Ion source beam adaptive method, apparatus, device, and storage medium

Abstract

This invention relates to the field of ion source beam technology, and particularly to an adaptive method, apparatus, device, and storage medium for ion source beams. The method first constructs an initial state space model, then obtains the current period input vector, the current period beam measurement value corresponding to the current period input vector, the current period state vector, and a candidate input sequence set. Based on the current period input vector and the current period beam measurement value, the physical parameter vector of the initial state space model is updated to obtain an updated state space model. Next, based on the updated state space model, a target control sequence is determined from the candidate input sequence set according to the current period state vector and the candidate input sequence set. Finally, a target control command is determined based on the target control sequence and output to the execution power supply of the ion source. This aims to suppress beam drift, improve the long-term dynamic stability of beam output, and adapt to the computing power requirements of industrial sites.

Description

Technical Field

[0001] This invention relates to the field of ion source beam technology, and in particular to an adaptive ion source beam method, apparatus, device, and storage medium. Background Technology

[0002] Ion implantation is a key doping process in semiconductor chip manufacturing. With the mass production of third-generation semiconductor devices such as silicon carbide, stringent requirements have been placed on the long-term stability of the ion implantation process. As the core component of an ion implanter, the long-term stability of the beam output of the ion source has become a major bottleneck restricting the performance of high-end equipment.

[0003] Currently, the technical solutions for improving the control performance of ion sources are mainly divided into two categories. One category is the classical control method based on static operating point tuning. This type of method mostly adopts proportional-integral-derivative control or a strategy combined with feedforward compensation. The controller parameters are tuned and fixed by the static or quasi-static model under specific operating conditions of the ion source. This method cannot characterize and compensate for the inherent dynamic inertia, nonlinear saturation, and hysteresis effects in the multi-physics coupling process inside the ion source, resulting in slow response or even instability when the system switches processes or encounters disturbances. It also cannot automatically adapt to slow time-varying characteristics such as cathode loss, small changes in gas flow, and coil thermal drift, causing unacceptable drift in the long-term operation of the beam. It is only suitable for short-term, constant operating conditions in laboratory environments. Another type is advanced control attempts based on data-driven modeling, which use deep neural network models to predict and optimize plasma process parameters. Although this avoids the difficulties of modeling complex mechanisms, the interpretability and credibility of the model are insufficient. When data is scarce or the operating conditions exceed the range of the training set, the behavior is difficult to predict. At the same time, the online adaptability is weak, the real-time update of complex model parameters is difficult, the huge amount of computation restricts the real-time performance of the control system, and the generalization ability and robustness are insufficient due to the neglect of physical mechanisms, making it difficult to cope with new processes or changes in equipment status. Summary of the Invention

[0004] In order to overcome the shortcomings of the prior art, the present invention aims to provide an ion source beam adaptive method, device, equipment and storage medium, which aims to suppress beam drift, improve the long-term dynamic stability of beam output and adapt to the computing power requirements of industrial sites.

[0005] The first aspect of this invention provides an adaptive beam method for an ion source, comprising: constructing an initial state space model; obtaining a current period input vector, a current period beam measurement value corresponding to the current period input vector, a current period state vector, and a candidate input sequence set; updating the physical parameter vector of the initial state space model based on the current period input vector and the current period beam measurement value to obtain an updated state space model; determining a target control sequence from the candidate input sequence set based on the updated state space model, according to the current period state vector and the candidate input sequence set; determining a target control command based on the target control sequence, and outputting the target control command to the execution power supply of the ion source.

[0006] Optionally, in a first implementation of the first aspect of the present invention, constructing the initial state space model includes: defining a state vector, an input vector, and a physical parameter vector of the initial state space model; constructing a state equation based on the law of conservation of plasma particle number, the law of conservation of energy, the state vector, the input vector, and the physical parameter vector; constructing an output equation based on the law of space charge confinement current, the state vector, the input vector, and the physical parameter vector; and integrating the state equation and the output equation to construct the initial state space model.

[0007] Optionally, in a second implementation of the first aspect of the present invention, the step of updating the physical parameter vector of the initial state space model based on the current period input vector and the current period beam measurement value to obtain the updated state space model includes: constructing a current period regression vector based on the current period input vector and the current period beam measurement value; obtaining the previous period covariance matrix and a preset forgetting factor; using an adaptive estimation algorithm to calculate the current period gain matrix based on the current period regression vector, the previous period covariance matrix and the forgetting factor; and updating the physical parameter vector of the initial state space model based on the current period gain matrix and the current period beam measurement value to obtain the updated state space model.

[0008] Optionally, in a third implementation of the first aspect of the present invention, the step of determining the target control sequence from the candidate input sequence set based on the updated state space model, according to the current period state vector and the candidate input sequence set, includes: performing beam prediction based on the updated state space model, according to the current period state vector and the candidate input sequence set, to obtain a set of beam prediction values ​​within a preset future period; and solving an optimization function based on the set of beam prediction values ​​to determine the target control sequence from the candidate input sequence set.

[0009] Optionally, in a fourth implementation of the first aspect of the present invention, the optimization function includes a beam tracking error sum of squares term and a control quantity change weighting term; the beam prediction value set includes multiple beam prediction value subsets, and the candidate input sequence set includes candidate input sequences corresponding one-to-one with each of the beam prediction value subsets; the step of solving the optimization function based on the beam prediction value set to determine the target control sequence from the candidate input sequence set includes: calculating the beam tracking error sum of squares term and the control quantity change weighting term based on each of the beam prediction value subsets and the corresponding candidate input sequences; performing a weighted summation of each beam tracking error sum of squares term and the corresponding control quantity change weighting term to obtain an optimization function value corresponding to each of the candidate input sequences; and selecting the candidate input sequence with the smallest optimization function value as the target control sequence.

[0010] Optionally, in a fifth implementation of the first aspect of the present invention, the step of calculating the sum of squares of the beam tracking error and the weighted term of the control quantity change based on each subset of beam prediction values ​​and the corresponding candidate input sequence includes: obtaining a preset control quantity change weight matrix and a preset beam setting value; calculating the sum of squares of the deviations of the beam prediction values ​​based on each subset of beam prediction values ​​and the beam setting value to obtain a sum of squares of beam tracking error corresponding to each candidate input sequence; calculating the input vector change in adjacent control cycles based on each candidate input sequence, and calculating the weighted term of the control quantity change based on each input vector change and the control quantity change weight matrix to obtain a weighted term of the control quantity change corresponding to each sum of squares of beam tracking error.

