Vacuum evaporation process operation parameter control optimization method and system for film thickness uniformity

Abstract

The application discloses a vacuum evaporation process operation parameter control optimization method and system for film thickness uniformity, relates to the field of vacuum coating process parameter optimization, and comprises the following steps: a forward calculation expression is established, and a required constant evaporation rate is obtained by backstepping; a nonlinear least square algorithm is used to initialize and correct geometric configuration parameters; a combined kernel Gaussian process regression model is constructed to obtain a mapping relationship from process operation parameters to evaporation rate and spatial deposition distribution characteristics; a self-adaptive inertia weight particle swarm optimization algorithm is used for global optimization, and a candidate optimal process operation parameter combination set is output; a criterion that the nominal performance is optimal and the disturbance resistance is the strongest is used to screen and determine final optimal process operation parameters; and process operation parameters of an evaporation process are dynamically adjusted based on real-time film thickness monitoring data. The application has the advantages that the difficulty that film thickness distribution is difficult to accurately control and process parameters are dependent on experience and trial and error in vacuum evaporation is effectively solved.

Description

Technical Field

[0001] This invention relates to the field of vacuum coating process parameter optimization, specifically to a method and system for controlling and optimizing the operating parameters of a vacuum evaporation process aimed at achieving uniform film thickness. Background Technology

[0002] Vacuum evaporation is a process technology that involves placing the material to be deposited in a vacuum environment for evaporation or sublimation, thereby depositing a thin film on the surface of a substrate. It is widely used in optical thin films, OLED display panels, semiconductor devices, and surface treatment of automotive parts. Film thickness uniformity is a core indicator of evaporation quality, representing the degree of consistency in deposition thickness across different points within the effective coating area of ​​the substrate. This indicator directly affects the optical performance, electrical performance, and lifespan of the product.

[0003] In actual production, film thickness uniformity is affected by a combination of factors. The first category is equipment geometry, including the type and spatial location of the evaporation source, the relative distance and orientation between the substrate and the evaporation source, and the kinematic parameters of the substrate's revolution and rotation. These factors determine the spatial transport path of the evaporating particles and the deposition geometric efficiency. The second category is process control factors, including evaporation power, evaporation rate, deposition chamber vacuum, and substrate temperature. These factors directly affect the evaporation characteristics of the evaporation source and the surface migration behavior of the deposited particles. The third category is system drift factors, including evaporation surface position drift caused by changes in the evaporation source material balance, deviations of the actual evaporation power from the set value due to heater aging, and slow fluctuations in the chamber vacuum. These factors are time-varying and uncertain, and are difficult to completely eliminate through offline calibration.

[0004] The setting and optimization of vacuum evaporation process parameters mainly suffer from the following technical defects:

[0005] First, it relies on the trial-and-error experience of operators or simple orthogonal experiments. Trial-and-error methods are time-consuming and costly, and when the parameter dimension is high, it is difficult to guarantee search efficiency, often only obtaining local feasible solutions rather than global optimal solutions. Second, although existing research has attempted to construct film thickness distribution models using numerical simulation methods, pure physical models are difficult to accurately describe complex non-ideal factors such as evaporation angle deviation and material adhesion coefficient changes, while pure data-driven models heavily rely on the number of samples and have poor interpretability. Both have their limitations, and no effective solution to integrate them has yet been found. Third, existing optimization methods usually only output a single optimal solution, without considering the impact of model prediction uncertainty and actual production condition fluctuations on the robustness of the solution, resulting in the optimization results often experiencing a sharp degradation in performance due to small perturbations in actual production. Fourth, offline optimization results are fixed once put into production, lacking a dynamic adjustment mechanism based on real-time monitoring data, and cannot effectively cope with equipment drift and environmental changes. Summary of the Invention

[0006] To address the aforementioned technical problems, this paper provides a method and system for optimizing the control parameters of a vacuum evaporation process to ensure film thickness uniformity. This technical solution solves the problems mentioned in the background art, which rely on the trial-and-error experience of operators or simple orthogonal experiments, and attempts to construct film thickness distribution models using numerical simulation methods. Existing optimization methods typically only output a single optimal solution, without considering the impact of model prediction uncertainties and fluctuations in actual production conditions on the robustness of the solution. Furthermore, offline optimization results are fixed once put into production, lacking a dynamic adjustment mechanism based on real-time monitoring data.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0008] A method for controlling and optimizing the operating parameters of a vacuum evaporation process aimed at achieving film thickness uniformity includes:

[0009] Based on Knudsen's cosine evaporation law with process-related angular distribution correction and the planetary motion law of the substrate, a forward calculation expression is established with constant evaporation rate and process-related angular distribution correction index as input and total periodic deposition thickness at each point on the substrate surface as output. Under the reference angular distribution condition, the constant evaporation rate required to achieve the target average film thickness is back-calculated from the arithmetic mean of the total periodic deposition thickness at the sampling points, which serves as the target evaporation rate for process control.

[0010] Based on multiple sets of measured film thickness data, a nonlinear least squares algorithm was used to initialize and correct the geometric configuration parameters, and the self-consistency between the parameters and the target evaporation rate was verified.

[0011] Multiple sets of process operating parameters and corresponding measured data of evaporation rate and film thickness distribution were collected. Using process operating parameters as input and evaporation rate and process correlation angular distribution correction index as output, a combined kernel Gaussian process regression model was constructed to obtain the mapping relationship between process operating parameters and evaporation rate and spatial deposition distribution characteristics.

[0012] Based on the predicted evaporation rate and the predicted process-related angular distribution correction index output by the combined kernel Gaussian process regression model, a fitness function is constructed with the goal of minimizing the relative deviation of film thickness uniformity. An adaptive inertial weighted particle swarm algorithm is then used for global optimization, and a set of candidate optimal process operating parameter combinations is output.

[0013] For each set of parameters in the candidate optimal process operation parameter combination set, the final optimal process operation parameter combination is determined based on the criteria of having the best nominal performance and the strongest anti-disturbance capability.

[0014] The optimal combination of process operating parameters is put into production operation, and the process operating parameters of the vapor deposition process are dynamically adjusted based on real-time film thickness monitoring data.

[0015] Preferably, the establishment of the forward calculation expression, which takes the constant evaporation rate and the process-related angular distribution correction index as input and the total periodic deposition thickness at each point on the substrate surface as output, includes:

[0016] K sampling points are predefined in the substrate's own coordinate system. The position of each sampling point relative to the geometric center of the substrate remains fixed, and the evaporation source coordinates and source point coordinates of the vacuum evaporation equipment chamber are obtained.

[0017] Based on the revolution angle and rotation angle of the substrate, as well as the ratio of the revolution and rotation angular velocities, the position coordinates of each sampling point are mapped to the real-time position coordinates under the fixed coordinate system of the chamber.

[0018] Based on the geometric parameters of the vacuum evaporation chamber and the type of evaporation source, and based on Knudsen's cosine evaporation law, a process-related angular distribution correction index is introduced to correct the actual spatial angular distribution of the evaporation flux. Combined with the inherent deposition property parameters of the material, an expression for the instantaneous deposition rate of the substrate sampling point is constructed to obtain the instantaneous deposition rate.

[0019] Based on the planetary motion law of the substrate, the instantaneous deposition rate within one revolution cycle is integrated over time to obtain the total periodic deposition thickness of each sampling point on the substrate surface within one complete revolution cycle.

[0020] Based on the total periodic deposition thickness at each sampling point on the substrate, the arithmetic mean of the total periodic deposition thickness at all sampling points is calculated as the average film thickness under this condition.

[0021] Based on the baseline angular distribution correction index corresponding to the baseline process parameters, and taking the target average film thickness as the average film thickness under this operating condition, the required constant evaporation rate is calculated as the target evaporation rate for process control.

[0022] The forward calculation expression, taking the constant evaporation rate and the process-related angular distribution correction index as input and the total periodic deposition thickness at each point on the substrate surface as output, is as follows:

[0023]

[0024] In the formula, Let the total periodic deposition thickness be the thickness at the j-th sampling point. For a constant mass evaporation rate, These are inherent depositional property parameters of the material. Let T be the solid density of the deposited thin film, and T be the orbital period. for The angle between the surface normal of the evaporation source at time t and the line connecting the source point to the j-th sampling point. for The angle between the substrate surface normal and the connecting line at the j-th sampling point at time j. for The straight-line distance from the evaporation source to the j-th sampling point at time step [j]. This is the correction index for the process-related angular distribution.

