A semiconductor clean room pollution intelligent monitoring and closed-loop control system and method
By using a smart cart equipped with a mass spectrometer and neural network technology in a semiconductor cleanroom, comprehensive, real-time, and accurate pollutant monitoring and closed-loop control were achieved, solving the problems of monitoring blind spots and response lag in existing technologies, and improving the stability of the production process and product yield.
Patent Information
- Application Number
- CN202611213079.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-11
- Publication Date
- 2026-11-06
AI Technical Summary
Existing environmental monitoring technologies for semiconductor cleanrooms cannot achieve comprehensive, real-time, and accurate monitoring and tracing of pollutants, resulting in delayed response to pollution incidents during production and affecting product yield.
The system employs a combination of mobile monitoring units, data sensing and transmission units, intelligent analysis units, and active control units. It utilizes a mass spectrometer mounted on an intelligent vehicle for real-time monitoring and employs GNN graph neural networks and LSTM neural networks for pollution source tracing and concentration prediction to achieve closed-loop control.
It achieves full coverage, real-time monitoring, and precise source tracing of pollutants in clean rooms, improving the efficiency of pollution incident handling and the stability of the production process, and reducing the risks of monitoring blind spots and response delays.
Smart Images
Figure CN123046209A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of semiconductor manufacturing and environmental monitoring technology, and in particular to a real-time monitoring system and method for volatile organic compounds (VOCs) and airborne molecular pollutants (AMCs) in semiconductor cleanrooms. Background Technology
[0002] Currently, the environmental monitoring system for semiconductor cleanrooms mainly adopts two traditional technologies: one is fixed-point sensors deployed in key areas, and the other is offline analysis after manual sample collection and delivery to the laboratory. While fixed-point sensors can provide continuous data, their coverage is limited, resulting in a large number of monitoring blind spots and an inability to capture dynamic contamination events throughout the cleanroom, especially inside production equipment and in airflow dead zones. On the other hand, the offline sampling and analysis process is cumbersome, often taking hours or even days from sampling and delivery to report generation, with extremely slow response times, making it impossible to provide timely early warnings for sudden contamination during the production process.
[0003] While existing monitoring technologies exist to address these issues, their limitations have led to a series of serious problems: First, traditional monitoring technologies cannot accurately trace the source of pollution incidents. When sensors alarm, engineers can only rely on experience to conduct a thorough search of the vast factory, resulting in low efficiency and poor accuracy. This forces many production machines to shut down, causing significant production losses. More importantly, as semiconductor processes advance to more advanced nodes such as 3nm and below, the requirements for the concentration of volatile organic compounds (VOCs) and metal pollutants in the environment have risen to extreme levels of ppb (parts per billion) or even ppt (parts per trillion). The detection accuracy, response speed, and source tracing capabilities of existing technologies can no longer meet these stringent requirements, causing some minute and instantaneous pollution to go undetected and unchecked, becoming a key factor affecting product yield. Summary of the Invention
[0004] To address the aforementioned problems, this invention provides an intelligent monitoring and closed-loop control system and method for contaminants in semiconductor cleanrooms, which enables intelligent monitoring that achieves full-area, real-time, and accurate source tracing and automatic control.
[0005] The present invention adopts the following technical solution: an intelligent monitoring and closed-loop control system for contaminants in a semiconductor cleanroom, comprising: a mobile monitoring unit, used to perform mobile inspections in the semiconductor cleanroom according to a preset inspection path or scheduling instructions, and to collect mass spectrometry monitoring data of volatile organic compounds and airborne molecular pollutants in the semiconductor cleanroom in real time. A data sensing and transmission unit, connected to the mobile monitoring unit, is used to receive the monitoring data collected by the mobile monitoring unit and transmit the monitoring data to the intelligent analysis unit in real time; The intelligent analysis unit, connected to the data sensing and transmission unit, is used to perform fusion analysis, pollution diagnosis, pollution source tracing and pollution prediction analysis on the received monitoring data, and generate control decisions corresponding to pollution events. An active control unit is connected to both the mobile monitoring unit and the intelligent analysis unit. It is used to control the mobile monitoring unit and the cleanroom environmental control equipment to perform corresponding control operations based on the control decisions generated by the intelligent analysis unit, and to complete closed-loop management of pollution events.
