trained hidden crack detection model to obtain a hidden crack detection result; The hidden crack detection model adopts an improved yolo series detection network with a multi-layer Swin Transformer module as the backbone network.

[0007] In order to solve the above technical problems, the technical solution adopted by the present invention is: A terminal for detecting hidden cracks in finished silicon wafers includes a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the following steps are implemented: S1, scanning the finished silicon wafer through laser with a preset wavelength, and capturing the scanned image of the surface of the finished silicon wafer with a camera to obtain a target image; S2, inputting the target image into a pre-trained hidden crack detection model to obtain a hidden crack detection result; The hidden crack detection model adopts an improved yolo series detection network with a multi-layer Swin Transformer module as the backbone network.

[0008] The beneficial effects of the present invention are as follows: the present invention relates to a method and terminal for detecting hidden cracks in silicon wafers, which utilizes a camera to capture images of a silicon wafer scanned by a laser of a specific wavelength, and can clearly display the location of hidden cracks on the surface of the silicon wafer through brightness differences based on the penetration of the laser and the absorption characteristics of the silicon wafer to the laser. Hidden crack detection is performed using an improved yolo series detection network with a multi-layer Swin Transformer module as a backbone network, and the self-attention mechanism of the multi-layer Swin Transformer is utilized to perform good screening of image features, thereby eliminating interference such as grid lines, scratches, and dirt, and highlighting real hidden cracks. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 A brief flow chart of a method for detecting hidden cracks in a silicon wafer according to an embodiment of the present invention; Figure 2 This is a structural diagram of a silicon wafer hidden crack detection terminal according to an embodiment of the present invention; Figure 3 This is a diagram showing an example of a network structure of a hidden crack detection model of a method for detecting hidden cracks in a silicon wafer according to an embodiment of the present invention; Figure 4 A schematic diagram of a process for detecting hidden cracks in a silicon wafer according to an embodiment of the present invention; Figure 5 An example diagram of a hidden crack of a method for detecting hidden cracks in a silicon wafer according to an embodiment of the present invention; Description of labels: 1. A silicon wafer hidden crack detection terminal; 2. A processor; 3. A memory. DETAILED DESCRIPTION

[0010] In order to explain the technical content, achieved objectives and effects of the present invention in detail, the following is an explanation in combination with the implementation modes and the accompanying drawings.

[0011] Please refer to Figure 1 , Figure 3 and Figure 4 , a method for detecting hidden cracks in silicon wafers, comprising the steps of: S1, scanning the finished silicon wafer through laser with a preset wavelength, and capturing the scanned image of the surface of the finished silicon wafer with a camera to obtain a target image; S2, inputting the target image into a pre-trained hidden crack detection model to obtain a hidden crack detection result; The hidden crack detection model adopts an improved yolo series detection network with a multi-layer Swin Transformer module as the backbone network.

[0012] From the above description, it can be seen that the beneficial effects of the present invention are: the present invention relates to a method for detecting hidden cracks in silicon wafers, which uses a camera to capture images of laser scanning silicon wafers with a specific wavelength, and based on the laser's penetration and the silicon wafer's absorption characteristics of the laser, can clearly display the location of hidden cracks on the surface of the silicon wafer through brightness differences, and uses an improved yolo series detection network with a multi-layer SwinTransformer module as the backbone network for hidden crack detection, and uses the self-attention mechanism of the multi-layer SwinTransformer to perform good screening of image features, which can eliminate interference such as grid lines, scratches, and dirt, and highlight the real hidden cracks.

[0013] Furthermore, the silicon wafer is a finished double-half-piece silicon wafer; The steps between step S1 and step S2 include: S11, locating and extracting the single-side silicon wafer area in the target image to obtain a single-side silicon wafer area image; Step S2 is specifically as follows: The single-side silicon wafer area image is input into a pre-trained hidden crack detection model to obtain a hidden crack detection result.

[0014] From the above description, it can be seen that the dicing (dividing into two) operation required in the process of manufacturing photovoltaic modules is carried out in advance, and the risk classification is moved from the back end to the front end, which is more conducive to the grading and reuse of damaged silicon wafers.

