Defect detection system and method for semiconductor sample

Abstract

The invention relates to the technical field of semiconductors, and discloses a semiconductor sample defect detection system which comprises an imaging module, a data acquisition module, a data processing module, an automation and integration module and a user interface module. According to the defect detection system and method for the semiconductor sample, by integrating the imaging modules of the optical microscope, the electron microscope, the X-ray diffractometer and the super-resolution microscope, details of the semiconductor sample can be comprehensively captured from macroscopic to microscopic and from the surface to a crystal structure, so that precise recognition of tiny or deep defects is ensured; the application of the deep learning model further improves the accuracy of image processing, through feature extraction and intelligent identification, even defects difficult to perceive by naked eyes have no place to hide, and the high-precision detection capability greatly reduces the risk that quality problems are only found in subsequent production links, reduces the possibility of reworking and scrapping, and improves the production efficiency. Therefore, the product quality and the production yield are improved.

Description

Technical Field

[0001] The present invention relates to the field of semiconductor technology, and in particular to a semiconductor sample defect detection system and method. Background Art

[0002] In the production process of semiconductor optoelectronic devices, problems such as crystal defects, device defects, and surface contamination may cause device performance degradation or even failure, thereby reducing production yield. Therefore, comprehensive characterization of optoelectronic device materials can detect problems as early as possible in the chip manufacturing process, thereby minimizing losses. In addition, through in-depth analysis and induction of test results, potential problems in the chip manufacturing process can be revealed, providing more precise guidance for process optimization.

[0003] At present, KOH etching technology is widely used in the field of semiconductor defect detection. This technology mainly focuses on the identification of surface defects, such as common problems such as particle contamination, tiny scratches and impurities. Although KOH etching technology plays a role in some aspects, it also exposes several limitations in practical applications. First, the detection speed of this technology is relatively slow, which may become a bottleneck in a large-scale production environment and affect the overall production efficiency. Secondly, the accuracy needs to be improved. Some subtle or deep defects may not be effectively detected, resulting in quality problems being discovered in subsequent production links, increasing the risk of rework and scrapping. With the rapid development of the semiconductor industry, higher requirements are placed on defect detection technology, including more efficient detection speed and higher accuracy. Therefore, we propose a defect detection system and method for semiconductor samples. Summary of the invention

[0004] 1. Technical issues to be resolved

[0005] In view of the shortcomings of the prior art, the present invention provides a defect detection system and method for semiconductor samples, which significantly improves the detection speed and accuracy and solves the problems that, first, the detection speed of this technology is relatively slow, which may become a bottleneck in a large-scale production environment and affect the overall production efficiency; secondly, the accuracy needs to be improved, and some subtle or deep defects may not be effectively detected, resulting in quality problems being discovered in subsequent production links, increasing the risk of rework and scrapping.

[0006] (II) Technical solution

[0007] In order to achieve the above-mentioned purpose of significantly improving the detection speed and accuracy, the present invention provides the following technical solution: a defect detection system for semiconductor samples, including an imaging module, a data acquisition module, a data processing module, an automation and integration module and a user interface module.

[0008] Preferably, the imaging module is responsible for acquiring a high-resolution image of a semiconductor sample, and includes four submodules: an optical microscope, an electron microscope, an X-ray diffractometer, and a super-resolution microscope. The process of the imaging module acquiring a high-resolution image of a semiconductor sample includes the following steps:

[0009] S11. Sample preparation and positioning: First, ultrapure water and nitrogen are used to purge to remove oil and dust on the surface of the semiconductor sample. Then, fluorescent dyes are used to enhance the visibility of specific structures of the sample.

[0010] S12. Application of imaging technology: Optical microscopes obtain macroscopic and microscopic structural images of semiconductor samples; electron microscopes obtain more detailed microscopic structural analysis images of semiconductor samples; X-ray diffractometers obtain crystal structure images of semiconductor samples; super-resolution microscopes can achieve single-molecule resolution imaging and capture the finest detail images of semiconductor samples.

[0011] Preferably, the data acquisition module includes a high-speed data acquisition card with a sampling rate of 100MS / s, an analog-to-digital conversion unit, a data encoding unit, a high-speed data interface, a data buffer unit and a timestamp generation unit.