[0025] Preferably, the step of initializing and correcting the geometric configuration parameters based on multiple sets of measured film thickness data using a nonlinear least squares algorithm, and verifying their consistency with the target evaporation rate, includes:

[0026] Based on the forward calculation expression, the set of geometric configuration parameters to be corrected is determined. The geometric configuration parameters to be corrected include at least: the three-dimensional spatial coordinates of the evaporation source in the chamber coordinate system, and the actual tilt angle and eccentricity of the substrate revolution axis.

[0027] Under the conditions of reference angular distribution and constant evaporation rate, a composite forward calculation function for the total periodic deposition thickness at each sampling point on the substrate surface is constructed based on the forward calculation expression, using geometric configuration parameters and revolution period parameters as inputs.

[0028] Multiple sets of coating test data under the benchmark process were collected. Each set of tests included the measured film thickness at each sampling point on the substrate surface, as well as the preliminary measured values ​​of the geometric configuration parameters under the current installation state.

[0029] The calibration process uses the center deposition rate measured by the built-in crystal oscillator monitoring module at a fixed position in the chamber under the reference operating conditions as the global calibration constraint, fixes the scaling reference of the evaporation rate, and eliminates the coupling ambiguity between geometric parameters and evaporation rate.

[0030] The initial measured values ​​of the geometric configuration parameters of each group of experiments are input into the composite forward calculation function, and the theoretical value of the total periodic deposition thickness of each sampling point is output. The sum of squares of the differences between the theoretical value and the measured value of the total periodic deposition thickness of each point is used as the objective function.

[0031] With the objective function as the goal, the Levenberg-Marquardt nonlinear least squares algorithm is used to iteratively solve the geometric configuration parameters to be corrected. The iteration is terminated when the decrease in the objective function value is less than a preset threshold or the number of iterations reaches a preset upper limit, and the optimal estimated value of the corrected geometric configuration parameters is output.

[0032] Substitute the corrected geometric configuration parameters into the composite forward calculation function to recalculate the arithmetic mean of the total periodic deposition thickness at all sampling points;

[0033] Determine whether the overall fitting error between the calculated and measured values ​​of the total periodic deposition thickness at each sampling point after correction is within the preset allowable range. If yes, it indicates that the corrected geometric configuration parameters are consistent with the calibrated evaporation rate, the verification is passed, the corrected geometric configuration parameters are output and the process proceeds to the next step; otherwise, optimize the initial values ​​and re-execute the geometric parameter correction until the fitting accuracy meets the requirements.

[0034] Preferably, the construction of the combined nuclear Gaussian process regression model to obtain the mapping relationship between process operating parameters and evaporation rate and spatial deposition distribution characteristics includes:

[0035] Based on the vacuum evaporation process flow, a set of process operating parameters that affect the evaporation rate and deposition distribution are determined. The process operating parameters include at least heating power, substrate temperature and chamber pressure.

[0036] The Latin hypercube sampling method was used to generate multiple sets of process operating parameters and conduct vacuum evaporation tests. The actual evaporation rate values ​​corresponding to each set of tests were collected in real time using the built-in crystal oscillator monitoring module of the evaporation equipment. The substrate film thickness data were collected by metallographic microscope and coulometer. The process correlation angle distribution correction index under the corresponding working conditions was obtained by nonlinear fitting.

[0037] A hybrid kernel function consisting of a weighted combination of radial basis kernels and linear kernels was constructed. A multi-task Gaussian process regression framework with shared kernels was adopted. The heating power, substrate temperature and chamber pressure vector were used as inputs, and the corresponding actual evaporation rate value and angular distribution correction exponent were used as outputs. The Gaussian process regression model was trained to obtain a mapping model from process operating parameters to evaporation rate and deposition distribution.

[0038] After training, for any new combination of process operating parameters, the Gaussian process regression model outputs the predicted mean and variance of the evaporation rate, as well as the predicted mean and variance of the process correlation angular distribution correction index, and uses the predicted mean as the predicted value of the evaporation characteristics and deposition distribution under that parameter combination.

[0039] Preferably, the construction of the fitness function and the use of an adaptive inertial weighted particle swarm optimization algorithm for global optimization, outputting a set of candidate optimal process operating parameter combinations, includes:

[0040] Determine the range constraints of the operating parameters to be optimized and the process feasibility constraints, and construct a set of constraints. The operating parameters to be optimized include at least heating power, substrate temperature and chamber pressure.

[0041] For any combination of candidate process operating parameters, the predicted evaporation rate and the predicted angular distribution correction index output by the combined kernel Gaussian process regression model are substituted into the forward formula to calculate the average film thickness. It is then determined whether the deviation between the average film thickness and the target average film thickness is within the allowable range. If it exceeds the allowable range, the candidate parameter combination is marked as an infeasible solution and discarded. If it is within the allowable range, the candidate parameter combination is marked as a feasible solution and proceeds to the next evaluation step.

[0042] Substitute the predicted evaporation rate and the predicted process-related angular distribution correction index of the selected values ​​into the corrected forward calculation expression to calculate the theoretical value of the total periodic deposition thickness at each sampling point on the substrate surface.

[0043] Based on the theoretical value of the total periodic deposition thickness at each sampling point on the substrate surface, the relative deviation of film thickness uniformity under this parameter combination is calculated, and the relative deviation value of uniformity is used as the fitness function value.

[0044] Initialize the particle swarm population, dynamically adjust the inertial weight of each generation of particles using an adaptive inertial weight strategy, and iterate to find the optimal value until the convergence condition is met.

[0045] During the iterative optimization process, the position vector of each particle is a combination of process operation parameters, and the velocity vector is the change of each parameter. The particle continuously updates its own position by tracking the individual historical best position and the global historical best position until it reaches the preset maximum number of iterations or the fitness value converges to the preset threshold.

[0046] After the algorithm converges, the top N combinations of process operating parameters with the best fitness are output as the set of candidate optimal combinations of process operating parameters. The fitness value of each group and the joint prediction variance of the combined kernel Gaussian process regression model under the parameter combination are recorded.

[0047] Preferably, the step of selecting and determining the final optimal combination of process operating parameters for each set of candidate optimal process operating parameter combinations, based on the criteria of having the best nominal performance and the strongest anti-disturbance capability, includes:

[0048] Extract the r-th group of process operating parameters and their corresponding combined kernel Gaussian process regression model joint prediction variance from the candidate optimal process operating parameter combination set;

[0049] The evaporation rate corresponding to the 95% confidence upper limit of the mean predicted by the combined kernel Gaussian process regression model, and the values ​​corresponding to the larger relative deviation of film thickness uniformity in the upper and lower limits of the 95% confidence interval of the angular distribution correction index, are used together as the worst-case evaporation characteristic estimate at this parameter point.

[0050] Substitute the worst-case evaporation characteristic estimate into the corrected forward calculation expression to calculate the total periodic deposition thickness at each sampling point under the worst-case condition and the relative deviation of film thickness uniformity under the worst-case condition.

[0051] Substitute the predicted mean of the combined nuclear Gaussian process regression model into the corrected forward calculation expression to calculate the total periodic deposition thickness at each sampling point under nominal conditions and the relative deviation of film thickness uniformity under nominal conditions.

[0052] Based on the relative deviation of film thickness uniformity under worst-case conditions and the relative deviation of film thickness uniformity under nominal conditions, a robustness comprehensive evaluation index is constructed.

[0053] Calculate the comprehensive evaluation index corresponding to each group of candidate parameters, and select the parameter combination corresponding to the minimum value of the comprehensive evaluation index as the final optimal process operation parameter combination.

[0054] Preferably, the step of putting the final optimal combination of process operating parameters into production operation, and dynamically adjusting the process operating parameters of the vapor deposition process based on real-time film thickness monitoring data, includes:

[0055] The final optimal combination of process operating parameters is sent as the initial set value to the vacuum evaporation equipment control system for execution.

[0056] In the actual vapor deposition process, the deposition rate at the center point of the evaporation source is monitored in real time by the built-in crystal oscillator monitoring module of the vapor deposition equipment, and the PID controller is used to maintain the evaporation rate at the target evaporation rate.

[0057] Every S production batches, the film thickness at each sampling point in the effective coating area of ​​the substrate is measured using a metallographic microscope and coulometric instrument.

