Three-dimensional flow field simulation method and system based on clean air conditioner

Abstract

The invention provides a three-dimensional flow field simulation method and system based on a clean air conditioner. The method comprises the following steps: constructing a grid historical database of an electronic plant; according to the grid historical database, an AI pre-judgment model is trained, and the AI pre-judgment model is based on a random forest algorithm superposition space attention mechanism; obtaining dynamic scene features of the electronic plant, inputting the dynamic scene features into the trained AI pre-judgment model, and outputting high-risk area coordinates and partition grid parameters; then generating a three-layer multi-scale grid of a macroscopic layer, a mesoscopic layer and a microscopic layer, and coupling and transmitting data by adopting an overlapping grid technology to obtain a grid; the method comprises the following steps: performing independence verification on a grid, and outputting a final grid file, specifically, through AI learning and a multi-scale technology, realizing distribution of grid resources according to needs, not only solving the contradiction that the traditional grid is too dense to waste computing power and too sparse to be insufficient in precision, but also capturing the association of microscopic airflow and macroscopic airflow.

Description

Technical Field

[0001] This invention belongs to the field of three-dimensional flow field simulation technology, and specifically relates to a three-dimensional flow field simulation method and system based on cleanroom air conditioning. Background Technology

[0002] Cleanroom air conditioning systems are typically installed in electronics factories to create and maintain a specific clean environment.

[0003] Typically, computational fluid dynamics (CFD) simulations, or three-dimensional flow field simulations, are performed on the clean areas of electronic manufacturing plants. On the one hand, the results of the three-dimensional flow field simulation can be connected to the cleanroom digital twin platform to achieve real-time linkage between CFD (Computational Fluid Dynamics) and digital twin. On the other hand, AR visualization can be used to guide construction and operation and maintenance.

[0004] However, the cleanroom of an electronics factory has extremely high requirements for the accuracy of CFD simulation. Traditional mesh generation relies on general scene databases and lacks characteristic data specific to electronics factories, such as the size of lithography machines, the layout of FFUs (fan filter units), and micro-vibration sources. This leads to blind mesh densification in high-risk areas (such as around the lithography machine workbench), resulting in insufficient accuracy or wasted computing power. Summary of the Invention

[0005] Based on this, the present invention provides a three-dimensional flow field simulation method and system based on cleanroom air conditioning, which aims to solve the problems in the prior art where relying on experience to refine the mesh leads to waste of computing power due to excessively dense meshes and insufficient accuracy due to excessively sparse meshes.

[0006] A first aspect of this invention provides a three-dimensional flow field simulation method based on cleanroom air conditioning, the method comprising: Construct a grid history database for electronic manufacturing plants. The grid history database includes scene characteristics, grid schemes, and error feedback for three types of scenarios: lithography machine workshop, wafer packaging workshop, and FFU dense area. Based on the grid historical database, an AI prediction model is trained, which is based on the random forest algorithm with spatial attention mechanism; The dynamic scene features of the electronic factory are obtained and input into the trained AI prediction model to output the coordinates of the high-risk area and the partition grid parameters. Based on the coordinates of the high-risk area, the partitioned grid parameters, and the accuracy requirements, a three-layer multi-scale grid of macroscopic, mesoscopic, and microscopic layers is generated. Overlapping grid technology is used to couple and transfer data to obtain the grid. The microscopic layer covers the lithography machine table surface, HEPA fiber gaps, and wafer surface. The independence of the mesh is verified, and the final mesh file is output.

[0007] Furthermore, the scene features include the number of lithography machines, the size of the lithography machines, the number of FFUs, the spacing between FFUs, the air volume of FFUs, the coordinates of the vacuum pumps, the vibration frequency of the vacuum pumps, the width of the wafer transport track, the coordinates of the VOCs pollution sources, the release rate of the VOCs pollution sources, the length and width of the clean area, and the floor height. The grid scheme includes the coordinates of high-risk areas, partitioned grid parameters, and the number of grid layers adapted to the vibration source. The error feedback includes the deviation between the simulated and measured wind speed on the lithography machine's worktable, the simulation error of VOCs concentration, and the time consumed in micro-vibration coupling calculation.

[0008] Furthermore, the macroscopic layer refers to the entire cleanroom, and the mesh is generated using ANSYS Meshing; The meso-layer consists of the 10m perimeter of the lithography machine and the FFU array area, with the mesh generated using ICEMCFD. The microlayer consists of the 0.1-0.5m gap between the lithography machine stage and the HEPA fiber, and the mesh is generated using Palabos+LAMMPS.

[0009] Furthermore, in the step of generating a three-layer multi-scale grid of macroscopic, mesoscopic, and microscopic layers based on the coordinates of the high-risk area, the partition grid parameters, and the accuracy requirements, and coupling and transmitting data using overlapping grid technology to obtain the grid, adjacent grid layers are intersected at the boundary to form a data transition zone, wherein the width of the intersecting region is the product of the prediction multiple and the size of the outer grid.

