Crystal ingot specific area identification and processing method and processing device

Abstract

The invention relates to the technical field of semiconductor crystal ingot processing, in particular to a method and a device for identifying and processing a specific area of a crystal ingot. The invention provides a method for identifying and processing a specific area of a crystal ingot, which comprises the following steps of: adopting a light source to irradiate, adopting a camera to collect a color pattern image, automatically identifying the specific area without cracks in the crystal ingot in combination with an area identification algorithm, and carrying out secondary processing on the area until the specific area disappears or reaches a processing standard. According to the method, the automation and precision of processing can be improved, manual intervention is reduced, the uniformity and integrity of crystal ingot cracks are ensured, and the stripping quality and production efficiency of wafers are improved. Meanwhile, the method effectively solves the technical problem of identifying and processing a specific area in the prior art, and has a relatively high industrial application value.

Description

Technical Field

[0001] The present invention relates to the technical field of semiconductor ingot processing, and in particular to a method and a processing device for identifying and processing a specific region of an ingot. Background Art

[0002] With the widespread application of materials such as silicon carbide in photovoltaics, semiconductors and other fields, the manufacturing and processing of ingots has become one of the key links. Laser processing is usually required during the production process of ingots to form a crack layer inside the ingot, so as to achieve efficient chip cutting in the subsequent stripping process. However, during the laser processing, due to the physical properties of the ingot material itself, uneven doping concentration and differences in processing parameters, cracks fail to form effectively in some areas or the crack connections are incomplete. These areas are called "specific areas". If they are not handled, they are prone to problems such as splitting and edge collapse during the subsequent chip stripping process, affecting product quality and production efficiency.

[0003] In the prior art, ingot processing usually relies on a combination of automatic control of the laser system and manual operation. Manual observation and judgment of color pattern images are subjective and uncertain, making it difficult to achieve high-precision and high-efficiency identification and processing. In addition, when identifying ingot cracks, the traditional light source and camera combination is limited by equipment accuracy and light conditions, and cannot accurately identify areas where cracks have not formed, resulting in specific areas that cannot be fully processed. To solve these problems, it is necessary to introduce a more accurate and intelligent method for identifying and processing specific areas of ingots. Summary of the invention

[0004] In order to solve the above-mentioned technical problems, the purpose of the present invention is to provide a method for identifying and processing specific areas of a crystal ingot. The method uses a light source for irradiation, a camera to collect color pattern images, and combines a specific area recognition algorithm to automatically identify specific areas in the crystal ingot where no cracks have formed, and perform secondary processing on these areas until the specific areas disappear or meet the processing standards. This method can improve the automation and accuracy of processing, reduce manual intervention, ensure the uniformity and integrity of the cracks in the crystal ingot, and improve the peeling quality and production efficiency of the wafer. At the same time, this method effectively solves the technical difficulties in identifying and processing specific areas in the prior art, and has a high industrial application value.

[0005] In order to achieve the above-mentioned purpose, the present invention adopts the following technical solutions: A method for identifying and processing a specific region of an ingot comprises the following steps: 1) After the previous laser processing of the ingot forms a peeling belt, a light source is used to illuminate the surface of the ingot and a camera is used to collect a color pattern image; 2) Preprocessing the collected color pattern images; 3) identifying a specific area in the ingot by a specific area recognition algorithm, wherein the specific area is an area where no color pattern is observed after the previous ingot laser processing; 4) Perform secondary processing on the specific area until the specific area disappears or meets the processing standards.

[0006] Preferably, the light source includes a surface light source or a strip light source, and the camera needs to be placed on the light path of the light from the light source after being reflected by the cracks of the ingot, and the spatial relationship needs to be fine-tuned according to the camera frame to ensure clear capture of the color pattern.

[0007] Preferably, the specific areas are areas that do not show color fringes after the previous laser processing because no cracks are formed or the cracks are not connected, and these areas tend to appear in small face areas or low resistivity areas of the ingot.

