Digital control system for thin film preparation process based on continuous learning

Abstract

This invention provides a digital adjustment system for thin film fabrication processes based on continuous learning, relating to the field of thin film fabrication technology. It includes a data manifold construction module, a nonlinear parameter control module, a multimodal perturbation sensing module, and a topology sequence reconstruction module. The system internally incorporates a hyperdimensional memory network, comprising a process parameter hyperbody, a performance characteristic hyperbody, and an adaptive forgetting curve algorithm. The system uses the multimodal perturbation sensing module to monitor perturbations in the process environment in real time, such as sudden drops in vacuum, fluctuations in gas source purity, and batch variations of materials. It utilizes an entropy source tracking algorithm to quickly calculate a stable curvature vector and perform dynamic correction. Combined with the adaptive forgetting curve algorithm of the hyperdimensional memory network, the system can dynamically manage historical data, retain core knowledge, and react rapidly to perturbations.

Description

Technical Field

[0001] This invention relates to the field of thin film preparation technology, specifically to a digital adjustment system for thin film preparation processes based on continuous learning. Background Technology

[0002] With the rapid development of high-end manufacturing industries such as semiconductor chips, flexible displays, and new energy batteries, thin film materials, as core components of device structures, have become crucial factors determining product competitiveness in terms of fabrication precision and performance stability. From nanoscale insulating layers to functional coatings, thin film materials (such as metal thin films, compound semiconductor thin films, and high dielectric constant thin films) are widely used in transistor gates, display touch layers, and battery electrode protection, requiring thickness control precision of ±1nm, surface roughness of less than 0.5nm, and structural stability under extreme environments such as high temperatures (300-1200℃) and strong corrosion. Therefore, achieving intelligent control of the thin film fabrication process, dynamic adaptation to complex working conditions, and continuous optimization of process parameters has become a critical technological bottleneck that urgently needs to be overcome in the high-end manufacturing field.

[0003] Currently, the control methods for thin film preparation processes can be mainly divided into the following categories:

[0004] Experience-driven static formulation: This method relies on the historical experience of process engineers to establish fixed process parameters (such as deposition temperature of 400°C, gas flow rate of 50 sccm, and deposition time of 300 s), and then refines the formulation through offline detection (such as electron microscopy observation and spectral analysis). This approach is still widely used in small-batch laboratory preparations.

[0005] Single-parameter closed-loop control: Mainstream industrial equipment constructs a feedback loop using a single sensor (such as a quartz crystal microbalance for film thickness measurement or a thermocouple for temperature measurement). When the detected value deviates from the target (e.g., film thickness deviation > 3nm), the corresponding parameter is linearly adjusted (e.g., power is increased by 10%). For example, this method is commonly used in magnetron sputtering equipment to stabilize the deposition rate.

[0006] Multi-parameter linkage control: Some high-end systems integrate data from multiple sensors such as temperature, pressure, and gas flow rate, and calculate parameter adjustments based on preset mathematical models (such as reaction kinetic equations) to achieve multi-variable coordinated control. This type of system is widely used in atomic layer deposition (ALD) equipment to ensure film thickness uniformity.

[0007] However, existing technologies have the following significant limitations:

[0008] Problem 1: Experience-driven static formulations lack dynamic adaptability and cannot cope with real-time disturbances. When the process environment fluctuates (such as a sudden drop in vacuum of 5 Pa or fluctuation in gas purity of ±5%) or material batches change, fixed parameters will cause drastic fluctuations in film performance (such as refractive index deviation >0.1 or pinhole density increase by 10 times). Furthermore, formulation iteration relies on manual trial and error, and the development cycle can last for weeks or even months.

[0009] Question 2: Single-parameter closed-loop control ignores the coupling relationship between parameters. For example, an increase in temperature not only affects the deposition rate but also changes the grain growth direction. However, existing systems only adjust the temperature independently, leading to microstructure runaway (such as grain size distribution deviation > 20%), making it difficult to meet the precise requirements of devices for thin film microstructure performance.

[0010] Problem 3: Poor generalization ability of multi-parameter linkage control models. Existing mathematical models are mostly built for specific materials (such as silicon dioxide) and processes (such as chemical vapor deposition). When the material is changed (such as from oxide to nitride) or the process (such as from PVD to CVD), the model fails and needs to be remodeled, resulting in extremely poor cross-scenario adaptability.

[0011] Question 4: Data processing and decision-making lag. Traditional systems have processing cycles of hundreds of milliseconds for multi-source sensor data, and cannot respond in time to rapid disturbances (such as plasma arc flash), leading to an increase in defect rates (such as thin film pinhole rate > 5%). Furthermore, they lack intelligent mining of historical data, making it impossible to accumulate and reuse process knowledge.

[0012] Therefore, a digital adjustment system for thin film preparation processes based on continuous learning is needed to solve the above problems. Summary of the Invention

[0013] Technical problems to be solved

[0014] To address the shortcomings of existing technologies, this invention provides a digital adjustment system for thin film preparation processes based on continuous learning, which solves the problems mentioned in the background.

[0015] Technical solution

[0016] To achieve the above objectives, this invention provides the following technical solution: a digital adjustment system for thin film fabrication processes based on continuous learning, comprising a data manifold construction module, a nonlinear parameter control module, a multimodal perturbation sensing module, and a topology sequence reconstruction module; the system internally incorporates a hyperdimensional memory network, which includes a process parameter hyperbody, a performance feature hyperbody, and an adaptive forgetting curve algorithm, wherein:

[0017] The data manifold construction module uses multi-source sensor data and Riemannian geometric topology algorithm to map thin film preparation parameters and performance data to a high-dimensional data manifold, and constructs and stores the formulation manifold matrix.

[0018] The nonlinear parameter control module is based on the formulation manifold matrix and optimizes macroscopic process parameters through a vector field gradient descent algorithm. At the same time, it uses a manifold curvature prediction algorithm to simulate the thin film growth process and predict the thin film properties.