[0010] Furthermore, in the step of generating a three-layer multi-scale grid (macro, meso, and micro) based on the coordinates of the high-risk area, the partitioned grid parameters, and accuracy requirements, and coupling and transmitting data using overlapping grid technology to obtain the grid, the data transmission in the three-layer multi-scale grid is bidirectional feedback. Specifically, the macro layer transmits the main duct wind speed and FFU inlet pressure to the meso layer, the meso layer transmits the airflow velocity around the lithography machine and the vacuum pump vibration frequency to the micro layer, and the micro layer transmits the wafer surface particle concentration and HEPA interception efficiency to the meso layer.

[0011] Furthermore, the step of obtaining the predicted multiple includes: A feature library for the impact of multiples is constructed. The feature library includes physical field gradient features, geometric complexity features, and error-sensitive features. The physical field gradient features include wind speed gradient and vibration frequency. The geometric complexity features include boundary curvature and mesh orthogonality. The error-sensitive features include historical simulation errors and cleanliness level. A multiple prediction model is constructed based on the random forest regression algorithm. The multiple prediction model is used to input the physical field gradient characteristics, geometric complexity characteristics, and error sensitivity characteristics, and outputs an initial prediction multiple. The optimization objective is to make the simulation error corresponding to the model prediction multiple less than a preset value, wherein the wind speed deviation is less than 2% and the HEPA efficiency deviation is less than 0.5%. The actual physical field gradient characteristics, geometric complexity characteristics, and error sensitivity characteristics are obtained and input into the trained multiple prediction model. The initial prediction multiple is output, and the multiple is dynamically corrected by real-time monitoring of coupling accuracy to obtain the predicted multiple. Specifically, real-time monitoring indicators include wind speed deviation, HEPA efficiency deviation, and vibration transmission error. A first mapping relationship is established between the condition of indicators exceeding the standard and the multiple adjustment strategy, and a second mapping relationship is established between the condition of indicators not exceeding the standard and the multiple adjustment strategy. Based on the real-time monitoring indicators, the first mapping relationship, and the second mapping relationship, the corresponding multiple adjustment strategy is output to complete the multiple correction.

[0012] Furthermore, based on the grid historical database, a process timing dimension is added, and a timing prediction model is constructed. The timing prediction model learns the correlation between process timing and grid requirements, and generates corresponding grid plans in advance for each process stage.

[0013] A second aspect of this invention provides a three-dimensional flow field simulation system based on cleanroom air conditioning, used to implement the three-dimensional flow field simulation method based on cleanroom air conditioning described in the first aspect, the system comprising: The construction module is used to build a grid history database for electronic factories. The grid history database includes scene characteristics, grid schemes, and error feedback for three types of scenarios: lithography machine workshop, wafer packaging workshop, and FFU dense area. The training module is used to train an AI prediction model based on the grid historical database. The AI ​​prediction model is based on a random forest algorithm with a spatial attention mechanism. The input module is used to acquire the dynamic scene features of the electronic factory and input the dynamic scene features into the trained AI prediction model, and output the coordinates of the high-risk area and the partition grid parameters. The coupling module is used to generate a three-layer multi-scale mesh of macroscopic, mesoscopic and microscopic layers according to the coordinates of the high-risk area, the partition mesh parameters and the accuracy requirements. The overlapping mesh technology is used to couple and transmit data to obtain the mesh. The microscopic layer covers the lithography machine table surface, HEPA fiber gaps and wafer surface. The adjustment module is used to perform independence verification on the mesh and output the final mesh file.

[0014] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the three-dimensional flow field simulation method based on cleanroom air conditioning provided in the first aspect.

[0015] A fourth aspect of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the three-dimensional flow field simulation method based on cleanroom air conditioning provided in the first aspect.

[0016] This invention provides a three-dimensional flow field simulation method and system based on cleanroom air conditioning. It constructs a historical grid database for an electronics factory, containing scene features, grid schemes, and error feedback for three types of scenarios: lithography machine workshops, wafer packaging workshops, and FFU (Fan Filter Unit) dense areas. Based on this database, an AI prediction model is trained, using a random forest algorithm with a spatial attention mechanism. Dynamic scene features of the electronics factory are acquired and input into the trained AI prediction model, outputting high-risk area coordinates and partitioned grid parameters. Based on the high-risk area coordinates, partitioned grid parameters, and accuracy requirements, a three-layer multi-scale grid (macro, meso, and micro) is generated. Overlapping grid technology is used to couple and transfer data to obtain the grid. The micro layer covers the lithography machine worktable, HEPA fiber gaps, and wafer surface. The grid's independence is verified, and the final grid file is output. Specifically, through AI learning and multi-scale technology, grid resources are allocated on demand, resolving the contradiction between excessively dense grids wasting computing power and excessively sparse grids lacking accuracy, while also capturing the correlation between micro and macro airflow. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the implementation of a three-dimensional flow field simulation method based on cleanroom air conditioning, as provided in Embodiment 1 of the present invention. Figure 2 This is a structural block diagram of a three-dimensional flow field simulation system based on cleanroom air conditioning, provided in Embodiment 3 of the present invention; Figure 3 This is a structural block diagram of an electronic device provided in Embodiment 4 of the present invention. Detailed Implementation