[0006] Furthermore, the mobile monitoring unit includes an intelligent vehicle and a mass spectrometer monitoring system installed on the intelligent vehicle; the intelligent vehicle adopts a clean AGV platform; The mass spectrometer monitoring system is mounted on the intelligent vehicle via a shock-absorbing platform and is used to collect real-time mass spectrometry data of volatile organic compounds and airborne molecular pollutants. Furthermore, the intelligent analysis unit includes a data fusion module, a pollution diagnosis module, a pollution source tracing module, and a predictive analysis module; The data fusion module is used to fuse the monitoring data collected by the mobile monitoring unit. The pollution diagnosis module is used to obtain the type and degree of pollution based on the fused data; The pollution source tracing module uses a GNN graph neural network to trace pollution sources and locate pollution locations. The predictive analysis module uses an LSTM neural network to predict the trend of pollutant concentration changes. A method for intelligent monitoring and closed-loop control of contaminants in a semiconductor cleanroom includes the following steps: S1. Real-time collection of pollutant data within the cleanroom via a mass spectrometer monitoring system mounted on an intelligent vehicle; S2. Determine if the pollutant concentration is abnormal. If it is abnormal, return to step S1. If it is normal, transmit the collected data to the intelligent analysis unit through the data sensing and transmission unit. S3. The intelligent analysis unit receives and integrates pollutant data, performs real-time diagnosis and source tracing analysis of pollution events, outputs the location coordinates and composition information of the pollution source, and obtains decision instructions based on the predicted pollutant diffusion trend. S4. The active control unit receives the decision instructions output by the intelligent analysis unit and executes the corresponding control operations, including confirming the movement of the intelligent vehicle, controlling and adjusting the clean room, or reporting maintenance information; then it returns to step S1 to form a complete closed-loop control.
[0007] Furthermore, in step S3, the intelligent analysis unit uses a GNN (Graph Neural Network) to trace the source of pollution and locate its location, including the following steps: S3.1 Construct the graphical structure of the cleanroom: G = (V, E); Where V represents the production equipment node; If E is an edge, and the spatial Euclidean distance between any two nodes is less than R, then an edge is established. R is the preset radius of influence; S3.2, Update the mass spectrometer concentration data by feature fusion: h v (l+1) = σ( Σ u∈N(v) α vu (l) W(l)h u (l) ); Where N(v) are the neighboring nodes of node V; W(l) is the trainable weight matrix in the l-th layer; σ is the activation function; α vu (l) is the attention coefficient, which represents the importance of node u to node v in message passing of the l-th layer; It is the feature vector of node u in the l-th layer; h v (l+1) It is the feature vector of node V in the (l+1)th layer; S3.3 Obtain the pollution source probability score for each device node according to the formula: p d =Softmax (MLP (h v (final) )); Among them, h v (final) It is the final embedding vector of node v; MLP is a multilayer perceptron, and its output is a scalar representing the probability that the node is a pollution source. S3.4, Take the pollution source probability score p d The largest node is then used to query the physical coordinates of that node that were pre-stored during graph construction, which is the location of the pollution source. Furthermore, the attention coefficient α is obtained. vu The calculation steps include the following: First, obtain the unnormalized attention score e. vu : e vu = LeakyReLU (aT [Wh v ||Wh u ]); Among them, a T It is a trainable attention weight transpose vector; Wh v It is the result of a linear transformation of node v; Wh u It is the result of a linear transformation of node u; LeakyReLU is an activation function; || indicates vector concatenation; Then, Softmax normalization is performed on all neighboring nodes u of node v: α vu =exp(e vu ) / Σ u∈N(v) exp(e vu ); Furthermore, predicting the diffusion trend of pollutant concentrations includes the following steps: Step 1: Obtain the concentration sequence c of each node v within the T time steps prior to the current time. t ; Step 2: Use a two-layer stacked LSTM network to obtain the 128-dimensional hidden layer states: h t =LSTM(x t ,h t-1 ,c t-1 ); Where, x t The input feature vector at sampling time t; Step 3: Obtain the predicted concentration sequence of node v for each node within a specified future time period: ; in, It is the VOC