Wafer process link pollution source identification method based on FOUP inner interlayer AMC distribution

Abstract

The invention discloses a wafer process link pollution source identification method based on FOUP inner interlayer AMC distribution, and relates to the technical field of pollution source identification, the method comprises the following steps: analyzing airflow disturbance and pollutant diffusion characteristics, formulating an external detector deployment strategy, and collecting pollutant concentration and associated equipment operation parameters in real time; establishing a space-time matrix of pollutant concentration, extracting gradient characteristics of pollutant concentration change by using wavelet transform, and obtaining initial position data of a pollution source in combination with operation parameters of associated equipment; constructing a space-time gridding input matrix, designing a double-branch deep learning model, generating a pollution source positioning probability graph, and obtaining position data of a pollution source; the contribution weight of each associated device to pollutant diffusion is evaluated, and a directional treatment strategy is formulated and executed. According to the invention, the position of the pollution source can be accurately positioned, and the accuracy and response speed of pollution source identification are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of pollution source identification, and more specifically, to a method for identifying pollution sources in the wafer manufacturing process based on the AMC distribution between the inner layers of a FOUP. Background Art

[0002] During the semiconductor manufacturing process, wafers are exposed to a complex and highly clean process environment and are extremely sensitive to the concentration of pollutants in the air. Among them, Airborne Molecular Contaminants (AMC) have become one of the important factors affecting the surface quality, device performance, and yield of wafers due to their small molecular weight, strong diffusibility, and difficulty in detection.

[0003] To ensure a clean environment for wafers during transportation and storage, a Front Opening Unified Pod (FOUP) is widely used as a wafer carrier and protection device. However, during the actual production process, due to reasons such as decreased equipment sealing, material volatilization, and process gas residues, AMC may invade the interior of the FOUP and undergo non-uniform diffusion between the internal spaces, resulting in pollution differences between different levels of the wafer, and thus triggering batch process defects.

[0004] Currently, the monitoring of AMC pollution inside the FOUP mainly relies on fixed-position detectors for fixed-point collection and analysis, lacking the ability to dynamically model the spatio-temporal distribution characteristics of pollutants and locate pollution sources. In addition, traditional monitoring methods fail to fully combine the internal structure characteristics of the FOUP and the airflow movement law, resulting in low accuracy of pollution source identification and slow response speed, and unable to meet the requirements of real-time and accurate pollution control in high-precision wafer manufacturing processes.

[0005] Regarding the problems in the related art, no effective solution has been proposed yet. Summary of the Invention

[0006] In view of the problems in the related art, the present invention proposes a method for identifying pollution sources in the wafer manufacturing process based on the AMC distribution between the inner layers of a FOUP to overcome the above-mentioned technical problems existing in the existing related technologies.

[0007] To this end, the specific technical solution adopted by the present invention is as follows: A method for identifying pollution sources in the wafer manufacturing process based on the AMC distribution between the inner layers of a FOUP, the method comprising: [[ID=2⑨]]S1. Analyze the airflow disturbance and pollutant diffusion characteristics according to the internal space of the transport box and the wafer layer placement, formulate an external detector deployment strategy, and collect the pollutant concentration and associated equipment operation parameters in real time; S2. Establish a spatio-temporal matrix of pollutant concentrations, extract the gradient features of pollutant concentration changes using wavelet transform, and combine with the operating parameters of associated devices to obtain preliminary location data of pollution sources; S3. Construct a spatio-temporal grid input matrix, design a dual-branch deep learning model, generate a pollution source location probability map, and obtain the location data of pollution sources; S4. Based on the location data of pollution sources, evaluate the contribution weights of each associated device to pollutant diffusion, and formulate and implement a targeted treatment strategy.

[0008] Optionally, analyze the airflow disturbance and pollutant diffusion characteristics according to the internal space of the transport box and the wafer-level placement, formulate an external detector deployment strategy, and collect pollutant concentrations and associated device operating parameters in real time, including: S11. Based on the internal space of the transport box, establish a parameterized geometric model, set fluid boundary conditions and run computational fluid dynamics simulations to obtain benchmark airflow fields and pollutant diffusion characteristic data for constructing a digital twin model; S12. Combine the wafer-level placement, design an initial external detector deployment strategy, and collect the initial pollutant concentrations, airflow velocities, and associated device operating parameters at each level to establish a spatio-temporally correlated pollution distribution baseline; S13. Combine the established pollution distribution baseline and the constructed digital twin model, and use the adversarial transfer learning algorithm to optimize the digital twin model; S14. Use the optimized digital twin model to predict the spatio-temporal distribution characteristics of pollutants, and combine with the multi-armed bandit algorithm to update the initial external detector deployment strategy to obtain the final external detector deployment strategy; S15. Implement the final external detector deployment strategy and collect pollutant concentrations and associated device operating parameters in real time.

