Semiconductor material defect detection method based on domestic computing card and related equipment

Abstract

The invention discloses a semiconductor material defect detection method based on a domestic computing card and related equipment, and relates to the field of image data processing. The method comprises the following steps: acquiring a microstructure image of a to-be-detected semiconductor material; based on a machine vision technology, performing identification processing on the microstructure image through a semiconductor microstructure identification model to obtain microstructure features of the semiconductor material; judging whether the semiconductor material has defects or not based on the microstructure characteristics; and if yes, marking the semiconductor material as an unqualified material. According to the method, the microstructure image of the material can be accurately acquired and analyzed, and key characteristics such as surface defects, lattice structure integrity and doping uniformity can be comprehensively evaluated, so that the defects of the semiconductor material can be quickly and automatically detected, and the detection accuracy and efficiency are effectively improved. According to the method, the domestic computing card is utilized, so that the dependence on an imported computing platform is reduced, the cost is reduced, and the autonomous controllability and safety of the technology are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of image data processing, and in particular to a semiconductor material defect detection method and related equipment based on domestic computing cards. Background Art

[0002] Defect detection of semiconductor materials is crucial for ensuring the performance, reliability, and safety of electronic devices. As the size of semiconductor devices continues to shrink, any tiny defect may lead to a decline in device performance or even failure, thus affecting the stability of the entire system. Therefore, accurately detecting and analyzing defects in semiconductor materials can identify potential problems in advance, prevent failures, reduce production costs, and ensure that the final products can meet strict quality standards.

[0003] In related technologies, semiconductor material defect detection technologies mainly rely on automated optical detection equipment, such as optical microscopes, electron microscopes, and automatic optical inspection (AOI) systems. These detection devices are often limited by the detection speed, accuracy, and insufficient ability to identify complex defects, resulting in low detection efficiency. Summary of the Invention

[0004] Aiming at the above technical problems and defects, the purpose of the present invention is to provide a semiconductor material defect detection method and related equipment based on domestic computing cards, which utilize domestic computing cards and adopt machine vision technology to improve the efficiency and accuracy of semiconductor material defect detection.

[0005] To achieve the above purpose, in the first aspect, the present invention provides a semiconductor material defect detection method based on domestic computing cards, which is applied to a server loaded with domestic computing cards. The method includes: obtaining a microscopic structure image of the semiconductor material to be detected; based on machine vision technology, performing identification processing on the microscopic structure image through a preset semiconductor microscopic structure identification model to obtain the microscopic structure characteristics of the semiconductor material, where the microscopic structure characteristics include surface defect characteristics, lattice structure integrity, and doping uniformity; judging whether there are defects in the semiconductor material based on the microscopic structure characteristics; if so, marking the semiconductor material as a non-conforming material.

[0006] By obtaining the microscopic structure image of the semiconductor material, the present invention can comprehensively and accurately understand the state of the material. By using machine vision technology to obtain microscopic structure features including surface defect characteristics, lattice structure integrity, and doping uniformity, the material can be evaluated from multiple dimensions. Based on these features, it is determined whether the material has defects, improving the accuracy and efficiency of detection. Finally, the unqualified materials are marked, facilitating subsequent processing and management, and contributing to improving the quality control level of semiconductor production. Moreover, the present invention uses domestic computing cards for high-speed computing, not only reducing the dependence on foreign imported technologies, but also enhancing the stability and security of the supply chain.

[0007] In combination with some embodiments of the first aspect, in some embodiments, determining whether the semiconductor material has defects based on the microscopic structure features includes: determining the surface defect score, lattice structure integrity score, and doping uniformity score of the semiconductor material according to the microscopic structure features; determining the defect probability of the semiconductor material according to the surface defect score, the lattice structure integrity score, and the doping uniformity score; and determining whether the semiconductor material has defects based on the defect probability.

[0008] Adopting the technical solution of the above embodiments, by calculating the surface defect score, lattice structure integrity score, and doping uniformity score, the qualitative analysis is transformed into quantitative data, making the defect judgment more accurate and objective. Through a preset defect quality inspection model, multiple microscopic structure features can be comprehensively considered to evaluate the overall defect probability of the material, providing a scientific and efficient evaluation method for the quality control of semiconductor materials, thereby effectively reducing the risk of device failure caused by material defects.

[0009] In combination with some embodiments of the first aspect, in some embodiments, determining the defect probability of the semiconductor material according to the surface defect score, the lattice structure integrity score, and the doping uniformity score includes: calculating the defect probability through a preset defect quality inspection model, and the defect quality inspection model includes:

[0010] wherein, D is the defect probability; x j is the input feature, including the surface defect score, the lattice structure integrity score, and the doping uniformity score; v ij is the weight from the input layer to the hidden layer; w i is the weight from the hidden layer to the output layer; bi is the bias term of the hidden layer; c is the bias term of the output layer; σis an activation function; m is the number of neurons in the input layer; n is the number of neurons in the hidden layer.

[0011] By adopting the technical solution of the above embodiment, a specific deep learning model is introduced to calculate the defect probability. This model can learn the complex non-linear relationship between the input features and the defect probability. By adjusting the weights and bias terms in the model, the accuracy of defect detection can be optimized, enabling the model to adapt to different detection scenarios and material types. This method improves the flexibility and adaptability of defect detection, making the defect detection process more intelligent, and contributing to the improvement of the quality of semiconductor materials and the reliability of production.

[0012] In combination with some embodiments of the first aspect, in some embodiments, before calculating the defect probability through a preset defect quality inspection model, it further includes: determining the material type of the semiconductor material; adjusting the parameters of the defect quality inspection model based on the material type.

