Batched semiconductor factory equipment digital twin model construction method and apparatus

Abstract

The application discloses a batched semiconductor factory equipment digital twin model construction method and device, and the method comprises the following steps: based on determined modeling requirements, matching a preset basic parameter template library, giving corresponding basic parameter templates, and determining a structured parameter set; analyzing equipment basic parameters and production process parameters, and combining a mapping relationship with a component library, traversing equipment components and logical components in the component library, determining each target equipment component and a plurality of target logical components; fusing the target equipment component and the plurality of target logical components to form a target standard component; based on factory layout parameters, linking the target standard component to construct a semiconductor factory equipment digital twin model. Through the processing behavior of disassembling the equipment, the basic parameter template library and the component library are combined to realize the rapid batch generation of the target standard component, and on the basis of guaranteeing the simulation effect of the semiconductor equipment, the modeling efficiency of the digital twin model is improved.

Description

Technical Field

[0001] This invention belongs to the technical field of model building, specifically relating to a method and apparatus for constructing digital twin models of semiconductor factory equipment in batches. Background Technology

[0002] One of the most crucial elements in building a semiconductor digital factory is the semiconductor equipment model. This model serves as the core carrier supporting production simulation and operation and maintenance optimization. Current digital twin modeling technologies in semiconductor digital factory construction suffer from several drawbacks, including limited modeling techniques, a lack of dedicated component libraries, and low modeling efficiency. Common modeling methods utilize BIM-like modeling software, prioritizing detailed geometric reconstruction without considering the internal process logic of the equipment (processing behavior, energy consumption, transmission, secondary distribution, etc.). These methods only support basic data transmission and cannot meet the multi-dimensional interactive needs of semiconductor factories regarding "production cycle time, energy consumption data, and secondary distribution networks." Current model building processes use general-purpose modeling component libraries, which cannot adapt to the processing logic and energy characteristics of semiconductor equipment, resulting in significant discrepancies between the model and the actual equipment's operational logic. Furthermore, relying on customized development of single-equipment models fails to meet the rapid modeling needs of multiple devices in a semiconductor factory.

[0003] Patent CN120124312A discloses an adaptive digital twin simulation device and method for semiconductor equipment energy consumption. The device includes a data acquisition module, a simulation model construction and operation module, and a parameter update module connected by signals. The data acquisition module acquires production information and process parameter information of semiconductor products, as well as real-time actual processing data. The simulation model construction and operation module constructs a process equipment energy consumption simulation model based on a preset process equipment configuration unit, integrates the semiconductor product production information and process parameter information, runs the process equipment energy consumption simulation model, and provides energy consumption simulation data in real time. The parameter update module is used to update the process parameter information in the process equipment energy consumption simulation model. By setting up the parameter update module to introduce a real-time data feedback mechanism, the process parameter information is dynamically updated, improving the accuracy and real-time performance of the process equipment energy consumption simulation model.

[0004] The aforementioned existing technologies achieve the accuracy and real-time performance of simulation models by constructing simulation models and combining them with real-time data feedback. However, they do not consider the versatility of modeling components or the modeling efficiency. How to improve the modeling efficiency of digital twin models of semiconductor devices while ensuring the simulation effect is a problem that needs to be solved. Summary of the Invention

[0005] To address the shortcomings of the existing technologies, this invention provides a method and apparatus for constructing batch digital twin models of semiconductor factory equipment. The method includes: based on defined modeling requirements, matching a preset basic parameter template library to provide corresponding basic parameter templates; determining a structured parameter set based on the basic parameter templates, wherein the structured parameter set includes equipment basic parameters, production process parameters, and factory layout parameters; analyzing the equipment basic parameters and production process parameters, and combining them with the mapping relationship with a component library, traversing the equipment components and logic components in the component library to determine each target equipment component and its corresponding multiple target logic components; fusing the target equipment components and their corresponding multiple target logic components to form target standard components; and linking the target standard components based on the factory layout parameters to construct a digital twin model of the semiconductor factory equipment.

[0006] By disassembling the processing behavior of the equipment, equipment components and logic components are formed. Combined with the basic parameter template library and component library, the target equipment components and their corresponding multiple target logic components are integrated to achieve rapid batch generation of target standard components and accurate matching of semiconductor equipment operating logic. This enables the batch construction of equipment twin models with information interaction capabilities, forming a digital twin model of semiconductor factory equipment. While ensuring the simulation effect of semiconductor equipment, the modeling efficiency of semiconductor equipment digital twin models is improved.

[0007] In a first aspect, the present invention provides a method for constructing digital twin models of semiconductor factory equipment in batches, specifically including the following steps: Based on the defined modeling requirements, a pre-defined basic parameter template library is matched to provide the corresponding basic parameter template; Based on the basic parameter template, a structured parameter set is determined, which includes basic equipment parameters, production process parameters, and factory layout parameters. Analyze the basic parameters of the equipment and the production process parameters, and combine them with the mapping relationship with the component library. Traverse the equipment components and logic components in the component library to determine each target equipment component and its corresponding multiple target logic components. Integrate the target device components and multiple corresponding target logic components to form the target standard component; Based on factory layout parameters, target standard components are linked to construct a digital twin model of semiconductor factory equipment.

[0008] Furthermore, the basic equipment parameters include equipment hardware configuration parameters, the production process parameters include equipment operating parameters, equipment energy consumption parameters, production process parameters, and commissioning parameters, and the factory layout parameters include pipeline location layout, factory building layout, and logistics route layout.

[0009] Furthermore, the component library includes device components and logic components. The device components include multiple cavity sub-components, and the logic components include processing logic components and energy-consuming logic components. The construction of the component library includes the following steps: Obtain the basic cavity parameters of each device, build multiple corresponding cavity sub-components, and form a device component; Determine the processing parameter interface and processing control logic interface, set the cavity invocation strategy, and construct the processing logic component. The cavity invocation strategy includes the processing parameters received by the processing parameter interface based on the processing control logic call received by the self-processing control logic interface. Build plant management sub-components for each energy consumption type, determine the association function between each plant management sub-component and the processing parameters, and combine them with the energy consumption logic to construct the energy consumption logic component; It gathers equipment components, processing logic components, and energy consumption logic components to build a component library.