[0018] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Several embodiments of the invention are illustrated in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

[0019] It should be noted that when a component is said to be "fixed to" another component, it can be directly on the other component or there may be an intervening component. When a component is said to be "connected to" another component, it can be directly connected to the other component or there may be an intervening component. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this document are for illustrative purposes only.

[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0021] Example 1 According to an embodiment of the present invention, a three-dimensional flow field simulation method based on cleanroom air conditioning is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0022] This embodiment provides a three-dimensional flow field simulation method based on cleanroom air conditioning, which can be used in electronic devices, such as computers. To achieve three-dimensional flow field simulation, the overall steps are as follows: First, geometric modeling is performed, transforming the physical space into a dynamic digital skeleton; then, mesh generation is performed, allocating precise computing resources to the digital skeleton; a physical model is constructed, injecting actual operating rules into the digital skeleton; boundary conditions are determined, and the solution is dynamically optimized and iterated based on these rules; the results are applied, transferring the iterative results to the physical space while simultaneously correcting deviations in the digital space to obtain the final simulation results. However, the mesh generation step often relies on experience for mesh refinement. Traditional mesh refinement uses fixed sizes; for example, for areas with large airflow gradients in cleanroom air conditioning, the mesh size is reduced to 1 / 3-1 / 5 of that in conventional areas, leading to a waste of computing power.

[0023] For this purpose, please refer to Figure 1 , Figure 1 The flowchart of a three-dimensional flow field simulation method based on cleanroom air conditioning provided in Embodiment 1 of the present invention is shown, specifically including steps S01 to S05.

[0024] Step S01: Construct a grid history database for the electronic factory. The grid history database includes scene characteristics, grid schemes, and error feedback for three types of scenarios: lithography machine workshop, wafer packaging workshop, and FFU dense area.

[0025] Specifically, several sets of cleanroom grid data for electronics manufacturing plants were first collected, covering three scenarios: lithography machine workshop, wafer packaging workshop, and FFU (Fan Filter Unit) dense area. For example, the lithography machine workshop scenario includes the layout of single / multiple lithography machines and wafer handling robot tracks; the wafer packaging workshop scenario includes multi-station packaging equipment and local VOCs (Volatile Organic Compounds) pollution sources; and the FFU dense area scenario includes ≥60 FFUs / m³. 2 Ultra-clean areas, such as the photoresist coating area.

[0026] Based on each set of grid data mentioned above, scene features, grid scheme, and error feedback are extracted. The scene features include the number of lithography machines, the size of the lithography machines, the number of FFUs, the spacing between FFUs, the air volume of FFUs, the coordinates of the vacuum pumps, the vibration frequency of the vacuum pumps, the width of the wafer transport track, the coordinates of the VOCs pollution sources, the release rate of the VOCs pollution sources, the length and width of the clean area, and the floor height. The grid scheme includes the coordinates of high-risk areas, partitioned grid parameters, and the number of grid layers adapted to the vibration source. The error feedback includes the deviation between the simulated and measured wind speed on the lithography machine's worktable, the simulation error of VOCs concentration, and the time consumed in micro-vibration coupling calculation.

[0027] Step S02: Train an AI prediction model based on the grid historical database. The AI ​​prediction model is based on the random forest algorithm with spatial attention mechanism.

[0028] In this embodiment of the invention, the PyTorch framework is used to build an AI prediction model. The spatial feature weights of the area surrounding the electronic factory equipment and the FFU area are set to twice that of the regular area. In addition, the input and output variables of the AI ​​prediction model are defined.

[0029] Understandably, the input variables are the number of lithography machines, the size of the lithography machines, the number of FFUs, the spacing between FFUs, the air volume of FFUs, the coordinates of the vacuum pump, the vibration frequency of the vacuum pump, the width of the wafer transport track, the coordinates of the VOCs pollution source, the release rate of the VOCs pollution source, the length, width and height of the clean area, and the output variables are the coordinates of the high-risk area, the zoning grid parameters and the number of vibration source adaptation grid layers.