[0013] By adopting the technical solution of the above embodiment, by determining the type of semiconductor material and adjusting the parameters of the defect quality inspection model accordingly, the defect detection process becomes more accurate. This method takes into account that different materials may have different defect sensitivities and quality standards. By adjusting the model parameters, the pertinence and effectiveness of defect detection can be improved, ensuring that the detection results of different types of materials are more in line with the actual application requirements, and improving the accuracy of defect detection and the qualified rate of materials.

[0014] In combination with some embodiments of the first aspect, in some embodiments, after marking the semiconductor material as a non-conforming material, it further includes: determining the identification mark of the non-conforming material; generating a sorting instruction according to the identification mark; sending the sorting instruction to the material sorting device so that the material sorting device sorts the non-conforming material to the non-conforming material concentration area.

[0015] By adopting the technical solution of the above embodiment, the automatic sorting process of non-conforming materials is realized. By generating the identification mark and sorting instruction of non-conforming materials and sending the instruction to the sorting device, non-conforming materials can be quickly and accurately separated from the production line. This automatic sorting mechanism improves the efficiency and accuracy of material handling, reduces manual intervention, lowers production costs, and at the same time ensures the continuity of the production line and the high efficiency of material management.

[0016] In combination with some embodiments of the first aspect, in some embodiments, after determining the identification mark of the non-conforming material, it further includes: generating a non-conforming material prompt message according to the identification mark; sending the non-conforming material prompt message to the administrator terminal.

[0017] Adopting the technical solution of the above embodiment, by generating a prompt message for unqualified materials and sending it to the administrator terminal, the rapid feedback and transparent management of unqualified material information are realized. The management personnel can timely understand the unqualified situation of the materials, quickly make a response, and take corresponding treatment measures. This timely information transmission mechanism improves the efficiency of problem handling, enhances the controllability of the production process, and helps to maintain the quality standards of semiconductor materials.

[0018] In combination with some embodiments of the first aspect, in some embodiments, after marking the semiconductor material as an unqualified material, it further includes: generating a quality inspection record of the semiconductor material; storing the quality inspection record of the semiconductor material in the data storage module.

[0019] Adopting the technical solution of the above embodiment, by generating and storing the quality inspection record of the semiconductor material, detailed data support is provided for quality control and subsequent analysis. These records contain the whole process information of material detection, which is convenient for tracking and auditing, and helps to continuously improve the quality control process. At the same time, these data can also be used to analyze the causes of material defects, optimize the production process, and improve the overall quality of the materials, which is of great significance for improving the quality management level of the semiconductor industry.

[0020] In a second aspect, an embodiment of the present invention provides a server, including: one or more processors and a memory; the memory is coupled to the one or more processors, and the memory is used to store computer program code, the computer program code includes computer instructions, and the one or more processors call the computer instructions to enable the server to execute the methods described in the first aspect or the second aspect, and any possible implementation manner in the first aspect or the second aspect.

[0021] In a third aspect, the present invention provides a computer-readable storage medium, including instructions, when the above instructions run on the above server, enabling the above server to execute the methods described in the first aspect or the second aspect, and any possible implementation manner in the first aspect or the second aspect.

[0022] In a fourth aspect, the present invention provides a computer program product including instructions, when the above computer program product runs on the above server, enabling the above server to execute the methods described in the first aspect or the second aspect, and any possible implementation manner in the first aspect or the second aspect.

[0023] It can be understood that the server provided in the second aspect above, the storage medium provided in the third aspect, and the computer program product provided in the fourth aspect are all used to execute the methods provided by the present invention. Therefore, the beneficial effects that can be achieved can refer to the beneficial effects in the corresponding methods, and will not be elaborated here.

[0024] One or more technical solutions provided by the present invention have at least the following technical effects or advantages: 1. Improve the accuracy and efficiency of defect detection: By integrating machine vision technology and deep learning models, the present invention realizes the automatic analysis of the microscopic structural characteristics of semiconductor materials and the precise calculation of defect probabilities. This method not only reduces the subjectivity and errors of manual inspection, but also significantly improves the speed and accuracy of defect detection, thus ensuring the high-quality standards of semiconductor materials, reducing the risk of device failure, and improving production efficiency.

[0025] 2. Realize the intelligence and automation of the defect detection process: Through the collaborative work of a preset defect quality inspection model and automated sorting equipment, the present invention realizes the full-process automation from defect detection to material sorting. This intelligent detection process reduces manual intervention, lowers labor costs, and improves the consistency and reliability of material handling. At the same time, the automated sorting mechanism ensures that unqualified materials can be identified and isolated in a timely manner, preventing unqualified products from flowing into subsequent production processes.

[0026] 3. Optimize quality control and production management: The present invention also includes generating and storing detailed semiconductor material quality inspection records, which provide important data support for quality control. By analyzing these data, the quality changes of materials can be traced, potential problems in the production process can be identified, and the production process can be optimized accordingly. In addition, the real-time prompt information of unqualified materials helps managers quickly respond to quality problems and take effective measures, thus maintaining the stability of the entire production process and the high-standard quality of materials. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The accompanying drawings here are incorporated into the specification and form a part of this specification, showing embodiments consistent with the present invention, and are used together with the specification to explain the principles of the present invention. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can obtain other drawings based on these drawings without creative efforts. In the drawings: Figure 1 is a schematic flowchart of a method for detecting defects in semiconductor materials based on domestic computing cards according to an embodiment of the present invention; Figure 2 is a schematic flowchart of another method for detecting defects in semiconductor materials based on domestic computing cards according to an embodiment of the present invention; Figure 3 is a schematic architecture diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0028] The terms used in the following embodiments of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the specification of the present invention, the singular forms "a", "an", "the above", "the", and "this" are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used in the present invention refers to any or all possible combinations including one or more of the listed items.

