Precise control method and system for mixing ratio of precursor gas of vapor deposition furnace

Abstract

The application provides a kind of gas deposition furnace precursor gas mixing proportion accurate control method and system, it is related to gas deposition technical field, method includes following steps: S1: precursor gas pretreatment and purity detection;S2: establish precursor gas physical property parameter library;S3: real-time acquisition gas deposition furnace operating condition data;S4: initial calculation based on the mixed proportion of multi-algorithm fusion;S5: precursor gas dynamic mixing execution;S6: mixed proportion feedback correction based on PID-Kalman filtering;S7: mixed process abnormal monitoring;S8: exception handling and recovery;S9: deposition process closed loop optimization.The application realizes the accurate stable control of mixed proportion by dynamically correcting gas physical property parameters, multi-target collaborative optimization mixed proportion, noise suppression type feedback correction and hierarchical exception handling, guarantees the consistency and reliability of film deposition quality.

Description

Technical Field

[0001] This invention relates to the field of vapor deposition technology, and in particular to a method and system for precise control of the mixing ratio of precursor gases in a vapor deposition furnace. Background Technology

[0002] Vapor deposition is a key technology for preparing high-performance thin film materials. The mixing ratio of precursor gases directly determines the composition, structure, and properties of the film; therefore, precise control of the mixing ratio is crucial for ensuring deposition quality. Existing methods for controlling precursor gas mixing have several shortcomings:

[0003] First, gas physical parameters often use fixed values ​​under standard conditions without dynamic correction based on the actual operating conditions (temperature, pressure) of the deposition furnace. This leads to a mismatch between physical parameters and actual mass transfer reaction characteristics, affecting the accuracy of mixing ratio calculations. Second, mixing ratio optimization often uses a single algorithm, which is prone to getting trapped in local optima and makes it difficult to balance multiple objectives such as deposition quality, gas utilization, and deposition rate stability. Third, the feedback correction stage often relies solely on PID control, failing to effectively suppress detection noise interference, resulting in correction lag or overshoot, affecting mixing stability. Fourth, anomaly monitoring is limited to a single dimension, and anomaly handling lacks targeted hierarchical strategies, making it difficult to quickly respond to and resolve multiple types of anomalies such as purity, temperature, and pressure, easily leading to deposition process interruptions or quality defects. These problems restrict the application of vapor deposition technology in the field of high-precision thin film preparation. Summary of the Invention

[0004] This invention provides a method and system for precise control of the mixing ratio of precursor gases in a vapor deposition furnace. It overcomes the shortcomings of existing methods for controlling the mixing ratio of precursor gases in vapor deposition furnaces, such as low precision, poor stability, insufficient adaptability, and delayed anomaly handling. By dynamically correcting gas physical property parameters, multi-objective collaborative optimization of the mixing ratio, noise suppression feedback correction, and graded anomaly handling, it achieves precise and stable control of the mixing ratio, ensuring the consistency and reliability of thin film deposition quality.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] A method for precisely controlling the mixing ratio of precursor gases in a vapor deposition furnace includes the following steps:

[0007] S1: Pre-treat the precursor gases and test their purity to obtain the purity test values ​​for each precursor gas;

[0008] S2: Collect the basic physical property parameters of the precursor gas that has passed the purity test, and establish a precursor gas physical property parameter library after working condition adaptation correction.

[0009] S3: Real-time acquisition of operating data of the vapor deposition furnace;

[0010] S4: Based on purity detection values, precursor gas property parameter database data, and operating condition data, a collaborative optimization calculation is carried out using an adaptive mutant krill swarm optimization algorithm, a multi-objective gray wolf optimization algorithm, and a quantum-behavioral cuckoo search algorithm. First, the adaptive mutant krill swarm optimization algorithm generates the optimal population, which is then used as the initial population for the multi-objective gray wolf optimization algorithm. The multi-objective gray wolf optimization algorithm generates a Pareto optimal solution set based on the initial population, which is then used as the initial nest population for the quantum-behavioral cuckoo search algorithm. During the collaborative process, the three algorithms achieve bidirectional interactive adjustment through parameter feedback. Finally, the quantum-behavioral cuckoo search algorithm outputs the optimal target flow rate for each precursor gas.

[0011] S5: Control the precursor gas to enter the mixing system for dynamic mixing according to the optimal target flow rate, and complete the mixing ratio regulation of the precursor gas.

[0012] In this specification, S4 also includes calculating the mixing ratio of each precursor gas based on the optimal target flow rate and the total target flow rate of the mixed gas, which is used to derive the target concentration of each precursor gas after mixing; S5 also includes collecting the actual concentration of each precursor gas after mixing.

[0013] Methods for precisely controlling the mixing ratio of precursor gases in a vapor deposition furnace also include:

[0014] S6: Based on the deviation between the actual concentration and the target concentration, the flow correction amount is obtained through feedback correction, the optimal target flow rate is adjusted, and the secondary control of the precursor gas mixing ratio is completed.

[0015] S7: Combine purity test values, operating condition data and actual concentration to monitor abnormalities in the mixing process and determine the type of abnormality;

[0016] S8: Execute the corresponding exception handling measures according to the exception type to restore the stability of the hybrid process;

[0017] S9: Based on the deposition rate and flow correction in the operating data, dynamically optimize the algorithm parameters of S4 and the feedback correction parameters of S6, and achieve closed-loop optimization and control of the precursor gas mixing ratio through iterative updates.

[0018] In this specification, the bidirectional interactive adjustment of parameters between the multi-objective gray wolf optimization algorithm and the adaptive mutant krill swarm optimization algorithm is as follows: the multi-objective gray wolf optimization algorithm calculates the fitness variance of the optimal layer population during the iteration process, and feeds this fitness variance back to the adaptive mutant krill swarm optimization algorithm to adjust the krill's movement step size in the adaptive mutant krill swarm optimization algorithm.

[0019] In this specification, the bidirectional interactive adjustment of parameters between the quantum behavior cuckoo search algorithm and the adaptive mutant krill swarm optimization algorithm is as follows: the quantum behavior cuckoo search algorithm calculates the search accuracy during the iteration process and feeds this search accuracy back to the adaptive mutant krill swarm optimization algorithm to adjust the mutation probability of the adaptive mutant krill swarm optimization algorithm.

[0020] In this specification, the bidirectional interactive adjustment of parameters between the quantum behavior cuckoo search algorithm and the multi-objective gray wolf optimization algorithm is as follows: the local optimal solution obtained by the quantum behavior cuckoo search algorithm during the iteration process is fed back to the multi-objective gray wolf optimization algorithm to correct the search direction of the gray wolf in the multi-objective gray wolf optimization algorithm.

[0021] In this manual, the adaptive mutant krill swarm optimization algorithm, the multi-objective gray wolf optimization algorithm, and the quantum behavior cuckoo search algorithm all have a preset maximum number of iterations. When the number of iterations of each algorithm reaches its preset maximum number of iterations, the iteration of the corresponding algorithm is stopped and the current optimal result is output.

[0022] In this specification, the correction of the basic physical property parameters in S2 is based on the actual operating temperature and pressure of the vapor deposition furnace, and the van der Waals equation is used to dynamically adjust the basic physical property parameters.

[0023] In this manual, the feedback correction in S6 uses a combination of PID controller and Kalman filter. First, the deviation between the actual concentration and the target concentration is smoothed by Kalman filter, and then the smoothed deviation is input into the PID controller to calculate the flow correction amount.

[0024] In this manual, the abnormal monitoring of the mixing process in S7 is based on multi-dimensional indicators, including purity, temperature, pressure, deposition rate, flow rate and concentration. Each indicator corresponds to a preset abnormal judgment standard. When any indicator meets the corresponding abnormal judgment standard, it is immediately judged as an abnormality in the mixing process and the abnormality type and related data are recorded.

[0025] A precise control system for the mixing ratio of precursor gases in a vapor deposition furnace, employing any one of the methods described above, wherein the precise control system for the mixing ratio of precursor gases in a vapor deposition furnace comprises:

[0026] The gas pretreatment and purity detection module is used to pretreatment the precursor gas and detect its purity to obtain the purity detection value of each precursor gas.

[0027] The physical property parameter library construction module is used to collect the basic physical property parameters of the precursor gas that has passed the purity test, and establish the precursor gas physical property parameter library after working condition adaptation correction.

[0028] The operating condition data acquisition module is used to acquire operating condition data of the vapor deposition furnace in real time.

