An epitaxial reaction chamber flow field real-time regulation method, system and computer

Abstract

The present application relates to the technical field of semiconductor, and provides an epitaxial reaction chamber flow field real-time regulation method, system and computer, the epitaxial reaction chamber flow field real-time regulation method includes: the three-dimensional modeling of epitaxial reaction chamber is carried out;Based on the physical field simulation data set and neural operator network, the real-time inference model is obtained by training;The real-time process parameters of epitaxial furnace are collected, and the real-time process parameters are input into the real-time inference model to obtain the real-time physical field of epitaxial reaction chamber;The target physical field is obtained, and the real-time physical field is compared with the target physical field to calculate the uniformity deviation;Gradient inversion is carried out based on the uniformity deviation to calculate the flow adjustment value of the showerhead area, and the flow adjustment value is used to control the mass flow controller of the showerhead area in real time.Through the above method, the epitaxial reaction chamber inside is micro-identified in the epitaxial growth process, and real-time response and active regulation are carried out.

Description

Technical Field

[0001] This invention relates to the field of semiconductor technology, and in particular to a method, system and computer for real-time control of flow field in an epitaxial reaction chamber. Background Technology

[0002] In semiconductor epitaxial growth processes, the gas flow field distribution and temperature uniformity within the epitaxial reaction chamber directly determine the thickness uniformity, doping concentration distribution, and wavelength consistency of the epitaxial layer. Nowadays, semiconductor epitaxy, represented by GaN LEDs and SiC power devices, has increasingly stringent requirements for the microscopic uniformity of its growth interface; for example, under mass production requirements, its on-wafer uniformity needs to reach over 99%.

[0003] For the control and optimization of epitaxial growth processes, the temperature setpoint or flow ratio of the next batch can be adjusted by analyzing the wavelength deviation of the previous batch or historical batches of epitaxial wafers. However, this method has a significant lag and cannot intervene in the micro-uniformity of the current growth process in real time. Moreover, it can only compensate for the system deviation at the macro level and cannot identify and eliminate the intra-wafer non-uniformity caused by local smoothness distortion in the reaction chamber. Therefore, existing technologies introduce CFD (Computational Fluid Dynamics) software to simulate the process in the reaction chamber in order to optimize the spray head structure and process parameters.

[0004] However, during the CFD simulation process in process development, the amount of computation required is enormous. A single high-fidelity simulation often takes several hours or even days and can only be used as an offline reference. Its statistical feedback still suffers from furnace secondary lag and cannot achieve millisecond-level real-time response and intervention to transient disturbances and micro-inhomogeneities during epitaxial growth. It cannot be used for online real-time control, and the simulated patterns can only provide flow field morphology classifications and cannot construct a complete physical field distribution, resulting in insufficient spatial resolution. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the present invention aims to provide a method, system, and computer for real-time control of the flow field in an epitaxial reaction chamber. This invention seeks to solve the technical problem that existing technologies lack the ability to identify the microscopic structures inside the epitaxial reaction chamber during epitaxial growth, thus failing to respond in real time and actively intervene.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solution: A method for real-time control of the flow field in an epitaxial reaction chamber includes the following steps: A three-dimensional model of the epitaxial reaction chamber was created to obtain a physical field simulation dataset. A neural operator network is obtained, and a real-time inference model is trained based on the physical field simulation dataset and the neural operator network. Real-time process parameters of the epitaxial furnace are collected and input into the real-time inference model to obtain the real-time physical field of the epitaxial reaction chamber. The target physical field is acquired, and the real-time physical field is compared with the target physical field to calculate the uniformity deviation. Gradient inversion is performed based on the uniformity deviation to calculate the flow adjustment value of the spray head area, which is used to control the mass flow controller of the spray head area in real time.