[0029] Hereinafter, the terms "first" and "second" are only used for descriptive purposes and cannot be construed as implying relative importance or implicitly indicating the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of the embodiments of the present invention, unless otherwise specified, the meaning of "a plurality" is two or more.

[0030] It should also be noted that, unless otherwise clearly specified and defined, in the embodiments of the present invention, terms such as "arrangement" and "connection" should be understood in a broad sense. For example, "connection" can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, and it can be the communication inside two components; it can be a wired communication connection or a wireless communication connection. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances. The embodiments of the present invention will be specifically described below.

[0031] The embodiments of the present invention provide a method for detecting semiconductor material defects based on domestic computing cards. By utilizing the high-performance computing capabilities of domestic computing cards, an efficient and reliable method for detecting semiconductor material defects is provided. This method can accurately acquire and analyze microscopic structure images of materials, comprehensively evaluate key features such as surface defects, lattice structure integrity, and doping uniformity, thereby realizing fast and automated detection of semiconductor material defects, effectively improving the accuracy and efficiency of detection.

[0032] Among them, domestic GPUs or processor chips can be considered, such as the Xiyun® C500 series, Huawei's Ascend series of AI processors, Cambricon's DaVinci architecture AI chips, Unigroup's FPGA products, Jingjiawei's JM5400 graphics processing chips, and Loongson's LoongArch architecture processors, etc. These domestic computing cards all have high enough performance to meet the performance requirements of complex data processing.

[0033] The following will be combined with Figure 1Specifically introduce the semiconductor material defect detection method of this embodiment. This method is applied to a server equipped with the above domestic computing card, and specifically includes the following steps: Step 201, obtain the microscopic structure image of the semiconductor material to be detected.

[0034] Specifically, the server is connected to a high-resolution imaging device, which includes a scanning electron microscope (SEM) or a transmission electron microscope (TEM). The server sends instructions through the control interface to adjust the parameters of the microscope, such as focal length, magnification, and electron beam intensity, and then captures the signals generated by the interaction between the sample and the electron beam. These signals are converted into digital image data and transmitted back to the server. The image processing software on the server further processes these data to ensure image clarity and contrast, and finally forms a high-quality microscopic structure image for analysis.

[0035] Step 202, based on machine vision technology, perform recognition processing on the microscopic structure image through a preset semiconductor microscopic structure recognition model to obtain the microscopic structure characteristics of the semiconductor material.

[0036] Among them, the microscopic structure characteristics may include but are not limited to surface defect characteristics, lattice structure integrity, and doping uniformity.

[0037] Surface defect characteristics refer to the irregularities on the surface of the semiconductor material, such as cracks, scratches, pits, or particle contaminants. These defects will seriously affect the electrical properties and mechanical strength of the material. Therefore, the detection of surface defect characteristics is a key step to ensure the quality of semiconductor materials.

[0038] Lattice structure integrity refers to the perfection degree of the crystal structure inside the semiconductor material, including the consistency of lattice constants and the distribution of defects such as dislocations and stacking faults in the lattice. The integrity of the lattice structure is directly related to the electron mobility of the material and the performance of the device. Therefore, evaluating the integrity of the lattice structure is crucial for the reliability of semiconductor materials.

[0039] Doping uniformity involves the consistency of the distribution of dopants in the semiconductor material. Dopants such as boron and phosphorus are used to change the conductivity of the material. Uneven doping will lead to inconsistent electrical properties of the material and affect the performance and stability of the device. Therefore, detecting doping uniformity is of great significance for manufacturing high-performance semiconductor devices.

[0040] In this embodiment, the server can utilize machine vision technology to determine the microscopic structure characteristics of semiconductor materials based on microscopic structure images. First, the image is preprocessed using image processing software, including denoising, enhancing contrast, and correcting chromatic aberration. Then, key features in the image, such as surface defects, lattice structure integrity, and doping uniformity, are extracted through machine vision algorithms. Specifically, the server runs a pre-trained semiconductor microscopic structure recognition model, which can identify and quantify surface defects such as cracks and scratches in the image, evaluate the integrity of the lattice and dislocation density, and analyze the distribution pattern of dopants.

[0041] Among them, the training process of the semiconductor microscopic structure recognition model is as follows: 1) Data collection and annotation: Collect a large number of microscopic structure images of semiconductor materials, which may be from different devices and conditions to ensure the generalization ability of the model. Manually annotate the key features in the image, such as the positions of cracks and scratches, the types and positions of lattice defects, and the distribution of dopants. These annotations will serve as the "ground truth" for training the model.

[0042] 2) Data preprocessing: Normalize the images, scale the pixel values to a unified range, which is usually from 0 to 1. Perform data augmentation operations such as cropping, rotating, or flipping the images to simulate different defect morphologies and lighting conditions and increase the robustness of the model. Divide the images into training set, validation set, and test set, usually in the ratio of 70%, 15%, 15%.

[0043] 3) Model architecture design: Select a deep learning architecture suitable for image recognition, such as ResNet, Inception, or a custom convolutional neural network (CNN). Determine the structural parameters of the network, such as the number of layers, convolutional kernel size, pooling layer, and fully connected layer.

[0044] 4) Loss function and optimizer selection: Select a suitable loss function according to the nature of the task, such as cross-entropy loss for classification tasks and mean squared error loss for regression tasks. Select an optimization algorithm, such as Adam, SGD, etc., to update the network weights.