[0009] Optionally, combining the established pollution distribution baseline and the constructed digital twin model, using the adversarial transfer learning algorithm to optimize the digital twin model includes: S131. Input the established pollution distribution baseline into the constructed digital twin model, simulate to obtain simulation data, and combine with the initial pollutant concentrations, airflow velocities, and device operating parameters to construct a dataset for adversarial transfer learning; S132. Based on the adversarial transfer learning algorithm, construct an adversarial transfer learning model, and the adversarial transfer learning model includes a convolutional encoder, a classifier, and a domain discriminator; S133. Use the convolutional encoder to extract the spatio-temporal features of pollutant concentrations in the dataset to generate feature vectors; S134. Use the domain discriminator to analyze the inter-domain differences of the feature vectors, and combine the convolutional neural network to identify the distribution differences between the simulation data and the initial pollutant concentration, air flow velocity, and equipment operation parameters. Calculate the domain classification loss based on the identification results; S135. According to the calculated domain classification loss and the adversarial training mechanism, optimize the classifier and the domain discriminator to obtain an optimized adversarial transfer learning model; S136. Combine the optimized adversarial transfer learning model with the digital twin model to optimize the parameters of the digital twin model.

[0010] Optionally, use the optimized digital twin model to predict the spatio-temporal distribution characteristics of pollutants, and combine the multi-armed bandit algorithm to update the initial external detector deployment strategy to obtain the final external detector deployment strategy, including: S141. Input the initial pollutant concentration, air flow velocity, and equipment operation parameters into the optimized digital twin model to predict the spatio-temporal distribution characteristics of the pollutant concentration in the internal space of the transport box; S142. Based on the designed initial external detector deployment strategy, use the nearest neighbor algorithm to calculate the spatio-temporal similarity between each external detector node, and identify several candidate external detector nodes that best match the current spatio-temporal distribution characteristics; S143. According to the identified candidate external detector nodes and the preset pollutant prediction function, calculate the average weight of each candidate external detector node, and evaluate the expected monitoring effect of each candidate external detector node according to the spatio-temporal distribution characteristics of the pollutant concentration; S144. Combine the multi-armed bandit algorithm to select the candidate external detector node with the best expected monitoring effect to update the initial external detector deployment strategy to obtain the final external detector deployment strategy.

[0011] Optionally, combine the multi-armed bandit algorithm to select the candidate external detector node with the best expected monitoring effect to update the initial external detector deployment strategy to obtain the final external detector deployment strategy, including: S1441. Map each candidate external detector node to a decision arm in the multi-armed bandit algorithm and construct a reward function based on the monitoring efficiency; S1442. Based on the expected monitoring effect of each candidate external detector node, initialize the reward function of each decision arm and set the initial confidence interval; S1443. According to the initial confidence interval of each decision arm, calculate the upper bound value of each initial confidence interval, and select the candidate external detector node with the highest upper bound value as the optimal external detector node through the exploration mechanism of the multi-armed bandit algorithm; S1444. Generate suggestions for the deployment of external detectors in the next stage based on the optimal external detector nodes, so as to update the external detector deployment strategy and obtain the final external detector deployment strategy.

[0012] Optionally, construct a spatio-temporal grid input matrix, design a dual-branch deep learning model, generate a pollution source location probability map, and obtain the location data of the pollution source, including: S31. Based on the preliminary location data of the pollution source and the pollutant concentration collected in real time, construct a spatio-temporal grid input matrix; S32. According to the convolutional neural network and the time attention mechanism, design a dual-branch deep learning model, and use the dual-branch deep learning model to extract the spatial features and time features of pollutants in the spatio-temporal grid input matrix; S33. Integrate the spatial features and time features of pollutants, and inject the integration result into a preset pollution source heat map through feature splicing and full connection layer compression to generate a pollution source location probability map; S34. Based on the generated pollution source location probability map and the Markov decision process, and combined with offline simulation training and greedy strategy, obtain the location data of the pollution source.

[0013] Optionally, integrate the spatial features and time features of pollutants, and inject the integration result into a preset pollution source heat map through feature splicing and full connection layer compression to generate a pollution source location probability map, including: S331. According to the newly deployed external detector nodes, align the extracted spatial features and time features of pollutants, and perform feature splicing to generate a joint feature matrix; S332. Input the generated joint feature matrix into the full connection layer, use non-linear dimensionality reduction technology for compression, and extract feature components through low-rank matrix factorization technology to obtain the compressed feature representation; S333. Perform weighted fusion on the compressed feature representation and the preset pollution source heat map, and use Gaussian filtering to smooth the fusion result to output the pollution source location probability map.