[0029] The multi-algorithm fusion computing module is used to conduct collaborative optimization calculations based on purity detection values, precursor gas property parameter database data, and operating condition data, employing an adaptive mutant krill swarm optimization algorithm, a multi-objective gray wolf optimization algorithm, and a quantum-behavioral cuckoo search algorithm. First, the adaptive mutant krill swarm optimization algorithm generates an optimal population, which serves as the initial population for the multi-objective gray wolf optimization algorithm. The multi-objective gray wolf optimization algorithm then generates a Pareto optimal solution set based on the initial population, which is used as the initial nest population for the quantum-behavioral cuckoo search algorithm. During the collaborative process, the three algorithms achieve bidirectional interactive adjustment through parameter feedback. Finally, the quantum-behavioral cuckoo search algorithm outputs the optimal target flow rate for each precursor gas. Based on the optimal target flow rate, the mixing ratio of each precursor gas is calculated according to the total target flow rate of the mixed gas, which is used to derive the target concentration of each precursor gas after mixing.

[0030] The gas mixing execution module is used to control the precursor gas to enter the mixing system for dynamic mixing according to the optimal target flow rate, complete the mixing ratio regulation of the precursor gas, and collect the actual concentration of each precursor gas after mixing.

[0031] The feedback correction module is used to obtain the flow correction amount based on the deviation between the actual concentration and the target concentration, adjust the optimal target flow rate, and complete the secondary control of the precursor gas mixing ratio.

[0032] The anomaly monitoring module is used to monitor anomalies in the mixing process by combining purity test values, operating condition data, and actual concentration, and to determine the type of anomaly.

[0033] The exception handling module is used to execute corresponding exception handling measures according to the exception type to restore the stability of the hybrid process;

[0034] The closed-loop optimization module is used to dynamically optimize the algorithm parameters of the multi-algorithm fusion calculation module and the feedback correction parameters of the feedback correction module based on the deposition rate and flow correction amount in the operating data, and realizes the closed-loop optimization and control of the precursor gas mixing ratio through iterative updates.

[0035] In summary, the present invention has at least the following beneficial effects:

[0036] 1. Improve the accuracy of mixing ratio control: By correcting gas property parameters according to working conditions and combining multi-objective optimization with multi-algorithm fusion, ensure that the mixing ratio calculation fits the actual deposition requirements, avoid local optima, and achieve accurate matching of the mixing ratio;

[0037] 2. Enhance the stability of the mixing process: The PID-Kalman filter composite feedback correction strategy is adopted to effectively suppress detection noise interference, improve the timeliness and stability of flow correction, and ensure the uniformity and stability of the mixed gas components;

[0038] 3. Improve system reliability: Construct a comprehensive anomaly monitoring system and a hierarchical anomaly handling mechanism to quickly identify and address anomalies of various types, such as purity, temperature, and pressure, thereby reducing the risk of interruption during the deposition process and ensuring continuous and stable operation.

[0039] 4. Improve adaptability to different working conditions: The closed-loop optimization mechanism dynamically adjusts the optimization model parameters and feedback correction parameters to adapt to parameter changes under different deposition conditions, broadens the applicability of the method, and ensures the consistency of deposition quality under different working conditions. Attached Figure Description

[0040] Figure 1 This is a schematic flowchart of the method for precisely controlling the mixing ratio of precursor gases in a vapor deposition furnace involved in this invention.

[0041] Figure 2 This is a schematic diagram of the process for precursor gas pretreatment and data preparation involved in this invention.

[0042] Figure 3 This is a schematic diagram of the multi-algorithm collaborative optimization calculation involved in this invention.

[0043] Figure 4 This is a schematic diagram of the hybrid execution and feedback correction process involved in this invention. Detailed Implementation

[0044] like Figure 1 As shown, this embodiment provides a method for precisely controlling the mixing ratio of precursor gases in a vapor deposition furnace, including:

[0045] S1: Pretreatment and Purity Detection of Precursor Gases

[0046] Each precursor gas required for the chemical vapor deposition furnace undergoes a systematic pretreatment process. The pretreatment process focuses on deep purification, employing a two-stage filtration system to efficiently remove impurities. The first stage uses a ceramic membrane filter element with a filtration precision of 0.1 μm, primarily trapping solid particles and metal oxide impurities larger than 0.1 μm in diameter. The second stage uses a molecular sieve adsorption device with 3A molecular sieves as the adsorbent, specifically removing moisture and trace polar impurities to ensure the gas dew point remains below -80°C. After pretreatment, the purity of each precursor gas is monitored in real-time using an online gas chromatograph equipped with a flame ionization detector (FID), achieving a resolution of 1 ppm and a detection frequency of once per second, continuously recording the purity values ​​for each precursor gas. ,in , This represents the total number of types of precursor gases. The value range is 0 to 100%.

[0047] Purity determination follows strict entry standards; if the purity test value of a certain type of precursor gas is... If the gas is not properly stored, the automatic switching valve will be activated to transfer it to a backup storage tank for temporary storage. Simultaneously, the batch number, detection time, and purity data of the gas will be recorded for subsequent traceability analysis. If the purity values ​​of all precursor gases meet the requirements... Then the purity test values ​​of each gas will be... The data is transmitted in real time via industrial Ethernet to step S4 as the basis for calculating the mixing ratio. It is also transmitted synchronously to step S7 for monitoring purity anomalies during the mixing process, ensuring the stability of raw material quality in subsequent mixing processes.

[0048] The process of precursor gas pretreatment and data preparation is as follows: Figure 2 As shown.

[0049] S2: Establish a database of precursor gas physical property parameters

[0050] A comprehensive collection of fundamental physical properties of all precursor gases that passed the purity test in step S1 was conducted, including a parameter system covering gas viscosity. diffusion coefficient molar mass saturated vapor pressure ,in The unit is Pa·s. Units are , The unit is g / mol. The unit is Pa. Parameter collection adopts a dual mechanism of standard data + operating condition correction. The standard data comes from authoritative chemical property handbooks, including the Chemical Property Handbook (Gas Volume) and the CRC Chemical and Physical Handbook, etc. Measured data under the standard condition of 25℃ and 101325Pa are selected first to ensure the accuracy of the basic data.

[0051] Considering the significant differences between the actual working environment of a vapor deposition furnace and its standard conditions, it is necessary to perform condition-adaptive corrections to the basic parameters using the van der Waals equation for the actual operating temperature range of 500℃~1500℃ and the pressure range of 1Pa~100Pa. The correction process fully considers the intermolecular attraction and volume effects of gas molecules to ensure a high degree of matching between the parameters and actual operating conditions. The viscosity correction formula is as follows: ;in This refers to the actual operating temperature of the vapor deposition furnace. For actual working pressure, this formula corrects for the effects of increased molecular motion due to increased temperature and increased intermolecular distance due to decreased pressure on viscosity through temperature exponent and pressure exponent terms, respectively.

[0052] The diffusion coefficient correction formula is as follows: The 1.75th power of temperature in this formula reflects the significant influence of temperature on the molecular diffusion kinetic energy, while the inverse proportional term of pressure reflects the widening effect of pressure reduction on the molecular diffusion path. The corrected diffusion coefficient is more in line with the gas mass transfer characteristics under low-pressure and high-temperature conditions.

[0053] The corrected formula for saturated vapor pressure is as follows: ;in This is the molar heat of vaporization of the gas, expressed in J / mol, which can be obtained by consulting a thermochemistry handbook. The universal gas constant is 8.314 J / (mol·K). This formula is derived based on the Clausius-Clapeyron equation and accurately describes the exponential effect of temperature on saturated vapor pressure.

[0054] The corrected physical properties of each precursor gas , , , A dynamic physical property parameter library is established, which has a real-time update function. When the real-time collected furnace temperature or pressure changes, the parameter library will automatically trigger a recalculation to ensure that the parameters match the operating conditions in real time.

[0055] S3: Real-time acquisition of vapor deposition furnace operating data

[0056] Throughout the operation of the vapor deposition furnace, a multi-dimensional data acquisition system was constructed. A high-precision sensor array was used to capture key parameters within the furnace in real time. The selection and installation of the sensor array adhered to industrial-grade measurement standards, ensuring the accuracy, stability, and real-time performance of the data acquisition. The temperature sensor, a platinum resistance PT1000, was installed at three evenly distributed points in the center of the furnace cavity and near the deposition substrate. Multi-point acquisition and averaging reduced measurement errors, achieving an accuracy of ±1℃ and a acquisition frequency of twice per second to obtain the actual operating temperature inside the furnace in real time. The pressure sensor is a capacitive vacuum pressure sensor, installed at the furnace exhaust port. Its measurement range covers 1 Pa to 100 Pa, with an accuracy of ±0.1 Pa and a sampling frequency of twice per second, continuously monitoring the actual working pressure inside the furnace. .

[0057] The deposition rate sensor employs a laser interferometer, mounted 5 cm directly above the deposition substrate surface. It calculates the deposition rate by monitoring the change in film thickness over time, achieving an accuracy of ±0.01 nm / s and a sampling frequency of once per second. The gas flow sensor uses a Coriolis mass flow meter, installed at the end of the supply pipeline for each precursor gas near the inlet of the mixing chamber. This flow meter is unaffected by changes in gas temperature and pressure, has a measurement accuracy of ±0.01 L / min, and a sampling frequency of twice per second to capture the real-time initial flow rate of each gas. .