[0045] 5) Model training: Use the training set data to train the model, calculate the predicted values through forward propagation, calculate the gradients through backward propagation, and update the weights. Regularly evaluate the model performance on the validation set to monitor overfitting, and adjust the model parameters or stop training in advance as needed.

[0046] 6) Performance evaluation and tuning: Evaluate the final performance of the model on an independent test set, using metrics such as accuracy, recall, and F1-score. Tune the model according to the test results, which may include adjusting the network structure, changing the loss function, or optimizing the optimizer parameters.

[0047] 7) Model deployment: Deploy the trained semiconductor microstructure recognition model to the server so that it can receive new microstructure images and perform real-time recognition.

[0048] Through the above process, the training of the semiconductor microstructure recognition model can be completed.

[0049] Step 203, based on the microstructure features, determine whether there are defects in the semiconductor material.

[0050] Specifically, the server analyzes the microstructure image of the semiconductor material by running a deep learning model (such as the semiconductor microstructure recognition model mentioned above). This model can identify and quantify surface defects such as cracks and scratches in the image, evaluate the integrity of the lattice structure, including dislocations and stacking faults, etc., and analyze the distribution pattern of dopants. The model converts these features into quantitative data and then compares them with pre-set thresholds or standards. If the quantitative data exceeds these thresholds, it indicates that the material has defects. For example, if the number or size of surface defects exceeds the allowed range, or the dislocation density of the lattice structure is higher than a specific standard, the model will mark the material as defective. In addition, the non-uniformity of the dopant distribution may also trigger a defect label.

[0051] In this embodiment, when the server analyzes the microstructure features of the semiconductor material, it compares each feature with pre-set standards, which are determined based on the standards of material performance and historical data. If any microstructure feature exceeds these standards, such as the size, number of surface defects, or the dislocation density in the lattice structure exceeds the standard range, or the distribution of dopants deviates significantly from the expected pattern, the server will determine that the semiconductor material has defects and proceed to step 204.

[0052] Conversely, if all microstructure features are within the standard range, or only a few microstructure features slightly exceed but are not sufficient to significantly affect the overall performance of the semiconductor material, then the server will determine that the semiconductor material has no defects and consider that the semiconductor material can continue to be used in the manufacturing process.

[0053] This judgment method provides a strict and flexible evaluation mechanism for the quality control of materials, reducing unnecessary waste of semiconductor materials.

[0054] Step 204, mark the semiconductor material as a non-conforming material.

[0055] Specifically, once the server determines through the above analysis that there are defects in the semiconductor materials, it will automatically execute the marking program to mark these materials as unqualified. This step involves the integration of the defect management system. The server will record the relevant information of the unqualified materials in the database and may trigger subsequent processing procedures, such as isolation, rework, or scrapping. Such automated marking not only improves production efficiency but also helps maintain the quality control standards of the entire production line, ensuring that only materials meeting the quality requirements can be used in subsequent manufacturing processes.

[0056] This embodiment adopts the above semiconductor material defect detection method based on domestic computing cards, which can improve the quality control level in the semiconductor manufacturing process. First, through automated image analysis and feature recognition, the subjectivity and errors of manual inspection are reduced, and the accuracy and consistency of defect detection are improved. Second, using domestic computing cards for high-speed computing not only reduces the dependence on foreign imported technologies but also enhances the stability and security of the supply chain. Especially in the context of current global supply chain fluctuations, this is particularly important. In addition, this method can evaluate and classify materials in real time, greatly improving production efficiency and reducing waste and cost losses caused by defective materials. At the same time, by accurately identifying and excluding defective materials, the reliability and performance of semiconductor devices are ensured.

[0057] This embodiment reduces the dependence on imported computing platforms, lowers the cost of the semiconductor material detection system, and improves the autonomy, controllability, and security of the system. Through the powerful processing capabilities of domestic computing cards, efficient image processing and defect detection are achieved, improving the accuracy and efficiency of semiconductor material quality detection.

[0058] Next, in combination with Figure 2 to further describe the method of this embodiment, which specifically includes the following steps: Step 301, obtain a microscopic structure image of the semiconductor material to be detected.

[0059] Step 302, perform recognition processing on the microscopic structure image through a preset semiconductor microscopic structure recognition model to obtain the microscopic structure characteristics of the semiconductor material.

[0060] Step 303, determine the surface defect score, lattice structure integrity score, and doping uniformity score of the semiconductor material according to the microscopic structure characteristics.

[0061] Specifically, the server uses image processing and machine learning algorithms to detect and quantify surface irregularities such as cracks and scratches in the image, thereby determining the surface defect score of the semiconductor material. Specifically, the server first preprocesses the high-resolution microstructure image, including denoising and enhancement, to improve the recognizability of defect features. Then, it calls the preset surface defect evaluation model. The surface defect evaluation model is trained with a large number of labeled defect images and can identify tiny cracks and scratches in the image. The surface defect evaluation model calculates parameters such as the number, size, depth, and distribution density of the defects, and converts these parameters into a quantified surface defect score, which reflects the severity of the surface defects.

[0062] Among them, the training process of the surface defect evaluation model includes: First, collect and label a large number of semiconductor material surface defect images, which contain various defect instances such as cracks and scratches. Then, using these labeled data, data scientists will select a suitable deep learning architecture, such as a convolutional neural network (CNN), and design the model to identify and locate the defects in the image. In the training stage, the model learns to extract features from the input image through forward propagation and adjusts the network weights through backpropagation to minimize the prediction error. Through multiple iterations of training, the model gradually learns to identify different types of surface defects and can quantify key parameters such as the number, size, depth, and distribution density of the defects. Finally, the model can synthesize these parameters into a quantified surface defect score.

