System for automatic analysis of sem images based on full database and process defect detection

Abstract

The application provides an automatic analysis and process defect detection system based on full-database SEM images, and relates to the technical field of semiconductor manufacturing, wherein the system realizes accurate classification of defect types and automatic identification of new defects through dynamic feature fusion (Sp1) of multi-scale SEM images and multi-channel acquisition devices, combined with adaptive feature extraction and density anomaly detection of a high-performance computing unit, dynamically adjusts the feature range and fusion weight, breaks through the limitations of traditional static detection, supports adaptive analysis of process context, improves the detection accuracy to 98%, reduces the misjudgment and missed judgment rate by 30%-50%, especially in advanced processes such as 3nm, can quickly identify new defects and update rules, significantly improves the process research and development efficiency and product reliability, and provides flexible and intelligent detection capability for semiconductor manufacturing.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of semiconductor manufacturing, specifically to an automatic analysis and process defect detection system for SEM images based on a full database. BACKGROUND

[0002] According to the Chinese patent No. "CN118571794B", a semiconductor automatic cleaning monitoring system and method based on image processing, belonging to the field of image processing, the number of surface transistors in the cleaning sub-area is obtained, and the cleaning sub-area importance evaluation strategy is introduced to evaluate the importance of the cleaning sub-area. The image abnormality evaluation result and the cleaning sub-area importance evaluation result are obtained to evaluate the degree of semiconductor contamination. According to the degree of semiconductor contamination, the automatic cleaning time is judged. According to the finished product image of the semiconductor wafer and the number of surface transistors, the degree of contamination of the semiconductor wafer is comprehensively judged, and then the cleaning time of the cleaning equipment is comprehensively and accurately analyzed, which improves the cleaning efficiency and avoids damage to the semiconductor wafer caused by cleaning.

[0003] The above patent document and prior art have the following technical problems when in use:

[0004] Problem one, in the existing semiconductor manufacturing, defect detection depends on fixed feature extraction and pre-defined classification model, which cannot adapt to the diversity of multi-scale SEM images, and has insufficient recognition ability for unknown defects in new processes, resulting in high misjudgment and omission rate, especially in advanced node processes such as 3nm, new defects occur frequently, which seriously affects the research and development efficiency and yield;

[0005] Problem two, in the prior art, defect root cause analysis is mostly offline statistics, which is disconnected with process adjustment, has long response time, cannot intervene in high-risk process steps in time, and leads to large yield fluctuation, especially in high-yield lines, delayed adjustment increases the scrap rate and cost. SUMMARY

[0006] Technical problems solved

[0007] In view of the deficiencies of the prior art, the present application provides an automatic analysis and process defect detection system for SEM images based on a full database, which solves the following problems:

[0008] 1. The traditional defect detection method lacks dynamic adaptability and new defect recognition ability;

[0009] 2. The defect root cause analysis and process optimization lack real-time and closed-loop control.

[0010] Technical scheme

[0011] To achieve the above object, the present application is realized by the following technical solutions: the automatic analysis and process defect detection system of SEM image based on full database, the automatic analysis and process defect detection system comprises the following steps:

[0012] Sp1: adaptive extraction and fusion of features from multi-scale scanning electron microscope (SEM) images to detect and classify defect types and support automatic identification of new defect types;

[0013] Sp2: based on the time-dependent features of defect type, process parameter and equipment state, dynamic analysis of the causal relationship between process steps and defects, and separation of the interaction effect between process steps;

[0014] Sp3: according to the causal relationship and multi-source data, predict the defect trend, cooperatively generate process parameter adjustment suggestion and equipment maintenance plan, and optimize the prediction and suggestion through feedback;

[0015] Sp4: build and update an interactive knowledge graph containing defect type, process step and equipment state in real time, support users to query defect root cause through path reasoning;

[0016] Sp5: seamlessly integrate the above analysis results with the semiconductor manufacturing process to realize real-time closed-loop control from defect detection to process optimization during semiconductor manufacturing.

[0017] Preferably, in the step Sp1, when adaptive extraction and fusion of features are performed, the feature extraction range and fusion weight are dynamically adjusted to generate comprehensive features suitable for different magnifications and process contexts, and the fusion weight is calculated based on image local statistical characteristics and process step type in real time.

[0018] Preferably, in the step Sp2, when the causal relationship between process steps and defects is dynamically analyzed, the historical data is encoded by time decay mechanism to generate time-varying feature vector, and the causal contribution of process steps to defects and the interaction effect intensity are calculated based on the time-varying feature vector.

[0019] Preferably, in the step Sp3, when the defect trend is predicted, the SEM image features, process parameters and equipment states are decomposed into long-term trend and periodic fluctuation, the future defect occurrence probability is predicted based on the decomposition result, and the priority of the cooperatively generated suggestion is adjusted according to the prediction probability.

[0020] Preferably, in the step Sp4, when the interactive knowledge graph is generated, multi-layer path search and confidence evaluation are supported, the optimal causal path from the defect type to the process step is dynamically generated, the graph structure and the correlation strength are incrementally updated according to new data, and the interactive knowledge graph further includes an abnormal pattern recognition function, which identifies abnormal process patterns by analyzing the correlation strength changes of the defect type and the process step in the graph, and automatically triggers process adjustment suggestions or equipment inspection instructions.

