A semiconductor radio frequency power supply control method, device, equipment and storage medium

Abstract

The present application relates to the technical field of semiconductor radio frequency power supply control, and particularly relates to a semiconductor radio frequency power supply control method, device, equipment and storage medium; residual error calculation is performed on a to-be-optimized parameter vector, control input and actual sampling output based on a preset residual error equation to obtain a residual error vector; a basic information matrix is constructed according to the residual error vector and the to-be-optimized parameter vector; the to-be-optimized parameter vector is updated according to a preset optimization algorithm, the basic information matrix, the residual error vector, a preset damping factor and a preset iteration step factor to obtain an updated parameter; a semiconductor radio frequency power supply control signal is generated according to the residual error vector, the updated parameter and a preset lookup table; original data is collected in multiple dimensions, system errors are filtered and collected through residual error solving, and a control signal is quickly generated in combination of residual error data and the preset lookup table, so that the control stability of the radio frequency power supply under a plasma load frequent fluctuation scene and the whole machine operation reliability are improved.

Description

Technical Field

[0001] This invention relates to the field of semiconductor radio frequency power supply control technology, and in particular to a semiconductor radio frequency power supply control method, apparatus, device and storage medium. Background Technology

[0002] Radio frequency (RF) power supplies used in semiconductor manufacturing equipment are widely applied in critical processes such as plasma etching and thin film deposition. Factors such as random and sudden changes in load impedance within the plasma cavity, high-frequency electromagnetic interference, and device nonlinear distortion continuously affect the accuracy of power supply control. Existing RF power supply control technologies generally suffer from narrow data acquisition coverage, limited modeling data sources, and a lack of systematic aggregation of multi-source disturbance errors, making residual distortion easily caused by noise. Conventional least-squares optimization is affected by insufficient system excitation and parameter redundancy, resulting in poor parameter identification. Furthermore, existing RF power supply control signals for semiconductor manufacturing equipment are computationally time-consuming, have delayed command response, and struggle to quickly adapt to dynamic load changes, resulting in insufficient power tracking accuracy and hindering the improvement of semiconductor process stability. Summary of the Invention

[0003] In order to overcome the shortcomings of the prior art, the present invention aims to provide a semiconductor radio frequency power supply control method, apparatus, device and storage medium.

[0004] A semiconductor radio frequency power supply control method includes: acquiring a parameter vector to be optimized, a control input, and an actual sampled output during an experiment; performing residual calculations on the parameter vector to be optimized, the control input, and the actual sampled output based on a preset residual equation to obtain a residual vector; constructing a basic information matrix based on the residual vector and the parameter vector to be optimized; updating the parameter vector to be optimized according to a preset optimization algorithm, the basic information matrix, the residual vector, a preset damping factor, and a preset iteration step size factor to obtain updated parameters; and generating a semiconductor radio frequency power supply control signal based on the residual vector, the updated parameters, and a preset lookup table.

[0005] Furthermore, the residual calculation based on the preset residual equation to obtain the residual vector includes: analyzing the parameter vector to be optimized and the control input according to the pre-trained prediction output model to obtain the model prediction output; and performing residual calculation on the model prediction output and the actual sampled output according to the residual equation to obtain the residual vector.

[0006] Furthermore, the step of constructing the basic information matrix based on the residual vector and the parameter vector to be optimized includes: performing gradient calculation on the residual vector and the parameter vector to be optimized to obtain the Jacobian matrix; and constructing the basic information matrix based on the Jacobian matrix and the transpose of the Jacobian matrix.

[0007] Furthermore, the step of updating the parameter vector to be optimized according to the preset optimization algorithm, the basic information matrix, the residual vector, the preset damping factor, and the preset iteration step size factor to obtain the updated parameters includes: calculating the residual vector according to the preset objective function to obtain the minimum residual sum of squares; calculating the minimum residual sum of squares, the basic information matrix, the damping factor, the preset identity matrix, and the residual vector according to the optimization algorithm to obtain the update amount; and updating the parameter vector to be optimized according to the preset boundary constraints, the update amount, and the iteration step size factor to obtain the updated parameters.

[0008] Furthermore, the step of generating the semiconductor radio frequency power supply control signal based on the residual vector, update parameters, and a preset lookup table includes: generating an optimal parameter vector based on the residual vector, update parameters, and a preset upper limit for the number of repetitions; and generating the semiconductor radio frequency power supply control signal based on the lookup table and the optimal parameter vector.

[0009] Furthermore, the step of generating the optimal parameter vector based on the residual vector, the update parameters, and the preset upper limit of repetition count includes: determining whether the norm of the residual vector is less than the preset norm threshold; if the norm of the residual vector is not less than the norm threshold, then returning to perform gradient calculation on the residual vector and the parameter vector to be optimized, until the number of repetitions of the step of performing gradient calculation on the residual vector and the parameter vector to be optimized is equal to the preset upper limit of repetition count, then the update parameters are used as the optimal parameter vector.