[0063] At the same time, the server evaluates the integrity of the lattice structure by analyzing the microstructure image and obtains the lattice structure integrity score. First, extract the regularity and symmetry features of the lattice arrangement, and then use the lattice structure evaluation model to identify and quantify lattice defects such as dislocations and stacking faults. The lattice structure evaluation model can also analyze the degree of lattice distortion, including the deviation of lattice constants and the situation of lattice distortion. After these features are converted into numerical data, according to the preset scoring criteria, which are based on the requirements for lattice integrity in materials science and historical data analysis, a comprehensive score is calculated, that is, the lattice structure integrity score. This score reflects the perfection degree of the internal lattice structure of the material. The lower the score, the more complete the lattice structure; the higher the score, the more lattice structure defects, thus providing a quantified evaluation basis for the quality control of semiconductor materials.

[0064] Among them, the training process of the lattice structure evaluation model involves collecting a large amount of lattice structure image data, which includes images of normal lattice structures and various lattice defects such as dislocations and stacking faults. These images are used for annotation to determine the type, location, and degree of lattice defects. Subsequently, a suitable deep learning architecture, such as a convolutional neural network, is selected to build a model that can extract features related to lattice integrity from the images. During the training phase, the model learns to identify lattice defects and distortions by learning the features in the images and adjusts the network parameters to minimize the prediction error. Model training also includes validation and testing phases to evaluate its accuracy and generalization ability. Finally, the trained model can convert the identified lattice features into numerical data and calculate a comprehensive score, namely the lattice structure integrity score, based on the requirements for lattice integrity in materials science and historical data analysis, thereby providing an accurate quantitative assessment for the quality control of semiconductor materials.

[0065] On the other hand, the server can also calculate the doping uniformity score based on the analysis of the distribution pattern and concentration consistency of dopants in the material. The server uses specific image processing algorithms to identify and isolate the doping regions, and then uses the doping uniformity evaluation model to evaluate the distribution uniformity of the dopants. The doping uniformity evaluation model can identify the local variations and overall patterns of dopant concentrations and compare these variations with the ideal uniform distribution. By calculating the deviation between the actual distribution and the ideal distribution, the uniformity of doping can be quantified. Finally, according to the preset scoring rules, which may be based on statistical metrics such as the standard deviation of dopant concentration, the uniformity index of the distribution, or other statistical measures, the doping uniformity score is calculated. This score reflects the distribution consistency of dopants in the material, with a higher score indicating more uniform doping and a lower score indicating uneven dopant distribution.

[0066] Among them, the training process of the doping uniformity evaluation model first requires collecting a series of image data on the distribution of dopants in semiconductor materials, which includes samples with different doping levels and distribution patterns. Then, these images are annotated manually or automatically to determine the actual distribution of the dopants. Next, a suitable machine learning model, such as a convolutional neural network (CNN), is selected and trained using these annotated data so that it can identify and understand the local variations and overall patterns of dopant distribution. During the training process, the model learns how to compare the actual doping distribution with the ideal uniform distribution and calculate the deviation between the two. Through multiple iterations and parameter adjustments, the model gradually improves its ability to identify and quantify doping uniformity. Finally, the model can automatically calculate the doping uniformity score according to the preset scoring rules, such as statistical metrics like the standard deviation of dopant concentration and the distribution uniformity index, providing an accurate assessment for the quality control of semiconductor materials.

[0067] Step 304: Determine the defect probability of the semiconductor material based on the surface defect score, the lattice structure integrity score, and the doping uniformity score.

[0068] Specifically, the server combines the surface defect score, the lattice structure integrity score, and the doping uniformity score to determine the defect probability of the semiconductor material. This process is achieved by integrating a comprehensive evaluation algorithm, which may use weighted summation or more complex statistical models, and assigns different weights according to the importance of each score's impact on the material properties. The server inputs each score into the algorithm, and the algorithm calculates a comprehensive defect probability value according to the preset logic and thresholds, such as the set defect tolerance and historical defect data. If this probability value exceeds the set risk threshold, the server determines that the semiconductor material has defects; conversely, if the probability value is below the threshold, the material is considered to meet the quality standards and can continue to be used in the subsequent production process. This method makes the defect judgment more scientific and objective, improving the accuracy and efficiency of defect detection.

[0069] In some embodiments, this step may specifically include: Calculate the defect probability through a preset defect quality inspection model, and the defect quality inspection model includes:

[0070] Where, D is the defect probability; x j are input features, including the surface defect score, the lattice structure integrity score, and the doping uniformity score; v ij are the weights from the input layer to the hidden layer; w i are the weights from the hidden layer to the output layer; bi is the bias term of the hidden layer; c is the bias term of the output layer; σ is the activation function; m is the number of neurons in the input layer; n is the number of neurons in the hidden layer.

[0071] Furthermore, in the above defect quality inspection model, x j (j = 1, 2, 3) represents each component in the input feature vector, and these components specifically include the above three key microstructure scores, namely the surface defect score, the lattice structure integrity score, and the doping uniformity score, which together constitute the input layer of the model: 1. Surface defect score x1: This feature reflects the degree of surface defects of semiconductor materials, including the quantity, size, and distribution density of surface irregularities such as cracks and scratches. This score is obtained through a surface defect assessment model, which quantifies the impact of surface defects on material quality.

[0072] 2. Lattice Structure Integrity Score x 2: This feature measures the integrity of the internal lattice structure of the material, involving dislocation density, stacking faults, and lattice distortion, etc. The lattice structure integrity score is provided by a lattice structure assessment model, which reveals the potential impact of lattice defects on material properties.

