A wafer electroplating process pre-research and simulation optimization method and system

By using pre-research and simulation optimization methods for wafer electroplating processes, and by utilizing verified process windows and a multi-dimensional quality scoring system, the problems of randomness and low efficiency in parameter adjustment during wafer electroplating process development were solved, achieving efficient and stable process parameter optimization and mass production stability.

CN123069664APending Publication Date: 2026-11-10NINGBO PRAITE SEMICONDUCTOR EQUIPMENT CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611232952.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-14
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing wafer electroplating process development suffers from problems such as strong randomness in parameter adjustment, low efficiency, high cost, one-sided quality evaluation, distorted simulation prediction, and reliance on manual operation for process window updates. It lacks quantifiable evaluation standards and adaptive iterative optimization mechanisms, resulting in long process development cycles, high costs, and unstable quality.

Method used

This paper presents a method for pre-research and simulation optimization of wafer electroplating process. By obtaining the verified process window, combining parameter sensitivity perturbation generation, multi-dimensional quality scoring system and iterative shrinkage strategy, candidate parameter combinations are generated, and simulation prediction and screening are performed to construct a stable and effective process parameter range. The process window is then updated through physical experiments to form a self-iterative closed-loop optimization.

Benefits of technology

Significantly reduce invalid parameter searches, improve process R&D efficiency, shorten development cycle, reduce costs, ensure mass production stability, improve process reuse efficiency, and achieve efficient and stable wafer electroplating process parameter optimization.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123069664A_ABST
    Figure CN123069664A_ABST
Patent Text Reader

Abstract

The application discloses a wafer electroplating process pre-research and simulation optimization method and system, and belongs to the technical field of semiconductor manufacturing and wafer electroplating process research and development. At least one verified wafer electroplating process window is obtained, the process window comprises a reference process parameter combination and a corresponding quality result; a forward and negative disturbance interval is set for the candidate parameter combination by taking the reference process parameter combination as the center and combining the sensitivity of the parameters to the plating layer quality; the candidate parameter combination is introduced into a wafer electroplating simulation model to obtain a predicted plating layer quality result; a feasible parameter set is obtained by screening according to three quality judgment standards of plating layer uniformity, surface defect rate and deep hole filling rate; the value distribution of each process parameter is counted, the edge discrete abnormal value is removed, and a stable and effective value interval is determined; a plurality of rounds of iterative search are performed in the effective value interval, and the candidate target process parameter is output, so that the invalid parameter search is reduced, and the wafer electroplating process pre-research efficiency and parameter optimization stability are improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the fields of semiconductor manufacturing and wafer electroplating process R&D technology, and in particular to a method and system for wafer electroplating process pre-research and simulation optimization, which can be applied to intelligent parameter optimization and multi-physics simulation pre-research of wafer electroplating processes such as advanced packaging TSV, RDL, and copper bumps. Background Technology

[0002] Wafer electroplating is a key process in advanced semiconductor manufacturing. Quality indicators such as coating thickness uniformity, defect rate, and deep-hole fill rate directly affect the stability of internal electrical connections and the reliability of subsequent packaging. The electroplating system involves strong coupling of multiple physical fields—electric field, flow field, and metal ion concentration field—while the micro- and nano-deep holes introduce significant scale-limiting effects. This results in a high-dimensional nonlinear mapping relationship between electroplating process parameters and coating quality that cannot be analyzed using electrochemical theory or empirical formulas.

[0003] The current mainstream process development approach in the industry relies entirely on manual trial and error by R&D personnel: R&D staff adjust process parameters such as voltage, current density, and solution flow rate piecemeal based on industry experience, verifying the coating effect through multiple batches of wafer electroplating experiments. This model has fundamental flaws: there is no unified, quantifiable, multi-indicator joint evaluation standard; parameter quality depends entirely on subjective human judgment. Furthermore, it lacks a standardized, adaptive parameter iteration optimization mechanism, allowing only single-point parameter verification and failing to systematically collect a large number of qualified samples to extract a stable and feasible complete process range. This leads to a series of industry pain points: lengthy process exploration cycles, each round of parameter tuning consumes large amounts of wafers and electroplating solutions, resulting in high costs for trial materials and manpower; the direction of manual parameter tuning is highly random, making it difficult to quickly converge to the optimal parameter combination, resulting in extremely low overall process development efficiency.

[0004] Furthermore, existing process development solutions still have many technical shortcomings in practical implementation: the parameter perturbation mechanism is rigid, using a fixed percentage perturbation without distinguishing parameter sensitivity, easily generating a large number of invalid simulation samples, and lacking an iterative adaptive shrinkage mechanism, resulting in huge computational consumption; the coating quality evaluation index is one-sided, relying solely on apparent indicators and failing to characterize the core causes of packaging failures such as voids inside through-holes, over-plating at the hole opening, and mass production process drift; simulation prediction efficiency is low and extrapolation is distorted, finite element simulation is extremely time-consuming, pure data-driven models lack electrochemical and physical constraints, and parameter out-of-range prediction accuracy is poor; the feasible parameter screening logic is rigid, and full threshold hard filtering is prone to empty feasible sets under high-dimensional parameters, lacking the ability to mine multi-objective compromise optimal solution sets; the stable process range extraction is coarse, simply removing discrete extreme values, failing to distinguish between single-sided / double-sided failure boundaries, and the output range has poor robustness; robustness verification only performs binary qualification judgment, failing to quantify the process's resistance to fluctuations and unable to optimize weak parameters in a targeted manner; process window updates rely on manual operation, failing to adaptively optimize the simulation model, scoring weights, and parameter sensitivity, making it difficult to form a self-iterative closed loop.

[0005] Therefore, there is an urgent need for a research and development obstacle to eliminate the nonlinear black box between wafer electroplating parameters and coating quality, and to provide a process pre-research method with quantifiable evaluation criteria and adaptive iterative optimization, so as to reduce the number of physical prototypes, shorten the process development cycle, and reduce the cost of research and development materials. Summary of the Invention

[0006] This invention addresses the research and development obstacles caused by the nonlinear black box relationship between wafer electroplating parameters and coating quality by providing a process pre-research method with quantifiable evaluation criteria and adaptive iterative optimization. The specific technical solution is as follows: On the one hand, a method for pre-research and simulation optimization of wafer electroplating process is provided, including: Obtain at least one verified wafer plating process window, the verified wafer plating process window including a baseline process parameter combination and a quality result corresponding to the baseline process parameter combination; Centered on the aforementioned baseline process parameter combination, positive and negative perturbation intervals are set according to the sensitivity of parameters to coating quality, and a perturbation generation method that integrates adaptive interval division, layered sampling and iterative shrinkage strategies is adopted to generate a batch of candidate parameter combinations. The candidate parameter combinations are imported into the wafer electroplating simulation model to obtain the predicted coating quality results corresponding to each candidate parameter combination. Construct a comprehensive quality scoring system that includes multi-dimensional orthogonal process indicators, and screen candidate parameter combinations based on the comprehensive quality score, retaining the set of feasible parameters that satisfy all constraints. The numerical distribution of each process parameter within the feasible parameter set is statistically analyzed, marginal discrete outliers are eliminated, and the stable and effective value range of each parameter is determined. Within the effective value range, perform multiple rounds of iterative search on the parameter combination, and in each round, filter the optimal parameters and narrow the search range until the iterative convergence condition is met, and output the candidate target process parameters.

