An intelligent scheduling system for electroplating production based on big data analysis

Abstract

The application discloses an intelligent scheduling system for electroplating production based on big data analysis, and belongs to the technical field of electroplating production. The system comprises a multi-parameter real-time monitoring unit, an instability risk quantization unit, a risk grade discrimination unit and a closed-loop correction control unit. The multi-parameter real-time monitoring unit is used for collecting multi-dimensional process parameters in the electroplating process in real time. The instability risk quantization unit is used for calculating a plating layer growth instability risk index based on the multi-dimensional process parameters. The risk grade discrimination unit is used for comparing and analyzing the plating layer growth instability risk index with a preset risk index threshold value, and generating a low-risk signal, a medium-risk signal or a high-risk signal. The closed-loop correction control unit is used for starting a first-level correction strategy in response to the medium-risk signal, and starting a second-level correction strategy in response to the high-risk signal, so as to dynamically adjust the electroplating process. The application lays a solid data foundation for precise prediction and control.

Description

Technical Field

[0001] This invention relates to the field of electroplating production, specifically to an intelligent scheduling system for electroplating production based on big data analysis. Background Technology

[0002] In high aspect ratio copper bump plating scenarios at the semiconductor wafer level, traditional scheduling strategies based on ampere-hour management and PID control can experience systematic misjudgments and positive feedback oscillations under certain conditions. This is mainly due to the complex deep contradiction between implicit influencing factors such as the asymmetry of the acoustic field spectrum of the circulating system, the residual stress tensor of the wafer substrate, and the heterogeneity index of the electric double-layer capacitance field, and the time-varying rate of change of the conformational entropy of the time-varying inhibitory molecules. These factors trigger an avalanche-like phase transition in the plating growth mode from conformal growth to destructive dendritic growth through nonlinear stochastic resonance under the coupling of the acoustic field, electromagnetic field, and stress field. This phase transition causes the classical Tafel polarization curve model to completely fail at the mesoscale and produces a counterintuitive system response that reduces current but accelerates abnormal growth, ultimately causing the bump coplanarity error in the critical chip area to deteriorate according to a double exponential function. Therefore, there is an urgent need for an intelligent scheduling system capable of identifying and resolving the above-mentioned complex coupling contradictions.

[0003] The information disclosed in the background section above is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent scheduling system for electroplating production based on big data analysis, so as to solve the problems mentioned in the background art.

[0005] The technical solution of the present invention includes a multi-parameter real-time monitoring unit, an instability risk quantification unit, a risk level discrimination unit, and a closed-loop correction control unit;

[0006] A multi-parameter real-time monitoring unit is used to collect multi-dimensional process parameters during the electroplating process in real time.

[0007] The instability risk quantification unit is used to calculate the coating growth instability risk index based on multidimensional process parameters;

[0008] The risk level discrimination unit is used to compare and analyze the coating growth instability risk index with the preset risk index threshold to generate low risk signal, medium risk signal or high risk signal.

[0009] The closed-loop correction control unit is used to activate the first-level correction strategy in response to a medium-risk signal and the second-level correction strategy in response to a high-risk signal, so as to dynamically adjust the electroplating process.

[0010] Preferably, the multidimensional process parameters include: the conformational entropy change rate of inhibitor molecules, the heterogeneity index of the electric double-layer capacitance field, the concentration of organic matter pyrolysis products, the asymmetry of the acoustic field spectrum, the local electromagnetic field leakage gradient, and the residual stress tensor of the wafer substrate.

[0011] Preferably, the specific operation of the instability risk quantification unit is as follows:

[0012] The heterogeneity index of the electric double-layer capacitor field, the concentration of organic matter pyrolysis products, and the asymmetry of the acoustic field spectrum are obtained from the multidimensional process parameters and coupled to determine the first risk component.

[0013] The conformational entropy change rate of the inhibitor molecule is obtained from the multidimensional process parameters and compared with the preset conformational collapse critical entropy change rate to determine the second risk component.

[0014] The local electromagnetic field leakage gradient is obtained from the multidimensional process parameters and identified as the third risk component;

[0015] By combining the first risk component, the second risk component, and the third risk component, and performing a weighted summation based on preset weighting coefficients, a coating growth instability risk index is generated.

[0016] Preferably, the specific operation of the risk level determination unit is as follows:

[0017] When the coating growth instability risk index is less than or equal to the preset risk index threshold, a low-risk signal is generated.

[0018] When the coating growth instability risk index is greater than the preset risk index threshold, but less than or equal to 1.5 times the preset risk index threshold, a medium risk signal is generated.

[0019] A high-risk signal is generated when the coating growth instability risk index is greater than 1.5 times the preset risk index threshold.

[0020] Preferably, the closed-loop correction control unit responds to low-risk signals and maintains the normal ampere-hour management and PID control strategy.