[0010] Furthermore, the step of generating the semiconductor radio frequency power supply control signal based on the lookup table and the optimal parameter vector includes: performing mapping analysis on the optimal parameter vector based on the lookup table to obtain the actual control quantity; and generating the semiconductor radio frequency power supply control signal based on the actual control quantity.

[0011] Furthermore, a semiconductor radio frequency power supply control device includes: a data acquisition module for acquiring the parameter vector to be optimized, control input, and actual sampled output during the experiment; a residual calculation module for performing residual calculations on the parameter vector to be optimized, control input, and actual sampled output based on a preset residual equation to obtain a residual vector; a matrix construction module for constructing a basic information matrix based on the residual vector and the parameter vector to be optimized; an update module for updating the parameter vector to be optimized based on a preset optimization algorithm, the basic information matrix, the residual vector, a preset damping factor, and a preset iteration step size factor to obtain updated parameters; and a signal generation module for generating a semiconductor radio frequency power supply control signal based on the residual vector, the updated parameters, and a preset lookup table.

[0012] Furthermore, a semiconductor radio frequency power supply control device includes: a memory and at least one processor, wherein the memory stores instructions; the at least one processor invokes the instructions in the memory to cause the semiconductor radio frequency power supply control device to perform the various steps of the semiconductor radio frequency power supply control method as described above.

[0013] Furthermore, a computer-readable storage medium stores instructions that, when executed by a processor, implement the various steps of the semiconductor radio frequency power supply control method described above.

[0014] In the technical solution of this invention, complete raw data is provided for the entire process algorithm operation by collecting the parameter vector to be optimized, control input, and actual sampled output from multiple dimensions; the residual vector is calculated by standardizing the residual equation to collect various system errors such as load disturbance and electromagnetic interference, and the error data is ensured to be reliable by filtering and noise reduction; a basic information matrix is ​​constructed based on the residual equation and the parameter vector to be optimized to characterize the sensitivity and nonlinearity of system parameters; the parameters are iteratively updated by relying on the damped optimization algorithm with step size constraint to improve the ill-conditioned defects of the basic information matrix, avoid parameter oscillation and local optima defects, and improve the parameter identification accuracy and operating condition adaptability; finally, the control signal is quickly generated by combining the optimized parameters, residual data, and preset lookup table, which simplifies real-time calculation, improves the control response speed, realizes the output following the target, improves the control stability and overall operation reliability of the RF power supply in the scenario of frequent fluctuation of plasma load, and improves the stability of semiconductor manufacturing process. Attached Figure Description

[0015] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which: Figure 1 This is a first flowchart of a semiconductor radio frequency power supply control method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a semiconductor radio frequency power supply control device provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of a semiconductor radio frequency power supply control device provided in an embodiment of the present invention. Detailed Implementation

[0016] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this invention 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 orders other than those 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 includes 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 apparatus.