[0007] Optionally, the parameters to be optimized include at least one of electroplating voltage, current density, electroplating solution flow rate, electroplating solution concentration, and electroplating solution pH value.

[0008] Optionally, the perturbation method for generating candidate parameter combinations adopts a triple fusion perturbation mechanism, which includes: The first mechanism is the sensitivity-adaptive asymmetric perturbation interval delineation, which is used to normalize the sensitivity weights based on the parameters output by the simulation model, dynamically adjust the positive and negative perturbation coefficients, automatically narrow the perturbation interval for high-sensitivity parameters, and broaden the exploration range for low-sensitivity parameters. The second mechanism is sensitivity-weighted hierarchical Latin hypercube sampling, which is used to allocate the number of hierarchical samples according to the parameter sensitivity. High-sensitivity parameters increase the sampling density, low-sensitivity parameters reduce redundant sampling, and multi-dimensional parameters are randomly paired to filter out illegal combinations of operating conditions. The third mechanism is the iterative decay truncated Cauchy contraction perturbation, which is used in each iteration to generate perturbation samples by superimposing the current optimal parameters as a new benchmark and using an exponential decay step size to superimpose the truncated Cauchy distribution. In the early stage of the iteration, a large step size is used for global search, and in the later stage, a small step size is used for fine convergence. Meanwhile, all parameter values ​​must not exceed the upper and lower limits of equipment operation, multiple parameters must meet the operating condition coupling constraints, and combinations that do not meet the safety and drug solution stability conditions must be eliminated.

[0009] Optionally, the quality judgment criteria for electroplating are five types of orthogonal process indicators, and adopt... Constrained Pareto multi-objective hierarchical screening, the five types of orthogonal process indices include: Global thickness relative error Characterizes the uniformity of the electric field distribution across the entire wafer; over-plating height at the aperture. , The thickness of the plating layer at the top of the through hole. For wafer planar coating thickness; via void volume ratio Characterizes the reliability of the filling inside the through hole; effective deep hole filling efficiency. ; Parameter drift sensitivity coefficient Drift≤Drift th Characterizes the quality stability under parameter fluctuations; Among them, the Constrained Pareto multi-objective hierarchical screening includes: The first-level hard filter removes severely out-of-specification samples, and the second-level solution solves the non-dominated Pareto front, dividing the set of feasible parameters for mass production into the set of compromise parameters for pre-research.

[0010] Optionally, the iterative convergence criterion includes setting a comprehensive quality score Q, represented as: ,in, , F represents the effective deep hole filling efficiency. The higher the Q value, the better the overall quality of the corresponding candidate parameter combination. The optimization objective is to maximize Q. Set an iteration termination condition, where the difference in the comprehensive score between two adjacent rounds satisfies... , This is the preset convergent minimum value; In each round, the optimal parameter combination is selected based on the comprehensive quality score, and the simulation is repeated to screen the parameters after narrowing the perturbation range.

[0011] Optionally, the method further includes a robust perturbation verification step, wherein the robust perturbation verification employs a Sobol global sensitivity robust hierarchical optimization mechanism, including: Multi-dimensional small-amplitude perturbations are superimposed on the candidate target process parameters. Based on the simulation results after perturbation, Sobol global sensitivity decomposition is performed to determine the sensitivity of each process parameter to the coating quality and locate the highly sensitive weak parameters. The process robustness score is calculated based on the global sensitivity decomposition results and the coating quality changes before and after the disturbance, and the robustness of the candidate target process parameters is classified according to the process robustness score. For candidate target process parameters with medium and low robustness, highly sensitive and weak parameters are fine-tuned and optimized, and robust perturbation verification is re-executed until the preset robustness requirements are met, and a high-robust mass production process solution is output.

[0012] Optionally, the method further includes: After completing the robust perturbation verification, a small batch of physical electroplating is carried out based on the candidate target process parameters to complete the process test, and the physical electroplating verification result data is written back to the verified wafer electroplating process window. The physical electroplating verification result data is written back using a Bayesian adaptive process window update mechanism to achieve three-layer adaptive optimization, including: The wafer electroplating simulation model is updated incrementally based on the physical electroplating verification results data, and the parameter sensitivity and perturbation coefficient are corrected simultaneously. Based on the residual distribution between simulation prediction results and actual measured results, the quality index threshold is adaptively adjusted. With the yield of physical packaging batches as the optimization objective, the weights of the indicators in the comprehensive quality score are adaptively optimized. The updated data obtained through the three-layer adaptive optimization is written back to the verified wafer electroplating process window to form a self-iterative closed-loop optimization system.

[0013] Optionally, the wafer electroplating simulation model is a MO-PI-GP multi-output physical information Gaussian process proxy model; The MO-PI-GP multi-output physical information Gaussian process proxy model adopts a multi-output regionalized covariance architecture to characterize the electrochemical coupling relationship between multiple coating quality indicators, and determines the parameter sensitivity corresponding to each process parameter through the ARD squared exponent kernel. The loss function of the MO-PI-GP multi-output physical information Gaussian process proxy model is embedded with limiting current, electric field uniformity and electrochemical physical constraint regularization terms corresponding to orifice deposition, which are used to apply electrochemical physical constraints to the model prediction process. The MO-PI-GP multi-output physical information Gaussian process surrogate model is used to batch output the predicted mean and prediction uncertainty variance of the quality indicators corresponding to each candidate parameter combination, and to perform incremental updates based on entity samples.

[0014] On the other hand, a wafer electroplating process pre-research and simulation optimization system is provided for executing wafer electroplating process pre-research and simulation optimization methods, the system comprising: The process window management module is used to acquire at least one verified wafer electroplating process window, wherein the verified wafer electroplating process window includes a baseline process parameter combination and a quality result corresponding to the baseline process parameter combination. The parameter perturbation generation module is used to set positive and negative perturbation intervals based on the reference process parameter combination and the sensitivity of the parameters to the coating quality, and to generate a batch of candidate parameter combinations by adopting a perturbation generation method that integrates adaptive interval division, layered sampling and iterative shrinkage strategies. The simulation prediction module is used to import the candidate parameter combinations into the wafer electroplating simulation model to obtain the predicted coating quality results corresponding to each candidate parameter combination. The quality screening module is used to construct a comprehensive quality scoring system that includes multi-dimensional orthogonal process indicators. Based on the comprehensive quality score, it completes the screening of candidate parameter combinations and retains the set of feasible parameters that meet all the constraints. The distributed calculation module is used to statistically analyze the numerical distribution of each process parameter within the feasible parameter set, eliminate marginal discrete outliers, and determine the stable and effective value range of each parameter. The iterative search module is used to perform multiple rounds of iterative search on parameter combinations within the effective value range. In each round, the optimal parameters are selected and the search range is narrowed until the iterative convergence condition is met, and candidate target process parameters are output.