[0073] 3. Doping Uniformity Score x 3: This feature evaluates the distribution uniformity of dopants in the material, including local concentration variations and overall distribution patterns. The doping uniformity score is calculated by a doping uniformity assessment model, which indicates the impact that doping non-uniformity may have on device performance.

[0074] These three input features together constitute the input layer of the neural network. Each feature is a neuron, and they pass their respective score values to the hidden layer of the network. These scores serve as the input to the model, enabling the neural network to comprehensively consider multiple factors and thus more accurately predict the defect probability of semiconductor materials. In this way, the model can learn the relationships between each feature and material defects and make predictions accordingly.

[0075] In this embodiment, the defect quality inspection model can be a neural network model based on a multi-layer perceptron (MLP) for calculating the defect probability of semiconductor materials. This model can accurately predict whether a semiconductor material has defects by learning the non-linear relationship between the input features (surface defect score, lattice structure integrity score, and doping uniformity score) and the defect probability.

[0076] The activation function in the formula σ usually adopts the Sigmoid function, for example σ ( x )=(1 + e −x ) -1 , where x represents the input variable of the activation function σ , and the activation function σ can map x to a value between 0 and 1, which can be interpreted as a probability or as the output of a binary classification problem in some cases. This ensures that the model can handle non-linear problems.

[0077] And the multi-layer structure (input layer, hidden layer, and output layer) enables the model to have sufficient complexity to capture the interactions between input features.

[0078] weight v ij 、 w i and the bias term b i a, b, and c are optimized through the training process, and they determine the sensitivity of the model to input features and the final prediction result.

[0079] This data-driven prediction method can effectively convert complex microstructural features into defect probabilities, providing an efficient and accurate tool for the quality control of semiconductor materials.

[0080] In some embodiments, before calculating the defect probability through a preset defect quality inspection model, it is also necessary to determine the material type of the semiconductor material, and then adjust the parameters of the defect quality inspection model based on the material type.

[0081] Because different types of semiconductor materials may have different quality standards and defect sensitivities. This process involves the analysis of material properties, which may include chemical composition analysis, crystal structure identification, etc., to ensure accurate identification of the material type.

[0082] Once the material type is determined, the parameters of the defect quality inspection model will be adjusted according to the characteristics and historical defect data of the material, which includes adjusting the weights of input features, the number of neurons in the hidden layer, the selection of activation functions, and other hyperparameters of the model.

[0083] In this process, it is first necessary to collect and analyze the historical defect data and performance indicators of a specific type of material to understand the sensitivity and impact of the material to different defects. Then, use this information to adjust the weights related to input features in the model. For example, if a certain semiconductor material is more sensitive to surface defects, then increase the weight of the surface defect score in the model. At the same time, it may be necessary to adjust the number of neurons and connection weights in the hidden layer, as well as the bias term in the output layer, to optimize the model's prediction ability for the defect probability of a specific semiconductor material. In addition, the model can be further fine-tuned by selecting different activation functions or optimization algorithms to ensure that the model can accurately capture the unique defect characteristics and patterns of the semiconductor material. Finally, through cross-validation and model evaluation, the optimal parameter settings are determined to ensure that the defect quality inspection model can provide accurate defect probability predictions for specific types of semiconductor materials.

[0084] This embodiment ensures that the defect quality inspection model can be optimized for specific types of materials through this adjustment, improving the accuracy and reliability of defect prediction, because different materials may have different tolerances and responses to the same defect, thus requiring customized model parameters to adapt to these differences.

[0085] Step 305: Determine whether there are defects in the semiconductor material based on the defect probability.

[0086] Specifically, compare the calculated defect probability with a preset risk threshold. If the defect probability exceeds this threshold, it indicates that the possibility of defects in the material is relatively high and sufficient to affect the performance and reliability of the material. Proceed to Step 306. At this time, the server will determine that the semiconductor material has defects and mark it as a non-conforming material. This threshold is set based on the requirements of materials science, historical defect data, and quality control standards, ensuring the accuracy and consistency of defect judgment.

[0087] If the defect probability is lower than the threshold, it is determined that there are no defects in the semiconductor material, and the semiconductor material is considered to be within an acceptable quality range and can continue to be used in the manufacturing process.

[0088] Such a method provides a scientific and quantitative decision-making basis, optimizing the quality control process of semiconductor materials.

[0089] Step 306: Mark the semiconductor material as a non-conforming material.

[0090] After that, proceed to Step 307, 310, or 312.

[0091] Step 307: Determine the identification mark of the non-conforming material.

[0092] Specifically, the server generates a unique identification mark for each semiconductor material determined to be non-conforming according to predefined rules or algorithms. This mark usually contains key data such as the basic information of the material, defect type, defect level, and detection time for subsequent tracking and management. The server encodes this information into an identification mark in the form of a barcode, QR code, or RFID tag, etc., ensuring that each non-conforming material can be accurately identified and recorded.

[0093] Step 308: Generate a sorting instruction based on the identification mark.

[0094] Specifically, the server generates a sorting instruction based on the identification mark of the non-conforming material. These instructions detail the handling method of the material, including moving the material to a specific location or container, and any special handling measures that may be required. The sorting instruction will contain the necessary operation codes and parameters to ensure that the sorting equipment can understand and execute these instructions.

[0095] Step 309: Send the sorting instruction to the material sorting equipment so that the material sorting equipment sorts the non-conforming material to the non-conforming material concentration area.