[0010] Furthermore, plant management sub-components for each energy consumption type are constructed, the association function between each plant management sub-component and processing parameters is determined, and energy consumption logic components are built in conjunction with energy consumption logic. Specifically, the following steps are included: Based on the energy consumption type, construct the corresponding plant management sub-components and basic energy consumption data; Provide the association function between each plant sub-component and the processing parameters; By integrating the correlation function and various processing parameters, the steady-state energy consumption components are determined; Determine the dynamic correction coefficients corresponding to the jumps in each processing parameter, and provide dynamic energy consumption breakdown; By superimposing the basic energy consumption base, steady-state energy consumption components, dynamic energy consumption components, and random error terms, the energy consumption function for each energy consumption type is given. Verify the energy consumption functions for each energy consumption type and construct energy consumption logic components.

[0011] Furthermore, using the energy function, it can be specifically expressed as: ; Among them, E total Let E be the energy function. base Basic energy consumption base, E process For steady-state energy consumption, E dynamic For dynamic energy consumption items, This is a random disturbance term.

[0012] The baseline energy consumption is the benchmark energy consumption under different states, such as standby, preparation, process, and maintenance states. The steady-state energy consumption component represents the steady-state energy consumption driven by process parameters. The dynamic energy consumption component represents the dynamic response of parameter change rate, such as transient heat capacity and inertia. The random disturbance component represents the change in energy consumption caused by factors such as equipment aging and environmental fluctuations.

[0013] Furthermore, the energy consumption functions for each energy consumption type are verified, specifically including the following steps: Construct a regularized loss function that integrates the least squares residual sum of squares and physical prior constraints; Obtain historical processing parameters, historical energy consumption data, and initial dynamic correction coefficients; set the learning rate and regularization coefficients. Based on the initial dynamic correction coefficients and historical processing parameters, the simulated energy consumption data is given through the energy consumption function; Error analysis is performed using a regularized loss function based on simulated energy consumption data and historical energy consumption data. Adjust the initial dynamic correction coefficients in the negative gradient direction, and continue to perform error analysis using the regularized loss function until convergence, thus obtaining the dynamic correction coefficients and completing the verification of the energy consumption function for each energy consumption type.

[0014] Furthermore, the regularization loss function is specifically expressed as: ; Where λ is the regularization coefficient. For the historical energy consumption data in the i-th iteration, Let M be the simulation energy consumption data in the i-th iteration, M be the iteration number, and θ be the dynamic correction coefficient.

[0015] Furthermore, the basic parameters of the equipment and the production process parameters are analyzed, and combined with the mapping relationship with the component library, the equipment components and logic components in the component library are traversed to determine each target equipment component and its corresponding multiple target logic components, specifically including: Based on production process parameters and commissioning parameters, we provide equipment selection strategies; Based on the device selection strategy, traverse the device components in the component library to determine each initial device component; Based on the basic and operating parameters of each piece of equipment, determine the production and operation mode of the corresponding equipment. Based on the production and operation mode of each piece of equipment, a cavity selection strategy for the corresponding equipment is given; Based on the cavity selection strategy, the target cavity sub-components for each device are given; The initial device components are updated based on the target cavity sub-components to form the target device component; Based on the processing parameters of the target cavity sub-component, the processing logic component and the power logic component are matched to determine multiple target logic components corresponding to each target device component.

[0016] Furthermore, based on the cavity selection strategy, the target cavity sub-components for each device are given, specifically including: Based on the real-time status of each cavity in each device, the number of available cavities in real time is given, and the real-time cavity queue is determined; Random probabilities are generated based on a Poisson distribution, and a target cavity is selected from the real-time cavity queue, specifically as follows: ; Among them, C selected For the target cavity, E direct,j Let E be the direct process energy consumption of cavity j, representing the standard energy consumption integral of cavity j performing the production process parameters corresponding to the target process. thermal,j E represents the energy consumption for the hot restart of cavity j, indicating the energy required for cavity j to recover from its current temperature to the process temperature specified in the production process parameters. switch,j The energy consumption penalty for process switching of cavity j is denoted by τ, where j is the cavity number in the real-time cavity queue, and τ is a noise weighting coefficient used to control the adjustment factor between the determinism of the optimization and the randomness of the exploration. To remove mean Poisson noise, Ω idle For real-time cavity queues; Based on the cavity sub-components corresponding to the target cavity, the target cavity sub-components for each device are given.

[0017] Furthermore, the direct process energy consumption of cavity j is specifically expressed as follows:

[0018] Among them, E direct,j T represents the direct process energy consumption of cavity j. process,j For cavity j, the process duration corresponding to the target process of the production process parameters is P. recipe (t) represents the process power curve of the target process, indicating the time-varying power setpoint in the production process parameters.

[0019] Furthermore, the energy consumption for the hot restart of cavity j is specifically expressed as follows: ; ; Among them, E thermal,j For the heat restart energy consumption of cavity j, c p The equivalent specific heat capacity, representing the overall specific heat capacity of the cavity thermal mass system, m chamber Equivalent thermal mass represents the total mass of the cavity walls, carrier, and wafer involved in the temperature cycle within the cavity. For the temperature rise requirement of cavity j, η represents the temperature difference when the current temperature is insufficient. heater The heater efficiency is represented by the efficiency coefficient for converting electrical energy into heat energy. This is a hot restart indicator function, indicating that a full restart is triggered when the temperature difference exceeds a temperature threshold δT; otherwise, a warm-start operation is initiated. process T represents the processing temperature in the production process parameters. current,j Let be the current temperature of cavity j.

[0020] Furthermore, the energy consumption penalty for process switching in cavity j is specifically expressed as follows: ; ; Among them, E switch,j For the energy consumption penalty of process switching of cavity j, P idle τ represents the standby power consumption, indicating the basic power consumption of the cavity in the idle state. purge,j Let γ be the purification steady-state time of cavity j, and γ represent the forced waiting time based on the lookup table of compatibility between the preceding and following gas formulations. match,j Let be the formula matching coefficient for cavity j. The closer the formula matching coefficient is to 1, the better the compatibility and the lower the switching cost. exp() is an exponential function, α is the gas difference sensitivity, used to control the decay rate of the formula matching coefficient with gas type differences, and D gas The gas atmosphere distance represents the weighted norm of the differences in gas type, flow rate, and pressure between the two target processes. (Recipe) last,j For the target process previously executed in cavity j, Recipe target The target process is represented by the process formula required for the wafer to be processed.