[0011] Optionally, in a sixth implementation of the first aspect of the present invention, determining the target control instruction based on the target control sequence includes: acquiring the first-step control quantity in the time domain and a preset safety constraint rule in the target control sequence; performing a validity check on the first-step control quantity in the time domain based on the safety constraint rule; and when the check passes, using the first-step control quantity in the time domain as the target control instruction.

[0012] A second aspect of the present invention provides an ion source beam adaptive device, comprising: a model building module for building an initial state space model; a data acquisition module for acquiring a current period input vector, a current period beam measurement value corresponding to the current period input vector, a current period state vector, and a candidate input sequence set; an update module for updating the physical parameter vector of the initial state space model based on the current period input vector and the current period beam measurement value to obtain an updated state space model; a target control sequence determination module for determining a target control sequence from the candidate input sequence set based on the updated state space model, according to the current period state vector and the candidate input sequence set; and an instruction output module for determining a target control instruction based on the target control sequence and outputting the target control instruction to the execution power supply of the ion source.

[0013] A third aspect of the present invention provides an ion source beam adaptive device, the ion source beam adaptive device comprising: a memory and at least one processor, the memory storing instructions; at least one of the processors calling the instructions in the memory to cause the ion source beam adaptive device to perform the steps of the ion source beam adaptive method described in any of the preceding claims.

[0014] A fourth aspect of the present invention provides a computer-readable storage medium storing instructions that, when executed by a processor, implement the steps of the ion source beam adaptive method described in any of the preceding claims.

[0015] In the technical solution of this invention, an initial state space model is first constructed. Then, the current period input vector, the current period beam measurement value corresponding to the current period input vector, the current period state vector, and the candidate input sequence set are obtained. Based on the current period input vector and the current period beam measurement value, the physical parameter vector of the initial state space model is updated to obtain an updated state space model. Next, based on the updated state space model, the target control sequence is determined from the candidate input sequence set according to the current period state vector and the candidate input sequence set. Finally, the target control command is determined based on the target control sequence and output to the execution power supply of the ion source. This aims to suppress beam drift, improve the long-term dynamic stability of beam output, adapt to the computing power requirements of industrial sites, solve the problem of insufficient long-term dynamic stability of ion source beams in large-scale mass production scenarios, and overcome the inherent defects of classic fixed parameter control and pure data-driven black box control. Attached Figure Description

[0016] Figure 1 A logic flowchart of the ion source beam adaptive method provided in an embodiment of the present invention;

[0017] Figure 2This is a schematic diagram of the structure of the ion source beam adaptive device provided in an embodiment of the present invention;

[0018] Figure 3 This is a schematic diagram of the structure of the ion source beam adaptive device provided in an embodiment of the present invention. Detailed Implementation

[0019] This invention provides an ion source beam adaptive method, apparatus, device, and storage medium. In this invention, the terms "first," "second," "third," "fourth," etc. (if present)," in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatuses.

[0020] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 One embodiment of the ion source beam adaptive method in this invention includes:

[0021] 101. Construct the initial state-space model;

[0022] In this embodiment, an initial state-space model is constructed with a parameterized nonlinear structure as its core to characterize the complete dynamic process of an interthermal cathode ion source from electrical input to beam output. First, the core vector definitions of the model are clarified, defining the specific connotations and compositions of the state vector, input control vector, and physical parameter vector. The state vector is defined as a two-dimensional vector composed of plasma electron density and electron temperature, intuitively reflecting the core dynamic characteristics of the plasma inside the ion source. The input control vector consists of filament current, arc voltage, and bias voltage, corresponding to the key output parameters of the power supply unit of the controlled object, directly related to the electrical input excitation state of the ion source. The physical parameter vector includes core parameters such as ionization rate coefficient, diffusion coefficient, and energy loss coefficient, quantifying the physical characteristics of plasma reaction and energy transfer inside the ion source. Based on the above vector definitions, a state equation is constructed by combining the plasma particle number conservation law and the energy conservation law. This equation characterizes the dynamic evolution process of the state vector through nonlinear functions, specifically covering the ionization source term, loss term, and diffusion term for electron density changes, as well as the energy input term and loss term for electron temperature changes, fully capturing the dynamic mechanism of multi-physics coupling inside the ion source. Simultaneously, an output equation is constructed based on the space charge-limited current law. A quantitative correlation is established between beam intensity and plasma electron density, bias voltage, and ion source geometry through nonlinear functions, mapping the correspondence between input excitation and beam output. Finally, the state equation and output equation are integrated to form a complete initial state-space model, achieving a mechanistic description of the ion source's dynamic operation. This effectively avoids the shortcomings of classical static models in characterizing system nonlinearity, dynamic inertia, and hysteresis effects, laying a solid mechanistic foundation for adaptive beam control of the ion source and contributing to improved long-term beam output stability and control accuracy.

[0023] 102. Obtain the current cycle input vector, the current cycle beam measurement value corresponding to the current cycle input vector, the current cycle state vector, and the candidate input sequence set;

[0024] In this embodiment, the current cycle input vector and the corresponding beam current measurement value are obtained through real-time acquisition. The current cycle input vector is a vector composed of the filament current, arc voltage, and bias voltage for the current control cycle, conforming to the definition of the input control vector in the state-space model. It is obtained by the sensing unit synchronously acquiring the output feedback values ​​of each power supply of the ion source. The current cycle beam current measurement value is the actual detected beam current value matching the input vector, captured in real-time by the beam current diagnostic device. The current cycle state vector is a two-dimensional vector composed of electron density and electron temperature. Based on the constructed state-space model, combined with the real-time acquired input and output data and the updated model parameter estimates, it is derived in real-time through a physical mechanism-driven state observation method, accurately reflecting the dynamic evolution state of the plasma inside the ion source. The candidate input sequence set is a set of multiple alternative control sequences constructed for rolling optimization. Based on the constraints of the input control quantity, within the maximum and minimum allowable values ​​of filament current, arc voltage, and bias voltage, as well as the limits on the change of control quantity, it is mainly determined based on the safe operation threshold of the ion source hardware, the process stability range of plasma discharge, and the industrial standard requirements of semiconductor ion implantation. It is generated for the finite prediction time domain. Each sequence contains candidate values ​​of the input vector for several future control cycles, providing a complete optimization space for solving the subsequent optimization problem.