[0012] Preferably, the specific process of the data acquisition module converting the image data of the semiconductor sample acquired by the imaging module into a processable digital signal includes the following steps:

[0013] S21, the high-speed data acquisition card captures the image data generated by the imaging module in real time, and converts the acquired analog image signal into digital image data through the analog-to-digital conversion unit;

[0014] S22, the data encoding unit encodes the digital image data into one of the formats of JPEG, TIFF and RAW, and then quickly transmits the digital image data to the data processing module through a high-speed data interface;

[0015] S23, before transmitting the digital image data to the data processing module, temporarily storing the image data through the data buffer unit, and adding a timestamp to each image data packet through the timestamp generation unit.

[0016] Preferably, the data processing module applies a deep learning model to process the data from the data acquisition module and performs tasks including image preprocessing, feature extraction and defect identification.

[0017] Preferably, the construction process of the deep learning model includes the following steps:

[0018] S31. Dataset preparation and preprocessing: Collect no less than 1 million labeled semiconductor sample images, including various typical and atypical defects, as training, validation and test data sets, and normalize the images, adjust the pixel value range, ensure the consistency of the data set, implement a class balance strategy, and solve the class imbalance problem in the data set through undersampling methods;

[0019] S32, Model architecture design and optimization: Select an improved model deep learning architecture based on ResNet50, introduce an attention mechanism to enhance the model's learning ability for subtle defect features, and improve the model's computing efficiency and storage efficiency through pruning, quantization, and lightweight design techniques;

[0020] S33, loss function and optimizer selection: define weighted cross entropy loss as the loss function, and select Adam optimizer, combined with learning rate decay strategy, to speed up convergence and avoid overfitting;

[0021] S34, Model training and validation: Implement early stopping mechanism and model periodic saving strategy, monitor the performance on the validation set to prevent overfitting, and use K-fold cross validation to evaluate the robustness of the model to ensure the generalization ability of the model on unseen data, and use batch normalization and dropout layers to enhance the training stability and generalization ability of the model;

[0022] S35, Model evaluation and tuning: Evaluate model performance on an independent test set, including accuracy, recall, F1 score, and ROC curve, and identify model weaknesses based on confusion matrix analysis, adjust model architecture and training strategies in a targeted manner, and find the best model configuration through Bayesian optimization;

[0023] S36. Model deployment and continuous learning: Integrate the trained model into the data processing module to ensure efficient processing of real-time data streams, and implement a model hot update mechanism to allow the model to be updated without affecting the normal operation of the system. Design a feedback loop to collect actual detection results for continuous iterative optimization and online learning of the model.

[0024] Preferably, the automation and integration module includes a six-axis robotic arm, a precision turntable and industrial robot simulation software, and the user interface module is based on the Web, provides real-time monitoring functions and supports remote access.

[0025] Preferably, the operation process of the automation and integration module includes the following steps:

[0026] S41, initialization and calibration phase: start the automation and integration module, perform self-check and calibration of the six-axis robot arm and the precision turntable, confirm the initial position of the precision turntable through the position encoder, and perform zero point calibration;

[0027] S42, sample loading and positioning: Use a six-axis robotic arm to accurately grab the semiconductor sample to be tested from the sample storage area according to preset instructions, and then use the multi-degree-of-freedom operation of the robotic arm to accurately place the sample on the precision turntable;

[0028] S43, virtual simulation verification: On the industrial robot simulation software platform, input the current sample type and detection requirements, start the 3D virtual environment modeling, and then perform motion trajectory planning to simulate the grabbing, moving and placing actions of the robot arm, as well as the rotation operation of the turntable. Through simulation verification, the optimized path is ensured to ensure smooth operation and no collision risk. Then, the simulation data is analyzed to predict the detection efficiency and system stability.

[0029] S44, automated detection process: According to the optimized path planning, the precision turntable starts to rotate precisely at the preset angle, and cooperates with the robotic arm to move the sample to the bottom of different imaging modules. The imaging modules perform all-round imaging of the sample in turn, including optical microscopy, electron microscopy, X-ray diffraction and super-resolution microscopy detection.

