AI adaptive control method and system for wafer heating

Abstract

The invention provides an AI adaptive control method and system for wafer heating, and the method comprises the steps: firstly constructing a dynamic state field which comprises a real-time operation state field of a heating device and an instantaneous state field of a wafer, and forming a space coupling relation through heat transfer; thirdly, performing multi-scale field analysis on the dynamic state field by utilizing a self-evolution control model, extracting a field feature interaction rule and generating a field feature coupling matrix; and then, based on the field feature coupling matrix and in combination with thermal evolution pre-judgment logic, generating dynamic regulation and control parameters which are in space-time adaptation with the dynamic state field, loading the dynamic regulation and control parameters to a heating execution system to trigger a heating process, synchronously capturing evolution trajectories of the real-time operation state field and the instantaneous state field, and fusing the evolution trajectories to form field evolution feedback. Finally, field evolution feedback is input into the self-evolution control model, a field correction factor is generated to iteratively optimize the dynamic regulation and control parameters, target regulation and control parameters matched with the current dynamic state field are obtained, and accurate self-adaptive control over wafer heating is achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of artificial intelligence, in particular to an AI adaptive control method and system for wafer heating. BACKGROUND

[0002] In the field of semiconductor manufacturing, wafer heating is a key link in the chip manufacturing process, and its heating precision and stability directly affect the quality and performance of the chip. The traditional wafer heating control method is mainly based on fixed control parameters and simple feedback mechanism. For example, the running of the heating device is controlled by a pre-set temperature curve, and only the deviation between the real-time temperature value fed back by the temperature sensor and the pre-set value is adjusted by PID (Proportion-Integral-Derivative). However, this method has obvious limitations. Since the wafer heating process is affected by many complex factors, such as the performance difference of different areas of the heating device, the non-uniformity of the physical properties of the wafer itself, and the dynamic changes in the heat transfer process, etc., it is difficult for fixed control parameters to adapt to the real-time changes in the heating process, resulting in uneven heating, large temperature fluctuations, etc., which in turn affects the yield and performance of the chip. Therefore, a control method that can adapt to the dynamic process of wafer heating is needed to improve the precision and stability of heating. SUMMARY

[0003] In view of the above-mentioned problems, in combination with the first aspect of the present application, the embodiments of the present application provide an AI adaptive control method for wafer heating, which comprises: constructing a dynamic state field of wafer heating, which includes a real-time running state field of the heating device and an instantaneous state field of the wafer, and the two state fields are spatially coupled through heat transfer; performing multi-scale field analysis on the dynamic state field by a self-evolution control model, extracting field feature interaction rules at different spatial scales, and generating a field feature coupling matrix; based on the field feature coupling matrix, combining the heat evolution prediction logic built-in the self-evolution control model, generating dynamic regulation and control parameters that are spatio-temporally adapted to the dynamic state field; loading the dynamic regulation and control parameters to the heating execution system to trigger the wafer heating process, synchronously capturing the evolution trajectory of the real-time running state field and the response trajectory of the instantaneous state field during the heating process, and fusing to form a field evolution feedback; inputting the field evolution feedback into the self-evolution control model, driving the self-evolution control model to generate a field correction factor based on the deviation between the field feature coupling matrix and the field evolution feedback, iteratively optimizing the dynamic regulation and control parameters through the field correction factor, and obtaining target regulation and control parameters that are adapted to the current dynamic state field.

[0004] In another aspect, embodiments of the present invention also provide an AI adaptive control system for wafer heating, including a processor and a machine-readable storage medium connected to the processor. The machine-readable storage medium is used to store programs, instructions, or code, and the processor is used to execute the programs, instructions, or code in the machine-readable storage medium to implement the above-described method.

[0005] Based on the above, this embodiment of the invention constructs a dynamic state field that includes the real-time operating state field of the heating device and the instantaneous state field of the wafer. By utilizing a self-evolutionary control model for multi-scale field analysis, it can comprehensively and accurately capture the interaction patterns of field features at different spatial scales during the heating process, generating a field feature coupling matrix. Based on this matrix, combined with dynamically adjusted parameters generated by thermal evolution prediction logic, spatiotemporal adaptation is achieved with the dynamic state field, effectively improving the accuracy of heating control. During the heating process, the evolution trajectories of the real-time operating state field and the instantaneous state field are simultaneously captured and fused to form field evolution feedback. Then, the self-evolutionary control model generates field correction factors to iteratively optimize the dynamic adjustment parameters. This allows for real-time and dynamic adjustment of heating parameters, quickly adapting to various changes during the heating process, ensuring the uniformity and stability of wafer heating, thereby significantly improving the quality and yield of chip manufacturing. Attached Figure Description

[0006] Figure 1 This is a schematic diagram of the execution flow of the AI ​​adaptive control method for wafer heating provided in an embodiment of the present invention.

[0007] Figure 2 This is a schematic diagram of exemplary hardware and software components of an AI adaptive control system for wafer heating provided in an embodiment of the present invention. Detailed Implementation

[0008] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating an AI adaptive control method for wafer heating according to an embodiment of the present invention. The following is a detailed description of the AI ​​adaptive control method for wafer heating.

[0009] Step S110: Construct a dynamic state field for wafer heating. The dynamic state field includes the real-time operating state field of the heating device and the instantaneous state field of the wafer. The real-time operating state field of the heating device and the instantaneous state field of the wafer form a spatial coupling relationship through heat transfer.

[0010] This embodiment forms the foundation of the entire wafer heating adaptive control method, aiming to establish a dynamic state field that comprehensively reflects the state of the heating device and the wafer, as well as their interaction. This provides data support for subsequent field analysis, control parameter generation, and feedback optimization. The core of constructing the dynamic state field lies in integrating the operating state of the heating device with the real-time state of the wafer through the physical relationship of heat transfer, forming a unified state field that can be updated in real time and contains spatial interaction information. Specifically, the real-time operating state field is a spatial representation of the operating parameters of the heating device in each region at the current moment, covering parameters directly related to heating capacity, such as the power output of each heating module, surface temperature, and resistance changes of heating elements. The instantaneous state field is a spatial representation of the thermal state of the wafer in each region at the current moment, covering parameters directly related to the wafer's thermal response, such as surface temperature, internal temperature gradient, and thermal stress distribution in each region of the wafer. Heat transfer, as a physical link connecting two state fields, has a spatial coupling relationship where changes in the heat output of a certain area of ​​the heating device are transferred to the corresponding area of ​​the wafer through heat conduction, heat radiation, or heat convection, causing changes in the thermal state of the corresponding area of ​​the wafer. In turn, changes in the thermal state of the corresponding area of ​​the wafer affect the operating state of the corresponding area of ​​the heating device through thermal feedback (such as heat radiation reflection and changes in thermal resistance). This interaction exhibits a specific distribution pattern in space, that is, the interaction intensity in areas where the heating device is in direct contact with or close to the wafer is higher than that in areas where it is not in contact or far away.

[0011] Step S111: Collect the state parameters of each area of ​​the heating device during operation, and map the collected state parameters into a grid according to the spatial distribution law to form a real-time operating state field with spatial gradient characteristics. Each grid cell corresponds to the state representation of a spatial location.

[0012] In this embodiment, the state parameters of the heating device include, but are not limited to, parameters directly related to the operating state of the heating device, such as the power output of each heating module, the surface temperature of each area, the current value and voltage value of the heating module, and the resistance value of the heating element. The data acquisition process is achieved through sensors distributed in different areas of the heating device. The sensor placement must cover all major heating areas and key structural parts of the heating device, specifically including the central heating area, edge heating areas, and connection points of heating modules. Each sensor corresponds to a specific spatial location on the heating device to ensure that the acquired state parameters comprehensively reflect the operating state of the heating device. After acquiring the state parameters, the device is divided into grids according to its physical spatial structure. The size of the grid cells needs to be determined based on the structural dimensions and control accuracy requirements of the heating device. Generally, the higher the control accuracy requirement, the smaller the grid cell size. The shape of the grid cells must be adapted to the physical structure of the heating device. For a circular heating device, a fan-shaped grid cell can be used; for a rectangular heating device, a rectangular grid cell can be used. Each grid cell corresponds to a specific spatial location on the heating device. The state parameters collected at that location are assigned to the corresponding grid cell. If multiple sensors are arranged within a grid cell, the average value of the state parameters collected by the multiple sensors is taken as the state representation of that grid cell, thus forming a real-time operating state field. For example, for a rectangular heating device, it can be divided into several rectangular grid cells of the same size. Each grid cell corresponds to a rectangular area on the heating device, and the state parameters within that area are the state representation of that grid cell.