[0025] 103. Update the physical parameter vector of the initial state space model based on the current period input vector and the current period beam measurement value to obtain the updated state space model;

[0026] In this embodiment, the initial state-space model is a nonlinear structure and cannot be directly used for online parameter identification. It needs to be discretized and linearized near the current operating point to remove nonlinear coupling characteristics, transforming it into a linear regression model suitable for parameter identification. This lays the structural foundation for accurate updates of subsequent physical parameters. The current period regression vector is constructed collaboratively from the current period input vector, the current period beam measurement value, and historical input-output data. It integrates recent system operation data, reflects the correlation between input excitation and beam output, and serves as the input carrier for parameter identification.

[0027] During parameter updates, the covariance matrix retained from the previous cycle is called upon, and a preset forgetting factor is introduced. A recursive least squares method with a forgetting factor is used as the adaptive estimation algorithm. Based on the current cycle's regression vector, the previous cycle's covariance matrix, and the forgetting factor, the current cycle's gain matrix is ​​solved through matrix operations. This matrix is ​​used to adjust the step size of parameter updates, balancing parameter convergence speed and stability. Subsequently, considering the deviation between the current cycle's beam measurement value and the model's prediction value based on historical parameters, the physical parameter vector of the initial state-space model is iteratively updated using the adaptive estimation algorithm. The initial physical parameter vector is updated to the parameter estimates for the current cycle. The forgetting factor is used to mitigate the influence of early outdated data, ensuring that parameter updates accurately track the system's time-varying characteristics. After the physical parameter vector is updated, it is substituted into the initial state-space model, replacing the original physical parameters, completing the model iteration, and obtaining the updated state-space model. By employing discretized linear processing, the physical interpretability of the mechanistic model and the feasibility of parameter identification are balanced. The application of an adaptive estimation algorithm enables real-time iteration of physical parameters, allowing the updated model to dynamically match the dynamic characteristics caused by cathode loss, coil thermal drift, slight variations in inlet flow, and multi-physics coupling within the ion source, thus overcoming the black-box limitations of purely data-driven models. Simultaneously, the updated state-space model exhibits higher prediction accuracy, providing a reliable prediction basis for subsequent rolling optimization, effectively improving the precision and stability of beam control, and supporting closed-loop adaptive control of the ion source beam.

[0028] 104. Based on the updated state space model, determine the target control sequence from the candidate input sequence set according to the current periodic state vector and the candidate input sequence set;

[0029] In this embodiment, the updated state-space model, through online parameter identification and iterative optimization, accurately matches the current dynamic operating characteristics and time-varying patterns of the ion source, serving as the core carrier for beam prediction and optimization. During the process, the current periodic state vector is used as the initial evolution benchmark. This vector contains plasma electron density and electron temperature, intuitively reflecting the real-time state of the plasma inside the ion source. Combined with each candidate control sequence from the candidate input sequence set, it is substituted into the updated state-space model for simulation and deduction, completing the beam prediction within a preset future finite time domain. This generates a set of beam prediction values ​​corresponding one-to-one with each candidate sequence, establishing a quantitative correlation between input and output. Subsequently, optimization calculations are performed based on an optimization function. This optimization function balances beam control accuracy and control stability, including a beam tracking error sum of squares term and a control quantity change weighting term. The beam tracking error sum of squares term minimizes the deviation between the beam prediction value and the preset beam setpoint, while the control quantity change weighting term adjusts the amplitude of the input control quantity change through a weight matrix, avoiding system fluctuations caused by parameter abrupt changes. By solving this constrained optimization problem, candidate input sequences that minimize the optimization function value are selected and determined as the target control sequence, completing the optimization decision within a single cycle. Prediction is performed based on the updated mechanistic model, preserving physical interpretability while avoiding the shortcomings of traditional fixed-parameter control in adapting to the time-varying characteristics of the system, thus improving the accuracy of beam prediction. The rolling optimization mode can update the optimization benchmark in real time in each control cycle, rapidly responding to changes in the coupling of multiple physical fields within the ion source and external disturbances, ensuring that the target control sequence always closely matches the actual operating state of the system.

[0030] 105. Determine the target control command based on the target control sequence, and output the target control command to the execution power supply of the ion source.

[0031] In this embodiment, the target control sequence is a set of candidate control parameters within a finite future time domain obtained through rolling optimization. The initial control quantity in the time domain corresponds to the optimal control parameter of the current control cycle. It is the core control benchmark that adapts to the current dynamic operating state of the ion source and can precisely regulate the beam output. Therefore, this initial control quantity is extracted first as a candidate for the target control command. To avoid equipment damage or process instability caused by abnormal control parameters, a preset safety constraint rule is introduced to verify its effectiveness. This rule conforms to the hardware safety limits and process stability requirements of the ion source, covering the maximum and minimum allowable ranges of filament current, arc voltage, and bias voltage, as well as the threshold for control quantity changes. The verification process focuses on checking whether the initial control quantity is within the safe and feasible domain and whether it meets the process requirements for stable plasma discharge and precise beam regulation. After verification confirms compliance, the initial control quantity in the time domain is formally determined as the target control command, completing the transformation from control decision to execution command. The maximum and minimum allowable ranges of filament current, arc voltage, and bias voltage, as well as the threshold values ​​for changes in control quantities between adjacent cycles, need to be comprehensively set based on the safe operating limits of the ion source hardware, the stable range of the plasma discharge process, the dynamic response characteristics of the power supply actuator, and the standard requirements of the semiconductor ion implantation process. After the target control command is generated, it is synchronously transmitted to the execution power supplies of the ion source, including the filament power supply, arc power supply, and bias power supply. Each execution power supply precisely adjusts the output current and voltage parameters according to the command, driving the interthermal cathode ion source to complete the corresponding operating state adjustment, thereby achieving closed-loop control of the beam output. By extracting and executing the initial control quantity, and in line with the real-time iterative characteristics of rolling optimization, the control commands can quickly respond to the time-varying characteristics of the ion source and external disturbances, ensuring the timeliness of control. The safety verification process constructs a protective barrier for control execution, effectively avoiding risks such as filament burnout, grid damage, and plasma discharge extinction caused by exceeding threshold parameters, thereby improving the reliability of equipment operation. The precise matching of commands and execution power supplies achieves seamless connection from control decision-making to engineering execution, ensuring that the beam output can stably approach the set value, reducing beam fluctuations and long-term drift, and further improving the accuracy and consistency of ion source beam control.