[0015] On the other hand, a computer-readable storage medium is also provided, which stores a computer program that, when executed by a processor, implements a method for pre-research and simulation optimization of wafer electroplating processes.

[0016] Compared with the prior art, the present invention has the following significant advantages.

[0017] Furthermore, instead of blindly traversing the entire parameter space, this invention expands the parameters by perturbation around a verified process window, which can significantly reduce the search for invalid parameters.

[0018] Furthermore, by using quality constraint screening and parameter range back deduction, this invention can automatically form an effective parameter range from parameter combinations that meet quality requirements, thereby improving the efficiency of narrowing down the development path of subsequent processes.

[0019] Furthermore, this invention uses a parameter shrinkage search mechanism to gradually narrow the parameter search range, thus avoiding the problem of random parameter adjustment direction in traditional manual trial and error methods.

[0020] Furthermore, this invention verifies parameter perturbations to ensure that the final process solution not only meets quality requirements but also has the ability to resist parameter fluctuations, thereby improving mass production stability.

[0021] Furthermore, this invention updates the process window by writing back the results of physical experiments, enabling subsequent process development to continuously utilize historical verification results and improving the efficiency of process development reuse. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the pre-research and simulation optimization method for wafer electroplating process in this application; Figure 2 This is an information flow diagram of the pre-research and simulation optimization method for wafer electroplating process in this application; Figure 3 This is a system architecture diagram for the pre-research and simulation optimization of wafer electroplating process in this application. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0024] In this article, "multiple" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. The character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0025] Example 1

[0026] See Figure 1 This embodiment provides a method for pre-research and simulation optimization of wafer electroplating process.

[0027] Instead of blindly exhaustively searching the entire wafer plating process parameter space, this method takes at least one verified wafer plating process window as a starting point and performs parameter perturbation expansion, simulation prediction, quality judgment screening, effective interval back-inference, and multiple rounds of iterative search in the vicinity of the process window to output candidate target process parameters.

[0028] Step S101: Obtain at least one verified wafer plating process window.

[0029] The validated wafer plating process window includes a baseline combination of process parameters and the corresponding quality results. Quality results can include coating uniformity, surface defect rate, deep hole filling rate, and other quality parameters.

[0030] Step S102: Centered on the baseline process parameter combination, positive and negative perturbation intervals are set according to the sensitivity of parameters to coating quality, and a perturbation generation method that integrates adaptive interval division, layered sampling and iterative shrinkage strategy is adopted to generate a batch of candidate parameter combinations.

[0031] Specifically, the system uses a baseline combination of process parameters as the central parameter set. Based on the sensitivity of different process parameters to quality indicators such as coating uniformity, surface defect rate, and deep hole filling rate, positive and negative perturbation intervals are set. Within the determined perturbation intervals, stratified sampling is performed according to the sensitivity of each process parameter, forming a batch of candidate parameter combinations. In subsequent iterations, the currently selected optimal parameter combination is used as the new perturbation center, and the corresponding perturbation range is gradually narrowed, causing the candidate parameter combinations to gradually concentrate from a wider range of exploration towards a more optimal parameter region. Through this perturbation generation method that integrates adaptive interval division, stratified sampling, and iterative narrowing, the effectiveness of candidate parameter combinations and the convergence efficiency of subsequent iterative searches are improved while ensuring the scope of parameter space exploration.

[0032] Step S103: Import the candidate parameter combinations into the wafer electroplating simulation model to obtain the predicted coating quality results corresponding to each candidate parameter combination.

[0033] Wafer electroplating simulation models can predict the coating quality under different combinations of candidate parameters based on equipment geometry, electric field distribution, flow field mass transfer, metal ion concentration distribution, and electrochemical reaction process, thereby obtaining predicted coating uniformity, predicted surface defect rate, and predicted deep hole filling rate.

[0034] Step S104: Construct a comprehensive quality scoring system that includes multi-dimensional orthogonal process indicators, and complete the screening of candidate parameter combinations based on the comprehensive quality score, retaining the set of feasible parameters that satisfy all indicators.

[0035] Specifically, the system selects independent process evaluation indicators from different quality dimensions based on the predicted coating quality results corresponding to candidate parameter combinations, and sets corresponding indicator constraints to construct a multi-dimensional orthogonal process indicator system. Based on the prediction results of each process indicator, the system performs a comprehensive quality evaluation of the candidate parameter combinations, obtaining a comprehensive quality score for each combination. The system then uses the constraints of each process indicator to complete the selection of candidate parameter combinations. Candidate parameter combinations that satisfy all process indicators are retained, forming a feasible parameter set for subsequent determination of stable and effective value ranges and iterative searches.

[0036] Step S105: Statistically analyze the numerical distribution of each process parameter within the feasible parameter set, eliminate marginal discrete outliers, and determine the stable and effective value range of each parameter.

[0037] Specifically, the system statistically analyzes the values ​​of each process parameter in the feasible parameter set, identifies the main distribution range, and removes marginal discrete outliers that are far from the main distribution, have poor stability, or are close to the failure boundary. The remaining parameter distribution range is taken as the stable and effective value range of the corresponding process parameter.

[0038] Step S106: Perform multiple rounds of iterative search on the parameter combination within the valid value range. In each round, the optimal parameter is selected and the search range is narrowed until the iterative convergence condition is met, and the candidate target process parameters are output.

[0039] Specifically, in each iteration, the system generates parameter combinations within the current search range and performs simulation predictions. Based on the comprehensive quality evaluation results, it selects the optimal parameters for the current round and then narrows the search range for the next round around these optimal parameters. This process is repeated until the iteration convergence condition is met, at which point candidate target process parameters are output.

[0040] Therefore, this embodiment improves the effectiveness of candidate parameter combinations by starting from a verified process window and combining parameter perturbation generation methods that integrate parameter sensitivity-based adaptive interval partitioning, hierarchical sampling, and iterative shrinkage. It also obtains predicted coating quality results corresponding to the candidate parameter combinations through a wafer electroplating simulation model. Furthermore, it constructs a comprehensive quality scoring system containing multi-dimensional orthogonal process indicators, combines the comprehensive quality score with indicator constraints to form a feasible parameter set, and then performs back-calculation of stable and effective value intervals and multiple rounds of iterative search. This reduces the traversal of invalid parameters, improves the accuracy of candidate parameter selection and the convergence efficiency of parameter search, and enhances the pre-research efficiency of wafer electroplating processes.

[0041] See Figure 2 In this embodiment, the system first reads the baseline process parameter combinations and corresponding quality results from the verified wafer electroplating process window and generates a set of candidate parameter combinations. After the candidate parameter combinations are input into the wafer electroplating simulation model, the predicted quality results are obtained. The predicted quality results are screened based on the comprehensive quality score of multi-dimensional orthogonal process indicators and the corresponding indicator constraints to form a set of feasible parameter combinations, and further determine the effective value range of the parameters to be optimized. Within the effective value range, an iterative shrinkage search is performed to obtain candidate target parameter combinations, and then parameter perturbation verification is performed. After the perturbation verification is passed, physical electroplating verification is performed, and finally the physical verification results are written back to the process window management module for updating the process window and subsequent parameter optimization process, thereby forming a closed loop of process pre-research.

