A silicon carbide crystal cutting system and cutting method based on 3D defect detection

Abstract

The present application relates to a kind of based on 3D defect detection silicon carbide crystal cutting system and cutting method, wherein a kind of based on 3D defect detection silicon carbide crystal cutting system includes, X-ray tomography CT scanning equipment, obtains the structure and defect information in crystal;Silicon carbide crystal fixing equipment is used to fix silicon carbide crystal;3D data processing module is connected with X-ray tomography CT scanning equipment, including defect data module, defect data contrast analysis module, image processing and generation module;Cutting data processing module, after receiving 3D defect data and image, after cutting data module processing calculation, form the best cutting path method;Cutting system is connected with cutting data processing module, according to the best cutting path, set cutting process parameters, finally form silicon carbide wafer;The present application obtains the best cutting route by 3D scanning defect and defect data analysis, forms low-defect high-yield wafer, adapts to different defect distribution, improves the precision and efficiency of cutting, reduces processing loss.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of silicon carbide cutting, and particularly relates to a silicon carbide crystal cutting system and method based on 3D defect detection. BACKGROUND

[0002] Since the mid-20th century, silicon material has been widely used in the field of integrated circuit chips in low-voltage, low-frequency, low-power transistors and detectors as a semiconductor material, including computers, mobile phone CPUs, GPUs, system-on-chip (SoC), etc. As the third generation of semiconductor materials, silicon carbide (SiC) and gallium nitride (GaN) are widely used in rail transit, new energy vehicles, wind power, etc. In terms of process, the production process of silicon carbide mainly includes the following steps: crystal ingot growth, slicing, polishing, epitaxy, photolithography, etching, ion implantation, and deposition. Because silicon carbide is one of the hardest substances, with a Mohs hardness of 9.2 and 9.3, in the process of manufacturing silicon carbide chips, iterative crystal ingot cutting technology and slicing technology are the main ways to improve productivity. Traditional multi-wire cutting technology is a process for cutting brittle and hard materials. The slurry wire cutting technology has been applied to most silicon carbide manufacturers, and the diamond wire cutting technology has become the mainstream iterative solution. The above methods not only have low cutting efficiency and high material loss, but also have high consumable cost. The latest laser cutting technology has high processing efficiency and low material loss, and is a research hotspot in the industry.

[0003] Patent CN118081117A discloses a new type of silicon carbide cutting system and method, which uses a femtosecond laser device for cutting. Because the femtosecond laser has the advantages of extremely short pulse width, extremely high peak power, low heat input, and large adjustable depth-diameter ratio, it can be used for cutting silicon carbide wafers and ingots. For wafer cutting, a grating and a galvanometer can be used to output multiple light beams for simultaneous cutting. For ingot cutting, a large depth-diameter ratio can be adjusted for multiple iterative cutting, supplemented by plasma etching equipment for shaping, cleaning, and real-time monitoring by vision and infrared for secondary cutting, which greatly improves cutting efficiency and reduces material loss. Patent CN117601293A discloses a method for cutting silicon carbide, which uses a heater in combination with a high-pressure water column to reduce the cost of hardware purchase and effectively cut silicon carbide. CN117283152A discloses a method for high-efficiency laser cutting of silicon carbide ingots, which adopts a double-wavelength laser coaxial synchronous single processing mode to realize the simultaneous space action of the modified laser and the heating laser on the SiC ingot, reduce the processing time and wafer peeling difficulty, and improve the peeling quality.

[0004] Patent CN116465912A discloses a method for rapid nondestructive testing of silicon carbide crystal quality, which performs full-slice scanning on the treated silicon carbide crystal through a specific X-ray source and scanning mode, and according to the local difference in X-ray absorption or scattering of different defects of the crystal material, the defect density of the silicon carbide crystal and the intuitive image of the distribution can be quickly obtained on the computed tomography imaging.

[0005] In summary: at present, the cutting of silicon carbide crystal is mainly through changing the cutting tool and improving the cutting method to improve the cutting efficiency and reduce the cutting loss, and the detection accuracy of silicon carbide crystal still needs to be further improved. SUMMARY

[0006] The purpose of this part is to summarize some aspects of the embodiments of the application and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this part and the abstract and title of the specification to avoid obscuring the purpose of this part, the abstract and the title, and such simplifications or omissions cannot be used to limit the scope of the application.

[0007] To solve the above technical problems, the application provides the following technical scheme: a silicon carbide crystal cutting system based on 3D defect detection, comprising,

[0008] X-ray tomography CT scanning device, for omnidirectional and high-resolution 3D scanning of silicon carbide crystal, obtaining the structure and defect information inside the crystal;

[0009] Silicon carbide crystal fixing device, for fixing silicon carbide crystal, with rotating and vertically moving functions, to ensure accurate control of the position of the crystal during scanning and cutting;

[0010] 3D data processing module, connected with the X-ray tomography CT scanning device, including defect data module, defect data comparison and analysis module, image processing and generation module; after the 3D data processing module obtains the X-ray data signal, the 3D defect image of the silicon carbide crystal is obtained;

[0011] Cutting data processing module, connected with the 3D data module, after receiving the 3D defect data and image, the best cutting path method is formed through the cutting data module processing and calculation;

[0012] Cutting system, connected with the cutting data processing module, according to the best cutting path, setting the cutting process parameters, finally forming the silicon carbide wafer.

[0013] As a preferred technical scheme of the silicon carbide crystal cutting system based on 3D defect detection, the X-ray tomography CT scanning device is a micro-focus high-energy X-ray tube, the accelerating voltage range of the X-ray light source is 200-350kV, the light source focal point size is ≤10 microns, the exposure time is ≥1000ms, and the spatial resolution is ≤80 microns.