[0015] Furthermore, the single-side silicon wafer area in the target image is located and extracted as follows: For the target image, a preliminary selected area in which the silicon wafer pixel value in the target image is greater than a preset threshold is selected by a sub-pixel threshold operator, and single-side silicon wafer areas corresponding to two half-wafers in the target image are respectively determined according to whether the area of ​​the largest area in the preliminary selected area is greater than the preset area threshold; The target image is cropped of the single-side silicon wafer region, and the cropped portion is scaled and cropped again to obtain a plurality of single-side silicon wafer region images that meet the input specifications of the hidden crack detection model.

[0016] From the above description, it can be seen that for the collected target image, the area where the silicon wafer is located is located by using a threshold value, so as to separate the single-sided silicon wafer area, and obtain the single-sided silicon wafer area image that is adapted to the input specifications of the hidden crack detection model by scaling and cropping.

[0017] Furthermore, the composition of the hidden crack detection model includes an input layer, a backbone network, an analysis network and an output layer; The backbone network includes a Patch Embedding layer, a Swin Transformer module, a PatchMerging module, a Swin Transformer module, a PatchMerging module and a Swin Transformer module which are arranged in sequence.

[0018] From the above description, it can be seen that a yolo series detection network improved by a backbone network based on Swin Transformer is constructed, and its backbone network structure is as described above.

[0019] Furthermore, the laser wavelength ranges from 1000 nm to 1020 nm.

[0020] From the above description, it can be seen that within the wavelength range of 1000nm to 1020nm, the laser has a good ability to penetrate the silicon wafer, can penetrate the silicon wafer more evenly, and produce obvious signal changes at the hidden cracks, thereby facilitating the detection of hidden cracks.

[0021] Please refer to Figure 2 A terminal for detecting hidden cracks in finished silicon wafers includes a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the following steps are implemented: S1, scanning the finished silicon wafer through laser with a preset wavelength, and capturing the scanned image of the surface of the finished silicon wafer with a camera to obtain a target image; S2, inputting the target image into a pre-trained hidden crack detection model to obtain a hidden crack detection result; The hidden crack detection model adopts an improved yolo series detection network with a multi-layer Swin Transformer module as the backbone network.

[0022] From the above description, it can be seen that the beneficial effects of the present invention are: the present invention relates to a silicon wafer hidden crack detection terminal, which uses a camera to capture pictures of a silicon wafer scanned by a laser of a specific wavelength, and can clearly display the location of hidden cracks on the surface of the silicon wafer through brightness differences based on the penetration of the laser and the absorption characteristics of the silicon wafer to the laser. It also uses an improved yolo series detection network with a multi-layer SwinTransformer module as the backbone network for hidden crack detection, and uses the self-attention mechanism of the multi-layer SwinTransformer to perform good screening of image features, which can eliminate interference such as grid lines, scratches, and dirt, and highlight the real hidden cracks.

[0023] Furthermore, the silicon wafer is a finished double-half-piece silicon wafer; The steps between step S1 and step S2 include: S11, locating and extracting the single-side silicon wafer area in the target image to obtain a single-side silicon wafer area image; Step S2 is specifically as follows: The single-side silicon wafer area image is input into a pre-trained hidden crack detection model to obtain a hidden crack detection result.

[0024] From the above description, it can be seen that the dicing (dividing into two) operation required in the process of manufacturing photovoltaic modules is carried out in advance, and the risk classification is moved from the back end to the front end, which is more conducive to the grading and reuse of damaged silicon wafers.

[0025] Furthermore, the single-side silicon wafer area in the target image is located and extracted as follows: For the target image, a preliminary selected area in which the silicon wafer pixel value in the target image is greater than a preset threshold is selected by a sub-pixel threshold operator, and single-side silicon wafer areas corresponding to two half-wafers in the target image are respectively determined according to whether the area of ​​the largest area in the preliminary selected area is greater than the preset area threshold; The target image is cropped of the single-side silicon wafer region, and the cropped portion is scaled and cropped again to obtain a plurality of single-side silicon wafer region images that meet the input specifications of the hidden crack detection model.