[0021] Furthermore, based on factory layout parameters, target standard components are linked to construct a digital twin model of semiconductor factory equipment, specifically including: The initial positions of each target standard component are determined by the factory building layout; Based on the pipeline network layout, analyze and match the equipment process parameters in the target standard components, and adjust the initial position of the target standard components. Based on the logistics route layout and combined with the target operation logic of the target standard components, the various target standard components are linked to form a digital twin model of semiconductor factory equipment.

[0022] Furthermore, this also includes updating the data twin model of semiconductor factory equipment, specifically including: Based on the energy consumption parameters of the semiconductor factory equipment data twin model, set energy consumption optimization targets; Based on the current structured parameter set, and combined with the energy optimization target, the structured parameter set is updated to obtain multiple sets of demand data. By running a data twin model of semiconductor factory equipment using multiple sets of demand data, the optimal set of demand data is obtained. By combining the optimal demand data set, update the production process parameters and / or factory layout parameters of the semiconductor factory equipment data twin model.

[0023] Secondly, the present invention also provides a batch-processing apparatus for constructing digital twin models of semiconductor factory equipment, employing any of the above-mentioned batch-processing methods for constructing digital twin models of semiconductor factory equipment, specifically including: The template acquisition module is used to match the preset basic parameter template library based on the determined modeling requirements and provide the corresponding basic parameter template. The parameter determination module is used to determine a structured parameter set based on the basic parameter template. The structured parameter set includes basic equipment parameters, production process parameters, and factory layout parameters. The component determination module is used to analyze the basic parameters of the equipment and the production process parameters, and combine them with the mapping relationship with the component library to traverse the equipment components and logic components in the component library to determine each target equipment component and its corresponding multiple target logic components. The component fusion module is used to fuse target device components and multiple corresponding target logic components to form a target standard component. The model building module is used to link target standard components based on factory layout parameters to build a digital twin model of semiconductor factory equipment.

[0024] The method and apparatus for constructing digital twin models of semiconductor factory equipment in batches provided by the present invention have at least the following beneficial effects: (1) By disassembling the processing behavior of the equipment, equipment components and logic components are formed. Combined with the basic parameter template library and component library, the target equipment components and corresponding multiple target logic components are integrated to realize the rapid batch generation of target standard components and accurately match the operating logic of semiconductor equipment. This enables the batch construction of equipment twin models with information interaction capabilities, forming a digital twin model of semiconductor factory equipment. While ensuring the simulation effect of semiconductor equipment, the modeling efficiency of the digital twin model of semiconductor equipment is improved.

[0025] (2) By building a component library, the equipment components and logic components in the component library are more in line with the actual operating logic of the equipment, and the processing logic and energy characteristics of different semiconductor equipment are adapted. During the model building process, the corresponding components can be directly retrieved from the component library, which improves the universality of the model building process and meets the rapid modeling needs of multiple equipment in semiconductor factories. Attached Figure Description

[0026] Figure 1 A flowchart illustrating the method for constructing a batch of digital twin models of semiconductor factory equipment provided in an embodiment of the present invention; Figure 2 A schematic diagram illustrating the process of building a component library according to an embodiment of the present invention; Figure 3 This is an architectural block diagram of the component library provided in an embodiment of the present invention; Figure 4 A schematic diagram illustrating the process of constructing a power-consuming logic component according to an embodiment of the present invention; Figure 5 This is a flowchart illustrating the process of determining the target device component and the target logic component according to an embodiment of the present invention; Figure 6 A schematic flowchart of the target cavity sub-assembly is provided for an embodiment of the present invention; Figure 7 This is a schematic diagram illustrating the process of constructing a digital twin model of semiconductor factory equipment according to an embodiment of the present invention; Figure 8 A structural block diagram of a device for constructing digital twin models of semiconductor factory equipment in batches, provided in an embodiment of the present invention.

[0027] Among them, 201 is the template acquisition module; 202 is the parameter determination module; 203 is the component determination module; 204 is the component fusion module; and 205 is the model construction module. Detailed Implementation

[0028] To better understand the above technical solutions, a detailed description of the solutions will be provided below in conjunction with the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0029] The terminology used in the embodiments of this invention is for the purpose of describing particular embodiments only and is not intended to limit the invention. The singular forms “a,” “the,” and “the” as used in the embodiments of this invention and the appended claims are also intended to include the plural forms, and “multiple” generally includes at least two unless the context clearly indicates otherwise.

[0030] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that an article or device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such an article or device. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the article or device that includes said element.

[0031] Digital twin technology has been gradually matured in industrial applications, but there is still a significant technological gap in the semiconductor industry: existing solutions mostly focus on "general industrial equipment" and do not address the specific needs of semiconductor equipment.

[0032] From an industry perspective, semiconductor equipment has three core characteristics that distinguish it from other industrial equipment: First, it has a modular structure, with the equipment divided into multiple functional modules, each with different processing logic and equipment behavior characteristics; second, it has strong operational coupling, with the wafer processing cycle directly determining the energy consumption intensity, requiring simultaneous simulation of the relationship between the two; and third, it has multi-dimensional interfaces, requiring the simultaneous possession of multiple interface information (interfacing with multiple systems and departments in the factory, such as the factory construction department, factory operation and maintenance department, production system, energy system, etc.), access to the virtual digital factory, and complex data interaction requirements.

[0033] The common methods for constructing digital twin models have the following problems: The model has limited functionality: it only completes the geometric reconstruction of the equipment's appearance and cannot simulate the process and energy consumption logic; or it only supports the simulation of equipment energy consumption and supports single data transmission, which cannot meet the needs of multi-system docking and interaction, and does not reflect the twin model. Poor compatibility of component library: The general industrial modeling component library is not designed for the characteristics of semiconductor equipment such as "processing timing, multi-medium energy use, and secondary pipeline association", and cannot accurately reproduce the equipment operation logic; Lack of batch modeling: Traditional modeling relies on manual customization, and the modeling cycle for a single device is long, making it difficult to meet the modeling needs of dozens to hundreds of devices in a semiconductor factory.