concentration value output by the LSTM network in the future Δth second; W out These are trainable parameters used to train the 128-dimensional hidden state h. t Mapped to a single number; h t It is the final 128-dimensional hidden state output by the LSTM network after T time steps; b is a trainable scalar parameter used to adjust the baseline value for predicting concentration; Step 4: Based on the predicted concentration sequence, obtain the location of the predicted concentration peak at each time point, connect them to form a trajectory, which is the predicted diffusion path, and the change of concentration at each node over time is the concentration change trend. Furthermore, the trend of concentration change was obtained, specifically: The slope is obtained from the formula: k = ; Δ: Values range from 1 to 30; The average of all 30 Δ values; The VOC concentration value in the future Δth second output by the LSTM network; The arithmetic mean of 30 predicted concentrations; If the slope is greater than 0, it indicates that the concentration is increasing; if the slope is less than 0, it indicates that the concentration is decreasing. Furthermore, decision instructions are generated based on concentration change trends, including: If the pollution source is located adjacent to sensitive equipment, and the VOC concentration value If so, it meets the requirements of Level I emergency, and a positive pressure air curtain should be established to isolate the cleanroom; If the location of the pollution source has been identified, it meets the Level II warning criteria, and the filtration speed of the fans in the contaminated area of the cleanroom should be increased. If VOC concentration value However, if the location of the pollution source cannot be determined, it meets the requirements for Level III monitoring. No adjustments will be made to the cleanroom, and the source tracing will continue.
[0008] The present invention also provides an electronic device, including a processor, a memory, and a computer program stored in the memory. When the processor executes the computer program, it implements the above-described method for intelligent monitoring and closed-loop control of contaminants in semiconductor cleanrooms.
[0009] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the above-described method for intelligent monitoring and closed-loop control of contaminants in a semiconductor cleanroom is implemented.
[0010] Compared with the prior art, the present invention has at least the following beneficial effects: (1) The present invention uses an intelligent vehicle equipped with a mass spectrometer monitoring system for mobile inspection to realize dynamic monitoring of pollutants in cleanrooms. Compared with fixed monitoring methods, it can expand the monitoring coverage, reduce monitoring blind spots, and improve the spatial monitoring capability of pollutants. (2) This invention uses GNN graph neural network to establish the spatial topology relationship of cleanroom equipment and combines it with real-time monitoring data to trace the source of pollution, which can realize the location of pollution source, improve the efficiency of pollution incident investigation, and provide a basis for equipment maintenance; (3) The present invention uses LSTM long short-term memory network to predict pollutant concentration time series data, which can obtain the pollutant concentration change trend in advance, provide predictive information for cleanroom pollution control, and improve the initiative of pollution event response; (4) This invention achieves closed-loop management of pollution monitoring, pollution analysis, pollution decision-making and pollution control by working together with the intelligent analysis unit and the active control unit, thereby improving the efficiency of cleanroom pollution incident handling and the level of automated management; (5) The present invention uses a mass spectrometer monitoring system installed on a smart car and combined with a shock-absorbing platform to reduce the impact of moving vibration and sampling adsorption on the detection results and improve the stability and real-time performance of monitoring data during the moving monitoring process. Attached Figure Description
[0011] Figure 1 This is a structural block diagram of the present invention; Figure 2 This is a flowchart of the present invention; Figure 3 This is a diagram showing the dynamic response test results of the mobile monitoring unit of this invention; Figure 4 This is a schematic diagram of the pollutant concentration prediction results of the present invention. Detailed Implementation
[0012] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be noted that the following embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the scope of protection of the present invention.