[0058] Input the current actual process operating parameters into the combined kernel Gaussian process regression model, and output the predicted value of the evaporation rate and the predicted value of the process correlation angular distribution correction index under the current operating conditions;

[0059] Substitute the predicted mean of the evaporation rate and the predicted value of the process correlation angle distribution correction index into the corrected forward calculation expression to calculate the theoretical value of the total periodic deposition thickness at each sampling point under the current process operating parameters.

[0060] The deviation between the measured film thickness at each sampling point and the theoretical total thickness of the periodic deposition is calculated point by point. The deviation distribution vector is obtained, and the root mean square error of the deviation distribution vector is calculated.

[0061] If the root mean square error of the deviation distribution vector is less than the dead zone threshold, the system is considered to be operating normally, the current set value is maintained, and normal production continues.

[0062] If the root mean square error of the deviation distribution vector is greater than or equal to the dead zone threshold, it is determined that the system has experienced inter-batch operating condition drift. Further analysis of the spatial distribution pattern of the deviation distribution vector is then conducted to determine the type of drift.

[0063] Furthermore, this solution proposes a vacuum evaporation process operating parameter control and optimization system for film thickness uniformity, used to implement the aforementioned method for controlling and optimizing vacuum evaporation process operating parameters for film thickness uniformity, including the following basic functional units:

[0064] The coordinate calibration unit is used to calibrate and standardize the position coordinates of each sampling point on the substrate using a 2.5D projector, providing a precise point reference for calculating the forward deposition thickness.

[0065] The geometric parameter calibration unit is used to calibrate the initial measurement values ​​of geometric configuration parameters such as the three-dimensional coordinates of the evaporation source chamber, the actual tilt angle and eccentricity of the substrate revolution axis using a coordinate measuring machine.

[0066] The film thickness distribution detection unit is used to obtain the measured value of the total periodic deposition thickness at each sampling point on the substrate surface by combining metallographic microscope and coulometric instrument, and output the in-plane film thickness distribution data.

[0067] The evaporation rate calibration unit is used to collect the real-time deposition rate at the center point of the evaporation source under the benchmark operating conditions, providing global scaling and calibration constraints for geometric parameter correction and eliminating the coupling ambiguity between geometric parameters and evaporation rate;

[0068] The real-time rate monitoring unit is used to collect the deposition rate at the center point of the evaporation source in real time based on the built-in crystal oscillator monitoring module of the evaporation equipment, and provide feedback signals to the PID controller to maintain the evaporation rate at the target evaporation rate.

[0069] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0070] This scheme constructs a forward deposition thickness analytical model coupled with Knudsen's cosine evaporation law and planetary motion, with angular distribution correction. The process-related angular distribution correction index is introduced as an optimizable variable into the control system. The Levenberg-Marquardt algorithm is then used to perform in-situ self-correction of geometric installation errors based on measured film thickness data, effectively eliminating parameter mismatch between the theoretical model and engineering practice. Furthermore, a combined kernel Gaussian process regression model is used to establish a high-fidelity mapping relationship between process operating parameters, evaporation rate, and spatial deposition distribution. This allows for simultaneous optimization of the target film thickness and minimization of relative deviations in film thickness uniformity. By introducing a robustness evaluation index based on prediction variance, the optimized process parameter combination maintains stable uniformity indicators even when facing model uncertainties. Finally, a fast and slow dual-timescale closed-loop control strategy is employed, achieving a leap from single-batch qualification to controllable batch consistency, effectively improving the engineering reliability and product quality consistency of the vacuum evaporation process. Attached Figure Description

[0071] Figure 1 This is a flowchart of the vacuum evaporation process operating parameter control and optimization method for film thickness uniformity according to the present invention;

[0072] Figure 2 The flowchart for constructing the fitness function and using the adaptive inertial weighted particle swarm optimization algorithm for global optimization, and outputting the candidate optimal process operation parameter combination set is provided in this invention. Detailed Implementation

[0073] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0074] Reference Figure 1 As shown, a method for controlling and optimizing the operating parameters of a vacuum evaporation process aimed at achieving film thickness uniformity includes:

[0075] Based on Knudsen's cosine evaporation law with process-related angular distribution correction and the planetary motion law of the substrate, a forward calculation expression is established with constant evaporation rate and process-related angular distribution correction index as input and total periodic deposition thickness at each point on the substrate surface as output. Under the reference angular distribution condition, the constant evaporation rate required to achieve the target average film thickness is back-calculated from the arithmetic mean of the total periodic deposition thickness at the sampling points, which serves as the target evaporation rate for process control.

[0076] Based on multiple sets of measured film thickness data, a nonlinear least squares algorithm was used to initialize and correct the geometric configuration parameters, and the self-consistency between the parameters and the target evaporation rate was verified.

[0077] Multiple sets of process operating parameters and corresponding measured data of evaporation rate and film thickness distribution were collected. Using process operating parameters as input and evaporation rate and process correlation angular distribution correction index as output, a combined kernel Gaussian process regression model was constructed to obtain the mapping relationship between process operating parameters and evaporation rate and spatial deposition distribution characteristics.

[0078] Based on the predicted evaporation rate and the predicted process-related angular distribution correction index output by the combined kernel Gaussian process regression model, a fitness function is constructed with the goal of minimizing the relative deviation of film thickness uniformity. An adaptive inertial weighted particle swarm algorithm is then used for global optimization, and a set of candidate optimal process operating parameter combinations is output.

[0079] For each set of parameters in the candidate optimal process operation parameter combination set, the final optimal process operation parameter combination is determined based on the criteria of having the best nominal performance and the strongest anti-disturbance capability.

[0080] The optimal combination of process operating parameters is put into production operation, and the process operating parameters of the vapor deposition process are dynamically adjusted based on real-time film thickness monitoring data.

[0081] The overall implementation logic of the vacuum evaporation process parameter control optimization method for film thickness uniformity includes six progressive execution steps, which successively complete modeling correction, proxy fitting, parameter optimization, robust screening, and online closed-loop full-process control. Specifically, it includes:

[0082] Forward deposition thickness modeling layer: Based on Knudsen's cosine evaporation law with process-related angular distribution correction and the planetary motion law of the substrate, a forward calculation expression that can accurately calculate the film thickness of a single cycle is built. Under the premise of fixed reference angular distribution, the constant evaporation rate that matches the target film thickness is solved by inversely using the average film thickness of multiple points in the plane, providing a unified benchmark for subsequent correction and optimization.

[0083] Geometric parameter self-calibration layer: Collect multiple sets of measured film thickness data of substrates, rely on the Levenberg-Marquardt nonlinear least squares algorithm to invert and correct geometric configuration deviations such as chamber coordinates and rotation axis tilt angle, and verify the consistency between the correction results and the reference evaporation rate through film thickness fitting error verification to eliminate calculation deviations caused by equipment installation errors;

[0084] Process-Deposition Characteristics Proxy Modeling Layer: A Latin hypercube experiment was designed to collect multiple sets of process parameters, real-time evaporation rate and in-plane film thickness distribution samples. With heating power, substrate temperature and chamber pressure as inputs and evaporation rate and angular distribution correction index as outputs, a shared kernel multi-task combined Gaussian process regression model was built to establish a high-precision mapping relationship between process parameters and deposition rate and film thickness distribution characteristics.

[0085] Global process parameter optimization layer: The fitness function is constructed with dual constraints of target average film thickness deviation and film thickness uniformity relative deviation. The piecewise adaptive inertial weighted particle swarm algorithm is used to carry out global optimization, and multiple sets of candidate process parameter combinations that take into account both the total film thickness target and excellent uniformity are selected. The joint prediction variance of each model is recorded simultaneously.

[0086] Robust process screening layer: For all candidate parameter combinations, the worst-case film thickness index is constructed based on the 95% confidence interval of the Gaussian process. A weighted comprehensive evaluation index is constructed in combination with the nominal operating condition uniformity. The set with the best nominal performance and the strongest stability under model perturbation is selected as the final production process parameters.

[0087] Production closed-loop dynamic control layer: The optimal process selected is sent to the evaporation equipment for execution. The evaporation rate is monitored in real time by the built-in crystal oscillator and quickly adjusted by PID. At the same time, the film thickness data is sampled in batches to diagnose the drift type. The proxy model is updated or the geometric parameters are recalibrated in a targeted manner to achieve adaptive dynamic optimization of the long-term production process.