[0096] Specifically, the server sends sorting instructions to the material sorting equipment. These instructions are transmitted to the sorting system through wireless networks, wired connections, or other communication interfaces, triggering the equipment to perform corresponding physical operations. After receiving the instructions, the sorting equipment will automatically identify and confirm the unqualified material according to the above identification mark, separate the unqualified material from the production line, and guide it to the designated centralized area for unqualified materials for subsequent processing, such as rework, recycling, or scrapping. This process realizes the automation of material management, improving production efficiency and the accuracy of material handling.

[0097] Among them, the material sorting equipment is an automated system specifically designed to classify semiconductor materials according to the sorting instructions sent by the server. This equipment usually integrates precise robotic arms, conveyor belts, identification systems (such as RFID readers, barcode scanners, or vision recognition systems), and control systems, which can identify the identification mark of each material and accurately sort the unqualified materials to the designated area according to the instructions. It can be seamlessly connected to the production management system to achieve efficient material tracking and quality control, ensuring the smooth operation of the production line and the reasonable allocation of materials.

[0098] Step 310, generate a prompt message for unqualified materials according to the identification mark.

[0099] Specifically, the server collects and integrates all key data related to the material, including the batch number of the material, the detected defect type, defect location, defect severity, and any other relevant quality control parameters. Then, the server converts these data into an easy-to-understand prompt message for unqualified materials, which is designed to provide managers with clear and accurate material status and defect details for taking corresponding measures.

[0100] Step 311, send the prompt message for unqualified materials to the administrator's terminal.

[0101] Specifically, the server sends the prompt message for unqualified materials to the administrator's terminal device, such as a computer, tablet, or smartphone, through the internal network or a dedicated communication line. This step ensures that managers can receive real-time notifications about unqualified materials, view detailed defect reports of the materials in a timely manner, and make decisions based on this information, such as initiating further investigations, adjusting the production process, or performing other quality management activities. This automated information transfer mechanism improves the response speed and processing efficiency, contributing to maintaining the integrity of the entire production and quality control process.

[0102] Step 312, generate a quality inspection record for semiconductor materials.

[0103] Specifically, the server integrates the data and analysis results related to the material quality inspection into a detailed record. This quality inspection record includes the basic information of the material, the microscopic structure images collected during the inspection, the surface defect score, the lattice structure integrity score, the doping uniformity score, as well as the final defect probability and the judgment result of whether the material is qualified. In addition, the record may also include the inspection time, operator information, inspection equipment information, and notes on any abnormal situations, providing complete documentation support for the quality control of semiconductor materials.

[0104] Step 313, store the semiconductor material quality inspection record in the data storage module.

[0105] This is to ensure the long-term preservation and secure backup of the quality inspection record, facilitating future queries, analysis, and audits. The server will transmit the record to a database or cloud storage service, and may also perform data encryption and access control to protect sensitive information and ensure data integrity. The storage module is designed to enable efficient retrieval and data sharing, allowing managers at different departments and levels to access the required quality inspection records according to their authorizations, thus supporting the continuous quality improvement and decision-making processes.

[0106] The embodiment of the present invention provides a semiconductor material defect detection method based on a domestic computing card, which realizes efficient and accurate quality control of semiconductor materials through a series of automated steps. First, this method uses the server of the domestic computing card to obtain the microscopic structure image of the semiconductor material, and extracts key microscopic structure features from it, including surface defect features, lattice structure integrity, and doping uniformity. These features are the basis for evaluating the material quality, and they are converted into defect probabilities through a preset defect quality inspection model, which is a deep learning network that can calculate the defect probability of the material according to the input features. Before calculating the defect probability, the system will also determine the material type and adjust the model parameters according to the material characteristics to meet the detection requirements of different materials.

[0107] When the defect probability exceeds the preset threshold, the system marks the material as unqualified and generates a unique identification label. This label is used to generate a sorting instruction to guide the automated material sorting equipment to separate and transfer the unqualified materials to a specific centralized area. At the same time, the system will also generate a prompt message for the unqualified materials and send it to the administrator terminal so that the management personnel can understand the situation in a timely manner and take corresponding measures.

[0108] In addition, this embodiment also includes generating and storing quality inspection records of semiconductor materials. These records detail the inspection processes and results of the materials, providing important data for quality control and subsequent analysis. Overall, this technical solution improves the efficiency and accuracy of semiconductor material defect detection through automated and intelligent means, reduces the reliance on manual inspection, enhances the controllability of the production process and the consistency of material quality, and is of great significance for enhancing the overall competitiveness of the semiconductor manufacturing industry.

[0109] The method provided in the above embodiment can be executed by a server, which is an electronic device. The following describes this electronic device in the embodiment of the present invention from the perspective of hardware processing. Please refer to Figure 3 , which is a schematic structural diagram of an entity device of the electronic device in the embodiment of the present invention.

[0110] It should be noted that Figure 3 the structure of the electronic device shown is only an example and should not impose any limitations on the functions and usage scope of the embodiments of the present invention.

[0111] As Figure 3 shown, the electronic device includes a Central Processing Unit (CPU) 401, which can perform various appropriate actions and processes according to the program stored in the Read-Only Memory (ROM) 402 or the program loaded from the storage section 408 into the Random Access Memory (RAM) 403, such as executing the method described in the above embodiment. In the RAM 403, various programs and data required for system operation are also stored. The CPU 401, ROM 402, and RAM 403 are connected to each other through a bus 404. The Input / Output (I / O) interface 405 is also connected to the bus 404.