[0014] As a preferred technical scheme of the silicon carbide crystal cutting system based on 3D defect detection, the defect data module includes a micro-tube defect data module, a carbon package defect data module and a polycrystal / multiform defect data module; the defect data comparison and analysis module classifies different defect data; and the image generation module gives different defect types different color expressions according to the comparison and analysis results, and the data within the normal range is represented by black.

[0015] As a preferred technical scheme of the silicon carbide crystal cutting system based on 3D defect detection, based on 3D defect image data, the cutting data processing module includes a defect distribution analysis module and a cutting path analysis module, and the cutting path analysis module of the cutting data processing module is provided with four silicon carbide defect level divisions, the A-class level defect index is <1000 / cm 2 , the B-class level defect index is 1000-3000 / cm 2 , the C-class level defect index is 3000-8000 / cm 2 , and the D-class defect index is >8000 / cm 2 .

[0016] The application also provides a cutting method based on the aforementioned silicon carbide crystal cutting system based on 3D defect detection, which comprises,

[0017] Step 1: performing surface cleaning treatment on the silicon carbide crystal;

[0018] Step 2: placing the silicon carbide crystal in a fixing device, setting the rotation or movement speed, emitting X-rays by the X-ray tomography CT scanning device, rotating or moving up and down the silicon carbide crystal, and obtaining X-ray data signals;

[0019] Step 3: inputting the X-ray data signals into a 3D data processing module, processing and analyzing the X data according to the differences in X-ray absorption or scattering of different defect parts, and finally forming a 3D defect image through an image generation module;

[0020] Step 4: inputting the 3D defect image data into a cutting data module, performing data analysis according to four silicon carbide defect levels, and obtaining the best cutting path;

[0021] Step 5: cutting the system according to the best cutting path, selecting appropriate cutting parameters, and finally obtaining silicon carbide wafers of different specifications.

[0022] As a preferred technical scheme of the silicon carbide crystal cutting method based on 3D defect detection, in step 4, the cutting data module performs data analysis according to the four silicon carbide defect levels, and sets the cutting path to pass through the defect dense area to the maximum extent, and the first cutting finally obtains different thickness silicon carbide wafers with minimized surface defects.

[0023] For the first cutting to obtain different thickness silicon carbide wafers with minimized surface defects, the cutting path analysis can be further performed, and the second cutting obtains silicon carbide wafers with minimized defects meeting the thickness requirement.

[0024] As a preferred technical scheme of the silicon carbide crystal cutting method based on 3D defect detection, in step 4, the cutting data module performs data analysis according to the four silicon carbide defect levels, and sets the cutting path to pass through the defect dense area to the maximum extent, and the first cutting finally obtains different thickness silicon carbide wafers with minimized surface defects.

[0025] As a preferred technical scheme of the silicon carbide crystal cutting method based on 3D defect detection, the X-ray data signal is input into the 3D data processing module, which specifically includes:

[0026] The FDK cone beam CT reconstruction algorithm is used to process the X-ray data, and a 3D model of the silicon carbide crystal is reconstructed, wherein the reconstructed voxel intensity I(x, y, z) represents the attenuation coefficient of each point in the crystal;

[0027] An artificial intelligence algorithm such as a deep learning neural network trained with labeled defect data is applied to analyze the 3D model and identify defect types such as microtubules, carbon inclusions, and polycrystalline regions, and the neural network outputs a defect classification probability P defect (x, y, z) for each voxel.

[0028] As a preferred technical scheme of the silicon carbide crystal cutting method based on 3D defect detection, the 3D defect image data is input into the cutting data processing module, which includes:

[0029] A machine learning algorithm trained with historical defects and cutting data is used to analyze the spatial distribution of defects and predict the best cutting plane;

[0030] The best cutting path is calculated by minimizing the total defect density in the resulting wafer, using the formula:

[0031]

[0032] wherein, is the total defect density, is the defect density of the i-th layer, is the thickness of the i-th layer;

[0033] determining the cutting parameters for each cutting path segment, including cutting depth, speed, and power settings, based on the defect data and material properties.

[0034] As a preferred technical scheme of the silicon carbide crystal cutting method based on 3D defect detection, cutting is performed according to the optimal cutting path, including:

[0035] inputting the cutting parameters from the cutting data processing module to the cutting system;

[0036] starting the cutting process, and the laser performs cutting according to the predetermined path, with an initial setting of a cutting line width of 0.15 mm, a cutting speed of 3 mm / s, and a laser peak power of 1.5 GW;

[0037] continuously monitoring the cutting process, and adjusting the cutting parameters such as laser power, focal point, and speed in real time according to the feedback of the 3D data processing module and the in-situ sensor to cope with the changes in defect distribution encountered during the cutting process.

[0038] The present application has the beneficial effects that by adjusting the parameters such as laser power and cutting speed in real time, the present application adapts to different defect distributions, improves the precision and efficiency of cutting, and reduces processing loss. BRIEF DESCRIPTION OF DRAWINGS

[0039] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor. Among them:

[0040] Figure 1 is a schematic diagram of the cutting system workflow of the present application;

[0041] Figure 2 is a schematic diagram of the cutting route entry point of the present application. DETAILED DESCRIPTION

[0042] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings of the specification.

[0043] In the following description, many specific details are set forth in order to provide a thorough understanding of the present application, but the present application can also be implemented in other ways different from those described herein, and those skilled in the art can make similar generalizations without departing from the connotation of the present application, therefore the present application is not limited to the specific embodiments disclosed below.

[0044] Second, the "one embodiment" or "an embodiment" described herein refers to a particular feature, structure, or characteristic described in one or more implementations. The appearances of "in one embodiment" or "in an embodiment" in various places in the specification are not necessarily all referring to the same embodiment nor are separate or alternative embodiments mutually exclusive of other embodiments.