[0007] Furthermore, the step of performing three-dimensional modeling of the epitaxial reaction chamber to obtain a physical field simulation dataset includes: A three-dimensional geometric model is constructed based on the epitaxial reaction chamber, which includes a spray head sub-model, a reaction chamber sub-model, a graphite base model, and an exhaust port model. The three-dimensional geometric model is meshed, and the mesh is refined according to the flow gradient; A set of boundary conditions is set for the three-dimensional geometric model, and a physical mode is set to form a simulation model. The physical mode is a turbulence mode, a chemical reaction mode, a radiation heat transfer mode, or a multi-component transport mode. Based on the simulation model, a physical field simulation dataset is formed through parametric scanning.

[0008] Furthermore, the step of forming a physical field simulation dataset based on the simulation model through parametric scanning includes: Set a process parameter variable group, which includes several variables, such as chamber pressure variable, base temperature variable, several zone gas flow rate variables, or total carrier gas flow rate variable. Set the variable range and scan parameters for the variable; Based on the process parameter variable group, several variable ranges, and several scanning parameters, the simulation model is subjected to parametric scanning simulation to obtain several scanning process parameter groups and several physical field simulation results corresponding to the several scanning process parameter groups. The several scanning process parameter groups and several physical field simulation results constitute a physical field simulation dataset.

[0009] Furthermore, the step of comparing the real-time physical field with the target physical field to calculate the uniformity deviation includes: A number of wafer surface sampling points are obtained, and based on the number of wafer surface sampling points, the point-by-point comparison results of the real-time physical field and the target physical field are obtained to obtain the position deviation term; The variance is calculated based on the real-time physical field to obtain the intra-chip uniformity term; Obtain the first weighting coefficient and the second weighting coefficient, and based on the first weighting coefficient, the position deviation term, the second weighting coefficient, and the intra-chip uniformity term, obtain the uniformity deviation.

[0010] Furthermore, the formula for the uniformity deviation is:

[0011] in, This indicates the uniformity deviation caused by the flow input variable. Indicates the flow input variable. This represents the first weighting coefficient. Indicates the number of sampling points on the wafer surface. Indicates in The real-time physical field at that location Represents spatial coordinates, Indicates in The target physical field at that location This represents the second weighting coefficient. Indicates variance.

[0012] Furthermore, the step of performing gradient inversion based on the uniformity deviation to calculate the flow adjustment value of the sprinkler head area includes: Automatic differentiation is performed based on the uniformity deviation to obtain several partition gradients; Obtain the learning rate and damping factor, and based on the partition gradient and the damping factor, obtain the adaptive step size; Based on the adaptive step size and the learning rate, the partition flow adjustment value is calculated, and several partition flow adjustment values ​​constitute the flow adjustment value of the sprinkler head area.

[0013] Furthermore, the partition traffic adjustment value is:

[0014] in, Indicates the first The first partition, the first The partition traffic adjustment value at any given time. This represents the truncation function. Indicates the first The first partition, the first The partition traffic adjustment value at any given time. Indicates the learning rate. Represents the partition gradient. Indicates adaptive step size, Indicates the damping factor. This represents the minimum feasible value. This represents the maximum feasible value.

[0015] Furthermore, after the step of performing gradient inversion based on the uniformity deviation to calculate the flow adjustment value of the spray head area, the flow adjustment value being used to control the mass flow controller of the spray head area in real time, the method further includes: Actual physical data are collected through in-situ monitoring sensors within the epitaxial reaction chamber; Obtain the residual network, construct a residual correction model based on the residual network, input the real-time physical field and the actual physical data into the residual correction model to output the sensor position correction amount, and use the sensor position correction amount as a supervision signal to adjust the input of the real-time inference model.

[0016] A real-time flow field control system for an epitaxial reaction chamber, employing the real-time flow field control method for an epitaxial reaction chamber as described in the above technical solution, the system comprising: The simulation module is used to perform three-dimensional modeling of the epitaxial reaction chamber in order to obtain a physical field simulation dataset. The training module is used to acquire a neural operator network and train a real-time inference model based on the physical field simulation dataset and the neural operator network. The inference module is used to collect real-time process parameters of the epitaxial furnace and input the real-time process parameters into the real-time inference model to obtain the real-time physical field of the epitaxial reaction chamber. The calculation module is used to acquire the target physical field and compare the real-time physical field with the target physical field to calculate the uniformity deviation. The control module is used to perform gradient inversion based on the uniformity deviation to calculate the flow adjustment value of the spray head area, and the flow adjustment value is used to control the mass flow controller of the spray head area in real time.