[0030] For example, the coordinates of the high-risk area may include the area around the lithography machine stage (0.5-1m), the FFU airflow intersection area, and the area within 1m of the vacuum pump; the partition grid parameters are 5-10mm around the lithography machine, 20-30mm in the FFU area, 8mm around the vacuum pump, and 50-80mm in the normal area; the vibration source adaptation grid layer number is 8-12 layers around the vacuum pump.

[0031] Step S03: Obtain the dynamic scene features of the electronic factory, input the dynamic scene features into the trained AI prediction model, and output the coordinates of the high-risk area and the partition grid parameters.

[0032] In this embodiment of the invention, the real-time position of the lithography machine and the operating status of the FFU are obtained from the MES system of the electronics factory, and the CAD drawings of the clean area are obtained from the geometric model. The drawings are parsed using Python and OpenCV to extract scene features. The obtained scene features are input into a trained AI prediction model, which outputs the coordinates of high-risk areas and partition grid parameters, and selects a mesh tool based on the partition grid parameters.

[0033] It should be noted that the periphery of the lithography machine uses ICEMCFD to generate unstructured tetrahedral meshes; the FFU region uses ANSYS Meshing to generate structured hexahedral meshes; and the vacuum pump vibration zone uses ANSYS Meshing to generate prism layer meshes.

[0034] Step S04: Based on the coordinates of the high-risk area, the partitioned grid parameters, and the accuracy requirements, a three-layer multi-scale grid consisting of a macroscopic layer, a mesoscopic layer, and a microscopic layer is generated. Overlapping grid technology is used to couple and transfer data to obtain the grid. The microscopic layer covers the lithography machine worktable, the HEPA fiber gaps, and the wafer surface.

[0035] First, the macro layer, meso layer, and micro layer in the electronics factory are defined. The macro layer is the entire cleanroom, and ANSY SMeshing is used to generate a structured grid. The meso-layer consists of the 10m perimeter of the lithography machine and the FFU array area, with the mesh generated using ICEMCFD. The microlayer consists of the 0.1-0.5m gap between the lithography stage and the HEPA (High Efficiency Particulate Air Filter) fiber. The mesh is generated using Palabos+LAMMPS, where Palabos generates the LBM mesh and LAMMPS generates the molecular dynamics mesh. Slip boundaries are provided on the wafer surface.

[0036] After defining the macro, meso, and micro layers, corresponding grids are formed in the macro, meso, and micro layers based on the output of the previous AI prediction model. Furthermore, overlapping grid technology is used to couple and transfer data. Specifically, adjacent grids are crossed and overlapped at the boundary to form a data transition zone. The width of the cross-overlapping zone is the product of the prediction multiple and the size of the outer grid.

[0037] It should be noted that the steps for obtaining the predicted multiple include: A feature library for the impact of multiples is constructed. This feature library includes physical field gradient features, geometric complexity features, and error-sensitive features. The physical field gradient features include wind speed gradient and vibration frequency. The geometric complexity features include boundary curvature and mesh orthogonality. The error-sensitive features include historical simulation errors and cleanliness level. Specifically, vibration frequency refers to the vibration frequency of the vacuum pump / lithography machine, boundary curvature refers to the radius of curvature of the equipment surface, mesh orthogonality refers to the mesh distortion at the boundary, historical simulation errors refer to the past wind speed deviation in this area, and cleanliness level refers to ISO Class 3 / 5 / 6. A multiple prediction model is constructed based on the random forest regression algorithm. This model takes into account the physical field gradient features, geometric complexity features, and error sensitivity features, and outputs an initial prediction multiple, i.e., the overlap region multiple. The optimization objective is to ensure that the simulation error corresponding to the model's prediction multiple is less than a preset value, specifically, wind speed deviation less than 2% and HEPA efficiency deviation less than 0.5%. In this embodiment, Python + Scikit-learn is used, and the number of decision trees (100) is adjusted through 5-fold cross-validation to ensure a prediction accuracy ≥ 92%. The actual physical field gradient characteristics, geometric complexity characteristics, and error sensitivity characteristics are obtained and input into the trained multiple prediction model. The initial prediction multiple is output, and the multiple is dynamically corrected by real-time monitoring of coupling accuracy to obtain the predicted multiple. Specifically, real-time monitoring indicators include wind speed deviation, HEPA efficiency deviation, and vibration transmission error. A first mapping relationship is established between the condition of indicators exceeding the standard and the multiple adjustment strategy, and a second mapping relationship is established between the condition of indicators not exceeding the standard and the multiple adjustment strategy. Based on the real-time monitoring indicators, the first mapping relationship, and the second mapping relationship, the corresponding multiple adjustment strategy is output to complete the multiple correction.