[0088] A forward calculation expression is established, taking a constant evaporation rate and a process-related angular distribution correction index as input, and the total periodic deposition thickness at each point on the substrate surface as output. This expression includes:

[0089] K sampling points are predefined in the substrate's own coordinate system. The position of each sampling point relative to the geometric center of the substrate remains fixed, and the evaporation source coordinates and source point coordinates of the vacuum evaporation equipment chamber are obtained.

[0090] Based on the revolution angle and rotation angle of the substrate, as well as the ratio of the revolution and rotation angular velocities, the position coordinates of each sampling point are mapped to the real-time position coordinates under the fixed coordinate system of the chamber.

[0091] Based on the geometric parameters of the vacuum evaporation chamber and the type of evaporation source, and based on Knudsen's cosine evaporation law, a process-related angular distribution correction index is introduced to correct the actual spatial angular distribution of the evaporation flux. Combined with the inherent deposition property parameters of the material, an expression for the instantaneous deposition rate of the substrate sampling point is constructed to obtain the instantaneous deposition rate.

[0092] Based on the planetary motion law of the substrate, the instantaneous deposition rate within one revolution cycle is integrated over time to obtain the total periodic deposition thickness of each sampling point on the substrate surface within one complete revolution cycle.

[0093] Based on the total periodic deposition thickness at each sampling point on the substrate, the arithmetic mean of the total periodic deposition thickness at all sampling points is calculated as the average film thickness under this condition.

[0094] Based on the baseline angular distribution correction index corresponding to the baseline process parameters, and taking the target average film thickness as the average film thickness under this operating condition, the required constant evaporation rate is calculated as the target evaporation rate for process control.

[0095] The forward calculation expression, taking the constant evaporation rate and the process-related angular distribution correction index as input and the total periodic deposition thickness at each point on the substrate surface as output, is as follows:

[0096]

[0097] In the formula, Let the total periodic deposition thickness be the thickness at the j-th sampling point. For a constant mass evaporation rate, These are inherent depositional property parameters of the material. Let T be the solid density of the deposited thin film, and T be the orbital period. for The angle between the surface normal of the evaporation source at time t and the line connecting the source point to the j-th sampling point. for The angle between the substrate surface normal and the connecting line at the j-th sampling point at time j. for The straight-line distance from the evaporation source to the j-th sampling point at time step [j]. This is the correction index for the process-related angular distribution.

[0098] Vacuum evaporation sources follow Knudsen's cosine evaporation law, which states that the flux density of evaporating particles in space is distributed according to the cosine of the normal direction of the evaporation source surface. That is, the flux is greatest in the normal direction, and the flux density is lower as the angle deviates from the normal. This inherent characteristic leads to a significantly higher deposition rate in the area of ​​the substrate facing the evaporation source than in the edge area under static substrate conditions, forming an intrinsic non-uniform distribution with a thick center and thin edges. To reduce this intrinsic non-uniformity, industrial evaporation equipment uses a planetary rotation mechanism to make the substrate rotate on its own axis and revolve around the central axis of the evaporation source during the coating process. The rotation causes the surface normal direction of each sampling point on the substrate to change periodically relative to the evaporation source, so that each sampling point is alternately facing and away from the evaporation source, averaging the angle dependence of single-point deposition over time. The revolution ensures that the substrate in different orientations in the chamber experiences an equivalent geometric transport path. After the two are superimposed, the geometric relationships such as the distance from any sampling point on the substrate to the evaporation source and the incident angle of evaporating particles change periodically with time within one revolution cycle. Based on the above principle, the instantaneous deposition rate within a complete revolution cycle is integrated over time, that is, the instantaneous deposition rate at each moment is accumulated and added together over time. The resulting integral value is the total periodic deposition thickness of the sampling point within one revolution cycle. The essence of this integration operation is to transform the non-uniform instantaneous deposition sequence in the dynamic motion process into a static equivalent accumulation result, thereby obtaining the total periodic deposition thickness that conforms to the actual working conditions.

[0099] The forward computation expression is derived in the following hierarchy:

[0100] Under constant evaporation rate and fixed angular distribution correction exponent, the instantaneous deposition rate at a single sampling point at any given time is linearly related to the evaporation rate:

[0101]

[0102] In the formula, for The instantaneous deposition rate at the j-th substrate sampling point at time j, where, These are inherent deposition property parameters of the material, characterizing the effective deposition ratio of incident evaporating particles on the substrate surface, encompassing the condensation coefficient and surface adhesion effect.

[0103] By integrating the instantaneous deposition rate over one revolution cycle, a forward calculation expression is constructed to obtain the total periodic deposition thickness of the j-th sampling point over one complete revolution cycle.

[0104] The arithmetic mean of the total periodic deposition thickness at all sampling points is used to obtain the average film thickness under this condition, which is then used to infer the target evaporation rate.

[0105]

[0106] In the formula, The arithmetic mean of the total periodic deposition thickness at all sampling points is taken. This represents the total number of sampling points;

[0107] Because the geometrical positions of each sampling point are different, their corresponding geometric integral constants differ from one another. Furthermore, the process-related angular distribution correction index changes the deposition flux weights of different angular paths, directly affecting the relative difference in the total periodic deposition thickness of each sampling point. Therefore, the total periodic deposition thickness of each sampling point is naturally different. This scheme uses a summation and averaging method to calculate the average film thickness of the entire region, and back-calculates the constant evaporation rate under the baseline angular distribution conditions. Its physical meaning is: to make the arithmetic mean of the total periodic deposition thickness of all sampling points in the entire region equal to the target average film thickness, rather than constraining the film thickness of each sampling point to be equal to the target value. The deviation of the film thickness of each sampling point from the target average value will be optimized in the subsequent optimization of process operation parameters.

[0108] Based on multiple sets of measured film thickness data, a nonlinear least squares algorithm was used to initialize and correct the geometric configuration parameters, and the self-consistency between the parameters and the target evaporation rate was verified, including:

[0109] Based on the forward calculation expression, the set of geometric configuration parameters to be corrected is determined. The geometric configuration parameters to be corrected include at least: the three-dimensional spatial coordinates of the evaporation source in the chamber coordinate system, and the actual tilt angle and eccentricity of the substrate revolution axis.

[0110] Under the conditions of reference angular distribution and constant evaporation rate, a composite forward calculation function for the total periodic deposition thickness at each sampling point on the substrate surface is constructed based on the forward calculation expression, using geometric configuration parameters and revolution period parameters as inputs.

[0111] Multiple sets of coating test data under the benchmark process were collected. Each set of tests included the measured film thickness at each sampling point on the substrate surface, as well as the preliminary measured values ​​of the geometric configuration parameters under the current installation state.

[0112] The calibration process uses the center deposition rate measured by the built-in crystal oscillator monitoring module at a fixed position in the chamber under the reference operating conditions as the global calibration constraint, fixes the scaling reference of the evaporation rate, and eliminates the coupling ambiguity between geometric parameters and evaporation rate.

[0113] The initial measured values ​​of the geometric configuration parameters of each group of experiments are input into the composite forward calculation function, and the theoretical value of the total periodic deposition thickness of each sampling point is output. The sum of squares of the differences between the theoretical value and the measured value of the total periodic deposition thickness of each point is used as the objective function.

[0114] With the objective function as the goal, the Levenberg-Marquardt nonlinear least squares algorithm is used to iteratively solve the geometric configuration parameters to be corrected. The iteration is terminated when the decrease in the objective function value is less than a preset threshold or the number of iterations reaches a preset upper limit, and the optimal estimated value of the corrected geometric configuration parameters is output.

[0115] Substitute the corrected geometric configuration parameters into the composite forward calculation function to recalculate the arithmetic mean of the total periodic deposition thickness at all sampling points;

[0116] Determine whether the overall fitting error between the calculated and measured values ​​of the total periodic deposition thickness at each sampling point after correction is within the preset allowable range. If yes, it indicates that the corrected geometric configuration parameters are consistent with the calibrated evaporation rate, the verification is passed, the corrected geometric configuration parameters are output and the process proceeds to the next step; otherwise, optimize the initial values ​​and re-execute the geometric parameter correction until the fitting accuracy meets the requirements.