[0017] A computer includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the real-time flow field control method for an epitaxial reaction chamber as described in the above technical solution.

[0018] Compared with the prior art, the beneficial effects of the present invention are as follows: By constructing the real-time inference model through the neural operator network, the physical field reconstruction time in the epitaxial reaction chamber can be shortened to the millisecond level, enabling online real-time identification and perception of the epitaxial growth process within the epitaxial reaction chamber; by calculating the uniformity deviation and its gradient relative to the control parameters, the inversion from the deviation value to the control parameter adjustment amount can be achieved, specifically, the flow adjustment value of the spray head area can be inverted, thereby enabling online real-time active control based on the flow adjustment value, and timely response and intervention to microscopic non-uniformity and transient disturbances in the epitaxial growth process; combined with real-time data from sensors, the model input is corrected, which helps to avoid the negative interference of disturbance factors such as graphite disk surface residues or equipment aging on epitaxial growth, and improves the wavelength hit rate of the epitaxial wafer. Attached Figure Description

[0019] Figure 1 This is a flowchart of the real-time flow field control method in the epitaxial reaction chamber according to the first embodiment of the present invention; Figure 2 This is a structural block diagram of the real-time flow field control system of the epitaxial reaction chamber in the second embodiment of the present invention; The following detailed description, in conjunction with the accompanying drawings, will further illustrate the present invention. Detailed Implementation

[0020] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Several embodiments of the invention are illustrated in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

[0021] It should be noted that when a component is said to be "fixed to" another component, it can be directly on the other component or there may be an intervening component. When a component is said to be "connected to" another component, it can be directly connected to the other component or there may be an intervening component. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this document are for illustrative purposes only.

[0022] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0023] Please see Figure 1The real-time flow field control method for the epitaxial reaction chamber in the first embodiment of the present invention includes the following steps: Step S10: Perform three-dimensional modeling of the epitaxial reaction chamber to obtain a physical field simulation dataset; Preferably, the CFD method is used for three-dimensional modeling, and the physical field simulation dataset is divided into a training set, a validation set, and a test set in a ratio of 7:2:1.

[0024] Step S10 includes: S110: Construct a three-dimensional geometric model based on the epitaxial reaction chamber. The three-dimensional geometric model includes a spray head sub-model, a reaction chamber sub-model, a graphite base model, and an exhaust port model. Understandably, key components and regions of the epitaxial reaction chamber are modeled.

[0025] S120: Mesh the three-dimensional geometric model and refine the mesh according to the flow gradient; Preferably, mesh refinement is performed in areas with large flow gradients. Specifically, mesh refinement is performed at the spray head outlet of the spray head sub-model and at the base edge of the graphite base sub-model.

[0026] S130: Set a set of boundary conditions for the three-dimensional geometric model and set a physical mode to form a simulation model. The physical mode is a turbulence mode, a chemical reaction mode, a radiation heat transfer mode, or a multi-component transport mode. Preferably, the set of boundary conditions includes the gas composition at the spray head inlet, the flow rate at the spray head inlet, the temperature at the spray head inlet, the thermal boundary conditions of the cavity wall, the base rotation speed, and the exhaust port pressure conditions. The physical mode is selected as the turbulence mode, that is, the simulation model is established based on the turbulence model.

[0027] S140: Based on the simulation model, a physical field simulation dataset is formed through parametric scanning.

[0028] S140 further includes: S1410: Set a process parameter variable group, which includes several variables, such as chamber pressure variable, base temperature variable, several zone gas flow rate variables, or total carrier gas flow rate variable. S1420: Set the variable range and scan parameters for the variable; S1430: Based on the process parameter variable group, several variable ranges and several scanning parameters, perform parameterized scanning simulation on the simulation model to obtain several scanning process parameter groups and several physical field simulation results corresponding to the several scanning process parameter groups. The several scanning process parameter groups and several physical field simulation results constitute a physical field simulation dataset.