[0030] A semiconductor sample defect detection method includes a semiconductor sample defect detection system, and further includes the following steps:

[0031] H1. Sample preparation: Use ultrapure water and nitrogen to purge to remove oil and dust on the surface of semiconductor samples;

[0032] H2, Automation and Integration: "Use industrial robot simulation software to simulate the operation of the six-axis robotic arm and optimize the path planning, and then automatically control the six-axis robotic arm of the simulation software to accurately grab the semiconductor sample and place it on the precision turntable;

[0033] H3. Imaging acquisition: Obtain macroscopic, microscopic and crystal structure images of semiconductor samples through optical microscope, electron microscope, X-ray diffractometer and super-resolution microscope;

[0034] H4, data acquisition: real-time image capture through high-speed data acquisition card, encoding after analog-to-digital conversion, and transmission to data processing module through high-speed interface;

[0035] H5. Image processing: Preprocess the image through deep learning model, extract features, identify and mark defects in the sample;

[0036] H6. Result analysis and reporting: Analyze the test results in detail, generate a report containing detailed information of the defects, provide treatment suggestions based on the severity of the defects, and display them through the user interface module.

[0037] (III) Beneficial effects

[0038] Compared with the prior art, the present invention provides a semiconductor sample defect detection system and method, which has the following beneficial effects:

[0039] 1. The semiconductor sample defect detection system and method achieves a sampling rate of 100MS / s by adopting a high-speed data acquisition card, which significantly improves the data capture speed. Combined with automation and integration modules, the system can process multiple samples continuously and quickly, greatly reducing the detection cycle and effectively avoiding the problem of slow detection speed becoming a bottleneck in a large-scale production environment. In addition, automated operation reduces human intervention and reduces production efficiency losses caused by operational delays, ensuring a smooth and efficient production process.

[0040] 2. The semiconductor sample defect detection system and method, by integrating the imaging modules of optical microscope, electron microscope, X-ray diffractometer and super-resolution microscope, can comprehensively capture the details of semiconductor samples from macro to micro, from surface to crystal structure, ensuring the accurate identification of subtle or deep defects. The application of deep learning model further improves the accuracy of image processing. Through feature extraction and intelligent recognition, even defects that are difficult to detect with the naked eye can be found. This high-precision detection capability greatly reduces the risk of quality problems being discovered in subsequent production links, reduces the possibility of rework and scrapping, and thus improves product quality and production yield.

[0041] 3. The semiconductor sample defect detection system and method, by integrating industrial robot simulation software, realizes the simulation and optimization of robot arm operation, enhances the flexibility and adaptability of the system, and allows users to quickly adjust detection parameters and path planning according to different production conditions and sample characteristics, ensuring that the system can maintain efficient and stable operation in a changing production environment. In addition, the system's Web-based user interface supports remote access and monitoring, allowing operators to flexibly manage and control the detection process from different locations, further improving the system's applicability and ease of operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 It is a schematic flow chart of a defect detection method for semiconductor samples of the present invention;

[0043] Figure 2 A schematic diagram of a process for obtaining a high-resolution image of a semiconductor sample by the imaging module of the present invention;

[0044] Figure 3 It is a schematic diagram of the data conversion process of the data acquisition module of the present invention;

[0045] Figure 4 A schematic diagram of the construction process of the deep learning model of the present invention;

[0046] Figure 5 It is a schematic diagram of the operation flow of the automation and integration module of the present invention. DETAILED DESCRIPTION

[0047] The following will be combined with the embodiments of the present invention and the accompanying drawings to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0048] See also Figure 1-5 , a defect detection system for semiconductor samples, includes an imaging module, a data acquisition module, a data processing module, an automation and integration module and a user interface module.

[0049] Furthermore, the imaging module is responsible for obtaining high-resolution images of semiconductor samples, which includes four submodules: optical microscope, electron microscope, X-ray diffractometer and super-resolution microscope. The process of the imaging module obtaining high-resolution images of semiconductor samples includes the following steps:

[0050] S11. Sample preparation and positioning: First, ultrapure water and nitrogen are used to purge to remove oil and dust on the surface of the semiconductor sample. Then, fluorescent dyes are used to enhance the visibility of specific structures of the sample.

[0051] S12. Application of imaging technology: Optical microscopes obtain macroscopic and microscopic structural images of semiconductor samples; electron microscopes obtain more detailed microscopic structural analysis images of semiconductor samples; X-ray diffractometers obtain crystal structure images of semiconductor samples; super-resolution microscopes can achieve single-molecule resolution imaging and capture the finest detail images of semiconductor samples.