[0042] Example 2

[0043] Based on Example 1, this example further illustrates the composition of the parameters to be optimized.

[0044] The parameters to be optimized include at least one of the following: electroplating voltage, current density, electroplating solution flow rate, electroplating solution concentration, and electroplating solution pH value.

[0045] Different parameters to be optimized have varying degrees of influence on coating uniformity, surface defect rate, and deep hole filling rate. Among them, electroplating voltage and current density mainly affect the metal deposition rate; electroplating solution flow rate affects the replenishment capacity of metal ions on the wafer surface and inside deep holes; electroplating solution concentration affects the supply capacity of depositable metal ions; and electroplating solution pH value affects the electrochemical reaction environment and deposition stability.

[0046] The system selects one or more parameters as optimization variables based on the object to be optimized, while keeping the values ​​of the remaining parameters in the baseline process parameter combination, thereby reducing the search dimensionality and improving the efficiency of process pre-research.

[0047] Example 3

[0048] Based on Example 1, this example further illustrates the triple fusion perturbation generation method of candidate parameter combination.

[0049] In this embodiment, the candidate parameter combination is generated using a triple perturbation mechanism that combines sensitivity-adaptive asymmetric perturbation interval delineation, sensitivity-weighted hierarchical Latin hypercube sampling, and iterative decay truncated Cauchy contraction perturbation. This allows the parameter perturbation range, parameter sampling density, and perturbation step size at different iteration stages to be adaptively adjusted according to parameter sensitivity and the iteration process.

[0050] The first mechanism is the sensitivity-adaptive asymmetric perturbation interval delineation. Specifically, for any parameter to be optimized, its current baseline value is used as the perturbation center. Based on the normalized sensitivity weight corresponding to the parameter output by the wafer electroplating simulation model, the positive and negative perturbation coefficients are determined respectively, and the upper limit of the positive perturbation and the lower limit of the negative perturbation are determined accordingly. For process parameters with high sensitivity, the perturbation coefficient is reduced accordingly to narrow the perturbation interval and avoid large fluctuations in coating quality indicators caused by small parameter changes; for process parameters with low sensitivity, the perturbation interval is expanded accordingly to increase the exploration range of the parameter space. The positive and negative perturbation coefficients are determined independently, so that the parameter can form an asymmetric perturbation interval according to the influence of different perturbation directions on coating quality.

[0051] The second mechanism is sensitivity-weighted hierarchical Latin hypercube sampling. Within the perturbation range of each process parameter determined by the first mechanism, the number of hierarchical samples is allocated according to the sensitivity of each process parameter, and parameter samples are generated using Latin hypercube sampling. Specifically, for process parameters with high sensitivity, the sampling density within the corresponding parameter range is increased to improve the parameter resolution capability in quality-sensitive areas; for process parameters with low sensitivity, the sampling density is appropriately reduced to decrease redundant samples. The sampled values ​​obtained from each process parameter are then combined in multiple dimensions to form a batch of candidate parameter combinations.

[0052] Furthermore, the candidate parameter combinations are filtered for operational condition validity. The values ​​of each process parameter must not exceed the upper and lower limits of the corresponding electroplating equipment, and it is determined whether the multi-parameter combination meets the preset operational condition coupling constraints, safe operation requirements, and electroplating solution stability conditions. Parameter combinations that do not meet the above conditions are eliminated, and only candidate parameter combinations that meet the corresponding conditions are input into the subsequent wafer electroplating simulation model.

[0053] The third mechanism is the iterative decay truncated Cauchy contraction perturbation. After completing the simulation prediction and quality evaluation of the current round of candidate parameter combinations, the optimal parameter combination obtained in the current round is used as the new benchmark for the next round of perturbation. A perturbation step size that decays exponentially with each iteration is used, combined with a truncated Cauchy distribution, to generate the next round of perturbation samples. Specifically, a relatively large perturbation step size is used in the early stages of iteration, allowing candidate parameters to be globally explored within a larger parameter space. As the number of iterations increases, the perturbation step size gradually decreases, causing candidate parameters to gradually concentrate in the currently optimal parameter region. In the later stages of iteration, a fine search with a small step size is performed. By truncating the perturbation generated by the Cauchy distribution, the generated parameters are kept within the currently allowable parameter range and the equipment's operating range.

[0054] Therefore, the first mechanism adaptively determines the exploration range of different parameters and different perturbation directions based on parameter sensitivity; the second mechanism adjusts the sampling density of different parameters based on parameter sensitivity to form multi-dimensional candidate parameter combinations; and the third mechanism dynamically adjusts the perturbation center and perturbation step size according to the iteration process, so that the parameter search gradually transitions from a large-scale exploration in the early stage to local fine convergence in the later stage. At the same time, the candidate parameter combinations are constrained and filtered by combining equipment operation, working condition coupling, safety and drug stability conditions, thereby reducing invalid parameter combinations, improving the effectiveness of candidate parameter spatial coverage and the convergence efficiency of iterative optimization.

[0055] Example 4

[0056] Based on Example 1, this example further illustrates the five types of orthogonal process indices and the specific method of constrained Pareto multi-objective hierarchical screening based on the five types of orthogonal process indices.

[0057] The system constructs five orthogonal process indices for the predicted coating quality results corresponding to the candidate parameter combinations. These indices include global thickness relative error, hole over-plating height, via void volume ratio, effective deep hole filling efficiency, and parameter drift sensitivity coefficient. The candidate parameter combinations are evaluated from five dimensions: wafer global coating distribution uniformity, via hole over-plating degree, via internal filling reliability, deep hole effective filling ability, and quality stability under parameter fluctuations.

[0058] Among them, the relative error of global thickness satisfy: In the formula, This represents the maximum coating thickness at different locations on the wafer. Minimum coating thickness, This represents the average coating thickness. The overall thickness relative error is also considered. Used to characterize the uniformity of the electric field distribution across the entire wafer. The smaller the value, the smaller the difference in coating thickness between different locations on the wafer, and the more uniform the coating distribution across the entire area.

[0059] Overplating height at the orifice satisfy: In the formula, The thickness of the plating layer at the top of the through hole. The thickness of the coating on the wafer plane. A preset over-plating height threshold is set for the via opening. The degree of local over-deposition in the via opening region is evaluated by the difference between the plating thickness at the top of the via and the plating thickness on the wafer plane. When this occurs, it indicates that the degree of overplating at the orifice is within the corresponding constraint range.

[0060] Percentage of through-hole volume satisfy: In the formula, This refers to the total volume of voids inside the through-hole that are not effectively filled by the electroplating material. This represents the total volume of the space to be filled inside the through-hole. This is a preset threshold for the percentage of void volume in through-holes. The percentage of void volume in through-holes is used to characterize the reliability of the filling inside the through-hole. The smaller the value, the fewer voids inside the through hole, and the higher the degree of complete filling.