[0013] Step S112: Collect the state parameters of each region of the wafer at the current moment, and perform spatial discretization processing according to the physical structure of the wafer to form an instantaneous state field containing the differences in regional states. The discrete unit corresponds one-to-one with the physical partition of the wafer.

[0014] In this embodiment, the wafer's state parameters mainly include surface temperature, internal temperature, thermal expansion, thermal stress, thermal conductivity, and other parameters related to the wafer's thermal state in various regions of the wafer. The data acquisition process is achieved through non-contact temperature sensors, thermal imaging equipment, infrared thermometers, etc., to avoid contamination or damage to the wafer. The sensor placement must cover all major areas of the wafer, specifically including the central region, edge regions, and regions near the wafer notch, with each sensor corresponding to a specific spatial location on the wafer. After acquiring the state parameters, spatial discretization is performed according to the wafer's physical structure. The division of discrete units must match the actual physical partitions of the wafer, specifically including the central physical partition, edge physical partitions, and physical partitions near the notch, with each physical partition corresponding to a discrete unit. The size of the discrete unit needs to be determined based on the wafer's diameter and control precision requirements; the higher the control precision requirement, the smaller the size of the discrete unit. The state parameters collected in each physical partition are assigned to the corresponding discrete unit. If multiple sensors are arranged in a physical partition, the average value of the state parameters collected by the multiple sensors is taken as the state characterization of the discrete unit, thereby forming an instantaneous state field. This instantaneous state field can clearly reflect the state differences between different regions of the wafer.

[0015] Step S113: Analyze the spatial overlap region between the real-time operating state field and the instantaneous state field, determine the main heat transfer path and transfer efficiency characteristics, and analyze the obtained main heat transfer path and transfer efficiency characteristics to reflect the spatial interaction intensity between the real-time operating state field and the instantaneous state field.

[0016] In this embodiment, the spatial range of the real-time operating state field corresponds to the physical space of the heating device, and the spatial range of the instantaneous state field corresponds to the physical space of the wafer. Since the wafer is placed on the heating device for heating, there is an overlapping area between the two spatial ranges, specifically the projection area of ​​the wafer on the heating device. When analyzing the spatial overlap area, it is necessary to match the spatial positions of the grid cells of the real-time operating state field and the discrete cells of the instantaneous state field. Specifically, the coordinates of the grid cells of the real-time operating state field and the coordinates of the discrete cells of the instantaneous state field are transformed to the same spatial coordinate system, and the spatial overlap area between each grid cell and each discrete cell is calculated. If the overlap area is greater than a preset threshold, it is considered that the grid cell and the discrete cell are in the same spatial position or have a direct spatial correspondence. The main path of heat transfer refers to the main channel through which heat is transferred from the heating device to the wafer, specifically including the area in direct contact between the heating device and the wafer, the main direction of heat radiation from the heating device to the wafer, and the main flow direction of heat convection between the heating device and the wafer. The spatial overlap area between the grid cells and discrete cells on the main path is large, and the changes in the state parameters are strongly correlated. Heat transfer efficiency characteristics reflect the efficiency of heat transfer along the main path, specifically including thermal conduction efficiency, thermal radiation efficiency, and thermal convection efficiency. Thermal conduction efficiency is calculated using the contact thermal resistance between the heating device and the wafer; the lower the contact thermal resistance, the higher the thermal conduction efficiency. Thermal radiation efficiency is calculated using the surface emissivity and surface temperature difference between the heating device and the wafer; the higher the surface emissivity and the greater the surface temperature difference, the higher the thermal radiation efficiency. Thermal convection efficiency is calculated using the air velocity and air thermal conductivity between the heating device and the wafer; the faster the air velocity and the higher the air thermal conductivity, the higher the thermal convection efficiency. Heat transfer efficiency characteristics are determined by analyzing the relationship between the state parameters of the heating device and the state parameters of the wafer. For example, under the same heating device power output, the faster the temperature rises in a certain region of the wafer, the higher the heat transfer efficiency of that region.

[0017] Step S114: Based on the main path and efficiency characteristics of heat transfer, construct the inter-field correlation function. The inter-field correlation function can quantify the influence of any grid cell in the real-time running state field on the corresponding discrete cell in the instantaneous state field.

[0018] In this embodiment, the construction of the inter-field correlation function needs to comprehensively consider the main path of heat transfer and the characteristics of heat transfer efficiency. First, the spatial correspondence between each grid cell in the real-time operating state field and each discrete cell in the instantaneous state field is determined. For grid cells and discrete cells on the main path, the spatial correspondence is direct; for grid cells and discrete cells on non-main paths, the spatial correspondence is indirect. Then, the heat transfer efficiency between each grid cell and each discrete cell is calculated. For directly corresponding grid cells and discrete cells, the heat transfer efficiency is the transfer efficiency on the main path; for indirectly corresponding grid cells and discrete cells, the heat transfer efficiency is the transfer efficiency on the non-main path. Finally, the inter-field correlation function is constructed based on the spatial correspondence and heat transfer efficiency. The form of the inter-field correlation function is the product of the state parameter of the grid cell and the heat transfer efficiency, that is, the value of the inter-field correlation function is equal to the state parameter of the grid cell multiplied by the corresponding heat transfer efficiency. The larger the value of the inter-field correlation function, the higher the influence of the grid cell on the discrete cell.

[0019] Step S1141: Extract the heat output characteristics of each grid cell in the real-time running state field and the heat reception characteristics of each discrete cell in the instantaneous state field. The heat output characteristics reflect the heat release capability of the corresponding grid cell, and the heat reception characteristics reflect the heat absorption capability of the corresponding discrete cell.

[0020] In this embodiment, the thermal output characteristics of each grid cell in the real-time operating state field include parameters such as the power density of the heating module corresponding to that grid cell, the surface temperature gradient, and the thermal emissivity of the heating element. These parameters are acquired by sensors distributed on and inside the heating module. The thermal receiving characteristics of each discrete cell in the instantaneous state field include parameters such as the thermal conductivity, specific heat capacity, and surface emissivity of the wafer region corresponding to that discrete cell. These parameters are acquired through a wafer material property database and non-contact detection equipment. The thermal output characteristics of each grid cell must correspond one-to-one with the spatial location of that grid cell, and the thermal receiving characteristics of each discrete cell must correspond one-to-one with the spatial location of that discrete cell.

[0021] Step S1142: Calculate the spatial distance and thermal resistance parameter between each grid cell and each discrete cell. The spatial distance and thermal resistance parameter together determine the degree of heat transfer attenuation.

[0022] In this embodiment, the spatial distance is calculated based on the spatial coordinate system of the heating device and the wafer. The center coordinates of each grid cell and the center coordinates of each discrete cell are substituted into the Euclidean distance formula. The calculation of the thermal resistance parameter needs to consider the medium type (such as air, vacuum, contact interface, etc.) between the grid cell and the discrete cell. Different medium types correspond to different thermal resistance coefficients. The thermal resistance parameter is equal to the thermal resistance coefficient of the medium multiplied by the spatial distance. For example, if the medium between the grid cell and the discrete cell is air, the thermal resistance coefficient of air is a fixed value, and the spatial distance is a calculated value, then the thermal resistance parameter is the product of the two.

[0023] Step S1143: Based on heat output characteristics, heat reception characteristics, spatial distance and thermal resistance parameters, construct basic correlation factors. The basic correlation factors initially quantify the influence potential of grid cells on discrete cells.

[0024] In this embodiment, the construction of the basic correlation factor requires integrating heat output characteristics, heat receiving characteristics, spatial distance, and thermal resistance parameters. The specific calculation logic is as follows: the basic correlation factor equals the value of the heat output characteristic multiplied by the value of the heat receiving characteristic, then divided by the product of the spatial distance and the thermal resistance parameter. For example, if the heat output characteristic value of a certain grid cell is A, the corresponding discrete cell's heat receiving characteristic value is B, the spatial distance between them is C, and the thermal resistance parameter is D, then the basic correlation factor is the result of A multiplied by B divided by the result of C multiplied by D.