[0013] like Figures 1-4 As shown, the present invention provides an intelligent monitoring and closed-loop control system for contaminants in a semiconductor cleanroom, comprising a mobile monitoring unit, a data sensing and transmission unit, an intelligent analysis unit, and an active control unit; wherein, the mobile monitoring unit is used to perform mobile inspections in the semiconductor cleanroom according to a preset inspection path or scheduling instructions, and to collect mass spectrometry monitoring data of volatile organic compounds and airborne molecular pollutants in the cleanroom in real time. The mobile monitoring unit includes a smart car and a mass spectrometer monitoring system installed on the smart car; The intelligent vehicle uses a clean AGV platform that meets ISO Class 3 cleanliness requirements; The mass spectrometer monitoring system is used for real-time mass spectrometry data acquisition of volatile organic compounds (VOCs) and airborne molecular pollutants. Preferably, the system employs a photoionization time-of-flight mass spectrometer (SPI-TOFMS). This system can be mounted on an intelligent vehicle (AGV) using an existing vibration-damping platform, actively offsetting vibrations caused by AGV movement and operation. This reduces the impact of mechanical vibrations on the stability of mass spectrometry detection during AGV movement, ensuring the mass spectrometer maintains extremely high detection stability and accuracy during movement. This enables real-time, ppt-level mass spectrometry data acquisition of VOCs and AMCs during movement. Air samples enter the mass spectrometer via an inert capillary tube for online detection, reducing the adsorption and retention of pollutants in the sampling pipeline and improving real-time detection capabilities. The mass spectrometer monitoring system uses online time-of-flight mass spectrometry (OTF-MS), which can obtain the signal response of each VOC within 1 second. The sample introduction system employs passivated capillary technology to reduce sample adsorption and shorten the signal response time. Figure 3 The response curve of the mass spectrometer monitoring system is shown. After the standard gas is introduced, the signal intensity of the mass spectrometer monitoring system soars from 10% of the steady-state peak value to 90% of the steady-state peak value. The signal response time is only 0.879s, and the curve slope is extremely steep, which proves that the ion response speed is excellent. The dynamic response speed of mass spectrometry ionization, ion transport and detector is very fast, realizing the rapid monitoring of VOCs. In the diagram, T 10 This indicates the moment when the signal intensity of the mass spectrometer monitoring system reaches 10% of the steady-state peak value, representing the time point when the mass spectrometer begins to detect the analyte and the signal starts to rise; T 90 This indicates the moment when the signal strength reaches 90% of its steady-state peak value; like Figure 3 As shown in the figure, the dynamic response test of the mobile monitoring unit of the intelligent monitoring and closed-loop control system for contaminants in a semiconductor cleanroom of the present invention demonstrates that the system can complete online detection and response of contaminants within a single second timescale, achieving rapid capture of transient contamination events in the cleanroom environment. Due to the high temporal resolution of the mobile monitoring unit, it can effectively reduce data deviations caused by contaminant diffusion and spatial migration during mobile sampling, improve the accuracy of contaminant spatial distribution analysis, and provide timely data support for subsequent dynamic contaminant propagation modeling, contaminant source localization, and predictive-driven active control.
[0014] The data sensing and transmission unit is connected to the mobile monitoring unit and is used to receive the monitoring data collected by the mobile monitoring unit and transmit the monitoring data to the intelligent analysis unit in real time. The intelligent analysis unit, connected to the data sensing and transmission unit, is used to perform fusion analysis, pollution diagnosis, pollution source tracing and pollution prediction analysis on the received monitoring data, and generate control decisions corresponding to pollution events. The active control unit is connected to both the intelligent analysis unit and the mobile monitoring unit. It is used to control the mobile monitoring unit and the cleanroom environmental control equipment to perform corresponding control operations based on the control decisions generated by the intelligent analysis unit, and to complete the closed-loop management of pollution events. The active control unit is essentially the "executor," primarily responsible for performing a series of intervention operations based on instructions output by the intelligent analysis unit. Specifically, it receives instructions from the intelligent analysis unit and executes at least one of the following operations: Accurate Confirmation: The intelligent vehicle is guided to the suspected location of the pollution source for close-range retesting and confirmation to avoid false alarms; Environmental control: The HVAC (heating, ventilation and air conditioning) system of the cleanroom is controlled through standard industrial interfaces (such as OPC UA, Modbus) to adjust the air pressure and airflow speed in specific areas in order to suppress the spread of pollutants; Enhanced purification: Control the FFU (fan filter unit) group to adjust its operating status and enhance local purification capabilities; Information reporting: Automatically sends alarm information to the plant monitoring system and accurately identifies pollutants suspected of exceeding the standard and their locations.