[0032] After the command is executed, the next control cycle begins, repeating the entire process of data acquisition, model parameter updating, target control sequence determination, target control command generation and output, forming a complete closed-loop adaptive control. This iterative cycle mechanism is the core of ensuring the long-term stable operation of the ion source. By continuously repeating the above process, the dynamic time-varying characteristics of the ion source can be captured in real time, including changes in system characteristics caused by cathode loss, coil thermal drift, slight changes in inlet flow rate, and multi-physics coupling effects. At the same time, it can quickly respond to the impact of external disturbances on beam output.

[0033] In this embodiment of the invention, constructing the initial state space model includes: defining the state vector, input vector, and physical parameter vector of the initial state space model; constructing a state equation based on the law of conservation of plasma particle number, the law of conservation of energy, the state vector, the input vector, and the physical parameter vector; constructing an output equation based on the law of space charge confinement current, the state vector, the input vector, and the physical parameter vector; and integrating the state equation and the output equation to construct the initial state space model.

[0034] In this embodiment, constructing the initial state-space model first requires defining the vectors. The state vector is defined as X=[n e ,T e ] T Where T represents the transpose operator in linear algebra, used to convert a row vector into a column vector to conform to standard arithmetic form, and n e T represents the plasma electron density. e The electron temperature and the electron temperature together characterize the core dynamic state of the plasma inside the ion source; the input control vector is defined as U=[I f V arc V bias ] T Where T represents the transpose operator in linear algebra, and I f Represents filament current, V arc Indicates arc voltage, V bias The bias voltage represents a key output parameter of the ion source power supply unit, directly determining the electrical input excitation level of the ion source; the physical parameter vector is defined as Θ=[θ1,θ2,...,θ...]. m ] T T represents the transpose operator in linear algebra, and each element θ in the vector represents the transpose operator. i (i=1,2,...,m) are all key coefficients obtained based on plasma physics theory or experimental calibration, such as ionization rate coefficient, diffusion coefficient, energy loss coefficient, etc., used to quantify the intensity or rate of specific physical processes inside the ion source.

[0035] Based on the above vector definition, and combining the plasma particle number conservation law and the energy conservation law, the state equation dX / dt=F(X,U,Θ) is constructed, where dX / dt represents the time derivative of the state vector X, reflecting the dynamic evolution rate of the plasma state over time. The nonlinear function F(·) contains two core components: one describes the electron density n. e The terms include the changing ionization source term, loss term, and diffusion term. The ionization source term characterizes the rate at which electrons collide with neutral particles to produce ions, the loss term characterizes the rate at which ions recombine or escape, and the diffusion term characterizes the transport rate of the plasma in space. Secondly, it describes the electron temperature T. eThe changing energy input and loss terms, the energy input terms characterize the rate at which electrons are injected into the ion source by processes such as arc discharge, and the loss terms characterize the rate at which electrons lose energy through collisions, radiation, etc. Through the coupling of these terms, the equation of state fully describes the dynamic process of multi-physics coupling inside the ion source.

[0036] Simultaneously, the output equation I is constructed based on the space charge-limited current law. beam =G(X,U,Θ), where I beam For the beam intensity output from the ion source, the nonlinear function G(·) establishes the relationship between beam intensity and plasma electron density n. e Bias voltage V bias The quantitative correlation between the ion source geometry and the space charge confinement current law indicates that when the ion extraction electric field is dominated by the space charge effect, the beam intensity is proportional to the plasma density and the 3 / 2 power of the extraction voltage. The function G(·) modifies this law by introducing geometric parameters in the physical parameter vector Θ to accurately map the actual beam output characteristics of the ion source.

[0037] Finally, the state equations and output equations are integrated to form a complete initial state-space model. This model is centered on the dynamic evolution of the state vector X, driven by the excitation of the input control vector, and ultimately mapped to the beam intensity I through the output equation. beam This approach enables a mechanistic description of the complete dynamic process of an ion source from electrical input to beam output. By defining vectors and constructing equations, the physical interpretability of the ion source operation is preserved, avoiding the black-box defects of purely data-driven models. The coupling of the state equation and the output equation fully captures the dynamic characteristics of multi-physics coupling within the ion source, providing a structured mechanistic framework for subsequent online parameter identification and rolling optimization.

[0038] In this embodiment of the invention, updating the physical parameter vector of the initial state space model based on the current period input vector and the current period beam measurement value to obtain the updated state space model includes: constructing a current period regression vector based on the current period input vector and the current period beam measurement value; obtaining the previous period covariance matrix and a preset forgetting factor; using an adaptive estimation algorithm to calculate the current period gain matrix based on the current period regression vector, the previous period covariance matrix, and the forgetting factor; and updating the physical parameter vector of the initial state space model based on the current period gain matrix and the current period beam measurement value to obtain the updated state space model.

[0039] In this embodiment, the initial nonlinear state-space model is first linearly discretized. Since the original model is a continuous-time nonlinear form and cannot be directly used for online parameter identification, a linear discretization algorithm is used near the current operating point for linear approximation. Specifically, the forward Euler discretization algorithm can be used to transform it into a linear regression model suitable for recursive least squares, providing a structured data foundation for subsequent parameter identification. Based on this, a regression vector φ(k) for the current period is constructed using the current period input vector, the current period beam measurement value, and historical input / output data. The historical input / output data is taken from the historical data cache unit. The historical input vector consists of filament current, arc voltage, and bias voltage data collected and stored in previous control cycles. The historical beam measurement value is the beam intensity detection data for the corresponding period, which can be quickly retrieved via the high-speed internal bus. This vector integrates the current and historical input / output data and is the core carrier for linking input / output with model parameters.