[0017] 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 a semiconductor radio frequency power supply control method according to the present invention includes: 101. Collect the parameter vector to be optimized, control input, and actual sampled output during the experiment; In this embodiment, the parameter vector to be optimized includes amplifier coefficients, matching network parameters, and control inputs including drive amplitude, frequency, and bias voltage; the actual sampled output includes power, voltage, or current; through the parameter vector to be optimized, control inputs, and actual sampled outputs during the experiment, the complete acquisition dimensions can provide a complete data source for subsequent residual calculations and parameter optimization, restore the actual operating conditions of the RF power supply, and ensure the data reliability of model identification and control. 102. Based on the preset residual equation, perform residual calculation on the parameter vector to be optimized, the control input, and the actual sampled output to obtain the residual vector; In this embodiment, based on the preset residual equation, the residual vector is calculated by combining the parameter vector to be optimized, the control input and the measured sampling data. This can quantify the deviation between the RF power supply model prediction and the actual operating conditions, intuitively collect various errors such as load fluctuation, electromagnetic interference, and device nonlinearity, and suppress the acquisition noise with periodic sampling and filtering methods to ensure the authenticity and reliability of the residual data. This provides a reliable error benchmark for subsequent Jacobian matrix solution and parameter iterative optimization, which facilitates the algorithm to dynamically correct the model parameters, improve the system identification accuracy, and enhance the control stability of the RF power supply under plasma load change conditions. 103. Construct the basic information matrix based on the residual vector and the parameter vector to be optimized; In this embodiment, a basic information matrix is ​​constructed by using the residual vector and the gradient information of the parameter vector to be optimized. This matrix can characterize the sensitivity correlation between the RF power supply system parameters and the output residual, effectively depicting the nonlinear dynamic characteristics of the system. Relying on the matrix structure to carry the coupling relationship between global error and parameters, it provides a stable and complete computational basis for subsequent damping correction calculations and parameter update calculations, improving the integrity and standardization of least squares iterative solutions, effectively improving the solution defects caused by multi-parameter redundancy and insufficient excitation, ensuring stable convergence of the subsequent parameter optimization process, and improving the identification and dynamic control of the RF power supply model. 104. Update the parameter vector to be optimized according to the preset optimization algorithm, basic information matrix, residual vector, preset damping factor and preset iteration step size factor to obtain the updated parameters; In this embodiment, the parameter vector to be optimized is updated by combining the optimization algorithm, the basic information matrix, the residual vector, the damping factor, and the iteration step size factor. The damping factor improves the ill-conditioned problem of the basic information matrix, ensuring the stability and reversibility of the iterative solution, and effectively adapting to the complex operating conditions of strong noise and sudden load changes in the RF power supply. The residual vector is used to feed back the system error, and the iteration step size factor is used to constrain the parameter update amplitude, avoiding parameter oscillation and local optima problems, ensuring smooth convergence of parameter iteration, effectively improving the accuracy of RF power supply parameter identification and iterative robustness, and enhancing the system's dynamic adaptability and control stability. 105. Generate semiconductor RF power supply control signals based on the residual vector, update parameters, and preset lookup table; In this embodiment, based on the updated parameters and residual verification results obtained from optimization iteration, and combined with a preset lookup table, the RF power supply control signal is quickly generated, eliminating the need for complex real-time calculations and improving the control command generation rate. The residual can provide real-time feedback on system deviations, and parameter optimization ensures that the input control quantity matches the actual operating conditions of the equipment. The lookup table simplifies the mapping logic from parameters to drive signals. Overall, the output power can track the set value, enhancing the control stability under plasma load fluctuation scenarios and effectively improving the real-time control performance and operational reliability of the semiconductor RF power supply. In this embodiment, complete raw data is provided for the entire process algorithm operation by collecting the parameter vector to be optimized, control input, and actual sampled output from multiple dimensions. The residual vector is calculated by standardizing the residual equation to collect various system errors such as load disturbance and electromagnetic interference. The error data is filtered and denoised to ensure reliability. A basic information matrix is ​​constructed based on the residual equation and the parameter vector to be optimized to characterize the sensitivity and nonlinearity of the system parameters. The parameters are iteratively updated by relying on the damped optimization algorithm with step size constraints to improve the ill-conditioned defects of the basic information matrix, avoid parameter oscillation and local optima defects, and improve the parameter identification accuracy and adaptability to operating conditions. Finally, the control signal is quickly generated by combining the optimized parameters, residual data, and preset lookup table to simplify real-time calculation, improve control response speed, achieve output following target, improve the control stability and overall reliability of the RF power supply under frequent fluctuations in plasma load, and improve the stability of semiconductor manufacturing process.

[0018] A second embodiment of a semiconductor radio frequency power supply control method according to the present invention includes: 201. Analyze the parameter vector to be optimized and the control input based on the pre-trained prediction output model to obtain the model's predicted output; In this embodiment, a semi-physical parameterized prediction model of the RF power supply is pre-established. Based on the principles of RF amplification and electromagnetic transmission, a mechanistic behavior model is constructed. The algebraic topology of the model is determined through offline experimental calibration or theoretical derivation, forming a predictive output model. This model integrates various nonlinear influencing terms such as gain saturation, impedance matching, and temperature coupling. During computation, the parameter vector to be optimized, θ, and the system control input, u, are input for analytical calculation. Simultaneously, a time-varying parameter online estimation method is used to continuously correct the model parameters, making the parameterized model nearly equivalent to the actual physical conditions of the RF power supply. Based on this, the model's predicted output is obtained. Subsequently, relying on the accurate prediction model, a control quantity is generated using model inverse mapping and feedback control algorithms to guide the actual power supply output to stably follow the set target output, providing predictive benchmark data for the next step of solving the residual vector using the residual formula. 202. Calculate the residuals between the model's predicted output and the actual sampled output based on the residual equation to obtain the residual vector; In this embodiment, the residual equation is used to characterize the deviation between the actual output of the RF power supply and the model's predicted output. It can intuitively reflect various system errors such as nonlinear distortion and load disturbance. The system periodically completes residual construction and optimization according to the sampling period, matches the 27.12MHz RF operating frequency, and uses Kalman filtering to reduce noise interference. u is set as the control input including drive amplitude, frequency, and bias voltage; y is the actual hardware sampling output of power, voltage, and current; and yd is the target expected output sent by the host computer. The residual equation is constructed as follows: , In the formula, To incorporate the predictive model of saturation gain, impedance matching, and temperature coupling, θ is the parameter vector to be optimized, consisting of amplifier coefficients and matching network parameters. e represents the measurement noise or disturbance term; the disturbance term mainly originates from the transient discharge impedance fluctuations of the load within the plasma reaction cavity; the measurement noise mainly originates from high-frequency electromagnetic interference within the high-frequency radio frequency, thermal noise from ADC acquisition quantization, and the inherent drift of the sensor conditioning circuit; based on the above formula, the real-time acquired data is substituted into the calculation to obtain the residual vector r required for this iteration, providing basic error data for subsequent Jacobian matrix solving and parameter iterative optimization; In this embodiment, a semi-physical parametric modeling approach is used to build a predictive model for the RF power supply. The model topology is determined based on RF amplification and electromagnetic transmission theory, and nonlinear factors such as gain saturation, impedance, and temperature are integrated. The model is continuously corrected with online parameter identification to reduce the deviation between the model and the actual system and ensure the accuracy and reliability of the predicted output. The residual equation is constructed based on the actual sampled output, the model predicted value, and the disturbance noise. The RF operating frequency is adapted by periodic sampling, and Kalman filtering is introduced to suppress data distortion caused by electromagnetic interference, sampling noise, and plasma load disturbances. The residual vector is calculated. The noise-reduced residual data can provide a reliable data source for subsequent Jacobian matrix solving and parameter iterative optimization, which facilitates the generation of control quantities based on model inverse mapping, so as to realize the actual power supply output tracking the set target and effectively improve the stability of RF power supply modeling and control.