[0025] Step S1144 introduces the weight coefficient of the main heat transfer path to modify the basic correlation factor, thereby strengthening the correlation factor between the grid cells and discrete cells on the main path.

[0026] In this embodiment, the weighting coefficient of the primary heat transfer path is determined according to the type of heat transfer path. For heat conduction paths, the weighting coefficient is set to a higher value; for heat radiation paths, the weighting coefficient is set to a medium value; and for heat convection paths, the weighting coefficient is set to a lower value. The specific value of the weighting coefficient is determined by the proportion of heat transfer contribution of different paths during historical heating. The weighting coefficient of the combination of grid cells and discrete cells on the primary path is the weighting coefficient of the corresponding path type; the weighting coefficient of the combination of grid cells and discrete cells on non-primary paths is 0.5 times the weighting coefficient of the corresponding path type. The corrected correlation factor is equal to the basic correlation factor multiplied by the corresponding weighting coefficient.

[0027] Step S1145: Based on the transmission efficiency characteristics, the modified correlation factor is dynamically adjusted. When the transmission efficiency value is large, the value of the correlation factor increases accordingly, ultimately forming an inter-field correlation function that can accurately quantify the degree of influence.

[0028] In this embodiment, the transfer efficiency characteristics include thermal conduction efficiency, thermal radiation efficiency, and thermal convection efficiency. These efficiency values ​​are calculated using parameters such as the temperature difference between the heating device and the wafer, and the medium flow rate, which are collected in real time. The dynamic adjustment logic is as follows: the final value of the inter-field correlation function is equal to the corrected correlation factor multiplied by the value of the transfer efficiency characteristic. For example, if the corrected correlation factor is E and the value of the transfer efficiency characteristic is F, then the final value of the inter-field correlation function is the result of E multiplied by F.

[0029] Step S115: Spatial coupling of the real-time running state field and the instantaneous state field is achieved through the inter-field correlation function, so that the corresponding units in the real-time running state field and the instantaneous state field form a dynamic correlation and jointly constitute a dynamic state field containing spatial coupling relationship.

[0030] In this embodiment, the state parameters of each grid cell in the real-time operating state field are substituted into the inter-field correlation function to calculate the influence of that grid cell on the corresponding discrete cell in the instantaneous state field. This influence value is then added to the state parameters of the corresponding discrete cell in the instantaneous state field to obtain the coupled discrete cell state parameters. If a discrete cell corresponds to multiple grid cells, the influence values ​​of the multiple grid cells on that discrete cell are added together, and then added to the initial state parameters of that discrete cell to obtain the coupled discrete cell state parameters. In this way, the corresponding cells in the real-time operating state field and the instantaneous state field form a dynamic correlation, jointly constituting the dynamic state field. For example, the state parameter of grid cell A in the real-time operating state field is a certain value, the corresponding heat transfer efficiency is a certain value, and the inter-field correlation function value is the product of the two; the initial state parameter of discrete cell B in the instantaneous state field is a certain value, and the coupled discrete cell B state parameters are the initial state parameter plus the inter-field correlation function value.

[0031] Step S120: Perform multi-scale field analysis on the dynamic state field through the self-evolutionary control model, extract the field feature interaction law under different spatial scales, and generate the field feature coupling matrix.

[0032] In this embodiment, the aim is to extract feature interaction patterns at different spatial scales from a dynamic state field. The self-evolutionary control model is a deep learning-based model with multi-scale analytical capabilities, capable of automatically learning and extracting features at different spatial scales.

[0033] Step S121: Input the dynamic state field into the multi-scale analysis module of the self-evolution control model. The multi-scale analysis module performs hierarchical analysis of the dynamic state field according to the preset spatial scale level. Each spatial scale level corresponds to a spatial scale range.

[0034] In this embodiment, the multi-scale analysis module is an important component of the self-evolutionary control model. Its preset spatial scale levels include, but are not limited to, macroscopic, mesoscopic, and microscopic scales. The macroscopic scale corresponds to a large spatial range, specifically the entire heating device or the entire wafer; the mesoscopic scale corresponds to a medium spatial range, specifically a heating module of the heating device or a physical partition of the wafer; and the microscopic scale corresponds to a small spatial range, specifically a heating element of the heating device or a grain of the wafer. After the dynamic state field is input into the multi-scale analysis module, the module performs layered analysis of the dynamic state field according to the preset spatial scale levels, extracting features at different spatial scales. For example, at the macroscopic scale level, features such as the total power output of the entire heating device and the average temperature of the entire wafer are extracted; at the mesoscopic scale level, features such as the power output value of the heating module and the temperature value of the physical partition of the wafer are extracted; and at the microscopic scale level, features such as the resistance value of the heating element and the temperature value of the wafer grain are extracted.

[0035] Step S122: At each spatial scale level, extract the scale features of the real-time running state field and the scale features of the instantaneous state field. The scale features include the field distribution uniformity and gradient change features at the corresponding spatial scale level.

[0036] In this embodiment, at the macroscopic level, the scale characteristics of the real-time operating state field include the uniformity of power output distribution and surface temperature distribution of the entire heating device. The calculation method is the difference between the maximum and minimum power output values ​​divided by the average value; the smaller the difference, the higher the distribution uniformity. The scale characteristics of the instantaneous state field include the uniformity of temperature distribution and thermal expansion distribution of the entire wafer. The calculation method is the difference between the maximum and minimum temperature values ​​divided by the average value; the smaller the difference, the higher the distribution uniformity. At the mesoscopic level, the scale characteristics of the real-time operating state field include the power output gradient and surface temperature gradient of the heating module. The calculation method is the difference in power output between the central and edge regions of the heating module divided by the distance; the larger the difference, the more pronounced the gradient change characteristics. The scale characteristics of the instantaneous state field include the temperature gradient and thermal stress gradient of the wafer's physical partitions. The calculation method is the temperature difference between the central and edge regions of the physical partition divided by the distance; the larger the difference, the more pronounced the gradient change characteristics. At the microscale level, the scale characteristics of the real-time operating state field include the resistance change gradient of the heating element and the surface temperature change gradient. The calculation method is to divide the resistance difference between one end and the other end of the heating element by the length. The larger the difference, the more obvious the gradient change characteristics. The scale characteristics of the instantaneous state field include the temperature change gradient of the wafer grains and the thermal expansion change gradient. The calculation method is to divide the temperature difference between one end and the other end of the grains by the length. The larger the difference, the more obvious the gradient change characteristics.

[0037] Step S123: Through the feature interaction module of the self-evolutionary control model, analyze the interaction mode between the real-time operating state field scale features and the instantaneous state field scale features at the same spatial scale level, and obtain the field feature interaction law at the corresponding spatial scale level.

[0038] In this embodiment, the feature interaction module is one of the core modules of the self-evolutionary control model, capable of analyzing the interaction patterns between different features. At the same spatial scale level, the scale features of the real-time operating state field and the instantaneous state field are input into the feature interaction module. This module analyzes the interaction patterns by calculating the correlation coefficient between the two. The correlation coefficient is calculated by dividing the covariance of the two features by the product of their standard deviations. The larger the absolute value of the correlation coefficient, the stronger the interaction between the two features. In this way, the field feature interaction laws at the corresponding spatial scale level are obtained. For example, at the macroscopic scale level, the correlation coefficient between the power output distribution uniformity of the real-time operating state field and the temperature distribution uniformity of the instantaneous state field is high, indicating a strong interaction between the two features.

[0039] Step S124: Perform cross-scale correlation analysis on the field feature interaction patterns at different spatial scale levels to identify the feature transmission paths between scales, i.e., the way in which the interaction patterns at one spatial scale level affect the interaction patterns at other spatial scale levels.