[0040] To proceed with the parameter update, it is necessary to obtain the covariance matrix P(k) of the previous period. 1) A preset forgetting factor λ is used, where the covariance matrix of the previous cycle is stored in the historical data unit of the core controller, reflecting the uncertainty of the parameter estimation in the previous cycle. The forgetting factor λ ranges from 0 to 1 and needs to be tuned according to the time-varying characteristics and measurement noise level of the ion source system. For slow time-varying processes such as cathode aging, a value close to 1 (e.g., 0.995) is usually selected. This value can be tuned through simulation or offline experiments, with the goal of improving the accuracy of system parameter tracking and the stability of closed-loop control. The recursive least squares method with forgetting factor is used as the adaptive estimation algorithm. First, the gain matrix K(k) = P(k) of the current cycle is calculated. 1)φ(k)[λ+φ T(k) P(k 1)φ(k)] 1, where k represents the current period, K(k) represents the gain matrix of the current period, and φ T(k) P(k) represents the transpose of the current periodic regression vector φ(k). 1) represents the covariance matrix of the previous period, and λ represents the forgetting factor.

[0041] Subsequently, the parameter update formula Θhat(k) = Θhat(k) is used. 1)+K(k)[I beam_m (k) φ T(k) Θhat(k 1)] Complete the physical parameter vector update, where k represents the current period, and Θhat(k) represents the updated physical parameter vector of the state-space model. 1) represents the physical parameter vector of the initial state-space model, K(k) represents the current periodic gain matrix, used to adjust the parameter update step size and balance convergence speed and stability, I beam_m (k) represents the current periodic beam measurement value, φ T(k) This represents the transpose of the current periodic regression vector φ(k), [I beam_m (k) φ T(k) Θhat(k 1)] represents the current periodic beam prediction error, which is the difference between the actual measured value and the model's prediction based on historical parameters.

[0042] Simultaneously, the covariance matrix is ​​updated using the formula P(k)=[I K(k)φ T(k) ]P(k 1) / λ completes the iterative update of the covariance matrix, where I is the identity matrix. Updating the covariance matrix is ​​the core step in achieving adaptive parameter estimation using the recursive least squares method. Its essence is to dynamically quantify the uncertainty of parameter estimates and provide a real-time basis for calculating the gain matrix in the next cycle, adapting to the control requirements of time-varying industrial systems like ion sources. The physical meaning of the covariance matrix P(k) is the covariance matrix of the parameter estimates Θhat(k). Its diagonal elements correspond to the variance of each physical parameter estimate, reflecting the confidence level of a single parameter estimate, while the off-diagonal elements reflect the correlation between different parameter estimates. During the recursive process, the uncertainty of the parameter estimates changes with each new set of input and output data. Therefore, it is necessary to iteratively update the covariance matrix to accurately characterize the confidence level of the current parameter estimate, providing a basis for calculating the gain matrix K(k+1) in the next cycle. From the perspective of the formula update mechanism, firstly, [I The K(k)φT(k) terms are paired with the historical covariance matrix P(k) 1) A correction is made because the introduction of new measurement data reduces the uncertainty of parameter estimation. This correction term reduces the trace (total variance) of the covariance matrix through matrix operations, reflecting the improvement in parameter estimation accuracy. Secondly, dividing by the forgetting factor λ (0 < λ ≤ 1) is the core design of the recursive least squares method with a forgetting factor. When λ < 1, the historical covariance matrix is ​​exponentially decayed, giving more weight to recent data and weakening the influence of outdated early data. For slow time-varying systems like ion sources, time-varying characteristics such as cathode aging and coil thermal drift can lead to a decrease in the effectiveness of early data. Through the decay effect of the forgetting factor, the update of the covariance matrix allows parameter estimation to quickly track the current dynamic characteristics of the system, avoiding estimation lag caused by fixed weights.

[0043] After the physical parameter vector is updated to Θhat(k), it is substituted into the initial state space model to obtain the updated state space model. The linear discretization algorithm balances the physical interpretability of the mechanistic model with the feasibility of parameter identification, avoiding the black box defect of a purely data-driven model. The recursive least squares method with a forgetting factor enables online real-time updates of physical parameters, allowing the model to dynamically track the time-varying characteristics caused by cathode loss, coil thermal drift, slight changes in inlet flow rate, and multi-physics coupling within the ion source. The accuracy and real-time nature of data acquisition ensure the timeliness and accuracy of parameter updates, while the iterative updates of the gain matrix and covariance matrix improve the convergence and stability of parameter identification. The updated state space model provides an accurate prediction carrier for subsequent rolling optimization, ensuring that the solution of the optimal control sequence closely matches the actual operating state of the system, effectively suppressing long-term beam drift, and improving the long-term stability and control accuracy of the ion source beam output.

[0044] In this embodiment of the invention, determining the target control sequence from the candidate input sequence set based on the updated state space model, according to the current period state vector and the candidate input sequence set, includes: performing beam prediction based on the updated state space model, according to the current period state vector and the candidate input sequence set, to obtain a set of beam prediction values ​​within a preset future period; and solving an optimization function based on the set of beam prediction values ​​to determine the target control sequence from the candidate input sequence set.

[0045] In this embodiment, the current periodic state vector includes plasma electron density and electron temperature, intuitively reflecting the real-time evolution state of the plasma inside the ion source. Using this as an initial benchmark, each set of candidate input sequences from the candidate input sequence set is input into the updated state space model. Each set of candidate input sequences contains input vectors for multiple consecutive control cycles within a preset future period. Each input vector consists of three components: filament current, arc voltage, and bias voltage. Through simulation, beam prediction within the preset future period (i.e., a finite future time domain) is completed, generating a subset of beam prediction values ​​corresponding one-to-one with each set of candidate input sequences. Each subset of beam prediction values ​​includes beam prediction values ​​for multiple consecutive control cycles within the preset future period. It should be noted that the setting of the preset future period must comprehensively consider multiple dimensions of factors, including the dynamic response characteristics of the ion source system, control cycle, process requirements, computational resource constraints, and engineering experience, to achieve synergistic optimization of control accuracy, dynamic response, and system stability. Subsequently, an optimization function is solved to perform optimization calculations for the target control sequence. This optimization function includes a sum of squared beam tracking errors and a weighted term for control quantity changes. For each candidate input sequence, the values ​​of these two terms are calculated separately and then summed with weights to obtain the corresponding optimization function value. Among all candidate input sequences, the sequence with the smallest optimization function value is selected as the target control sequence. This sequence ensures that the beam output in future cycles is as close as possible to the set value, meeting the accuracy requirements of the ion implantation process, while also ensuring that the variation range of the control quantity is within a reasonable range, avoiding damage to core components such as the cathode and grid, and guaranteeing long-term stable operation of the equipment. Beam prediction based on the updated mechanism model retains physical interpretability and avoids the shortcomings of traditional fixed parameter control in adapting to the time-varying characteristics of the system, thus improving the accuracy of beam prediction. The dual-term design of the optimization function balances beam control accuracy and control stability.