[0019] A third embodiment of a semiconductor radio frequency power supply control method according to the present invention includes: 301. Calculate the gradient of the residual vector and the parameter vector to be optimized to obtain the Jacobian matrix; In this embodiment, the Jacobian matrix is ​​solved by calculating the gradient of the residual vector with respect to the parameter vector to be optimized. The Jacobian matrix expression is: , In the formula, the Jacobian matrix is The first-order partial derivative matrix realizes the local linearization of the nonlinear system at the current parameter point and is used to characterize the sensitive correlation between parameters and residuals; the matrix internal elements Quantize the j-th parameter to be optimized Small changes affect the i-th path residual component The degree of influence; in actual calculations, three methods can be used to complete the gradient solution: analytical differentiation, numerical difference, and automatic differentiation, and the resulting Jacobian matrix is ​​obtained. It fully reflects the sensitivity of residual changes caused by parameter disturbances, and provides the original operation matrix for the subsequent construction of basic information matrix; 302. The basic information matrix is ​​constructed based on the Jacobian matrix and the transpose of the Jacobian matrix; In this embodiment, the Jacobian matrix J and its transpose matrix Jᵀ are multiplied to generate the basic information matrix JᵀJ. This matrix serves as the computational carrier for subsequent least squares iterative solutions. As the coefficient matrix of the undamped Gauss-Newton algorithm, it is used to construct the solution equation for parameter update quantities. It also receives the sensitivity data from the Jacobian matrix and provides key matrix support for the subsequent introduction of the damping factor correction equation and the solution of the original parameter update quantities. In this embodiment, the Jacobian matrix is ​​solved by performing gradient calculations on the residual vector and the parameter vector to be optimized. This quantifies the sensitivity relationship between small perturbations of each parameter and the residual components of the system. The local linearization of the nonlinear system of the RF power supply is achieved using the first-order partial derivative matrix, effectively adapting to the nonlinear characteristics of the system. Simultaneously, multiple solution methods are supported, including analytical differentiation, numerical difference, and automatic differentiation, adapting to the computational needs of different operating conditions and ensuring the flexibility and accuracy of gradient solutions. A basic information matrix is ​​constructed based on the Jacobian matrix and its transpose, fully inheriting the sensitivity characteristics of the parameters and residuals. This provides a stable and reliable computational carrier for least-squares iterative operations. This basic information matrix construction method lays a solid data foundation for subsequent damping correction and parameter update calculations, effectively improving the accuracy of parameter identification and the stability of iterative solutions in the RF power supply system, and ensuring the convergence of subsequent optimization algorithms and the realization of high-precision power supply control.