[0058] All data collected by the sensors are converted into standard 4-20mA current signals by isolated signal converters, and then preprocessed by the industrial controller, including outlier removal and smoothing filtering.

[0059] S4: Initial calculation of the mixing ratio based on multi-algorithm fusion

[0060] With the core optimization objectives of achieving thin film deposition quality standards, maximizing gas utilization, and stabilizing deposition rate, combined with purity test values... ( , (Total number of precursor gas types) and physical property parameter database data , , , and operating condition data , , A three-algorithm fusion optimization model was constructed, combining an adaptive mutant krill swarm optimization algorithm, a multi-objective gray wolf optimization algorithm, and a quantum-behavioral cuckoo search algorithm, to achieve accurate initial calculation of the mixing ratio. The three algorithms employ a fusion architecture of bidirectional interaction, layer-by-layer optimization, and feedback iteration. The multi-algorithm collaborative optimization calculation process is as follows: Figure 3 As shown, the specific process is as follows:

[0061] 4.1 Adaptive Mutant Krill Population Optimization Algorithm (AM-KHA): Population Initialization Optimization; The core function of the Adaptive Mutant Krill Population Optimization Algorithm is to generate an initial population with high diversity and quality, avoiding subsequent algorithms from getting trapped in local optima. Simultaneously, it dynamically adjusts the population's evolutionary direction through an adaptive mutation strategy. Its model construction, training, and application process are as follows:

[0062] 4.1.1 Construction of Adaptive Mutant Krill Population Optimization Algorithm Model

[0063] 1. Fitness Function Construction and Alignment with Optimization Objectives: Based on the multi-objective optimization requirements, a single-objective fitness function for the adaptive mutant krill population optimization algorithm is constructed. The three optimization objectives are integrated through weights, as shown in the following formula: ;in The fitness function value is the adaptive mutant krill population optimization algorithm. The smaller the value, the better the corresponding flow solution. In the adaptive mutant krill swarm optimization algorithm, the first... The second iteration The position vector of a single krill. , The krill population size for the adaptive mutant krill population optimization algorithm; This represents the current iteration number; , , These are weighting coefficients with a sum of 1, and can be adjusted according to process requirements; The target deposition rate; The total target flow rate of the mixed gas is set according to the furnace cavity volume and deposition rate requirements. For the first The second iteration Each krill corresponds to the first Solution for the flow rate of a certain gas; For the first The standard deviation of the gas flow rate for each krill reflects the uniformity of the flow rate distribution.

[0064] 2. Krill Location Update Model with Step Size Adjustment: Adaptive Mutant Krill Population Optimization Algorithm. Krill location updates are jointly determined by foraging behavior, social behavior, and random behavior. Simultaneously, the population fitness variance feedback from the multi-objective gray wolf optimization algorithm is introduced to adjust the step size, achieving algorithmic interaction. The formula is as follows: ;in For the first The second iteration The updated position vector of a single krill; For the first The step size for the next iteration, and the initial value. The subsequent feedback from the multi-objective gray wolf optimization algorithm is the population fitness variance. Adaptive adjustment, the interaction formula is as follows: ; The multi-objective gray wolf optimization algorithm The variance of the gray wolf population fitness in the next iteration; This represents the maximum variance of the population fitness in the multi-objective gray wolf optimization algorithm, preset to 0.5. The larger, The smaller the krill, the more gently it moves, thus avoiding population fluctuations.

[0065] Foraging behavior vector The formula for pointing to the current optimal krill position is as follows: ; Foraging coefficient; For the first The optimal krill position in the next iteration of the adaptive mutation krill swarm optimization algorithm is the krill position with the minimum fitness function value.

[0066] Social Behavior Vector The formula reflecting the group cooperation among krill is as follows: ; The social cooperation coefficient; For the first The second iteration The position vector of a single krill.

[0067] Random Behavior Vector The formula used to increase population diversity is as follows: ; For random behavior coefficients; To follow a standard normal distribution 3D random vector.

[0068] 3. Adaptive mutation model, adjusted by feedback from the quantum behavior cuckoo search algorithm: To avoid premature convergence of the population, a mutation operation is introduced, with mutation probability... Search accuracy fed back by the cuckoo search algorithm based on quantum behavior Adaptive adjustment enables two-way interaction, as shown in the following formula: ;in , These are the upper and lower limits of the mutation probability; The Cuckoo Search Algorithm for Quantum Behavior The search precision of the next iteration, ranging from 0 to 1, increases with precision. The smaller.

[0069] The mutation operation formula is as follows: ;in The coefficient of variation; , For the first The maximum and minimum allowable flow rates of a gas are determined by the hardware performance of the gas supply system. A uniformly distributed random number between 0 and 1. When the probability value of the randomly generated number is... At that time, the first time for the krill Perform mutation operations on each flow component.

[0070] 4.1.2 Training of the Adaptive Mutant Krill Population Optimization Algorithm Model; 1. Initialization Parameters: Setting the Krill Population Size Maximum number of iterations for the adaptive mutant krill swarm optimization algorithm Initial movement step size Minimum mutation probability Maximum mutation probability Foraging coefficient Social cooperation coefficient Random behavior coefficient Coefficient of variation Based on gas flow constraints Randomly generate the initial krill population The flow component of each krill is within the constraints. 2. Iterative training process: The first step is to calculate the fitness of each krill in the initial population. Selecting the initial optimal krill The second step is to determine the current iteration number. Is it less than If so, execute the position update formula and calculate. The third step is to receive feedback from the multi-objective gray wolf optimization algorithm. ,Adjustment ; The cuckoo search algorithm that receives feedback on quantum behavior ,Adjustment The fourth step is to perform an adaptive mutation operation on the updated population to generate a mutated population. The fifth step is to calculate the fitness of the population after mutation and update the optimal krill. Step 6 Increment by 1, return to step 2, and continue until the maximum number of iterations is reached. .

[0071] 4.1.3 Application of the Adaptive Mutant Krill Population Optimization Algorithm Model; After training, the optimal population of the adaptive mutant krill population optimization algorithm is output. ,in To determine the gray wolf pack size in the multi-objective gray wolf optimization algorithm, take the top... The optimal number of krill individuals is selected to match the population size requirement of the multi-objective gray wolf optimization algorithm. This population is directly used as the initial gray wolf population for the multi-objective gray wolf optimization algorithm, as shown in the following formula: Simultaneously, the mean population fitness of the adaptive mutant krill population optimization algorithm is... The feedback is fed into the multi-objective gray wolf optimization algorithm as an initial fitness reference for the gray wolf pack. After the training of the adaptive mutant krill population optimization algorithm is completed, the first... The optimal location for krill.

[0072] 4.2 Multi-Objective Gray Wolf Optimization Algorithm (MOGWO): Generation of Multi-Objective Optimal Solution Sets; The core function of the Multi-Objective Gray Wolf Optimization Algorithm is to handle multi-objective optimization problems. It generates Pareto optimal solution sets through hierarchical sorting, balancing the three major objectives of thin film deposition quality, gas utilization rate, and deposition rate stability. Simultaneously, it receives feedback information from the Adaptive Mutant Krill swarm optimization algorithm and the quantum-behavioral cuckoo search algorithm, dynamically adjusting the search strategy. Its model construction, training, and application process are as follows:

[0073] 4.2.1 Construction of Multi-Objective Gray Wolf Optimization Algorithm Model

[0074] 1. Multi-objective function construction, preserving multi-objective characteristics: The multi-objective Grey Wolf optimization algorithm adopts a multi-objective optimization framework. The objective function is a vector form of three independent sub-objectives, which do not require weight integration and are directly related to the core optimization objective. The formula is as follows: ;in The multi-objective function vector of the multi-objective gray wolf optimization algorithm; In the multi-objective gray wolf optimization algorithm, the first The second iteration The position vector of Sekiro: Shadows Die Twice. ; For the target of sedimentation rate deviation, the smaller the value, the better; The smaller the value, the higher the gas utilization rate; For flow rate fluctuation targets, the smaller the value, the more stable the deposition rate. For the first The second iteration Sekiro: Shadows Die Twice corresponds to the first Solution for the flow rate of a certain gas.

[0075] 2. Gray Wolf Position Update Model: The multi-objective gray wolf optimization algorithm guides the swarm search through three levels of gray wolves (α, β, δ), while incorporating local optimal solutions from the quantum behavior of the cuckoo search algorithm to correct the search direction, achieving bidirectional interaction. The position update formula is as follows: ;in For the first The second iteration The updated position vector of Sekiro: Shadows Die Twice; , , The first The positions of the gray wolves in the next iteration, namely α (optimal), β (second best), and δ (third best), are obtained by Pareto sorting. The basic search vector is determined by the gray wolf's encirclement behavior, as shown in the following formula: ; , The convergence factor decreases linearly with the number of iterations. This represents the maximum number of iterations for the multi-objective gray wolf optimization algorithm. This is the optimal position for the current gray wolf pack; Uniformly distributed between 0 and 1 3D random vector. The local optimum solution is fed back by the cuckoo search algorithm based on quantum behavior to adjust the feedback vector. The decision is used to correct the gray wolf's search direction and improve search efficiency. The interaction formula is as follows: ; This is a feedback coefficient, which can be adjusted according to process requirements; The Cuckoo Search Algorithm for Quantum Behavior The local optimal location of the bird's nest in the next iteration.