[0014] Optionally, based on the generated pollution source location probability map and the Markov decision process, and combined with offline simulation training and greedy strategy, obtain the location data of the pollution source, including: S341. Based on the Markov decision process, use the pollution source location probability map as the environmental state, the pollution source verification operation as the agent action, and combined with the spatial features and time features of pollutants, establish a state transition rule; S342. According to the established state transition rule, construct a multi-objective reward function; S343. Generate a virtual pollution source scenario dataset using the optimized digital twin model, construct a convolutional policy network based on the dataset, and combine a multi-objective reward function to perform offline simulation training through the proximal policy optimization algorithm; S344. According to the action value output by the trained convolutional policy network, use the greedy policy to select the optimal pollution source verification operation, and update the pollution source localization probability map in combination with the data feedback from the external detector; S345. Based on the updated pollution source localization probability map, generate a verification path, and optimize the verification path in combination with the entropy threshold to obtain the location data of the pollution source.

[0015] Optionally, the multi-objective reward function includes: operation cost penalty, successful localization bonus, information gain reward, probability density reward, and distance penalty.

[0016] Optionally, based on the location data of the pollution source, evaluate the contribution weights of each associated device to the pollutant diffusion, and formulate and execute a targeted treatment strategy, including: S41. Combine the location data of the pollution source, the pollutant concentration collected in real time, and the operating parameters of the associated devices to construct a causal graph model; S42. Based on the constructed causal graph model, use causal inference technology to evaluate the influence degree of each associated device on the pollutant diffusion process, obtain the contribution weights, and formulate and execute a targeted treatment strategy based on the contribution weights.

[0017] The beneficial effects of the present invention are as follows: 1. The present invention combines the airflow in the transportation box, the diffusion characteristics of pollutants, and the operating parameters of related devices to achieve real-time monitoring and collection of environmental data; by extracting and analyzing the pollutant concentration gradient characteristics in these data, the location of the pollution source can be accurately determined, thereby improving the accuracy and response speed of pollution source identification and preventing the decline of product or environmental quality caused by the failure to detect pollution in time.

[0018] 2. The present invention can accurately identify the pollution source by evaluating the contribution degree of each associated device to the pollutant diffusion based on the pollution source localization result, and formulate a targeted treatment strategy accordingly; by combining the pollution source location information and the device operating status, the device operating parameters can be adjusted in real time, thereby reducing the risk of pollutant diffusion. Description of the Drawings

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0020] Figure 1 It is a flowchart of a method for identifying pollution sources in the wafer manufacturing process based on the AMC distribution between the inner layers of the FOUP according to an embodiment of the present invention. Detailed implementation manners

[0021] To further illustrate each embodiment, the present invention provides accompanying drawings, which are part of the disclosure of the present invention. They are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these contents, those of ordinary skill in the art should be able to understand other possible implementation manners and the advantages of the present invention.

[0022] According to an embodiment of the present invention, a method for identifying pollution sources in the wafer manufacturing process based on the AMC distribution between the inner layers of the FOUP is provided.

[0023] Now, the present invention will be further described in conjunction with the accompanying drawings and detailed implementation manners. As Figure 1 shown, the method for identifying pollution sources in the wafer manufacturing process based on the AMC distribution between the inner layers of the FOUP according to an embodiment of the present invention includes: S1. Analyze the airflow disturbance and pollutant diffusion characteristics according to the internal space of the transport cassette and the wafer layer placement, formulate an external detector deployment strategy, and collect the pollutant concentration and the operating parameters of associated equipment in real time.

[0024] In this optional embodiment, analyzing the airflow disturbance and pollutant diffusion characteristics according to the internal space of the transport cassette and the wafer layer placement, formulating an external detector deployment strategy, and collecting the pollutant concentration and the operating parameters of associated equipment in real time includes: S11. Based on the internal space of the transport cassette, establish a parameterized geometric model, set the fluid boundary conditions and run a computational fluid dynamics simulation to obtain the reference airflow field and pollutant diffusion characteristic data, so as to construct a digital twin model.

[0025] It should be added that establishing a parameterized geometric model based on the internal space of the transport cassette, setting the fluid boundary conditions and running a computational fluid dynamics simulation to obtain the reference airflow field and pollutant diffusion characteristic data, so as to construct a digital twin model includes: First, establish a parametric geometric model based on the internal space of the transport box to clarify the spatial dimensions and connection relationships. Secondly, set key parameters such as fluid inlet / outlet boundary conditions, initial pollutant concentration distribution, and material surface characteristics. Then, run computational fluid dynamics (CFD) simulation calculations. By solving the Navier-Stokes equations and pollutant transport equations, obtain the spatio-temporal distribution data of the velocity field, pressure field, and concentration field. Next, perform grid processing and feature extraction on the simulation results to establish a database containing key parameters such as air velocity and pollutant diffusion coefficient. Finally, align the simulation data with the monitoring data of real-time external detectors, and construct a digital twin model that can be updated in real time through dynamic calibration to provide a benchmark reference for subsequent pollution source identification.

[0026] S12. Combine the wafer-level placement, design an initial external detector deployment strategy, and collect the initial pollutant concentration, air velocity, and associated equipment operation parameters at each level to establish a spatio-temporal correlated pollution distribution baseline.