[0052] Specifically, the imaging module realizes comprehensive high-resolution imaging of semiconductor samples from macro to micro, from surface to crystal structure. This multimodal imaging technology not only improves the accuracy and resolution of detection, but also enhances the visibility of specific structures through sample pretreatment, such as the use of ultrapure water and nitrogen cleaning and fluorescent dye labeling. In particular, the application of super-resolution microscopy achieves single-molecule resolution, making even the smallest defects nowhere to hide. The automated design of the entire imaging process reduces human errors, ensures high efficiency and repeatability of detection, and provides strong technical support for defect detection of semiconductor samples.

[0053] Furthermore, the data acquisition module includes a high-speed data acquisition card with a sampling rate of 100MS / s, an analog-to-digital conversion unit, a data encoding unit, a high-speed data interface, a data buffer unit and a timestamp generation unit.

[0054] Furthermore, the specific process of the data acquisition module converting the image data of the semiconductor sample acquired by the imaging module into a processable digital signal includes the following steps:

[0055] S21, the high-speed data acquisition card captures the image data generated by the imaging module in real time, and converts the acquired analog image signal into digital image data through the analog-to-digital conversion unit;

[0056] S22, the data encoding unit encodes the digital image data into one of the formats of JPEG, TIFF and RAW, and then quickly transmits the digital image data to the data processing module through a high-speed data interface;

[0057] S23, before transmitting the digital image data to the data processing module, temporarily storing the image data through the data buffer unit, and adding a timestamp to each image data packet through the timestamp generation unit.

[0058] Specifically, the 100MS / s high-speed data acquisition card ensures real-time capture and high-precision conversion of image data, significantly improving the speed and efficiency of data processing. The addition of the analog-to-digital conversion unit enables analog signals to be quickly converted into digital signals, laying the foundation for subsequent processing. The data encoding unit supports encoding in multiple formats, such as JPEG, TI FF, and RAW, which increases data compatibility and application flexibility. The combination of high-speed data interface and data buffer unit not only speeds up data transmission, but also provides temporary storage to cope with fluctuations in data flow. The introduction of the timestamp generation unit adds time information to the data packet, which helps data synchronization and subsequent analysis.

[0059] Furthermore, the data processing module applies a deep learning model to process the data from the data acquisition module and perform tasks including image preprocessing, feature extraction, and defect identification.

[0060] Furthermore, the deep learning model construction process includes the following steps:

[0061] S31. Dataset preparation and preprocessing: Collect no less than 1 million labeled semiconductor sample images, including various typical and atypical defects, as training, validation and test data sets, and normalize the images, adjust the pixel value range, ensure the consistency of the data set, implement a class balance strategy, and solve the class imbalance problem in the data set through undersampling methods;

[0062] S32, Model architecture design and optimization: Select an improved model deep learning architecture based on ResNet50, introduce an attention mechanism to enhance the model's learning ability for subtle defect features, and improve the model's computing efficiency and storage efficiency through pruning, quantization, and lightweight design techniques;

[0063] S33, loss function and optimizer selection: define weighted cross entropy loss as the loss function, and select Adam optimizer, combined with learning rate decay strategy, to speed up convergence and avoid overfitting;

[0064] S34, Model training and validation: Implement early stopping mechanism and model periodic saving strategy, monitor the performance on the validation set to prevent overfitting, and use K-fold cross validation to evaluate the robustness of the model to ensure the generalization ability of the model on unseen data, and use batch normalization and dropout layers to enhance the training stability and generalization ability of the model;

[0065] S35, Model evaluation and tuning: Evaluate model performance on an independent test set, including accuracy, recall, F1 score, and ROC curve, and identify model weaknesses based on confusion matrix analysis, adjust model architecture and training strategies in a targeted manner, and find the best model configuration through Bayesian optimization;

[0066] S36. Model deployment and continuous learning: Integrate the trained model into the data processing module to ensure efficient processing of real-time data streams, and implement a model hot update mechanism to allow the model to be updated without affecting the normal operation of the system. Design a feedback loop to collect actual detection results for continuous iterative optimization and online learning of the model.