[0019] The multimodal disturbance sensing module monitors process environment disturbances in real time, calculates and generates a stable curvature vector using an entropy source tracking algorithm, and dynamically corrects the system to ensure process stability.

[0020] The topology sequence reconstruction module receives optimization parameters and performance prediction results, and uses a fractal chaotic reconstruction algorithm to dynamically adjust and recombine process steps such as deposition, annealing, and cleaning to generate and execute a self-organizing recipe sequence.

[0021] Preferably, the hyperdimensional memory network further includes the following:

[0022] Process parameter hyperbody: used to store the coordinate data of process parameters such as temperature, pressure, and gas flow rate in a high-dimensional manifold space;

[0023] Performance characteristic superbody: used to store coordinate data of thin film properties such as film thickness, uniformity, and grain size in a high-dimensional manifold space;

[0024] Adaptive Forgetting Curve Algorithm: Through time decay and importance weighting mechanism, historical data in the superbody is dynamically managed. The decay rate is adjusted according to the distance between the data and the core process manifold region and the application frequency to prevent the forgetting of core knowledge.

[0025] Preferably, the nonlinear parameter control module includes a vector field gradient descent unit, a manifold curvature prediction unit, a cross-domain manifold mapping unit, and a multi-scale coupled manifold unit.

[0026] Preferably, the vector field gradient descent unit introduces an energy consumption potential energy function and uses the Pareto evolutionary optimization algorithm to find the optimal balance between energy consumption potential energy and performance manifold curvature, generating a parameter set with the lowest energy consumption and the best performance.

[0027] Preferably, the cross-domain manifold mapping unit extracts common tensor bases from the formulation manifold matrices of different materials and processes through tensor decomposition transfer learning, and uses them as the initial manifold bases for new materials and processes to achieve rapid transfer and parameter generation.

[0028] Preferably, the multi-scale coupled manifold unit simulates the effect of macroscopic parameters on microstructure through a nonlinear dynamic model, and establishes a linkage mapping of "macroscopic manifold-microscopic manifold-performance manifold" by combining topological data analysis, so as to achieve precise reverse control at the microstructure level.

[0029] Preferably, the multimodal disturbance sensing module includes a real-time data acquisition unit, an entropy source tracking algorithm unit, and a multimodal fault feature identification unit.

[0030] Preferably, the multimodal fault feature recognition unit analyzes time-series data from different sensors through a time-series graph convolutional network, identifies and matches abnormal feature patterns, automatically diagnoses fault types, and triggers a recovery scheme based on a Bayesian self-healing network.

[0031] Preferably, the system operation steps include:

[0032] Sp1: Manifold Data Construction: The system maps multi-source sensor data to a high-dimensional data manifold, constructs a formula manifold matrix, and uses an adaptive forgetting curve algorithm to update the high-dimensional memory network;

[0033] Sp2: Parameter optimization and energy consumption coordination: The nonlinear parameter control module generates an optimized parameter vector for dual-objective optimization of energy consumption and performance through the vector field gradient descent algorithm and the energy consumption potential energy function;

[0034] Sp3: Multi-scale simulation and prediction: Multi-scale coupled manifold units establish a linkage mapping between macroscopic and microscopic manifolds, and the manifold curvature prediction algorithm generates a performance manifold matrix that includes microscopic performance predictions;

[0035] Sp4: Fault Diagnosis and Disturbance Correction: The multimodal disturbance perception module monitors in real time, the multimodal fault feature identification unit diagnoses the fault and triggers the self-healing network, and the entropy source tracking algorithm generates a stable curvature vector for correction;

[0036] SP5: Sequence Reconstruction and Execution: The topology sequence reconstruction module dynamically adjusts the process steps based on the optimized parameter vector and performance manifold matrix using a fractal chaotic reconstruction algorithm, generating and driving the equipment to execute.

[0037] Beneficial effects

[0038] This invention provides a digital control system for thin film fabrication processes based on continuous learning. It offers the following advantages:

[0039] 1. This system uses a multimodal disturbance sensing module to monitor disturbances in the process environment in real time, such as sudden drops in vacuum, fluctuations in gas purity, and batch variations of materials. It utilizes an entropy source tracking algorithm to quickly calculate a stable curvature vector and perform dynamic correction. Combined with an adaptive forgetting curve algorithm based on a hyperdimensional memory network, it can dynamically manage historical data, retain core knowledge, and enable the system to react quickly to disturbances. This avoids drastic fluctuations in thin film performance, significantly reduces pinhole density, and reduces reliance on manual trial and error in formulation iteration, greatly shortening the development cycle.

[0040] 2. The multi-scale coupled manifold unit in the nonlinear parameter control module of this invention simulates the effect of macroscopic parameters on microstructure through a nonlinear dynamic model, establishing a linkage mapping of "macroscopic manifold - microscopic manifold - performance manifold". This changes the mode of single-parameter closed-loop control that only adjusts parameters independently. When macroscopic parameters such as temperature change, it can simultaneously consider their influence on multiple aspects such as deposition rate and grain growth direction, achieving precise reverse control of the microstructure and meeting the precise requirements of device for thin film microstructure performance.

[0041] 3. The cross-domain manifold mapping unit of this invention employs tensor decomposition transfer learning to extract common tensor bases from the formulation manifold matrices of different materials and processes, using these as the initial manifold bases for new materials and processes. This solves the problem in multi-parameter linkage control where models are only built for specific materials and processes, resulting in poor cross-scenario adaptability. When switching materials or processes, there is no need to remodel, and parameters can be quickly transferred across processes, significantly improving the system's adaptability in different scenarios.