[0034] Based on this, the present invention provides a method for constructing digital twin models of semiconductor factory equipment in batches, specifically including: based on determined modeling requirements, matching a preset basic parameter template library to provide corresponding basic parameter templates; determining a structured parameter set according to the basic parameter templates, wherein the structured parameter set includes equipment basic parameters, production process parameters, and factory layout parameters; analyzing the equipment basic parameters and production process parameters, and combining them with the mapping relationship with the component library, traversing the equipment components and logic components in the component library to determine each target equipment component and its corresponding multiple target logic components; merging the target equipment components and their corresponding multiple target logic components to form target standard components; and linking the target standard components based on the factory layout parameters to construct a digital twin model of semiconductor factory equipment.

[0035] By dissecting the processing behavior of equipment, designing component libraries, basic parameter template libraries, and multiple interfaces, we can achieve rapid batch generation of target standard components, accurately match the operating logic of semiconductor equipment, seamlessly integrate with the digital factory system, and have information interaction capabilities. This enables the batch construction of equipment twin models with information interaction capabilities, forming a digital twin model of semiconductor factory equipment.

[0036] like Figure 1 As shown in the figure, this embodiment of the invention provides a method for constructing digital twin models of semiconductor factory equipment in batches. The specific steps are as follows: S101: Based on the defined modeling requirements, match the preset basic parameter template library and provide the corresponding basic parameter template.

[0037] It's important to understand that modeling requirements vary depending on the application scenario. A pre-defined basic parameter template library can be used to match the modeling needs. In subsequent processes, this basic parameter template is used to collect and process structured parameter sets.

[0038] In a specific example, modeling requirements include those from equipment manufacturers and design institutes. For equipment manufacturers, the basic parameter templates consist of parameterized information about the physical machine, including hardware configuration (number / type of modules, dimensional parameters), operating parameters (processing cycle time range, energy consumption ratings), a process recipe library (including workstation operation sequences and chemical configurations for each recipe), and IES production deployment tables. For design institutes, the basic parameter templates include 2D factory layout drawings (pipeline location layout, building layout, logistics path layout), basic energy consumption parameters for the machines (pressure / flow thresholds for water / drainage / gas / exhaust), and FDC water usage setting tables. Different modeling requirements correspond to slightly different basic parameter templates; the basic parameter template library is pre-built through correlation analysis between modeling requirements and basic parameters.

[0039] By defining the basic parameter template, we can uniformly receive multi-source parameterized information related to semiconductor devices, providing a standardized input source for subsequent modeling and covering data source requirements under different modeling needs.

[0040] S102: Determine the structured parameter set based on the basic parameter template.

[0041] The structured parameter set includes basic equipment parameters, manufacturing process parameters, and factory layout parameters. Basic equipment parameters are a set of static physical quantities describing the semiconductor equipment, while manufacturing process parameters are a set of dynamic operational quantities describing the semiconductor equipment under different processes. Factory layout parameters are a set of data describing the spatial location of the semiconductor equipment in the factory coordinate system.

[0042] Specifically, the basic equipment parameters include equipment hardware configuration parameters, the production process parameters include equipment operating parameters, equipment energy consumption parameters, production process parameters, and commissioning parameters, and the factory layout parameters include pipeline location layout, factory building layout, and logistics route layout.

[0043] In a specific example, basic equipment parameters include the semiconductor equipment model, dimensions, weight, power capacity, temperature range, and humidity range. Equipment operating parameters include cavity vacuum, RF power, wafer temperature, process time, and turntable speed. Equipment energy consumption parameters include peak power consumption, average power consumption, cooling water flow rate, cooling water inlet temperature, and nitrogen consumption. Production process parameters include etching gas, pressure settings, and target linewidth. Production parameters include capacity, first-wafer yield, stabilization cycle, changeover time, and annual uptime. Piping network layout includes process gas riser coordinates, valve box numbers, vacuum pump locations, cooling water inlet elevation, and exhaust outlet diameter. Factory building layout includes column spacing, live load, return air duct width, fire compartment area, and roof load. Logistics path layout includes overhead crane track elevation, loading aisle width, elevator car dimensions, temporary storage area, and path width.

[0044] Through basic parameter templates, multi-format data, including Excel spreadsheets, JSON files, CAD drawings, and CSV files, is input from preset standardized interfaces. These preset standardized interfaces include a production launch interface, a Recipe interface, and a water usage setting interface. The production launch interface receives batch production plans, the Recipe interface receives recipe numbers and operational logic, and the water usage setting interface receives workstation water usage status, time, and flow rate. The input multi-source data is cleaned, transformed, and serialized, outliers are removed, and the format is standardized to output standardized data compatible with the modeling component library, thus determining the structured parameter set.

[0045] In a specific example, the 3σ rule is used to remove outlier data (e.g., the energy consumption of a semiconductor device is 45kW, exceeding the mean by 3 times the standard deviation, and is therefore identified as outlier and removed). If missing data is found, it is automatically supplemented based on the parameter statistics of similar semiconductor devices. After data cleaning, the data undergoes a structured transformation. Unstructured information in 2D drawings is extracted using image recognition algorithms, converting pipeline locations, building boundaries, and logistics paths into 3D coordinate parameters, and machine dimensions into geometric parameters recognizable by the model. Time-series data in different formats (e.g., timestamps in IES production reports and second-level records in FDC data) are standardized into the "yyyy-MM-dd HH:mm:ss" format. Data is sorted according to the three dimensions of "device ID-timestamp-parameter type" to generate a standardized dataset, resulting in a structured dataset.

[0046] S103: Analyze the basic parameters of the equipment and the production process parameters, and combine them with the mapping relationship with the component library. Traverse the equipment components and logic components in the component library to determine each target equipment component and its corresponding multiple target logic components.