[0021] Preferably, the automatic analysis and process defect detection system includes a data alignment unit for aligning multi-scale SEM images with process parameters and equipment state data in time and space dimensions, and embedding process context identifiers in the aligned data to enhance the context relevance of the analysis.

[0022] Preferably, in the step Sp4, the automatic identification of new defect types is achieved through density anomaly detection in the feature space, and the identified new defect types are associated with the process parameter change trend to generate preliminary cause hypotheses of the new defects, and the cause hypotheses are automatically updated after being verified by comparing with historical data.

[0023] Preferably, in the step Sp2, when the dynamic analysis of the causal relationship between the process step and the defect is performed, the influence of a small perturbation of the process parameter on the defect probability is quantified, and a priority adjustment sequence of the process step is generated based on the quantified results to minimize the defect occurrence rate.

[0024] Preferably, in the step Sp3, the collaborative generation of process parameter adjustment suggestions and equipment maintenance plans includes: constructing a multi-objective optimization model based on the defect trend prediction results and the equipment operating state, simultaneously optimizing the defect rate reduction and maintenance cost control, and dynamically adjusting the model weights to adapt to changes in production requirements during the optimization process.

[0025] Preferably, the hardware components of the automatic analysis and process defect detection system include:

[0026] A multi-channel SEM image acquisition device is configured with multiple magnification lenses and a real-time data transmission interface, which is used to acquire multi-scale SEM images and transmit them to the database;

[0027] A high-performance computing unit integrates multi-core GPU and special-purpose acceleration chips to support real-time computing of adaptive feature fusion, time-varying causal analysis, and defect trend prediction;

[0028] A process parameter and equipment state monitor includes a sensor array and a time synchronization module, which is used to acquire process parameters and equipment operating state in real time and store them in alignment with SEM image data;

[0029] An interactive display terminal is equipped with a high-resolution touch screen and a map rendering engine for presenting the interactive knowledge graph and supporting user path reasoning operations.

[0030] An adaptive storage server is configured with a hierarchical cache structure and a dynamic partition module for dynamically allocating storage space according to SEM image data volume and process analysis requirements, and supporting fast retrieval and update.

[0031] A closed-loop control interface unit integrates a programmable logic controller (PLC) and a process equipment communication protocol for real-time transmission of the process parameter adjustment suggestions and equipment maintenance plans to semiconductor manufacturing equipment, and receiving execution feedback signals to optimize subsequent analysis.

[0032] Advantages

[0033] The present application provides an automatic analysis and process defect detection system based on full-database SEM images.

[0034] Advantages

[0035] 1. The system uses dynamic feature fusion (Sp1) of multi-scale SEM images and a multi-channel acquisition device, combined with adaptive feature extraction and density anomaly detection of a high-performance computing unit, to achieve accurate classification of defect types and automatic identification of new defects, dynamically adjust feature range and fusion weight, break through the limitations of traditional static detection, support adaptive analysis of process context, improve detection accuracy to 98%, reduce misjudgment and missed judgment rate by 30%-50%, especially in advanced processes such as 3nm, can quickly identify new defects and update rules, significantly improve process development efficiency and product reliability, and provide flexible and intelligent detection capabilities for semiconductor manufacturing.

[0036] 2. The system integrates time-varying causal analysis (Sp2), defect trend prediction (Sp3), and real-time control (Sp5), relying on monitors, storage servers, and closed-loop control interface units, forming a closed-loop system from root cause analysis to process optimization, disturbance simulation and multi-objective optimization, real-time adjustment of process parameters and feedback optimization, surpassing traditional offline analysis, response speed up to seconds, yield improvement of 5%-10%, for example, rapidly reducing defect rate from 5% to 2%, reducing production cost, ensuring manufacturing stability, and providing efficient control for high-yield lines. BRIEF DESCRIPTION OF DRAWINGS

[0037] Fig. 1 System operation steps of the present application;

[0038] Fig. 2 System hardware architecture diagram of the present application. DETAILED DESCRIPTION

[0039] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Specific Implementation Example 1:

[0041] like Figs. 1-2 As shown, the automatic analysis and process defect detection system based on a full database of SEM images includes the following steps:

[0042] Sp1: Adaptive Feature Extraction and Defect Detection Classification: The system first adaptively extracts and fuses features from multi-scale scanning electron microscope (SEM) images to detect and classify defect types and supports automatic identification of new defect types. It begins by acquiring images at different magnifications using a multi-channel SEM image acquisition device (e.g., low magnification for macroscopic defect detection, high magnification for nanoscale defect analysis). These images are then transferred to a full database for storage. Next, the system dynamically adjusts the feature extraction range and fusion weights to generate comprehensive features suitable for different magnifications and process contexts. Specifically, the feature extraction range is dynamically adjusted based on the local statistical properties of the image (e.g., grayscale variance, entropy value). For example, the range is narrowed in high-detail areas to capture minute defects, and expanded in smooth areas to identify large-scale anomalies. The fusion weights are based on the local statistical properties of the image and the type of process step (e.g., photolithography or etching). The system calculates multi-scale features into a comprehensive feature vector through a weighting mechanism (e.g., weighted averaging based on process context). Based on this vector, the system classifies defects (e.g., bridging, fracture, particle deposition) and identifies new defect types through density anomaly detection in the feature space. For example, if a feature cluster aggregates outside a predefined category and its density exceeds the normal range, the system marks it as a new defect and correlates it with the trend of process parameter changes (e.g., temperature fluctuations) to generate preliminary causal hypotheses (e.g., "high temperature causes particle deposition"). These hypotheses are verified by comparison with historical data, and the defect classification rules are automatically updated to ensure system adaptability and efficient conversion from raw SEM images to defect types. The process reduces misjudgments through adaptability and dynamism, while providing accurate data for subsequent analysis. The workflow consists of image acquisition, feature extraction and fusion, defect classification and new type identification, and rule updating.