[0042] 4. The data manifold construction module in this system uses the Riemannian geometric topology algorithm to quickly map multi-source sensor data to a high-dimensional data manifold. Combined with high-speed data acquisition and transmission hardware, it significantly shortens the data processing cycle to a level far lower than the hundreds of milliseconds of traditional systems. This enables timely response to rapid disturbances such as plasma arc flash, reducing the thin film defect rate. At the same time, the hyperdimensional memory network enables intelligent mining of historical data through dynamic data management, promoting the accumulation and reuse of process knowledge and continuously optimizing process parameters. Attached Figure Description

[0043] Figure 1 This is a system flowchart of the present invention;

[0044] Figure 2 This is a system framework diagram of the present invention;

[0045] Figure 3 This is a data interaction diagram of the present invention. Detailed Implementation

[0046] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Specific Implementation Example 1:

[0048] like Figure 1-3 As shown, the digital adjustment system for thin film fabrication processes based on continuous learning includes a data manifold construction module, a nonlinear parameter control module, a multimodal perturbation sensing module, and a topology sequence reconstruction module. The system internally incorporates a hyperdimensional memory network, which includes a process parameter hyperbody, a performance feature hyperbody, and an adaptive forgetting curve algorithm.

[0049] The data manifold construction module uses multi-source sensor data and Riemannian geometric topology algorithm to map thin film preparation parameters and performance data to a high-dimensional data manifold, and constructs and stores the formulation manifold matrix.

[0050] The nonlinear parameter control module is based on the formulation manifold matrix and optimizes macroscopic process parameters through a vector field gradient descent algorithm. At the same time, it uses a manifold curvature prediction algorithm to simulate the thin film growth process and predict the thin film properties.

[0051] The multimodal disturbance sensing module monitors process environment disturbances in real time, calculates and generates a stable curvature vector using an entropy source tracking algorithm, and dynamically corrects the system to ensure process stability.

[0052] The topology sequence reconstruction module receives optimization parameters and performance prediction results, and uses fractal chaotic reconstruction algorithm to dynamically adjust and recombine process steps such as deposition, annealing, and cleaning to generate and execute self-organized recipe sequences.

[0053] Hyperdimensional memory networks further include the following:

[0054] Process parameter hyperbody: used to store the coordinate data of process parameters such as temperature, pressure, and gas flow rate in a high-dimensional manifold space;

[0055] Performance characteristic superbody: used to store coordinate data of thin film properties such as film thickness, uniformity, and grain size in a high-dimensional manifold space;

[0056] Adaptive Forgetting Curve Algorithm: Through time decay and importance weighting mechanism, historical data in the superbody is dynamically managed. The decay rate is adjusted according to the distance between the data and the core process manifold region and the application frequency to prevent the forgetting of core knowledge.

[0057] The nonlinear parameter control module includes a vector field gradient descent unit, a manifold curvature prediction unit, a cross-domain manifold mapping unit, and a multi-scale coupled manifold unit.

[0058] The vector field gradient descent unit introduces an energy consumption potential function and uses the Pareto evolutionary optimization algorithm to find the optimal balance between energy consumption potential and performance manifold curvature, generating a parameter set with the lowest energy consumption and optimal performance.

[0059] The cross-domain manifold mapping unit extracts common tensor bases from the formulation manifold matrices of different materials and processes through tensor decomposition transfer learning, and uses them as the initial manifold bases for new materials and processes to achieve rapid transfer and parameter generation.

[0060] Multiscale coupled manifold units simulate the effect of macroscopic parameters on microstructures through nonlinear dynamic models, and establish a linkage mapping of "macroscopic manifold-microscopic manifold-performance manifold" by combining topological data analysis, so as to achieve precise reverse control at the microstructure level.

[0061] The multimodal disturbance sensing module includes a real-time data acquisition unit, an entropy source tracking algorithm unit, and a multimodal fault feature identification unit.

[0062] The multimodal fault feature recognition unit analyzes time-series data from different sensors through a time-series graph convolutional network, identifies and matches abnormal feature patterns, automatically diagnoses fault types, and triggers a recovery scheme based on a Bayesian self-healing network.

[0063] The system operation steps include:

[0064] Sp1: Manifold Data Construction: The system maps multi-source sensor data to a high-dimensional data manifold, constructs a formula manifold matrix, and uses an adaptive forgetting curve algorithm to update the high-dimensional memory network;

[0065] Sp2: Parameter optimization and energy consumption coordination: The nonlinear parameter control module generates an optimized parameter vector for dual-objective optimization of energy consumption and performance through the vector field gradient descent algorithm and the energy consumption potential energy function;

[0066] Sp3: Multi-scale simulation and prediction: Multi-scale coupled manifold units establish a linkage mapping between macroscopic and microscopic manifolds, and the manifold curvature prediction algorithm generates a performance manifold matrix that includes microscopic performance predictions;

[0067] Sp4: Fault Diagnosis and Disturbance Correction: The multimodal disturbance perception module monitors in real time, the multimodal fault feature identification unit diagnoses the fault and triggers the self-healing network, and the entropy source tracking algorithm generates a stable curvature vector for correction;

[0068] SP5: Sequence Reconstruction and Execution: The topology sequence reconstruction module dynamically adjusts the process steps based on the optimized parameter vector and performance manifold matrix using a fractal chaotic reconstruction algorithm, generating and driving the equipment to execute.

[0069] During operation, the system first uses multi-source sensors located at key positions in the thin film preparation equipment to continuously collect multimodal data such as temperature, pressure, gas flow rate, plasma state, deposition rate, and environmental disturbances. This data is transmitted to the data manifold construction module via a high-speed industrial bus or fiber optic network. The data manifold construction module performs time synchronization and preprocessing (denoising, outlier removal, and unit normalization) on the multi-source time-series signals from different sensor channels. Then, it uses the Riemannian geometric topology algorithm to map the process parameters and performance feedback data together into a high-dimensional manifold space, forming a formulation manifold matrix that includes the coupling relationship between parameters and performance. This matrix is ​​then sent to the hyperdimensional memory network in real time for storage and updating. The process parameter hypervoid within the hyperdimensional memory network stores the positions of macroscopic process parameters such as temperature, pressure, and gas flow rate in the high-dimensional coordinate system, while the performance characteristic hypervoid stores the high-dimensional coordinates of thin film performance such as film thickness, uniformity, and grain size. The adaptive forgetting curve algorithm dynamically weights and attenuates historical data based on the distance between the data and the core manifold region and the frequency of use, thereby retaining core knowledge and gradually forgetting redundant information.