[0047] Furthermore, the component library includes device components and logic components. The device components include multiple cavity sub-components, and the logic components include processing logic components and energy-consuming logic components. For the construction of the component library, refer to Figure 2 Specifically, it includes the following steps: Obtain the basic cavity parameters of each device, build multiple corresponding cavity sub-components, and form a device component; Determine the processing parameter interface and processing control logic interface, set the cavity invocation strategy, and construct the processing logic component. The cavity invocation strategy includes the processing parameters received by the processing parameter interface based on the processing control logic call received by the self-processing control logic interface. Build plant management sub-components for each energy consumption type, determine the association function between each plant management sub-component and the processing parameters, and combine them with the energy consumption logic to construct the energy consumption logic component; It gathers equipment components, processing logic components, and energy consumption logic components to build a component library.

[0048] In one specific implementation, refer to Figure 3Based on the basic parameters of the cavities in various devices within a semiconductor factory, a 3D model of each cavity is constructed, forming corresponding cavity sub-components. These cavity sub-components are combined to form the 3D model of the device, resulting in a device component. Through in-cavity invocation strategies, processing parameters obtained from the processing parameter interface and processing control logic obtained from the processing control logic interface are merged to form executable processing logic, resulting in a processing logic component. Simultaneously, based on the plant management sub-components that deliver different energy types, the association function between each plant management sub-component and the processing parameters is determined. Combined with the energy consumption logic for each energy type, an energy consumption logic component is constructed to determine the energy consumption situation corresponding to each energy type. The device components, processing logic components, and energy consumption logic components are then merged and summarized to form a component library. In a specific example, the processing logic component sets a chamber to process 4 wafers simultaneously, with each processing session lasting 2 minutes and followed by a 3-minute rinse. The 2-minute processing time is the processing parameter, and the processing control logic is to process 4 wafers and perform a 3-minute rinse after each processing session.

[0049] It's important to understand that each piece of equipment in a semiconductor factory includes at least one cavity, and the processing is completed within these cavities. Energy type refers to the category of all energy and material flows that semiconductor equipment obtains from the plant system to complete specified processes, maintain normal operation, and ensure processing accuracy and yield. Energy types include water, air, nitrogen, oxygen, hydrogen, argon, helium, carbon dioxide, specialty gases, vacuum, exhaust ventilation, and electrical energy.

[0050] Furthermore, plant management sub-components for each energy consumption type are constructed, the association function between each plant management sub-component and processing parameters is determined, and energy consumption logic components are built in conjunction with energy consumption logic, referring to... Figure 4 Specifically, it includes the following steps: Based on the energy consumption type, construct the corresponding plant management sub-components and basic energy consumption data; Provide the association function between each plant sub-component and the processing parameters; By integrating the correlation function and various processing parameters, the steady-state energy consumption components are determined; Determine the dynamic correction coefficients corresponding to the jumps in each processing parameter, and provide dynamic energy consumption breakdown; By superimposing the basic energy consumption base, steady-state energy consumption components, dynamic energy consumption components, and random error terms, the energy consumption function for each energy consumption type is given. Verify the energy consumption functions for each energy consumption type and construct energy consumption logic components.

[0051] Using the energy function, it can be specifically expressed as: ; Among them, E total Let E be the energy function. baseBasic energy consumption base, E process For steady-state energy consumption, E dynamic For dynamic energy consumption items, This is a random disturbance term.

[0052] The baseline energy consumption is the benchmark energy consumption under different states, such as standby, preparation, process, and maintenance states. The steady-state energy consumption component represents the steady-state energy consumption driven by process parameters. The dynamic energy consumption component represents the dynamic response of parameter change rate, such as transient heat capacity and inertia. The random disturbance component represents the change in energy consumption caused by factors such as equipment aging and environmental fluctuations.

[0053] It is understandable that different energy consumption types result in different operating parameters for the corresponding plant sub-components, leading to different baseline energy consumption figures. The baseline energy consumption figure is only related to the energy consumption type and the corresponding plant sub-component. Based on the correlation function between the plant sub-component and processing parameters, substituting the determined processing parameters into the correlation function yields the energy consumption generated by the plant sub-component under the corresponding processing conditions, resulting in a steady-state energy consumption component. The steady-state energy consumption component is related to the energy consumption type and processing parameters of the plant sub-component. Since processing parameters may change during processing, these changes can cause variations in energy consumption. By analyzing the different abrupt changes in processing parameters, corresponding dynamic correction coefficients can be obtained, thus providing a dynamic energy consumption component. The baseline energy consumption figure, steady-state energy consumption component, dynamic energy consumption component, and random error term are superimposed to give the energy consumption function for each energy consumption type. Finally, the energy consumption function for each energy consumption type is verified, constructing the energy consumption logic component.

[0054] Furthermore, the energy consumption functions for each energy consumption type are verified, specifically including the following steps: Construct a regularized loss function that integrates the least squares residual sum of squares and physical prior constraints; Obtain historical processing parameters, historical energy consumption data, and initial dynamic correction coefficients; set the learning rate and regularization coefficients. Based on the initial dynamic correction coefficients and historical processing parameters, the simulated energy consumption data is given through the energy consumption function; Error analysis is performed using a regularized loss function based on simulated energy consumption data and historical energy consumption data. Adjust the initial dynamic correction coefficients in the negative gradient direction, and continue to perform error analysis using the regularized loss function until convergence, thus obtaining the dynamic correction coefficients and completing the verification of the energy consumption function for each energy consumption type.

[0055] The regularization loss function is specifically expressed as: ; Where λ is the regularization coefficient. For the historical energy consumption data in the i-th iteration, Let M be the simulation energy consumption data in the i-th iteration, M be the iteration number, and θ be the dynamic correction coefficient.

[0056] In one specific implementation, historical processing parameters are first acquired, such as cavity pressure, gas flow rate, processing time, and number of wafers for a certain cavity sub-component. Historical energy consumption data is also acquired, such as electricity consumption, cooling water flow rate, and nitrogen mass flow rate corresponding to the processing period. The initial dynamic correction coefficient is pre-set based on historical experience. Simultaneously, a regularization coefficient and learning rate are set. Then, based on the energy consumption function, combined with the initial dynamic correction coefficient and historical processing parameters, simulated energy consumption data is calculated. Next, a regularization error analysis is performed on the historical and simulated energy consumption data using a regularization loss function. This process is repeated until the function value corresponding to the regularization loss function converges. At this point, iteration stops, and the final dynamic correction coefficient is output, completing the verification of the energy consumption function.