[0117] The forward calculation expression is based on theory, but in actual calculations, geometric quantities such as distance and angle are affected by installation positioning errors or mechanical vibrations, causing deviations in their parameter values. This results in a discrepancy between the final calculated total periodic deposition thickness and the measured value, affecting subsequent process adjustments and optimizations. Therefore, this scheme requires initial calibration before the process begins. Geometric configuration parameters are corrected by inverting measured film thickness data to eliminate the impact of the aforementioned geometric deviations on the calculation of the total periodic deposition thickness, ensuring that subsequent film thickness uniformity optimization is based on accurate geometric parameter benchmarks. The composite forward calculation function refers to the process function that calculates the distance and angle of each sampling point at each time point using geometric configuration parameters, and substitutes these values ​​into the forward calculation expression to obtain the arithmetic mean of the total periodic deposition thickness of all sampling points. For the same process, without changing the process operating parameters and measurement configuration, the number of subsequent experiments can be dynamically adjusted based on the convergence rate of the objective function in the preceding experiments. If the process operating parameters or measurement configuration are changed, multiple experiments must be repeated to complete the calibration of the geometric configuration parameters.

[0118] A combined kernel Gaussian process regression model was constructed to obtain the mapping relationship between process operating parameters and evaporation rate and spatial deposition distribution characteristics, including:

[0119] Based on the vacuum evaporation process flow, a set of process operating parameters that affect the evaporation rate and deposition distribution are determined. The process operating parameters include at least heating power, substrate temperature and chamber pressure.

[0120] The Latin hypercube sampling method was used to generate multiple sets of process operating parameters and conduct vacuum evaporation tests. The actual evaporation rate values ​​corresponding to each set of tests were collected in real time using the built-in crystal oscillator monitoring module of the evaporation equipment. The substrate film thickness data were collected by metallographic microscope and coulometer. The process correlation angle distribution correction index under the corresponding working conditions was obtained by nonlinear fitting.

[0121] A hybrid kernel function consisting of a weighted combination of radial basis kernels and linear kernels was constructed. A multi-task Gaussian process regression framework with shared kernels was adopted. The heating power, substrate temperature and chamber pressure vector were used as inputs, and the corresponding actual evaporation rate value and angular distribution correction exponent were used as outputs. The Gaussian process regression model was trained to obtain a mapping model from process operating parameters to evaporation rate and deposition distribution.

[0122] After training, for any new combination of process operating parameters, the Gaussian process regression model outputs the predicted mean and variance of the evaporation rate, as well as the predicted mean and variance of the process correlation angular distribution correction index, and uses the predicted mean as the predicted value of the evaporation characteristics and deposition distribution under that parameter combination.

[0123] In the vacuum evaporation process, although the built-in crystal oscillator monitoring module can monitor the evaporation rate in real time and reflect its deviation from the target value, the measured value and the target value can only indicate that the rate has deviated, but cannot determine whether this deviation is caused by improper heating power setting, substrate temperature fluctuation or chamber pressure change. It also cannot reflect the regulatory effect of process parameters on evaporation angle distribution and thus on film thickness uniformity. This solution establishes a quantitative mapping relationship between process operating parameters and evaporation rate and deposition spatial distribution by constructing a Gaussian process regression model. When the measured rate deviates from the target value, the contribution of each process operating parameter to the deviation can be calculated in reverse, thereby providing a clear adjustment direction and magnitude for the feedback control of the equipment.

[0124] The expression for the hybrid kernel function is:

[0125]

[0126] In the formula, Sample of process operating parameters and The covariance function values ​​between them Let i be the vector of process operating parameters for the i-th group of experiments. Let j be the vector of process operating parameters for the j-th group of experiments. The signal variance of the radial basis kernel. Input dimensions include heating power, substrate temperature, and chamber pressure. For the first The length scale parameter of each input dimension The weighting coefficients are those of the linear kernel. for The transpose of, where, , , The hyperparameters are solved by maximizing the log-marginal likelihood function.

[0127] Reference Figure 2 As shown, a fitness function is constructed, and an adaptive inertia weighted particle swarm optimization algorithm is used for global optimization. The output set of candidate optimal process operating parameter combinations includes:

[0128] Determine the range constraints of the operating parameters to be optimized and the process feasibility constraints, and construct a set of constraints. The operating parameters to be optimized include at least heating power, substrate temperature and chamber pressure.

[0129] For any combination of candidate process operating parameters, the predicted evaporation rate and the predicted angular distribution correction index output by the combined kernel Gaussian process regression model are substituted into the forward formula to calculate the average film thickness. It is then determined whether the deviation between the average film thickness and the target average film thickness is within the allowable range. If it exceeds the allowable range, the candidate parameter combination is marked as an infeasible solution and discarded. If it is within the allowable range, the candidate parameter combination is marked as a feasible solution and proceeds to the next evaluation step.

[0130] Substitute the predicted evaporation rate and the predicted process-related angular distribution correction index of the selected values ​​into the corrected forward calculation expression to calculate the theoretical value of the total periodic deposition thickness at each sampling point on the substrate surface.

[0131] Based on the theoretical value of the total periodic deposition thickness at each sampling point on the substrate surface, the relative deviation of film thickness uniformity under this parameter combination is calculated, and the relative deviation value of uniformity is used as the fitness function value.

[0132] Initialize the particle swarm population, dynamically adjust the inertial weight of each generation of particles using an adaptive inertial weight strategy, and iterate to find the optimal value until the convergence condition is met.

[0133] During the iterative optimization process, the position vector of each particle is a combination of process operation parameters, and the velocity vector is the change of each parameter. The particle continuously updates its own position by tracking the individual historical best position and the global historical best position until it reaches the preset maximum number of iterations or the fitness value converges to the preset threshold.

[0134] After the algorithm converges, the top N combinations of process operating parameters with the best fitness are output as the set of candidate optimal combinations of process operating parameters. The fitness value of each group and the joint prediction variance of the combined kernel Gaussian process regression model under the parameter combination are recorded.

[0135] By combining the nuclear Gaussian process regression model, a mapping relationship between process operating parameters and evaporation rate and deposition distribution is established. The average film thickness is constrained within the target range to ensure that the total film thickness meets the standard. Under this premise, the combination of process operating parameters that makes the spatial distribution of the total deposition thickness of each sampling point most uniform is sought. Since the geometric transport paths experienced by each sampling point in one revolution cycle are naturally different, and the process correlation angle distribution correction index will change the deposition flux weight of different angle paths, there will be a certain degree of dispersion between the total deposition thickness of each point. The goal of this scheme is to compress this dispersion to the minimum, thereby meeting the film thickness uniformity requirements of the final product.

[0136] The relative deviation of film thickness uniformity is defined as:

[0137]

[0138] In the formula, For the relative deviation of film thickness uniformity, The total thickness of the maximum periodic deposition. The minimum total thickness of the periodic deposition. Let J be the total periodic deposition thickness at the j-th sampling point;

[0139] The formula for calculating adaptive inertia weight is:

[0140]

[0141] In the formula, Let the inertia weight be the weight for the k-th iteration. For minimum inertia weight, For maximum inertia weight, Let $\frac{ ... This represents the globally optimal fitness value up to the k-th iteration. This represents the maximum fitness value of the population in the k-th iteration.

[0142] For each set of candidate optimal process operating parameter combinations, the final optimal process operating parameter combination is determined based on the criteria of having the best nominal performance and the strongest disturbance resistance. This includes:

[0143] Extract the r-th group of process operating parameters and their corresponding combined kernel Gaussian process regression model joint prediction variance from the candidate optimal process operating parameter combination set;

[0144] The evaporation rate corresponding to the 95% confidence upper limit of the mean predicted by the combined kernel Gaussian process regression model, and the values ​​corresponding to the larger relative deviation of film thickness uniformity in the upper and lower limits of the 95% confidence interval of the angular distribution correction index, are used together as the worst-case evaporation characteristic estimate at this parameter point.

[0145] Substitute the worst-case evaporation characteristic estimate into the corrected forward calculation expression to calculate the total periodic deposition thickness at each sampling point under the worst-case condition and the relative deviation of film thickness uniformity under the worst-case condition.

[0146] Substitute the predicted mean of the combined nuclear Gaussian process regression model into the corrected forward calculation expression to calculate the total periodic deposition thickness at each sampling point under nominal conditions and the relative deviation of film thickness uniformity under nominal conditions.