[0046] In this embodiment of the invention, the optimization function includes a beam tracking error sum of squares term and a control quantity change weighting term; the beam prediction value set includes multiple beam prediction value subsets, and the candidate input sequence set includes candidate input sequences corresponding one-to-one with each of the beam prediction value subsets; the step of solving the optimization function based on the beam prediction value set to determine the target control sequence from the candidate input sequence set includes: calculating the beam tracking error sum of squares term and the control quantity change weighting term based on each beam prediction value subset and the corresponding candidate input sequence; performing a weighted summation of each beam tracking error sum of squares term and the corresponding control quantity change weighting term to obtain an optimization function value corresponding to each candidate input sequence; and selecting the candidate input sequence with the smallest optimization function value as the target control sequence.

[0047] In this embodiment, the optimization function, balancing beam control accuracy and system operational stability, consists of a beam tracking error sum of squares term and a control quantity change weighting term. These two terms work together to provide a quantitative evaluation standard for selecting the target control sequence. The beam tracking error sum of squares term focuses on the accuracy of the beam output, primarily measuring the deviation between the predicted beam value and the beam setpoint required by the process. By accumulating the squares of deviations at each step within a finite future time domain, it strengthens the suppression of persistent deviations. The control quantity change weighting term emphasizes system operational stability and equipment safety, constraining the abrupt changes in the input control quantity. It adjusts the weighting of changes in control parameters such as filament current, arc voltage, and bias voltage through a weight matrix, balancing the adjustment sensitivity of each parameter. This prevents sudden changes in control quantity from causing severe plasma fluctuations within the ion source, avoids equipment risks such as cathode overload and grid sputtering damage, and simultaneously reduces the adjustment load on the power supply actuator, extending the service life of core components.

[0048] There is a one-to-one correspondence between the beam prediction value set and the candidate input sequence set. Each candidate input sequence corresponds to a complete set of control parameters within a finite future time domain. Substituting these parameters into the updated state-space model yields a corresponding subset of beam prediction values. All subsets of beam prediction values ​​are integrated to form the beam prediction value set, which fully reflects the future evolution trend of the beam under different control schemes. For each candidate input sequence and its corresponding subset of beam prediction values, two indicators are calculated: first, the sum of squared beam tracking errors, which quantifies the beam tracking capability by accumulating the squared deviations between all beam prediction values ​​and the beam setpoint corresponding to the candidate sequence; and second, the weighted control quantity change, which quantifies the operational stability and equipment adaptability by weighting the control quantity changes in each control cycle of the candidate sequence using a weight matrix. The two calculation results are then weighted and summed to obtain the optimized function value corresponding to each candidate input sequence. This value comprehensively reflects the beam control accuracy and system operational stability of the candidate sequence; the smaller the value, the better the sequence can meet both core requirements. Finally, the candidate input sequence with the smallest optimization function value was selected as the target control sequence, ensuring that the selected control scheme possesses both excellent beam tracking performance and stable operation of the ion source system. The dual-term design of the optimization function overcomes the limitations of control solely focused on precision, achieving synergistic optimization of precision and stability. This avoids system fluctuations caused by solely pursuing precision while preventing insufficient precision due to overemphasis on stability. By correlating candidate sequences with beam prediction values, a comprehensive quantitative evaluation basis is provided for each control scheme, eliminating the subjectivity and uncertainty of empirical selection. The selection of the target control sequence based on the principle of minimizing the optimization function value ensures the optimality and scientific rigor of the control strategy, adapts to the time-varying characteristics of multi-physics coupling in the ion source, and can quickly respond to system changes such as cathode aging and thermal drift, effectively suppressing long-term beam drift and improving the long-term consistency and control precision of the beam output.

[0049] In this embodiment of the invention, the step of calculating the sum of squares of the beam tracking error and the weighted term of the control quantity change based on each subset of beam prediction values ​​and the corresponding candidate input sequence includes: obtaining a preset control quantity change weight matrix and a preset beam setting value; calculating the sum of squares of the deviations of the beam prediction values ​​based on each subset of beam prediction values ​​and the beam setting value to obtain a sum of squares of beam tracking error corresponding to each candidate input sequence; calculating the input vector change in adjacent control cycles based on each candidate input sequence, and calculating the weighted term of the control quantity change based on each input vector change and the control quantity change weight matrix to obtain a weighted term of the control quantity change corresponding to each sum of squares of beam tracking error.

[0050] In this embodiment, the core of the optimization process is to solve the minimization problem of the optimization function. First, a preset control quantity change weight matrix and a preset beam current setting value are obtained. The beam current setting value is a core process parameter of the semiconductor ion implantation process, usually derived from the process formula or production instructions issued by the host computer. It clarifies the target beam current intensity value that the ion source needs to output and serves as the benchmark for the entire control process. The control quantity change weight matrix is ​​a diagonal matrix, and its element values ​​need to be pre-tuned according to specific process requirements. Its core function is to balance beam tracking performance with the smoothness of control actions. If the element values ​​in R are too large, the optimization algorithm will tend to generate control instructions with gradual changes to reduce the impact on core components such as the filament and gate, protect the equipment, and improve process stability. If the element values ​​are too small, the control instructions are allowed to change more quickly to improve the system's dynamic response speed and adapt to scenarios where the beam current needs to be adjusted rapidly.

[0051] Based on this, the sum of squares of the beam tracking error is calculated for each beam prediction value and beam setting value. This corresponds to the first part of the optimization function. Each candidate input sequence corresponds to a subset containing beam prediction values ​​for the next N periods. For this subset, the deviation between the beam prediction value and the beam setting value at each step within the preset future period (length N) is calculated. All deviations are squared and summed to obtain the sum of squares of the beam tracking error for the candidate sequence. The specific calculation expression is as follows:

[0052] ,

[0053] Where N represents the preset future period, and i ranges from 1 to N, representing the 1st to Nth future periods starting from the current period k. Indicates the beam current setting value. This represents the beam prediction value for the next k+i periods in the current period k. The k to the right of the vertical line indicates that the prediction is based on the state and parameters of the current period k. The sum of squares of the beam tracking error term is used to quantify the tracking ability of the candidate sequence to the beam setpoint. The smaller the value, the closer the beam output in the next N steps is to the setpoint, and the higher the control accuracy.