[0026] From the above description, it can be seen that for the collected target image, the area where the silicon wafer is located is located by using a threshold value, so as to separate the single-sided silicon wafer area, and obtain the single-sided silicon wafer area image that is adapted to the input specifications of the hidden crack detection model by scaling and cropping.

[0027] Furthermore, the composition of the hidden crack detection model includes an input layer, a backbone network, an analysis network and an output layer; The backbone network includes a Patch Embedding layer, a Swin Transformer module, a PatchMerging module, a Swin Transformer module, a PatchMerging module and a Swin Transformer module which are arranged in sequence.

[0028] From the above description, it can be seen that a yolo series detection network improved by a backbone network based on Swin Transformer is constructed, and its backbone network structure is as described above.

[0029] Furthermore, the laser wavelength ranges from 1000 nm to 1020 nm.

[0030] From the above description, it can be seen that within the wavelength range of 1000nm to 1020nm, the laser has a good ability to penetrate the silicon wafer, can penetrate the silicon wafer more evenly, and produce obvious signal changes at the hidden cracks, thereby facilitating the detection of hidden cracks.

[0031] A silicon wafer hidden crack detection method and terminal of the present invention are suitable for the hidden crack detection of silicon wafers, and are particularly suitable for the hidden crack detection of silicon wafers in the finished product stage.

[0032] Please refer to Figure 1 and Figure 3 , Embodiment 1 of the present invention is: A method for detecting hidden cracks in silicon wafers, comprising the steps of: S1, scanning the finished silicon wafer through laser with a preset wavelength, and capturing the scanned image of the surface of the finished silicon wafer with a camera to obtain a target image; The laser wavelength ranges from 1000nm to 1020nm.

[0033] In this embodiment, a 1006 nm wavelength laser light source is used.

[0034] In this embodiment, the laser with a specific wavelength penetrates the silicon wafer, and a 4K line scan camera mounted on the machine platform is used to capture a 2D image of the surface of the silicon wafer to obtain a target image.

[0035] S2. Input the target image into a pre-trained hidden crack detection model to obtain a hidden crack detection result.

[0036] In this embodiment, the model is run and the running result is given, for example, the hidden crack area is identified and a rejection signal is given.

[0037] The hidden crack detection model adopts an improved yolo series detection network with a multi-layer Swin Transformer module as the backbone network; The hidden crack detection model comprises an input layer, a backbone network, an analysis network and an output layer; The backbone network includes a Patch Embedding layer, a Swin Transformer module, a PatchMerging module, a Swin Transformer module, a PatchMerging module and a Swin Transformer module which are arranged in sequence.

[0038] In this embodiment, taking into account the characteristics of the hidden crack defects themselves showing diverse shapes and different positions, a modified yolo series detection network based on the backbone network of Swin Transformer is constructed. In order to solve the problem of excessive memory and computing power consumption when processing images in the traditional Transformer model, Swin Transformer adopts a layered window mechanism to process images while retaining the global image receptive field. In this embodiment, a multi-layer Swin Transformer is constructed as the backbone network. The first layer is the Patch Embedding layer, which divides the image into image blocks and connects them in series into a 1×16 image matrix, followed by a Swin Transformer module. The subsequent modules are repeated superpositions of PatchMerging modules and Swin Transformer modules. The network structure of the entire hidden crack detection model is as follows: Figure 3 As shown in the figure, b is the batch size of the network training line, i.e., batch size; BottleNeckCSP is the attention network layer; Conv is the convolution layer; UpSample is the upsampling layer; and Concat is the connection layer.

[0039] In this embodiment, for the training of the hidden crack detection model, 10,000 training image data (including 7,000 hidden crack defects and 3,000 non-defective images) are sorted out and input into the network for training. The training batch size is 6, and the Adam optimizer is used to train the detection model required in this article.