[0147] Based on the relative deviation of film thickness uniformity under worst-case conditions and the relative deviation of film thickness uniformity under nominal conditions, a robustness comprehensive evaluation index is constructed.

[0148] Calculate the comprehensive evaluation index corresponding to each group of candidate parameters, and select the parameter combination corresponding to the minimum value of the comprehensive evaluation index as the final optimal process operation parameter combination.

[0149] The combined kernel Gaussian process regression model, while outputting the predicted mean of evaporation rate and distribution correction index, also provides a joint prediction variance. This variance reflects the overall prediction uncertainty of the model at this parameter combination. This scheme constructs the worst-case evaporation characteristics at a 95% confidence level to evaluate the worst uniformity performance of candidate parameters under the influence of model prediction uncertainty. The comprehensive evaluation index considers both nominal performance and disturbance resistance, and gives higher weight to disturbance resistance. This ensures that the final parameter combination selected is not only theoretically optimal but also has sufficient robustness to model prediction bias, avoiding misjudging parameter combinations with large prediction variance due to data sparsity as the optimal solution. This improves the reliability and engineering feasibility of the optimization results in actual production.

[0150] The expression for the robustness comprehensive evaluation index is:

[0151]

[0152] In the formula, For the first A comprehensive evaluation index for candidate parameters. For the first Relative deviation of film thickness uniformity under nominal operating conditions and parameters. For the first Relative deviation of film thickness uniformity under worst-case conditions with set parameters This represents the minimum relative deviation of film thickness uniformity under nominal operating conditions for all parameter groups in the candidate set. This represents the minimum relative deviation of film thickness uniformity under the worst-case operating condition for all parameter groups in the candidate set. It is a very small positive number, used to prevent the denominator from being zero. The nominal operating condition weighting coefficient satisfies... ;

[0153] In the expression, the comprehensive evaluation index Used for relative sorting and filtering within a candidate set, its physical meaning is: to measure the... The combined deviation of the candidate parameters from their respective best achievable levels under both nominal and worst-case operating conditions. It should be noted that... This does not reflect the absolute severity of film thickness deviation because the absolute severity has already been determined by the relative deviation of film thickness uniformity in the first stage. The evaluation and screening processes have been completed; the parameter sets entering the second stage of evaluation have all reached low and similar levels of absolute deviation. The core issue at this point is "which set of parameters can maintain optimal performance under the influence of model prediction uncertainty and operating condition fluctuations," rather than repeatedly evaluating the magnitude of the absolute deviation. The division of labor between the two stages is clear: the first stage is responsible for screening candidate solutions with excellent absolute uniformity, and the second stage is responsible for selecting the final solution with the strongest resistance to disturbances from among them. Therefore... When the minimum value is taken, the corresponding parameter set approaches its own optimum under both operating conditions, thus achieving the best trade-off between nominal performance and disturbance rejection capability.

[0154] Determining the value involves: using the maximization of the historical batch pass rate as the optimization objective, and employing a grid search method to optimize within the feasible interval. The process involves iterating and optimizing the data. Specifically, existing historical production data is divided into a training set and a validation set, and a comprehensive evaluation metric is performed on the training set. The calculation and selection of candidate parameter combinations are performed, and the actual batch pass rate corresponding to the selected parameter combinations is statistically analyzed on the validation set. The batch pass rate corresponding to the highest pass rate in the validation set is then selected. The value is taken as the optimal value.

[0155] The final optimal combination of process operating parameters is put into production operation. Based on real-time film thickness monitoring data, the process operating parameters of the vapor deposition process are dynamically adjusted, including:

[0156] The final optimal combination of process operating parameters is sent as the initial set value to the vacuum evaporation equipment control system for execution.

[0157] In the actual vapor deposition process, the deposition rate at the center point of the evaporation source is monitored in real time by the built-in crystal oscillator monitoring module of the vapor deposition equipment, and the PID controller is used to maintain the evaporation rate at the target evaporation rate.

[0158] Every S production batches, the film thickness at each sampling point in the effective coating area of ​​the substrate is measured using a metallographic microscope and coulometric instrument.

[0159] Input the current actual process operating parameters into the combined kernel Gaussian process regression model, and output the predicted value of the evaporation rate and the predicted value of the process correlation angular distribution correction index under the current operating conditions;

[0160] Substitute the predicted mean of the evaporation rate and the predicted value of the process correlation angle distribution correction index into the corrected forward calculation expression to calculate the theoretical value of the total periodic deposition thickness at each sampling point under the current process operating parameters.

[0161] The deviation between the measured film thickness at each sampling point and the theoretical total thickness of the periodic deposition is calculated point by point. The deviation distribution vector is obtained, and the root mean square error of the deviation distribution vector is calculated.

[0162] If the root mean square error of the deviation distribution vector is less than the dead zone threshold, the system is considered to be operating normally, the current set value is maintained, and normal production continues.

[0163] If the root mean square error of the deviation distribution vector is greater than or equal to the dead zone threshold, it is determined that the system has experienced inter-batch operating condition drift. Further analysis of the spatial distribution pattern of the deviation distribution vector is then conducted to determine the type of drift.

[0164] This scheme employs a dual-timescale control strategy, decoupling the maintenance of instantaneous stability of the evaporation rate from in-plane uniformity drift compensation. Specifically, it combines PID control with batch-level offline sampling. Fast timescale control ensures a constant evaporation rate for each batch, guaranteeing film thickness accuracy for each batch. Slow timescale control uses film thickness distribution data from offline sampling as feedback, calculating the deviation vector between the actual and theoretical deposition thickness point-by-point. This preserves spatial distribution information to diagnose drift types, employing differentiated compensation strategies for different types of drift. Furthermore, local neighborhood re-optimization is performed to rapidly restore uniformity while maintaining a constant evaporation rate, balancing control accuracy and engineering feasibility, and avoiding production disturbances caused by frequent adjustments.

[0165] The dead zone threshold is determined based on statistical analysis: Under stable equipment operation, at least 30 batches of normal production film thickness measurement data are continuously collected. The root mean square error (RMSE) of the deviation between the measured film thickness and the theoretical value is calculated for each batch, forming a statistical sample set of deviations under normal operating conditions. The mean of this set plus three times the standard deviation is used as the dead zone threshold. In subsequent production, an exponentially weighted moving average strategy is used to dynamically update the mean and standard deviation. For scenarios lacking historical data, a hierarchical analysis method is used, where process experts compare the severity of two types of misjudgments: "adjustment mistakenly triggered due to normal fluctuations" and "adjustment missed due to actual drift." The initial value of the confidence coefficient is determined through a judgment matrix, ensuring the system can function normally before initial operation.

[0166] Further analysis of the spatial distribution pattern of the deviation distribution vector to determine the drift type follows these rules:

[0167] If the deviation distribution vector shows an overall unidirectional shift, that is, the deviation directions of all sampling points are consistent and the magnitudes are similar, it is preferentially determined to be an overall drift of the evaporation rate between batches. The process operating parameters and the corresponding actual evaporation rate values ​​in the measured data of the current batch are added to the training sample set of the combined kernel Gaussian process regression model to incrementally update and optimize the model.

[0168] If the deviation distribution vector exhibits an asymmetric distribution, i.e. the deviation amplitude changes monotonically along the radial or circumferential direction of the substrate, it is determined that the thermal radiation boundary conditions in the cavity or the local airflow field distribution near the substrate have undergone asymmetric changes. In the fitness function of subsequent optimization, a spatial distribution deviation penalty term constructed from the mean square value of the deviation at each sampling point is introduced as a new constraint condition.

[0169] If the deviation distribution vector exhibits an axisymmetric distribution, that is, the deviation amplitude is rotationally symmetric with respect to the substrate's revolution axis, it is determined that the geometric position of the evaporation source or the substrate tilt angle has drifted. The measured film thickness data of the current batch is added to the correction dataset, and the geometric configuration parameter correction is re-executed.

[0170] After updating the parameters according to the drift type, the adaptive inertial weighted particle swarm optimization algorithm is re-executed within a reliable neighborhood of ±5% of each parameter, starting from the current actual process operating parameters. This generates the updated optimal combination of process operating parameters, which is then sent to the vacuum evaporation equipment control system for execution after operator confirmation, thereby achieving slow time-varying closed-loop uniformity control of the evaporation process.