[0038] For example, when the wind speed deviation is greater than 2%, the multiplier adjustment strategy is multiplier = current multiplier × 1.2; when the HEPA efficiency deviation is greater than 0.5%, the multiplier adjustment strategy is multiplier = current multiplier × 1.1; when the vibration transmission error is greater than 5%, the multiplier adjustment strategy is multiplier = current multiplier × 1.3 (upper limit 5 times); when all indicators are less than the threshold and the multiplier is greater than 2 times, the multiplier adjustment strategy is multiplier = current multiplier × 0.9. It can be understood that the adjusted multiplier should not exceed the preset upper and lower limits.

[0039] More specifically, the data transmission in the three-layer multi-scale grid is bidirectional feedback. The macro layer transmits the main duct wind speed and FFU inlet pressure to the meso layer, the meso layer transmits the airflow velocity around the lithography machine and the vacuum pump vibration frequency to the micro layer, and the micro layer transmits the wafer surface particle concentration and HEPA interception efficiency to the meso layer.

[0040] Understandably, the meso-level layer needs to calculate the FFU outlet velocity and the airflow around the equipment, which requires knowing the FFU's input conditions. This means that the main duct velocity and FFU inlet pressure are transmitted to the meso-level layer through the macro-level layer. The micro-level layer needs to calculate the micro-airflow on the wafer surface, which requires obtaining the airflow velocity around the lithography machine, which is the basic airflow on the wafer surface. Vacuum pump vibration will disturb this airflow, so the vacuum pump vibration frequency is taken into consideration. Since the meso-level layer needs to verify the overall cleanliness around the equipment, the particle concentration on the wafer surface is the final indicator of whether the cleanliness meets the standards, and the HEPA interception efficiency determines the overall particle removal rate of the meso-level layer.

[0041] Furthermore, to verify the coupling effect and ensure that the inter-layer wind speed deviation is <2% and the simulated-measured deviation of HEPA interception efficiency is <0.5%, it is understandable that the cleanliness and thermal environment of electronic factories depend on airflow distribution. If the inter-layer wind speed transmission is inaccurate, all subsequent simulations will be distorted. HEPA is the cleanliness defense line of electronic factories. To achieve ISO Class 5, electronic factories rely on the interception efficiency of HEPA for 0.1-0.3μm particles (≥99.999%). If the interception efficiency simulation is inaccurate, the particle concentration calculation for the entire workshop will be wrong.

[0042] Step S05: Perform independence verification on the mesh and output the final mesh file.

[0043] Specifically, based on the current grid, three sets of fused grids with different densities are generated. The wind speed error between two adjacent grids is verified to be <2% and the particle concentration error is <1%. The grid with the middle density is selected as the final grid.

[0044] In summary, the three-dimensional flow field simulation method based on cleanroom air conditioning in the above embodiments of the present invention constructs a grid history database of an electronic factory, which includes scene features, grid schemes, and error feedback for three types of scenarios: lithography machine workshop, wafer packaging workshop, and FFU dense area. Based on the grid history database, an AI prediction model is trained, which is based on a random forest algorithm with a spatial attention mechanism. Dynamic scene features of the electronic factory are acquired and input into the trained AI prediction model, outputting high-risk area coordinates and partitioned grid parameters. Based on the high-risk area coordinates, the partitioned grid parameters, and accuracy requirements, a three-layer multi-scale grid (macro, meso, and micro) is generated. Overlapping grid technology is used to couple and transfer data to obtain the grid. The micro layer covers the lithography machine worktable, HEPA fiber gaps, and wafer surface. The grid is then validated for independence, and the final grid file is output. Specifically, through AI learning and multi-scale technology, grid resources are allocated on demand, resolving the contradiction between excessively dense grids wasting computing power and excessively sparse grids lacking accuracy, while also capturing the correlation between micro and macro airflow.

[0045] Example 2 Embodiment 2 of the present invention also provides a three-dimensional flow field simulation method based on cleanroom air conditioning. The difference from Embodiment 1 of the present invention is that, based on the grid historical database, a process sequence dimension is added, and a timing prediction model is constructed. The timing prediction model learns the correlation between process sequence and grid requirements, and generates corresponding grid plans in advance for each process stage. It is understood that the processes in electronic factories have a clear sequence. For example, in the photolithography process: wafer loading → coating → exposure → development → inspection, the pollution sources, airflow requirements, and equipment status are different in each stage. The solution in Embodiment 2 of the present invention can avoid simulation interruptions caused by temporary grid adjustments during process switching.

[0046] Specifically, based on existing historical data, complete timing data of the photolithography / packaging process is collected, and key parameters of each stage are marked. For example, in the coating stage, the photoresist solvent evaporation rate and coating machine rotation speed are included, and the corresponding grid in the 0.3m area around the coating machine needs to be densified.