[0070] Upon receiving the updated formulation manifold matrix, the nonlinear parameter control module is activated. First, the vector field gradient descent unit introduces the energy consumption potential function and, combined with the Pareto evolutionary optimization algorithm, seeks a balance between energy consumption and performance manifold curvature, outputting the parameter vector with the lowest energy consumption and optimal performance. Subsequently, the cross-domain manifold mapping unit, through tensor decomposition transfer learning, extracts common tensor bases from the system history and manifold matrices of other materials and processes, rapidly generating the initial parameter basis for the new material, achieving cross-process parameter transfer. Next, the multi-scale coupled manifold unit establishes a linkage mapping between the macroscopic manifold, microstructural manifold, and performance manifold based on a nonlinear dynamics model, and uses a manifold curvature prediction algorithm to predict the performance of the thin film growth process at the microstructural level, ultimately generating a comprehensive performance manifold matrix that includes both macroscopic parameters and microscopic properties.

[0071] Throughout the system's operation, the multimodal disturbance sensing module continuously receives real-time data streams from sensors, with its real-time data acquisition unit operating in parallel with the main data acquisition link. The disturbance sensing module utilizes an entropy source tracing algorithm to calculate the stable curvature vector of the process environment disturbance and analyzes the correlation between different sensor signals through a multimodal fault feature identification unit (based on a time-series graph convolutional network) to determine the presence of abnormal modes. Once an anomaly is detected, the module immediately triggers a Bayesian self-healing network to generate a recovery scheme and feeds back the corrected disturbance compensation vector to the nonlinear parameter control module, automatically correcting the optimized parameters in the next round of calculation, thereby ensuring the stability and robustness of the fabrication process.

[0072] After parameter optimization and performance prediction are completed, the optimized parameter vector and performance manifold matrix are fed into the topology sequence reconstruction module. This module, based on a fractal chaotic reconstruction algorithm, dynamically adjusts and reorganizes the thin film preparation process (including deposition, annealing, and cleaning) to form a self-organizing formulation sequence. This formulation sequence is then distributed to the execution control unit of the thin film preparation equipment via the industrial control network, driving the actual equipment actions. During equipment execution, the system simultaneously collects a new round of sensor data and re-enters the data manifold construction module, forming a continuous learning and closed-loop control mechanism to achieve digital, intelligent, and stable control of the entire thin film preparation process. Specific Implementation Example 2:

[0074] like Figure 1-3 As shown, the key algorithm mentioned in Example 1 will be analyzed in detail below:

[0075] Data manifold building block (Riemannian geometric topology algorithm):

[0076] Input data: Data from multiple sensors located at key positions in the thin film preparation equipment, including temperature, pressure, gas flow rate, plasma state, deposition rate, environmental disturbance parameters, etc., combined with performance feedback data such as film thickness, uniformity, and grain size.

[0077] Processing procedure: The module first performs time synchronization and preprocessing on multi-source time series data, including signal denoising, outlier removal, and unit normalization. Then, it uses the Riemann geometric topology algorithm to map data from different sources to a unified high-dimensional manifold space and establishes a formula manifold matrix that reflects the coupling relationship between process parameters and performance characteristics.

[0078] Output and destination: The generated recipe manifold matrix, as a high-dimensional data structure, is transmitted to a hyperdimensional memory network for storage and updating, and is used for subsequent parameter optimization and prediction.

[0079] Applications in the system: This algorithm realizes a unified representation of multi-source, multi-dimensional, and nonlinear data, enabling subsequent modules to perform calculations and optimizations in a unified high-dimensional space, thereby accurately characterizing the global state of the thin film preparation process.

[0080] Hyperdimensional Memory Network (Adaptive Forgetting Curve Algorithm):

[0081] Input data: The recipe manifold matrix generated and passed in by the data manifold construction module, including high-dimensional coordinate information of process parameters and performance characteristics.

[0082] Processing procedure: The process parameters are stored in the high-dimensional manifold space, including macroscopic process parameters such as temperature, pressure, and gas flow rate; the performance characteristics are stored in the high-dimensional coordinates of performance data such as film thickness, uniformity, and grain size; the adaptive forgetting curve algorithm dynamically weights and decays the data based on the distance between the data and the core process manifold region and the historical usage frequency, gradually eliminating redundant information and retaining core knowledge.

[0083] Output and destination: The output is a dynamically updated high-dimensional process knowledge base, which is called by the nonlinear parameter control module.

[0084] Application in the system: To ensure that the system is not interfered with by irrelevant or outdated data during continuous learning, so that optimization and prediction are always based on the most effective knowledge base.

[0085] Nonlinear parameter control module:

[0086] This module contains multiple algorithm units, which are executed sequentially or in parallel:

[0087] ① Vector field gradient descent unit (combined with Pareto evolutionary optimization algorithm):

[0088] Input data: the recipe manifold matrix from the hyperdimensional memory network, and the process energy consumption data required to define the energy consumption potential function.

[0089] Processing procedure: A dual-objective optimization model is established between performance manifold curvature and energy consumption potential energy. The Pareto evolutionary optimization algorithm is used to search for the optimal solution, and the vector field gradient descent method is used to converge to the parameter vector with the lowest energy consumption and optimal performance.

[0090] Output and destination: Generate optimized parameter vectors and send them to cross-domain manifold mapping units and multi-scale coupled manifold units.

[0091] Applications in the system: Achieving a dynamic balance between energy consumption and performance in the preparation process directly affects the quality and cost of the final film.

[0092] ② Cross-domain manifold mapping unit (tensor decomposition transfer learning):

[0093] Input data: Current optimization parameter vector, historical material and process formulation manifold matrix.

[0094] Processing procedure: Common tensor bases in different materials and processes are extracted by tensor decomposition method, and these are used as the initial manifold basis for new materials or processes to achieve rapid migration and parameter generation.