[0040] In this embodiment, cross-scale correlation analysis is implemented through the cross-scale analysis module of the self-evolutionary control model. This module can analyze the feature transfer relationships between different spatial scale levels. First, the field feature interaction patterns of different spatial scale levels are input into the cross-scale analysis module. The module identifies feature transfer paths by calculating the correlation coefficient between different scale levels. The correlation coefficient is calculated as the product of the covariance of the field feature interaction patterns of the two scale levels and the standard deviation of the field feature interaction patterns of the two scale levels. The larger the absolute value of the correlation coefficient, the more obvious the feature transfer path between the two scale levels. Then, the direction of feature transfer is determined based on the magnitude of the correlation coefficient. If the correlation coefficient between the macro-scale level and the meso-scale level is positive, the feature transfer direction is from the macro-scale level to the meso-scale level; if the correlation coefficient is negative, the feature transfer direction is from the meso-scale level to the macro-scale level. In this way, the feature transfer path between scales is identified.

[0041] Step S1241: Transform the field feature interaction patterns of each spatial scale level into standardized feature vectors. The dimension of the feature vectors corresponds to the main interaction parameters at the corresponding spatial scale level.

[0042] In this embodiment, the field feature interaction rules at each spatial scale level include parameters such as the correlation coefficient between the real-time operating state field and the instantaneous state field at that scale, the synchronicity of feature changes, and the direction of gradient propagation. These parameters are arranged in a fixed order to form a standardized feature vector. For example, the feature vector at the macroscopic scale level includes parameters such as the correlation coefficient between overall power output and overall temperature, and the gradient propagation direction between the central and edge regions; the feature vector at the microscopic scale level includes parameters such as the correlation coefficient between the power of a single heating element and the temperature of a single grain, and the thermal diffusion coefficient between grains. Each dimension of the feature vector corresponds to a specific interaction parameter, and the values ​​of the parameters need to be normalized to ensure that the parameters of different dimensions are on the same order of magnitude.

[0043] Step S1242: Calculate the correlation degree between feature vectors at different spatial scale levels. The correlation degree reflects the consistency of the interaction patterns between the two spatial scale levels in terms of parameter change trends.

[0044] In this embodiment, the correlation degree is calculated using the Pearson correlation coefficient method. The correlation degree is obtained by substituting the feature vectors from two different spatial scales into the Pearson correlation coefficient formula. For example, to calculate the correlation degree between macroscopic and mesoscopic feature vectors, the corresponding dimension parameters of the two vectors are multiplied pairwise and summed, then divided by the product of the magnitudes of the two vectors. The correlation degree ranges from -1 to 1. The larger the absolute value of the correlation degree, the more consistent the interaction patterns and trends between the two spatial scales.

[0045] Step S1243: Based on the correlation degree, select scale pairs with significant correlation. Each scale pair contains two spatial scale levels with strong mutual influence.

[0046] In this embodiment, the screening threshold for significant association is set to 0.7, meaning that scale pairs with an absolute association degree greater than 0.7 are considered to have a significant association. The screening process involves iterating through all possible scale pair combinations, comparing the association degree of each combination with the threshold, and including combinations that meet the criteria in the list of significantly associated scale pairs. For example, if the association degree between the macroscale and mesoscale is 0.8, and the association degree between the mesoscale and microscale is 0.6, then the combination of the macroscale and mesoscale is included in the list, while the combination of the mesoscale and microscale is not.

[0047] Step S1244: For each scale pair, analyze how changes in the interaction patterns of one spatial scale level trigger changes in the interaction patterns of another spatial scale level, and record the triggering conditions and propagation delays of the changes.

[0048] In this embodiment, the analysis of the interaction pattern changes is achieved by comparing the time sequence of feature vector changes at two scale levels. Specifically, the logic is as follows: A starting point is selected, and the time when the feature vector at the first scale level changes is observed. Then, the time when the feature vector at the second scale level changes is observed. The time difference between the two is the propagation delay. The trigger condition is that the change in the feature vector at the first scale level reaches a preset threshold, which is determined based on the fluctuation range of the feature vector in historical data. For example, if the starting point is t0, the macro-scale feature vector changes at time t1 and the change reaches the threshold, and the meso-scale feature vector changes at time t2, then the propagation delay is the result of t2 minus t1, and the trigger condition is that the change in the macro-scale feature vector reaches the threshold.

[0049] Step S1245: Based on the changing triggering conditions and transmission delay, obtain the direction and temporal relationship of feature transmission between scales, and form a feature transmission path that includes transmission direction, triggering threshold and delay duration.

[0050] In this embodiment, the propagation direction is determined based on the order of feature vector changes. If the feature vector change at the first scale level precedes that at the second scale level, the propagation direction is from the first scale level to the second scale level. The trigger threshold is the feature vector change threshold at the first scale level, and the delay duration is the calculated propagation delay value. For example, if the feature vector change at the macro scale precedes that at the meso scale, the propagation direction is from macro to meso, the trigger threshold is the feature vector change threshold at the macro scale, and the delay duration is the calculated propagation delay value.

[0051] Step S125: Based on the field feature interaction rules and cross-scale feature transmission paths at each spatial scale level, construct a field feature coupling matrix. The elements in the field feature coupling matrix represent the interaction strength and transmission coefficient between the real-time operating state field scale features and the instantaneous state field scale features at different spatial scales.

[0052] In this embodiment, the construction of the field feature coupling matrix needs to comprehensively consider the field feature interaction patterns at each spatial scale level and the cross-scale feature transmission path. First, the dimensions of the field feature coupling matrix are determined, with rows and columns corresponding to different spatial scale levels, and the number of rows and columns equal to the number of spatial scale levels. Then, the elements on the diagonal of the matrix are determined based on the field feature interaction patterns at each spatial scale level. These diagonal elements represent the correlation coefficients of the field feature interaction patterns at the corresponding spatial scale level. The larger the absolute value of the correlation coefficient, the stronger the interaction between the real-time operating state field scale features and the instantaneous state field scale features at that spatial scale level. Finally, the elements on the off-diagonal of the matrix are determined based on the cross-scale feature transmission path. These off-diagonal elements represent the correlation coefficients between two corresponding spatial scale levels. The larger the absolute value of the correlation coefficient, the higher the feature transmission coefficient between the two spatial scale levels. In this way, the field feature coupling matrix is ​​constructed.

[0053] Step S130: Based on the field feature coupling matrix and combined with the thermal evolution prediction logic built into the self-evolution control model, generate dynamic control parameters that are spatiotemporally adapted to the dynamic state field.

[0054] In this embodiment, the aim is to generate control parameters that can adapt to the spatiotemporal changes of the dynamic state field based on the field feature coupling matrix and thermal evolution prediction logic, so as to achieve precise control of the wafer heating process.

[0055] Step S131: Analyze the interaction strength and transmission coefficient in the field feature coupling matrix to obtain the key coupling terms that play a dominant role in the wafer heating effect. The key coupling terms correspond to the field feature interaction relationships that have a significant impact.

[0056] In this embodiment, when analyzing the field feature coupling matrix, it is necessary to calculate the absolute value of each element in the matrix. The field feature interaction relationship corresponding to the element with the larger absolute value is the key coupling term. For example, if the absolute value of a certain element in the field feature coupling matrix is ​​large, it indicates that the interaction strength or transmission coefficient between the real-time operating state field scale feature and the instantaneous state field scale feature at the spatial scale level corresponding to that element is large, and the impact on the wafer heating effect is more significant. Therefore, the field feature interaction relationship corresponding to that element is the key coupling term.

[0057] Step S132: Call the prediction module of the self-evolution control model and input the key coupling terms into the built-in thermal evolution prediction logic of the self-evolution control model. The thermal evolution prediction logic is constructed based on the field evolution law in the historical heating process and can predict the evolution trend of field characteristics over time.

[0058] In this embodiment, the prediction module of the self-evolutionary control model incorporates thermal evolution prediction logic, which is trained based on a large amount of historical heating process data. This historical heating process data includes dynamic state fields, field characteristic coupling matrices, control parameters, and corresponding wafer heating effects under different heating conditions. After inputting key coupling terms into the thermal evolution prediction logic, the logic predicts the evolution trend of these key coupling terms over time based on the field evolution patterns observed during historical heating processes. Examples of key coupling term values ​​over time include the time it takes for a certain threshold to be reached.

[0059] Step S133: Generate the future evolution trajectory of key coupling terms through thermal evolution prediction logic. The future evolution trajectory includes the expected coupling states at different time points.