[0047] Furthermore, based on the original random forest and spatial attention mechanisms, an LSTM time series prediction module is superimposed. The current process stage and remaining time are input. For example, in the glue application stage, there are 5 minutes left. The output is the high-risk area change and grid adjustment plan for the next stage. In addition, the process execution module of the electronic factory MES system is connected through the OPCUA protocol to obtain the current process progress in real time. The AI ​​prediction model with the LSTM time series prediction module updates the grid plan every 30 seconds. When the process is switched, the pre-generated grid parameters are directly called.

[0048] Furthermore, when the AI ​​prediction model predicts the grid plan for the next process stage, it directly triggers the adjustment of the multi-scale meso-level elastic grid. Specifically, the grid around the meso-level equipment (such as a 10m range around the lithography machine) is defined as an elastic cell. The grid nodes can move with the change of equipment position. Understandably, when the wafer handling robot moves, the corresponding grid nodes change accordingly. It should be noted that when the wafer handling robot approaches the preset area (such as 1m away from the FFU), the elastic cells are automatically densified; when the wafer handling robot moves away, the elastic cells are automatically sparsed to avoid wasting computing power.

[0049] Meanwhile, as the elastic grid cells move, the interlayer overlap area automatically adjusts its size. For example, when the robot moves from point A to point B, the meso-layer elastic grid moves with the wafer handling robot, and the overlap area between the macro-layer and the meso-layer automatically expands from 30mm to 40mm (ensuring uninterrupted data transmission). The micro-layer (the area around the wafer carried by the wafer handling robot) grid moves synchronously with the meso-layer.

[0050] Example 3 Please see Figure 2 , Figure 2 This is a structural block diagram of a three-dimensional flow field simulation system based on cleanroom air conditioning, provided in Embodiment 3 of the present invention. This three-dimensional flow field simulation system 200 based on cleanroom air conditioning is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the apparatus described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0051] Specifically, the three-dimensional flow field simulation system 200 based on cleanroom air conditioning includes: a construction module 21, a training module 22, an input module 23, a coupling module 24, and an adjustment module 25, wherein: Module 21 is used to build a grid history database for electronic factories. The grid history database includes scene characteristics, grid schemes, and error feedback for three types of scenarios: lithography machine workshop, wafer packaging workshop, and FFU dense area. The scene characteristics include the number of lithography machines, lithography machine size, number of FFUs, FFU spacing, FFU air volume, vacuum pump coordinates, vacuum pump vibration frequency, wafer handling track width, VOCs pollution source coordinates, VOCs pollution source release rate, clean area length, width, and floor height. The grid scheme includes the coordinates of high-risk areas, partitioned grid parameters, and the number of grid layers adapted to the vibration source. The error feedback includes the deviation between the simulated and measured wind speed on the lithography machine table, the simulation error of VOCs concentration, and the calculation time for micro-vibration coupling. Training module 22 is used to train an AI prediction model based on the grid historical database. The AI ​​prediction model is based on the random forest algorithm with spatial attention mechanism. Input module 23 is used to acquire dynamic scene features of the electronic factory, input the dynamic scene features into the trained AI prediction model, and output the coordinates of high-risk areas and partition grid parameters. The coupling module 24 is used to generate a three-layer multi-scale mesh of macroscopic, mesoscopic and microscopic layers according to the coordinates of the high-risk area, the partition mesh parameters and the accuracy requirements. The overlapping mesh technology is used to couple and transfer data to obtain the mesh. The microscopic layer covers the lithography machine table surface, HEPA fiber gaps and wafer surface. The macroscopic layer is the entire clean room and the mesh is generated using ANSYS Meshing. The meso-layer consists of the 10m perimeter of the lithography machine and the FFU array area, with the mesh generated using ICEMCFD. The microlayer consists of the 0.1-0.5m gap between the lithography machine stage and HEPA fibers, and the mesh is generated using Palabos+LAMMPS. Adjacent mesh layers are overlapped at the boundary to form a data transition region. The width of the overlapped region is the product of the prediction factor and the size of the outer mesh layer. Specifically, the steps for obtaining the prediction factor include: A feature library for the impact of multiples is constructed. The feature library includes physical field gradient features, geometric complexity features, and error-sensitive features. The physical field gradient features include wind speed gradient and vibration frequency. The geometric complexity features include boundary curvature and mesh orthogonality. The error-sensitive features include historical simulation errors and cleanliness level. A multiple prediction model is constructed based on the random forest regression algorithm. The multiple prediction model is used to input the physical field gradient characteristics, geometric complexity characteristics, and error sensitivity characteristics, and outputs an initial prediction multiple. The optimization objective is to make the simulation error corresponding to the model prediction multiple less than a preset value, wherein the wind speed deviation is less than 2% and the HEPA efficiency deviation is less than 0.5%. The actual physical field gradient characteristics, geometric complexity characteristics, and error sensitivity characteristics are obtained and input into the trained multiple prediction model. The initial prediction multiple is output, and the multiple is dynamically corrected by real-time monitoring of coupling accuracy to obtain the predicted multiple. Specifically, real-time monitoring indicators include wind speed deviation, HEPA efficiency deviation, and vibration transmission error. A first mapping relationship between the condition of indicators exceeding the standard and the multiple adjustment strategy, and a second mapping relationship between the condition of indicators not exceeding the standard and the multiple adjustment strategy are established. Based on the real-time monitoring indicators, the first mapping relationship, and the second mapping relationship, the corresponding multiple adjustment strategy is output to complete the multiple correction. Data transmission in the three-layer multi-scale grid is bidirectional feedback. The macro layer transmits the main duct wind speed and FFU inlet pressure to the meso layer, the meso layer transmits the airflow velocity around the lithography machine and the vacuum pump vibration frequency to the micro layer, and the micro layer transmits the wafer surface particle concentration and HEPA interception efficiency to the meso layer. Adjustment module 25 is used to perform independence verification on the mesh and output the final mesh file.