[0015] The intelligent analysis unit includes a data fusion module, a pollution diagnosis module, a pollution source tracing module, and a predictive analysis module; The data fusion module is used to fuse and process the monitoring data collected by the mobile monitoring unit; The pollution diagnosis module is used to obtain the type and degree of pollution based on the fused data; The pollution source tracing module uses a graph neural network (GNN) to trace pollution sources and locate pollution locations. It can integrate and model the physical spatial structure of the cleanroom (such as equipment layout and airflow field) with real-time monitoring data, and trace pollution events back to specific emission equipment or locations. The positioning accuracy can reach within 0.5 meters, and accurately identify the components of pollutants. The predictive analysis module uses an LSTM neural network to predict the trend of pollutant concentration changes, enabling the analysis of time-series data on pollutant concentrations and predicting their future diffusion paths and concentration change trends, thus achieving a leap from "post-event source tracing" to "pre-event prediction".
[0016] A method for intelligent monitoring and closed-loop control of contaminants in a semiconductor cleanroom includes the following steps: S1. Real-time collection of pollutant data within the cleanroom via a mass spectrometer monitoring system mounted on an intelligent vehicle; S2. Determine if the pollutant concentration is abnormal. If it is abnormal, return to step S1. If it is normal, transmit the collected data to the intelligent analysis unit through the data sensing and transmission unit. S3. The intelligent analysis unit receives and integrates pollutant data, performs real-time diagnosis and source tracing analysis of pollution events, outputs the location coordinates and composition information of the pollution source, and obtains decision instructions based on the predicted pollutant diffusion trend. In step S3, the intelligent analysis unit uses a GNN (Graph Neural Network) to trace the source of pollution and locate its location, including the following steps: S3.1 Construct the graphical structure of the cleanroom: G = (V, E); Where V represents the production equipment node; If E is an edge, and the spatial Euclidean distance between any two nodes is less than R, then an edge is established. R is the preset radius of influence; S3.2, Update the mass spectrometer concentration data by feature fusion: h v (l+1) = σ( Σ u∈N(v) α vu (l) W(l) h u (l) ); Where N(v) are the neighboring nodes of node V; W(l) is the trainable weight matrix in the l-th layer; σ is the activation function (such as ReLU); α vu (l) is the attention coefficient, which represents the importance of node u to node v in message passing of the l-th layer; It is the feature vector of node u in the l-th layer; h v (l+1) It is the feature vector of node V in the (l+1)th layer; S3.3 Obtain the pollution source probability score for each device node according to the formula: p d =Softmax (MLP (h v (final) )); Among them, h v (final) It is the final embedding vector of node v; MLP is a multilayer perceptron, and its output is a scalar representing the probability that the node is a pollution source. S3.4, Take the pollution source probability score p d The largest node is then used to query the physical coordinates of that node that were pre-stored during graph construction, which is the location of the pollution source. Furthermore, the attention coefficient α is obtained. vu The calculation steps include the following: First, obtain the unnormalized attention score e. vu : e vu = LeakyReLU (a T [Wh v ||Wh u ]); Among them, a T It is a trainable attention weight transpose vector; Wh v It is the result of a linear transformation of node v; Wh u It is the result of a linear transformation of node u; LeakyReLU is an activation function; || indicates vector concatenation; Then, Softmax normalization is performed on all neighboring nodes u of node v: α vu =exp(e vu ) / Σ u∈N(v) exp(e vu ); By integrating pollutant composition information from mass spectrometry detection with the spatial topology information of the cleanroom, compared with the traditional method of judging solely based on concentration gradient, it can effectively avoid positioning deviations caused by airflow disturbances due to local turbulence or personnel movement, and significantly improve the accuracy of source tracing in complex clean environments.
[0017] After locating the pollution source, the system also needs to predict the diffusion trend of pollutants to assist the active control unit in making scientific decisions. This embodiment uses a two-layer stacked long short-term memory (LSTM) network to predict the time-series changes in VOC concentration at each monitoring node in the clean room.