[0061] The effective deep hole filling efficiency F satisfies: In the formula, To effectively fill the height, This corresponds to the total depth of the through hole. This sets a preset threshold for effective deep hole filling efficiency. A higher effective deep hole filling efficiency F indicates a higher degree of effective filling by the electroplated material along the depth direction of the through-hole; when... When the value is 1, it indicates that the corresponding candidate parameter combination satisfies the deep hole filling efficiency constraint.

[0062] The parameter drift sensitivity coefficient (Drift) is used to characterize the stability of quality results when process parameters fluctuate, and satisfies the following: Drift≤Drift th In the formula, Drift th A preset parameter drift sensitivity threshold is set. Specifically, a preset range of parameter perturbation is applied to the process parameters in the candidate parameter combination, and the parameter drift sensitivity coefficient Drift is determined based on the degree of change in the corresponding quality indicators before and after the perturbation. The smaller the Drift, the smaller the degree of change in the corresponding quality results when the process parameters fluctuate, and the higher the quality stability of the candidate parameter combination.

[0063] After obtaining the above five types of orthogonal process indicators, the system uses a constrained Pareto multi-objective hierarchical screening method to screen candidate parameter combinations.

[0064] Specifically, a first-level hard filter is first performed to identify severely out-of-standard samples based on the preset constraints corresponding to the five types of orthogonal process indicators, and candidate parameter combinations that severely exceed the corresponding indicator constraints are eliminated to avoid obviously infeasible candidate parameter combinations from entering the subsequent multi-objective screening process.

[0065] For the candidate parameter combinations retained after the first-level hard filtering, a second-level undominated Pareto front solution is performed. The system uses five orthogonal process indices as multi-objective evaluation dimensions to compare the dominance relationships among different candidate parameter combinations. When no other candidate parameter combination can achieve better results on at least one process indice and other process indices do not deteriorate, the corresponding candidate parameter combination is determined as an undominated parameter combination. The undominated Pareto front is formed from all undominated parameter combinations.

[0066] Furthermore, the five types of process indicator constraints are categorized based on the satisfaction of each candidate parameter combination in the non-dominated Pareto front. Candidate parameter combinations that satisfy all process indicators are included in the mass production feasible parameter set. Candidate parameter combinations that do not fully satisfy the mass production constraints but are located on the non-dominated Pareto front and have compromise optimization value among different process indicators are included in the pre-research compromise parameter set for subsequent parameter adjustment and iterative optimization.

[0067] Therefore, by using five orthogonal process indicators—global thickness relative error, orifice overplating height, through-hole void volume ratio, effective deep hole filling efficiency, and parameter drift sensitivity coefficient—candidate parameter combinations are comprehensively evaluated from different quality dimensions. A first-level hard filter eliminates severely out-of-specification samples, and a second-level non-dominated Pareto front solution retains candidate parameter combinations with multi-objective comprehensive optimization value. This avoids the direct elimination of parameter combinations with compromise optimization value due to single-quality indicator evaluation or simple hard threshold screening of all indicators, and provides a parameter basis for subsequent determination of stable and effective value ranges and iterative optimization.

[0068] Example 5

[0069] Based on Example 1, this example further illustrates the iterative convergence determination method based on comprehensive quality score.

[0070] During the parameter iterative search process, the system constructs a comprehensive quality scoring model based on the five types of orthogonal process indicators in Example 4, and comprehensively evaluates the candidate parameter combinations generated in each iteration.

[0071] Specifically, the overall quality score Q is set as follows: .

[0072] Among them, w T w Top w V w F w D The evaluation weights are respectively: relative thickness error across the entire area, over-plating height at the orifice, proportion of void volume in through-holes, effective deep hole filling efficiency, and parameter drift sensitivity coefficient, and: The higher the Q value, the better the overall quality of the corresponding candidate parameter combination. The system uses Q maximization as the parameter selection objective during the iteration process.

[0073] In each iteration, the system generates candidate parameter combinations based on the current search range and obtains the corresponding predicted quality results through the wafer electroplating simulation model. Further, based on the above comprehensive quality score, the system calculates the score value of each candidate parameter combination, selects the parameter combination with the highest score as the optimal parameter combination for the current round, and continues to execute the next round of simulation screening after narrowing the perturbation range around the optimal parameter combination.

[0074] Furthermore, the system sets an iteration termination condition when the difference in the overall quality score between two adjacent rounds meets the following criteria: ,in, This represents the overall quality score corresponding to the optimal parameter combination obtained in the k-th iteration. This represents the overall quality score corresponding to the optimal parameter combination obtained in the (k-1)th iteration. This is the preset convergent minimum value.

[0075] When the above iteration termination condition is met, the system determines that the parameter search process has reached a convergence state, stops the iterative search, and outputs the candidate target process parameter with the highest current comprehensive quality score.

[0076] Therefore, by constructing a comprehensive quality scoring system that integrates five types of orthogonal process indicators, the parameter optimization process can simultaneously consider the wafer's global thickness uniformity, the degree of overplating at the aperture, the reliability of through-hole filling, the deep hole filling efficiency, and the stability of parameter fluctuations. Furthermore, by combining Q-maximization optimization with the convergence judgment of score changes, stable convergence output of candidate target process parameters can be achieved.

[0077] Example 6

[0078] Based on Examples 1 to 5, this example further illustrates a robust hierarchical optimization method based on Sobol global sensitivity.

[0079] After obtaining candidate target process parameters that meet the iterative convergence conditions, the system does not directly determine the candidate target process parameters as the mass production process scheme, but further performs robust perturbation verification on them to evaluate the ability of the candidate target process parameters to maintain the stability of coating quality when the actual process parameters fluctuate.

[0080] Specifically, the system adopts the Sobol global sensitivity robust hierarchical optimization mechanism, which superimposes a preset range of multi-dimensional small-amplitude perturbations on at least one of the candidate target process parameters, namely electroplating voltage, current density, electroplating solution flow rate, electroplating solution concentration, and electroplating solution pH value, to form multiple sets of robustness verification parameter combinations distributed around the candidate target process parameters. Each robustness verification parameter combination is then input into the wafer electroplating simulation model to obtain the corresponding coating quality prediction results.

[0081] Based on the robustness verification parameter combinations and the corresponding coating quality prediction results, the system performs Sobol global sensitivity decomposition to calculate the first-order sensitivity index and the total effect sensitivity index for each process parameter. The first-order sensitivity index characterizes the contribution of independent changes in a single process parameter to the coating quality change, while the total effect sensitivity index characterizes the contribution of the corresponding process parameter's own changes and the interactions between this parameter and other process parameters to the coating quality change.

[0082] The system compares the Sobol sensitivity index corresponding to each process parameter and identifies process parameters with high sensitivity and whose fluctuations are likely to cause significant changes in at least one coating quality indicator as highly sensitive weak parameters, thereby locating the main parameters affecting mass production stability in the candidate target process parameter combination.