[0052] Furthermore, in some optional embodiments of the present invention, the construction module 21 adds a process timing dimension to the grid historical database and constructs a timing prediction model. The timing prediction model learns the correlation between process timing and grid requirements and generates corresponding grid plans for each process stage in advance.

[0053] Example 4 In another aspect, the present invention also proposes an electronic device, please refer to [link to relevant documentation]. Figure 3The image shows an electronic device according to Embodiment 4 of the present invention, including a memory 20, a processor 10, and a computer program 30 stored in the memory and executable on the processor. When the processor 10 executes the computer program 30, it implements the three-dimensional flow field simulation method based on cleanroom air conditioning as described above.

[0054] In some embodiments, the processor 10 may be a central processing unit (CPU), controller, microcontroller, microprocessor or other data processing chip, used to run program code stored in memory 20 or process data, such as executing access restriction programs.

[0055] The memory 20 includes at least one type of readable storage medium, such as flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 20 can be an internal storage unit of an electronic device, such as the hard disk of the electronic device. In other embodiments, the memory 20 can also be an external storage device of the electronic device, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc. Furthermore, the memory 20 can include both internal and external storage units of the electronic device. The memory 20 can be used not only to store application software and various types of data of the electronic device, but also to temporarily store data that has been output or will be output.

[0056] It should be pointed out that, Figure 3 The structure shown does not constitute a limitation on the electronic device. In other embodiments, the electronic device may include fewer or more components than shown, or combine certain components, or have different component arrangements.

[0057] This invention also proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the three-dimensional flow field simulation method based on cleanroom air conditioning as described above.

[0058] Those skilled in the art will understand that the logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can mean any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0059] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0060] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0061] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0062] The above embodiments merely illustrate several implementation methods of the present invention, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this patent should be determined by the appended claims.

Claims

1. A three-dimensional flow field simulation method based on cleanroom air conditioning, characterized in that, The method includes: Construct a grid history database for electronic manufacturing plants. The grid history database includes scene characteristics, grid schemes, and error feedback for three types of scenarios: lithography machine workshop, wafer packaging workshop, and FFU dense area. Based on the grid historical database, an AI prediction model is trained, which is based on the random forest algorithm with spatial attention mechanism; The dynamic scene features of the electronic factory are obtained and input into the trained AI prediction model to output the coordinates of the high-risk area and the partition grid parameters. Based on the coordinates of the high-risk area, the partitioned grid parameters, and the accuracy requirements, a three-layer multi-scale grid of macroscopic, mesoscopic, and microscopic layers is generated. Overlapping grid technology is used to couple and transfer data to obtain the grid. The microscopic layer covers the lithography machine table surface, HEPA fiber gaps, and wafer surface. The independence of the mesh is verified, and the final mesh file is output.

2. The three-dimensional flow field simulation method based on cleanroom air conditioning according to claim 1, characterized in that, The scene features include the number of lithography machines, the size of the lithography machines, the number of FFUs, the spacing between FFUs, the air volume of FFUs, the coordinates of the vacuum pumps, the vibration frequency of the vacuum pumps, the width of the wafer transport track, the coordinates of the VOCs pollution sources, the release rate of the VOCs pollution sources, the length and width of the clean area, and the floor height. The grid scheme includes the coordinates of high-risk areas, partitioned grid parameters, and the number of grid layers adapted to the vibration source. The error feedback includes the deviation between the simulated and measured wind speed on the lithography machine's worktable, the simulation error of VOCs concentration, and the time consumed in micro-vibration coupling calculation.