[0020] A fourth embodiment of a semiconductor radio frequency power supply control method according to the present invention includes: 401. Calculate the residual vector according to the preset objective function to obtain the minimum residual sum of squares; In this embodiment, the objective function expression is: In the formula, Represents the L2 norm. For the parameter vector to be optimized The input error objective function is physically the sum of squared residuals between the measured output of the RF power supply and the predicted output of the model. An optimization criterion is constructed using the squared L2 norm of the residual vector. The residual vector is substituted into the calculation to minimize the sum of squared residuals. The calculation result determines the optimization direction of the parameter iteration. Subsequently, the update amount and iteration step size factor will be combined to complete the iterative update of the parameter vector to be optimized. 402. Calculate the minimum residual sum of squares, basic information matrix, damping factor, preset identity matrix, and residual vector according to the optimization algorithm to obtain the update amount; In this embodiment, the optimization direction is determined based on the aforementioned objective function of minimizing the sum of squared residuals, and the least squares optimization algorithm is selected to update the parameters. The solution; the expression for the basic least squares solution formula for undamped foundations is: , During the operation of the RF power supply, due to insufficient dynamic excitation or the existence of multi-parameter redundancy, the Jacobian matrix J and the current residual vector r can be used in the optimization algorithm to form a highly efficient nonlinear least squares search direction, thereby achieving rapid convergence of control parameters. In the iterative calculation of the least squares optimization algorithm, norm minimization of the multidimensional residual vector can naturally suppress and smooth zero-mean random noise in mathematical processing, preventing ill-conditioned bias in the estimation of the parameter vector θ to be optimized. Affected by the multi-parameter redundancy and insufficient system excitation of the RF power supply, the basic information matrix JᵀJ is prone to ill-conditioned singularity (i.e., close to singularity or with extremely high condition number), leading to non-convergence in the inversion calculation. Therefore, a non-negative scalar damping factor λ and a unit matrix of the same dimension are introduced. The solution equations were modified, and the Levenberg-Marquardt damped least squares optimization algorithm was used to update the parameters. The solution is given by the following expression: ; By introducing a damping term to modify the inverse matrix, the strict symmetric positive definiteness and invertibility of the matrix are mathematically ensured. This improves the numerical convergence stability and robustness of the control algorithm under strong noise and dynamic plasma load abrupt changes. The damping factor enables smooth switching between the Gauss-Newton method and the gradient descent method, guaranteeing the positive definiteness and invertibility of the modified matrix, and suppressing the solution failure problem caused by plasma load abrupt changes and strong environmental noise. Finally, the original parameter update is obtained through calculation. ; 403. Update the parameter vector to be optimized according to the preset boundary constraints, update amount and iteration step size factor to obtain the updated parameters; In this embodiment, the update of the parameter vector to be optimized follows an iterative formula, which is as follows: , In the formula, The parameter vector to be optimized in the k-th iteration refers to the model parameters of the RF power supply (amplifier nonlinear coefficients, matching network parameters, compensation factors, etc.) before the iteration begins, and is the iterative baseline parameter; For the first The parameter vector updated in the next iteration is the new parameter that has been corrected in this round of iteration and is used for the next round of calculation. The scaling correction magnitude of the iteration step size factor is used to fine-tune the search step size, avoiding oscillations caused by excessive single parameter changes, ensuring smooth convergence of parameter iterations, and guaranteeing the smoothness of the convergence process. This step pre-configures parameter boundary constraints such as power upper limit and voltage range, and uses the projection method or penalty term to correct parameters that exceed the limit. At the same time, it sets an upper limit for the single parameter update magnitude to prevent the algorithm from falling into a local optimum due to excessive iteration step size, ensuring that the optimization result approaches the global optimum. When the system residual value is abnormally large and the optimization algorithm cannot work normally, the system automatically switches to the conservative backup control mode of the PID feedback control algorithm, corrects the original parameters based on multiple constraints, and finally outputs updated parameters that are adapted to the actual physical conditions of the RF power supply. In this embodiment, by constructing a residual sum of squares objective function, the direction of parameter iteration optimization is locked. Minimizing the residual norm effectively suppresses random noise in the system, preventing ill-conditioned bias in parameter estimation. The LM damped least squares algorithm optimizes the parameter update amount, effectively solving the problem of ill-conditioned singularities in the fundamental matrix caused by insufficient system excitation and multi-parameter redundancy. Flexible switching of the optimization algorithm via damping factors ensures stable and invertible matrix solving, improving the algorithm's convergence robustness under complex conditions of strong noise and sudden plasma load changes. Simultaneously, combining the iteration step size factor with fine-tuning of the parameter update amplitude, along with power and voltage boundary constraints and update amplitude limits, avoids parameter oscillations and local optima, ensuring smooth iteration convergence and parameters closely matching the actual operating conditions of the equipment. Furthermore, the solution adds a residual abnormal PID fallback control mechanism, effectively mitigating the risk of algorithm failure and comprehensively improving the stability and reliability of RF power supply parameter identification accuracy.

[0021] A fifth embodiment of a semiconductor radio frequency power supply control method according to the present invention includes: 501. Generate the optimal parameter vector based on the residual vector, update parameters, and the preset upper limit of repetition count; In this embodiment, the optimal parameter vector is selected by combining the accuracy threshold of the residual vector and the upper limit of the number of repetitions. The iteration can be terminated in advance when the residual reaches the target to ensure the accuracy of parameter fitting; the calculation is terminated when the iteration exceeds the limit to avoid the algorithm dead loop and waste of computing power. The optimal parameter vector obtained in this way fits the actual working conditions of the equipment, provides a reliable basis for the subsequent conversion of actual control quantities, and improves the stability and real-time performance of RF power supply control. 502. Generate semiconductor RF power supply control signals based on the lookup table and the optimal parameter vector; In this embodiment, a lookup table is used to quickly map and convert the optimal parameter vector into the actual control quantity, simplifying the calculation process from parameters to control quantity and improving the efficiency of control signal generation. When the residual vector is abnormal and cannot obtain effective parameters, the PID feedback control algorithm can be switched to standby control to avoid system loss of control. The actual control quantity is converted into an RF drive signal to realize power regulation, taking into account both control accuracy and operational safety. In this embodiment, the optimal parameter vector is obtained by constraining the residual threshold and the upper limit of the number of repetitions. The iteration is terminated early when the residual vector meets the accuracy requirements, ensuring a high degree of matching between the parameters and the actual operating conditions of the RF power supply. The calculation is terminated when the iteration reaches the limit, preventing the algorithm from consuming hardware resources due to invalid calculations. A lookup table is used to quickly map the optimal parameter vector to the actual control quantity, simplifying the calculation process and accelerating the generation speed of semiconductor RF power supply control signals. In scenarios where parameter optimization fails or residual is abnormal, the PID feedback control algorithm is automatically switched to backup control to avoid the risk of equipment runaway. It not only relies on parameter optimization to achieve high-precision output of the RF power supply, but also improves the system's fault tolerance through fallback control, effectively improving the real-time performance and operational stability of power supply control.