[0076] 3. Pareto ranking and stratification model for selecting the optimal gray wolf: A non-dominated ranking method is used to stratify the gray wolf population. The non-dominated relationship is defined as: for two gray wolves... and ,like For all sub-targets All of them are true, and there exists at least one sub-objective that satisfies Then it is called Dominate The stratification rule is as follows: individuals not controlled by any gray wolves are classified as stratum 1 (optimal stratum), denoted as... After removing individuals from layer 1, individuals without a dominance relationship among the remaining individuals are classified into layer 2, denoted as... This process continues until all gray wolves have completed their stratification, meaning the set of all individual gray wolves equals the union of the sets of individuals in each stratum.

[0077] To ensure the diversity of the solution set, the crowding degree of the gray wolves in each layer is calculated, reflecting the degree of dispersion of individuals in the target space, as shown in the following formula: ;in For the first The second iteration The crowding of only gray wolves; , The first The maximum and minimum values ​​of each sub-objective; , The first Only, the first The position vector of the gray wolf.

[0078] 4. Population fitness variance model, fed back to adaptive mutant krill population optimization algorithm: calculate the... The fitness variance of the gray wolf population in the next iteration, based on the individuals in the first layer. For example, represents the stability of the optimal group, and the formula is as follows: ;in This refers to the number of gray wolves in the first layer. The variance represents the mean deviation of the deposition rate of individuals in the first layer. This variance will be fed back into the adaptive variation krill swarm optimization algorithm to adjust the movement step size. .

[0079] 4.2.2 Training of the multi-objective gray wolf optimization algorithm model; 1. Initialization parameters: setting the gray wolf pack size Maximum number of iterations for the multi-objective gray wolf optimization algorithm Initial convergence factor Feedback coefficient ; Receive the initial gray wolf population output from the adaptive mutant krill swarm optimization algorithm and the mean fitness of the population 2. Iterative training process: The first step is to calculate the multi-objective function value of each gray wolf in the initial swarm. The first step is to perform Pareto sorting and stratification to select α, β, and δ gray wolves; the second step is to determine the current iteration number. Is it less than If so, then calculate the convergence factor. and search vector The third step is to receive the local optimal solution from the cuckoo search algorithm based on quantum behavior. Calculate the feedback adjustment vector Perform gray wolf location update and generate The fourth step is to perform Pareto sorting and stratification on the updated gray wolf population, and calculate the crowding degree of individuals in each stratum; the fifth step is to update the α, β, and δ gray wolves based on the stratification results and crowding degree, prioritizing individuals with higher strata and greater crowding degree; the sixth step is to calculate the fitness variance of individuals in the first stratum. The feedback is given to the adaptive mutant krill population optimization algorithm; the seventh step is to... Increment by 1, return to step 2, and continue until the maximum number of iterations is reached. .

[0080] 4.2.3 Application of the Multi-Objective Gray Wolf Optimization Algorithm Model; After training, output the Pareto optimal solution set of the first layer of the multi-objective gray wolf optimization algorithm. ,in To determine the nest population size of the cuckoo search algorithm based on quantum behavior, take the first... The optimal gray wolf individuals are selected to match the nest size requirement of the quantum behavior cuckoo search algorithm. This solution set is directly used as the initial nest population for the quantum behavior cuckoo search algorithm, as shown in the following formula: Simultaneously, the mean of the multi-objective function of the Pareto solution set is... The cuckoo search algorithm, fed back to quantum behavior, serves as the initial target reference for the nest population. , , , After the multi-objective gray wolf optimization algorithm has been trained, the first The optimal position for the gray wolf.

[0081] 4.3 Quantum Behavior-Based Cuckoo Search Algorithm (QB-CS): Fine-grained Search for Local Optimum Solutions; The core function of the quantum behavior-based cuckoo search algorithm is to perform a fine-grained search on the Pareto optimal solution set generated by the multi-objective gray wolf optimization algorithm, thereby improving the accuracy of the flow solution. Simultaneously, it expands the search range through quantum behavior, avoiding local optima. Its model construction, training, and application process are as follows:

[0082] 4.3.1 Construction of the Cuckoo Search Algorithm Model Based on Quantum Behavior

[0083] 1. Objective function construction, inheriting the multi-objective optimization objective: The quantum behavior cuckoo search algorithm uses the weighted sum of the multi-objective functions of the multi-objective gray wolf optimization algorithm as the fitness function to ensure consistency with the overall optimization objective. The formula is as follows: ;in The fitness function value of the cuckoo search algorithm based on quantum behavior is given; the smaller the value, the better the corresponding flow solution. In the cuckoo search algorithm for quantum behavior, the first... The second iteration The location vectors of each bird's nest. , The nest population size of the cuckoo search algorithm based on quantum behavior; weights , , Consistent with the adaptive mutant krill population optimization algorithm, ensuring consistency of objectives.

[0084] 2. Quantum Behavior Bird Nest Location Update Model: The quantum behavior-based cuckoo search algorithm introduces quantum behavior into the cuckoo search, controlling the search accuracy through the quantum step size. The formula is as follows: ;in For the first The second iteration The updated position vectors of the bird's nests; For the first The optimal nest position of the cuckoo search algorithm with sub-iterative quantum behavior, that is, the nest position with the minimum fitness function value; The quantum step size is determined by the Pareto solution set range of the multi-objective gray wolf optimization algorithm, ensuring that the search is performed within the effective solution space, as shown in the following formula: That is, the maximum fluctuation range of all gas flow rates; For the first The second iteration The bird's nest corresponds to the first Solution for the flow rate of a certain gas; It is the natural logarithm function; This is a sign function; it outputs 1 when the input value is greater than or equal to 0, and -1 when the input value is less than 0.

[0085] 3. Improved Lévy flight model, balancing global and local search: To avoid getting trapped in local optima, the quantum behavior cuckoo search algorithm retains the improved Lévy flight model for generating new nests, as shown in the following formula: ;in For the first The second iteration The location vector of the new bird's nest; For Levi's flight coefficient; For Lévy distribution random numbers, With fixed parameters, the calculation formula is as follows: ; It is a gamma function; , These are random numbers that follow a standard normal distribution.

[0086] 4. Search accuracy model, fed back to the adaptive mutant krill swarm optimization algorithm: The search accuracy of the cuckoo search algorithm with quantum behavior is defined as the rate of change of the optimal nest fitness, as shown in the following formula: This accuracy reflects the convergence of the quantum behavior cuckoo search algorithm, and is fed back to the adaptive mutation krill swarm optimization algorithm to adjust the mutation probability. .

[0087] 5. Bird's Nest Elimination and Replacement Model: Setting the probability of discovering the Bird's Nest When the probability value is randomly generated At that time, a replacement operation is performed on the existing bird's nest. A new bird's nest is generated by improving Levi's flight to replace the original one, as shown in the following formula: .

[0088] 4.3.2 Training of the Cuckoo Search Algorithm Model Based on Quantum Behavior; 1. Initialization Parameters: Setting the Nest Population Size Maximum number of iterations for the cuckoo search algorithm based on quantum behavior Levi's flight coefficient Lévy distribution parameters Probability of discovering the Bird's Nest ; Receive the initial nest population output from the multi-objective gray wolf optimization algorithm and the mean of multi-objective functions 2. Iterative training process: The first step is to calculate the fitness of the initial nest population. Select the initial optimal bird's nest The second step is to determine the current iteration number. Is it less than If so, then perform quantum behavior to update the nest location and generate the updated population. The third step is to perform improved Levi flight on the updated population to generate a new nesting population. The fourth step involves performing nest elimination and replacement operations to generate the final updated population. The fifth step is to calculate the fitness of the final updated population and update the optimal nest. Step 6: Calculate the search accuracy The feedback is fed into the adaptive mutant krill swarm optimization algorithm; the local optimum solution is then used to optimize the local optimum. Feedback is fed into the multi-objective gray wolf optimization algorithm to adjust the gray wolf's position; the seventh step will... Increment by 1, return to step 2, and continue until the maximum number of iterations is reached. .

[0089] 4.3.3 Application of the quantum behavior-based cuckoo search algorithm model; After training, output the globally optimal nest location of the cuckoo search algorithm based on quantum behavior. This position represents the optimal target flow rate for each precursor gas. ,in The position of the Bird's Nest in the globally optimal location. The flow rate of a gas.