[0008] Preferably, when a bar light source is used, the image stitching comprises the following steps: a) First, determine the stitching starting point by collecting the coordinates of the center point of the first image; b) driving the crystal ingot or bar light source and the camera to move by a motor, and the moving distance is less than or equal to a predetermined distance d; c) Multiple images are stitched together through coordinate cropping to finally form a complete color pattern image.

[0009] As a further preferred embodiment, during the stitching process, image fusion processing is performed for each stitching point position to eliminate stitching marks and ensure the continuity and integrity of the color pattern information.

[0010] Preferably, the specific area identification comprises the following steps: a) Segment the preprocessed color pattern image using a deep learning image segmentation algorithm; b) The image segmentation model is trained using a fully convolutional neural network (FCN), using a labeled image dataset for model training, and the model is optimized by adjusting hyperparameters such as learning rate and batch size.

[0011] Preferably, the OpenCV library is used for image preprocessing during the model training phase, and the PyTorch framework is used for model training. After the training is completed, the unlabeled color pattern image is input into the model to obtain the segmented specific area result.

[0012] Preferably, in the specific area processing step, the following parameters are adjusted according to the process requirements: the light intensity of the light source; the depth of laser processing; the speed and spacing of laser scribing; the path of laser scribing; by adjusting the above parameters, the specific area is processed to ensure the generation and connection of cracks, avoiding the risk of splitting or edge collapse during subsequent peeling.

[0013] Preferably, after the processing of the specific area is completed, the color pattern image of the ingot is collected and identified again to determine whether the specific area has completely disappeared. If not, the identification-processing cycle steps are repeated until the processing is completed.

[0014] Furthermore, the present invention also discloses a device for identifying and processing a specific region of an ingot for implementing the method, the device comprising: The detection module composed of a light source and a camera collects the image of the color pattern of the crystal ingot; A specific area recognition module based on deep learning, used to segment and identify specific areas; Specific area processing module, adjust laser processing according to process parameters; The result display module returns the recognition and processing results to the client through network communication.

[0015] The present invention adopts the above-mentioned technical solution, mainly through the steps of light source irradiation, camera acquisition of color pattern images, image preprocessing, specific area recognition algorithm and secondary processing, to solve the technical problem that specific areas cannot be effectively identified and processed during ingot processing in the prior art. The technical effects of the present invention are reflected in the following aspects: 1. Improved the accuracy of specific area recognition: By using light source illumination and camera to collect color pattern images, combined with a specific spatial relationship configuration, the color pattern information on the surface of the ingot can be clearly captured, especially the characteristic image of crack reflection light. Through image preprocessing technology (such as image stitching, cropping, etc.) and specific area recognition algorithms, specific areas where no cracks are formed or cracks are not connected after laser processing can be accurately identified, thereby greatly improving the recognition accuracy.

[0016] 2. Realize intelligent secondary processing of specific areas: After identifying a specific area, the present invention uses secondary processing to adjust laser process parameters (such as light intensity, depth, scribing path, etc.) to perform targeted processing on the specific area to ensure that a complete crack is formed in the area until the area disappears or meets the preset processing standards. This intelligent secondary processing process effectively reduces the incidence of problems such as cracking and edge collapse, and ensures the processing quality of the ingot.

[0017] 3. Reduced human intervention and improved production efficiency: Traditional identification and repair of specific areas of crystal ingots often rely on manual observation and manual correction, while the present invention greatly reduces dependence on manual operation through automated image acquisition, recognition and secondary processing, reduces errors caused by subjective judgment, and improves the stability and consistency of the processing process, thereby significantly improving production efficiency.

[0018] 4. Optimized wafer stripping quality: The method of the present invention can ensure the uniformity and integrity of the cracks inside the ingot, and avoid wafer stripping defects caused by unformed or unconnected cracks, such as chipping and edge collapse. After secondary processing, specific areas can better meet the requirements of subsequent wafer stripping, thereby improving the overall quality of the wafer.