[0045] Third, the present application is described in detail in conjunction with the schematic diagram. In the detailed description of the embodiments of the present application, the cross-sectional view of the device structure is partially enlarged without the general proportion for the convenience of description, and the schematic diagram is only an example which should not limit the scope of protection of the present application. In addition, the three-dimensional spatial dimensions of length, width and depth should be included in actual production.

[0046] Embodiment 1

[0047] With reference to Figure 1 and 2 , the present embodiment provides a silicon carbide crystal cutting system based on 3D defect detection, comprising:

[0048] An X-ray tomography CT scanning device is used to perform full-range and high-resolution 3D scanning on the silicon carbide crystal to obtain the structure and defect information inside the crystal;

[0049] A silicon carbide crystal fixing device is used to fix the silicon carbide crystal and has rotation and vertical up-and-down movement functions to ensure that the position of the crystal is accurately controllable during scanning and cutting;

[0050] A 3D data processing module is connected with the X-ray tomography CT scanning device and includes a defect data module for collecting and storing data of microtubule defects, carbon inclusion defects, and polycrystalline / multitype defects;

[0051] That is, the original data obtained by X-ray CT scanning is collected and stored.

[0052] A defect data comparison and analysis module uses an artificial intelligence algorithm to classify and analyze the defect data;

[0053] That is, the original data is preprocessed, segmented, and classified to identify different types of defects;

[0054] An image processing and generation module generates a 3D defect image of different colors representing different defect types.

[0055] A cutting data processing module is connected with the 3D data processing module and includes:

[0056] A defect distribution analysis module uses a machine learning algorithm to analyze the spatial distribution of defects; a trained convolutional neural network (CNN) model is used to perform deep learning on the defect data to predict the best cutting path.

[0057] The cutting path optimization module uses a deep learning algorithm to calculate the optimal cutting path to maximize the yield of low-defect wafers.

[0058] The cutting system, connected to the cutting data processing module, can perform cutting according to the optimal cutting path and adjust cutting parameters in real time based on feedback from the 3D data processing module.

[0059] The relationship between the modules is as follows:

[0060] The X-ray tomography CT scanning device obtains the internal structure data of the silicon carbide crystal, and the data is transmitted to the 3D data processing module.

[0061] The 3D data processing module processes the data to generate a 3D defect image and calculates the optimal cutting path through the artificial intelligence optimization module, and sends the results to the cutting data processing module.

[0062] The cutting data processing module generates specific cutting parameters based on the optimal cutting path and transmits them to the cutting system.

[0063] The cutting system performs cutting according to the parameters and returns real-time feedback information (such as cutting progress, abnormal conditions) to the cutting data processing module to achieve closed-loop control.

[0064] Further, the X-ray tomography CT scanning device is a micro-focus high-energy X-ray tube, the acceleration voltage range of the X-ray light source is 200-350kV, the light source focal point size is ≤10 microns, the exposure time is ≥1000ms, and the spatial resolution is ≤80 microns.

[0065] Further, the 3D data processing module uses a deep learning algorithm to classify defects with higher precision.

[0066] Further, the cutting data processing module uses a neural network trained with historical defect data and cutting results to optimize the cutting path.

[0067] Further, the cutting system is a laser cutting system that can adjust cutting parameters in real time based on feedback from the 3D data processing module.

[0068] The defect data module includes microtubule defect data module, carbon inclusion defect data module, and polycrystalline / multiform defect data module; the defect data comparison and analysis module classifies different defect data; the image generation module assigns different colors to different defect types based on the comparison and analysis results, and normal data is represented by black.

[0069] Based on the 3D defect image data, the cutting data processing module contains a defect distribution analysis module and a cutting path analysis module. The cutting path analysis module of the cutting data processing module is provided with four silicon carbide defect level divisions. The A-class level defect index is <1000 / cm 2 , the B-class level defect index is 1000-3000 / cm 2 , the C-class level defect index is 3000-8000 / cm 2 , and the D-class defect index is >8000 / cm 2 .

[0070] Through the implementation of the following technical steps, the present application realizes efficient and high-quality cutting of silicon carbide crystals. The specific analysis is as follows:

[0071] Use of X-ray tomography CT scanning equipment:

[0072] By introducing a high-energy micro-focus X-ray CT scanning device into the system, full-range and high-resolution 3D scanning of the silicon carbide crystal is realized, and the internal structure and defect information of the crystal are obtained. By using an X-ray tube with an acceleration voltage range of 200-350kV and a focus size ≤10 microns, the depth and accuracy of the scanning are ensured. This step allows accurate understanding of the internal defect distribution of the crystal before cutting, including microtubules, carbon inclusions, and polycrystalline regions, providing an accurate data basis for subsequent defect identification and cutting path optimization. Non-destructive detection of internal defects of the crystal is realized, avoiding the cutting quality problems caused by the lack of internal defect information in traditional methods, significantly improving the yield of wafers.

[0073] Design of silicon carbide crystal fixing equipment:

[0074] By using a silicon carbide crystal fixing device with rotation and vertical up-and-down movement functions, accurate positioning and controllable movement of the crystal during scanning and cutting are realized. By accurately controlling the position and angle of the crystal, the X-ray CT scanning equipment can obtain information inside the crystal from multiple angles and positions. This step ensures the integrity and accuracy of the scanning data, and ensures the consistency of the cutting path with the optimization scheme during the cutting process, avoiding cutting deviations caused by positioning errors and improving cutting precision and efficiency.

[0075] Application of 3D data processing module:

[0076] Use of defect data module

[0077] By collecting and storing data on microtubule defects, carbon inclusion defects, and polycrystalline / multitype defects, systematic management and analysis of different types of defects is realized. Comprehensive data support is provided for subsequent defect classification and cutting path optimization.