[0090] 4.4 Mixing ratio calculation and result output; Optimal target flow rate output by the cuckoo search algorithm based on quantum behavior. Combined with the total target flow rate of the mixed gas The mixing ratio of each precursor gas is calculated using the following formula: ;in For the first The mixing ratio of the precursor gases is such that the sum of the mixing ratios of all gases is 1, satisfying the normalization requirement. The calculated target flow rate is then... The data is transmitted in real time to step S5 for dynamic mixing of the precursor gas, ensuring the accuracy and stability of the mixing ratio.

[0091] 4.5 Summary of the core contributions of the three algorithms: The adaptive mutant krill swarm optimization algorithm provides a high-quality foundation for subsequent optimization through a highly diverse initial population. The adaptive mutation strategy, combined with the search accuracy feedback of the quantum behavior-based cuckoo search algorithm, dynamically balances population diversity and convergence speed, avoiding premature convergence. The multi-objective gray wolf optimization algorithm efficiently handles multi-objective optimization problems. Pareto sorting generates a balanced optimal solution set, while the local optimal solution feedback of the quantum behavior-based cuckoo search algorithm corrects the search direction and improves the quality of the solution set. The quantum behavior-based cuckoo search algorithm utilizes quantum behavior to achieve local fine-grained search, significantly improving the accuracy of the flow solution. Its search accuracy is fed back to the adaptive mutant krill swarm optimization algorithm to optimize the mutation strategy, forming a closed-loop fusion system of population optimization, multi-objective optimization, and local fine-grained refinement. The three work together to ensure the global optimality of the mixing ratio while taking into account the control accuracy and stability, solving the optimization bottleneck of a single algorithm in multi-objective, high-constraint scenarios.

[0092] S5: Dynamic mixing of precursor gases

[0093] The high-precision gas mixing system of the vapor deposition furnace is activated. This system consists of a mixing chamber, an electromagnetic flow control valve, a static mixer, and component concentration sensors. These components work together to achieve precise mixing of the precursor gas. The mixing chamber has a cylindrical structure made of Hastelloy alloy, with polished inner walls to prevent gas adsorption and residue. Built-in baffles, arranged in a spiral pattern, guide the gas into turbulent motion, promoting initial mixing. The static mixer is connected in series at the outlet of the mixing chamber and contains 10-15 sets of cross-type mixing elements made of corrosion-resistant ceramic. These elements achieve deep and uniform mixing of the gas through segmentation, recombination, and rotation, ensuring that the component uniformity error of the mixed gas is less than ±0.1%.

[0094] Each electromagnetic flow control valve corresponds to a specific gas supply pipeline for each precursor gas. The valve employs a direct-acting design with a response time of less than 50ms and an opening adjustment accuracy of ±0.1%, based on the target flow rate transmitted in step S4. The industrial controller outputs a pulse width modulation signal to precisely control the valve opening, ensuring that each precursor gas flows according to a preset target flow rate. The gas enters the mixing chamber stably. An online component concentration sensor is installed at the output end of the mixed gas. This sensor, based on laser Raman spectroscopy, can simultaneously detect the concentration of multiple gas components with an accuracy of ±0.001 mol / L and a sampling frequency of 2 times per second, real-time acquiring the actual concentration of each precursor gas in the mixed gas. .

[0095] Hybrid execution and feedback correction process as follows Figure 4 As shown.

[0096] S6: Hybrid proportional feedback correction based on PID-Kalman filtering

[0097] The mixing ratio feedback correction stage employs a PID-Kalman filter composite control strategy. First, a Kalman filter eliminates concentration detection noise interference, and then a PID controller achieves precise flow correction, ensuring dynamic stability of the mixing ratio. Firstly, based on the mixing ratio output from step S4... and total target flow rate of mixed gas The target concentration was derived by combining the ideal gas law. The derivation process fully considers the actual temperature inside the furnace. and pressure The effect is shown in the following formula: ;in The universal gas constant is taken as 8.314 J / (mol·K). The target concentration is calculated using this formula. It can accurately reflect the concentration requirements of the mixing ratio.

[0098] Calculate the actual concentration With target concentration deviation Because sensor noise and electromagnetic interference are unavoidable factors during concentration detection, the deviation signal contains redundant noise components, which need to be smoothed using Kalman filtering. Kalman filtering employs a recursive estimation method, and its state equation and observation equation are defined as follows: ; ;in For the first Smoothed deviation at time intervals, Assuming the deviation signal is a stationary sequence, The noise is the process noise, which follows a mean of 0 and a variance of . Gaussian distribution, For the first The observation bias at time is , , To observe the noise, it follows a mean of 0 and a variance of . The Gaussian distribution.

[0099] By employing a two-step iterative process of prediction and update using Kalman filtering, noise suppression of the deviation signal is achieved, resulting in a smoothed deviation. Input to a PID controller. The PID controller uses a positional control algorithm, and its output is the flow correction. The control formula is as follows: ;in , , These are the initial parameters for the PID controller, which can be dynamically adjusted according to actual operating conditions. To control the cycle, This represents the current control cycle number. (Proportional term) Fast response deviation, integral term Eliminating static error, differential term Suppressing overshoot, the three factors work together to achieve accurate calculation of flow correction.

[0100] Based on flow correction Calculate the corrected target flow rate At the same time, flow constraint judgment is introduced to ensure satisfy If the flow exceeds the constraints, the boundary value will be automatically used as the final corrected flow. The corrected target flow will then be... The data is transmitted in real time to step S5 to adjust the opening of the electromagnetic flow control valve, achieving dynamic correction of the mixing ratio; simultaneously, the correction amount is... The data is transferred to step S9 to provide data support for the closed-loop optimization of the deposition process, ensuring that the mixing ratio is always maintained within the optimal range.

[0101] S7: Monitoring for Abnormalities in the Mixing Process

[0102] Construct a comprehensive anomaly monitoring system for mixing processes, based on the purity detection value of step S1. Operating data for step S3 , , , and the actual concentration in step S5. A multi-dimensional anomaly judgment indicator system was established, with each indicator having a clearly defined anomaly threshold. By comparing the collected data with the threshold in real time, the anomaly could be quickly identified and judged.

[0103] The purity anomaly threshold is set to This threshold is based on the stringent requirements of the deposition process for raw material purity, ensuring that only high-purity gases enter the mixing process; the temperature anomaly threshold is set to... ,in To ensure the target operating temperature is pre-set according to the deposition process, excessive temperature fluctuations will affect gas properties and reactivity; the pressure anomaly threshold is set as follows: , The target working pressure is preset, and unstable pressure will cause fluctuations in gas flow rate and diffusion rate.

[0104] The abnormal deposition rate threshold is set to , The target deposition rate is preset, and fluctuations in the deposition rate directly reflect the rationality of the mixing ratio; the abnormal flow rate threshold is set as follows. This is used to monitor the deviation between the actual flow rate and the target flow rate, preventing flow control failure; the concentration anomaly threshold is set to... This directly reflects the control precision of the mixing ratio.

[0105] The anomaly monitoring system adopts a real-time data processing mode, which judges all indicators once per second. If any indicator meets the abnormal conditions, it is immediately judged as an anomaly in the mixing process. At the same time, it records the anomaly type, including purity anomaly, temperature anomaly, pressure anomaly, deposition rate anomaly, flow rate anomaly, concentration anomaly, as well as detailed information such as the time of anomaly occurrence, abnormal value, and normal threshold, forming an anomaly log.

[0106] S8: Exception Handling and Recovery

[0107] A tiered and categorized anomaly handling mechanism is established. Based on the anomaly type determined in step S7, targeted handling measures are implemented to ensure rapid resolution of anomalies and restoration of stable operation of the mixing process. If a purity anomaly is detected, the industrial controller immediately issues a command to close the gas supply valve for that type of precursor gas and simultaneously activates the gas supply valve of the backup gas tank. The gas in the backup gas tank is qualified gas that has passed the purity test in step S1. The switching process adopts a seamless switching design with a switching time of less than 1 second to avoid interruption of the mixed gas supply or sudden changes in composition. At the same time, an audible and visual alarm signal is issued, with an alarm sound level greater than 85dB and an alarm indicator flashing red, prompting the operator to replace the original gas tank in time. After the operator replaces the gas tank, the system automatically performs the purity test procedure of step S1 on the gas in the new gas tank. If the test is qualified, the normal mixing process is automatically restored. If the test is unqualified, the alarm continues to be issued and the gas supply from the backup gas tank is maintained.

[0108] If an abnormal temperature is detected, the system adjusts the heating power of the vapor deposition furnace. The heating system uses PID temperature control, dynamically adjusting the heating power based on the deviation between the current temperature and the target temperature. Large deviations result in rapid heating or cooling with high power, while small deviations require fine-tuning with low power. During the adjustment process, temperature data is collected every 10 seconds to monitor the temperature trend in real time until... The normal mixing process is restored, and the temperature adjustment is fed back to step S3 to update the operating data.