[0019] 5. Strong adaptability: The present invention is not only suitable for conventional ingot materials, but can also flexibly adapt to the processing requirements of different ingots by adjusting laser processing parameters and recognition algorithms according to the characteristics of different materials (such as conductive SiC, etc.), thus having broad industrial application prospects.

[0020] In summary, the present invention improves the processing quality of ingot cracks through automated and intelligent specific area identification and processing methods, reduces manual intervention, significantly improves production efficiency, and ensures the quality and stability during chip peeling, which has important technical advantages and application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is an example of the angle between the cleavage plane of conductive SiC and the crystal surface.

[0022] Figure 2 Schematic diagram of the crack generation process of conductive SiC.

[0023] Figure 3 Schematic diagram of light reflected during crack generation in conductive SiC.

[0024] Figure 4 Example Ⅰ of the spatial relationship among the light source, camera and crystal ingot.

[0025] Figure 5 Example II of the spatial relationship among the light source, camera, and crystal ingot.

[0026] Figure 6 The system flow chart of the method of the present invention is shown in FIG.

[0027] Figure 7 Schematic diagram of image stitching of strip light sources.

[0028] Figure 8 It is a structural block diagram of a crystal ingot characteristic area identification and processing device of the present invention. DETAILED DESCRIPTION

[0029] The following is a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0030] The present invention is based on a method for identifying and processing a specific area of ​​a crystal ingot after the stripping zone is formed by laser processing of the previous crystal ingot. Figure 1 As shown in the figure, after the previous laser processing of the ingot forms a peeling band, the cleavage plane of the conductive SiC and the crystal surface have an angle (for example, 4°), and the intersection line between the two is Crystal orientation. Figure 2 During laser processing, the laser generates modified points inside the SiC ingot and generates cracks along the cleavage plane. The cracks will not extend infinitely along the cleavage plane, but will connect with the previous and next cracks after extending to a certain length. Figure 3 After the cracks are connected, a surface light source, a line light source or other light source is used to illuminate the surface of the ingot. The light reflected by the crack surface will be reflected with the perpendicular line of the cleavage plane as the normal.

[0031] like Figure 4 , 5 An example of the spatial relationship between the light source, camera, and ingot is shown. After the laser forms a crack layer inside the ingot, in order for the camera to capture the light reflected from the crack, the following conditions need to be met: the camera needs to be placed along the propagation direction of the reflected light from the ingot crack, and fine-tuned according to the camera's frame. Because the actual crack lengths inside the ingot after laser processing are different, the intensity of the light reflected from the cracks is different, and the camera or the naked eye observes the result as color fringes of varying brightness.

[0032] For this reason, the specific areas defined in this application are defined as follows: After the previous laser processing of the ingot, due to factors such as inappropriate selection of various process parameters and uneven doping concentration of the ingot itself, cracks have formed in some areas and color lines are observed, but there may still be some areas where no color lines are observed because there are no cracks or the cracks are not connected. These areas where no color lines are observed because there are no cracks or the cracks are not connected are defined as specific areas. For conductive SiC, specific areas tend to appear in and near the small face area. Because in the production process of conductive SiC, the small face area and the non-small face area have different growth modes and different physical properties such as resistivity, and the absorption of lasers is also very different. After laser processing, the small face area is prone to have no cracks or the cracks are not connected, and no color lines are observed.

[0033] like Figure 6 As shown, a method for identifying and processing a specific region of an ingot specifically comprises the following steps: 1. The front laser processing forms a peeling belt First, the front-end laser processing is performed on the surface of the ingot. The laser generates modified points inside the ingot and generates cracks along the cleavage plane to form a peeling zone. In this process, the cracks will not expand indefinitely, but will connect with the previous and next cracks after an appropriate length to form a complete crack layer to facilitate the subsequent wafer peeling. However, in some areas, due to the influence of material properties or processing parameters, complete cracks may not be formed, and "specific areas" will be generated at this time.