[0040] Please refer to Figure 4 and Figure 5 , Embodiment 2 of the present invention is: A method for detecting hidden cracks in a silicon wafer, which differs from the first embodiment in that, in the present embodiment, the silicon wafer is a finished double-half-wafer silicon wafer; The steps between step S1 and step S2 include: S11, locating and extracting the single-side silicon wafer area in the target image to obtain a single-side silicon wafer area image; The specific steps for locating and extracting the single-sided silicon wafer area in the target image are as follows: For the target image, a preliminary selected area in which the silicon wafer pixel value in the target image is greater than a preset threshold is selected by a sub-pixel threshold operator, and single-side silicon wafer areas corresponding to two half-wafers in the target image are respectively determined according to whether the area of ​​the largest area in the preliminary selected area is greater than the preset area threshold; The target image is cropped of the single-side silicon wafer region, and the cropped portion is scaled and cropped again to obtain a plurality of single-side silicon wafer region images that meet the input specifications of the hidden crack detection model.

[0041] In this embodiment, a 4K line scan camera is used to acquire a target image, and the acquired image has a horizontal resolution of 4096 and a vertical resolution of 4096. The acquired image has a completely black background and a gray silicon wafer surface.

[0042] For the target image M, first locate the area where the silicon wafer is located through the threshold. That is, through the sub-pixel threshold operator, select the area where the silicon wafer pixel value is greater than 30, calculate whether the area with the largest area in the above areas is greater than Max_Area_Thresh, and locate the left silicon wafer area L_SoloarRegion and the right silicon wafer area R_SolarRegion respectively. If L_SoloarRegion or R_SolarRegion is empty, the corresponding area is an empty film, and the operation of this area is adjusted to the last step, returning to the left or right empty film, and prompting an error; this article will then assume that one side of the silicon wafer area is not empty, the left silicon wafer area L_SoloarRegion, or the right silicon wafer area R_SolarRegion is not empty, and simplify the name to SolarRegion to execute the next step.

[0043] Please refer to Figure 5 For the target image, the hidden crack itself is one or more connected irregular line segments inside the silicon wafer, the color is not a single color, and the shapes are different. In this embodiment, in order to better detect the hidden crack, the image where the SolarRegion of each half-chip is located is first cropped out, and the image size is rescaled to a new image of 1280×640 (1280 high, 640 wide). Next, two 640×640 images are cropped, and a total of four 640×640 images are obtained when the two half-chips are not empty. Input: SolarImg ; ResizeImg = Resize(SolarImg,1280, 640); SplitImg = Split(ResizeImg, 640, 640); For the input target image M, firstly, the image is cropped according to SolarRegion to obtain the image SolarImg where the silicon wafer is located. The SolarImg image is scaled to 1280×640 (1280 high, 640 wide). Next, the image is cropped to divide it into two 640×640 silicon wafers as the input of the next detection model.

[0044] Step S2 is specifically as follows: The single-side silicon wafer area image is input into a pre-trained hidden crack detection model to obtain a hidden crack detection result.

[0045] Please refer to Figure 2 , Embodiment 3 of the present invention is: A finished product segment silicon wafer hidden crack detection terminal 1 comprises a processor 2, a memory 3 and a computer program stored in the memory 3 and executable on the processor 2. When the processor 2 executes the computer program, the steps of a finished product segment silicon wafer hidden crack detection method of the above embodiment 1 or 2 are implemented.

[0046] In summary, the present invention provides a method and terminal for detecting hidden cracks in silicon wafers. The camera is used to capture images of a silicon wafer scanned by a laser of a specific wavelength. Based on the penetration of the laser and the absorption characteristics of the silicon wafer to the laser, the location of the hidden cracks on the surface of the silicon wafer can be clearly displayed through brightness differences. The improved yolo series detection network with a multi-layer Swin Transformer module as the backbone network is used for hidden crack detection. The self-attention mechanism of the multi-layer Swin Transformer is used to well screen the image features, which can eliminate interference such as grid lines, scratches, and dirt, and highlight the real hidden cracks.