[0078] Artificial intelligence algorithm of defect data comparative analysis module:

[0079] Classify and analyze defect data using artificial intelligence algorithms such as deep learning neural networks. Through training a model with labeled defect data, accurately identify different defect types in the crystal. The neural network outputs the defect classification probability P of each voxel (x, y, z). defect (x,y,z). Achieve high-precision identification and classification of defects, improve the accuracy and efficiency of defect identification.

[0080] Visualization of image processing and generation module:

[0081] Generate 3D defect images of different colors representing different defect types, and visually display the spatial distribution of defects. This step helps technicians quickly understand the defect situation inside the crystal, assists decision-making, and improves the operability of the system.

[0082] Function of cutting data processing module:

[0083] Machine learning algorithm of defect distribution analysis module:

[0084] Use convolutional neural network (CNN) trained with historical defects and cutting data to analyze the spatial distribution of defects and predict the best cutting plane. Through deep learning algorithms, consider factors such as defect type, density, and location to calculate the optimal cutting path. Achieve intelligent and optimized cutting solutions, maximize the yield of low-defect wafers.

[0085] Deep learning algorithm of cutting path optimization module:

[0086] Use deep learning algorithms to calculate the best cutting path, aiming to minimize the total defect density in the resulting wafer. Use the formula:

[0087]

[0088] Where, is the total defect density, is the defect density of the i-th layer, is the thickness of the i-th layer. Through mathematical modeling, achieve the optimization of the cutting path and improve the quality of the wafer.

[0089] Real-time feedback and adjustment of cutting system:

[0090] The cutting system can perform cutting according to the optimal cutting path and adjust the cutting parameters in real time based on the feedback from the 3D data processing module. During the cutting process, the cutting state is continuously monitored, and parameters such as laser power, focal point position, and cutting speed are adjusted in real time according to the changes in defect distribution. Closed-loop control is achieved, improving the stability and reliability of the cutting process and ensuring the quality of the final wafer.

[0091] The adoption of high-performance X-ray CT equipment:

[0092] By using a micro-focus high-energy X-ray tube with an acceleration voltage range of 200-350 kV, a focal point size of ≤10 microns, an exposure time of ≥1000 ms, and a spatial resolution of ≤80 microns, the depth and resolution of the scan are improved. It can detect tiny defects and improve the accuracy of defect identification, providing high-quality data support for subsequent cutting optimization.

[0093] Application of deep learning algorithm:

[0094] The 3D data processing module uses a deep learning algorithm to classify defects with higher accuracy. Through deep learning, the recognition ability of complex defects is improved, reducing false positives and improving the intelligence level of the system.

[0095] Cutting path optimization based on historical data:

[0096] The cutting data processing module uses a neural network trained on historical defect data and cutting results to optimize the cutting path. Using historical data, the model can more accurately predict the optimal cutting scheme, reducing trial and error costs and improving cutting efficiency and wafer quality.

[0097] Real-time adjustment of laser cutting system: The cutting system is a laser cutting system that can adjust cutting parameters in real time based on feedback from the 3D data processing module. By adjusting parameters such as laser power and cutting speed in real time, it adapts to different defect distributions, improving cutting accuracy and efficiency and reducing processing loss.

[0098] Example 2

[0099] The application also discloses a cutting method of a silicon carbide crystal cutting system based on 3D defect detection, which comprises the following steps: 1. surface cleaning treatment of the silicon carbide crystal;

[0100] Step 2: Place the silicon carbide crystal in the fixing device, set the rotation or movement speed, and perform X-ray tomography CT scanning to emit X-rays. The silicon carbide crystal rotates or moves up and down, and the X-ray data signal is obtained.

[0101] Step 3: The X-ray data signal is input into a 3D data processing module, which processes and analyzes the X data based on the differences in X-ray absorption or scattering at different defect sites, and finally forms a 3D defect image through an image generation module;

[0102] Step 4: The 3D defect image data is input into a cutting data module, which analyzes the data based on the four silicon carbide defect levels to obtain the best cutting path;

[0103] Step 5: The cutting system cuts according to the best cutting path and selects appropriate cutting parameters to finally obtain silicon carbide wafers of different specifications.

[0104] Further, the cutting data module analyzes the data based on the four silicon carbide defect levels, and at the same time, sets the cutting path to pass through the defect dense area to the greatest extent, and the first cutting finally obtains silicon carbide wafers of different thicknesses with minimized surface defects;

[0105] For the silicon carbide wafers of different thicknesses with minimized surface defects obtained by the first cutting, the cutting path can be further analyzed, and the second cutting obtains silicon carbide wafers with minimized defects that meet the thickness requirements;

[0106] The cutting data module analyzes the data based on the four silicon carbide defect levels, and at the same time, sets the final cutting silicon carbide wafer thickness to be consistent, with a thickness of 0.35mm or 0.5mm, to finally obtain the best cutting path with consistent wafer thickness.

[0107] Further, in step 4, the cutting data module analyzes the data based on the four silicon carbide defect levels, and at the same time, sets the final cutting silicon carbide wafer thickness to be inconsistent, with a thickness of 0.35mm and 0.5mm, to finally obtain the best cutting path with inconsistent wafer thickness.