[0054] Subsequently, the change in input vector between adjacent control cycles is calculated based on each candidate input sequence, and the weighted term of the control quantity change is calculated in conjunction with the control quantity change weight matrix. The specific calculation expression is as follows:

[0055]

[0056] Where N represents the preset future period, and i ranges from 0 to N. 1. i=0 corresponds to the change in input control quantity ΔU(k) ​​in the current period k (i.e., the change in period k). The input vector difference of 1), the input change ΔU(k+1) corresponding to period k+1 at i=1, until i=N 1 corresponds to period k+N The input variation is 1. Each candidate input sequence contains N future periods (periods k to k+N). 1) is the input control vector, therefore N can be obtained. The change in the input vector over an adjacent cycle reflects the dynamic range of the control quantity. This change is then weighted with the control quantity change weight matrix R. Each component of ΔU(k+i) (change in filament current, change in arc voltage, and change in bias voltage) is multiplied by the corresponding weight element in R and then squared and summed to obtain the control quantity change weighting term for the candidate sequence. The purpose of the control quantity change weighting term is to quantify the control stability of the candidate sequence. The smaller the value, the smoother the change in control quantity, which can avoid violent plasma fluctuations caused by sudden changes in control quantity, reduce the risk of cathode overload and grid sputtering damage, and at the same time reduce the adjustment load on the power supply actuator and extend the service life of core components.

[0057] The objective function value J of each candidate input sequence is obtained by summing the sum of squared beam tracking errors and the weighted term of control quantity changes. The specific calculation expression is as follows:

[0058] ,

[0059] in, This indicates the optimization function value. This represents the sum of squares of the beam tracking error. This represents the weighted term for changes in control quantity. The objective function value reflects the beam control accuracy and system stability of the candidate sequence. By selecting the candidate input sequence with the smallest objective function value as the target control sequence, we can ensure the beam output accuracy in the next N steps while avoiding equipment risks caused by sudden changes in control quantity, thus achieving synergistic optimization of accuracy and stability.

[0060] In this embodiment of the invention, determining the target control instruction based on the target control sequence includes: acquiring the first-step time-domain control quantity and a preset safety constraint rule in the target control sequence; validating the first-step time-domain control quantity based on the safety constraint rule; and using the first-step time-domain control quantity as the target control instruction when the validation passes.

[0061] In this embodiment, the target control sequence obtained by rolling optimization includes multiple control instructions within a finite future time domain. The first control quantity in the time domain is the control instruction corresponding to the current cycle, which corresponds to the first element in the target control sequence. Before outputting this control quantity to each execution power supply of the ion source, a preset safety constraint rule needs to be obtained. This rule is usually formulated based on the safe operation threshold and process stability requirements of the ion source hardware, covering the upper and lower limits of filament current, arc voltage, and bias voltage, as well as the rate of change constraint of control quantities in adjacent cycles. Based on this safety constraint rule, the validity of the first control quantity in the time domain is verified. Specifically, it includes verifying whether the absolute value of the control quantity is within the safe range allowed by the hardware, and whether the change amplitude of the control quantity is within the dynamic response capability range of the equipment. If the verification passes, it indicates that the control quantity will not cause damage to core components such as the filament and grid, nor will it cause violent plasma fluctuations. At this time, the first control quantity in the time domain is output as the target control instruction to each execution power supply to drive the ion source to adjust its operating state. If the verification fails, a fault-tolerant processing mechanism is triggered, such as correcting the control quantity based on the constraint conditions or switching to a safe backup instruction to avoid equipment failure and process interruption. By introducing a safety constraint verification mechanism, dual protection of hardware safety and process stability is added on the basis of rolling optimization. This avoids the problem of control quantity exceeding limits that may be caused by relying solely on optimization algorithms, effectively protecting core components such as filaments and gates from damage and extending equipment lifespan. The verification process and the constraints of the optimization stage form a closed loop, ensuring the engineering executability of control commands and improving the reliability and stability of system operation. The design of the fault-tolerant processing mechanism further enhances the system's anti-interference capability, reduces the production risks caused by abnormal control quantities, ensures the continuous and stable operation of the semiconductor ion implantation process, and provides solid safety support for large-scale mass production.

[0062] The above describes the ion source beam adaptive method in the embodiments of the present invention. The following describes the ion source beam adaptive device in the embodiments of the present invention. Please refer to [link / reference]. Figure 2One embodiment of the ion source beam adaptive device in this invention includes:

[0063] Model building module 201: Used to build the initial state space model;

[0064] Data acquisition module 202: used to acquire the current cycle input vector, the current cycle beam measurement value corresponding to the current cycle input vector, the current cycle state vector, and the candidate input sequence set;

[0065] Update module 203: used to update the physical parameter vector of the initial state space model based on the current period input vector and the current period beam measurement value, so as to obtain the updated state space model;

[0066] Target control sequence determination module 204: used to determine a target control sequence from the candidate input sequence set based on the updated state space model, according to the current periodic state vector and the candidate input sequence set;

[0067] Command output module 205: used to determine the target control command based on the target control sequence and output the target control command to the execution power supply of the ion source.

[0068] Based on the same ideas as the methods in the above embodiments, the apparatus provided in this application can implement the methods in the above embodiments.

[0069] above Figure 2 The ion source beam adaptive device in the embodiments of the present invention will be described in detail from the perspective of modular functional entities. The ion source beam adaptive device in the embodiments of the present invention will be described in detail from the perspective of hardware processing.

[0070] Figure 3This is a schematic diagram of the structure of an ion source beam adaptive device 300 provided in an embodiment of the present invention. The ion source beam adaptive device 300 can vary significantly due to different configurations or performance. It may include one or more central processing units (CPUs) 310 (e.g., one or more processors) and a memory 320, and one or more storage media 330 (e.g., one or more mass storage devices) storing application programs 333 or data 332. The memory 320 and storage media 330 can be temporary or persistent storage. The program stored in the storage media 330 may include one or more modules (not shown in the diagram), each module may include a series of instruction operations on the ion source beam adaptive device 300. Furthermore, the processor 310 may be configured to communicate with the storage media 330 and execute the series of instruction operations in the storage media 330 on the ion source beam adaptive device 300 to implement the steps of the ion source beam adaptive method provided in the above-described method embodiments.