[0109] If an abnormal pressure is detected, adjust the power of the vacuum pumping system in the furnace cavity. The vacuum pumping system uses frequency conversion control, which adjusts the pumping rate by changing the operating frequency of the vacuum pump, thereby controlling the pressure inside the furnace. During the recovery process, pressure data is collected every 10 seconds until... The normal mixing process is restored, and the pressure adjustment is fed back to step S3 to update the operating data.

[0110] If an abnormal deposition rate is detected, the system automatically triggers the S4 step of the mixing ratio recalculation process, based on the current deposition rate. Based on operating data and physical property parameters, the target flow rate was re-optimized. and will new Transfer to step S5 for execution until... .

[0111] If an abnormal flow rate is detected, the system first checks the operating status of the electromagnetic flow control valve. It determines if a fault exists by analyzing the valve's opening and current signals. If a valve is faulty, it automatically switches to the backup flow control valve. The backup valve is the same model as the main valve, and the switching time is less than 0.5 seconds. If signal interference is detected, the system automatically strengthens signal shielding measures, including activating the backup signal transmission channel and increasing the signal amplification factor. After repair, the flow sensor is recalibrated to ensure... Restore the normal mixing process.

[0112] If the concentration is determined to be abnormal, the system automatically increases the proportional gain of the PID controller in step S6. To 1.2, shorten the control cycle. To 0.2s, the flow correction speed was accelerated, and the sampling frequency of the component concentration sensor was increased to 4 times per second to improve the feedback response speed, until... Restore the initial parameters of the PID controller.

[0113] After all anomalies are handled, the results, including the anomaly handling time, handling measures, and post-handling parameter values, are recorded in the anomaly log and fed back to step S3 to update the operating data, ensuring the continuous and stable operation of the hybrid process.

[0114] S9: Closed-loop optimization of the deposition process

[0115] Construct a closed-loop optimization system based on full-process data, and continuously analyze the deposition rate collected in step S3. and the flow correction amount in step S6 The system dynamically optimizes the mixing ratio calculation model and feedback correction parameters to improve the long-term stability of control accuracy. Each optimization cycle lasts 10 minutes, and within each cycle, the system calculates the average deposition rate. ,in For the first The deposition rate of each sample is calculated, with 1200 representing the number of samples collected within 10 minutes; the standard deviation of the flow correction is also calculated. ,in For the first The amount of traffic correction each time. This represents the average flow correction amount within that period.

[0116] Parameters are optimized and adjusted based on the calculation results. This indicates that the current mixing ratio is insufficient to support the deposition rate, and the optimization priority of the deposition rate needs to be increased. Therefore, the weight coefficients in the S4 step optimization model are adjusted. Adjusted to 0.6, with a corresponding reduction. up to 0.2, To 0.2, the mixing ratio calculation focuses more on achieving the deposition rate target; if This indicates that the flow correction amount fluctuates significantly and the mixing ratio is not stable enough, requiring a reduction in the response sensitivity of the PID controller. Therefore, the proportional gain of the PID controller should be adjusted. Decrease by 0.1, integral coefficient Increasing it by 0.02 makes the flow correction smoother and reduces fluctuations.

[0117] The updated optimized model parameters, including weighting coefficients and PID controller parameters, are fed back to steps S4 and S6, respectively, to achieve dynamic optimization of the mixing ratio calculation model and feedback correction parameters. Simultaneously, at the end of each optimization cycle, the system evaluates the optimization effect, calculating indicators such as the optimized deposition rate deviation and flow correction fluctuation. If the indicators do not meet expectations, the parameters are adjusted further; if the indicators meet the requirements, the current parameters are maintained. This forms a complete closed-loop control system of data acquisition, ratio calculation, mixing execution, feedback correction, and model optimization, continuously improving the control accuracy and long-term stability of the precursor gas mixing ratio, ensuring the consistency of thin film deposition quality.

[0118] In some embodiments, the Chaos Mapping Improved Weed Optimization Algorithm (CM-IWO) is incorporated. By leveraging the randomness, ergodicity, and sensitivity of chaotic sequences, the CM-IWO dynamically adjusts the reproduction, diffusion, and competition strategies of the weed population, generating an initial / intermediate optimized population with better diversity and more even distribution. Simultaneously, it establishes a bidirectional interaction mechanism with the Adaptive Mutant Krill Population Optimization Algorithm (AM-KHA), the Multi-Objective Gray Wolf Optimization Algorithm (MOGWO), and the Quantum Behavior Cuckoo Search Algorithm (QB-CS). This provides a high-quality population foundation for other algorithms and receives feedback parameters from other algorithms to optimize its own optimization strategy, breaking through the optimization bottleneck of a single algorithm and further improving the global optimality and accuracy stability of the mixed ratio calculation.

[0119] I. Construction of the Chaotic Mapping Improved Weed Optimization Algorithm Model

[0120] (1) Fitness function construction; The fitness function of CM-IWO inherits the global optimization objectives: achieving the target thin film deposition quality, maximizing gas utilization, and stabilizing the deposition rate. It adopts a weighted sum form, with the weights logically consistent with other algorithms but distinguished by subscripts to ensure the consistency of objectives. The formula is as follows:

[0121] ;in The fitness function value of CM-IWO; the smaller the value, the better the corresponding flow solution. : CM-IWO The second iteration The position vector of each weed. , The weed population size of CM-IWO; , , The target weight coefficients of CM-IWO sum to 1, which is suitable for the optimization characteristics of chaotic populations. Real-time deposition rate; Target deposition rate; : No. The second iteration The weeds correspond to the first Solution for the flow rate of a certain gas; Total target flow rate of the mixed gas; : No. The standard deviation of the gas flow rate for each weed species reflects the uniformity of the flow rate distribution. Total number of precursor gases.

[0122] (2) Chaotic mapping model; Logistic chaotic mapping is used to generate chaotic sequences to dynamically adjust the reproduction number and diffusion step size of weeds, avoiding premature convergence of the population. The formula is as follows: ;in : No. A chaotic variable, with a range of values. ; : No. One chaotic variable; Chaos control parameters, initial values This ensures the mapping is in a chaotic state, and is subsequently dynamically adjusted through interaction with QB-CS; Initialization: Initial value of the chaotic sequence. Randomly generated To avoid getting stuck in a fixed point.

[0123] (3) Weed reproduction model: The number of weeds reproduces dynamically with the chaotic sequence and their own fitness. The better the fitness and the larger the chaotic sequence value, the more weeds reproduce, ensuring the transmission of genes of high-quality individuals. The formula is as follows: ;in : No. The second iteration The number of weeds that reproduce; , Minimum and maximum number of weed reproductions; : No. The second iteration The chaotic variables corresponding to each weed are randomly assigned from the chaotic sequence; : No. The second iteration Fitness value of individual weeds; , : No. Maximum and minimum fitness values ​​of the CM-IWO population in the next iteration; : The floor function, which rounds down to an integer.

[0124] (4) Weed diffusion model; the diffusion step size of weed seeds is dynamically adjusted by the chaotic sequence and the population fitness variance fed back by AM-KHA, balancing global exploration and local development, as shown in the following formula: ;in : No. The second iteration Seed dispersal step length of a single weed; Initial maximum diffusion step size; : The maximum number of iterations for CM-IWO; AM-KHA feedback Variance of population fitness in each iteration; : The preset maximum value of the fitness variance of the AM-KHA population (consistent with AM-KHA); : Natural exponential function.

[0125] Seed diffusion location update formula: ;in : No. The second iteration The first of the weeds The position vector of each seed. ; : Follows a standard normal distribution A 3D random vector controls the direction of diffusion.

[0126] (5) Competitive elimination model; after each iteration, the parent weeds and offspring seeds are merged, sorted in ascending order according to their fitness function values, and the top-ranked weeds are retained. The optimal individual is selected as the population for the next iteration, as shown in the following formula: ;in : No. The parent population of the next iteration CM-IWO; : No. The set of offspring seeds generated in the next iteration; Set merge operator; Sort by fitness in ascending order and take the top results. Individual.

[0127] (6) Bidirectional interaction model with other algorithms; CM-IWO establishes bidirectional interaction with AM-KHA, MOGWO, and QB-CS respectively, and achieves collaborative optimization through parameter feedback. The interaction process and formula are as follows:

[0128] ①CM-IWO and AM-KHA interact bidirectionally; AM-KHA provides population fitness reference for CM-IWO, and CM-IWO provides high-diversity population supplements for AM-KHA, while adjusting their optimization strategies at the same time.