[0108] The cutting method can be specifically as follows:

[0109] Step (1) Clean the surface of the silicon carbide crystal with ethanol or isopropanol to remove contaminants, and then dry with nitrogen to prevent oxidation;

[0110] Step (2) Fix the silicon carbide crystal on a crystal fixing device with rotation and up-down movement functions, set the rotation speed to 1 revolution per minute, start the X-ray tomography CT scanning equipment, emit X-rays while the crystal is rotating, and obtain X-ray data signals from multiple angles;

[0111] Step (3) Input the X-ray data signal into a 3D data processing module, which specifically includes:

[0112] The X-ray data is processed using the Feldkamp-Davis-Kress (FDK) cone beam CT reconstruction algorithm to reconstruct a 3D model of the silicon carbide crystal, where the reconstructed voxel intensity I(x, y, z) represents the attenuation coefficient at each point in the crystal;

[0113] A 3D model is analyzed using an artificial intelligence algorithm such as a deep learning neural network trained on labeled defect data to identify defect types such as micropipes, carbon inclusions, and poly regions. The neural network outputs a defect classification probability P for each voxel defect (x, y, z);

[0114] The defect density D for different regions is calculated using the formula D = N / A, where N is the number of defects detected within an area A;

[0115] Defects are classified into four predefined levels based on the defect density per unit area:

[0116] Level A: D < 1000 / cm 2 ;

[0117] Level B: D is between 1000-3000 / cm 2 ;

[0118] Level C: D is between 3000-8000 / cm 2 ;

[0119] Level D: D > 8000 / cm 2 ;

[0120] A 3D defect image is generated, with different defect types and levels represented by different colors for easy visualization;

[0121] Step (4) inputs the 3D defect image data into a cutting data processing module, which specifically includes:

[0122] A machine learning algorithm (such as a convolutional neural network CNN) trained on historical defects and cutting data is used to analyze the spatial distribution of defects and predict the optimal cutting plane;

[0123] The optimal cutting path is calculated by minimizing the total defect density in the resulting wafer using the formula:

[0124]

[0125] where, is the total defect density, is the defect density of the i-th layer, is the thickness of the i-th layer;

[0126] determining cutting parameters for each cutting path segment, including cutting depth, speed, and power settings, based on defect data and material properties;

[0127] Step (5) sets up a cutting system (a laser cutting system capable of real-time parameter adjustment) to perform cutting according to the optimal cutting path, specifically including:

[0128] Inputting cutting parameters from the cutting data processing module to the cutting system;

[0129] Starting the cutting process, and the laser cuts according to the predetermined path, with an initial setting of a cutting line width of 0.15 mm, a cutting speed of 3 mm / s, and a laser peak power of 1.5 GW;

[0130] Continuously monitoring the cutting process and adjusting laser power, focal point, and speed in real time according to feedback from the 3D data processing module and in-situ sensors to cope with changes in defect distribution during the cutting process;

[0131] Step (6) collects the cut silicon carbide wafer, as the optimized cutting path and real-time parameter adjustment minimize wafer defects, meeting or exceeding market thickness (e.g., 0.35 mm) and quality requirements, reaching A-level defect levels.

[0132] In step (3), the 3D data processing module classifies defects into four levels according to defect density and uses different colors to represent them.

[0133] In step (4), the cutting data processing module sets the cutting path to pass through defect-dense areas to minimize defects in the resulting wafer.

[0134] In step (5), the cutting system adjusts cutting speed, power, and other parameters in real time according to feedback from the 3D data processing module to ensure optimal cutting quality.

[0135] After the initial cutting, the cut wafer is subjected to secondary 3D defect detection and analysis, and if necessary, secondary cutting is performed to further minimize defects and meet market thickness requirements.

[0136] Further, in step (3), X-ray data signal processing and defect identification are required.

[0137] In this step, the data signal obtained by the X-ray CT scanning device enters the 3D data processing module, and the specific process is as follows:

[0138] Data preprocessing:

[0139] Reconstructing the sequence of two-dimensional cross-sectional images obtained by X-ray CT scanning to form a three-dimensional volume data.

[0140] De-noising: Apply Gaussian filter or median filter to remove noise generated during scanning.

[0141] Defect type identification:

[0142] Voxel classification: Divide the three-dimensional data into voxels, each representing a tiny cubic region in the crystal.

[0143] Feature extraction: Calculate the gray value, texture features, and other parameters for each voxel.

[0144] Neural network classification:

[0145] Model input: Input the features of each voxel into a pre-trained convolutional neural network (CNN) model.

[0146] Model output: The CNN model outputs the probability value of each voxel belonging to a specific defect type, denoted as where i, j, k are the positions of the voxel in three-dimensional space, and C is the defect type (such as microtubule, carbon package, polycrystalline region, etc.).

[0147] Defect type determination:

[0148] For each voxel, select the defect type with the highest probability as its classification result.

[0149] In step (4), the defect density calculation and the prediction of the best cutting path are required, including,

[0150] Defect density calculation:

[0151] Divide the crystal into several macro regions (such as cutting planes).

[0152] For each region, count the number of defect voxels N and calculate the defect density D:

[0153]

[0154] where A is the area of the cutting plane of the region.

[0155] Spatial distribution analysis:

[0156] Use the defect density data to draw a defect spatial distribution map to show the defect intensity of different regions.

[0157] Prediction of the best cutting plane:

[0158] Data set preparation: Combine historical defect distribution data and cutting effect data to form a training set to train the convolutional neural network model.

[0159] Model input: Defect spatial distribution data of the current crystal.

[0160] Model output: Predicted position and angle of the optimal cutting plane that minimizes defects and maximizes quality of the cut wafer.

[0161] Cutting parameter determination: Based on the prediction of the optimal cutting plane, calculate the specific coordinates of the cutting path.

[0162] Determine cutting parameters such as cutting speed v, laser power P, pulse frequency f, etc., considering the material properties of the crystal and the performance of the cutting equipment.

[0163] In step (5), perform cutting according to the optimal cutting path, including,

[0164] Cutting path loading:

[0165] Input the optimal cutting path and cutting parameters into the control unit of the cutting system.

[0166] Cutting parameter setting:

[0167] Adjust the output power, focusing position, and moving speed of the laser according to the cutting parameters.

[0168] Real-time feedback control: During the cutting process, sensors such as optical sensors and temperature sensors of the cutting system monitor the cutting state in real time.

[0169] If abnormalities such as cutting deviation or overheating are detected, adjust the cutting parameters through the feedback mechanism to ensure cutting quality.