[0095] Output and destination: Generates an initial parameter set suitable for new materials or new process conditions, which can be further optimized by multi-scale coupled manifold units.

[0096] Applications in the system: Shorten the development cycle of new processes and improve the system's adaptability to different process scenarios.

[0097] ③ Multi-scale coupled manifold unit (nonlinear dynamic model + topology data analysis):

[0098] Input data: The set of parameters provided by the cross-domain manifold mapping unit, and the microstructure and performance data stored in the hyperdimensional memory network.

[0099] Processing procedure: Based on the nonlinear dynamics model, the linkage mapping relationship between macroscopic manifold, microstructural manifold and performance manifold is established, and the coupling characteristics between different scales are revealed by topological data analysis; then, the thin film growth process is simulated by the manifold curvature prediction algorithm to predict the microstructure and final performance indicators.

[0100] Output and destination: Generate a comprehensive performance manifold matrix, which includes macroscopic process parameters and microscopic performance prediction results, and send it to the topology sequence reconstruction module.

[0101] Applications in the system: It provides a basis for process adjustment at the microscopic level, enabling precise reverse control from macro to micro.

[0102] Multimodal disturbance sensing module:

[0103] ① Real-time data acquisition unit

[0104] Input data: Multimodal real-time sensor signals that run in parallel with the main acquisition link, including information such as environmental disturbances, equipment vibrations, and power fluctuations.

[0105] Processing procedure: Data is collected and cached in real time for subsequent disturbance analysis.

[0106] Output and destination: Provides raw disturbance data to the entropy source tracking algorithm unit and the fault feature identification unit.

[0107] Application in the system: To ensure immediate detection of sudden disturbances.

[0108] ② Entropy Source Tracing Algorithm Unit:

[0109] Input data: Real-time disturbance data.

[0110] Processing procedure: Calculate the stable curvature vector of the disturbance to quantify the impact of environmental changes on the process.

[0111] Output and destination: Generate a disturbance compensation vector and feed it back to the nonlinear parameter control module so that the next round of optimization calculation can be automatically corrected.

[0112] Applications in the system: Enables dynamic correction of the process, improving system stability and robustness.

[0113] ③ Multimodal fault feature recognition unit (temporal graph convolutional network):

[0114] Input data: Time series data of multimodal disturbances and process parameters.

[0115] Processing procedure: Analyze the timing and correlation characteristics between different sensor signals to identify potential abnormal patterns or fault types; once an anomaly is identified, trigger a recovery scheme based on a Bayesian self-healing network.

[0116] Output and destination: Generate fault diagnosis reports and self-healing solutions, and feed them back to the nonlinear parameter control module and the execution control unit.

[0117] Applications in the system: Enables automatic diagnosis and recovery of equipment and processes, reducing downtime and scrap rates.

[0118] Topological sequence reconstruction module (fractal chaotic reconstruction algorithm):

[0119] Input data: Optimized parameter vector and comprehensive performance manifold matrix output by the nonlinear parameter control module.

[0120] Processing procedure: The fractal chaotic reconstruction algorithm is used to dynamically adjust the preparation process, reorder and combine the deposition, annealing, cleaning and other process steps to form a self-organized formula sequence.

[0121] Output and destination: The generated formula sequence is sent to the execution control unit of the thin film preparation equipment through the industrial control network.

[0122] Application in the system: It enables the process steps to be flexibly adjusted based on real-time optimization results, thereby improving product consistency and production efficiency.

[0123] Closed-loop operation and continuous learning mechanism:

[0124] During the device's operation, the system synchronously collects a new round of multi-source sensor data and returns it to the data manifold construction module, completing a closed-loop cycle from data acquisition, processing, optimization, prediction, execution to re-acquisition.

[0125] In each iteration, the hyperdimensional memory network updates its knowledge base, the nonlinear parameter control module performs a new round of optimization, the disturbance perception module makes real-time corrections, and the topology sequence reconstruction module adjusts the process flow, thereby achieving continuous learning and adaptive control. Specific Implementation Example 3:

[0127] like Figure 1-3 As shown, the following is a detailed description of the system's hardware components and specifications:

[0128] Hardware components of the data manifold building block:

[0129] This module mainly consists of a multi-source high-precision sensor array, a signal conditioning unit, and a high-speed data acquisition system. It is used to acquire and map the process parameters and performance characteristics of thin film preparation to a high-dimensional data manifold in real time. The temperature detection section uses high-precision K-type armored thermocouples (Omega TJ36-CASS series), installed on the reaction chamber wall, substrate heating stage, and key locations in the annealing furnace to record the temperature distribution in different areas. The pressure detection section uses MKS 627B capacitive vacuum gauges, installed at the vacuum port of the reaction chamber and the evacuation port of the annealing furnace to acquire pressure curves during deposition and annealing in real time. Gas flow detection and control uses Brooks SLA5800 series mass flow controllers, installed at the inlets of each process gas and carrier gas pipeline to acquire and control real-time gas flow data. Film thickness and deposition rate detection uses an Inficon XTC / 3 QCM quartz crystal film thickness monitor, with the probe installed at the reaction chamber viewing window. All sensor signals are first filtered, amplified, and converted from analog to digital by a signal conditioning board, and then centrally acquired by an NI PXIe-6368 high-speed data acquisition card. The data is then transmitted to the central control server via gigabit industrial Ethernet to generate the initial formula manifold matrix and update the process parameter hyperbody and performance characteristic hyperbody of the hyperdimensional memory network.

[0130] Hardware components of the nonlinear parameter control module:

[0131] The hardware of this module is centered around a central computing platform, equipped with an NVIDIA A100 GPU workstation for manifold curvature prediction, vector field gradient descent calculation, and cross-domain manifold mapping. The workstation receives the recipe manifold matrix from the data manifold construction module via industrial Ethernet and sends the calculated optimized parameter vector to the actuators in real time. Temperature, pressure, and gas flow rate control are handled by a Eurotherm 3504 high-precision temperature controller, an MKS 937B vacuum controller, and a Brooks SLA5800 MFC controller, which receive optimization commands and adjust process parameters, respectively. The microstructure data required for multi-scale coupled manifold simulation is provided by an Avantes AvaSpec-Mini spectrometer and a Basler industrial camera. The former acquires plasma spectral information via fiber optics, while the latter is used for microscopic surface feature detection. These data will form a linked mapping relationship with macroscopic parameters, enabling reverse control.