[0129] AM-KHA→CM-IWO:AM-KHA will be the first mean fitness of the population in the next iteration Feedback is sent to CM-IWO to adjust its breeding stock. The formula is: ; For the first The reproductive base of the next iteration CM-IWO; The mean fitness of the AM-KHA population; This is the preset maximum value for AM-KHA fitness;

[0130] CM-IWO→AM-KHA: CM-IWO will be the first Chaotic entropy of the next iteration Feedback is given to AM-KHA to adjust the mutation probability of AM-KHA, using the formula: Chaotic entropy calculation: , The number of samples in the chaotic sequence. Avoid logarithmic inconsistencies; For the first The chaotic entropy of the next iteration of CM-IWO reflects the diversity of chaotic sequences and takes values... .

[0131] ②CM-IWO and MOGWO interact bidirectionally; MOGWO provides CM-IWO with multi-objective optimization directions, and CM-IWO supplements MOGWO with Pareto optimal solution candidate individuals;

[0132] MOGWO→CM-IWO: MOGWO will Mean crowding of the Pareto optimal solution set in the next iteration Feedback is given to CM-IWO to adjust the chaotic weights for the diffusion step size, using the following formula: ; ; The chaos weights are used to adjust the degree of influence of the chaotic sequence on the diffusion step size. The mean of the crowding in the MOGWO Pareto optimal solution set; : The preset maximum value for MOGWO congestion;

[0133] CM-IWO→MOGWO: CM-IWO will be the first Local optimal solution in the next iteration Feedback is sent to MOGWO to correct the position of MOGWO's α Grey Wolf. The formula is: ; For the first The optimal weed location (minimum fitness) in the next iteration of CM-IWO. This is the position of MOGWO's α Grey Wolf.

[0134] ③ CM-IWO and QB-CS interact bidirectionally; QB-CS provides local optimization accuracy feedback for CM-IWO, and CM-IWO expands the initial bird nest population diversity for QB-CS;

[0135] QB-CS→CM-IWO:QB-CS will be the first Search accuracy of the next iteration Feedback is sent to CM-IWO to adjust the chaos control parameters. ,formula: ; For the search accuracy of QB-CS; Initial values ​​of chaos control parameters;

[0136] CM-IWO→QB-CS: CM-IWO will be the first The optimal population before the next iteration individual Feedback is given to QB-CS to expand the initial nesting population of QB-CS, using the following formula: ; For CM-IWO The 5 weed locations with the best fitness in the next iteration; This is the initial nesting population of QB-CS.

[0137] II. Training the Improved Weed Optimization Algorithm Model Using Chaotic Mapping

[0138] Initialization parameters; weed population size Maximum number of iterations Reproduction parameters: , Diffusion parameters: Initial maximum diffusion step size Chaos parameters: Initial chaos control parameters Initial value of chaotic sequence Population initialization: based on gas flow rate constraints Randomly generate the initial weed population. The flow component of each weed is within the constraints.

[0139] Iterative training process; 1. Calculate the fitness value of each weed in the initial population. ( ), to select the initial optimal weeds 2. The mean initial population fitness received from AM-KHA feedback. Calculate the initial breeding population ; Receive initial search accuracy from QB-CS feedback Adjusting the chaos control parameters 3. Determine the current iteration number. Is it less than ( Starting from 0), if so, perform the following steps: a. Generate the first... Chaotic sequence of the next iteration (Based on the Logistic mapping formula); b. Calculate the number of reproductions for each weed. (Based on the chaotic adaptive reproduction formula); c. Generate offspring seeds: For each weed, according to the reproduction number And the diffusion position update formula, to generate offspring seeds Merge all offspring seeds to obtain d. Population merging and competitive elimination: merging parent populations and offspring seeds Calculate the fitness values ​​of all individuals, sort them in ascending order, and retain the top 40 best individuals. e. Update the optimal weeds f. Inter-algorithm interaction: Calculating chaotic entropy Feedback is given to AM-KHA to adjust the mutation probability. The mean crowding of the Pareto optimal solution set received from MOGWO feedback. Adjusting the chaotic weights of the diffusion step size Search accuracy received from QB-CS feedback Adjusting the chaos control parameters ;Will Feedback was sent to MOGWO to correct the position of the α Grey Wolf. 4. Increment by 1, return to step 3, until... Stop iterating.

[0140] III. Application of Chaotic Mapping in Improving Weed Optimization Algorithm Model

[0141] After training, output the globally optimal population of CM-IWO. (Sorted in ascending order of fitness), and integrated into the collaborative process of the original three algorithms in the following manner:

[0142] 1. Supplement the initial population of AM-KHA: Use the top of the optimal population of CM-IWO... individual The AM-KHA population was merged with the initial AM-KHA population to obtain an expanded initial AM-KHA population. This will enhance the diversity of the initial AM-KHA population.

[0143] 2. Supplement the Pareto optimal solution set of MOGWO: Add the top-ranked population of the CM-IWO optimal set... individual Adding the Pareto optimal solution set of MOGWO The expanded Pareto optimal solution set is obtained. , as the initial nesting population of QB-CS;

[0144] 3. Provide closed-loop optimization data: Show the optimal fitness value change curve of CM-IWO The data is transmitted to S9 as a reference for optimizing multiple algorithm parameters. The smaller the value, the higher the efficiency of population optimization, and the number of iterations of other algorithms can be adjusted accordingly.

[0145] IV. Core Contributions of CM-IWO to the Fusion System

[0146] 1. Enhanced population diversity: By dynamically adjusting the reproduction number and diffusion step size through Logistic chaotic mapping, the generated population covers a wider solution space, avoiding AM-KHA and MOGWO from getting trapped in local optima. This is especially beneficial for multi-constraint optimization scenarios under complex working conditions of vapor deposition furnaces, improving global optimization capabilities.

[0147] 2. Two-way interactive collaboration: Establish parameter feedback mechanisms with the original three algorithms in pairs. It not only receives feedback information such as the accuracy and crowding of other algorithms to optimize its own strategy, but also provides high-quality populations and local optimal solutions for other algorithms, forming an enhanced collaborative system of chaotic population optimization → multi-objective optimization → local refinement.

[0148] 3. Improved stability of optimization: The ergodicity of chaotic sequences makes the population distribution more uniform, reducing optimization fluctuations caused by insufficient population diversity of a single algorithm. Combined with the competitive elimination mechanism of CM-IWO, it ensures the continuous transmission of the best individuals, ultimately improving the long-term stability of mixing ratio control and ensuring the consistency of thin film deposition quality.

[0149] Summary of inter-algorithm fusion and synergy: With the addition of CM-IWO, the original unidirectional population transfer process of AM-KHA→MOGWO→QB-CS is upgraded to a closed-loop fusion system of multi-algorithm bidirectional interaction + population complementarity. The core connections are as follows:

[0150] 1. Population transmission: ;

[0151] 2. Parameter feedback: AM-KHA↔CM-IWO: MOGWO↔CM-IWO: QB-CS↔CM-IWO: Existing interaction: , , ;

[0152] 3. Final Output: QB-CS outputs the optimal target flow based on the expanded initial population. This ensures the accuracy, stability, and global optimality of the mixing ratio calculation.

[0153] A precise control system for the mixing ratio of precursor gases in a vapor deposition furnace, employing any one of the methods described above, wherein the precise control system for the mixing ratio of precursor gases in a vapor deposition furnace comprises:

[0154] The gas pretreatment and purity detection module is used to pretreatment the precursor gas and detect its purity to obtain the purity detection value of each precursor gas.

[0155] The physical property parameter library construction module is used to collect the basic physical property parameters of the precursor gas that has passed the purity test, and establish the precursor gas physical property parameter library after working condition adaptation correction.

[0156] The operating condition data acquisition module is used to acquire operating condition data of the vapor deposition furnace in real time.

[0157] The multi-algorithm fusion computing module is used to conduct collaborative optimization calculations based on purity detection values, precursor gas property parameter database data, and operating condition data, employing an adaptive mutant krill swarm optimization algorithm, a multi-objective gray wolf optimization algorithm, and a quantum-behavioral cuckoo search algorithm. First, the adaptive mutant krill swarm optimization algorithm generates an optimal population, which serves as the initial population for the multi-objective gray wolf optimization algorithm. The multi-objective gray wolf optimization algorithm then generates a Pareto optimal solution set based on the initial population, which is used as the initial nest population for the quantum-behavioral cuckoo search algorithm. During the collaborative process, the three algorithms achieve bidirectional interactive adjustment through parameter feedback. Finally, the quantum-behavioral cuckoo search algorithm outputs the optimal target flow rate for each precursor gas. Based on the optimal target flow rate, the mixing ratio of each precursor gas is calculated according to the total target flow rate of the mixed gas, which is used to derive the target concentration of each precursor gas after mixing.

[0158] The gas mixing execution module is used to control the precursor gas to enter the mixing system for dynamic mixing according to the optimal target flow rate, complete the mixing ratio regulation of the precursor gas, and collect the actual concentration of each precursor gas after mixing.

[0159] The feedback correction module is used to obtain the flow correction amount based on the deviation between the actual concentration and the target concentration, adjust the optimal target flow rate, and complete the secondary control of the precursor gas mixing ratio.