[0171] Furthermore, based on the same inventive concept as the aforementioned method for controlling and optimizing the operating parameters of a vacuum evaporation process aimed at film thickness uniformity, this solution proposes a system for controlling and optimizing the operating parameters of a vacuum evaporation process aimed at film thickness uniformity, comprising:

[0172] The coordinate calibration unit is used to calibrate and standardize the position coordinates of each sampling point on the substrate using a 2.5D projector, providing a precise point reference for calculating the forward deposition thickness.

[0173] The geometric parameter calibration unit is used to calibrate the initial measurement values ​​of geometric configuration parameters such as the three-dimensional coordinates of the evaporation source chamber, the actual tilt angle and eccentricity of the substrate revolution axis using a coordinate measuring machine.

[0174] The film thickness distribution detection unit is used to obtain the measured value of the total periodic deposition thickness at each sampling point on the substrate surface by combining metallographic microscope and coulometric instrument, and output the in-plane film thickness distribution data.

[0175] The evaporation rate calibration unit is used to collect the real-time deposition rate at the center point of the evaporation source under the benchmark operating conditions, providing global scaling and calibration constraints for geometric parameter correction and eliminating the coupling ambiguity between geometric parameters and evaporation rate;

[0176] The real-time rate monitoring unit is used to collect the deposition rate at the center point of the evaporation source in real time based on the built-in crystal oscillator monitoring module of the evaporation equipment, and provide feedback signals to the PID controller to maintain the evaporation rate at the target evaporation rate.

[0177] In summary, the advantages of this invention are as follows: by geometric mechanism modeling, in-situ parameter correction based on measured data, and global optimization and robustness screening driven by Gaussian process regression, an end-to-end closed-loop self-optimizing control system from process parameters to film thickness uniformity is constructed, which effectively solves the difficulties in accurately controlling film thickness distribution and process parameters relying on empirical trial and error in vacuum evaporation.

[0178] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. A method for controlling and optimizing the operating parameters of a vacuum evaporation process aimed at achieving uniform film thickness, characterized in that, include: Based on Knudsen's cosine evaporation law with process-related angular distribution correction and the planetary motion law of the substrate, a forward calculation expression is established with constant evaporation rate and process-related angular distribution correction index as input and total periodic deposition thickness at each point on the substrate surface as output. Under the reference angular distribution condition, the constant evaporation rate required to achieve the target average film thickness is back-calculated from the arithmetic mean of the total periodic deposition thickness at the sampling points, which serves as the target evaporation rate for process control. Based on multiple sets of measured film thickness data, a nonlinear least squares algorithm was used to initialize and correct the geometric configuration parameters, and the self-consistency between the parameters and the target evaporation rate was verified. Multiple sets of process operating parameters and corresponding measured data of evaporation rate and film thickness distribution were collected. Using process operating parameters as input and evaporation rate and process correlation angular distribution correction index as output, a combined kernel Gaussian process regression model was constructed to obtain the mapping relationship between process operating parameters and evaporation rate and spatial deposition distribution characteristics. Based on the predicted evaporation rate and the predicted process-related angular distribution correction index output by the combined kernel Gaussian process regression model, a fitness function is constructed with the goal of minimizing the relative deviation of film thickness uniformity. An adaptive inertial weighted particle swarm algorithm is then used for global optimization, and a set of candidate optimal process operating parameter combinations is output. For each set of parameters in the candidate optimal process operation parameter combination set, the final optimal process operation parameter combination is determined based on the criteria of having the best nominal performance and the strongest anti-disturbance capability. The optimal combination of process operating parameters is put into production operation, and the process operating parameters of the vapor deposition process are dynamically adjusted based on real-time film thickness monitoring data.

2. The method for controlling and optimizing the operating parameters of a vacuum evaporation process for film thickness uniformity according to claim 1, characterized in that, The forward calculation expression, which takes a constant evaporation rate and a process-related angular distribution correction index as input and the total periodic deposition thickness at each point on the substrate surface as output, includes: K sampling points are predefined in the substrate's own coordinate system. The position of each sampling point relative to the geometric center of the substrate remains fixed, and the evaporation source coordinates and source point coordinates of the vacuum evaporation equipment chamber are obtained. Based on the revolution angle and rotation angle of the substrate, as well as the ratio of the revolution and rotation angular velocities, the position coordinates of each sampling point are mapped to the real-time position coordinates under the fixed coordinate system of the chamber. Based on the geometric parameters of the vacuum evaporation chamber and the type of evaporation source, and based on Knudsen's cosine evaporation law, a process-related angular distribution correction index is introduced to correct the actual spatial angular distribution of the evaporation flux. Combined with the inherent deposition property parameters of the material, an expression for the instantaneous deposition rate of the substrate sampling point is constructed to obtain the instantaneous deposition rate. Based on the planetary motion law of the substrate, the instantaneous deposition rate within one revolution cycle is integrated over time to obtain the total periodic deposition thickness of each sampling point on the substrate surface within one complete revolution cycle. Based on the total periodic deposition thickness at each sampling point on the substrate, the arithmetic mean of the total periodic deposition thickness at all sampling points is calculated as the average film thickness under this condition. Based on the baseline angular distribution correction index corresponding to the baseline process parameters, and taking the target average film thickness as the average film thickness under this operating condition, the required constant evaporation rate is calculated as the target evaporation rate for process control. The forward calculation expression, taking the constant evaporation rate and the process-related angular distribution correction index as input and the total periodic deposition thickness at each point on the substrate surface as output, is as follows: In the formula, Let the total periodic deposition thickness be the thickness at the j-th sampling point. For a constant mass evaporation rate, These are inherent deposition property parameters of the material. Let T be the solid density of the deposited thin film, and T be the orbital period. for The angle between the surface normal of the evaporation source at time t and the line connecting the source point to the j-th sampling point. for The angle between the substrate surface normal and the connecting line at the j-th sampling point at time j. for The straight-line distance from the evaporation source to the j-th sampling point at time step [j]. This is the correction index for the process-related angular distribution.

3. The method for controlling and optimizing the operating parameters of a vacuum evaporation process for film thickness uniformity according to claim 2, characterized in that, The process of initializing and correcting the geometric configuration parameters based on multiple sets of measured film thickness data using a nonlinear least squares algorithm, and verifying their consistency with the target evaporation rate, includes: Based on the forward calculation expression, the set of geometric configuration parameters to be corrected is determined. The geometric configuration parameters to be corrected include at least: the three-dimensional spatial coordinates of the evaporation source in the chamber coordinate system, and the actual tilt angle and eccentricity of the substrate revolution axis. Under the conditions of reference angular distribution and constant evaporation rate, a composite forward calculation function for the total periodic deposition thickness at each sampling point on the substrate surface is constructed based on the forward calculation expression, using geometric configuration parameters and revolution period parameters as inputs. Multiple sets of coating test data under the benchmark process were collected. Each set of tests included the measured film thickness at each sampling point on the substrate surface, as well as the preliminary measured values ​​of the geometric configuration parameters under the current installation state. The calibration process uses the center deposition rate measured by the built-in crystal oscillator monitoring module at a fixed position in the chamber under the reference operating conditions as the global calibration constraint, fixes the scaling reference of the evaporation rate, and eliminates the coupling ambiguity between geometric parameters and evaporation rate. The initial measured values ​​of the geometric configuration parameters of each group of experiments are input into the composite forward calculation function, and the theoretical value of the total periodic deposition thickness of each sampling point is output. The sum of squares of the differences between the theoretical value and the measured value of the total periodic deposition thickness of each point is used as the objective function. With the objective function as the goal, the Levenberg-Marquardt nonlinear least squares algorithm is used to iteratively solve the geometric configuration parameters to be corrected. The iteration is terminated when the decrease in the objective function value is less than a preset threshold or the number of iterations reaches a preset upper limit, and the optimal estimated value of the corrected geometric configuration parameters is output. Substitute the corrected geometric configuration parameters into the composite forward calculation function to recalculate the arithmetic mean of the total periodic deposition thickness at all sampling points; Determine whether the overall fitting error between the calculated and measured values ​​of the total periodic deposition thickness at each sampling point after correction is within the preset allowable range. If yes, it indicates that the corrected geometric configuration parameters are consistent with the calibrated evaporation rate, the verification is passed, the corrected geometric configuration parameters are output and the process proceeds to the next step; otherwise, optimize the initial values ​​and re-execute the geometric parameter correction until the fitting accuracy meets the requirements.