[0018] Furthermore, predicting the diffusion trend of pollutant concentrations includes the following steps: Step 1: Obtain the concentration sequence {c} of each node v within T (e.g., T=60) time steps (1 second per step) prior to the current time. t-59 , ..., c t}; Step 2: Use a two-layer stacked LSTM network to obtain the 128-dimensional hidden layer states: h t =LSTM(x t ,h t-1 ,c t-1 ) Where, x t The input feature vector at sampling time t; h t-1 It is the hidden state at time t-1; Step 3: Obtain the predicted concentration sequence of node v within a future set time period (e.g., 30 seconds):
[0019] in, It is the VOC concentration value output by the LSTM network in the future Δth second; W out These are trainable parameters used to train the 128-dimensional hidden state h. t Mapped to a single number; h tIt is the final 128-dimensional hidden state output by the LSTM network after T time steps; b is a trainable scalar parameter used to adjust the baseline value for predicting concentration; Step 4: Based on the predicted concentration sequence, obtain the location of the predicted concentration peak at each time point, connect them to form a trajectory, which is the predicted diffusion path, and the change of concentration at each node over time is the concentration change trend. Furthermore, the trend of concentration change was obtained, specifically: The slope is obtained from the formula: k = ; Δ: Values from 1 to 30, representing the future second number; The average of all 30 Δ values, i.e., (1+2+...+30) / 30 = 15.5; The VOC concentration value in the future Δth second output by the LSTM network; The arithmetic mean of 30 predicted concentrations; If k > 0, it indicates that the concentration is increasing; if k < 0, it indicates that the concentration is decreasing. Furthermore, decision instructions are generated based on concentration change trends, including: If the pollution source is located adjacent to sensitive equipment, and the VOC concentration value If so, it meets the requirements of Level I emergency, and a positive pressure air curtain should be established to isolate the cleanroom; If the location of the pollution source has been identified, it meets the Level II warning criteria, and the filtration speed of the fans in the contaminated area of the cleanroom should be increased. If VOC concentration value However, if the location of the pollution source cannot be determined, it meets the requirements of Level III monitoring. No adjustments will be made to the cleanroom, and the source tracing will continue with more intensive sampling. like Figure 4 As shown, the predicted curve obtained through experimental testing has a high degree of consistency with the actual detection curve, verifying that the LSTM prediction model can predict the trend of cleanroom contamination in advance.
[0020] S4. The active control unit receives the decision instructions output by the intelligent analysis unit and executes the corresponding control operations, including confirming the movement of the intelligent vehicle, controlling and adjusting the clean room, or reporting maintenance information. Then, it records the entire process of the control method event, updates the data of the GNN graph neural network and LSTM neural network, and returns to step S1 to form a complete closed-loop control.
[0021] The present invention also provides an electronic device, including a processor, a memory, and a computer program stored in the memory. When the processor executes the computer program, it implements the above-described method for intelligent monitoring and closed-loop control of contaminants in semiconductor cleanrooms.
[0022] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the above-described method for intelligent monitoring and closed-loop control of contaminants in a semiconductor cleanroom is implemented.
[0023] During system operation, the mobile monitoring unit continuously acquires the pollution status of the cleanroom space, the data sensing and transmission unit updates environmental data in real time, the intelligent analysis unit continuously optimizes the pollution propagation model and prediction model, and the active control unit dynamically adjusts the clean environment parameters based on the analysis results, thereby forming a continuously operating intelligent closed-loop management system. Specifically, this invention uses an intelligent vehicle equipped with a high-performance mass spectrometer monitoring system to achieve full-area, blind-spot-free monitoring of the cleanroom, completely eliminating monitoring blind spots. Secondly, through the established intelligent analysis unit, a minute-level rapid response from pollution detection to precise source tracing is achieved, transforming passive alarms into proactive predictions. Ultimately, a complete closed-loop control logic can be formed, which can automatically link with the plant management system to suppress and isolate pollution, thereby significantly improving the environmental control level and production yield of advanced semiconductor production lines, fundamentally solving the problems of lag and passivity in control.