[0029] Preferably, the chamber pressure variable ranges from 10 Torr to 760 Torr, the base temperature variable ranges from 500°C to 1200°C, the total carrier gas flow rate variable ranges from 5 slm to 50 slm, and the gas flow rate variables of several zones, such as the gas flow rate variable of the three-source zone ranging from 0.01 slm to 0.5 slm and the gas flow rate variable of the five-source zone ranging from 0.1 slm to 2 slm, are defined as the number of scans or the scan step size corresponding to different variables. Specifically, by setting the scan parameters, the combination of variables within the process parameter variable group can form more than 5000 sets of operating conditions to cover the key areas of the process parameters.

[0030] Step S20: Obtain the neural operator network, and train a real-time inference model based on the physical field simulation dataset and the neural operator network; Preferably, the Fourier Neural Operator (FNO) is used as the core network architecture of the neural operator network, which understandably benefits the learning of operator mappings in the frequency domain and has generalization ability unaffected by resolution.

[0031] Furthermore, the real-time inference model includes an input projection layer, several Fourier integral operator layers, local convolutional layers, activation functions, and an output projection layer. The input projection layer maps the process parameter vector and low-resolution physical field grid to a high-dimensional latent space. The L-layer Fourier integral operator layer transforms the spatial field to the frequency domain through the fast Fourier transform of each layer, applies a learnable parameterized kernel function in the frequency domain, and then transforms it back to the spatial domain through inverse Fourier transform, realizing global information interaction across spatial locations. The local convolutional layers are used to superimpose local convolution operations to capture local fine features. The activation function adopts the SiLU (Sigmoid Linear Unit) activation function, which is beneficial to avoid gradient vanishing and accelerate convergence. The output projection layer maps the latent space field back to the physical field space. The trained real-time inference model can be deployed to the embedded control system of the epitaxial device.

[0032] Step S30: Collect the real-time process parameters of the epitaxial furnace and input the real-time process parameters into the real-time inference model to obtain the real-time physical field of the epitaxial reaction chamber; Preferably, the real-time process parameters include the flow setpoints of each partition's MFC (Mass Flow Controller), chamber pressure, base temperature, and carrier gas flow rate. These parameters are input into the real-time inference model, and after forward propagation calculation, the distribution of the reconstructed flow field and temperature field within the extended reaction chamber at the current moment can be output. The real-time physical field is the reconstructed flow field or temperature field. The inference time is 5 milliseconds to 50 milliseconds. The inference process of the real-time physical field can be considered as a complete forward process of a neural operator network. Based on the chamber pressure and temperature setpoints, the real-time physical field is mapped using a neural operator.

[0033] Understandably, by constructing the real-time inference model through the neural operator network, the physical field reconstruction time within the epitaxial reaction chamber can be shortened to the millisecond level, enabling online real-time identification and perception of the epitaxial growth process within the epitaxial reaction chamber.

[0034] Step S40: Obtain the target physical field, compare the real-time physical field with the target physical field to calculate the uniformity deviation; Preferably, the uniformity deviation is calculated by constructing a uniformity deviation loss function.

[0035] Step S40 includes: S410: Acquire several wafer surface sampling points, and based on the several wafer surface sampling points, obtain the point-by-point comparison results of the real-time physical field and the target physical field to obtain the position deviation term; The position deviation term is used to measure the degree of deviation between the predicted value and the target value at each point.

[0036] S420: Calculate the variance based on the real-time physical field to obtain the intra-chip uniformity term; Using the variance of the predicted physical field within the wafer range as the intra-wafer uniformity term is beneficial for minimizing the difference between the wafer center and the edge.

[0037] S430: Obtain the first weighting coefficient and the second weighting coefficient, and obtain the uniformity deviation based on the first weighting coefficient, the position deviation term, the second weighting coefficient, and the intra-chip uniformity term.

[0038] The first weighting coefficient and the second weighting coefficient can be adjusted according to process requirements.