[0067] Specifically, the application of deep learning models has significantly improved the intelligence and automation level of image processing. The model construction process forms a complete closed loop from data set preparation to preprocessing, architecture design, optimization, training, verification, evaluation, tuning, and deployment and continuous learning. This process not only ensures the model's high recognition ability for defects in semiconductor sample images, but also improves the model's computing and storage efficiency by introducing attention mechanisms and optimization techniques. The selection of loss functions and optimizers, as well as the application of early stopping mechanisms and K-fold cross-validation effectively prevent overfitting and enhance the generalization ability of the model. The use of Bayesian optimization further optimizes the model configuration. In addition, the model's hot update and online learning mechanisms ensure that the system can continuously adapt to new data and requirements and achieve continuous performance improvement.

[0068] Furthermore, the automation and integration module includes a six-axis robotic arm, a precision turntable and industrial robot simulation software, and the user interface module is web-based, providing real-time monitoring capabilities and supporting remote access.

[0069] Furthermore, the operation process of the automation and integration module includes the following steps:

[0070] S41, initialization and calibration phase: start the automation and integration module, perform self-check and calibration of the six-axis robot arm and the precision turntable, confirm the initial position of the precision turntable through the position encoder, and perform zero point calibration;

[0071] S42, sample loading and positioning: Use a six-axis robotic arm to accurately grab the semiconductor sample to be tested from the sample storage area according to preset instructions, and then use the multi-degree-of-freedom operation of the robotic arm to accurately place the sample on the precision turntable;

[0072] S43, virtual simulation verification: On the industrial robot simulation software platform, input the current sample type and detection requirements, start the 3D virtual environment modeling, and then perform motion trajectory planning to simulate the grabbing, moving and placing actions of the robot arm, as well as the rotation operation of the turntable. Through simulation verification, the optimized path is ensured to ensure smooth operation and no collision risk. Then, the simulation data is analyzed to predict the detection efficiency and system stability.

[0073] S44, automated detection process: According to the optimized path planning, the precision turntable starts to rotate precisely at the preset angle, and cooperates with the robotic arm to move the sample to the bottom of different imaging modules. The imaging modules perform all-round imaging of the sample in turn, including optical microscopy, electron microscopy, X-ray diffraction and super-resolution microscopy detection.

[0074] Specifically, the multi-degree-of-freedom operation of the six-axis robotic arm ensures the accurate grasping and placement of samples, while the calibration and precise rotation of the precision turntable provide the necessary conditions for all-round imaging. The use of industrial robot simulation software optimizes the operating path of the robotic arm through virtual environment modeling and motion trajectory planning, reduces the risk of collision, and improves detection efficiency and system stability. In addition, the Web-based design of the user interface module supports real-time monitoring and remote access, enhancing the flexibility of the system and the convenience of operation.

[0075] A semiconductor sample defect detection method includes a semiconductor sample defect detection system, and further includes the following steps:

[0076] H1. Sample preparation: Use ultrapure water and nitrogen to purge to remove oil and dust on the surface of semiconductor samples;

[0077] H2, Automation and Integration: "Use industrial robot simulation software to simulate the operation of the six-axis robotic arm and optimize the path planning, and then automatically control the six-axis robotic arm of the simulation software to accurately grab the semiconductor sample and place it on the precision turntable;

[0078] H3. Imaging acquisition: Obtain macroscopic, microscopic and crystal structure images of semiconductor samples through optical microscope, electron microscope, X-ray diffractometer and super-resolution microscope;

[0079] H4, data acquisition: real-time image capture through high-speed data acquisition card, encoding after analog-to-digital conversion, and transmission to data processing module through high-speed interface;

[0080] H5. Image processing: Preprocess the image through deep learning model, extract features, identify and mark defects in the sample;

[0081] H6. Result analysis and reporting: Analyze the test results in detail, generate a report containing detailed information of the defects, provide treatment suggestions based on the severity of the defects, and display them through the user interface module.

[0082] In summary, the semiconductor sample defect detection system and method, by adopting a high-speed data acquisition card, achieves a sampling rate of 100MS / s, significantly improving the data capture speed. Combined with automation and integration modules, the system can process multiple samples continuously and quickly, greatly reducing the detection cycle and effectively avoiding the problem of slow detection speed becoming a bottleneck in a large-scale production environment. In addition, automated operation reduces human intervention and reduces production efficiency losses caused by operational delays, ensuring a smooth and efficient production process.