[0027] S13. Combine the established pollution distribution baseline with the constructed digital twin model, and use the adversarial transfer learning algorithm to optimize the digital twin model.

[0028] In this optional embodiment, combining the established pollution distribution baseline with the constructed digital twin model and using the adversarial transfer learning algorithm to optimize the digital twin model includes: S131. Input the established pollution distribution baseline into the constructed digital twin model, simulate and obtain simulation data, and combine the initial pollutant concentration, air velocity, and equipment operation parameters to construct a dataset for adversarial transfer learning. S132. Based on the adversarial transfer learning algorithm, construct an adversarial transfer learning model, which includes a convolutional encoder, a classifier, and a domain discriminator. S133. Use the convolutional encoder to extract the spatio-temporal features of the pollutant concentration in the dataset to generate feature vectors. S134. Use the domain discriminator to analyze the inter-domain differences of the feature vectors, and combine the convolutional neural network to identify the distribution differences between the simulation data and the initial pollutant concentration, air velocity, and equipment operation parameters. Calculate the domain classification loss based on the identification results. S135. According to the calculated domain classification loss and the adversarial training mechanism, optimize the classifier and the domain discriminator to obtain an optimized adversarial transfer learning model. S136. Combine the optimized adversarial transfer learning model with the digital twin model to optimize the parameters of the digital twin model.

[0029] It should be noted that by inputting the pollution distribution baseline into the digital twin model and combining it with the adversarial transfer learning algorithm, the distribution difference between the simulation data and the actual data can be effectively overcome, and the accuracy of pollution source identification can be improved; by extracting spatio-temporal features through a convolutional encoder and generating feature vectors, the domain discriminator is further used to analyze the differences between data to ensure the adaptability of the model to different data sources; by optimizing the classifier and the domain discriminator through adversarial training, the sensitivity of the model to pollutant concentration, air flow velocity, and equipment operation parameters is improved, the generalization ability of the model is enhanced, and real-time optimization and accurate prediction of the digital twin model in a dynamic environment are ensured; finally, the precise optimization of the digital twin model is achieved, making the identification and treatment of pollution sources more efficient, reducing the costs of experiments and actual monitoring, and improving the stability of the production process and the product quality.

[0030] S14. Predict the spatio-temporal distribution characteristics of pollutants using the optimized digital twin model, and update the initial external detector deployment strategy in combination with the multi-armed bandit algorithm to obtain the final external detector deployment strategy.

[0031] In this optional embodiment, predicting the spatio-temporal distribution characteristics of pollutants using the optimized digital twin model and updating the initial external detector deployment strategy in combination with the multi-armed bandit algorithm to obtain the final external detector deployment strategy includes: S141. Input the initial pollutant concentration, air flow velocity, and equipment operation parameters into the optimized digital twin model to predict the spatio-temporal distribution characteristics of the pollutant concentration in the internal space of the transport box; S142. Based on the designed initial external detector deployment strategy, use the nearest neighbor algorithm to calculate the spatio-temporal similarity between each external detector node, and identify several candidate external detector nodes that best match the current spatio-temporal distribution characteristics; S143. According to the identified candidate external detector nodes and the preset pollutant prediction function, calculate the average weight of each candidate external detector node, and evaluate the expected monitoring effect of each candidate external detector node according to the spatio-temporal distribution characteristics of the pollutant concentration; S144. Combine the multi-armed bandit algorithm to select the candidate external detector node with the best expected monitoring effect to update the initial external detector deployment strategy and obtain the final external detector deployment strategy.

[0032] It should be noted that by inputting the initial pollutant concentration, air flow velocity, and equipment operation parameters into the optimized digital twin model, accurate prediction of the spatio-temporal distribution of pollutant concentration inside the transport box is achieved, providing a reliable basis for the deployment of external detectors; the spatio-temporal similarity between external detector nodes is calculated using the nearest neighbor algorithm to effectively identify candidate nodes that best match the current distribution characteristics, ensuring the scientific nature of the external detector layout; by calculating the average weight of candidate nodes and evaluating their expected monitoring effects, and combining with the multi-armed bandit algorithm to select the optimal nodes, the external detector deployment strategy is dynamically updated, significantly improving the representativeness and accuracy of monitoring data, reducing the use of redundant external detectors, lowering the deployment cost, while enhancing the real-time and accuracy of pollution source identification, providing strong support for efficient and accurate pollution control.

[0033] In this alternative embodiment, combining with the multi-armed bandit algorithm, selecting the candidate external detector node with the best expected monitoring effect to update the initial external detector deployment strategy and obtain the final external detector deployment strategy includes: S1441. Map each candidate external detector node to a decision arm in the multi-armed bandit algorithm and construct a reward function based on monitoring efficiency. S1442. Initialize the reward function of each decision arm based on the expected monitoring effect of each candidate external detector node and set the initial confidence interval. S1443. According to the initial confidence interval of each decision arm, calculate the upper bound value of each initial confidence interval, and through the exploration mechanism of the multi-armed bandit algorithm, select the candidate external detector node with the highest upper bound value as the optimal external detector node. S1444. Generate a deployment recommendation for the external detector in the next stage based on the optimal external detector node to update the external detector deployment strategy and obtain the final external detector deployment strategy.