[0039] The formula for the uniformity deviation is:

[0040] in, This indicates the uniformity deviation caused by the flow input variable. Indicates the flow input variable. This represents the first weighting coefficient. Indicates the number of sampling points on the wafer surface. Indicates in The real-time physical field at that location Represents spatial coordinates, Indicates in The target physical field at that location This represents the second weighting coefficient. Indicates variance.

[0041] Understandably, the above formula is the uniformity deviation function. When the process requires absolute thickness uniformity, the first weighting coefficient is increased, and when the process requires intra-wafer homogeneity, the second weighting coefficient is increased.

[0042] Step S50: Perform gradient inversion based on the uniformity deviation to calculate the flow adjustment value of the spray head area. The flow adjustment value is used to control the mass flow controller of the spray head area in real time.

[0043] Understandably, by calculating the uniformity deviation and its gradient relative to the control parameters, the inversion from the deviation value to the control parameter adjustment amount can be achieved. Specifically, the flow adjustment value of the spray head area can be inverted, thereby enabling online real-time active control based on the flow adjustment value, and timely response and intervention to microscopic non-uniformity and transient disturbances in the epitaxial growth process.

[0044] Step S50 includes: S510: Perform automatic differentiation based on the uniformity deviation to obtain several partition gradients; Preferably, the real-time inference model is regarded as an end-to-end differentiable function. The gradient of the uniformity deviation with respect to the control parameters can be directly calculated using the backpropagation algorithm. Specifically, the gradient is calculated through an automatic differentiation mechanism. The spray head area is divided into multiple zones. The zone gradient represents the degree of influence of each tiny unit increase in zone flow rate on the overall flow field uniformity deviation. A negative gradient indicates that increasing the zone flow rate helps to reduce the deviation, while a positive gradient indicates that the zone flow rate should be reduced.

[0045] S520: Obtain the learning rate and damping factor, and obtain the adaptive step size based on the partition gradient and the damping factor; Preferably, the learning rate is set according to the response speed of MFC, and the value range is 0.001~0.01 to ensure that the hardware can follow the changes in instructions well. The damping factor is used to prevent control oscillations caused by large gradients, and its value range is 0.1~1. When the gradient is large, the adaptive step size will automatically decrease to avoid oscillations. When the gradient is small, the adaptive step size will maintain a large step size to accelerate convergence.

[0046] S530: Based on the adaptive step size and the learning rate, calculate the partition flow adjustment value, and several partition flow adjustment values ​​constitute the flow adjustment value of the sprinkler head area.

[0047] Preferably, upper and lower limits are set for the flow adjustment value, and a truncation function is used to limit the flow within the physically feasible range of MFC.

[0048] The partition traffic adjustment value is:

[0049] in, Indicates the first The first partition, the first The partition traffic adjustment value at any given time. This represents the truncation function. Indicates the first The first partition, the first The partition traffic adjustment value at any given time. Indicates the learning rate. Represents the partition gradient. Indicates adaptive step size, Indicates the damping factor. This represents the minimum feasible value. This represents the maximum feasible value.

[0050] Following step S50, the method further includes: S540: Collect actual physical data through the in-situ monitoring sensor in the epitaxial reaction chamber; Preferably, the in-situ monitoring sensor can be a thermocouple, an infrared thermometer, or a laser Doppler velocimeter, etc., to collect actual temperature and flow rate data.

[0051] S550: Obtain the residual network, construct a residual correction model based on the residual network, input the real-time physical field and the actual physical data into the residual correction model to output the sensor position correction amount, and use the sensor position correction amount as a supervision signal to adjust the input of the real-time inference model.

[0052] Preferably, the residual network is a 3-5 layer fully connected network. The lightweight residual network makes real-time predictions and uses the correction amount at the sensor distribution location as a supervision signal for online incremental learning. This is beneficial to improve long-term operating accuracy, correct model input, avoid the negative interference of disturbance factors such as graphite disk surface residues or equipment aging on epitaxial growth, and improve the wavelength hit rate of the epitaxial wafer.