[0170] Cutting execution: The cutting system cuts the silicon carbide crystal according to the set path and parameters, obtaining high-quality wafers.

[0171] Defect identification examples:

[0172] Microtubule defects:

[0173] Features: In CT images, they appear as long, thin cavities or low-density areas.

[0174] Identification: The CNN model identifies these features and outputs a higher probability of microtubule defects .

[0175] Carbon inclusion defects:

[0176] Features: In CT images, they appear as small particles or clumps of high density.

[0177] Identification: The CNN model identifies these high-density areas and outputs a higher probability of carbon inclusion defects .

[0178] Polycrystalline region:

[0179] Features: irregular texture and grain boundaries in CT images.

[0180] Identification: CNN model outputs higher polycrystalline region defect probability through texture analysis

[0181] Defect density calculation example:

[0182] For a certain region, count the number of microtubule defect voxels N 微管 =500, area A=1 cm 2 Then the microtubule defect density is:

[0183]

[0184] Optimal cutting plane prediction example:

[0185] Input: defect spatial distribution data of the current crystal.

[0186] Model prediction: CNN model calculates the total defect density of different cutting planes, and selects the plane with the smallest total defect density as the optimal cutting plane.

[0187] Output: position z=z0 and angle θ=θ0 of the optimal cutting plane.

[0188] Cutting parameter determination example: according to the characteristics of the crystal material and the performance of the cutting equipment, set the cutting speed v=2 mm / s, laser power P=1.5 GW, and pulse frequency f=500 kHz.

[0189] Cutting execution example: the cutting system starts cutting according to the set path and parameters.

[0190] Real-time monitoring of the temperature T and stress σ of the cutting area, if T exceeds the set value T max Orσ exceeds the set valueσ max Then reduce the laser power or slow down the cutting speed.

[0191] The invention achieves the following effects:

[0192] Defect identification: X-ray CT scanning combined with CNN model can accurately identify different types of defects, which has been widely used in the fields of materials science and medical imaging.

[0193] Defect density calculation: by counting the classification results of voxels, the defect density is calculated, which is a common method.

[0194] Optimal Cutting Plane Prediction: Utilizing machine learning models, the optimal cutting plane is predicted based on the spatial distribution of defects, effectively improving wafer quality.

[0195] Cutting Parameter Determination and Execution: Combining material properties and equipment performance, cutting parameters are set, and through real-time feedback control, cutting quality is ensured, which is a key technology in modern precision machining.

[0196] Benefits of Method Steps:

[0197] Step (1): Clean the surface of the silicon carbide crystal with ethanol or isopropanol to remove contaminants, then dry with nitrogen to prevent oxidation, ensuring the cleanliness of the crystal surface and preventing oxidation from affecting subsequent scanning and cutting, improving data accuracy and cutting quality.

[0198] Step (2): Fix the silicon carbide crystal on a crystal fixing device with rotation and up-down movement functions, set the rotation speed to 1 revolution per minute, realize multi-angle and multi-direction scanning. By obtaining X-ray data at different angles, the data source for three-dimensional reconstruction is enriched, and the accuracy of the reconstructed model is improved.

[0199] Step (3): Use the Feldkamp-Davis-Kress (FDK) cone beam CT reconstruction algorithm to process X-ray data and reconstruct a 3D model of the silicon carbide crystal, accurately reflecting the internal structure and defect information of the crystal. Apply a deep learning neural network trained with labeled defect data to accurately identify and classify defect types such as microtubes, carbon inclusions, and polycrystalline regions, and calculate the defect density D of each region. By classifying defect density, an intuitive defect distribution map is provided, providing a basis for cutting path optimization.

[0200] Step (4): Use machine learning algorithms to analyze the spatial distribution of defects and predict the optimal cutting plane. By minimizing the total defect density in the resulting wafer, the optimal cutting path is calculated. Based on defect data and material properties, determine the cutting parameters for each cutting path segment to ensure the relevance and efficiency of the cutting process.

[0201] Step (5): Set up the cutting system and execute the cutting according to the optimal cutting path. Through real-time monitoring and adjustment, the stability and efficiency of the cutting process are ensured. Through optimized cutting path and parameter adjustment, the defects in the wafer are minimized, improving the quality of the wafer and meeting or exceeding market thickness and quality requirements.

[0202] Step (6): Collect the cut silicon carbide wafer. Due to the optimized cutting path and real-time parameter adjustment, the wafer defects are minimized to reach A-level defect levels. If necessary, secondary 3D defect detection and secondary cutting can be performed to further improve wafer quality and ensure high-quality products.

[0203] In summary, the present application combines 3D defect detection, artificial intelligence algorithms, machine learning, and real-time adjusted laser cutting technology to achieve efficient and high-precision cutting of silicon carbide crystals. The steps are closely linked, and through accurate defect detection, intelligent cutting path optimization, and real-time cutting parameter adjustment, the yield of low-defect wafers is maximized. The system and method significantly improve wafer quality, reduce production costs, and improve production efficiency, with important application value and broad market prospects.

[0204] To verify the effectiveness and advantages of the proposed 3D defect detection-based silicon carbide crystal cutting system and method, the following specific experiments were designed and conducted.

[0205] Experimental preparation and detailed implementation process:

[0206] A high-purity silicon carbide (SiC) single crystal with a diameter of 200 mm and a thickness of 20 mm was selected as the experimental sample. First, the surface of the SiC crystal was cleaned with anhydrous ethanol, and thoroughly wiped to remove organic matter and particulate contaminants on the surface. Then, high-purity nitrogen gas was used to purge the crystal surface to ensure thorough drying and prevent oxidation and water vapor from affecting subsequent detection.