[0022] A sixth embodiment of a semiconductor radio frequency power supply control method according to the present invention includes: 601. Determine whether the norm of the residual vector is less than a preset norm threshold; In this embodiment, the residual vector uses the L2 norm to represent the overall cumulative root mean square error of each sampling channel of the system, with a preset norm threshold (e.g., 10⁻³~10⁻). 4 The norm is the convergence accuracy index pre-set for the algorithm. By comparing the norm of the residual vector with the norm threshold, the first level of convergence judgment is completed. If the norm of the residual vector is less than the norm threshold, it means that the model parameter fitting accuracy meets the standard and the iteration can be terminated directly. If the norm threshold standard is not met, it means that there is still room for improvement in the optimization of the parameter vector to be optimized, and the parameters need to be continuously iterated and updated to provide a judgment basis for starting the cyclic iteration process. 602. If the norm of the residual vector is not less than the norm threshold, then return to perform gradient calculation on the residual vector and the parameter vector to be optimized, until the number of repetitions of the step of performing gradient calculation on the residual vector and the parameter vector to be optimized is equal to the preset upper limit of repetitions, then update the parameters as the optimal parameter vector; In this embodiment, the repetition determination mechanism uses the residual vector norm as the criterion to prioritize parameter fitting and power control accuracy. Simultaneously, an upper limit on the number of repetitions is added as a hard constraint, effectively preventing infinite algorithm loops and excessive controller computational resources, ensuring the continuous and stable operation of the RF control system. The entire determination logic balances control accuracy and equipment computing power. Under complex conditions of variable plasma load and strong electromagnetic interference, it can optimize model parameters as much as possible while preventing program crashes, improving algorithm robustness and overall system reliability. The updated parameters are used as the optimal parameter vector. After the number of repetitions reaches the upper limit, the updated parameters for the current period are selected as the optimal parameter vector. The optimal parameter is the combination of parameters for a nonlinear system model that minimizes error and best matches the physical operating conditions of the RF power supply system within the current control cycle. This combination is then fed down to the lookup table mapping stage to calculate the actual control quantity. Radio frequency power supply regulation is achieved by relying on optimal parameters; In this embodiment, a convergence determination mechanism based on the norm of the residual vector and the upper limit of the number of repetitions is adopted. The L2 norm of the residual is used to measure the overall error of the system, and the iteration is exited preferentially through the norm threshold. When the norm of the residual vector reaches the standard, the optimal parameter vector is output quickly, which effectively ensures the modeling of the RF power supply. When the residual cannot converge quickly, the iteration continues and the number of iterations of the steps is counted. The calculation is terminated based on the preset upper limit of the number of repetitions of the steps to avoid the algorithm's infinite iteration occupying hardware computing power and to ensure the real-time operation of the control system. The optimal parameter vector obtained by dual-condition screening is the model parameter with the highest matching degree and the smallest error with the RF power supply entity under the current operating conditions. It can be directly connected to the subsequent lookup table to convert the actual control quantity. The entire determination logic takes into account both control accuracy and program operation stability. It achieves power supply regulation by relying on residual optimization and improves the robustness of the algorithm by relying on the number of repetitions fallback mechanism, which is suitable for complex operating conditions such as variable plasma load and external interference.

[0023] A seventh embodiment of a semiconductor radio frequency power supply control method according to the present invention includes: 701. Perform mapping analysis on the optimal parameter vector based on the lookup table to obtain the actual control quantity; In this embodiment, a lookup table pre-stores a one-to-one mapping relationship between the optimal parameter vector and the actual control quantity of the RF power supply, relying on the optimal parameter vector output from the previous iteration convergence. Perform lookup table mapping conversion; only when the residual vector norm is less than the preset convergence threshold and the parameters have converged normally in iteration, can an effective optimal parameter vector be obtained. Under the premise of performing table lookup mapping, if the residual value exceeds the limit or the system disturbance is abnormal and parameter optimization cannot be completed during the calculation process, the table lookup mapping process is skipped and the backup control logic of the PID feedback control algorithm is directly triggered; after one-to-one mapping and conversion through the lookup table, the actual control quantity adapted to the current power supply condition is output. This provides quantization input parameters for the generation of radio frequency control signals; 702. Generate semiconductor radio frequency power supply control signals based on actual control quantities; In this embodiment, after obtaining the actual control quantity obtained from the lookup table, the actual control quantity is converted into level, amplitude, and frequency format based on the RF power supply hardware drive standard and power amplifier input specifications. This is then converted into a drive control signal that can be recognized by the power amplifier module and the matching network hardware, and sent to the RF power supply execution unit to complete power regulation, voltage regulation, and current regulation. In this embodiment, a lookup table is used to achieve a fast mapping and conversion from the optimal parameter vector to the actual control quantity. When the parameter iteration converges and the residual meets the accuracy requirements, the actual control quantity can be efficiently calculated and converted into an RF power supply control signal adapted to the hardware, ensuring the regulation of power, voltage, and current. At the same time, an anomaly fallback mechanism is set up. When the residual exceeds the standard or the parameter optimization fails, the backup control mechanism of the PID feedback control algorithm is automatically switched to avoid the runaway problem caused by algorithm anomalies. Overall, the reliability of operation is taken into account, the real-time performance and adaptability of RF power supply control are improved, and the control failure caused by load fluctuations and electromagnetic interference is effectively resisted.