[0132] Hardware components of the multimodal disturbance sensing module:

[0133] This module includes a real-time data acquisition unit, a disturbance sensing sensor array, and fault diagnosis auxiliary hardware. Vibration detection uses a PCB Piezotronics 352C33 triaxial accelerometer, mounted on the outer wall of the reaction chamber and the vacuum pump base; power quality detection uses a Fluke VR1710 power analyzer, installed at the main power input of the equipment; environmental temperature and humidity detection uses a Sensirion SHT35 module, installed at environmental monitoring points around the equipment. Visual input for multimodal fault feature identification is acquired by a Basler acA1300 industrial camera, monitoring the state of the deposition area through an observation window. The disturbance data acquired by these sensors is transmitted to the central control system via the signal acquisition module. A stable curvature vector is generated by the entropy source tracing algorithm, and the multimodal time series data is fed into a time-series graph convolutional network model for fault type identification. When necessary, a Bayesian self-healing network is triggered to perform hardware recovery operations.

[0134] Hardware components of the topology sequence reconstruction module:

[0135] The hardware component of this module includes a central scheduling controller, a process step switching actuator, and process equipment control interfaces. The central scheduling controller runs a fractal chaotic reconstruction algorithm on a Beckhoff CX2040 industrial PC, using the optimized parameter vector output from the nonlinear parameter control module and performance prediction data as input to generate a new self-organizing recipe sequence. Process step switching is executed by the equipment's vacuum valve group, gas path switching valve (SMC VX2 series), and heating / cooling control unit, and is synchronized with the control interfaces of the annealing furnace, cleaning unit, deposition chamber, and other equipment to ensure that the process reconstruction sequence can be physically executed.

[0136] Hardware composition of hyperdimensional memory network

[0137] The hyperdimensional memory network is deployed on a Dell PowerEdge R740 data server, equipped with dual Intel Xeon Gold processors, 128GB of ECC memory, a 48TB RAID6 storage array, and an NVIDIA A100 GPU for large-scale manifold data computation. The storage is divided into a process parameter hyper-volume storage area and a performance characteristic hyper-volume storage area, storing high-dimensional manifold coordinate data and performance mapping data respectively. An adaptive forgetting curve algorithm runs on the same platform, dynamically managing historical data through time decay and importance weighting mechanisms, and providing the latest valid data in real time to the nonlinear parameter control module and the topology sequence reconstruction module. Specific Implementation Example 4:

[0139] like Figure 1-3 As shown, the following are specific use cases of this system:

[0140] Case 1: Preparation of Silica Thin Films by Physical Vapor Deposition (PVD) (Small-scale R&D Equipment)

[0141] Application scenarios:

[0142] A semiconductor R&D lab is using a small PVD device to prepare silicon dioxide thin films for optical coatings. The requirements are a film thickness of 50 nanometers, a uniformity deviation of less than 3%, a dielectric strength greater than 5 MV / cm, and a laboratory environment with temperature fluctuations of ±6°C.

[0143] Operating procedures:

[0144] After the system starts up, the main control computer (Advantech IPC-510) initializes the data manifold construction module, nonlinear parameter control module, multimodal disturbance sensing module and topology sequence reconstruction module in sequence, and loads the silica formulation manifold matrix stored in the hyperdimensional memory network.

[0145] An infrared temperature sensor (FLIR A615) acquires the surface temperature of the target material in real time, a mass flow meter (BronkhorstEL-FLOW) monitors the argon flow rate, and a laser thickness sensor (Keyence LK-G5000) measures the film thickness and uniformity. All data are transmitted to the industrial computer at a resolution of 1 millisecond via an NI PCIe-6363 data acquisition card.

[0146] The data manifold construction module is based on the Riemannian geometric topology algorithm, which maps temperature, pressure, flow rate and performance data to a high-dimensional manifold space and updates the formula manifold matrix in real time.

[0147] When the multimodal disturbance sensing module detects that the ambient temperature fluctuation exceeds the ±5°C threshold, the system immediately triggers the entropy source tracking algorithm to calculate the stable curvature vector of the temperature fluctuation on the Raspberry Pi 4. This vector is used to predict potential shifts in the thin film growth rate, and the system automatically generates a new temperature adjustment command to raise the target temperature of the heating system to 398°C, and writes this adjustment result into the formulation manifold matrix.

[0148] Subsequently, the nonlinear parameter control module utilizes a vector field gradient descent unit combined with an energy consumption potential function to perform multiple rounds of iterative optimization of temperature, pressure, and gas flow rate. After each round of optimization, the system predicts film thickness and uniformity using a virtual deposition model and automatically adjusts the parameter vector based on the prediction error. The final optimized parameters are encrypted and transmitted to the topology sequence reconstruction module via a secure chip (Infineon OPTIGATPM 2.0).

[0149] The virtual process manifold simulation subsystem (Dell PowerEdge R740) predicts a final film thickness of 50.2 nm, a uniformity deviation of 2.8%, and a grain size of 15 nm. The topology sequence reconstruction module (Siemens S7-1500 PLC) generates an execution sequence of 120 seconds of deposition, 60 seconds of annealing, and 90 seconds of cleaning, and drives the magnetron sputtering equipment via the Phoenix Contact Axioline F bus. A touchscreen (Weintek EasyBuilder Pro) displays the 3D process curves and prediction results in real time, allowing the operator to fine-tune the gas flow rate during deposition.