[0034] 2. Collection of color pattern images After the front-end laser processing is completed, a suitable light source (such as a surface light source or a strip light source) is set to illuminate the surface of the ingot, and the camera is used to collect the color pattern image of the ingot. A certain spatial relationship needs to be formed between the camera, the light source and the ingot to ensure that the light reflected by the crack can be accurately captured by the camera after the light source illuminates the ingot. Especially when the light source is a strip light source, the ingot or the light source and the camera are driven by a motor to gradually collect the entire color pattern information of the ingot.

[0035] 3. Preprocessing of color pattern images The collected color pattern images need to be preprocessed, including image stitching and cropping. The strip light source can only illuminate part of the ingot, and the color pattern image is composed of multiple local images, which need to be stitched together through a stitching algorithm.

[0036] like Figure 7 As shown in FIG. 1 , when a strip light source is used, only a part of the surface of the ingot can be illuminated due to the range of the light source, such as Figure 7 The rectangular area shown in the first image in a. At this time, the camera can only detect the color pattern in the rectangular area, while the rest of the ingot is a darker color. Therefore, it is necessary to pre-process the ingot photos taken under strip light by image stitching, so as to show the color pattern on the entire surface of the ingot, which can be used for recognition and calculation of subsequent algorithms.

[0037] The image stitching process includes: Figure 7 As shown in a, the world coordinate starting point is confirmed by the coordinates of the center point of the first image and the position of the splicing point, and the ingot is driven by the motor (it can also drive the bar light source and camera to move). The condition to ensure that the color pattern information is not lost is that the moving distance each time is ≤ d, Figure 7 b is an example of movement = d. After completing the image acquisition of all positions, calculate the position of the stitching point of each image acquisition area, crop and stitch, and then obtain a complete color pattern image of the ingot. Figure 7 Middle c.

[0038] 4. Identification of specific areas The pre-processed color pattern image will be analyzed by a specific area recognition algorithm. The specific area refers to the area where cracks are not formed or the cracks are not connected after the previous laser processing, and these areas do not show color pattern features in the color pattern image. In order to improve the accuracy of recognition, the present invention uses a deep learning image segmentation algorithm, such as a fully convolutional neural network (FCN), to automatically recognize specific areas. The algorithm first trains the color pattern image with a large amount of historical data, and then inputs the newly collected image into the trained model to automatically identify the specific area and output the result.

[0039] The goal of specific area recognition is to segment specific areas and non-specific areas. The specific area recognition algorithm of the present invention basically includes four stages: image annotation, training environment construction, model training and model prediction; In the image annotation stage, specific areas are manually or automatically marked for previously saved complete ingot images after splicing; During the training environment construction phase, install necessary software libraries and dependent libraries according to the actual segmentation algorithm used to build the training environment; In the model training phase, the labeled images are input into the corresponding image segmentation model for training; In the model prediction stage, the ingot color pattern images that have not been used for training are input into the trained model to obtain the segmentation results.

[0040] In this embodiment, this patent uses a fully convolutional neural network as a segmentation model. First, multiple complete ingot splicing images are collected, and specific areas in the image are accurately annotated at the pixel level using manual and automated tools; according to the selected image segmentation algorithm, the corresponding training environment is built. The CNN image segmentation model is mainly selected using the PyTorch deep learning framework and the OpenCV image processing library, and the annotated image data set is divided into a training set and a validation set, and the model is input for training. During the training process, the model performance is optimized by adjusting hyperparameters such as learning rate, batch size, etc.

[0041] The specific steps for using the trained model to predict the new ingot color pattern image are as follows: 1) Input the ingot image to be predicted; 2) Preprocess the image, including scaling, normalization, etc.; 3) Input the preprocessed image into the trained model; 4) The model outputs the segmentation result, i.e. the location and shape of a specific area; 5) Post-process the segmentation results, including boundary smoothing, small area filtering, etc.; 6) Superimpose the processed results on the original image to intuitively display the specific identified areas.