[0207] The treated SiC crystal was fixed on a crystal fixing device with high-precision rotation and up-down movement functions. The rotation speed of the device was set to 1 revolution per minute (1 rpm) to ensure that comprehensive internal information of the crystal could be obtained from different angles during scanning. The micro-focus high-energy X-ray tomography scanning equipment was started, and the device parameters were set as follows: accelerating voltage 250 kV, focal size less than 10 microns, exposure time 1000 ms, and spatial resolution up to 80 microns. While the crystal was rotating, X-rays were irradiated from different angles to obtain complete X-ray data signals.

[0208] The collected X-ray data signals were input into the 3D data processing module. First, the Feldkamp-Davis-Kress (FDK) cone beam CT reconstruction algorithm was used to reconstruct the two-dimensional cross-sectional image sequence into a three-dimensional volume data of the SiC crystal. To improve data quality, a Gaussian filtering method was used to denoise the reconstructed three-dimensional data, eliminating random noise and artifacts generated during scanning.

[0209] Next, the three-dimensional body data is divided into voxels, each representing a tiny cubic region in the crystal. For each voxel, the gray value, texture features, and other parameters are extracted, which reflect the subtle differences in the internal structure of the crystal. The extracted features are input into a pre-trained convolutional neural network (CNN) model, which is trained based on a large amount of labeled defect data and has high accuracy in defect classification.

[0210] The CNN model outputs a probability value P for each voxel belonging to a specific defect type defect (x, y, z), where x, y, z are the positions of the voxel in three-dimensional space, and the defect types include microtubules, carbon packages, and polycrystalline regions. For each voxel, the defect type with the highest probability is selected as its classification result.

[0211] The crystal is divided into several macroscopic regions according to a thickness of 0.5 mm, and the defect density D of different regions is calculated using the formula D = N / A for each region, where N is the number of defects detected within an area A. According to the defect density per unit area, the defects are divided into four levels:

[0212] Level A: D < 1000 / cm 2 ;

[0213] Level B: D is between 1000-3000 / cm 2 ;

[0214] Level C: D is between 3000-8000 / cm 2 ;

[0215] Level D: D > 8000 / cm 2 .

[0216] Using the above results, a 3D defect image is generated, with different defect types and levels represented by different colors, making it easy to visually observe the spatial distribution of internal defects in the crystal.

[0217] The 3D defect image data is input into a cutting data processing module. Using a CNN algorithm trained with historical defects and cutting data, the spatial distribution of defects is analyzed, and the optimal cutting plane is predicted. The optimal cutting path is calculated by minimizing the total defect density in the resulting wafer, with the objective function being:

[0218]

[0219] where D 总 is the total defect density, Di is the defect density of the i-th layer, and hi is the thickness of the i-th layer. Based on the defect data and material properties, the cutting parameters for each cutting path segment are determined, including cutting depth, speed, and laser power settings.

[0220] The cutting system is set up for a laser cutting system that can adjust parameters in real time. The cutting parameters are input from the cutting data processing module to the cutting system, initially set to a cutting line width of 0.15 mm, a cutting speed of 3 mm / s, and a laser peak power of 1.5 GW. Start the cutting process, and the laser cuts according to the predetermined optimal cutting path. During the cutting process, in-situ sensors (such as optical sensors, temperature sensors) monitor the cutting state in real time, and if abnormalities (such as cutting deviation, overheating, etc.) are detected, the cutting parameters such as laser power, focal point position and speed are adjusted through a feedback mechanism to ensure cutting quality.

[0221] Finally, the cut silicon carbide wafer is collected. Due to the use of optimized cutting path and real-time parameter adjustment, the defects of the wafer are minimized, while meeting or exceeding market quality requirements.

[0222] Table 1: Performance comparison of silicon carbide wafers cut by traditional method and the method of the present application

[0223] Wafer No. Cutting method Average total dislocations (per cm2) of the wafer 2 )]]> Wafer yield 1 Conventional method 4560 80 2 Conventional method 3870 82 3 Conventional method 4210 85 4 Inventive method 1530 95 5 Inventive method 1800 96 6 Inventive method 1720 97

[0224] As can be seen from the data in Table 1, the wafers cut by the method of the present application (wafer numbers 4-6) are significantly better than the wafers cut by the traditional method (wafer numbers 1-3) in various performance indicators.

[0225] First, in terms of average total dislocation, the average defect density of wafers cut by the traditional method is 4560, 3780 and 4210 per cm 2 , indicating that there are many defects inside the wafer. The average defect density of the wafer cut by the method of the present application is significantly reduced, at 1530, 1800 and 1720 per cm 2 , which shows that the method of the present application can effectively detect and avoid the defect area inside the crystal, significantly improving the quality of the wafer.

[0226] In terms of wafer yield, the yield of the traditional method is between 80% and 85%, while the yield of the method of the present application is increased to more than 95%. This means that the method of the present application significantly reduces the waste rate in the production process while significantly improving the quality of the wafer, improving production efficiency and reducing production cost.

[0227] In summary, the method of the present application realizes accurate identification and positioning of internal defects in silicon carbide crystals by introducing 3D defect detection and artificial intelligence optimization technology. Using machine learning algorithms such as CNN, the optimal cutting path can be predicted to maximize the avoidance of defect-rich areas. During the cutting process, the cutting parameters are adjusted in real time through real-time monitoring and feedback to ensure the stability and consistency of the cutting quality.

[0228] In contrast, the conventional cutting method lacks effective detection of internal defects of the crystal and optimization of the cutting path, resulting in high wafer defect density, low wafer yield, and limiting the application and market competitiveness of the silicon carbide wafer.

[0229] Therefore, the data of the present embodiment sufficiently prove the significant advantages of the method of the present application in improving wafer quality, optimizing cutting efficiency, and reducing production cost, embodying the innovation and practical value of the present application, and having an important promoting effect on the production and application of the silicon carbide wafer.