[0112] The following components are connected to the I / O interface 405: an input section 406 including an audio input device, a button switch, etc.; an output section 407 including a liquid crystal display (LCD), an audio output device, an indicator light, etc.; a storage section 408 including a hard disk, etc.; and a communication section 409 including a network interface card such as a LAN (Local Area Network) card, a modem, etc. The communication section 409 performs communication processing via a network such as the Internet. A drive 410 is also connected to the I / O interface 405 as needed. A removable medium 411, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 410 as needed so that a computer program read therefrom can be installed into the storage section 408 as needed.

[0113] Specifically, according to an embodiment of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, an embodiment of the present invention includes a computer program product including a computer program carried on a computer-readable medium, the computer program including a computer program for performing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 409, and / or installed from the removable medium 411. When the computer program is executed by a central processing unit (CPU) 401, various functions defined in the present invention are executed.

[0114] It should be noted that specific examples of the computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a flash memory, an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present invention, the computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0115] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of systems, methods, and computer program products according to various embodiments of the present invention. Among them, each block in the flowchart or block diagram may represent a module, a program segment, or a part of code, and the above-mentioned module, program segment, or part of code contains one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order from that marked in the accompanying drawings.

[0116] Specifically, the electronic device of this embodiment includes a processor and a memory. The memory is coupled to one or more processors. The memory is used to store computer program code, and the computer program code includes computer instructions. One or more processors call the computer instructions to cause the electronic device to execute the method provided in the above-mentioned embodiment.

[0117] On the other hand, the present invention also provides a computer-readable storage medium. This storage medium may be included in the electronic device described in the above-mentioned embodiment; or it may exist separately and not be assembled into the electronic device. The above storage medium carries one or more computer programs. When the above one or more computer programs are executed by a processor of the electronic device, the electronic device is caused to implement the method provided in the above-mentioned embodiment.

[0118] As mentioned above, the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the various embodiments of the present invention.

[0119] As used in the above embodiments, depending on the context, the term "when..." may be interpreted to mean "if...", or "after...", or "in response to determining...", or "in response to detecting...". Similarly, depending on the context, the phrase "when determining..." or "if detecting (the stated condition or event)" may be interpreted to mean "if determining...", or "in response to determining...", or "when detecting (the stated condition or event)", or "in response to detecting (the stated condition or event)".

[0120] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by relevant hardware instructed by a computer program. This program can be stored in a computer-readable storage medium. When this program is executed, it can include the processes of the above method embodiments. The foregoing storage medium includes: various media that can store program codes, such as ROM or random access memory RAM, magnetic disks, or optical discs.

Claims

1. A method for detecting semiconductor material defects based on domestic computing cards, characterized in that, Applied to a server equipped with domestic computing cards, the method includes: Obtain a microscopic structure image of the semiconductor material to be detected; Based on machine vision technology, perform recognition processing on the microscopic structure image through a preset semiconductor microscopic structure recognition model to obtain the microscopic structure characteristics of the semiconductor material, where the microscopic structure characteristics include surface defect characteristics, lattice structure integrity, and doping uniformity; Judge whether there are defects in the semiconductor material based on the microscopic structure characteristics; If so, mark the semiconductor material as a non-conforming material.

2. The method according to claim 1, wherein The judging whether there are defects in the semiconductor material based on the microscopic structure characteristics includes: Determine the surface defect score, lattice structure integrity score, and doping uniformity score of the semiconductor material according to the microscopic structure characteristics; Determine the defect probability of the semiconductor material according to the surface defect score, the lattice structure integrity score, and the doping uniformity score; Judge whether there are defects in the semiconductor material based on the defect probability.

3. The method according to claim 2, wherein The determining the defect probability of the semiconductor material according to the surface defect score, the lattice structure integrity score, and the doping uniformity score includes: Calculate the defect probability through a preset defect quality inspection model, and the defect quality inspection model includes: ； Among them, D is the defect probability; x j is the input feature, including the surface defect score, the lattice structure integrity score, and the doping uniformity score; v ij is the weight from the input layer to the hidden layer; w i is the weight from the hidden layer to the output layer; bi is the bias term of the hidden layer; c is the bias term of the output layer; σ is the activation function; m is the number of neurons in the input layer; n is the number of neurons in the hidden layer.

4. The method according to claim 3, wherein Before calculating the defect probability through the preset defect quality inspection model, it further includes: Determine the material type of the semiconductor material; Adjust the parameters of the defect quality inspection model based on the material type.

5. The method according to any one of claims 1-4, characterized in that, After marking the semiconductor material as a non-conforming material, it further includes: Determine the identification mark of the non-conforming material; Generate a sorting instruction according to the identification mark; Send the sorting instruction to the material sorting device so that the material sorting device sorts the non-conforming material to the non-conforming material concentration area.

6. The method according to claim 5, wherein After determining the identification mark of the non-conforming material, it further includes: Generate a non-conforming material prompt message according to the identification mark; Send the non-conforming material prompt message to the administrator terminal.

7. The method according to claim 1, wherein After marking the semiconductor material as a non-conforming material, it further includes: Generate a semiconductor material quality inspection record; Store the semiconductor material quality inspection record in the data storage module.

8. A server, characterized in that, Includes one or more processors and a memory; The memory is coupled to the one or more processors, and the memory is used to store computer program code. The computer program code includes computer instructions, and the one or more processors call the computer instructions to cause the server to execute the method according to any one of claims 1-7.

9. A computer-readable storage medium storing computer instructions, characterized in that, When the computer instructions run on the server, the server is caused to execute the method according to any one of claims 1-7.

10. A computer program product, characterized in that, When the computer program product runs on the server, the server is caused to execute the method according to any one of claims 1-7.

Images