[0021] Preferably, the primary correction strategy is nonlinear current pulse modulation, and the closed-loop correction control unit generates the corrected time-varying current and controls the electroplating power supply accordingly.

[0022] Preferably, the specific operation of the closed-loop correction control unit in generating the corrected time-varying current is as follows:

[0023] The risk deviation is determined by dividing the difference between the coating growth instability risk index and the preset risk index threshold by the preset risk index threshold.

[0024] Determine the adaptive decay factor based on the risk deviation.

[0025] By combining the adaptive attenuation factor, the preset base DC current, the pulse amplitude modulation depth, and the pulse modulation frequency, a corrected time-varying current is generated.

[0026] Preferably, the secondary correction strategy is intelligent fine-tuning of the plating solution composition. The closed-loop correction control unit determines the required increase in inhibitor concentration and controls the injection of inhibitor into the plating solution.

[0027] Preferably, the specific operation of the closed-loop correction control unit in determining the required increase in inhibitor concentration is as follows:

[0028] Obtain the conformational entropy change rate of the inhibitor molecule from the multidimensional process parameters, and divide it by the preset conformational collapse critical entropy change rate to determine the collapse urgency.

[0029] When the collapse urgency is greater than 1, the incremental concentration of inhibitor that needs to be supplemented is determined by combining the collapse urgency with the preset supplementation coefficient.

[0030] This invention provides an improved intelligent scheduling system for electroplating production based on big data analysis, which has the following improvements and advantages compared with the prior art:

[0031] 1. This invention establishes an unprecedented high-dimensional, high-fidelity process perception system. Existing technologies are usually limited to monitoring macroscopic process parameters and cannot reach the root cause of system instability. This invention uses a multi-parameter real-time monitoring unit to collect multi-dimensional process parameters covering microscopic, mesoscopic, macroscopic, and implicit physical field influences in real time. These parameters specifically include the conformational entropy change rate of inhibitor molecules, the heterogeneity index of the electric double-layer capacitance field, the concentration of organic matter decomposition products, the asymmetry of the acoustic field spectrum, the local electromagnetic field leakage gradient, and the residual stress tensor of the wafer substrate. This multi-dimensional data acquisition enables the system to penetrate the fog of macroscopic phenomena and directly capture the core contradiction that triggers the avalanche-like phase transition of the coating growth mode, laying a solid data foundation for achieving accurate prediction and control.

[0032] 2. This invention achieves a qualitative leap from phenomenon tracking to mechanism prediction. Existing technologies rely on macroscopic models for state assessment, which suffers from significant lag. This invention establishes an instability risk quantification unit, which transforms real-time monitored multidimensional process parameters into a coating growth instability risk index that can proactively predict failure risks through a nonlinear coupling process. This quantification process is not a simple parameter superposition, but rather generates the index by determining a first risk component reflecting interface inhomogeneity and external disturbances, a second risk component quantifying the conformational stability of inhibitors, and a third risk component characterizing the influence of electromagnetic fields, and then performing a weighted summation of the three. This method elevates the assessment of system stability from macroscopic phenomenon tracking to the level of mesoscopic mechanism prediction, making early intervention possible.

[0033] 3. This invention constructs an intelligent, precise, and efficient control logic from hierarchical decision-making to closed-loop correction. Traditional control strategies often have slow responses and limited measures, easily leading to misjudgments or over-intervention. This invention achieves precise risk classification and dynamic response through the collaborative work of a risk level discrimination unit and a closed-loop correction control unit. The risk level discrimination unit generates low-risk, medium-risk, or high-risk signals based on the comparison between the coating growth instability risk index and a preset risk index threshold. The closed-loop correction control unit executes differentiated strategies accordingly: in response to a low-risk signal, the system maintains conventional ampere-hour management and PID control strategies, ensuring production continuity and economy; in response to a medium-risk signal, the system initiates a first-level correction strategy, namely nonlinear current pulse modulation, which generates a corrected time-varying current combined with an adaptive attenuation factor to finely perturb the system without significantly affecting production efficiency, effectively suppressing the deterioration of the instability trend; in response to a high-risk signal, the system initiates a second-level correction strategy, namely intelligent fine-tuning of the plating solution composition, which determines and supplements a precise dose of inhibitor concentration increment by calculating the collapse urgency, chemically reversing the system's runaway trend. Attached Figure Description

[0034] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0035] Figure 1 This is a flowchart of the system of the present invention. Detailed Implementation

[0036] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0037] Example 1

[0038] Please see Figure 1 This invention provides an intelligent scheduling system for electroplating production based on big data analysis, including a multi-parameter real-time monitoring unit, an instability risk quantification unit, a risk level discrimination unit, and a closed-loop correction control unit;

[0039] A multi-parameter real-time monitoring unit is used to collect multi-dimensional process parameters during the electroplating process in real time.

[0040] The instability risk quantification unit is used to calculate the coating growth instability risk index based on multidimensional process parameters;

[0041] The risk level discrimination unit is used to compare and analyze the coating growth instability risk index with the preset risk index threshold to generate low risk signal, medium risk signal or high risk signal.