[0071] The ion source beam adaptive device 300 may also include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input / output interfaces 360, and / or one or more operating systems 331, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will understand that... Figure 3 The illustrated ion source beam adaptive device structure does not constitute a limitation on the ion source beam adaptive device, and may include more or fewer components than illustrated, or combine certain components, or have different component arrangements.

[0072] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when executed on a computer, cause the computer to perform the steps of the ion source beam adaptive method.

[0073] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system, device, or unit described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0074] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0075] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An adaptive method for ion source beam, characterized in that, include: Define the state vector, input vector, and physical parameter vector of the initial state-space model. The state vector is defined as a two-dimensional vector composed of plasma electron density and electron temperature. The input vector consists of filament current, arc voltage, and bias voltage. The physical parameter vector includes ionization rate coefficient, diffusion coefficient, and energy loss coefficient. A state equation is constructed based on the plasma particle number conservation law, energy conservation law, the state vector, the input vector, and the physical parameter vector. An output equation is constructed based on the space charge confinement current law, the state vector, the input vector, and the physical parameter vector. The state equation and the output equation are then integrated to construct the initial state-space model. Obtain the current cycle input vector, the current cycle beam measurement value corresponding to the current cycle input vector, the current cycle state vector, and the candidate input sequence set; The current cycle regression vector is constructed based on the current cycle input vector and the current cycle beam measurement value; the previous cycle covariance matrix and a preset forgetting factor are obtained; an adaptive estimation algorithm is used to calculate the current cycle gain matrix based on the current cycle regression vector, the previous cycle covariance matrix and the forgetting factor; and the physical parameter vector of the initial state space model is updated based on the current cycle gain matrix and the current cycle beam measurement value to obtain the updated state space model. Based on the updated state space model, beam prediction is performed according to the current period state vector and the candidate input sequence set to obtain a set of beam prediction values ​​for a preset future period; an optimization function is solved based on the set of beam prediction values ​​to determine the target control sequence from the candidate input sequence set; Based on the target control sequence, a target control command is determined and output to the execution power supply of the ion source.

2. The ion source beam adaptive method according to claim 1, characterized in that, The optimization function includes a beam tracking error sum of squares term and a control quantity change weighting term; the beam prediction value set includes multiple beam prediction value subsets, and the candidate input sequence set includes candidate input sequences that correspond one-to-one with each of the beam prediction value subsets. The step of solving the optimization function based on the beam prediction value set to determine the target control sequence from the candidate input sequence set includes: The sum of squares of the beam tracking error and the weighted term of the control quantity change are calculated based on each subset of the beam prediction values ​​and the corresponding candidate input sequence. The sum of squared beam tracking errors for each term and the corresponding weighted term of control variable change are weighted and summed to obtain the optimized function value corresponding to each candidate input sequence. The candidate input sequence with the smallest optimization function value is selected as the target control sequence.

3. The ion source beam adaptive method according to claim 2, characterized in that, The calculation of the sum of squared beam tracking errors and the weighted term of control quantity changes based on each subset of beam prediction values ​​and the corresponding candidate input sequence includes: Obtain the preset control quantity change weight matrix and the preset beam setting value; Based on each subset of the beam prediction values ​​and the beam setpoint, the sum of squared deviations of the beam prediction values ​​is calculated to obtain a beam tracking error sum of squares term corresponding one-to-one with each candidate input sequence; The input vector change in adjacent control cycles is calculated based on each candidate input sequence, and a control quantity change weighting term is calculated based on each input vector change and the control quantity change weighting matrix, resulting in a control quantity change weighting term that corresponds one-to-one with each beam tracking error sum of squares term.

4. The ion source beam adaptive method according to claim 1, characterized in that, The step of determining the target control command based on the target control sequence includes: Obtain the first-step time-domain control quantity and preset safety constraint rules from the target control sequence; The validity of the first-step control quantity in the time domain is verified based on the aforementioned security constraint rules. When the verification passes, the first step control quantity in the time domain is used as the target control command.

5. An ion source beam adaptive device, characterized in that, include: The model construction module defines the state vector, input vector, and physical parameter vector of the initial state-space model. The state vector is defined as a two-dimensional vector composed of plasma electron density and electron temperature. The input vector consists of filament current, arc voltage, and bias voltage. The physical parameter vector includes ionization rate coefficient, diffusion coefficient, and energy loss coefficient. Based on the plasma particle number conservation law, energy conservation law, the state vector, the input vector, and the physical parameter vector, a state equation is constructed. Based on the space charge confinement current law, the state vector, the input vector, and the physical parameter vector, an output equation is constructed. The state equation and the output equation are then integrated to construct the initial state-space model. Data acquisition module: used to acquire the current cycle input vector, the current cycle beam measurement value corresponding to the current cycle input vector, the current cycle state vector, and the candidate input sequence set; Update module: used to construct the current cycle regression vector based on the current cycle input vector and the current cycle beam measurement value; obtain the previous cycle covariance matrix and a preset forgetting factor; use an adaptive estimation algorithm to calculate the current cycle gain matrix based on the current cycle regression vector, the previous cycle covariance matrix and the forgetting factor; and update the physical parameter vector of the initial state space model based on the current cycle gain matrix and the current cycle beam measurement value to obtain the updated state space model; Target control sequence determination module: used to perform beam prediction based on the updated state space model, according to the current period state vector and the candidate input sequence set, to obtain a set of beam prediction values ​​in a preset future period; and to solve an optimization function based on the set of beam prediction values ​​to determine the target control sequence from the candidate input sequence set; Command output module: used to determine the target control command based on the target control sequence and output the target control command to the execution power supply of the ion source.

6. An ion source beam adaptive device, characterized in that, The ion source beam adaptive device includes: a memory and at least one processor, wherein the memory stores instructions; At least one of the processors invokes the instructions in the memory to cause the ion source beam adaptive device to perform the steps of the ion source beam adaptive method as described in any one of claims 1-4.

7. A computer-readable storage medium storing instructions thereon, characterized in that, When the instructions are executed by the processor, they implement the steps of the ion source beam adaptive method as described in any one of claims 1-4.

Images