[0034] It should be noted that by mapping the candidate external detector nodes to the decision arms of the multi-armed bandit algorithm and constructing a reward function based on monitoring efficiency, the expected monitoring effects of each external detector node can be effectively evaluated, realizing the dynamic optimization of the external detector deployment strategy; initializing the reward function and confidence interval of each decision arm ensures the balance between exploration and exploitation of the algorithm. By calculating the upper bound value of the confidence interval and selecting the optimal external detector node, the pertinence and efficiency of the external detector deployment are improved; generating deployment recommendations based on the optimal nodes and continuously updating the deployment strategy, finally obtaining the optimal external detector deployment plan, improving the coverage and accuracy of pollution monitoring, reducing the use of redundant external detectors, lowering the deployment cost, enhancing the real-time response ability of the system, and providing a reliable data basis for accurate pollution source identification and efficient treatment.

[0035] S15. Execute the final external detector deployment strategy and collect pollutant concentration and associated equipment operation parameters in real time.

[0036] S2. Establish a spatio-temporal matrix of pollutant concentration, extract the gradient features of pollutant concentration changes using wavelet transform, and combine with the operation parameters of associated equipment to obtain preliminary location data of the pollution source.

[0037] It should be noted that establishing a spatio-temporal matrix of pollutant concentration, extracting the gradient features of pollutant concentration changes using wavelet transform, and combining with the operation parameters of associated equipment to obtain preliminary location data of the pollution source includes: First, collect the pollutant concentration and organize it into a spatio-temporal matrix according to time and space distribution to ensure that the data at each moment and each internal space can accurately reflect the changes in pollutants; then, use wavelet transform to process the spatio-temporal matrix and extract the gradient features of pollutant concentration changes to capture its fluctuation trends in time and space; next, combine these gradient features with equipment operation parameters (such as air flow velocity, equipment status, etc.) to analyze the influence of different factors on pollutant diffusion, and obtain preliminary location data of the pollution source through model fusion method, providing a basis for subsequent pollution source location and treatment.

[0038] S3. Construct a spatio-temporal grid input matrix, design a two-branch deep learning model, generate a pollution source location probability map, and obtain the location data of the pollution source.

[0039] In this alternative embodiment, constructing a spatio-temporal grid input matrix, designing a two-branch deep learning model, generating a pollution source location probability map, and obtaining the location data of the pollution source includes: S3-1. Based on the preliminary location data of the pollution source and the pollutant concentration collected in real time, construct a spatio-temporal grid input matrix.

[0040] S3-2. According to the convolutional neural network and time attention mechanism, design a two-branch deep learning model, and use the two-branch deep learning model to extract the spatial and time features of pollutants in the spatio-temporal grid input matrix.

[0041] It should be noted that according to the convolutional neural network and time attention mechanism, designing a two-branch deep learning model, and using the two-branch deep learning model to extract the spatial and time features of pollutants in the spatio-temporal grid input matrix includes: First, construct a spatio-temporal grid input matrix, where the pollutant concentration and spatial information at each time step are integrated into input features. Then, design a dual-branch deep learning model. The first branch uses a convolutional neural network to process spatial features and extracts the spatial distribution features of pollutants through multiple convolutional layers. The second branch uses a temporal attention mechanism to weight the pollutant concentration at each time step and learn the features of important moments in the time dimension, thereby capturing the temporal variation law of pollutants. Then, fuse the output features of the two branches, combine the spatial features and temporal features, and finally form a complete spatio-temporal feature representation for subsequent tasks (such as pollution source identification). Through this model, the variation laws of pollutants in space and time can be captured simultaneously, improving the accuracy and stability of pollution source localization.

[0042] S33. Fuse the spatial features and temporal features of pollutants, and inject the fusion result into a preset pollution source heat map through feature concatenation and full connection layer compression to generate a pollution source localization probability map.

[0043] In this optional embodiment, fusing the spatial features and temporal features of pollutants and injecting the fusion result into a preset pollution source heat map through feature concatenation and full connection layer compression to generate a pollution source localization probability map includes: S331. According to the newly deployed external detector nodes, align the extracted spatial features and temporal features of pollutants and perform feature concatenation to generate a joint feature matrix. S332. Input the generated joint feature matrix into a full connection layer, compress it using nonlinear dimensionality reduction technology, and extract feature components through low-rank matrix factorization technology to obtain a compressed feature representation. S333. Perform weighted fusion of the compressed feature representation and the preset pollution source heat map, and smooth the fusion result using Gaussian filtering to output a pollution source localization probability map.