[0053] Preferably, the epitaxial reaction chamber in this embodiment is an MOCVD reaction chamber. The spray head area is divided into three independent control areas: a central area, an intermediate ring area, and an outer ring area. Each area is equipped with an independent MFC. The epitaxial material grown is a GaN-based blue LED structure with a target wavelength of 450 nm and an allowable fluctuation of 1 nm. A CFD model is established for the MOCVD reaction chamber. The process parameter variables specifically include TMGa flow rate, NH3 flow rate, chamber pressure, and substrate temperature. The range of TMGa flow rate is 0.05 slm to 0.3 slm, the range of NH3 flow rate is 5 slm to 30 slm, and the range of chamber pressure is 100 Torr to 500 Torr. The Torr and base temperature range is 1000℃~1100℃. Parameter scanning generates 8000 sets of operating condition data. The real-time inference model has a total of 4 Fourier integral operator layers and a hidden dimension of 128. The concentration and temperature distribution of TMGa are used as prediction targets. The training rounds are 200. The trained real-time inference model is deployed to the device controller. The single inference time is 15ms. Specifically, in a certain batch of N-type GaN growth, the wafer edge growth rate was found to be low. The calculated uniformity deviation value was 0.038. Further calculations showed that the partition gradient of the central region was 0.002, and the partition gradient of the middle ring region was -0.015. The gradient of the outer ring region is -0.042. Based on the gradients of the multiple regions, the effect of flow rate adjustment on the reactant concentration at the growth interface can be analyzed. Specifically, the flow rate in the central region has little effect on the edge, increasing the flow rate in the middle ring region is beneficial to improving the uniformity of the transition region, and increasing the flow rate in the outer ring region is beneficial to increasing the edge concentration. The learning rate is 0.005, the damping factor is 0.5, and the adjustment amounts for the three regions are calculated to be 0, 0.000075, and 0.00021, respectively. The adjusted flow rates are input into the model again in the next control cycle to verify whether the deviation has decreased. After 8 iterations, the deviation converges to below 0.005, and the intra-chip uniformity increases from 97.2% to 99.3%.

[0054] Please see Figure 2 The second embodiment of the present invention provides a real-time flow field control system for an epitaxial reaction chamber, which applies the real-time flow field control method for an epitaxial reaction chamber as described in the first embodiment above. The system includes: Simulation module 10 is used to perform three-dimensional modeling of the epitaxial reaction chamber to obtain a physical field simulation dataset; The simulation module 10 includes: The first unit is used to construct a three-dimensional geometric model based on the epitaxial reaction chamber. The three-dimensional geometric model includes a spray head sub-model, a reaction chamber sub-model, a graphite base model, and an exhaust port model. The second unit is used to mesh the three-dimensional geometric model and refine the mesh according to the flow gradient. The third unit is used to set a set of boundary conditions for the three-dimensional geometric model and set a physical mode to form a simulation model. The physical mode is a turbulence mode, a chemical reaction mode, a radiation heat transfer mode, or a multi-component transport mode. The fourth unit is used to form a physical field simulation dataset based on the simulation model through parametric scanning.

[0055] The fourth unit is specifically used for: Set a process parameter variable group, which includes several variables, such as chamber pressure variable, base temperature variable, several zone gas flow rate variables, or total carrier gas flow rate variable. Set the variable range and scan parameters for the variable; Based on the process parameter variable group, several variable ranges, and several scanning parameters, the simulation model is subjected to parametric scanning simulation to obtain several scanning process parameter groups and several physical field simulation results corresponding to the several scanning process parameter groups. The several scanning process parameter groups and several physical field simulation results constitute a physical field simulation dataset.

[0056] Training module 20 is used to acquire a neural operator network and train a real-time inference model based on the physical field simulation dataset and the neural operator network. The inference module 30 is used to collect real-time process parameters of the epitaxial furnace and input the real-time process parameters into the real-time inference model to obtain the real-time physical field of the epitaxial reaction chamber. The calculation module 40 is used to acquire the target physical field and compare the real-time physical field with the target physical field to calculate the uniformity deviation. The computing module 40 includes: The fifth unit is used to acquire several wafer surface sampling points, and based on the several wafer surface sampling points, to obtain the point-by-point comparison results of the real-time physical field and the target physical field to obtain the position deviation term; The sixth unit is used to calculate the variance based on the real-time physical field to obtain the intra-chip uniformity term; The seventh unit is used to obtain the first weighting coefficient and the second weighting coefficient, and to obtain the uniformity deviation based on the first weighting coefficient, the position deviation term, the second weighting coefficient and the intra-chip uniformity term.