[0024] The intelligent monitoring and closed-loop control system provided by this invention has the following significant advantages and positive effects: 1. Through the high-precision mass spectrometer monitoring system mounted on the intelligent vehicle, the system of this invention can conduct full-area patrol monitoring of the cleanroom according to the preset path or task instructions, completely covering the airflow dead corners and dense equipment areas that existing fixed sensors cannot reach, which can expand the spatial coverage of pollution detection, reduce the number of fixed detection points, improve the flexibility of cleanroom pollution monitoring, and truly realize the construction of a high-resolution spatial pollution map without dead corners. 2. By integrating spatial relationships and pollution propagation patterns using a GNN model, precise location of pollution sources can be achieved; 3. Predict future trends of pollutants using LSTM models to achieve early warning of pollution risks; 4. By combining the three-level control strategy from Level I to Level III, the cleanroom environmental parameters can be actively adjusted to improve the stability of the semiconductor manufacturing process and the product yield.
[0025] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.
[0026] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.
Claims
1. A smart monitoring and closed-loop control system for contaminants in a semiconductor cleanroom, characterized in that: include: The mobile monitoring unit is used to conduct mobile inspections in the semiconductor cleanroom according to a preset inspection path or scheduling instructions, and to collect mass spectrometry monitoring data of volatile organic compounds and airborne molecular pollutants in the semiconductor cleanroom in real time. A data sensing and transmission unit, connected to the mobile monitoring unit, is used to receive the monitoring data collected by the mobile monitoring unit and transmit the monitoring data to the intelligent analysis unit in real time; The intelligent analysis unit, connected to the data sensing and transmission unit, is used to perform fusion analysis, pollution diagnosis, pollution source tracing and pollution prediction analysis on the received monitoring data, and generate control decisions corresponding to pollution events. An active control unit is connected to both the mobile monitoring unit and the intelligent analysis unit. It is used to control the mobile monitoring unit and the cleanroom environmental control equipment to perform corresponding control operations based on the control decisions generated by the intelligent analysis unit, and to complete closed-loop management of pollution events.
2. The intelligent monitoring and closed-loop control system for contaminants in a semiconductor cleanroom according to claim 1, characterized in that: The mobile monitoring unit includes an intelligent vehicle and a mass spectrometer monitoring system installed on the intelligent vehicle; the intelligent vehicle adopts a clean AGV platform; The mass spectrometer monitoring system is mounted on the intelligent vehicle via a shock-absorbing platform and is used to collect real-time mass spectrometry data of volatile organic compounds and airborne molecular pollutants.
3. The intelligent monitoring and closed-loop control system for contaminants in a semiconductor cleanroom according to claim 1, characterized in that: The intelligent analysis unit includes a data fusion module, a pollution diagnosis module, a pollution source tracing module, and a predictive analysis module; wherein, The data fusion module is used to fuse the monitoring data collected by the mobile monitoring unit. The pollution diagnosis module is used to obtain the type and degree of pollution based on the fused data; The pollution source tracing module uses a GNN graph neural network to trace pollution sources and locate pollution locations. The predictive analysis module uses an LSTM neural network to predict the trend of pollutant concentration changes.
4. A method for intelligent monitoring and closed-loop control of contaminants in a semiconductor cleanroom, characterized in that: Includes the following steps: S1. Real-time collection of pollutant data within the cleanroom via a mass spectrometer monitoring system mounted on an intelligent vehicle; S2. Determine if the pollutant concentration is abnormal. If it is abnormal, return to step S1. If it is normal, transmit the collected data to the intelligent analysis unit through the data sensing and transmission unit. S3. The intelligent analysis unit receives and integrates pollutant data, performs real-time diagnosis and source tracing analysis of pollution events, outputs the location coordinates and composition information of the pollution source, and obtains decision instructions based on the predicted pollutant diffusion trend. S4. The active control unit receives the decision instructions output by the intelligent analysis unit and executes the corresponding control operations, including controlling the movement of the intelligent vehicle for confirmation, controlling and adjusting the clean room, or reporting maintenance information. Then return to step S1 to form a complete closed-loop control.