[0043] Sp2: Dynamic process causal relationship analysis: After obtaining the defect type, the system dynamically analyzes the causal relationship between the process steps and the defects based on the time-dependent characteristics of the defect type, process parameters, and equipment state, and separates the interaction effects between process steps. Starting from the data alignment unit, the multi-scale SEM images are aligned with the process parameters (such as temperature, pressure) and equipment state (such as vacuum degree, voltage stability) in time and space dimensions, and the process context identifier (such as "etching after lithography") is embedded in the aligned data to enhance the context relevance of the analysis. The aligned data enters the analysis stage, and the system encodes the historical data through a time decay mechanism to generate time-varying feature vectors. Specifically, recent data is given a higher weight to reflect the immediate impact of process changes, while long-term data provides a long-term trend reference. Based on these time-varying feature vectors, the system calculates the causal contribution of process steps to defects and the interaction effect strength. For example, through covariance analysis, it determines the direct impact of a step (such as etching) on bridge defects, and through tensor decomposition, it separates the combined effect of lithography and etching interaction; In addition, the system quantifies the impact of a small perturbation in process parameters (such as increasing the etching temperature by 0.5°C) on the probability of defects (such as the probability rising from 5% to 7%), and generates a priority adjustment sequence for process steps (such as "adjust the etching temperature first") based on the quantification results to minimize the defect occurrence rate. Based on quantitative data, ensure the pertinence of adjustment. The function of step Sp2 is to reveal the dynamic causal chain between defects and processes. The process provides accurate root cause positioning through time-dependent analysis and perturbation simulation. The workflow is data alignment, time-varying feature generation, causal and interaction analysis, perturbation quantification and adjustment sequence generation.

[0044] Sp3: Defect trend prediction and optimization suggestion generation: Based on the causal relationship results of Sp2, the system predicts defect trends based on multi-source data, collaboratively generates process parameter adjustment suggestions and equipment maintenance plans, and optimizes prediction and suggestions through feedback. Starting from the integration of multi-source data, including SEM image features, process parameters and equipment status, the system decomposes these data into long-term trends (such as slow rise of defect rate over time) and periodic fluctuations (such as fluctuations caused by equipment aging every week), which is achieved through dynamic frequency analysis (such as adaptive Fourier decomposition based on process cycle). Based on the decomposition results, the system predicts the probability of defect occurrence in the future period (such as the next week), for example, predicting that the probability of a certain defect type will rise from 3% to 5%. According to the predicted probability, the system adjusts the priority of the suggestions generated collaboratively, and the suggestions corresponding to high-probability defects are processed first; the collaborative generation process is achieved by constructing a multi-objective optimization model, which optimizes both defect rate reduction (such as reducing the defect rate to below 2%) and maintenance cost control (such as limiting maintenance frequency), and dynamically adjusts the model weight in optimization to adapt to changes in production demand (for example, high peak period prioritizes defect rate, low valley period considers cost), specific suggestions include process parameter adjustment (such as "reduce photolithography exposure time by 5%") and equipment maintenance plan (such as "clean etching cabin"), these suggestions are transmitted to manufacturing equipment through closed-loop control interface unit for execution, the feedback signal (such as defect rate change) received by the system is used to optimize the prediction model and suggestion generation logic, for example, if a certain adjustment does not achieve the expected effect, the system will adjust the prediction weight or optimization target, predict defects and provide optimization scheme, the process ensures the scientificity and practicality of the suggestions through decomposition prediction and multi-objective optimization, the workflow is data decomposition, trend prediction, optimization model construction, suggestion generation and execution, feedback optimization.

[0045] Sp4: Interactive knowledge graph construction and updating: The system then constructs and updates an interactive knowledge graph containing defect types, process steps, and equipment states in real time, supporting users in querying defect root causes through path reasoning. With the analysis results of Sp1-Sp3 as input, the system takes defect types (such as bridging), process steps (such as etching), and equipment states (such as equipment temperature) as nodes, and constructs a multi-layer correlation graph based on causal contribution and interaction strength. The graph supports the dynamic generation of optimal causal paths through multi-layer path search and confidence assessment, such as the path from "bridge defects" to "etching temperature too high", with the confidence quantified by correlation strength (such as 0.9). The graph structure and correlation strength are incrementally updated according to new data (such as newly added defects or process adjustments), avoiding global recalculation to improve efficiency. In addition, the graph has an abnormal pattern recognition function, which identifies abnormal process patterns (such as device aging) by analyzing the correlation strength changes between defect types and process steps (such as a sudden increase in the correlation strength of a certain step to 0.95), and automatically triggers process adjustment suggestions (such as "reduce etching power") or equipment inspection instructions (such as "check the vacuum pump"). Users can operate the graph through an interactive display terminal, such as zooming in on a certain defect node to view related process paths, or filtering high-confidence paths for decision-making. The function provides intuitive root cause queries and abnormal early warnings, and the process realizes knowledge visualization and proactivity through dynamic construction and abnormal identification. The workflow is data input, graph construction, path reasoning and updating, abnormal identification and triggering.