[0044] It should be added that by aligning and concatenating the spatial and temporal features of pollutants extracted by the newly deployed external detector nodes, a joint feature matrix can be generated, realizing the effective fusion of multi-source data. By inputting the joint feature matrix into a full connection layer and using nonlinear dimensionality reduction technology for compression, the data dimension can be effectively reduced, improving the calculation efficiency while retaining key feature information. The low-rank matrix factorization technology further extracts feature components, optimizes the feature representation, and ensures the accuracy and precision of pollution source localization. Finally, perform weighted fusion of the compressed feature representation and the pollution source heat map, and use Gaussian filtering for smoothing to reduce noise interference and output an accurate pollution source localization probability map, providing efficient and reliable support for the real-time monitoring and treatment of pollution sources.

[0045] S34. Based on the generated pollution source location probability map and the Markov decision process, and combined with offline simulation training and the greedy strategy, obtain the location data of the pollution source.

[0046] In this alternative embodiment, based on the generated pollution source location probability map and the Markov decision process, and combined with offline simulation training and the greedy strategy, obtaining the location data of the pollution source includes: S341. Based on the Markov decision process, take the pollution source location probability map as the environmental state, the pollution source verification operation as the agent action, and combined with the spatial and temporal characteristics of the pollutant, establish the state transition rule.

[0047] S342. According to the established state transition rule, construct a multi-objective reward function.

[0048] In this alternative embodiment, the multi-objective reward function includes: operation cost penalty, successful location heavy reward, information gain reward, probability density reward, and distance penalty.

[0049] S343. Use the optimized digital twin model to generate a virtual pollution source scenario dataset, construct a convolutional policy network based on the dataset, and combined with the multi-objective reward function, perform offline simulation training through the proximal policy optimization algorithm.

[0050] S344. According to the action value output by the trained convolutional policy network, use the greedy strategy to select the optimal pollution source verification operation, and update the pollution source location probability map in combination with the external detector feedback data.

[0051] S345. Based on the updated pollution source location probability map, generate a verification path, and optimize the verification path in combination with the entropy threshold to obtain the location data of the pollution source.

[0052] It should be noted that by combining the Markov decision process with the digital twin model, the problem of pollution source location can be transformed into an intelligent decision-making task, improving the accuracy and efficiency of pollution source location; by constructing a multi-objective reward function, the model can comprehensively consider multiple factors such as operation cost, location success rate, information gain, probability density, and verification distance, optimize the pollution source verification strategy, and avoid the limitations of single-objective optimization; using the optimized digital twin model to generate a virtual data set, and based on this, constructing a convolutional policy network, and performing offline training through the proximal policy optimization algorithm to ensure the robustness and stability of the model in various environments; after training, using the greedy strategy to select the optimal operation, combining the external detector feedback data to dynamically update the pollution source location probability map, realizing the generation of an accurate pollution source verification path, and further optimizing the verification path through the entropy value threshold, finally obtaining the accurate pollution source location data; improving the efficiency and accuracy of pollution source location, being able to adapt to changing environments in real time and optimize the decision-making path, and significantly reducing the cost of pollution source identification.

[0053] S4. Based on the location data of the pollution source, evaluate the contribution weights of each associated device to the pollutant diffusion, and formulate and implement a targeted treatment strategy.

[0054] In this alternative embodiment, based on the location data of the pollution source, evaluating the contribution weights of each associated device to the pollutant diffusion and formulating and implementing a targeted treatment strategy includes: S41. Combine the location data of the pollution source, the pollutant concentration collected in real time, and the operating parameters of the associated devices to construct a causal graph model; S42. Based on the constructed causal graph model, use causal inference technology to evaluate the influence degree of each associated device on the pollutant diffusion process, obtain the contribution weights, and formulate and implement a targeted treatment strategy based on the contribution weights.

[0055] It should be noted that by combining the pollution source location data, the real-time pollutant concentration, and the device operating parameters to construct a causal graph model, the causal relationship between the pollution source and the device can be clarified, revealing the source and propagation path of the pollutant diffusion; using causal inference technology to evaluate the influence degree of each associated device on the pollutant diffusion can accurately identify which devices have a significant contribution to the pollutant diffusion, so as to obtain the contribution weight of each device; this information can help formulate more accurate and effective targeted treatment strategies, optimize the device operation or adjust the process, reduce the pollutant diffusion risk, and improve the treatment efficiency; it can achieve the precision and intelligence of pollution source control, improve the accuracy and response speed of pollution management, and reduce the pollution treatment cost in the production process.

[0056] In summary, by means of the above technical solutions of the present invention, through the combination of the airflow in the transport box, the diffusion characteristics of pollutants, and the operating parameters of related equipment, the real-time monitoring and collection of environmental data are realized; by extracting and analyzing the characteristics of pollutant concentration gradients in these data, the location of the pollution source can be accurately determined, thereby improving the accuracy and response speed of pollution source identification and preventing the decline of product or environmental quality caused by the failure to detect pollution in a timely manner. By evaluating the contribution degree of each associated device to pollutant diffusion based on the pollution source location result, the pollution source can be accurately identified, and a targeted treatment strategy can be formulated accordingly; by combining the pollution source location information with the equipment operating status, the equipment operation parameters can be adjusted in real time, thereby reducing the risk of pollutant diffusion.