[0160] The anomaly monitoring module is used to monitor anomalies in the mixing process by combining purity test values, operating condition data, and actual concentration, and to determine the type of anomaly.

[0161] The exception handling module is used to execute corresponding exception handling measures according to the exception type to restore the stability of the hybrid process;

[0162] The closed-loop optimization module is used to dynamically optimize the algorithm parameters of the multi-algorithm fusion calculation module and the feedback correction parameters of the feedback correction module based on the deposition rate and flow correction amount in the operating data, and realizes the closed-loop optimization and control of the precursor gas mixing ratio through iterative updates.

Claims

1. A method for precisely controlling the mixing ratio of precursor gases in a vapor deposition furnace, characterized in that, Includes the following steps: S1: Pre-treat the precursor gases and test their purity to obtain the purity test values ​​for each precursor gas; S2: Collect the basic physical property parameters of the precursor gas that has passed the purity test, and establish a precursor gas physical property parameter library after working condition adaptation correction. S3: Real-time acquisition of operating data of the vapor deposition furnace; S4: Based on purity detection values, precursor gas property parameter database data, and operating condition data, a collaborative optimization calculation is carried out using an adaptive mutant krill swarm optimization algorithm, a multi-objective gray wolf optimization algorithm, and a quantum-behavioral cuckoo search algorithm. First, the adaptive mutant krill swarm optimization algorithm generates the optimal population, which is then used as the initial population for the multi-objective gray wolf optimization algorithm. The multi-objective gray wolf optimization algorithm generates a Pareto optimal solution set based on the initial population, which is then used as the initial nest population for the quantum-behavioral cuckoo search algorithm. During the collaborative process, the three algorithms achieve bidirectional interactive adjustment through parameter feedback. Finally, the quantum-behavioral cuckoo search algorithm outputs the optimal target flow rate for each precursor gas. S5: Control the precursor gas to enter the mixing system for dynamic mixing according to the optimal target flow rate, and complete the mixing ratio regulation of the precursor gas.

2. The method for precisely controlling the mixing ratio of precursor gas in a vapor deposition furnace according to claim 1, characterized in that, S4 also includes calculating the mixing ratio of each precursor gas based on the optimal target flow rate and the total target flow rate of the mixed gas, which is used to derive the target concentration of each precursor gas after mixing; S5 also includes collecting the actual concentration of each precursor gas after mixing. Methods for precisely controlling the mixing ratio of precursor gases in a vapor deposition furnace also include: S6: Based on the deviation between the actual concentration and the target concentration, the flow correction amount is obtained through feedback correction, the optimal target flow rate is adjusted, and the secondary control of the precursor gas mixing ratio is completed. S7: Combine purity test values, operating condition data and actual concentration to monitor abnormalities in the mixing process and determine the type of abnormality; S8: Execute the corresponding exception handling measures according to the exception type to restore the stability of the hybrid process; S9: Based on the deposition rate and flow correction in the operating data, dynamically optimize the algorithm parameters of S4 and the feedback correction parameters of S6, and achieve closed-loop optimization and control of the precursor gas mixing ratio through iterative updates.

3. The method for precisely controlling the mixing ratio of precursor gases in a vapor deposition furnace according to claim 1, characterized in that, The bidirectional interactive adjustment of parameters between the multi-objective gray wolf optimization algorithm and the adaptive mutant krill swarm optimization algorithm is as follows: the multi-objective gray wolf optimization algorithm calculates the fitness variance of the optimal layer population during the iteration process, and feeds this fitness variance back to the adaptive mutant krill swarm optimization algorithm to adjust the krill's movement step size in the adaptive mutant krill swarm optimization algorithm.

4. The method for precisely controlling the mixing ratio of precursor gas in a vapor deposition furnace according to claim 1, characterized in that, The bidirectional interactive adjustment of parameters between the quantum behavior cuckoo search algorithm and the adaptive mutation krill swarm optimization algorithm is as follows: the quantum behavior cuckoo search algorithm calculates the search accuracy during the iteration process and feeds the search accuracy back to the adaptive mutation krill swarm optimization algorithm to adjust the mutation probability of the adaptive mutation krill swarm optimization algorithm.

5. The method for precisely controlling the mixing ratio of precursor gas in a vapor deposition furnace according to claim 1, characterized in that, The bidirectional interactive adjustment of parameters between the quantum behavior cuckoo search algorithm and the multi-objective gray wolf optimization algorithm is as follows: the local optimal solution obtained by the quantum behavior cuckoo search algorithm during the iteration process is fed back to the multi-objective gray wolf optimization algorithm to correct the search direction of the gray wolf in the multi-objective gray wolf optimization algorithm.

6. The method for precisely controlling the mixing ratio of precursor gas in a vapor deposition furnace according to claim 1, characterized in that, The adaptive mutant krill swarm optimization algorithm, the multi-objective gray wolf optimization algorithm, and the quantum behavior cuckoo search algorithm all have a preset maximum number of iterations. When the number of iterations of each algorithm reaches its preset maximum number of iterations, the iteration of the corresponding algorithm is stopped and the current optimal result is output.

7. The method for precisely controlling the mixing ratio of precursor gas in a vapor deposition furnace according to claim 1, characterized in that, The correction of the basic physical property parameters in S2 is based on the actual operating temperature and pressure of the vapor deposition furnace, and the van der Waals equation is used to dynamically adjust the basic physical property parameters.

8. The method for precisely controlling the mixing ratio of precursor gas in a vapor deposition furnace according to claim 1, characterized in that, In S6, feedback correction uses a combination of PID controller and Kalman filter. First, the deviation between the actual concentration and the target concentration is smoothed by Kalman filter. Then, the smoothed deviation is input into the PID controller to calculate the flow correction amount.

9. The method for precisely controlling the mixing ratio of precursor gas in a vapor deposition furnace according to claim 1, characterized in that, In S7, anomaly monitoring of the mixing process is carried out based on multi-dimensional indicators, including purity, temperature, pressure, deposition rate, flow rate and concentration. Each indicator corresponds to a preset anomaly judgment standard. When any indicator meets the corresponding anomaly judgment standard, it is immediately judged as an anomaly in the mixing process and the anomaly type and related data are recorded.

10. A system for precisely controlling the mixing ratio of precursor gases in a vapor deposition furnace, characterized in that, The method for precisely controlling the mixing ratio of precursor gases in a vapor deposition furnace according to any one of claims 2 to 9, wherein the precise control system for the mixing ratio of precursor gases in a vapor deposition furnace comprises: The gas pretreatment and purity detection module is used to pretreatment the precursor gas and detect its purity to obtain the purity detection value of each precursor gas. The physical property parameter library construction module is used to collect the basic physical property parameters of the precursor gas that has passed the purity test, and establish the precursor gas physical property parameter library after working condition adaptation correction. The operating condition data acquisition module is used to acquire operating condition data of the vapor deposition furnace in real time. The multi-algorithm fusion computing module is used to conduct collaborative optimization calculations based on purity detection values, precursor gas property parameter database data, and operating condition data, employing an adaptive mutant krill swarm optimization algorithm, a multi-objective gray wolf optimization algorithm, and a quantum-behavioral cuckoo search algorithm. First, the adaptive mutant krill swarm optimization algorithm generates an optimal population, which serves as the initial population for the multi-objective gray wolf optimization algorithm. The multi-objective gray wolf optimization algorithm then generates a Pareto optimal solution set based on the initial population, which is used as the initial nest population for the quantum-behavioral cuckoo search algorithm. During the collaborative process, the three algorithms achieve bidirectional interactive adjustment through parameter feedback. Finally, the quantum-behavioral cuckoo search algorithm outputs the optimal target flow rate for each precursor gas. Based on the optimal target flow rate, the mixing ratio of each precursor gas is calculated according to the total target flow rate of the mixed gas, which is used to derive the target concentration of each precursor gas after mixing. The gas mixing execution module is used to control the precursor gas to enter the mixing system for dynamic mixing according to the optimal target flow rate, complete the mixing ratio regulation of the precursor gas, and collect the actual concentration of each precursor gas after mixing. The feedback correction module is used to obtain the flow correction amount based on the deviation between the actual concentration and the target concentration, adjust the optimal target flow rate, and complete the secondary control of the precursor gas mixing ratio. The anomaly monitoring module is used to monitor anomalies in the mixing process by combining purity test values, operating condition data, and actual concentration, and to determine the type of anomaly. The exception handling module is used to execute corresponding exception handling measures according to the exception type to restore the stability of the hybrid process; The closed-loop optimization module is used to dynamically optimize the algorithm parameters of the multi-algorithm fusion calculation module and the feedback correction parameters of the feedback correction module based on the deposition rate and flow correction amount in the operating data, and realizes the closed-loop optimization and control of the precursor gas mixing ratio through iterative updates.

Images