[0083] Furthermore, the system combines the degree of change of each quality index relative to the corresponding quality index of the candidate target process parameter under multi-dimensional small-amplitude perturbation conditions with the Sobol global sensitivity decomposition results to calculate the process robustness score corresponding to the candidate target process parameter. The process robustness score is used to characterize the ability of the candidate target process parameter to resist process parameter fluctuations and maintain the stability of coating quality. The smaller the quality change caused by parameter perturbation and the fewer the highly sensitive weak parameters, the higher the corresponding process robustness.

[0084] The system classifies candidate target process parameters into high, medium, and low robustness levels based on a preset robustness scoring threshold. For candidate target process parameters that achieve a high robustness level, they are output as high-robust mass production process solutions. For candidate target process parameters at the medium or low robustness levels, highly sensitive and weak parameters are identified based on the Sobol global sensitivity decomposition results, and targeted fine-tuning is performed on these highly sensitive and weak parameters.

[0085] After completing the targeted fine-tuning, the fine-tuned parameter combination is used as the new parameter combination to be verified. Multi-dimensional small perturbations are then superimposed, and simulation prediction, Sobol global sensitivity decomposition, and process robustness score calculation are performed. When the fine-tuned parameter combination meets the preset robustness requirements, it is output as a highly robust mass production process scheme; if it does not meet the preset robustness requirements, targeted fine-tuning and robust perturbation verification continue based on the highly sensitive and weak parameters.

[0086] Therefore, by using Sobol global sensitivity decomposition, it is possible not only to determine whether the candidate target process parameters are stable under parameter fluctuations, but also to further identify the highly sensitive weak parameters that cause quality fluctuations. Targeted optimization of these weak parameters is then carried out, transforming robustness verification from a simple pass / fail judgment into an optimization process of "global sensitivity decomposition, weak parameter location, robustness scoring, robustness grading, targeted fine-tuning, and re-verification," thereby improving the resilience of the output mass production process solution to parameter fluctuations.

[0087] Example 7

[0088] Based on Example 6, this example further illustrates the physical electroplating verification and the Bayesian adaptive process window update method based on the physical electroplating verification results.

[0089] After the candidate target process parameters have been robustly verified by disturbance and a highly robust mass production process scheme has been obtained, the system conducts small-batch physical electroplating based on the candidate target process parameters to verify the process parameters obtained by simulation optimization in a physical process.

[0090] Specifically, the system controls the wafer plating equipment to perform small-batch physical electroplating according to the candidate target process parameters, and performs quality inspection on the plated wafers to obtain physical measurement results corresponding to the simulated predicted quality indicators. The system matches the simulated prediction results corresponding to the candidate target process parameters with the physical measurement results, and establishes a correlation between the candidate target process parameters, the simulated prediction results, and the physical measurement results to form physical electroplating verification result data.

[0091] The physical electroplating verification results data are written back to the verified wafer electroplating process window using a Bayesian adaptive process window update mechanism. Based on the newly added physical samples, the statistical relationship between process parameters and quality results in the original process window is updated, so that the physical measured data gradually participates in the subsequent simulation prediction, parameter perturbation and quality evaluation process.

[0092] Specifically, the Bayesian adaptive process window update mechanism includes three layers of adaptive optimization.

[0093] The first layer involves adaptive optimization of the wafer electroplating simulation model and parameter perturbation mechanism. The system uses physical electroplating verification results as new physical samples to incrementally update the wafer electroplating simulation model, allowing the model to correct the predictive relationship between process parameters and coating quality indicators based on the new physical samples. Furthermore, based on the updated wafer electroplating simulation model, the system redetermines the parameter sensitivity corresponding to each process parameter, and synchronously corrects the positive perturbation coefficient, negative perturbation coefficient, and corresponding perturbation range based on the updated parameter sensitivity. This enables the generation range of subsequent candidate parameter combinations to be dynamically adjusted according to the physical process verification results.

[0094] The second layer involves adaptive optimization of quality indicator thresholds. The system calculates the residuals between simulation predictions and actual measured results under the same process parameters, and forms a residual distribution based on the residuals corresponding to multiple entity samples. Based on the residual distribution, the prediction deviation and fluctuation of the wafer electroplating simulation model for different quality indicators are determined, and the constraint thresholds corresponding to the quality indicators are adaptively corrected accordingly, ensuring that the quality constraints used in subsequent candidate parameter combination selection remain dynamically consistent with the actual process results.

[0095] The third layer involves adaptive optimization of the weights of the comprehensive quality scoring indicators. The system obtains the batch yield corresponding to different combinations of process parameters after physical packaging, and establishes the correlation between five types of orthogonal process indicators, the comprehensive quality score, and the batch yield of physical packaging. Using the batch yield of physical packaging as the optimization target, the weights of each quality indicator in the comprehensive quality score are adjusted so that the quality indicators with a high degree of influence on the batch yield of physical packaging receive evaluation weights commensurate with their influence in the subsequent comprehensive quality evaluation.

[0096] After completing the above three-layer adaptive optimization, the system writes back the updated wafer electroplating simulation model data, parameter sensitivity and perturbation coefficients, quality index thresholds, and comprehensive quality score index weights to the validated wafer electroplating process window. The updated process window serves as a new validated process window for subsequent wafer electroplating process pre-research and is used for the next round of candidate parameter combination generation, simulation prediction, quality screening, and parameter iterative optimization.

[0097] Therefore, by verifying the physical electroplating, writing back the physical data, updating the simulation model and disturbance mechanism, updating the quality index threshold, updating the comprehensive scoring weight, and having the updated process window re-participate in the optimization, the physical process data can continuously correct the simulation prediction and parameter optimization process, forming a self-iterative closed-loop optimization system based on physical data feedback.

[0098] Example 8

[0099] Based on Example 1, this example further illustrates the specific implementation method of the wafer electroplating simulation model.

[0100] In this embodiment, the wafer electroplating simulation model adopts the MO-PI-GP multi-output physical information Gaussian process surrogate model to establish the nonlinear mapping relationship between wafer electroplating process parameters and multiple coating quality indicators, and outputs the prediction results and prediction uncertainties of the corresponding quality indicators for batch candidate parameter combinations.

[0101] Specifically, the MO-PI-GP multi-output physical information Gaussian process surrogate model takes at least one of the following candidate parameter combinations as model input: electroplating voltage, current density, electroplating solution flow rate, electroplating solution concentration, and electroplating solution pH value. It takes the quality information corresponding to the global thickness relative error, orifice overplating height, through-hole void volume ratio, effective deep hole filling efficiency, and parameter drift sensitivity coefficient as model output to achieve joint prediction of process parameters and multi-dimensional coating quality results.

[0102] The MO-PI-GP multi-output physical information Gaussian process surrogate model employs a multi-output regionalized covariance architecture to jointly model the covariance relationships between different quality indicators, thereby characterizing the coupling relationships between different quality indicators during wafer electroplating caused by electrochemical reactions, electric field distribution, mass transfer, and deposition processes. By sharing relevant information between different outputs, the model can improve the consistency of joint predictions of multiple quality indicators by utilizing information contained in other related quality indicators while predicting a single quality indicator.