[0057] Furthermore, the basic parameters of the equipment and the production process parameters are analyzed, and combined with the mapping relationship with the component library, the equipment components and logic components in the component library are traversed to determine each target equipment component and its corresponding multiple target logic components, referring to... Figure 5 Specifically, it includes: Based on production process parameters and commissioning parameters, we provide equipment selection strategies; Based on the device selection strategy, traverse the device components in the component library to determine each initial device component; Based on the basic and operating parameters of each piece of equipment, determine the production and operation mode of the corresponding equipment. Based on the production and operation mode of each piece of equipment, a cavity selection strategy for the corresponding equipment is given; Based on the cavity selection strategy, the target cavity sub-components for each device are given; The initial device components are updated based on the target cavity sub-components to form the target device component; Based on the processing parameters of the target cavity sub-component, the processing logic component and the power logic component are matched to determine multiple target logic components corresponding to each target device component.

[0058] Furthermore, based on the cavity selection strategy, the target cavity sub-components for each device are given, with reference to... Figure 6 Specifically, it includes: Based on the real-time status of each cavity in each device, the number of available cavities in real time is given, and the real-time cavity queue is determined; Random probabilities are generated based on a Poisson distribution, and a target cavity is selected from the real-time cavity queue, specifically as follows: ; Among them, Cselected For the target cavity, E direct,j Let E be the direct process energy consumption of cavity j, representing the standard energy consumption integral of cavity j performing the production process parameters corresponding to the target process. thermal,j E represents the energy consumption for the hot restart of cavity j, indicating the energy required for cavity j to recover from its current temperature to the process temperature specified in the production process parameters. switch,j The energy consumption penalty for process switching of cavity j is denoted by τ, where j is the cavity number in the real-time cavity queue, and τ is a noise weighting coefficient used to control the adjustment factor between the determinism of the optimization and the randomness of the exploration. To remove mean Poisson noise, Ω idle For real-time cavity queues; Based on the cavity sub-components corresponding to the target cavity, the target cavity sub-components for each device are given.

[0059] Furthermore, the direct process energy consumption of cavity j is specifically expressed as follows:

[0060] Among them, E direct,j T represents the direct process energy consumption of cavity j. process,j For cavity j, the process duration corresponding to the target process of the production process parameters is P. recipe (t) represents the process power curve of the target process, indicating the time-varying power setpoint in the production process parameters.

[0061] Furthermore, the energy consumption for the hot restart of cavity j is specifically expressed as follows: ; ; Among them, E thermal,j For the heat restart energy consumption of cavity j, c p The equivalent specific heat capacity, representing the overall specific heat capacity of the cavity thermal mass system, m chamber Equivalent thermal mass represents the total mass of the cavity walls, carrier, and wafer involved in the temperature cycle within the cavity. For the temperature rise requirement of cavity j, η represents the temperature difference when the current temperature is insufficient. heater The heater efficiency is represented by the efficiency coefficient for converting electrical energy into heat energy. This is a hot restart indicator function, indicating that a full restart is triggered when the temperature difference exceeds a temperature threshold δT; otherwise, a warm-start operation is initiated. process T represents the processing temperature in the production process parameters. current,j Let be the current temperature of cavity j.

[0062] Furthermore, the energy consumption penalty for process switching in cavity j is specifically expressed as follows: ; ; Among them, E switch,j For the energy consumption penalty of process switching of cavity j, P idle τ represents the standby power consumption, indicating the basic power consumption of the cavity in the idle state. purge,j Let γ be the purification steady-state time of cavity j, and γ represent the forced waiting time based on the lookup table of compatibility between the preceding and following gas formulations. match,j Let be the formula matching coefficient for cavity j. The closer the formula matching coefficient is to 1, the better the compatibility and the lower the switching cost. exp() is an exponential function, α is the gas difference sensitivity, used to control the decay rate of the formula matching coefficient with gas type differences, and D gas The gas atmosphere distance represents the weighted norm of the differences in gas type, flow rate, and pressure between the two target processes. (Recipe) last,j For the target process previously executed in cavity j, Recipe target The target process is represented by the process formula required for the wafer to be processed.

[0063] By constructing a cost model that includes direct process energy consumption, hot restart penalty, and process switching energy consumption penalty, the selection of the target cavity is optimized. Based on deterministic optimization, the exploration capability is maintained by superimposing mean-reduced Poisson noise, thus achieving a Pareto balance between optimal energy consumption and system robustness.

[0064] Noise processing is performed by removing the mean Poisson noise to ensure that the noise does not systematically bias the optimization direction, only introducing minor perturbations to break the symmetry. A hot restart indicator function is used to avoid unnecessary penalty calculations triggered by minor temperature fluctuations, thus improving numerical stability. The formula matching coefficient adopts an exponential decay form, ensuring that the switching penalty increases non-linearly and sharply as gas differences increase, conforming to empirical patterns regarding actual purification difficulty.

[0065] In one specific implementation, based on the collected production process parameters and commissioning parameters, an equipment selection strategy is provided. This strategy allows for the selection of initial equipment components from a component library that meet the production process and commissioning parameters. By analyzing the basic and operational parameters of each piece of equipment, the production operation mode of that equipment can be determined. Production operation modes include discrete and continuous, or parallel and serial. If a piece of equipment operates in a discrete mode, it means that the cavities within that equipment operate in parallel; conversely, if it operates in a continuous mode, it means that the cavities within that equipment operate serially, meaning there is an inherent processing sequence between the cavities. Based on the production operation mode of each piece of equipment, a cavity selection strategy is provided, allowing selection of a single cavity or all cavities within the equipment simultaneously. According to the cavity selection strategy, the target cavity sub-components for each piece of equipment are determined, and the initial equipment components are updated to form the target equipment components. In a specific example, based on the cavity selection strategy, various cavities or cavity groups that meet the production process parameters and commissioning parameters can be selected from the equipment to obtain target cavity sub-components. The initial equipment components contain all available cavities. The initial equipment components are updated using the target cavity sub-components, by removing unused cavities to form the target equipment components. Finally, the processing parameters of the target cavity sub-components are used to match processing logic components and energy-consuming logic components to determine multiple target logic components corresponding to each target equipment component. The target logic components include complete operational logic that allows the target equipment components to operate according to the energy-consuming logic in the energy-consuming logic component and the processing logic in the processing logic component.