[0060] In this embodiment, the thermal evolution prediction logic generates the future evolution trajectory of the key coupling items based on the predicted evolution trend of the key coupling items over time. The future evolution trajectory is a time series containing the expected coupling states at different time points, and the expected coupling state at each time point consists of the expected value of the key coupling item. For example, for a time point t, the expected coupling state is the expected value of the key coupling item at that time point.

[0061] Step S134: Based on the future evolution trajectory, obtain the control requirements corresponding to each time node. The control requirements are reflected in the adjustment direction of the heating device required to maintain the expected coupling state.

[0062] In this embodiment, the difference between the expected coupling state at each time point in the future evolution trajectory and the corresponding coupling state in the current dynamic state field is first analyzed. The difference value is equal to the expected coupling state value minus the current coupling state value. Then, the adjustment direction of the heating device is determined according to the sign of the difference value. If the difference value is positive, the power output of the corresponding area of ​​the heating device needs to be increased; if the difference value is negative, the power output of the corresponding area of ​​the heating device needs to be decreased. Finally, the adjustment range of the heating device is determined according to the magnitude of the difference value; the larger the difference value, the larger the adjustment range.

[0063] Step S1341: Extract the expected coupling state parameters for each time node from the future evolution trajectory. The extracted expected coupling state parameters represent the ideal state of the key coupling terms at the corresponding time node.

[0064] In this embodiment, the future evolution trajectory is a sequence of key coupling terms changing over time. The expected coupling state parameters at each time point include the numerical range, rate of change, and correlation with other coupling terms of the key coupling term at that time point. These parameters are generated through thermal evolution prediction logic. The extraction process requires mapping each time point in the future evolution trajectory to its corresponding expected coupling state parameters to ensure the completeness of the parameters at each time point.

[0065] Step S1342: Compare the predicted state parameters and expected coupled state parameters of the current dynamic state field at the corresponding time node to obtain the deviation direction and deviation range between the two.

[0066] In this embodiment, the predicted state parameters of the current dynamic state field at the corresponding time node are generated by the prediction module of the self-evolutionary control model. The comparison process requires pairwise comparison of the corresponding dimensions of the predicted state parameters and the expected coupled state parameters. The deviation direction is the positive or negative value of the result of subtracting the expected coupled state parameter from the predicted state parameter, and the deviation range is the absolute value of the difference between the predicted state parameter and the expected coupled state parameter. For example, if the predicted state parameter is G and the expected coupled state parameter is H, then the deviation direction is the positive or negative value of the result of subtracting H from G, and the deviation range is the absolute value of the result of subtracting H from G.

[0067] Step S1343: Based on the deviation direction and deviation range, analyze which areas of the heating device need to be adjusted to reduce the deviation. The areas that need to be adjusted are related to the spatial distribution characteristics of the dynamic state field.

[0068] In this embodiment, the analysis process needs to consider the spatial distribution characteristics of the dynamic state field. Specifically, the logic is as follows: if the deviation direction is positive and the deviation range is large, the power output of the region corresponding to the coupling term in the heating device needs to be adjusted; if the deviation direction is negative and the deviation range is large, the power input of the region corresponding to the coupling term in the heating device needs to be adjusted. The region to be adjusted is determined through the mapping relationship between the coupling term and the region of the heating device. This mapping relationship is established based on the correlation data between the coupling term and the region adjustment during historical heating processes. For example, if the deviation direction is positive and the deviation range is large, and the coupling term corresponds to the central region of the heating device, then the power output of the central region needs to be adjusted.

[0069] Step S1344: For the area that needs to be adjusted, the direction of adjustment is obtained by combining the characteristics of the corresponding area in the real-time operating state field. The direction of adjustment includes enhancing or weakening the operating intensity of the corresponding area of ​​the heating device.

[0070] In this embodiment, the characteristics of the corresponding region in the real-time operating state field include parameters such as the power density, temperature gradient, and aging degree of the heating element in that region. These parameters are obtained through real-time sensor data. The logic for determining the adjustment direction is as follows: if the power density of the region is lower than a preset threshold, the adjustment direction is to enhance; if the power density of the region is higher than the preset threshold, the adjustment direction is to weaken. For example, if the power density of the region to be adjusted is lower than the preset threshold, the adjustment direction is to enhance the operating intensity of that region.

[0071] Step S1345 integrates the areas and directions to be adjusted to form the control requirements corresponding to each time node. The control requirements can directly guide the specific operation of the heating device to maintain the expected coupling state.

[0072] In this embodiment, the integration of control requirements requires mapping the areas to be adjusted to the directions of adjustment one by one, forming structured control instructions. These instructions include area identifiers, adjustment directions, and preliminary ranges of adjustment magnitudes. For example, if the area to be adjusted is the central region and the adjustment direction is enhancement, then the control requirement is "enhance the operational intensity of the central region".

[0073] Step S135: The control requirements are converted into specific parameter values. The converted parameter values ​​change dynamically with time nodes and correspond to the spatial characteristics of the dynamic state field, together forming dynamic control parameters that are spatiotemporally adapted to the dynamic state field.

[0074] In this embodiment, when converting the control requirements into specific parameter values, it is necessary to consider the control characteristics of the heating device and the spatial characteristics of the dynamic state field. For example, if the control requirement is to increase the power output of a certain area of ​​the heating device, then the control requirement needs to be converted into an increase in the power output of the heating module in that area; if the control requirement is to decrease the power output of a certain area of ​​the heating device, then the control requirement needs to be converted into a decrease in the power output of the heating module in that area. The converted parameter values ​​need to change dynamically with time nodes to adapt to the spatiotemporal changes of the dynamic state field, and at the same time correspond to the spatial characteristics of the dynamic state field. For example, the spatial distribution of the parameter values ​​needs to match the spatial characteristics of the dynamic state field.

[0075] Step S140: The dynamic control parameters are loaded into the heating execution system to trigger the wafer heating process. The evolution trajectory of the real-time operating state field and the response trajectory of the instantaneous state field are captured simultaneously during the heating process and fused to form field evolution feedback.

[0076] In this embodiment, the purpose is to apply the generated dynamic control parameters to the heating execution system, trigger the wafer heating process, and capture the state changes during the heating process in real time to form feedback information.

[0077] Step S141: The dynamic control parameters are allocated according to the spatial control logic of the heating execution system, so that each parameter corresponds to a specific spatial area of ​​the heating device, and the parameters can be accurately applied to the corresponding area of ​​the dynamic state field.

[0078] In this embodiment, the spatial control logic of the heating execution system refers to the method and rules by which the heating execution system controls different spatial regions of the heating device. Specifically, it includes the control priority and control range of each heating region of the heating device. Dynamic control parameters are allocated according to this spatial control logic. For example, the power output increase parameter is allocated to the central heating region of the heating device, and the power output decrease parameter is allocated to the edge heating regions of the heating device. Each parameter corresponds to a specific spatial region on the heating device, thereby enabling each parameter to precisely act on the corresponding region of the dynamic state field.

[0079] Step S142: The allocated dynamic control parameters are loaded into the heating execution system, triggering the heating execution system to execute the wafer heating process according to the time change law and spatial allocation logic of the parameters.

[0080] In this embodiment, the allocated dynamic control parameters are input into the heating execution system. Based on the time-varying patterns and spatial allocation logic of the parameters, the heating execution system controls different heating modules of the heating device to heat according to the corresponding parameter values, thereby triggering the wafer heating process. For example, based on the time-varying patterns of the parameters, the heating execution system increases the power output of the central heating region by a certain value at the first time node and decreases the power output of the edge heating region by a certain value at the second time node, thus achieving precise control of the wafer heating process.

[0081] Step S143: During the wafer heating process, the spatial distributed acquisition system is activated to capture the spatial distribution changes of the real-time operating state field at preset time intervals, forming a real-time operating state field evolution trajectory containing a time series.

[0082] In this embodiment, the spatially distributed acquisition system consists of sensors distributed across different areas of the heating device. The sensor placement must correspond to the grid cells of the real-time operating state field, with each sensor corresponding to one grid cell. During the wafer heating process, the spatially distributed acquisition system is activated to collect state parameters of each area of ​​the heating device at preset time intervals. The collected state parameters are arranged in chronological order to form the real-time operating state field evolution trajectory. For example, if the preset time interval is a fixed duration, the state parameters of each area of ​​the heating device are collected every fixed duration. The collected state parameters are arranged in chronological order to form a time series, which is the real-time operating state field evolution trajectory.