[0103] Furthermore, the MO-PI-GP multi-output physical information Gaussian process surrogate model uses the ARD squared exponent to set corresponding scale parameters for different input process parameters, and determines the parameter normalization sensitivity weights based on the scale parameters corresponding to each process parameter. The sensitivity weights corresponding to different process parameters are used to characterize the degree of influence of their changes on the model's prediction quality results, and the parameter sensitivity is output to the candidate parameter combination generation process, which is used for sensitivity adaptive asymmetric perturbation interval delineation and sensitivity weighted hierarchical Latin hypercube sampling in Example 3.

[0104] To reduce the likelihood of predictions that contradict the physical laws of wafer electroplating generated by purely data-driven surrogate models in areas of parameter extrapolation or insufficient sample size, the MO-PI-GP multi-output physical information Gaussian process surrogate model further incorporates electrochemical physical constraints. During model training, limiting current constraints, electric field uniformity constraints, and aperture deposition constraints are constructed as physical constraint regularization terms and embedded into the model loss function. This ensures that the model parameter training process is simultaneously constrained by both sample fitting errors and electrochemical physical constraints. When the model prediction deviates from the corresponding electrochemical physical constraints, the model loss is increased through the corresponding regularization term, thereby suppressing predictions that do not conform to the physical laws of wafer electroplating.

[0105] For each candidate parameter combination input to the model, the MO-PI-GP multi-output physical information Gaussian process surrogate model outputs the predicted mean and predicted uncertainty variance of the corresponding quality index. The predicted mean characterizes the predicted result of the quality index corresponding to the current candidate parameter combination and is used for subsequent comprehensive quality evaluation and candidate parameter combination selection. The predicted uncertainty variance characterizes the uncertainty of the model regarding the corresponding prediction result, providing uncertainty information for subsequent candidate parameter combination selection, stable process range determination, and parameter iterative search.

[0106] Furthermore, after new physical samples are generated from the small-batch physical electroplating in Example 7, the process parameters and corresponding physical quality results from the new physical samples are added to the model sample set. Incremental updates are then performed based on the existing MO-PI-GP multi-output physical information Gaussian process surrogate model to correct the model's representation of the mapping relationship between process parameters and coating quality indicators. After the model update, the parameter sensitivity, the predicted mean of quality indicators, and the prediction uncertainty variance are re-output, and the updated results are used for subsequent candidate parameter generation and process parameter optimization.

[0107] Therefore, the MO-PI-GP multi-output physical information Gaussian process surrogate model achieves joint modeling of multiple quality indicators through a multi-output regionalized covariance architecture, obtains parameter sensitivity through the ARD squared exponential kernel, improves the physical rationality of the prediction results by embedding electrochemical physical constraints in the model loss function, and simultaneously outputs the predicted values ​​and prediction uncertainties of quality indicators. It can also use entity samples for incremental updates, thus providing a unified simulation prediction basis for parameter adaptive perturbation, quality evaluation, robustness verification, and process window closed-loop update.

[0108] On the other hand, a wafer electroplating process pre-research and simulation optimization system is provided, which is used for wafer electroplating process pre-research and simulation optimization methods, including: The process window management module is used to acquire at least one verified wafer electroplating process window, wherein the verified wafer electroplating process window includes a baseline process parameter combination and a quality result corresponding to the baseline process parameter combination. The parameter perturbation generation module is used to set positive and negative perturbation intervals based on the reference process parameter combination and the sensitivity of the parameters to the coating quality, and to generate a batch of candidate parameter combinations by adopting a perturbation generation method that integrates adaptive interval division, layered sampling and iterative shrinkage strategies. The simulation prediction module is used to import the candidate parameter combinations into the wafer electroplating simulation model to obtain the predicted coating quality results corresponding to each candidate parameter combination. The quality screening module is used to construct a comprehensive quality scoring system that includes multi-dimensional orthogonal process indicators. Based on the comprehensive quality score, it completes the screening of candidate parameter combinations and retains the set of feasible parameters that meet all the constraints. The distributed calculation module is used to statistically analyze the numerical distribution of each process parameter within the feasible parameter set, eliminate marginal discrete outliers, and determine the stable and effective value range of each parameter. The iterative search module is used to perform multiple rounds of iterative search on parameter combinations within the effective value range. In each round, the optimal parameters are selected and the search range is narrowed until the iterative convergence condition is met, and candidate target process parameters are output.

[0109] This embodiment provides a computer-readable storage medium. The computer-readable storage medium can be a hard disk, solid-state drive, flash memory, memory card, server storage space, or other media capable of storing programs.

[0110] When the program is executed by the processor, it implements the method described in any one of Embodiments 1 to 8. Specifically, after the processor executes the program, it can read the verified wafer electroplating process window, generate a candidate parameter combination set around the baseline process parameter combination, call the wafer electroplating simulation model to obtain the predicted quality result, screen feasible parameter combinations according to quality constraints, reverse-engineer the effective value range, perform a shrinkage search within the effective value range, perform perturbation verification on the candidate target parameter combination, and receive the physical electroplating verification result to update the process window.

[0111] In one implementation, the program can also be connected to the experimental data system of the wafer electroplating equipment, enabling the physical electroplating verification results to be automatically written into the process window management module. In another implementation, the program can also run as offline process pre-research software, allowing R&D personnel to import the physical electroplating verification results and complete the write-back update.

[0112] Therefore, in this embodiment, the above-mentioned method program is carried on a computer-readable storage medium. This embodiment facilitates the deployment of the wafer electroplating process pre-research and simulation optimization method on R&D workstations, process servers, or enterprise internal process platforms, thereby improving the replicability and engineering implementation capability of the method.

[0113] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0114] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware, or by a program instructing related hardware. The program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk. The above are merely optional embodiments of this application and are not intended to limit the application. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for pre-research and simulation optimization of wafer electroplating process, characterized in that, include: Obtain at least one verified wafer plating process window, the verified wafer plating process window including a baseline process parameter combination and a quality result corresponding to the baseline process parameter combination; Centered on the aforementioned baseline process parameter combination, positive and negative perturbation intervals are set according to the sensitivity of parameters to coating quality, and a perturbation generation method that integrates adaptive interval division, layered sampling and iterative shrinkage strategies is adopted to generate a batch of candidate parameter combinations. The candidate parameter combinations are imported into the wafer electroplating simulation model to obtain the predicted coating quality results corresponding to each candidate parameter combination. Construct a comprehensive quality scoring system that includes multi-dimensional orthogonal process indicators, and screen candidate parameter combinations based on the comprehensive quality score, retaining the set of feasible parameters that satisfy all constraints. The numerical distribution of each process parameter within the feasible parameter set is statistically analyzed, marginal discrete outliers are eliminated, and the stable and effective value range of each parameter is determined. Within the effective value range, perform multiple rounds of iterative search on the parameter combination, and in each round, filter the optimal parameters and narrow the search range until the iterative convergence condition is met, and output the candidate target process parameters.