[0057] The formula for the uniformity deviation is:

[0058] in, This indicates the uniformity deviation caused by the flow input variable. Indicates the flow input variable. This represents the first weighting coefficient. Indicates the number of sampling points on the wafer surface. Indicates in The real-time physical field at that location Represents spatial coordinates, Indicates in The target physical field at that location This represents the second weighting coefficient. Indicates variance.

[0059] The control module 50 is used to perform gradient inversion based on the uniformity deviation to calculate the flow adjustment value of the spray head area, and the flow adjustment value is used to control the mass flow controller of the spray head area in real time.

[0060] The control module 50 includes: The eighth unit is used to perform automatic differentiation based on the uniformity deviation to obtain several partition gradients; The ninth unit is used to obtain the learning rate and damping factor, and to obtain the adaptive step size based on the partition gradient and the damping factor. The tenth unit is used to calculate the partition flow adjustment value based on the adaptive step size and the learning rate, and several partition flow adjustment values ​​constitute the flow adjustment value of the sprinkler head area.

[0061] The partition traffic adjustment value is:

[0062] in, Indicates the first The first partition, the first The partition traffic adjustment value at any given time. This represents the truncation function. Indicates the first The first partition, the first The partition traffic adjustment value at any given time. Indicates the learning rate. Represents the partition gradient. Indicates adaptive step size, Indicates the damping factor. This represents the minimum feasible value. This represents the maximum feasible value.

[0063] The eleventh unit is used to collect actual physical data through in-situ monitoring sensors within the extended reaction chamber; The twelfth unit is used to obtain the residual network, construct a residual correction model based on the residual network, input the real-time physical field and the actual physical data into the residual correction model to output the sensor position correction amount, and use the sensor position correction amount as a supervision signal to adjust the input of the real-time inference model.

[0064] The third embodiment of the present invention provides a computer, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the real-time flow field control method of the epitaxial reaction chamber as described in the first embodiment.

[0065] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0066] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for real-time control of the flow field in an epitaxial reaction chamber, characterized in that, Includes the following steps: A three-dimensional model of the epitaxial reaction chamber was created to obtain a physical field simulation dataset. A neural operator network is obtained, and a real-time inference model is trained based on the physical field simulation dataset and the neural operator network. Real-time process parameters of the epitaxial furnace are collected and input into the real-time inference model to obtain the real-time physical field of the epitaxial reaction chamber. The target physical field is acquired, and the real-time physical field is compared with the target physical field to calculate the uniformity deviation. Gradient inversion is performed based on the uniformity deviation to calculate the flow adjustment value of the spray head area, which is used to control the mass flow controller of the spray head area in real time.

2. The method for real-time control of the flow field in the epitaxial reaction chamber according to claim 1, characterized in that, The step of performing three-dimensional modeling of the epitaxial reaction chamber to obtain a physical field simulation dataset includes: A three-dimensional geometric model is constructed based on the epitaxial reaction chamber, which includes a spray head sub-model, a reaction chamber sub-model, a graphite base model, and an exhaust port model. The three-dimensional geometric model is meshed, and the mesh is refined according to the flow gradient; A set of boundary conditions is set for the three-dimensional geometric model, and a physical mode is set to form a simulation model. The physical mode is a turbulence mode, a chemical reaction mode, a radiation heat transfer mode, or a multi-component transport mode. Based on the simulation model, a physical field simulation dataset is formed through parametric scanning.

3. The method for real-time control of the flow field in the epitaxial reaction chamber according to claim 2, characterized in that, The step of generating a physical field simulation dataset based on the simulation model through parametric scanning includes: Set a process parameter variable group, which includes several variables, such as chamber pressure variable, base temperature variable, several zone gas flow rate variables, or total carrier gas flow rate variable. Set the variable range and scan parameters for the variable; Based on the process parameter variable group, several variable ranges, and several scanning parameters, the simulation model is subjected to parametric scanning simulation to obtain several scanning process parameter groups and several physical field simulation results corresponding to the several scanning process parameter groups. The several scanning process parameter groups and several physical field simulation results constitute a physical field simulation dataset.