[0024] The above describes a semiconductor radio frequency power supply control method according to an embodiment of the present invention. The following describes a semiconductor radio frequency power supply control device according to an embodiment of the present invention. Please refer to [link / reference]. Figure 2 One embodiment of the semiconductor radio frequency power supply control device of the present invention includes: Data acquisition module 1 is used to acquire the vector of parameters to be optimized, control inputs, and actual sampled outputs during the experiment. The residual calculation module 2 is used to perform residual calculations based on the preset residual equations of the parameter vector to be optimized, the control input, and the actual sampled output to obtain the residual vector. Matrix construction module 3 is used to construct the basic information matrix based on the residual vector and the parameter vector to be optimized; Update module 4 is used to update the parameter vector to be optimized according to the preset optimization algorithm, basic information matrix, residual vector, preset damping factor and preset iteration step size factor to obtain updated parameters; Signal generation module 5 is used to generate semiconductor RF power supply control signals based on residual vectors, update parameters, and preset lookup tables; In this embodiment, complete raw data is provided for the entire process algorithm operation by collecting the parameter vector to be optimized, control input, and actual sampled output from multiple dimensions. The residual vector is calculated by standardizing the residual equation to collect various system errors such as load disturbance and electromagnetic interference. The error data is filtered and denoised to ensure reliability. A basic information matrix is ​​constructed based on the residual equation and the parameter vector to be optimized to characterize the sensitivity and nonlinearity of the system parameters. The parameters are iteratively updated by relying on the damped optimization algorithm with step size constraints to improve the ill-conditioned defects of the basic information matrix, avoid parameter oscillation and local optima defects, and improve the parameter identification accuracy and adaptability to operating conditions. Finally, the control signal is quickly generated by combining the optimized parameters, residual data, and preset lookup table to simplify real-time calculation, improve control response speed, achieve output following target, improve the control stability and overall reliability of the RF power supply under frequent fluctuations in plasma load, and improve the stability of semiconductor manufacturing process.

[0025] Figure 3This is a schematic diagram of a semiconductor radio frequency power control device 900 provided in an embodiment of the present invention. The semiconductor radio frequency power control device 900 can vary significantly due to different configurations or performance characteristics. It may include one or more central processing units (CPUs) 910 (e.g., one or more processors) and a memory 920, and one or more storage media 930 (e.g., one or more mass storage devices) storing application programs 933 or data 932. The memory 920 and storage media 930 can be temporary or persistent storage. The program stored in the storage media 930 may include one or more modules (not shown in the diagram), each module including a series of instruction operations on the semiconductor radio frequency power control device 900. Furthermore, the processor 910 may be configured to communicate with the storage media 930 and execute a series of instruction operations in the storage media 930 on the semiconductor radio frequency power control device 900 to implement the steps of the semiconductor radio frequency power control method provided in the above-described method embodiments.

[0026] A semiconductor radio frequency power supply control device 900 may further include one or more power supplies 940, one or more wired or wireless network interfaces 950, one or more input / output interfaces 960, and / or one or more operating systems 931, such as Windows Server, MacOSX, Unix, Linux, FreeBSD, etc. Those skilled in the art will understand that... Figure 3 The illustrated semiconductor radio frequency power control device structure does not constitute a limitation on a semiconductor radio frequency power control device, and may include more or fewer components than illustrated, or combine certain components, or have different component arrangements.

[0027] 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 a semiconductor radio frequency power supply control method.

[0028] 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.

[0029] 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.

[0030] 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. A semiconductor radio frequency power supply control method, characterized in that, include: Collect the vector of parameters to be optimized, control inputs, and actual sampled outputs during the experiment; The parameter vector to be optimized includes amplifier coefficients and matching network parameters; the control inputs include drive amplitude, frequency, and bias voltage; and the actual sampled outputs include power, voltage, or current. The residuals are calculated based on the preset residual equations of the parameter vector to be optimized, the control input, and the actual sampled output to obtain the residual vector. The residual calculation based on the preset residual equation, the parameter vector to be optimized, the control input, and the actual sampled output are used to obtain the residual vector, including: The pre-trained prediction output model is analyzed to obtain the model's predicted output based on the parameter vector to be optimized and the control input. The residuals are calculated based on the model's predicted output and the actual sampled output according to the residual equation to obtain the residual vector; The basic information matrix is ​​constructed based on the residual vector and the parameter vector to be optimized. The parameter vector to be optimized is updated based on the preset optimization algorithm, basic information matrix, residual vector, preset damping factor and preset iteration step size factor to obtain the updated parameters. The semiconductor RF power supply control signal is generated based on the residual vector, update parameters, and a preset lookup table. The lookup table mapping is performed only if the residual vector norm is less than the preset convergence threshold and the parameters have converged normally to obtain a valid optimal parameter vector. If the residual value exceeds the limit or the system disturbance is abnormal and parameter optimization cannot be completed during the operation, the lookup table mapping process is skipped and the backup control logic of the PID feedback control algorithm is directly triggered. After the lookup table is mapped one by one, the actual control quantity adapted to the current power supply condition is output to provide quantized input parameters for the generation of RF control signals.