[0083] Step S144: Simultaneously capture the spatial response change of the instantaneous state field under heating, record the state change of each discrete unit over time, and form the response trajectory of the instantaneous state field.

[0084] In this embodiment, the instantaneous state field is captured using non-contact temperature sensors, thermal imaging devices, infrared thermometers, etc., similar to the devices used to acquire wafer state parameters. Each device corresponds to a discrete unit on the wafer. During the wafer heating process, the spatial response changes of the instantaneous state field are captured synchronously, recording the state changes of each discrete unit over time, such as the temperature changes of each discrete unit over time. The recorded state changes are arranged in chronological order to form the response trajectory of the instantaneous state field.

[0085] Step S145: Align the real-time running state field evolution trajectory and the instantaneous state field response trajectory according to the time axis, extract the field feature correlation relationship corresponding to the same time mark, and fuse them to form a field evolution feedback that can reflect the dynamic interaction between fields.

[0086] In this embodiment, a time stamp is first added to each data point in the real-time running state field evolution trajectory and the instantaneous state field response trajectory. The time stamp format is year-month-day-hour-minute-second, accurate to the second. Then, the data points in the two trajectories are aligned according to the time stamps, i.e., data points with the same time stamp are found. For the aligned data points, the field characteristics of the real-time running state field and the instantaneous state field are extracted, and the correlation between them is analyzed. For example, the correlation coefficient between the two is calculated. The larger the absolute value of the correlation coefficient, the stronger the correlation between the two. These correlations are arranged in chronological order to form a field evolution feedback.

[0087] Step S1451: Add precise time stamps to each data point in the real-time running state field evolution trajectory and the instantaneous state field response trajectory. The precision of the time stamps is consistent with the acquisition interval.

[0088] In this embodiment, the addition of time stamps is achieved through the clock synchronization module of the acquisition system. The clock synchronization module ensures that the time stamps of the real-time running state field evolution trajectory and the instantaneous state field response trajectory are based on the same clock source. The format of the time stamp is year-month-day-hour-minute-second-millisecond, and the acquisition interval is a preset fixed duration. The time stamp of each data point is the clock value at the time of data acquisition. For example, if the acquisition interval is 1 millisecond, then the time stamp of each data point is accurate to the millisecond.

[0089] Step S1452: Using the time marker as a reference, pair up data points with the same time marker in the two trajectories to form time-aligned data point pairs.

[0090] In this embodiment, the pairing process requires traversing all data points in the real-time running state field evolution trajectory and the instantaneous state field response trajectory, extracting data points with the same time stamp and combining them into data point pairs. Each data point pair includes both real-time running state field data points and instantaneous state field data points. For example, a data point with time stamp 't' in the real-time running state field evolution trajectory is paired with a data point with time stamp 't' in the instantaneous state field response trajectory to form a data point pair.

[0091] Step S1453: For each time-aligned data point pair, calculate the correlation coefficient between the real-time running state field data and the instantaneous state field data in the corresponding data point. The correlation coefficient quantifies the interaction strength between the real-time running state field features and the instantaneous state field features corresponding to the same time mark.

[0092] In this embodiment, the correlation coefficient is calculated using the Pearson correlation coefficient method. The specific logic is as follows: the real-time operating state data and the instantaneous state data in the data point pair are treated as two variables, and substituted into the Pearson correlation coefficient formula to obtain the correlation coefficient. For example, if the real-time operating state data in the data point pair is I and the instantaneous state data is J, then the correlation coefficient is the covariance of I and J divided by the product of the standard deviations of I and J.

[0093] Step S1454: Arrange all time-aligned data point pairs and their corresponding correlation coefficients in chronological order to form a correlation sequence of field characteristics changing over time.

[0094] In this embodiment, the sorting process requires arranging time-aligned data point pairs and their corresponding correlation coefficients according to the order of their time markers, forming an ordered correlation sequence. The correlation sequence includes time markers, data point pairs, correlation coefficients, etc. For example, data point pairs and their correlation coefficients with time marker t1 are placed before data point pairs and their correlation coefficients with time marker t2, if t1 is less than t2.

[0095] Step S1455: Extract trend change features from the correlation sequence, including the increasing or decreasing trend of the correlation coefficient and abrupt change points. The extracted trend change features reflect the changing pattern of the dynamic interaction between fields.

[0096] In this embodiment, the extraction of trend change features is achieved using the sliding window method. The specific logic is as follows: a fixed-size sliding window is set, the correlation sequence is traversed, and the average value of the correlation coefficient within each window is calculated. If the average value shows an upward trend, the trend is considered increasing; if the average value shows a downward trend, the trend is considered decreasing. A mutation point is the point in time when the value of the correlation coefficient suddenly changes. The criterion for determining a mutation point is that the change in the correlation coefficient exceeds a preset threshold. For example, if the average value within the sliding window shows an upward trend, the trend is considered increasing; if the change in the correlation coefficient at a certain moment exceeds a preset threshold, that moment is considered a mutation point.

[0097] Step S1456 integrates time-aligned data point pairs, correlation coefficient sequences, and trend change characteristics to form a complete field evolution feedback that reflects the dynamic interaction between fields.

[0098] In this embodiment, the integration process requires combining time-aligned data point pairs, correlation coefficient sequences, and trend change characteristics in chronological order to form structured field evolution feedback data. This data includes time stamps, data point pairs, correlation coefficients, increasing / decreasing trends, and abrupt change points. For example, a record at a specific moment in the field evolution feedback data includes a time stamp, the corresponding data point pair, correlation coefficients, an increasing / decreasing trend (increased), and whether a sudden change point is present.

[0099] Step S150: Input the field evolution feedback into the self-evolutionary control model, drive the self-evolutionary control model to generate a field correction factor based on the deviation between the field feature coupling matrix and the field evolution feedback, and iteratively optimize the dynamic control parameters through the field correction factor to obtain the target control parameters that are adapted to the current dynamic state field.

[0100] In this embodiment, the aim is to optimize the dynamic control parameters based on field evolution feedback in order to improve the adaptability of the control parameters and achieve precise control of the wafer heating process.

[0101] For example, in step S151, the field evolution feedback and field feature coupling matrix are input into the deviation analysis module of the self-evolution control model. The deviation analysis module calculates the feature deviation of the field evolution feedback and field feature coupling matrix at the same spatial scale and time node.

[0102] In this embodiment, the deviation analysis module is an important component of the self-evolutionary control model, capable of calculating the deviation between the field evolution feedback and the field feature coupling matrix. First, the field evolution feedback and the field feature coupling matrix are input into the deviation analysis module. This module compares and analyzes the characteristics of both based on the same spatial scale and time node, calculating the feature deviation between them. The feature deviation is calculated by subtracting the corresponding value in the field feature coupling matrix from the field feature value in the field evolution feedback. For example, for the same spatial scale level and the same time node, if the field feature value in the field evolution feedback is a certain value and the corresponding value in the field feature coupling matrix is ​​a certain value, the feature deviation is the difference between the two.

[0103] Step S152: Based on the feature bias, obtain the bias contribution of each scale level and time node. The bias contribution reflects the influence weight of the bias of the corresponding scale level and time node on the overall adaptability.

[0104] In this embodiment, the calculation of the deviation contribution needs to comprehensively consider factors such as the magnitude of the feature deviation and the importance of the feature in the dynamic state field. First, the absolute value of the feature deviation at each scale level and time node is calculated. Then, a weight is assigned to each feature deviation according to its importance in the dynamic state field; the more important the feature deviation, the larger its weight. Finally, the absolute value of the feature deviation is multiplied by its corresponding weight to obtain the deviation contribution. For example, for a scale level and time node with a large feature deviation and high importance in the dynamic state field, its deviation contribution is large.

[0105] In step S153, the correction factor generation module of the self-evolutionary control model assigns corresponding correction weights to different scale levels and time nodes based on the deviation contribution. When the deviation contribution value is large, the correction weight of the corresponding scale level and time node increases accordingly.