[0083] In addition, the defect detection system and method for semiconductor samples, by integrating the imaging modules of optical microscopes, electron microscopes, X-ray diffractometers and super-resolution microscopes, can comprehensively capture the details of semiconductor samples from macro to micro, from surface to crystal structure, ensuring the accurate identification of subtle or deep defects. The application of deep learning models further improves the accuracy of image processing. Through feature extraction and intelligent recognition, even defects that are difficult to detect with the naked eye can be found. This high-precision detection capability greatly reduces the risk of quality problems being discovered in subsequent production links, reduces the possibility of rework and scrapping, and thus improves product quality and production yield.

[0084] Furthermore, the semiconductor sample defect detection system and method, by integrating industrial robot simulation software, realizes the simulation and optimization of robot arm operation, enhances the flexibility and adaptability of the system, and users can quickly adjust the detection parameters and path planning according to different production conditions and sample characteristics, ensuring that the system can maintain efficient and stable operation in a changing production environment. In addition, the system's Web-based user interface supports remote access and monitoring, allowing operators to flexibly manage and control the detection process from different locations, further improving the applicability of the system and the ease of operation, and solving the problems that, first, the detection speed of this technology is relatively slow, which may become a bottleneck in a large-scale production environment and affect the overall production efficiency; secondly, the accuracy needs to be improved, and some subtle or deep defects may not be effectively detected, resulting in quality problems being discovered in subsequent production links, increasing the risk of rework and scrapping.

[0085] The relevant modules involved in this system are all hardware system modules or functional modules that combine computer software programs or protocols with hardware in the prior art. The computer software programs or protocols involved in the functional modules are themselves technologies that are well known to those skilled in the art and are not improvements of this system. The improvements of this system are the interaction or connection relationships between the modules, that is, improvements to the overall structure of the system to solve the corresponding technical problems to be solved by this system.

[0086] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A semiconductor sample defect detection system, characterized in that: It includes an imaging module, a data acquisition module, a data processing module, an automation and integration module and a user interface module.

2. A semiconductor sample defect detection system according to claim 1, characterized in that: The imaging module is responsible for obtaining a high-resolution image of a semiconductor sample, and includes four submodules: an optical microscope, an electron microscope, an X-ray diffractometer, and a super-resolution microscope. The process of obtaining a high-resolution image of a semiconductor sample by the imaging module includes the following steps: S11. Sample preparation and positioning: First, ultrapure water and nitrogen are used to purge to remove oil and dust on the surface of the semiconductor sample. Then, fluorescent dyes are used to enhance the visibility of specific structures of the sample. S12. Application of imaging technology: Optical microscopes obtain macroscopic and microscopic structural images of semiconductor samples; electron microscopes obtain more detailed microscopic structural analysis images of semiconductor samples; X-ray diffractometers obtain crystal structure images of semiconductor samples; super-resolution microscopes can achieve single-molecule resolution imaging and capture the finest detail images of semiconductor samples.

3. A semiconductor sample defect detection system according to claim 1, characterized in that: The data acquisition module includes a high-speed data acquisition card with a sampling rate of 100MS / s, an analog-to-digital conversion unit, a data encoding unit, a high-speed data interface, a data buffer unit and a time stamp generation unit.

4. A semiconductor sample defect detection system according to claim 3, characterized in that: The specific process of the data acquisition module converting the image data of the semiconductor sample acquired by the imaging module into a processable digital signal includes the following steps: S21, the high-speed data acquisition card captures the image data generated by the imaging module in real time, and converts the acquired analog image signal into digital image data through the analog-to-digital conversion unit; S22, the data encoding unit encodes the digital image data into one of the formats of JPEG, TIFF and RAW, and then quickly transmits the digital image data to the data processing module through a high-speed data interface; S23, before transmitting the digital image data to the data processing module, temporarily storing the image data through the data buffer unit, and adding a timestamp to each image data packet through the timestamp generation unit.

5. The semiconductor sample defect detection system according to claim 1, characterized in that: The data processing module applies a deep learning model to process the data from the data acquisition module and performs tasks including image preprocessing, feature extraction and defect identification.