4. The method for real-time control of the flow field in the epitaxial reaction chamber according to claim 1, characterized in that, The step of comparing the real-time physical field with the target physical field to calculate the uniformity deviation includes: A number of wafer surface sampling points are obtained, and based on the number of wafer surface sampling points, the point-by-point comparison results of the real-time physical field and the target physical field are obtained to obtain the position deviation term; The variance is calculated based on the real-time physical field to obtain the intra-chip uniformity term; Obtain the first weighting coefficient and the second weighting coefficient, and based on the first weighting coefficient, the position deviation term, the second weighting coefficient, and the intra-chip uniformity term, obtain the uniformity deviation.

5. The method for real-time control of the flow field in the epitaxial reaction chamber according to claim 4, characterized in that, The formula for the uniformity deviation is: in, This indicates the uniformity deviation caused by the flow input variable. Indicates the flow input variable. This represents the first weighting coefficient. Indicates the number of sampling points on the wafer surface. Indicates in The real-time physical field at that location Represents spatial coordinates, Indicates in The target physical field at that location This represents the second weighting coefficient. Indicates variance.

6. The method for real-time control of the flow field in the epitaxial reaction chamber according to claim 1, characterized in that, The step of performing gradient inversion based on the uniformity deviation to calculate the flow adjustment value of the sprinkler head area includes: Automatic differentiation is performed based on the uniformity deviation to obtain several partition gradients; Obtain the learning rate and damping factor, and based on the partition gradient and the damping factor, obtain the adaptive step size; Based on the adaptive step size and the learning rate, the partition flow adjustment value is calculated, and several partition flow adjustment values ​​constitute the flow adjustment value of the sprinkler head area.

7. The method for real-time control of the flow field in the epitaxial reaction chamber according to claim 6, characterized in that, The partition traffic adjustment value is: in, Indicates the first The first partition, the first The partition traffic adjustment value at any given time. This represents the truncation function. Indicates the first The first partition, the first The partition traffic adjustment value at any given time. Indicates the learning rate. Represents the partition gradient. Indicates adaptive step size, Indicates the damping factor. This represents the minimum feasible value. This represents the maximum feasible value.

8. The method for real-time control of the flow field in the epitaxial reaction chamber according to claim 1, characterized in that, After the step of performing gradient inversion based on the uniformity deviation to calculate the flow adjustment value of the spray head area, the flow adjustment value being used to control the mass flow controller of the spray head area in real time, the method further includes: Actual physical data are collected through in-situ monitoring sensors within the epitaxial reaction chamber; Obtain the residual network, construct a residual correction model based on the residual network, input the real-time physical field and the actual physical data into the residual correction model to output the sensor position correction amount, and use the sensor position correction amount as a supervision signal to adjust the input of the real-time inference model.

9. A real-time flow field control system for an epitaxial reaction chamber, employing the real-time flow field control method for an epitaxial reaction chamber as described in any one of claims 1 to 8, characterized in that, The system includes: The simulation module is used to perform three-dimensional modeling of the epitaxial reaction chamber in order to obtain a physical field simulation dataset. The training module is used to acquire a neural operator network and train a real-time inference model based on the physical field simulation dataset and the neural operator network. The inference module is used to collect real-time process parameters of the epitaxial furnace and input the real-time process parameters into the real-time inference model to obtain the real-time physical field of the epitaxial reaction chamber. The calculation module is used to acquire the target physical field and compare the real-time physical field with the target physical field to calculate the uniformity deviation. The control module is used to perform gradient inversion based on the uniformity deviation to calculate the flow adjustment value of the spray head area, and the flow adjustment value is used to control the mass flow controller of the spray head area in real time.

10. A computer comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the real-time flow field control method for the epitaxial reaction chamber as described in any one of claims 1 to 8.

Images