[0230] It should be understood that, during the development of any actual implementation, numerous implementation decisions can be made, as in any engineering or design project. Such development efforts might be complex and time-consuming, but would be a routine undertaking for those of ordinary skill in the art having the benefit of this disclosure, without undue experimentation.

[0231] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application but not limit the present application, and although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present application, and all should be covered in the scope of the claims of the present application.

Claims

1. A method for cutting silicon carbide crystals based on 3D defect detection, characterized in that: include, X-ray computed tomography (CT) scanning equipment is used to perform all-round, high-resolution 3D scanning of silicon carbide crystals to obtain information on the internal structure and defects of the crystals. Silicon carbide crystal fixing equipment is used to fix silicon carbide crystals. It has rotation and vertical movement functions to ensure that the position of the crystal is precise and controllable during scanning and cutting. The 3D data processing module is connected to the X-ray computed tomography (CT) scanning equipment and includes a defect data module, a defect data comparison and analysis module, and an image processing and generation module. After obtaining the X-ray data signal, the 3D data processing module obtains a 3D defect image of the silicon carbide crystal. The cutting data processing module is connected to the 3D data module. After receiving 3D defect data and images, the cutting data module processes and calculates to form the optimal cutting path method. The cutting system, connected to the cutting data processing module, sets the cutting process parameters according to the optimal cutting path, and finally forms a silicon carbide wafer; The X-ray computed tomography (CT) scanning equipment uses a micro-focus high-energy X-ray tube. The accelerating voltage range of the X-ray source is 200–350 kV, the focal spot size is ≤10 micrometers, the exposure time is ≥1000 ms, and the spatial resolution is ≤80 micrometers. The defect data module includes a microtube defect data module, a carbon-encapsulated defect data module, and a polycrystalline / polymorphic defect data module; the defect data comparison and analysis module classifies different defect data; the image generation module assigns different colors to different defect types based on the comparison and analysis results, with data within the normal range represented by black; Based on 3D defect image data, the cutting data processing module includes a defect distribution analysis module and a cutting path analysis module. The cutting path analysis module of the cutting data processing module has four silicon carbide defect levels, with Class A defects having a defect index of <1000 defects / cm. 2 Class B level defect index: 1000-3000 defects / cm 2 Class C defect index: 3000-8000 defects / cm 2 Class D defect index > 8000 defects / cm 2 ; The cutting method based on the above cutting system follows these steps: Step 1: Perform surface cleaning treatment on the silicon carbide crystal; Step 2: Place the silicon carbide crystal on the fixed device, set the rotation or movement speed, the X-ray tomography CT scan emits X-rays, the silicon carbide crystal rotates or moves up and down, and the X-ray data signal is acquired; Step 3: The X-ray data signal is input into the 3D data processing module. Based on the differences in X-ray absorption or scattering at different defect locations, the X-ray data is processed and analyzed, and finally, a 3D defect image is generated by the image generation module. Step 4: Input 3D defect image data into the cutting data module. The cutting data module analyzes the data based on the four silicon carbide defect levels to obtain the optimal cutting path. Step 5: The cutting system cuts according to the optimal cutting path, selects appropriate cutting parameters, and finally obtains silicon carbide wafers of different specifications; In step 4, the cutting data module performs data analysis based on four silicon carbide defect levels, and sets the cutting path to pass through the defect-dense area to the maximum extent. The first cut ultimately yields silicon carbide wafers of different thicknesses with minimized surface defects. For silicon carbide wafers of different thicknesses with minimized surface defects obtained from the first cut, further cutting path analysis can be performed to obtain silicon carbide wafers with minimized defects that meet the thickness requirements from the second cut.

2. The silicon carbide crystal cutting method based on 3D defect detection according to claim 1, characterized in that: In step 4, the cutting data module performs data analysis based on the four silicon carbide defect levels, and sets the final cut silicon carbide wafer thickness to be consistent, either 0.35mm or 0.5mm, thus obtaining the optimal cutting path with consistent wafer thickness.

3. The silicon carbide crystal cutting method based on 3D defect detection according to claim 2, characterized in that: The X-ray data signal is input into the 3D data processing module, which specifically includes: The FDK cone-beam CT reconstruction algorithm was used to process X-ray data and reconstruct a 3D model of silicon carbide crystal, where the reconstructed voxel intensity I(x,y,z) represents the attenuation coefficient of each point in the crystal. Using a deep learning neural network artificial intelligence algorithm trained on labeled defect data, a 3D model is analyzed to identify defect types such as microtubules, carbon encapsulation, and polycrystalline regions. The neural network outputs the defect classification probability P for each voxel. defect (x,y,z).

4. The silicon carbide crystal cutting method based on 3D defect detection according to claim 3, characterized in that: The 3D defect image data input and cutting data processing module includes: The spatial distribution of defects is analyzed using machine learning algorithms trained with historical defect and cutting data to predict the optimal cutting plane. The optimal cutting path is calculated by minimizing the total defect density in the resulting wafer, using the formula: in, The total defect density, Let be the defect density of the i-th layer. Let be the thickness of the i-th layer; The cutting parameters for each cutting path segment are determined based on defect data and material properties, including cutting depth, speed, and power settings.

5. The silicon carbide crystal cutting method based on 3D defect detection according to claim 3 or 4, characterized in that: Cutting is performed according to the optimal cutting path, including: The cutting parameters are input from the cutting data processing module into the cutting system; The cutting process is initiated, and the laser cuts along a predetermined path. The initial settings are a cutting line width of 0.15 mm, a cutting speed of 3 mm / s, and a laser peak power of 1.5 GW. The cutting process is continuously monitored, and the laser power, focus, and speed are adjusted in real time based on feedback from the 3D data processing module and in-situ sensors to cope with changes in defect distribution encountered during the cutting process.

Images