2. The method according to claim 1, characterized in that, The parameters to be optimized include at least one of the following: electroplating voltage, current density, electroplating solution flow rate, electroplating solution concentration, and electroplating solution pH value.

3. The method according to claim 1, characterized in that, The perturbation method for generating candidate parameter combinations adopts a triple fusion perturbation mechanism, which includes: The first mechanism is the sensitivity-adaptive asymmetric perturbation interval delineation, which is used to normalize the sensitivity weights based on the parameters output by the simulation model, dynamically adjust the positive and negative perturbation coefficients, automatically narrow the perturbation interval for high-sensitivity parameters, and broaden the exploration range for low-sensitivity parameters. The second mechanism is sensitivity-weighted hierarchical Latin hypercube sampling, which is used to allocate the number of hierarchical samples according to the parameter sensitivity. High-sensitivity parameters increase the sampling density, low-sensitivity parameters reduce redundant sampling, and multi-dimensional parameters are randomly paired to filter out illegal combinations of operating conditions. The third mechanism is the iterative decay truncated Cauchy contraction perturbation, which is used in each iteration to generate perturbation samples by superimposing the current optimal parameters as a new benchmark and using an exponential decay step size to superimpose the truncated Cauchy distribution. In the early stage of the iteration, a large step size is used for global search, and in the later stage, a small step size is used for fine convergence. Meanwhile, all parameter values ​​must not exceed the upper and lower limits of equipment operation, multiple parameters must meet the operating condition coupling constraints, and combinations that do not meet the safety and drug solution stability conditions must be eliminated.

4. The method according to claim 1, characterized in that, The quality judgment criteria for electroplating are five categories of orthogonal process indicators, and adopt... Constrained Pareto multi-objective hierarchical screening, the five types of orthogonal process indices include: Global thickness relative error Characterizes the uniformity of the electric field distribution across the entire wafer; over-plating height at the aperture. , The thickness of the plating layer at the top of the through hole. For wafer planar coating thickness; via void volume ratio Characterizes the reliability of the filling inside the through hole; effective deep hole filling efficiency. ; Parameter drift sensitivity coefficient Drift≤Drift th Characterizes the quality stability under parameter fluctuations; Among them, the Constrained Pareto multi-objective hierarchical screening includes: The first-level hard filter removes severely out-of-specification samples, and the second-level solution solves the non-dominated Pareto front, dividing the set of feasible parameters for mass production into the set of compromise parameters for pre-research.

5. The method according to claim 4, characterized in that, The iterative convergence criterion includes setting a comprehensive quality score Q, represented as: ,in, , F represents the effective deep hole filling efficiency. The higher the Q value, the better the overall quality of the corresponding candidate parameter combination. The optimization objective is to maximize Q. Set an iteration termination condition, where the difference in the comprehensive score between two adjacent rounds satisfies... , This is the preset convergent minimum value; In each round, the optimal parameter combination is selected based on the comprehensive quality score, and the simulation is repeated to screen the parameters after narrowing the perturbation range.

6. The method according to claim 1, characterized in that, The method further includes a robust perturbation verification step, which employs a Sobol global sensitivity robust hierarchical optimization mechanism, including: Multi-dimensional small-amplitude perturbations are superimposed on the candidate target process parameters. Based on the simulation results after perturbation, Sobol global sensitivity decomposition is performed to determine the sensitivity of each process parameter to the coating quality and locate the highly sensitive weak parameters. The process robustness score is calculated based on the global sensitivity decomposition results and the coating quality changes before and after the disturbance, and the robustness of the candidate target process parameters is classified according to the process robustness score. For candidate target process parameters with medium and low robustness, highly sensitive and weak parameters are fine-tuned and optimized, and robust perturbation verification is re-executed until the preset robustness requirements are met, and a high-robust mass production process solution is output.

7. The method according to claim 6, characterized in that, The method further includes: After completing the robust perturbation verification, a small batch of physical electroplating is carried out based on the candidate target process parameters to complete the process test, and the physical electroplating verification result data is written back to the verified wafer electroplating process window. The physical electroplating verification result data is written back using a Bayesian adaptive process window update mechanism to achieve three-layer adaptive optimization, including: The wafer electroplating simulation model is updated incrementally based on the physical electroplating verification results data, and the parameter sensitivity and perturbation coefficient are corrected simultaneously. Based on the residual distribution between simulation prediction results and actual measured results, the quality index threshold is adaptively adjusted. With the yield of physical packaging batches as the optimization objective, the weights of the indicators in the comprehensive quality score are adaptively optimized. The updated data obtained through the three-layer adaptive optimization is written back to the verified wafer electroplating process window to form a self-iterative closed-loop optimization system.

8. The method according to claim 1, characterized in that, The wafer electroplating simulation model is a MO-PI-GP multi-output physical information Gaussian process proxy model. The MO-PI-GP multi-output physical information Gaussian process proxy model adopts a multi-output regionalized covariance architecture to characterize the electrochemical coupling relationship between multiple coating quality indicators, and determines the parameter sensitivity corresponding to each process parameter through the ARD squared exponent kernel. The loss function of the MO-PI-GP multi-output physical information Gaussian process proxy model is embedded with limiting current, electric field uniformity and electrochemical physical constraint regularization terms corresponding to orifice deposition, which are used to apply electrochemical physical constraints to the model prediction process. The MO-PI-GP multi-output physical information Gaussian process surrogate model is used to batch output the predicted mean and prediction uncertainty variance of the quality indicators corresponding to each candidate parameter combination, and to perform incremental updates based on entity samples.

9. A pre-research and simulation optimization system for wafer electroplating processes, characterized in that, The system is used to perform the wafer electroplating process pre-research and simulation optimization method according to any one of claims 1 to 8, the system comprising: The process window management module is used to acquire at least one verified wafer electroplating process window, wherein the verified wafer electroplating process window includes a baseline process parameter combination and a quality result corresponding to the baseline process parameter combination. The parameter perturbation generation module is used to set positive and negative perturbation intervals based on the reference process parameter combination and the sensitivity of the parameters to the coating quality, and to generate a batch of candidate parameter combinations by adopting a perturbation generation method that integrates adaptive interval division, layered sampling and iterative shrinkage strategies. The simulation prediction module is used to import the candidate parameter combinations into the wafer electroplating simulation model to obtain the predicted coating quality results corresponding to each candidate parameter combination. The quality screening module is used to construct a comprehensive quality scoring system that includes multi-dimensional orthogonal process indicators. Based on the comprehensive quality score, it completes the screening of candidate parameter combinations and retains the set of feasible parameters that meet all the constraints. The distributed calculation module is used to statistically analyze the numerical distribution of each process parameter within the feasible parameter set, eliminate marginal discrete outliers, and determine the stable and effective value range of each parameter. The iterative search module is used to perform multiple rounds of iterative search on parameter combinations within the effective value range. In each round, the optimal parameters are selected and the search range is narrowed until the iterative convergence condition is met, and candidate target process parameters are output.

10. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the wafer electroplating process pre-research and simulation optimization method according to any one of claims 1 to 8.