[0047] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent transformations made using the contents of the present invention's specification and drawings, or directly or indirectly applied in related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for detecting hidden cracks in silicon wafers, characterized in that: Includes steps: S1, scanning the finished silicon wafer through laser with a preset wavelength, and capturing the scanned image of the surface of the finished silicon wafer with a camera to obtain a target image; S2, inputting the target image into a pre-trained hidden crack detection model to obtain a hidden crack detection result; The hidden crack detection model adopts an improved yolo series detection network with a multi-layer Swin Transformer module as the backbone network.

2. A method for detecting hidden cracks in finished silicon wafers according to claim 1, characterized in that: The silicon wafer is a finished double-half-piece silicon wafer; The steps between step S1 and step S2 include: S11, locating and extracting the single-side silicon wafer area in the target image to obtain a single-side silicon wafer area image; Step S2 is specifically as follows: The single-side silicon wafer area image is input into a pre-trained hidden crack detection model to obtain a hidden crack detection result.

3. The method for detecting hidden cracks in finished silicon wafers according to claim 2, characterized in that: The specific steps for locating and extracting the single-sided silicon wafer area in the target image are as follows: For the target image, a preliminary selected area in which the silicon wafer pixel value in the target image is greater than a preset threshold is selected by a sub-pixel threshold operator, and single-side silicon wafer areas corresponding to two half-wafers in the target image are respectively determined according to whether the area of ​​the largest area in the preliminary selected area is greater than the preset area threshold; The target image is cropped of the single-side silicon wafer region, and the cropped portion is scaled and cropped again to obtain a plurality of single-side silicon wafer region images that meet the input specifications of the hidden crack detection model.

4. The method for detecting hidden cracks in finished silicon wafers according to claim 1, characterized in that: The hidden crack detection model comprises an input layer, a backbone network, an analysis network and an output layer; The backbone network includes a Patch Embedding layer, a Swin Transformer module, a PatchMerging module, a Swin Transformer module, a PatchMerging module and a Swin Transformer module which are arranged in sequence.

5. The method for detecting hidden cracks in finished silicon wafers according to claim 1, characterized in that: The laser wavelength ranges from 1000nm to 1020nm.

6. A finished product segment silicon wafer hidden crack detection terminal, comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the following steps are implemented: S1, scanning the finished silicon wafer through laser with a preset wavelength, and capturing the scanned image of the surface of the finished silicon wafer with a camera to obtain a target image; S2, inputting the target image into a pre-trained hidden crack detection model to obtain a hidden crack detection result; The hidden crack detection model adopts an improved yolo series detection network with a multi-layer Swin Transformer module as the backbone network.

7. A finished product segment silicon wafer hidden crack detection terminal according to claim 6, characterized in that: The silicon wafer is a finished double-half-piece silicon wafer; The steps between step S1 and step S2 include: S11, locating and extracting the single-side silicon wafer area in the target image to obtain a single-side silicon wafer area image; Step S2 is specifically as follows: The single-side silicon wafer area image is input into a pre-trained hidden crack detection model to obtain a hidden crack detection result.

8. The finished product segment silicon wafer hidden crack detection terminal according to claim 7, characterized in that: The specific steps for locating and extracting the single-sided silicon wafer area in the target image are as follows: For the target image, a preliminary selected area in which the silicon wafer pixel value in the target image is greater than a preset threshold is selected by a sub-pixel threshold operator, and single-side silicon wafer areas corresponding to two half-wafers in the target image are respectively determined according to whether the area of ​​the largest area in the preliminary selected area is greater than the preset area threshold; The target image is cropped of the single-side silicon wafer region, and the cropped portion is scaled and cropped again to obtain a plurality of single-side silicon wafer region images that meet the input specifications of the hidden crack detection model.

9. The finished product segment silicon wafer hidden crack detection terminal according to claim 6, characterized in that: The hidden crack detection model comprises an input layer, a backbone network, an analysis network and an output layer; The backbone network includes a Patch Embedding layer, a Swin Transformer module, a PatchMerging module, a Swin Transformer module, a PatchMerging module and a Swin Transformer module which are arranged in sequence.

10. The finished product segment silicon wafer hidden crack detection terminal according to claim 6, characterized in that: The laser wavelength ranges from 1000nm to 1020nm.

Images