[0042] 5. Secondary processing of specific areas After identifying a specific area, secondary processing is required to ensure the formation and connection of cracks to avoid chip breakage or edge cracking during the subsequent stripping process. Secondary processing is performed using lasers, and the following parameters are adjusted according to actual process requirements: light intensity of the light source; depth of laser processing; speed of laser scribing; distance between laser lines; path of laser scribing. After the processing is completed, the color pattern image is collected and identified again to determine whether the specific area has disappeared. If the color pattern is complete, the processing is completed; if the specific area still does not meet the requirements, the identification and secondary processing process is repeated until the specific area disappears or meets the processing standards.

[0043] like Figure 8 As shown, a crystal ingot characteristic area recognition and processing device of the present invention is composed of a specific area detection module, a specific area recognition module, a specific area processing module and a result display module.

[0044] Specific area detection module: This module consists of a light source and a camera that form a certain spatial relationship with the ingot. The camera needs to be placed in the propagation direction of the reflected light after the light from the light source is reflected by the cracks in the ingot. When the light source is a large enough surface light source, the surface light source is used to illuminate the entire ingot and take pictures; when the light source is a strip light source, the ingot is driven by a motor (it can also drive the strip light source and the camera to move), each time moving a distance ≤d until the entire ingot is photographed, cropped and spliced ​​to obtain a complete color pattern image of the ingot.

[0045] Specific area recognition module: Train and deploy the trained model on the server, input the captured color pattern image into the trained model through socket communication, obtain the segmentation result and return it to the client.

[0046] Result display module: After identifying a specific area, the recognition result is transmitted back to the client through socket communication and displayed.

[0047] Specific area processing module: According to the actual situation, the process parameters including light intensity, processing depth, laser line spacing, laser marking speed, laser marking path, etc. are adjusted, and additional processing of specific areas is performed according to the process parameter adjustment results.

[0048] Experimental example In order to prove the effectiveness of the method for identifying and processing specific areas of a crystal ingot of the present invention, we designed a series of experiments to test the effect of the present invention in actual crystal ingot processing and compared it with the traditional method.

[0049] 1. Experimental object: Conductive SiC standard ingot material was used for the experiment. The ingot material itself has an uneven doping concentration, making it difficult to form a complete crack through the initial laser processing.

[0050] 2. Test equipment: using a light source, camera system, deep learning segmentation algorithm, and laser processing equipment with the functions of the present invention.

[0051] 3. Experimental content: Through the color pattern image acquisition, preprocessing, specific area recognition and secondary processing process of the present invention, the specific area is processed and compared with the traditional processing method that relies on manual recognition and repair.

[0052] 4. Evaluation criteria: Recognition accuracy: The percentage of specific areas that can be accurately identified.

[0053] Processing Time: The time required to complete the identification and secondary processing of specific areas of each ingot.

[0054] Splitting rate: The probability of each ingot to split or break during the subsequent peeling process.

[0055] Product qualification rate: the proportion of processed chips that meet standard quality requirements.

[0056] Table 1 Experimental data

[0057] 6. Experimental results analysis 1) Recognition accuracy: The method of the present invention uses a high-precision camera and deep learning segmentation algorithm to achieve a recognition accuracy of up to 99.5% for specific areas, which is better than the 93.4% of traditional manual observation. Manual observation is limited by lighting conditions and the experience of the observer, and it is easy to miss or misjudge some specific areas. The present invention can perform high-precision image processing on the surface of the ingot, greatly improving the recognition accuracy.

[0058] 2) Processing time: The average processing time for each ingot using the method of the present invention is 5 minutes, which is 15 minutes shorter than that of the traditional method. The reason is that the present invention reduces the manual judgment and intervention process through automated image recognition and processing flow, and speeds up the processing speed of specific areas.