3. The three-dimensional flow field simulation method based on cleanroom air conditioning according to claim 2, characterized in that, The macroscopic layer refers to the entire cleanroom, and the mesh is generated using ANSYS Meshing. The meso-layer consists of the 10m perimeter of the lithography machine and the FFU array area, with the mesh generated using ICEMCFD. The microlayer consists of the 0.1-0.5m gap between the lithography machine stage and the HEPA fiber, and the mesh is generated using Palabos+LAMMPS.

4. The three-dimensional flow field simulation method based on cleanroom air conditioning according to claim 3, characterized in that, In the step of generating a three-layer multi-scale grid (macro, meso, and micro) based on the coordinates of the high-risk area, the partitioned grid parameters, and accuracy requirements, and coupling and transmitting data using overlapping grid technology to obtain the grid, adjacent grid layers are intersected at the boundary to form a data transition zone. The width of the intersecting region is the product of the prediction multiple and the size of the outer grid.

5. The three-dimensional flow field simulation method based on cleanroom air conditioning according to claim 4, characterized in that, In the step of generating a three-layer multi-scale grid (macro, meso, and micro) based on the coordinates of the high-risk area, the partitioned grid parameters, and accuracy requirements, and coupling and transmitting data using overlapping grid technology to obtain the grid, the data transmission in the three-layer multi-scale grid is bidirectional feedback. Specifically, the macro layer transmits the main duct wind speed and FFU inlet pressure to the meso layer, the meso layer transmits the airflow velocity around the lithography machine and the vacuum pump vibration frequency to the micro layer, and the micro layer transmits the wafer surface particle concentration and HEPA interception efficiency to the meso layer.

6. The three-dimensional flow field simulation method based on cleanroom air conditioning according to claim 5, characterized in that, The steps for obtaining the predicted multiple include: A feature library for the impact of multiples is constructed. The feature library includes physical field gradient features, geometric complexity features, and error-sensitive features. The physical field gradient features include wind speed gradient and vibration frequency. The geometric complexity features include boundary curvature and mesh orthogonality. The error-sensitive features include historical simulation errors and cleanliness level. A multiple prediction model is constructed based on the random forest regression algorithm. The multiple prediction model is used to input the physical field gradient characteristics, geometric complexity characteristics, and error sensitivity characteristics, and outputs an initial prediction multiple. The optimization objective is to make the simulation error corresponding to the model prediction multiple less than a preset value, wherein the wind speed deviation is less than 2% and the HEPA efficiency deviation is less than 0.5%. The actual physical field gradient characteristics, geometric complexity characteristics, and error sensitivity characteristics are obtained and input into the trained multiple prediction model. The initial prediction multiple is output, and the multiple is dynamically corrected by real-time monitoring of coupling accuracy to obtain the predicted multiple. Specifically, real-time monitoring indicators include wind speed deviation, HEPA efficiency deviation, and vibration transmission error. A first mapping relationship is established between the condition of indicators exceeding the standard and the multiple adjustment strategy, and a second mapping relationship is established between the condition of indicators not exceeding the standard and the multiple adjustment strategy. Based on the real-time monitoring indicators, the first mapping relationship, and the second mapping relationship, the corresponding multiple adjustment strategy is output to complete the multiple correction.

7. The three-dimensional flow field simulation method based on cleanroom air conditioning according to claim 6, characterized in that, Based on the historical grid database, a new process timing dimension is added, and a timing prediction model is constructed. The timing prediction model learns the correlation between process timing and grid requirements, and generates corresponding grid plans in advance for each process stage.

8. A three-dimensional flow field simulation system based on cleanroom air conditioning, characterized in that, For implementing the three-dimensional flow field simulation method based on cleanroom air conditioning as described in any one of claims 1-7, the system comprises: The construction module is used to build a grid history database for electronic factories. The grid history database includes scene characteristics, grid schemes, and error feedback for three types of scenarios: lithography machine workshop, wafer packaging workshop, and FFU dense area. The training module is used to train an AI prediction model based on the grid historical database. The AI ​​prediction model is based on a random forest algorithm with a spatial attention mechanism. The input module is used to acquire the dynamic scene features of the electronic factory and input the dynamic scene features into the trained AI prediction model, and output the coordinates of the high-risk area and the partition grid parameters. The coupling module is used to generate a three-layer multi-scale mesh of macroscopic, mesoscopic and microscopic layers according to the coordinates of the high-risk area, the partition mesh parameters and the accuracy requirements. The overlapping mesh technology is used to couple and transmit data to obtain the mesh. The microscopic layer covers the lithography machine table surface, HEPA fiber gaps and wafer surface. The adjustment module is used to perform independence verification on the mesh and output the final mesh file.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the three-dimensional flow field simulation method based on cleanroom air conditioning as described in any one of claims 1-7.

10. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the three-dimensional flow field simulation method based on cleanroom air conditioning as described in any one of claims 1-7.

Images