[0066] S104: Integrate the target device components and the corresponding multiple target logic components to form the target standard component.

[0067] The target logic component and the target device component are integrated. The target device component is processed according to the complete operating logic in the target logic component to form the target standard component.

[0068] S105: Based on factory layout parameters, link target standard components to build a digital twin model of semiconductor factory equipment.

[0069] Specifically, based on factory layout parameters, target standard components are linked to construct a digital twin model of semiconductor factory equipment, referring to... Figure 7 Specifically, it includes: The initial positions of each target standard component are determined by the factory building layout; Based on the pipeline network layout, analyze and match the equipment process parameters in the target standard components, and adjust the initial position of the target standard components. Based on the logistics route layout and combined with the target operation logic of the target standard components, the various target standard components are linked to form a digital twin model of semiconductor factory equipment.

[0070] In one specific implementation, each target standard component corresponds to a device. Based on the factory layout parameters, the initial positions of each target standard component can be obtained. Each target standard component includes device process parameters, representing the process requirements of each device, such as gas type, flow rate, pressure, and vibration level. The pipeline network layout includes the access points for each process requirement in the semiconductor factory. Based on the device process parameters of the target standard components, the initial positions are fine-tuned to ensure that the process requirements of each target standard component do not conflict, guaranteeing the normal operation of each target standard component. The logistics path layout represents the overall processing and operational logic chain of the semiconductor factory. Finally, based on the logistics path layout and the target operational logic within the target standard components, the target operational logic between each target standard component is connected, ultimately forming a digital twin model of the semiconductor factory equipment.

[0071] The method for constructing digital twin models of semiconductor factory equipment in batches also includes updating the semiconductor factory equipment data twin model, specifically including: Based on the energy consumption parameters of the semiconductor factory equipment data twin model, set energy consumption optimization targets; Based on the current structured parameter set, and combined with the energy optimization target, the structured parameter set is updated to obtain multiple sets of demand data. By running a data twin model of semiconductor factory equipment using multiple sets of demand data, the optimal set of demand data is obtained. By combining the optimal demand data set, update the production process parameters and / or factory layout parameters of the semiconductor factory equipment data twin model.

[0072] Understandably, the aforementioned content constructs a semiconductor factory equipment data twin model, including multiple target standard components, based on a series of data such as modeling requirements. This data twin model can process and operate according to the collected parameters, but its operation is not optimal at this stage. Based on the model's operation, the structured parameter set can be updated to obtain the required data set, and the model's operating parameters can be adjusted to optimize its operation.

[0073] In a specific example, the current structured parameter set includes equipment process parameters (RF power, cavity pressure, gas flow rate, processing time), factory layout parameters (cooling water riser number, pipe diameter, supply margin), and energy consumption parameters (electricity 2.1MW, cooling water 380 m³ / h, nitrogen 4500 SLM). Energy optimization targets are set based on actual conditions; for example, these targets include total cooling water ≤ 350 m³ / h, electrical power ≤ 2.0MW (avoiding transformer capacity expansion), and capacity loss ≤ 2%. By minimizing the peak values ​​of energy consumption parameters, and constraining wafer output decline to less than 2%, the data in the structured parameter set is updated and optimized, resulting in multiple sets of demand data. Based on each set of demand data, a semiconductor factory equipment data twin model is run. Combining the energy optimization targets, the optimal demand data set is selected from the multiple sets of demand data, and the production process parameters and / or factory layout parameters of the semiconductor factory equipment data twin model are updated.

[0074] Reference Figure 8 This invention provides a device for constructing digital twin models of semiconductor factory equipment in batches, comprising: The template acquisition module 201 is used to match the preset basic parameter template library based on the determined modeling requirements and provide the corresponding basic parameter template. The parameter determination module 202 is used to determine a structured parameter set based on the basic parameter template. The structured parameter set includes basic equipment parameters, production process parameters, and factory layout parameters. The component determination module 203 is used to analyze the basic parameters of the equipment and the production process parameters, and in combination with the mapping relationship with the component library, traverse the equipment components and logic components in the component library to determine each target equipment component and its corresponding multiple target logic components. The component fusion module 204 is used to fuse the target device components and multiple corresponding target logic components to form a target standard component. Model building module 205 is used to link target standard components based on factory layout parameters to build a digital twin model of semiconductor factory equipment.

[0075] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the described module can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0076] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention. Clearly, those skilled in the art can make various alterations and modifications to the invention without departing from its spirit and scope. Thus, if these modifications and variations of the invention fall within the scope of the claims and their equivalents, the invention is also intended to include these modifications and variations.

Claims

1. A method for constructing digital twin models of semiconductor factory equipment in batch production, characterized in that, Specifically, the steps include the following: Based on the defined modeling requirements, a pre-defined basic parameter template library is matched to provide the corresponding basic parameter template; Based on the basic parameter template, a structured parameter set is determined, which includes basic equipment parameters, production process parameters, and factory layout parameters. Analyze the basic parameters of the equipment and the production process parameters, and combine them with the mapping relationship with the component library. Traverse the equipment components and logic components in the component library to determine each target equipment component and its corresponding multiple target logic components. Integrate the target device components and multiple corresponding target logic components to form the target standard component; Based on factory layout parameters, target standard components are linked to construct a digital twin model of semiconductor factory equipment.

2. The method for constructing a batch of digital twin models of semiconductor factory equipment as described in claim 1, characterized in that, The basic parameters of the equipment include the hardware configuration parameters of the equipment; the production process parameters include the operating parameters of the equipment, the energy consumption parameters of the equipment, the production process parameters, and the commissioning parameters; and the factory layout parameters include the layout of pipeline network locations, the layout of factory buildings, and the layout of logistics routes.