[0046] Sp5: Real-time closed-loop control and integration with semiconductor manufacturing: Finally, the system seamlessly integrates the above analysis results with the semiconductor manufacturing process for real-time closed-loop control from defect detection to process optimization. Relying on the closed-loop control interface unit, the process parameter adjustment suggestions and maintenance plans of Sp3 are transmitted to manufacturing equipment (such as lithography machines, etching machines) in real time through programmable logic controllers (PLCs) and communication protocols, while the abnormal triggering instructions of Sp4 are sent to the equipment management system. After the manufacturing equipment executes, the feedback data (such as the adjusted defect rate, equipment state) are fed back to the system through the interface unit, updating the entire database and optimizing subsequent analysis. For example, if a suggestion reduces the defect rate to the target value, the system records the successful case to strengthen the prediction model. If it does not meet expectations, adjust the causal analysis or optimize the weights. The system supports this process through an adaptive storage server, with a hierarchical cache structure and dynamic partitioning module that allocates storage space according to data volume and analysis requirements, ensuring fast retrieval and real-time updating, and realizing seamless integration of analysis and manufacturing. The process forms a closed loop through real-time transmission and feedback, and the workflow is result transmission, device execution, feedback collection, and system optimization. Ultimately, it realizes continuous control from defect detection to process improvement in semiconductor manufacturing.

[0047] The operation scheme of the system is based on a full database, and a closed-loop process from SEM image analysis to process optimization is formed through five steps: Sp1 extracts features from multi-scale images and detects defects, laying the foundation for subsequent analysis; Sp2 locates the root cause of the defect through time-varying causal analysis, providing a direction for process improvement; Sp3 predicts trends and generates optimization suggestions to actively prevent defects; Sp4 builds a knowledge graph to support query and abnormality warning; Sp5 integrates the results into the manufacturing process to realize real-time control; the entire workflow is as follows: image acquisition and data alignment, feature extraction and defect classification, causal analysis and root cause positioning, trend prediction and optimization suggestion, graph construction and abnormal triggering, manufacturing execution and feedback optimization, through dynamic (feature fusion, time-varying analysis), synergy (prediction and optimization) and closed-loop (real-time control and feedback) to realize efficient defect management, which significantly reduces manual intervention compared to traditional methods and improves the yield and efficiency of semiconductor manufacturing. Embodiment Two:

[0049] As shown in Figs. 1-2 According to the content in the above embodiments, the following content is further disclosed:

[0050] The hardware composition of the automatic analysis and process defect detection system includes:

[0051] Multi-channel SEM image acquisition device: The multi-channel SEM image acquisition device is configured with multiple magnification lenses and a real-time data transmission interface, used for acquiring multi-scale SEM images and transmitting them to the full database, and is the starting point of the system operation, directly supporting step Sp1 (adaptive feature extraction and defect detection classification), and its operation logic is based on multi-channel parallel acquisition, which simultaneously acquires multi-scale images through different magnification lenses (such as low magnification lenses to capture wafer macro defects and high magnification lenses to focus on nanoscale details), and relies on electron beam scanning and signal detection technology to convert the sample surface topography into digital images. The operation process starts with SEM scanning of the wafer in the manufacturing process. The device acquires images in channels according to the preset magnification (such as 500x and 5000x), and transmits the image stream to the full database through the real-time data transmission interface (such as high-speed Ethernet or optical fiber), ensuring that Sp1 can obtain original data in real time. For example, when detecting wafer surface bridging defects, low magnification images provide overall distribution, and high magnification images reveal specific morphology. The multi-scale images collected provide diversified input for subsequent feature extraction and fusion. The process is as follows: sample scanning, multi-channel image acquisition, real-time transmission, database storage, efficient acquisition capability ensures the basis for dynamic adjustment of feature extraction range in Sp1, and ensures the comprehensiveness and accuracy of defect detection.

[0052] High-performance computing unit: integrated multi-core GPU and dedicated acceleration chip, used to support real-time computing of adaptive feature fusion in Sp1, time-varying causal analysis in Sp2, and defect trend prediction in Sp3, is the core computing engine of the system, based on parallel computing and task allocation, multi-core GPU handles large-scale matrix operations of image feature extraction and fusion, dedicated acceleration chip optimizes complex model computing in time-varying analysis and prediction, working principle realizes high-throughput data processing through hardware acceleration, for example, in Sp1, GPU parallel computing local statistical properties of multi-scale images (such as gray variance), acceleration chip optimizes real-time adjustment of fusion weight; in Sp2, the chip supports time decay coding and causal contribution calculation; in Sp3, GPU drives the iteration of multi-objective optimization model; the running process is: receiving full database transmission craters (SEM images, process parameters, equipment status), allocating computing tasks (feature fusion, causal analysis, trend prediction), parallel processing, outputting results (such as comprehensive feature vector, causal strength, optimization suggestion), transmitting to storage or display, high-performance computing capability ensures the real-time performance of Sp1-Sp3, for example, in Sp3 defect trend prediction, the unit can complete trend decomposition and optimization suggestion generation in seconds, laying the foundation for real-time closed-loop control.