[0106] In this embodiment, the correction factor generation module is an important component of the self-evolutionary control model, capable of generating correction weights based on the deviation contribution. This module first normalizes the deviation contributions, ensuring the sum of the normalized deviation contributions equals 1. Then, it assigns the normalized deviation contributions as correction weights to the corresponding scale levels and time nodes. For example, larger correction weights are assigned to scale levels and time nodes with larger deviation contributions, allowing for focused correction of deviations at these scale levels and time nodes during subsequent parameter optimization.

[0107] Step S154: Combine the specific values ​​of the correction weights and feature biases to generate a field correction factor. The field correction factor includes the specific correction amount for each scale level and time node.

[0108] In this embodiment, the generation of the field correction factor requires combining the correction weight with the specific numerical values ​​of the feature bias. First, the correction weight for each scale level and time node is multiplied by the corresponding feature bias to obtain the correction amount for that scale level and time node. Then, the correction amounts for all scale levels and time nodes are combined to form the field correction factor. For example, for a scale level and time node, if the correction weight is a certain value and the feature bias is a certain value, then the correction amount for that scale level and time node is the product of the two.

[0109] Step S155: The field correction factor is input into the optimization module of the dynamic control parameters. The optimization module makes targeted adjustments to the dynamic control parameters according to the correction amount, while maintaining the spatiotemporal correlation characteristics of the parameters during the adjustment process.

[0110] In this embodiment, the optimization module for dynamic control parameters is a crucial component of the self-evolutionary control model, capable of optimizing the dynamic control parameters based on the field correction factor. This optimization module first decomposes the field correction factor into correction amounts at different scale levels and time nodes. Then, it adjusts the dynamic control parameters accordingly based on these correction amounts. For example, for a given scale level and time node, if the correction amount is positive, the corresponding control parameter value is increased; if the correction amount is negative, the corresponding control parameter value is decreased. During the adjustment process, the spatiotemporal correlation characteristics of the parameters must be maintained; that is, the adjusted parameters must match the spatiotemporal changes of the dynamic state field.

[0111] Step S156: The adjusted dynamic control parameters can offset the influence of characteristic deviations, form a precise fit with the current dynamic state field, and ultimately become the target control parameters that fit the current dynamic state field.

[0112] In this embodiment, the adjusted dynamic control parameters are verified to offset the characteristic deviation between the field evolution feedback and the field characteristic coupling matrix, ensuring a precise fit between the dynamic control parameters and the current dynamic state field. The verification process involves loading the adjusted dynamic control parameters into the heating execution system and observing whether the state changes during wafer heating meet expectations. If they meet expectations, the adjusted dynamic control parameters become the target control parameters; if they do not meet expectations, steps S150 to S156 are repeated until the target control parameters that meet expectations are obtained.

[0113] The above steps achieve adaptive control of the wafer heating process, enabling real-time adjustment of control parameters based on changes in the dynamic state field. This improves the uniformity and stability of wafer heating, thereby enhancing wafer manufacturing quality. The application of the self-evolutionary control model gives this invention strong self-learning and adaptive capabilities, allowing it to adapt to different heating conditions and wafer types, thus exhibiting high versatility and practicality.

[0114] Figure 2 The illustration shows exemplary hardware and software components of an AI adaptive control system 100 for wafer heating that can implement the ideas of this application, according to some embodiments of this application. For example, a processor 120 may be used in the AI ​​adaptive control system 100 for wafer heating and to perform the functions described in this application.

[0115] The AI ​​adaptive control system 100 for wafer heating can be a general-purpose server or a special-purpose server, both of which can be used to implement the AI ​​adaptive control method for wafer heating of this application. Although only one server is shown in this application, for convenience, the functions described in this application can be implemented in a distributed manner on multiple similar platforms to balance the load.

[0116] For example, the AI ​​adaptive control system 100 for wafer heating may include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and various forms of storage media 140, such as a disk, ROM, or RAM, or any combination thereof. Exemplarily, the AI ​​adaptive control system 100 for wafer heating may also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. The methods of this application can be implemented according to these program instructions. The AI ​​adaptive control system 100 for wafer heating also includes an I / O interface 150 between the computer and other input / output devices.

[0117] For ease of explanation, only one processor is described in the AI ​​adaptive control system 100 for wafer heating. However, it should be noted that the AI ​​adaptive control system 100 for wafer heating in this application may also include multiple processors, and therefore the steps performed by one processor described in this application may also be performed jointly or individually by multiple processors. For example, if the processor of the AI ​​adaptive control system 100 for wafer heating performs steps A and B, it should be understood that steps A and B may also be performed jointly by two different processors or individually by one processor. For example, the first processor performs step A, the second processor performs step B, or the first processor and the second processor jointly perform steps A and B.

[0118] Furthermore, this embodiment of the invention also provides a readable storage medium, wherein computer-executable instructions are preset in the readable storage medium, and when the processor executes the computer-executable instructions, the above-mentioned AI adaptive control method for wafer heating is implemented.

[0119] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. An AI adaptive control method for wafer heating, characterized in that, The method includes: A dynamic state field for wafer heating is constructed, which includes the real-time operating state field of the heating device and the instantaneous state field of the wafer. The real-time operating state field of the heating device and the instantaneous state field of the wafer form a spatial coupling relationship through heat transfer. The dynamic state field is analyzed at multiple scales using a self-evolutionary control model to extract the field feature interaction patterns at different spatial scales and generate a field feature coupling matrix. Based on the field feature coupling matrix, and combined with the thermal evolution prediction logic built into the self-evolution control model, dynamic control parameters that are spatiotemporally adapted to the dynamic state field are generated. The dynamic control parameters are loaded into the heating execution system to trigger the wafer heating process, and the evolution trajectory of the real-time operating state field and the response trajectory of the instantaneous state field are captured simultaneously during the heating process, and fused to form field evolution feedback. The field evolution feedback is input into the self-evolutionary control model, which drives the self-evolutionary control model to generate a field correction factor based on the deviation between the field feature coupling matrix and the field evolution feedback. The dynamic control parameters are iteratively optimized through the field correction factor to obtain the target control parameters that are adapted to the current dynamic state field.

2. The adaptive control method for wafer heating according to claim 1, characterized in that, The dynamic state field for constructing wafer heating includes a real-time operating state field of the heating device and an instantaneous state field of the wafer. These two state fields are spatially coupled through heat transfer, including: The state parameters of each area of ​​the heating device are collected during operation. The collected state parameters are then mapped into a grid according to the spatial distribution law to form a real-time operating state field with spatial gradient characteristics. Each grid cell corresponds to the state representation of a spatial location. The state parameters of each region of the wafer are collected at the current moment, and spatial discretization is performed according to the physical structure of the wafer to form an instantaneous state field containing the differences in regional states. The discrete unit corresponds one-to-one with the physical partition of the wafer. The spatial overlap region between the real-time operating state field and the instantaneous state field is analyzed to determine the main heat transfer path and transfer efficiency characteristics. The obtained main heat transfer path and transfer efficiency characteristics reflect the spatial interaction strength between the real-time operating state field and the instantaneous state field. Based on the main paths and efficiency characteristics of heat transfer, an inter-field correlation function is constructed. The inter-field correlation function can quantify the degree of influence of any grid cell in the real-time running state field on the corresponding discrete cell in the instantaneous state field. By using the inter-field correlation function, the real-time operating state field and the instantaneous state field are spatially coupled, so that the corresponding units in the real-time operating state field and the instantaneous state field form a dynamic correlation and together constitute a dynamic state field containing spatial coupling relationship.

3. The adaptive control method for wafer heating according to claim 1, characterized in that, The process of performing multi-scale field analysis on the dynamic state field using a self-evolutionary control model to extract field feature interaction patterns at different spatial scales and generate a field feature coupling matrix includes: The dynamic state field is input into the multi-scale analysis module of the self-evolution control model. The multi-scale analysis module performs hierarchical analysis of the dynamic state field according to the preset spatial scale level, and each spatial scale level corresponds to a spatial scale range. At each spatial scale level, the scale features of the real-time operating state field and the scale features of the instantaneous state field are extracted. The scale features include the field distribution uniformity and gradient change features at the corresponding spatial scale level. By using the feature interaction module of the self-evolutionary control model, the interaction mode between the real-time operating state field scale features and the instantaneous state field scale features under the same spatial scale level is analyzed, and the field feature interaction law under the corresponding spatial scale level is obtained. Cross-scale correlation analysis is conducted on the interaction patterns of field features at different spatial scales to identify the feature transmission paths between scales, i.e., how the interaction patterns at one spatial scale affect the interaction patterns at other spatial scales. Based on the field feature interaction patterns and cross-scale feature transmission paths at each spatial scale level, a field feature coupling matrix is ​​constructed. The elements in the field feature coupling matrix represent the interaction strength and transmission coefficient between the real-time operating state field scale features and the instantaneous state field scale features at different spatial scales.