3. The method for constructing a batch of digital twin models of semiconductor factory equipment as described in claim 1, characterized in that, The component library includes device components and logic components. The device components include multiple cavity sub-components, and the logic components include processing logic components and energy-consuming logic components. The construction of the component library includes the following steps: Obtain the basic cavity parameters of each device, build multiple corresponding cavity sub-components, and form a device component; Determine the processing parameter interface and processing control logic interface, set the cavity invocation strategy, and construct the processing logic component. The cavity invocation strategy includes the processing parameters received by the processing parameter interface based on the processing control logic call received by the self-processing control logic interface. Build plant management sub-components for each energy consumption type, determine the association function between each plant management sub-component and the processing parameters, and combine them with the energy consumption logic to construct the energy consumption logic component; It gathers equipment components, processing logic components, and energy consumption logic components to build a component library.

4. The method for constructing a batch of digital twin models of semiconductor factory equipment as described in claim 3, characterized in that, Build plant management sub-components for each energy consumption type, determine the association function between each plant management sub-component and processing parameters, and construct the energy consumption logic component in conjunction with the energy consumption logic. The specific steps include the following: Based on the energy consumption type, construct the corresponding plant management sub-components and basic energy consumption data; Provide the association function between each plant sub-component and the processing parameters; By integrating the correlation function and various processing parameters, the steady-state energy consumption components are determined; Determine the dynamic correction coefficients corresponding to the jumps in each processing parameter, and provide dynamic energy consumption breakdown; By superimposing the basic energy consumption base, steady-state energy consumption components, dynamic energy consumption components, and random error terms, the energy consumption function for each energy consumption type is given. Verify the energy consumption functions for each energy consumption type and construct energy consumption logic components.

5. The method for constructing a batch of digital twin models of semiconductor factory equipment as described in claim 4, characterized in that, The energy consumption function for each energy consumption type is verified, specifically including the following steps: Construct a regularized loss function that integrates the least squares residual sum of squares and physical prior constraints; Obtain historical processing parameters, historical energy consumption data, and initial dynamic correction coefficients; set the learning rate and regularization coefficients. Based on the initial dynamic correction coefficients and historical processing parameters, the simulated energy consumption data is given through the energy consumption function; Error analysis is performed using a regularized loss function based on simulated energy consumption data and historical energy consumption data. Adjust the initial dynamic correction coefficients in the negative gradient direction, and continue to perform error analysis using the regularized loss function until convergence, thus obtaining the dynamic correction coefficients and completing the verification of the energy consumption function for each energy consumption type.

6. The method for constructing a batch of digital twin models of semiconductor factory equipment as described in claim 3, characterized in that, Analyze the basic parameters of the equipment and the production process parameters, and combine them with the mapping relationship with the component library. Traverse the equipment components and logic components in the component library to determine each target equipment component and its corresponding multiple target logic components, specifically including: Based on production process parameters and commissioning parameters, we provide equipment selection strategies; Based on the device selection strategy, traverse the device components in the component library to determine each initial device component; Based on the basic and operating parameters of each piece of equipment, determine the production and operation mode of the corresponding equipment. Based on the production and operation mode of each piece of equipment, a cavity selection strategy for the corresponding equipment is given; Based on the cavity selection strategy, the target cavity sub-components for each device are given; The initial device components are updated based on the target cavity sub-components to form the target device component; Based on the processing parameters of the target cavity sub-component, the processing logic component and the power logic component are matched to determine multiple target logic components corresponding to each target device component.

7. The method for constructing a batch of digital twin models of semiconductor factory equipment as described in claim 6, characterized in that, Based on the cavity selection strategy, the target cavity sub-components for each device are given, specifically including: Based on the real-time status of each cavity in each device, the number of available cavities in real time is given, and the real-time cavity queue is determined; Random probabilities are generated based on a Poisson distribution, and a target cavity is selected from the real-time cavity queue, specifically as follows: ； Among them, C selected For the target cavity, E direct,j E represents the direct process energy consumption of cavity j. thermal,j For the heat restart energy consumption of cavity j, E switch,j The energy consumption penalty for process switching of cavity j is denoted by τ, where j is the cavity number in the real-time cavity queue, and τ is the noise weighting coefficient. To remove mean Poisson noise, Ω idle For real-time cavity queues; Based on the cavity sub-components corresponding to the target cavity, the target cavity sub-components for each device are given.

8. The method for constructing a batch of digital twin models of semiconductor factory equipment as described in claim 2, characterized in that, Based on factory layout parameters, target standard components are linked to construct a digital twin model of semiconductor factory equipment, specifically including: The initial positions of each target standard component are determined by the factory building layout; Based on the pipeline network layout, analyze and match the equipment process parameters in the target standard components, and adjust the initial position of the target standard components. Based on the logistics route layout and combined with the target operation logic of the target standard components, the various target standard components are linked to form a digital twin model of semiconductor factory equipment.

9. The method for constructing a batch of digital twin models of semiconductor factory equipment as described in claim 1, characterized in that, This also includes updating the data twin model of semiconductor factory equipment, specifically including: Based on the energy consumption parameters of the semiconductor factory equipment data twin model, set energy consumption optimization targets; Based on the current structured parameter set, and combined with the energy optimization target, the structured parameter set is updated to obtain multiple sets of demand data. By running a data twin model of semiconductor factory equipment using multiple sets of demand data, the optimal set of demand data is obtained. By combining the optimal demand data set, update the production process parameters and / or factory layout parameters of the semiconductor factory equipment data twin model.

10. A device for constructing digital twin models of semiconductor factory equipment in batch production, characterized in that, The method for constructing a batch-scale digital twin model of semiconductor factory equipment as described in any one of claims 1-9 includes: The template acquisition module is used to match the preset basic parameter template library based on the determined modeling requirements and provide the corresponding basic parameter template. The parameter determination module is used to determine a structured parameter set based on the basic parameter template. The structured parameter set includes basic equipment parameters, production process parameters, and factory layout parameters. The component determination module is used to analyze the basic parameters of the equipment and the production process parameters, and combine them with the mapping relationship with the component library to traverse the equipment components and logic components in the component library to determine each target equipment component and its corresponding multiple target logic components. The component fusion module is used to fuse target device components and multiple corresponding target logic components to form a target standard component. The model building module is used to link target standard components based on factory layout parameters to build a digital twin model of semiconductor factory equipment.

Images