[0053] Process parameter and equipment status monitor: contains sensor array and time synchronization module, used to collect process parameters (such as temperature, pressure) and equipment running status (such as voltage stability) in real time, and store them in alignment with SEM image data, directly supporting Sp2 (dynamic process causal relationship analysis) and Sp3 (defect trend prediction), its running logic is to monitor manufacturing equipment through distributed sensor network, time synchronization module ensures the time consistency of data, based on physical quantity detection (such as thermocouple temperature measurement, pressure sensor pressure measurement) and signal digitization; the running process is: sensor array is deployed on manufacturing equipment (such as lithography machine, etching machine), real-time acquisition of parameters and status (such as etching cabin temperature 45℃, vacuum degree 10^-6 Torr), time synchronization module aligns time stamp (such as synchronization to millisecond level with SEM image acquisition), transmission to full database, provides time-dependent feature input for Sp2, for example, through time decay mechanism to encode temperature change history, generate time-varying feature vector; in Sp3, equipment status (such as voltage fluctuation) is decomposed into periodic fluctuation, to predict potential defect trend, high-precision acquisition and alignment capability of the monitor ensures the context relevance of multi-source data, for example, in the analysis of bridge defects, synchronized temperature and SEM image data reveal the correlation between high temperature and defects.

[0054] Interactive display terminal: equipped with high-resolution touch screen and atlas rendering engine, used to present interactive knowledge graph in Sp4 and support user path reasoning operation, is the key interface for user interaction with the system, its running logic is based on graphics rendering and touch response, the working principle is to convert the associated data of defect type, process step and equipment state into visual atlas through atlas rendering engine, the touch screen supports user operation (such as zoom, path selection), the running process is: receiving Sp4 analysis results (such as causal path, correlation strength), rendering engine building multi-layer knowledge graph (such as "bridge defect, etching temperature" path), displaying on touch screen, user interaction (such as clicking node to query root cause), output reasoning result (such as confidence 0.9), in Sp4, the terminal presents dynamic updated atlas, users can identify abnormal patterns (such as sudden increase of correlation strength) through multi-layer path search, and trigger adjustment suggestions, the interaction supports users to make quick decisions, for example, after identifying abnormal process patterns, the terminal automatically displays "reduce etching power" suggestion, ensuring that the abnormal warning and root cause query functions of Sp4 are intuitive and efficient.

[0055] Adaptive storage server: configured with hierarchical cache structure and dynamic partitioning module, used to dynamically allocate storage space according to SEM image data volume and process analysis requirements, and support fast retrieval and update, is the data management core of Sp1-Sp5, its running logic is based on adaptive resource allocation, hierarchical cache (high-speed SSD and low-speed HDD) stores data according to access frequency, dynamic partitioning module adjusts storage layout according to analysis tasks (such as Sp1 feature extraction, Sp3 trend prediction), the working principle is to optimize data throughput through intelligent scheduling; the running process is: receiving multi-source data (SEM image, process parameter, analysis result), hierarchical cache storage (such as high-frequency access feature vector stored in SSD), dynamic partitioning adjustment (such as allocating more space for Sp3 prediction), supporting fast retrieval (such as millisecond-level extraction of historical defect data), updating data (such as Sp4 atlas incremental update), ensuring fast access of multi-scale images in Sp1, coding efficiency of historical data in Sp2, real-time data support of prediction model in Sp3, and immediate update of feedback data in Sp5, for example, in closed-loop control, the server quickly stores execution feedback to optimize subsequent suggestions.

[0056] Closed-loop control interface unit: integrated programmable logic controller (PLC) and process equipment communication protocol, used to transmit process parameter adjustment suggestions and equipment maintenance plans in Sp3 to semiconductor manufacturing equipment in real time, and receive execution feedback signals to optimize subsequent analysis, is the execution bridge of Sp5 (real-time closed-loop control), its operation logic is based on instruction conversion and feedback cycle, the working principle is to convert analysis results (such as "reduce exposure time by 5%") into device executable instructions through PLC, and communication protocols (such as Modbus, OPCUA) ensure seamless connection with manufacturing equipment; the operation flow is: receiving Sp3 suggestions and Sp4 trigger instructions, PLC generating control signals, transmitting to equipment (such as adjusting parameters of lithography machine), equipment executing and feeding back results (such as reducing defect rate to 2%), returning to the system to update analysis, realizing closed-loop control of Sp5, for example, after executing "cleaning etching cabin", the feedback data shows that the defect is reduced, the system optimizes the prediction model accordingly, ensuring the continuity and real-time of process improvement.

[0057] The operation logic and workflow of hardware components closely cooperate with Sp1-Sp5: multi-channel SEM image acquisition device provides multi-scale images for Sp1, high-performance computing unit drives real-time analysis of Sp1-Sp3, process parameter and equipment state monitor supports multi-source data input of Sp2 and Sp3, interactive display terminal presents atlas and interactive functions of Sp4, adaptive storage server guarantees data management of Sp1-Sp5, and closed-loop control interface unit realizes real-time control of Sp5; the whole process is: image acquisition, calculation analysis, data monitoring, atlas presentation, storage management, control execution and feedback, forming a closed-loop system from defect detection to process optimization, ensuring efficient and accurate semiconductor manufacturing optimization. Specific embodiment three:

[0059] As Figs. 1-2 shown, according to the content in the above specific embodiments, the following content is further disclosed:

[0060] To further verify the feasibility of the present application, compare and analyze the existing commonly used systems or methods for defect detection of semiconductor surface images, and design experimental content to verify the distinguishing features of the present scheme, the specific content is as follows:

[0061] Existing commonly used systems or methods:

[0062] Deep learning-based defect detection systems (e.g., DeepSEM-Net, CN110672644A): Use convolutional neural networks (CNN) or their variants (e.g., ResNet) to extract fixed features from SEM images, classify predefined defect types (e.g., bridging, cracking), rely on single magnification images, fixed feature extraction range, need manual annotation of new defects, lack of multi-scale feature fusion and dynamic adjustment capability, low recognition efficiency for new defects, detection accuracy about 90%, slow response to process changes, suitable for conventional defect detection in stable processes;

[0063] Statistical analysis and offline optimization systems (e.g., SEMVision G10, US20200372635A1): Analyze SEM image defect distribution through statistical methods, generate reports for manual process adjustment; root cause analysis based on historical data, optimization suggestions require additional process generation, no real-time causal analysis and closed-loop control, long response time (hours to days), limited yield improvement (about 2%-5%), unable to intervene actively, suitable for low-frequency defect analysis and post-optimization;

[0064] This technical solution (SEM image automatic analysis and process defect detection system based on full database)

[0065] Distinctive feature 1: Dynamic adaptive defect detection and automatic discovery of new defects (Sp1): Through multi-scale feature adaptive fusion (dynamic adjustment of extraction range and fusion weight) and density anomaly detection, support automatic identification and rule updating of new defect types, combined with multi-channel SEM acquisition device and high-performance computing unit, break through the limitation of fixed feature extraction, adapt to process changes and new defects, detection accuracy up to 98%, false negative rate reduced by 30%-50%;

[0066] Distinctive feature 2: Real-time closed-loop control and process optimization (Sp2-Sp5): Integrate time-varying causal analysis (Sp2), trend prediction and optimization suggestions (Sp3), and closed-loop control (Sp5), rely on monitors, storage servers, and interface units, achieve second-level response and dynamic optimization, surpass offline analysis, yield improvement 5%-10%, response speed greatly improved, build a closed-loop system from root cause analysis to process adjustment.

[0067] System feature differences are shown in Table 1 below:

[0068]

[0069] Table 1

[0070] Experimental objectives: Verify the two distinctive features of this technical solution:

[0071] Dynamic adaptive defect detection and automatic discovery of new defects: Compare detection accuracy, new defect recognition rate, and false negative rate;

[0072] Real-time closed-loop control and process optimization: compare response time, yield improvement range, and process adjustment efficiency;

[0073] Experimental subjects:

[0074] Samples: 3nm process wafers containing known defects (such as bridging, particle deposition) and unknown new defects;

[0075] Comparison systems: DeepSEM-Net (representing deep learning methods), SEMVision G10 (representing statistical analysis methods), and the current system;

[0076] The experimental steps are as follows:

[0077] Step 1: Defect detection experiment:

[0078] Input: 1000 multi-scale SEM images (500x and 5000x magnification, 500 each), containing known defects and new defects;

[0079] Operation: DeepSEM-Net uses single-scale CNN detection, SEMVision G10 performs statistical classification, and the current system performs Sp1 dynamic feature fusion and density anomaly detection;

[0080] Record: detection accuracy (percentage of correctly classified defects), new defect recognition rate (percentage of recognized new defects), and misjudgment rate (percentage of incorrect classification);

[0081] Step 2: Real-time optimization experiment:

[0082] Input: wafer production line data containing defects (SEM images, process parameters, equipment status), simulating high-risk process steps (such as excessively high etching temperature);

[0083] Operation: DeepSEM-Net only detects, SEMVision G10 generates an offline report, and the current system performs Sp2-Sp5 (causal analysis, prediction, closed-loop control) to adjust parameters (such as reducing temperature by 5℃);

[0084] Record: response time (from defect detection to adjustment completion), yield improvement (change in defect rate before and after adjustment), and adjustment efficiency (number of optimizations per unit time);

[0085] Experimental parameters:

[0086] Hardware environment: The current system uses multi-channel SEM acquisition devices (lens magnification 500x-5000x), high-performance computing units (GPU+acceleration chips), monitors, etc.; the comparison systems use standard SEM equipment and general servers;

[0087] Running condition: production line simulation environment, 1000 images per hour, process parameter variation frequency 10 times per hour;

[0088] Repetition: 3 times, average value.

[0089] The experimental results are shown in Table 2 below:

[0090]

[0091] Table 2

[0092] Result analysis:

[0093] Defect detection experiment:

[0094] Detection accuracy: due to multi-scale feature fusion and dynamic adjustment, the accuracy of the system reaches 98.1%, which is better than DeepSEM-Net (90.2%) and SEMVisionG10 (87.5%);

[0095] New defect recognition rate: the system recognizes 85.3% new defects (density anomaly detection), while the comparative system has no such ability, verifying the breakthrough of distinguishing feature 1;

[0096] Misjudgment rate: the misjudgment rate of the system is only 2.4%, which is much lower than that of the comparative system (8.5% and 10.2%), benefiting from adaptability and new defect rule update;

[0097] Real-time optimization experiment:

[0098] Response time: the system responds in seconds (3 seconds), far exceeding DeepSEM-Net (120 seconds) and SEMVisionG10 (3600 seconds), reflecting the advantage of closed-loop control;

[0099] Yield improvement: the system improves by 7.5%, which is better than the comparative system (2.1% and 3.8%), due to the synergistic effect of time-varying analysis and optimization suggestion;

[0100] Adjustment efficiency: the system adjusts 12 times per hour, while the comparative system has little real-time adjustment, verifying the high efficiency of distinguishing feature 2.