4. The adaptive control method for wafer heating according to claim 1, characterized in that, The process of generating dynamic control parameters that are spatiotemporally adapted to the dynamic state field, based on the field feature coupling matrix and combined with the built-in thermal evolution prediction logic of the self-evolutionary control model, includes: By analyzing the interaction strength and transfer coefficient in the field feature coupling matrix, the key coupling terms that play a dominant role in the wafer heating effect are obtained, and the key coupling terms correspond to the field feature interaction relationships that have a significant impact. The prediction module of the self-evolution control model is invoked, and the key coupling terms are input into the thermal evolution prediction logic built into the self-evolution control model. The thermal evolution prediction logic is constructed based on the field evolution law in the historical heating process and can predict the evolution trend of field characteristics over time. The future evolution trajectory of key coupling terms is generated by the thermal evolution prediction logic. The future evolution trajectory includes the expected coupling state at different time points. Based on the future evolution trajectory, the control requirements corresponding to each time point are obtained. The control requirements are reflected in the adjustment direction of the heating device required to maintain the expected coupling state. The control requirements are transformed into specific parameter values. The transformed parameter values ​​change dynamically with time nodes and correspond to the spatial characteristics of the dynamic state field, together forming dynamic control parameters that are spatiotemporally adapted to the dynamic state field.

5. The adaptive control method for wafer heating according to claim 1, characterized in that, The process of loading the dynamic control parameters into the heating execution system to trigger the wafer heating process, simultaneously capturing the evolution trajectory of the real-time operating state field and the response trajectory of the instantaneous state field during the heating process, and fusing them to form field evolution feedback includes: The dynamic control parameters are allocated according to the spatial control logic of the heating execution system, so that each parameter corresponds to a specific spatial area of ​​the heating device, and the parameters can be accurately applied to the corresponding area of ​​the dynamic state field. The allocated dynamic control parameters are loaded into the heating execution system, triggering the heating execution system to execute the wafer heating process according to the time change pattern and spatial allocation logic of the parameters; During the wafer heating process, a spatially distributed acquisition system is activated to capture the spatial distribution changes of the real-time operating state field at preset time intervals, forming a real-time operating state field evolution trajectory containing a time series. The instantaneous state field is captured simultaneously to capture the spatial response changes of the instantaneous state field under heating, and the state of each discrete unit is recorded over time to form the response trajectory of the instantaneous state field. The real-time operating state field evolution trajectory and the instantaneous state field response trajectory are aligned along the time axis, and the field feature correlations corresponding to the same time markers are extracted and fused to form a field evolution feedback that can reflect the dynamic interaction between fields.

6. The adaptive control method for wafer heating according to claim 2, characterized in that, Based on the main heat transfer path and transfer efficiency characteristics, an inter-field correlation function is constructed. This function quantifies the influence of any grid cell in the real-time operating state field on the corresponding discrete cell in the instantaneous state field, including: Extract the heat output characteristics of each grid cell in the real-time running state field and the heat reception characteristics of each discrete cell in the instantaneous state field. The heat output characteristics reflect the heat release capability of the corresponding grid cell, and the heat reception characteristics reflect the heat absorption capability of the corresponding discrete cell. Calculate the spatial distance and thermal resistance parameter between each grid cell and each discrete cell. The spatial distance and thermal resistance parameter together determine the degree of heat transfer attenuation. Based on heat output characteristics, heat reception characteristics, spatial distance and thermal resistance parameters, a basic correlation factor is constructed. The basic correlation factor initially quantifies the potential influence of grid cells on discrete cells. By introducing weighting coefficients for the main heat transfer path, the basic correlation factor is modified, thereby strengthening the correlation factor between grid cells and discrete cells on the main path. By combining the transmission efficiency characteristics, the modified correlation factor is dynamically adjusted. When the transmission efficiency value is large, the value of the correlation factor increases accordingly, ultimately forming an inter-field correlation function that can accurately quantify the degree of influence.

7. The adaptive control method for wafer heating according to claim 3, characterized in that, The cross-scale correlation analysis of field feature interaction patterns at different spatial scales identifies feature transmission paths between scales, i.e., the influence of interaction patterns at one spatial scale on interaction patterns at other spatial scales, including: The field feature interaction patterns at each spatial scale level are transformed into standardized feature vectors, and the dimensions of the feature vectors correspond to the main interaction parameters at the corresponding spatial scale level. Calculate the correlation degree between feature vectors at different spatial scales. The correlation degree reflects the consistency of the interaction patterns between the two spatial scales in terms of parameter variation trends. Based on the correlation degree, scale pairs with significant correlations are selected, and each scale pair contains two spatial scale levels with strong mutual influence. For each scale pair, analyze how changes in the interaction patterns at one spatial scale level trigger changes in the interaction patterns at another spatial scale level, and record the triggering conditions and propagation delays of the changes. Based on the changing triggering conditions and transmission delays, the direction and temporal relationship of feature transmission between scales are obtained, forming a feature transmission path that includes the transmission direction, triggering threshold, and delay duration.

8. The adaptive control method for wafer heating according to claim 4, characterized in that, Based on the future evolution trajectory, the control requirements corresponding to each time point are obtained. These control requirements reflect the adjustment direction of the heating device needed to maintain the expected coupling state, including: The expected coupling state parameters for each time point are extracted from the future evolution trajectory. The extracted expected coupling state parameters represent the ideal state of the key coupling terms at the corresponding time point. By comparing the predicted state parameters and expected coupled state parameters of the current dynamic state field at the corresponding time node, the direction and range of the deviation between the two can be obtained. Based on the direction and range of the deviation, we analyze which areas of the heating device need to be adjusted to reduce the deviation. The areas that need to be adjusted are related to the spatial distribution characteristics of the dynamic state field. For the area that needs adjustment, the direction of adjustment is obtained by combining the characteristics of the corresponding area in the real-time operating state field. The direction of adjustment includes strengthening or weakening the operating intensity of the corresponding area of ​​the heating device. By integrating the areas and directions that need adjustment, the control requirements corresponding to each time point are formed. These control requirements can directly guide the specific operation of the heating device to maintain the expected coupling state.

9. The adaptive control method for wafer heating according to claim 5, characterized in that, The process of aligning the real-time operational state field evolution trajectory and the instantaneous state field response trajectory along the time axis, extracting the field feature correlations corresponding to the same time markers, and fusing them to form a field evolution feedback that can reflect the dynamic interaction between fields includes: Add precise time stamps to each data point in the real-time running state field evolution trajectory and the instantaneous state field response trajectory, with the accuracy of the time stamps consistent with the acquisition interval; Based on the time marker, data points with the same time marker in two trajectories are paired to form time-aligned data point pairs; For each time-aligned data point pair, calculate the correlation coefficient between the real-time running state field data and the instantaneous state field data in the corresponding data point. The correlation coefficient quantifies the interaction strength between the real-time running state field features and the instantaneous state field features corresponding to the same time mark. All time-aligned data point pairs and their corresponding correlation coefficients are arranged in chronological order to form a correlation sequence of field characteristics changing over time. Extract trend change features from the correlation sequence, including the increasing and decreasing trends of the correlation coefficient and abrupt change points. The extracted trend change features reflect the changing patterns of dynamic interactions between fields. By integrating time-aligned data point pairs, correlation coefficient sequences, and trend change characteristics, a complete field evolution feedback reflecting the dynamic interaction between fields is formed.

10. An AI adaptive control system for wafer heating, characterized in that, The AI ​​adaptive control system for wafer heating includes a processor and a memory, the memory and the processor being connected. The memory is used to store programs, instructions or code, and the processor is used to execute the programs, instructions or code in the memory to implement the AI ​​adaptive control method for wafer heating as described in any one of claims 1-9.

Images