4. The method for controlling and optimizing the operating parameters of a vacuum evaporation process for film thickness uniformity according to claim 3, characterized in that, The construction of the combined kernel Gaussian process regression model yields the mapping relationship between process operating parameters and evaporation rate and spatial deposition distribution characteristics, including: Based on the vacuum evaporation process flow, a set of process operating parameters that affect the evaporation rate and deposition distribution are determined. The process operating parameters include at least heating power, substrate temperature and chamber pressure. The Latin hypercube sampling method was used to generate multiple sets of process operating parameters and conduct vacuum evaporation tests. The actual evaporation rate values ​​corresponding to each set of tests were collected in real time using the built-in crystal oscillator monitoring module of the evaporation equipment. The substrate film thickness data were collected by metallographic microscope and coulometer. The process correlation angle distribution correction index under the corresponding working conditions was obtained by nonlinear fitting. A hybrid kernel function consisting of a weighted combination of radial basis kernels and linear kernels was constructed. A multi-task Gaussian process regression framework with shared kernels was adopted. The heating power, substrate temperature and chamber pressure vector were used as inputs, and the corresponding actual evaporation rate value and angular distribution correction exponent were used as outputs. The Gaussian process regression model was trained to obtain a mapping model from process operating parameters to evaporation rate and deposition distribution. After training, for any new combination of process operating parameters, the Gaussian process regression model outputs the predicted mean and variance of the evaporation rate, as well as the predicted mean and variance of the process correlation angular distribution correction index, and uses the predicted mean as the predicted value of the evaporation characteristics and deposition distribution under that parameter combination.

5. The method for controlling and optimizing the operating parameters of a vacuum evaporation process for film thickness uniformity according to claim 4, characterized in that, The process of constructing a fitness function and using an adaptive inertial weighted particle swarm optimization algorithm for global optimization, outputting a set of candidate optimal process operating parameter combinations, includes: Determine the range constraints of the operating parameters to be optimized and the process feasibility constraints, and construct a set of constraints. The operating parameters to be optimized include at least heating power, substrate temperature and chamber pressure. For any combination of candidate process operating parameters, the predicted evaporation rate and the predicted angular distribution correction index output by the combined kernel Gaussian process regression model are substituted into the forward formula to calculate the average film thickness. It is then determined whether the deviation between the average film thickness and the target average film thickness is within the allowable range. If it exceeds the allowable range, the candidate parameter combination is marked as an infeasible solution and discarded. If it is within the allowable range, the candidate parameter combination is marked as a feasible solution and proceeds to the next evaluation step. Substitute the predicted evaporation rate and the predicted process-related angular distribution correction index of the selected values ​​into the corrected forward calculation expression to calculate the theoretical value of the total periodic deposition thickness at each sampling point on the substrate surface. Based on the theoretical value of the total periodic deposition thickness at each sampling point on the substrate surface, the relative deviation of film thickness uniformity under this parameter combination is calculated, and the relative deviation value of uniformity is used as the fitness function value. Initialize the particle swarm population, dynamically adjust the inertial weight of each generation of particles using an adaptive inertial weight strategy, and iterate to find the optimal value until the convergence condition is met. During the iterative optimization process, the position vector of each particle is a combination of process operation parameters, and the velocity vector is the change of each parameter. The particle continuously updates its own position by tracking the individual historical best position and the global historical best position until it reaches the preset maximum number of iterations or the fitness value converges to the preset threshold. After the algorithm converges, the top N combinations of process operating parameters with the best fitness are output as the set of candidate optimal combinations of process operating parameters. The fitness value of each group and the joint prediction variance of the combined kernel Gaussian process regression model under the parameter combination are recorded.

6. The method for controlling and optimizing the operating parameters of a vacuum evaporation process for film thickness uniformity according to claim 5, characterized in that, The process of selecting and determining the final optimal combination of process operating parameters for each set of candidate optimal process operating parameters, based on the criteria of having the best nominal performance and the strongest anti-disturbance capability, includes: Extract the r-th group of process operating parameters and their corresponding combined kernel Gaussian process regression model joint prediction variance from the candidate optimal process operating parameter combination set; The evaporation rate corresponding to the 95% confidence upper limit of the mean predicted by the combined kernel Gaussian process regression model, and the values ​​corresponding to the larger relative deviation of film thickness uniformity in the upper and lower limits of the 95% confidence interval of the angular distribution correction index, are used together as the worst-case evaporation characteristic estimate at this parameter point. Substitute the worst-case evaporation characteristic estimate into the corrected forward calculation expression to calculate the total periodic deposition thickness at each sampling point under the worst-case condition and the relative deviation of film thickness uniformity under the worst-case condition. Substitute the predicted mean of the combined nuclear Gaussian process regression model into the corrected forward calculation expression to calculate the total periodic deposition thickness at each sampling point under nominal conditions and the relative deviation of film thickness uniformity under nominal conditions. Based on the relative deviation of film thickness uniformity under worst-case conditions and the relative deviation of film thickness uniformity under nominal conditions, a robustness comprehensive evaluation index is constructed. Calculate the comprehensive evaluation index corresponding to each group of candidate parameters, and select the parameter combination corresponding to the minimum value of the comprehensive evaluation index as the final optimal process operation parameter combination.

7. The method for controlling and optimizing the operating parameters of a vacuum evaporation process for film thickness uniformity according to claim 6, characterized in that, The step of putting the final optimal combination of process operating parameters into production operation, and dynamically adjusting the process operating parameters of the vapor deposition process based on real-time film thickness monitoring data, includes: The final optimal combination of process operating parameters is sent as the initial set value to the vacuum evaporation equipment control system for execution. In the actual vapor deposition process, the deposition rate at the center point of the evaporation source is monitored in real time by the built-in crystal oscillator monitoring module of the vapor deposition equipment, and the PID controller is used to maintain the evaporation rate at the target evaporation rate. Every S production batches, the film thickness at each sampling point in the effective coating area of ​​the substrate is measured using a metallographic microscope and coulometric instrument. Input the current actual process operating parameters into the combined kernel Gaussian process regression model, and output the predicted value of the evaporation rate and the predicted value of the process correlation angular distribution correction index under the current operating conditions; Substitute the predicted mean of the evaporation rate and the predicted value of the process correlation angle distribution correction index into the corrected forward calculation expression to calculate the theoretical value of the total periodic deposition thickness at each sampling point under the current process operating parameters. The deviation between the measured film thickness at each sampling point and the theoretical total thickness of the periodic deposition is calculated point by point. The deviation distribution vector is obtained, and the root mean square error of the deviation distribution vector is calculated. If the root mean square error of the deviation distribution vector is less than the dead zone threshold, the system is considered to be operating normally, the current set value is maintained, and normal production continues. If the root mean square error of the deviation distribution vector is greater than or equal to the dead zone threshold, it is determined that the system has experienced inter-batch operating condition drift. Further analysis of the spatial distribution pattern of the deviation distribution vector is then conducted to determine the type of drift.

8. A vacuum evaporation process operating parameter control and optimization system for film thickness uniformity, characterized in that, The method for controlling and optimizing operating parameters of a vacuum evaporation process to achieve film thickness uniformity as described in any one of claims 1-7 includes the following basic functional units: The coordinate calibration unit is used to calibrate and standardize the position coordinates of each sampling point on the substrate using a 2.5D projector, providing a precise point reference for calculating the forward deposition thickness. The geometric parameter calibration unit is used to calibrate the initial measurement values ​​of geometric configuration parameters such as the three-dimensional coordinates of the evaporation source chamber, the actual tilt angle and eccentricity of the substrate revolution axis using a coordinate measuring machine. The film thickness distribution detection unit is used to obtain the measured value of the total periodic deposition thickness at each sampling point on the substrate surface by combining metallographic microscope and coulometric instrument, and output the in-plane film thickness distribution data. The evaporation rate calibration unit is used to collect the real-time deposition rate at the center point of the evaporation source under the benchmark operating conditions, providing global scaling and calibration constraints for geometric parameter correction and eliminating the coupling ambiguity between geometric parameters and evaporation rate; The real-time rate monitoring unit is used to collect the deposition rate at the center point of the evaporation source in real time based on the built-in crystal oscillator monitoring module of the evaporation equipment, and provide feedback signals to the PID controller to maintain the evaporation rate at the target evaporation rate.

Images