2. The semiconductor radio frequency power supply control method as described in claim 1, characterized in that, The basic information matrix constructed based on the residual vector and the parameter vector to be optimized includes: Gradient calculations are performed on the residual vector and the parameter vector to be optimized to obtain the Jacobian matrix; The basic information matrix is ​​constructed based on the Jacobian matrix and its transpose.

3. The semiconductor radio frequency power supply control method as described in claim 2, characterized in that, The step of updating the parameter vector to be optimized according to a preset optimization algorithm, a basic information matrix, a residual vector, a preset damping factor, and a preset iteration step size factor to obtain updated parameters includes: The residual vector is calculated based on the preset objective function to obtain the minimum residual sum of squares; The update amount is obtained by calculating the minimum residual sum of squares, basic information matrix, damping factor, preset identity matrix and residual vector according to the optimization algorithm; The parameter vector to be optimized is updated according to the preset boundary constraints, update amount and iteration step size factor to obtain the updated parameters.

4. The semiconductor radio frequency power supply control method as described in claim 3, characterized in that, The step of generating semiconductor RF power supply control signals based on residual vectors, update parameters, and a preset lookup table includes: Generate the optimal parameter vector based on the residual vector, update parameters, and the preset upper limit for the number of repetitions; The semiconductor RF power supply control signal is generated based on the lookup table and the optimal parameter vector.

5. A semiconductor radio frequency power supply control method as described in claim 4, characterized in that, The step of generating the optimal parameter vector based on the residual vector, update parameters, and a preset upper limit for the number of repetitions includes: Determine whether the norm of the residual vector is less than a preset norm threshold; If the norm of the residual vector is not less than the norm threshold, then return to perform gradient calculation on the residual vector and the parameter vector to be optimized, until the number of times the step of performing gradient calculation on the residual vector and the parameter vector to be optimized is equal to the preset upper limit of the number of repetitions, then update the parameters as the optimal parameter vector.

6. The semiconductor radio frequency power supply control method as described in claim 4, characterized in that, The step of generating semiconductor radio frequency power supply control signals based on a lookup table and an optimal parameter vector includes: The optimal parameter vector is mapped and analyzed based on a lookup table to obtain the actual control quantity; The semiconductor radio frequency power supply control signal is generated based on the actual control quantity.

7. A semiconductor radio frequency power supply control device, characterized in that, include: The data acquisition module is used to collect the vector of parameters to be optimized, control inputs, and actual sampled outputs during the experiment. The parameter vector to be optimized includes amplifier coefficients and matching network parameters; the control inputs include drive amplitude, frequency, and bias voltage; and the actual sampled outputs include power, voltage, or current. The residual calculation module is used to perform residual calculations based on the preset residual equations of the parameter vector to be optimized, the control input, and the actual sampled output to obtain the residual vector. The specific steps include: The pre-trained prediction output model is analyzed to obtain the model's predicted output based on the parameter vector to be optimized and the control input. The residuals are calculated based on the model's predicted output and the actual sampled output according to the residual equation to obtain the residual vector; The matrix construction module is used to construct the basic information matrix based on the residual vector and the parameter vector to be optimized. The update module is used to update the parameter vector to be optimized based on the preset optimization algorithm, basic information matrix, residual vector, preset damping factor and preset iteration step size factor to obtain the updated parameters. The signal generation module is used to generate semiconductor RF power supply control signals based on the residual vector, update parameters, and a preset lookup table. The lookup table mapping is performed only if the residual vector norm is less than the preset convergence threshold and the parameters have converged normally to obtain a valid optimal parameter vector. If the residual value exceeds the limit or the system disturbance is abnormal and parameter optimization cannot be completed during the operation, the lookup table mapping process is skipped and the backup control logic of the PID feedback control algorithm is directly triggered. After the lookup table is mapped one by one, the actual control quantity adapted to the current power supply condition is output to provide quantized input parameters for the generation of RF control signals.

8. A semiconductor radio frequency power supply control device, characterized in that, The semiconductor radio frequency power control device includes: a memory and at least one processor, wherein the memory stores instructions; at least one processor invokes the instructions in the memory to cause the semiconductor radio frequency power control device to perform the various steps of the semiconductor radio frequency power control method as described in any one of claims 1-6.

9. A computer-readable storage medium storing instructions thereon, characterized in that, When the instructions are executed by the processor, they implement the various steps of the semiconductor radio frequency power supply control method as described in any one of claims 1-6.

Images