5. The method for intelligent monitoring and closed-loop control of contaminants in a semiconductor cleanroom according to claim 4, characterized in that: In step S3, the intelligent analysis unit uses a GNN (Graph Neural Network) to trace the source of pollution and locate its location, including the following steps: S3.1 Construct the graphical structure of the cleanroom: G = (V, E); Where V represents the production equipment node; If E is an edge, and the spatial Euclidean distance between any two nodes is less than R, then an edge is established. R is the preset radius of influence; S3.2, Update the mass spectrometer concentration data by feature fusion: h v (l+1) = σ( Σ u∈N(v) α vu (l) W(l)h u (l) ); Where N(v) are the neighboring nodes of node V; W(l) is the trainable weight matrix in the l-th layer; σ is the activation function; α vu (l) is the attention coefficient, which represents the importance of node u to node v in message passing of the l-th layer; It is the feature vector of node u in the l-th layer; h v (l+1) It is the feature vector of node V in the (l+1)th layer; S3.3 Obtain the pollution source probability score for each device node according to the formula: p d =Softmax (MLP (h v (final) )); Among them, h v (final) It is the final embedding vector of node v; MLP is a multilayer perceptron, and its output is a scalar representing the probability that the node is a pollution source. S3.4, Take the pollution source probability score p d The largest node is then identified by querying its pre-stored physical coordinates during graph construction, which indicate the location of the pollution source.
6. The method for intelligent monitoring and closed-loop control of contaminants in a semiconductor cleanroom according to claim 5, characterized in that: Obtain the attention coefficient α vu The calculation steps include the following: First, obtain the unnormalized attention score e. vu : e vu = LeakyReLU (a T [Wh v ||Wh u ]); Among them, a T It is a trainable attention weight transpose vector; Wh v It is the result of a linear transformation of node v; Wh u It is the result of a linear transformation of node u; LeakyReLU is an activation function; || indicates vector concatenation; Then, Softmax normalization is performed on all neighboring nodes u of node v: α vu =exp(e vu ) / Σ u∈N(v) exp(e vu )。 7. The method for intelligent monitoring and closed-loop control of contaminants in a semiconductor cleanroom according to claim 4, characterized in that: Predicting pollutant concentration diffusion trends involves the following steps: Step 1: Obtain the concentration sequence c of each node v within the T time steps prior to the current time. t ; Step 2: Use a two-layer stacked LSTM network to obtain the 128-dimensional hidden layer states: h t =LSTM(x t ,h t-1 ,c t-1 ); Where, x t The input feature vector at sampling time t; Step 3: Obtain the predicted concentration sequence of node v for each node within a specified future time period: ； in, It is the VOC concentration value output by the LSTM network in the future Δth second; W out These are trainable parameters used to train the 128-dimensional hidden state h. t Mapped to a single number; h t It is the final 128-dimensional hidden state output by the LSTM network after T time steps; b is a trainable scalar parameter used to adjust the baseline value for predicting concentration; Step 4: Based on the predicted concentration sequence, obtain the location of the predicted concentration peak at each time point, connect them to form a trajectory, which is the predicted diffusion path, and the change of concentration at each node over time is the concentration change trend.
8. The method for intelligent monitoring and closed-loop control of contaminants in a semiconductor cleanroom according to claim 7, characterized in that: The concentration change trend was obtained, specifically: The slope is obtained from the formula: k = ; Δ: Values range from 1 to 30; The average of all 30 Δ values; The VOC concentration value in the future Δth second output by the LSTM network; The arithmetic mean of 30 predicted concentrations; If the slope is greater than 0, it indicates that the concentration is increasing; if the slope is less than 0, it indicates that the concentration is decreasing.
9. The method for intelligent monitoring and closed-loop control of contaminants in a semiconductor cleanroom according to claim 7, characterized in that: Decision instructions are generated based on concentration change trends, including: If the pollution source is located adjacent to sensitive equipment, and the VOC concentration value If so, it meets the requirements of Level I emergency, and a positive pressure air curtain should be established to isolate the cleanroom; If the location of the pollution source has been identified, it meets the Level II warning criteria, and the filtration speed of the fans in the contaminated area of the cleanroom should be increased. If VOC concentration value However, if the location of the pollution source cannot be determined, it meets the requirements for Level III monitoring. No adjustments will be made to the cleanroom, and the source tracing will continue.
10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the intelligent monitoring and closed-loop control method for contaminants in semiconductor cleanrooms as described in any one of claims 4 to 9.
11. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the intelligent monitoring and closed-loop control method for contaminants in semiconductor cleanrooms as described in any one of claims 4 to 9.