6. A semiconductor sample defect detection system according to claim 5, characterized in that: The construction process of the deep learning model includes the following steps: S31. Dataset preparation and preprocessing: Collect no less than 1 million labeled semiconductor sample images, including various typical and atypical defects, as training, validation and test data sets, and normalize the images, adjust the pixel value range, ensure the consistency of the data set, implement a class balance strategy, and solve the class imbalance problem in the data set through undersampling methods; S32, Model architecture design and optimization: Select an improved model deep learning architecture based on ResNet50, introduce an attention mechanism to enhance the model's learning ability for subtle defect features, and improve the model's computing efficiency and storage efficiency through pruning, quantization, and lightweight design techniques; S33, loss function and optimizer selection: define weighted cross entropy loss as the loss function, and select Adam optimizer, combined with learning rate decay strategy, to speed up convergence and avoid overfitting; S34, Model training and validation: Implement early stopping mechanism and model periodic saving strategy, monitor the performance on the validation set to prevent overfitting, and use K-fold cross validation to evaluate the robustness of the model to ensure the generalization ability of the model on unseen data, and use batch normalization and dropout layers to enhance the training stability and generalization ability of the model; S35, Model evaluation and tuning: Evaluate model performance on an independent test set, including accuracy, recall, F1 score, and ROC curve, and identify model weaknesses based on confusion matrix analysis, adjust model architecture and training strategies in a targeted manner, and find the best model configuration through Bayesian optimization; S36. Model deployment and continuous learning: Integrate the trained model into the data processing module to ensure efficient processing of real-time data streams, and implement a model hot update mechanism to allow the model to be updated without affecting the normal operation of the system. Design a feedback loop to collect actual detection results for continuous iterative optimization and online learning of the model.

7. A semiconductor sample defect detection system according to claim 1, characterized in that: The automation and integration module includes a six-axis robotic arm, a precision turntable and industrial robot simulation software, and the user interface module is based on the Web, provides real-time monitoring functions and supports remote access.

8. A semiconductor sample defect detection system according to claim 7, characterized in that: The operation process of the automation and integration module includes the following steps: S41, initialization and calibration phase: start the automation and integration module, perform self-check and calibration of the six-axis robot arm and the precision turntable, confirm the initial position of the precision turntable through the position encoder, and perform zero point calibration; S42, sample loading and positioning: Use a six-axis robotic arm to accurately grab the semiconductor sample to be tested from the sample storage area according to preset instructions, and then use the multi-degree-of-freedom operation of the robotic arm to accurately place the sample on the precision turntable; S43, virtual simulation verification: On the industrial robot simulation software platform, input the current sample type and detection requirements, start the 3D virtual environment modeling, and then perform motion trajectory planning to simulate the grabbing, moving and placing actions of the robot arm, as well as the rotation operation of the turntable. Through simulation verification, the optimized path is ensured to ensure smooth operation and no collision risk. Then, the simulation data is analyzed to predict the detection efficiency and system stability. S44, automated detection process: According to the optimized path planning, the precision turntable starts to rotate precisely at the preset angle, and cooperates with the robotic arm to move the sample to the bottom of different imaging modules. The imaging modules perform all-round imaging of the sample in turn, including optical microscopy, electron microscopy, X-ray diffraction and super-resolution microscopy detection.

9. A method for defect detection of a semiconductor sample, characterized in that: A defect detection system for a semiconductor sample according to claims 1 to 8, further comprising the following steps: H1. Sample preparation: Use ultrapure water and nitrogen to purge to remove oil and dust on the surface of semiconductor samples; H2, Automation and Integration: "Use industrial robot simulation software to simulate the operation of the six-axis robotic arm and optimize the path planning, and then automatically control the six-axis robotic arm of the simulation software to accurately grab the semiconductor sample and place it on the precision turntable; H3. Imaging acquisition: Obtain macroscopic, microscopic and crystal structure images of semiconductor samples through optical microscope, electron microscope, X-ray diffractometer and super-resolution microscope; H4, data acquisition: real-time image capture through high-speed data acquisition card, encoding after analog-to-digital conversion, and transmission to data processing module through high-speed interface; H5. Image processing: Preprocess the image through deep learning model, extract features, identify and mark defects in the sample; H6. Result analysis and reporting: Analyze the test results in detail, generate a report containing detailed information of the defects, provide treatment suggestions based on the severity of the defects, and display them through the user interface module.

Images