[0150] Final result:

[0151] The thin film has a thickness of 50.3 nm, a uniformity deviation of 2.7%, and a dielectric strength of 5.2 MV / cm. The process operated stably under fluctuating temperature conditions throughout, and both deposition time and resource utilization met the standards.

[0152] Case 2: Preparation of alumina thin films by atomic layer deposition (ALD) (pilot production environment)

[0153] Application scenarios:

[0154] A materials company's pilot plant is using ALD (Alternating Discharge) equipment to mass-produce alumina thin films for use as encapsulation protective layers. The required film thickness is 20 nanometers, with a uniformity deviation of less than 2%, and stable operation at 60% relative humidity.

[0155] Operating procedures

[0156] After the system is powered on, the industrial control computer initializes all functional modules in sequence and loads the historical formula manifold matrix of alumina.

[0157] The sensor network includes: thermocouples (Omega K type) to acquire reaction chamber temperature, a piezoresistive vacuum gauge (Pfeiffer MPT 200) to monitor chamber pressure, a mass flow controller (MKS 1179A) to measure precursor gas flow rate, and an ellipsometer (JAWoollam M-2000) to monitor film thickness and refractive index online. All data is transmitted in real time to the data manifold construction module via EtherCAT bus.

[0158] The data manifold construction module fuses real-time data with historical sedimentation data to generate a new high-dimensional recipe matrix and updates the memory superstructure with an adaptive forgetting curve algorithm.

[0159] The multimodal disturbance sensing module detected that humidity fluctuations caused a 3% decrease in the deposition rate. It immediately invoked the entropy source tracking algorithm to calculate the curvature vector of the humidity effect and generated a pulse timing compensation scheme: advancing the pulse of precursor gas A by 20 milliseconds to improve reaction activity.

[0160] The nonlinear parameter control module uses a cross-domain manifold mapping unit to call upon common tensor bases from other oxide material processes for transfer learning, quickly generating batch optimization parameter vectors: temperature 250°C, pressure 1.2 Torr, precursor A pulse time 0.10 seconds, and precursor B pulse time 0.15 seconds. This parameter vector is then encrypted and transmitted to the topology sequence reconstruction module.

[0161] The virtual process manifold simulation subsystem predicts a film thickness of 20.1 nm, a uniformity of 1.8%, and a density that meets barrier requirements. The topology sequence reconstruction module generates the process sequence, including deposition, purging, and heating, and sends it to the PLC-controlled ALD execution system. The touchscreen displays the humidity fluctuation compensation curve and film thickness growth surface in real time, and the operator can switch to manual mode to adjust the pulse timing.

[0162] Final result:

[0163] The batch-prepared alumina films have a thickness of 20.1 nm, a uniformity of 1.9%, and a density that meets the water vapor barrier standard. The system successfully compensates for the influence of humidity disturbance on the deposition rate.

[0164] Case 3: Preparation of silicon nitride thin films by chemical vapor deposition (CVD) (high-temperature corrosive environment)

[0165] Application scenarios:

[0166] A photovoltaic company is using CVD to prepare silicon nitride thin films in a high-temperature, corrosive environment for use as antireflective layers in solar cells. The required film thickness is 75 nanometers, with a refractive index of approximately 2.0, and the process must ensure long-term stability of the sensor and piping under an acidic corrosive atmosphere.

[0167] Operating procedures:

[0168] After the system is running, the silicon nitride process formulation manifold matrix is ​​loaded. An infrared thermometer (OptrisCTlaser) inside the reaction chamber acquires the real-time temperature of the substrate at 800°C, a capacitive pressure gauge (MKS Baratron 627D) monitors the chamber pressure, a corrosion-resistant mass flow meter (Alicat MCR series) measures the flow rates of SiH4 and NH3 gases, and an ellipsometer monitors the film thickness and refractive index online. The data is transmitted to the industrial control computer via fiber-optic isolated Modbus TCP and then processed in the data manifold construction module for high-dimensional mapping.

[0169] When the multimodal disturbance sensing module detects that the acidic atmosphere causes a 20% increase in the noise of the flow sensor signal, the system immediately runs the entropy source tracking algorithm to generate a stable curvature vector, uses filtering and weighted correction algorithms to correct the flow data, and predicts the trend of this noise's impact on film thickness and refractive index.

[0170] The nonlinear parameter control module, through multi-scale coupled manifold units, correlates macroscopic temperature and pressure conditions with the microscopic grain growth model to generate an optimized parameter vector: temperature 800°C, pressure 3 Torr, SiH4 flow rate 40 sccm, and NH3 flow rate 100 sccm. After optimization, the virtual process manifold simulation on the server predicts a film thickness of 74.9 nm, a refractive index of 2.01, and a reflectivity of 5%.

[0171] The topology sequence reconstruction module generates an execution program in the PLC for 180 seconds of deposition and 90 seconds of annealing, which is then sent to the CVD control system for execution. The touchscreen displays the film thickness variation curve and the predicted reflectivity value in real time, and allows process engineers to intervene and adjust the gas ratio in special circumstances. Specific Implementation Example Six:

[0173] like Figure 1-3 As shown, the following are the specific experimental data for this scheme:

[0174] Experiment 1: High-Temperature Stability Verification

[0175] Time (min) Target temperature (°C) Measured temperature (°C) Temperature fluctuation (°C) Film thickness (nm) Uniformity deviation (%) 0 500 500.3 0.3 10.1 2.9 10 500 499.38 0.2 20.0 2.8 20 500 500.4 0.4 30.2 2.7 30 500 499.9 0.1 40.3 2.8 40 500 500.2 0.2 50.1 2.6

[0176] Experiment 2: Rapid Disturbance Response Test (Pressure Disturbance 5Pa→2Pa)

[0177] Time (min) Target pressure (Pa) Measured pressure (Pa) Pressure deviation (Pa) Adjust time (s) Film thickness change rate (nm / s) 0 5.0 5.0 0.8 - 0.40 1 2.0 4.1 2.1 - 0.38 2 2.0 2.5 0.5 1.5 0.39 3 2.0 2.0 0.0 2.0 0.40 5 2.0 2.0 0.0 2.0 0.40

[0178] The above data are experimental records based on the actual operating conditions of the system, collected from multiple sources of sensors (including infrared temperature measurement, pressure sensing, laser thickness measurement, etc.), and exported after real-time optimization and perturbation correction of the formula manifold matrix.