[0057] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for identifying pollution sources in the wafer manufacturing process based on the AMC distribution between the inner layers of the FOUP, characterized in that The method includes: S1. Analyze the airflow disturbance and pollutant diffusion characteristics according to the internal space of the transport box and the wafer layer placement, formulate an external detector deployment strategy, and collect the pollutant concentration and associated equipment operation parameters in real time; S2. Establish a spatio-temporal matrix of the pollutant concentration, extract the gradient characteristics of the pollutant concentration change using wavelet transform, and combine with the operation parameters of the associated equipment to obtain the preliminary location data of the pollution source; S3. Construct a spatio-temporal grid input matrix, design a dual-branch deep learning model, generate a pollution source location probability map, and obtain the location data of the pollution source; S4. Based on the location data of the pollution source, evaluate the contribution weights of each associated equipment to the pollutant diffusion, and formulate and implement a targeted treatment strategy.

2. The method for identifying pollution sources in the wafer manufacturing process based on the AMC distribution between the inner layers of the FOUP according to claim 1, wherein The step of analyzing the airflow disturbance and pollutant diffusion characteristics according to the internal space of the transport box and the wafer layer placement, formulating an external detector deployment strategy, and collecting the pollutant concentration and associated equipment operation parameters in real time includes: S11. Based on the internal space of the transport box, establish a parameterized geometric model, set the fluid boundary conditions and run computational fluid dynamics simulation to obtain the reference airflow field and pollutant diffusion characteristic data, so as to construct a digital twin model; S12. Combine the wafer layer placement, design an initial external detector deployment strategy, and collect the initial pollutant concentration, airflow velocity and associated equipment operation parameters at each layer to establish a spatio-temporal associated pollution distribution baseline; S13. Combine the established pollution distribution baseline and the constructed digital twin model, and optimize the digital twin model using the adversarial transfer learning algorithm; S14. Use the optimized digital twin model to predict the spatio-temporal distribution characteristics of pollutants, and combine with the multi-armed bandit algorithm to update the initial external detector deployment strategy to obtain the final external detector deployment strategy; S15. Implement the final external detector deployment strategy, and collect the pollutant concentration and associated equipment operation parameters in real time.

3. The method for identifying pollution sources in the wafer manufacturing process based on the AMC distribution between the inner layers of the FOUP according to claim 2, wherein The step of combining the established pollution distribution baseline and the constructed digital twin model, and optimizing the digital twin model using the adversarial transfer learning algorithm includes: S131. Input the established pollution distribution baseline into the constructed digital twin model, simulate and obtain simulation data, and combine with the initial pollutant concentration, airflow velocity and equipment operation parameters to construct a dataset for adversarial transfer learning; S132. Based on the adversarial transfer learning algorithm, construct an adversarial transfer learning model, and the adversarial transfer learning model includes a convolutional encoder, a classifier and a domain discriminator; S133. Use the convolutional encoder to extract the spatio-temporal characteristics of the pollutant concentration in the dataset to generate a feature vector; S134. Use the domain discriminator to analyze the inter-domain differences of the feature vectors, and combine with the convolutional neural network to identify the distribution differences between the simulation data and the initial pollutant concentration, airflow velocity and equipment operation parameters, and calculate the domain classification loss based on the identification results; S135. Optimize the classifier and the domain discriminator according to the calculated domain classification loss and the adversarial training mechanism to obtain an optimized adversarial transfer learning model; S136. Combine the optimized adversarial transfer learning model and the digital twin model to optimize the parameters of the digital twin model.

4. The method for identifying pollution sources in the wafer manufacturing process based on the AMC distribution between the inner layers of the FOUP according to claim 3, wherein Predicting the spatio-temporal distribution characteristics of pollutants based on the optimized digital twin model, and updating the initial external detector deployment strategy by combining the multi-armed bandit algorithm to obtain the final external detector deployment strategy includes: S141. Input the initial pollutant concentration, air flow velocity, and equipment operation parameters into the optimized digital twin model to predict the spatio-temporal distribution characteristics of the pollutant concentration in the internal space of the transport box; S142. Based on the designed initial external detector deployment strategy, use the nearest neighbor algorithm to calculate the spatio-temporal similarity between each external detector node, and identify several candidate external detector nodes that best match the current spatio-temporal distribution characteristics; S143. According to the identified candidate external detector nodes and the preset pollutant prediction function, calculate the average weight of each candidate external detector node, and evaluate the expected monitoring effect of each candidate external detector node according to the spatio-temporal distribution characteristics of the pollutant concentration; S144. Combine the multi-armed bandit algorithm to select the candidate external detector node with the best expected monitoring effect to update the initial external detector deployment strategy and obtain the final external detector deployment strategy.