[0101] Through multi-scale image input and process parameter variation, the real production line environment is simulated to ensure the credibility of the results. The comparative system represents the current mainstream technology (deep learning and statistical analysis), which is directly related to the distinguishing features of the system. The experimental results show that the system surpasses existing methods in accuracy, adaptability and response speed, and is particularly suitable for efficient defect management of advanced processes. Specific embodiment four:

[0103] As Figs. 1-2 shown, according to the contents in the above specific embodiments, the following contents are further disclosed:

[0104] To further verify the effect of the system in application, the following are two application cases:

[0105] Application Case 1: Detection and optimization of nanoscale particle deposition defects in 3nm process:

[0106] Background and problem: In the research and development of 3nm process, a semiconductor foundry found unknown nanoscale particle deposition defects on the wafer surface. The traditional DeepSEM-Net system relies on a single scale CNN and only identifies predefined defects (such as bridging). The detection accuracy is only 90%, and it cannot identify new defects, resulting in a high false negative rate of 15%, affecting yield and research and development progress.

[0107] Application technical solution:

[0108] Sp1: Dynamic adaptive defect detection: Multi-channel SEM image acquisition device collects wafer images at 500x and 5000x magnification, the system dynamically adjusts the feature extraction range (high magnification focus particle details), and the fusion weight is calculated based on image entropy and process context (such as deposition step) in real time; High-performance computing unit identifies particle deposition as a new defect type through density anomaly detection, and associates it to the deposition temperature change, and updates the classification rules;

[0109] Sp2-Sp3: Cause analysis and optimization: Process parameter monitor collects deposition chamber temperature (50℃), the system confirms that the temperature is too high as the root cause through time-varying analysis, predicts that the particle defect trend will rise to 10%, and generates the suggestion "reduce temperature to 45℃";

[0110] Sp5: Closed-loop control: The closed-loop control interface unit transmits the suggestion to the deposition equipment, and the defect rate is reduced to 2% after execution;

[0111] Effect: Dynamic adaptive detection identifies 85% new particle defects, accuracy reaches 98%, false negative rate is reduced to 2%, which is better than DeepSEM-Net's 0% new defect recognition rate and 15% false negative rate, research and development cycle is shortened by 30%, yield is improved by 7%, and the breakthrough of new defect automatic discovery is verified;

[0112] Application Case 2: Real-time control of etching step bridging defects in 5nm process:

[0113] Background and problem: In the mass production of 5nm process, the bridging defect rate after etching step rises to 5%, the traditional SEMVisionG10 system only provides offline statistical report, the response time is 4 hours, and the yield is improved by only 3% after manual adjustment of etching parameters, which cannot meet the demand of high production line;

[0114] Application technical solution:

[0115] Sp2: Time-varying causal analysis: The process parameter monitor collects etching temperature (60°C) and pressure in real time, the system generates time-varying features through time decay coding, and the disturbance simulation (temperature rise 1°C) quantifies the bridge probability to 7%, confirming that temperature is the key factor;

[0116] Sp3: Trend prediction and optimization: The high-performance computing unit decomposes data and predicts that the defect rate will rise to 8% within 24 hours, and cooperatively generates the suggestion "reduce temperature to 55°C and clean the etching chamber", and the adaptive storage server supports fast data updates;

[0117] Sp5: Closed-loop control: The closed-loop control interface unit transmits the suggestion to the etching machine, and completes the adjustment within 3 seconds, and the feedback shows that the defect rate is reduced to 1.5%;

[0118] Sp4: Knowledge graph: The interactive display terminal presents the "bridge→high temperature" path, and the user confirms the root cause and triggers subsequent checks;

[0119] Effect: Real-time closed-loop control response time is 3 seconds, yield is improved by 7.5%, adjustment efficiency is 12 times / hour, far exceeding the 4-hour response and 0.5 times / hour adjustment of SEMVision G10, production interruption is reduced by 50%, cost is reduced by 10%, and the efficiency of closed-loop optimization is reflected;

[0120] Case summary: Case 1 demonstrates dynamic adaptive detection and new defect discovery (Sp1), solves the problem of unknown defects through multi-scale fusion and density detection, and significantly improves accuracy and adaptability, and case 2 verifies real-time closed-loop control (Sp2-Sp5), optimizes the process through time-varying analysis and rapid intervention, and greatly improves response speed and yield. The combination of the two cases and hardware support fully reflects the distinguishing features and effects of the technical scheme, and provides a breakthrough solution for semiconductor manufacturing.

[0121] It should be noted that in this paper, relationship terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between the entities or operations. Moreover, the terms "include", "contain" or any other variant thereof are intended to cover non-exclusive inclusion, so that the process, method, article or equipment including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or equipment. Without more limitations, the element defined by the statement "including a reference structure" does not exclude the presence of another identical element in the process, method, article or equipment including the element.