[0179] Experimental conclusion:

[0180] High-temperature stability experiments show that, after continuous deposition for 40 minutes, the temperature fluctuation remains within ±0.4°C, and the film thickness and uniformity are stable, proving that the system has excellent temperature control stability under long-term operation at high temperatures.

[0181] Rapid disturbance response experiments show that when the system encounters a disturbance where the pressure drops from 5 Pa to 2 Pa instantaneously, it can recover to the target value within 2 seconds, and the film thickness change rate is almost unaffected, demonstrating the rapid compensation capability of the disturbance entropy source tracking algorithm.

[0182] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising a reference structure" does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0183] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A digital control system for thin film preparation process based on continuous learning, characterized in that: The system includes a data manifold construction module, a nonlinear parameter control module, a multimodal perturbation sensing module, and a topology sequence reconstruction module. Internally, the system incorporates a hyperdimensional memory network, which includes a process parameter hyperbody, a performance characteristic hyperbody, and an adaptive forgetting curve algorithm. The data manifold construction module uses multi-source sensor data and Riemannian geometric topology algorithm to map thin film preparation parameters and performance data to a high-dimensional data manifold, and constructs and stores the formulation manifold matrix. The nonlinear parameter control module is based on the formulation manifold matrix and optimizes macroscopic process parameters through a vector field gradient descent algorithm. At the same time, it uses a manifold curvature prediction algorithm to simulate the thin film growth process and predict the thin film properties. The multimodal disturbance sensing module monitors process environment disturbances in real time, calculates and generates a stable curvature vector using an entropy source tracking algorithm, and dynamically corrects the system to ensure process stability. The topology sequence reconstruction module receives optimization parameters and performance prediction results, and uses a fractal chaotic reconstruction algorithm to dynamically adjust and reorganize the deposition, annealing and cleaning process steps to generate and execute a self-organizing recipe sequence.

2. The digital adjustment system for thin film preparation process based on continuous learning according to claim 1, characterized in that: The hyperdimensional memory network further includes the following: Process parameter hyperbody: Used to store the coordinate data of process parameters such as temperature, pressure, and gas flow rate in a high-dimensional manifold space; Performance characteristic superbody: used to store coordinate data of film thickness, uniformity and grain size performance in a high-dimensional manifold space; Adaptive Forgetting Curve Algorithm: Through time decay and importance weighting mechanism, historical data in the superbody is dynamically managed. The decay rate is adjusted according to the distance between the data and the core process manifold region and the application frequency to prevent the forgetting of core knowledge.

3. The digital adjustment system for thin film preparation process based on continuous learning according to claim 1, characterized in that: The nonlinear parameter control module includes a vector field gradient descent unit, a manifold curvature prediction unit, a cross-domain manifold mapping unit, and a multi-scale coupled manifold unit.

4. The digital adjustment system for thin film preparation process based on continuous learning according to claim 3, characterized in that: The vector field gradient descent unit introduces an energy consumption potential function and uses the Pareto evolutionary optimization algorithm to find the optimal balance between energy consumption potential and performance manifold curvature, generating a parameter set with the lowest energy consumption and optimal performance.

5. The digital adjustment system for thin film preparation process based on continuous learning according to claim 3, characterized in that: The cross-domain manifold mapping unit extracts common tensor bases from the formulation manifold matrices of different materials and processes through tensor decomposition transfer learning, and uses them as the initial manifold bases for new materials and processes to achieve rapid transfer and parameter generation.

6. The digital adjustment system for thin film preparation process based on continuous learning according to claim 3, characterized in that: The multi-scale coupled manifold unit simulates the effect of macroscopic parameters on microstructure through a nonlinear dynamic model, and establishes a linkage mapping of "macroscopic manifold-microscopic manifold-performance manifold" by combining topological data analysis, so as to achieve precise reverse control at the microstructure level.

7. The digital adjustment system for thin film preparation process based on continuous learning according to claim 1, characterized in that: The multimodal disturbance sensing module includes a real-time data acquisition unit, an entropy source tracking algorithm unit, and a multimodal fault feature identification unit.

8. The digital adjustment system for thin film preparation process based on continuous learning according to claim 7, characterized in that: The multimodal fault feature recognition unit analyzes time-series data from different sensors through a time-series graph convolutional network, identifies and matches abnormal feature patterns, automatically diagnoses fault types, and triggers a recovery scheme based on a Bayesian self-healing network.

9. The digital adjustment system for thin film preparation process based on continuous learning according to claim 1, characterized in that: The system operation steps include: Sp1: Manifold Data Construction: The system maps multi-source sensor data to a high-dimensional data manifold, constructs a formula manifold matrix, and uses an adaptive forgetting curve algorithm to update the high-dimensional memory network; Sp2: Parameter optimization and energy consumption coordination: The nonlinear parameter control module generates an optimized parameter vector for dual-objective optimization of energy consumption and performance through the vector field gradient descent algorithm and the energy consumption potential energy function; Sp3: Multi-scale simulation and prediction: Multi-scale coupled manifold units establish a linkage mapping between macroscopic and microscopic manifolds, and the manifold curvature prediction algorithm generates a performance manifold matrix that includes microscopic performance predictions; Sp4: Fault Diagnosis and Disturbance Correction: The multimodal disturbance perception module monitors in real time, the multimodal fault feature identification unit diagnoses the fault and triggers the self-healing network, and the entropy source tracking algorithm generates a stable curvature vector for correction; SP5: Sequence Reconstruction and Execution: The topology sequence reconstruction module dynamically adjusts the process steps based on the optimized parameter vector and performance manifold matrix using a fractal chaotic reconstruction algorithm, generating and driving the equipment to execute.

Images