[0059] 3) Split rate: The present invention significantly reduces the split rate, which is only 1.0% in the experiment, while the traditional method has a split rate of 7.5%. This is because the present invention can accurately identify and reprocess the area where no cracks have formed, ensuring the uniformity and integrity of the cracks and avoiding the generation of splits and edge collapse during the wafer peeling process.

[0060] 4) Product qualification rate: The qualified rate of wafer processing products of the present invention reaches 99.0%, which is significantly better than the 92.5% of the traditional method. The higher qualified rate is due to the higher recognition accuracy and the reduction of the cracking rate, which ensures the quality of the final processed wafer.

[0061] The above is a description of the embodiments of the present invention. Through the above description of the disclosed embodiments, professionals and technicians in the field can implement or use the present invention. Various modifications to these embodiments will be apparent to professionals and technicians in the field. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown in this article, but will conform to the widest range consistent with the principles and novelties disclosed herein.

Claims

1. A method for identifying and processing a specific region of an ingot, characterized in that: The following steps are involved: 1) After the previous laser processing of the ingot forms a peeling belt, a light source is used to illuminate the surface of the ingot and a camera is used to collect a color pattern image; 2) Preprocessing the collected color pattern images; 3) identifying a specific area in the ingot by a specific area recognition algorithm, wherein the specific area is an area where no color pattern is observed after the previous ingot laser processing; 4) Perform secondary processing on the specific area until the specific area disappears or meets the processing standards.

2. The method according to claim 1, characterized in that: The light source includes a surface light source or a strip light source. The camera needs to be placed on the light path of the light from the light source after being reflected by the cracks of the ingot, and the spatial relationship needs to be fine-tuned according to the camera frame to ensure clear capture of the color pattern.

3. The method according to claim 1, characterized in that: The specific areas are areas that do not show color fringes after the previous laser processing because no cracks are formed or the cracks are not connected, and these areas tend to appear in small face areas or low resistivity areas of the ingot.

4. The method according to claim 1, characterized in that: When using a bar light source, image stitching consists of the following steps: a) First, determine the stitching starting point by collecting the coordinates of the center point of the first image; b) driving the crystal ingot or bar light source and the camera to move by a motor, and the moving distance is less than or equal to a predetermined distance d; c) Multiple images are stitched together through coordinate cropping to finally form a complete color pattern image.

5. The method according to claim 4, characterized in that: During the stitching process, image fusion processing is performed on each stitching point to eliminate stitching marks and ensure the continuity and integrity of the color pattern information.

6. The method according to claim 1, characterized in that: The specific area identification comprises the following steps: a) Segment the preprocessed color pattern image using a deep learning image segmentation algorithm; b) The image segmentation model is trained using a fully convolutional neural network (FCN), using a labeled image dataset for model training, and the model is optimized by adjusting hyperparameters such as learning rate and batch size.

7. The method according to claim 6, characterized in that: In the model training phase, the OpenCV library is used for image preprocessing, and the PyTorch framework is used for model training. After the training is completed, the unlabeled color pattern image is input into the model to obtain the segmentation results of the specific area.

8. The method according to claim 1, characterized in that: During the processing step of a specific area, the following parameters are adjusted according to the process requirements: the intensity of the light source; the depth of laser processing; the speed and spacing of laser scribing; and the path of laser scribing. By adjusting the above parameters, the specific area is processed to ensure the generation and connection of cracks, avoiding the risk of cracks or edge collapse during subsequent peeling.

9. The method according to claim 8, characterized in that: After the processing of the specific area is completed, the color pattern image of the crystal ingot is collected and recognized again to determine whether the specific area has completely disappeared. If not, the recognition-processing cycle steps are repeated until the processing is completed.

10. A device for identifying and processing a specific region of an ingot for implementing the method according to any one of claims 1 to 9, characterized in that: The device comprises: The detection module composed of a light source and a camera collects the image of the color pattern of the crystal ingot; A specific area recognition module based on deep learning, used to segment and identify specific areas; Specific area processing module, adjust laser processing according to process parameters; The result display module returns the recognition and processing results to the client through network communication.

Images