5. The method for identifying pollution sources in the wafer manufacturing process based on the AMC distribution between the inner layers of the FOUP according to claim 4, wherein, The combination of the multi-armed bandit algorithm to select the candidate external detector node with the best expected monitoring effect to update the initial external detector deployment strategy and obtain the final external detector deployment strategy includes: S1441. Map each candidate external detector node to a decision arm in the multi-armed bandit algorithm and construct a reward function based on the monitoring efficiency; S1442. Initialize the reward function of each decision arm based on the expected monitoring effect of each candidate external detector node and set the initial confidence interval; S1443. According to the initial confidence interval of each decision arm, calculate the upper bound value of each initial confidence interval, and select the candidate external detector node with the highest upper bound value as the optimal external detector node through the exploration mechanism of the multi-armed bandit algorithm; S1444. Generate the external detector deployment suggestion for the next stage based on the optimal external detector node to update the external detector deployment strategy and obtain the final external detector deployment strategy.

6. The method for identifying pollution sources in the wafer manufacturing process based on the AMC distribution between the inner layers of the FOUP according to claim 1, wherein, The construction of the spatio-temporal grid input matrix, the design of the dual-branch deep learning model, the generation of the pollution source location probability map, and the acquisition of the location data of the pollution source include: S31. Based on the preliminary location data of the pollution source and the pollutant concentration collected in real time, construct a spatio-temporal grid input matrix; S32. According to the convolutional neural network and the time attention mechanism, design a dual-branch deep learning model, and use the dual-branch deep learning model to extract the spatial and time features of the pollutants in the spatio-temporal grid input matrix; S33. Fuse the spatial and time features of the pollutants, and inject the fusion result into the preset pollution source heat map through feature splicing and full connection layer compression to generate the pollution source location probability map; S34. Based on the generated pollution source location probability map and the Markov decision process, and combined with offline simulation training and greedy strategy, obtain the location data of the pollution source.

7. The method for identifying pollution sources in the wafer manufacturing process based on the AMC distribution between the inner layers of the FOUP according to claim 6, wherein, Fusing the spatial and temporal characteristics of the pollutants, and injecting the fusion result into a preset heat map of pollution sources through feature splicing and full connection layer compression to generate a pollution source location probability map, which includes: S331. Align the extracted spatial and temporal characteristics of the pollutants according to the newly deployed external detector nodes, and perform feature splicing to generate a joint feature matrix; S332. Input the generated joint feature matrix into a full connection layer, compress it using nonlinear dimensionality reduction technology, and extract feature components through low-rank matrix factorization technology to obtain a compressed feature representation; S333. Perform weighted fusion of the compressed feature representation with the preset heat map of pollution sources, and smooth the fusion result using Gaussian filtering to output a pollution source location probability map.

8. The method for identifying pollution sources in the wafer manufacturing process based on the AMC distribution between the inner layers of the FOUP according to claim 7, wherein, Based on the generated pollution source location probability map and Markov decision process, and combining offline simulation training and greedy strategy to obtain the location data of the pollution source, which includes: S341. Based on the Markov decision process, use the pollution source location probability map as the environmental state, the pollution source verification operation as the agent action, and combine the spatial and temporal characteristics of the pollutants to establish a state transition rule; S342. Construct a multi-objective reward function according to the established state transition rule; S343. Use the optimized digital twin model to generate a virtual pollution source scenario data set, construct a convolutional policy network based on the data set, and combine the multi-objective reward function to perform offline simulation training through the proximal policy optimization algorithm; S344. According to the action value output by the trained convolutional policy network, select the optimal pollution source verification operation using the greedy strategy, and update the pollution source location probability map in combination with the feedback data of the external detector; S345. Based on the updated pollution source location probability map, generate a verification path, and optimize the verification path in combination with the entropy threshold to obtain the location data of the pollution source.

9. The method for identifying pollution sources in the wafer manufacturing process based on the AMC distribution between the inner layers of the FOUP according to claim 8, wherein The multi-objective reward function includes: operation cost penalty, successful location heavy reward, information gain reward, probability density reward, and distance penalty.

10. The method for identifying pollution sources in the wafer manufacturing process based on the AMC distribution between the inner layers of the FOUP according to claim 1, wherein, Based on the location data of the pollution source, evaluate the contribution weights of each associated device to the pollutant diffusion, and formulate and execute a directional control strategy, which includes: S41. Combine the location data of the pollution source, the pollutant concentration collected in real time, and the operating parameters of the associated devices to construct a causal graph model; S42. Based on the constructed causal graph model, use causal inference technology to evaluate the influence degree of each associated device on the pollutant diffusion process, obtain the contribution weights, and formulate and execute a directional control strategy based on the contribution weights.

Images