[0122] While embodiments of the application have been shown and described, it is to be understood that the embodiments described are merely exemplary of the principles and application of the present application. Numerous modifications and adaptions can be effected without departing from the spirit and scope of the present application, which is not limited to the exact construction and arrangement described. It is intended, therefore, to cover all modifications and adaptions that fall within the scope of the claims and their equivalents.

Claims

1. A system for automatic analysis of SEM images and process defect detection based on full database, characterized in that: The automatic analysis and process defect detection system comprises the following steps: Sp1: adaptively extracting and fusing features from multi-scale scanning electron microscope (SEM) images to detect and classify defect types and support automatic identification of new defect types, which is achieved by density anomaly detection in feature space; when adaptively extracting and fusing features in Sp1, the feature extraction range is dynamically adjusted according to the local statistical characteristics of the image, and the range is reduced in high-detail areas to capture small defects and expanded in smooth areas to identify large-scale anomalies, comprehensive features suitable for different magnifications and process contexts are generated, and the fusion weight is calculated in real time based on the local statistical characteristics of the image and the process step type, including photolithography or etching; Sp2: dynamically analyzing the causal relationship between process steps and defects based on time-dependent features of defect types, process parameters and device states, and separating the interaction effects between process steps; Sp3: predicting defect trends according to the causal relationship and multi-source data, cooperatively generating process parameter adjustment suggestions and device maintenance plans, and optimizing the prediction and suggestions through feedback; Sp4: constructing and updating an interactive knowledge graph containing defect types, process steps and device states in real time, supporting users to query defect root causes through path reasoning; in Sp4, the interactive knowledge graph supports multi-layer path search and confidence evaluation, dynamically generates the optimal causal path from defect type to process step, and updates the graph structure and correlation strength according to new data increments, and the interactive knowledge graph also includes an abnormal pattern recognition function, which identifies abnormal process patterns by analyzing the correlation strength changes of defect types and process steps in the graph, and automatically triggers process adjustment suggestions or device inspection instructions; Sp5: seamlessly integrating the above analysis results into the semiconductor manufacturing process to realize real-time closed-loop control from defect detection to process optimization during semiconductor manufacturing.

2. The full database based automatic analysis of SEM images and process defect detection system of claim 1, wherein: When dynamically analyzing the causal relationship between process steps and defects in Sp2, historical data is encoded through a time decay mechanism to generate time-varying feature vectors, and the causal contribution of process steps to defects and the interaction effect strength are calculated based on the time-varying feature vectors.

3. The full database based automatic analysis of SEM images and process defect detection system of claim 1, wherein: When predicting defect trends in Sp3, SEM image features, process parameters and device states are decomposed into long-term trends and periodic fluctuations, future defect occurrence probabilities are predicted based on the decomposition results, and the priority of the cooperatively generated suggestions is adjusted according to the predicted probabilities.

4. The full database based automatic analysis of SEM images and process defect detection system of claim 1, wherein: The automatic analysis and process defect detection system comprises a data alignment unit for aligning multi-scale SEM images with process parameters and device state data in time and space dimensions, and embedding process context identifiers in the aligned data to enhance the context relevance of the analysis.

5. The full database based automatic analysis of SEM images and process defect detection system of claim 1, wherein: The identified new defect type is associated with the trend of process parameter changes to generate a preliminary cause hypothesis of the new defect, and the cause hypothesis is verified by comparing with historical data to automatically update the defect classification rules of the system.

6. The full database based automatic analysis of SEM images and process defect detection system of claim 1, wherein: In the step Sp2, the dynamic analysis process step and the cause-effect relationship between the defects are analyzed by simulating a slight disturbance of the process parameters, quantifying the impact of the disturbance on the probability of defects, and generating a priority adjustment sequence of the process steps based on the quantification results to minimize the defect occurrence rate.

7. The full database based automatic analysis of SEM images and process defect detection system of claim 1, wherein: In the step Sp3, the process parameter adjustment suggestions and the equipment maintenance plan are generated in coordination, including constructing a multi-objective optimization model based on the defect trend prediction results and the equipment operating state, simultaneously optimizing the defect rate reduction and the maintenance cost control, and dynamically adjusting the model weights in the optimization process to adapt to changes in production requirements.

8. The full database based automatic analysis of SEM images and process defect detection system of claim 1, wherein: The hardware components of the automatic analysis and process defect detection system include: A multi-channel SEM image acquisition device configured with multiple magnification lenses and real-time data transmission interfaces for acquiring multi-scale SEM images and transmitting them to the full database; A high-performance computing unit integrating multi-core GPUs and special-purpose acceleration chips for supporting real-time computing of adaptive feature fusion, time-varying cause-effect analysis, and defect trend prediction; A process parameter and equipment state monitor containing a sensor array and a time synchronization module for real-time acquisition of process parameters and equipment operating state, and storage aligned with SEM image data; An interactive display terminal equipped with a high-resolution touch screen and a graph rendering engine for presenting the interactive knowledge graph and supporting user path reasoning operations; An adaptive storage server configured with a hierarchical cache structure and a dynamic partitioning module for dynamically allocating storage space based on SEM image data volume and process analysis requirements, and supporting fast retrieval and update; A closed-loop control interface unit integrating a programmable logic controller (PLC) and a process equipment communication protocol for real-time transmission of the process parameter adjustment suggestions and the equipment maintenance plan to semiconductor manufacturing equipment, and receiving execution feedback signals for optimization of subsequent analysis.

Images