[0042] The closed-loop correction control unit is used to activate the first-level correction strategy in response to a medium-risk signal and the second-level correction strategy in response to a high-risk signal, so as to dynamically adjust the electroplating process.

[0043] This embodiment provides an intelligent scheduling system for electroplating production based on big data analysis. The system aims to solve the core technical challenges in the high aspect ratio copper bump electroplating scenario in semiconductor wafer-level packaging, namely, the avalanche-like phase transition of the coating growth mode caused by the contradiction between adsorption locking and conformation collapse under the coupling effect of the three fields of acoustic field, electromagnetic field and stress field, and the counterintuitive system response that reduces the current but accelerates abnormal growth.

[0044] In this embodiment, the system operates as a complete and self-consistent technical closed loop, including a multi-parameter real-time monitoring unit, an instability risk quantification unit, a risk level discrimination unit, and a closed-loop correction control unit.

[0045] The purpose of the multi-parameter real-time monitoring unit is to establish a comprehensive and multi-dimensional profile of the electroplating process status, providing real-time and accurate data input for subsequent risk quantification. In this embodiment, the unit deploys a variety of high-precision sensors inside and outside the electroplating tank to collect multi-dimensional process parameters in real time during the electroplating process. These parameters include not only conventional macroscopic process parameters, but also microscopic, mesoscopic, and latent influencing factors that are crucial for revealing the deep-seated mechanisms of system instability.

[0046] The core purpose of the instability risk quantification unit is to transform multidimensional and heterogeneous monitoring parameters into a single, quantifiable index that can proactively predict the risk of coating growth failure. In this embodiment, the unit receives multidimensional process parameters from a multi-parameter real-time monitoring unit and calculates the coating growth instability risk index based on a nonlinear model that deeply integrates multi-physics coupling mechanisms. The construction of this index is a key innovation of this invention that distinguishes it from traditional macroscopic models based on Tafel polarization curves. It elevates system stability assessment from tracking macroscopic phenomena to predicting mesoscopic mechanisms.

[0047] The selection of these parameters and the construction of the model are precisely to characterize the complex effects under the coupling of the acoustic field, electromagnetic field, and stress field, and the resulting contradictory process of inhibitor molecule adsorption-lock-conformation collapse, through quantitative indicators; for example, the conformational entropy change rate of the inhibitor molecule. It directly quantifies the urgency of the core conformational collapse contradiction; the electric double-layer capacitor field heterogeneity index This reflects the inhomogeneity of the interfacial energy, which is a mesoscopic manifestation of the stress field; while the asymmetry of the sound field spectrum... and local electromagnetic field leakage gradient These serve as external disturbance sources, representing the effects of the acoustic field and electromagnetic field on the system, respectively.

[0048] The purpose of the risk level discrimination unit is to provide clear and graded decision-making basis for subsequent closed-loop correction control. In this embodiment, the unit compares and analyzes the coating growth instability risk index calculated by the instability risk quantification unit with one or more preset risk index thresholds in real time. Based on the comparison results, the unit generates low-risk, medium-risk, or high-risk signals, thereby mapping the continuously changing risk index to discrete control levels with clear operational directions.

[0049] The purpose of the closed-loop correction control unit is to execute precise and dynamic intervention strategies based on the output signal of the risk level judgment unit, pulling the electroplating process back from the unstable critical state to the stable growth range. In this embodiment, the unit acts as an intelligent decision-making and execution center, initiating a first-level correction strategy in response to a medium-risk signal and a second-level correction strategy in response to a high-risk signal, so as to dynamically adjust the electroplating process. This hierarchical and closed-loop control logic ensures the timeliness and appropriateness of system intervention, avoiding the lag and misjudgment of traditional control strategies.

[0050] Through the collaborative work of the above four units, this invention constructs a complete intelligent scheduling closed loop from real-time perception to quantitative prediction, and then to hierarchical decision-making and closed-loop control. It can effectively identify and respond to nonlinear random resonance caused by multi-physics coupling, provide early warning and intervene in potential avalanche phase transitions in coating growth modes, and solve the counterintuitive system response problem in the prior art where reducing the current accelerates abnormal growth. It significantly improves the stability, yield and control accuracy of the coplanarity of bumps in key chip areas during the high aspect ratio copper bump electroplating process.

[0051] Example 2

[0052] Multidimensional process parameters include: inhibitor molecule conformational entropy change rate, electric double-layer capacitance field heterogeneity index, organic matter pyrolysis product concentration, acoustic field spectrum asymmetry, local electromagnetic field leakage gradient, and wafer substrate residual stress tensor.

[0053] Based on the technical solution of Example 1, this example further defines and explains the specific composition of the multidimensional process parameters to ensure the comprehensiveness and accuracy of the risk assessment model. The multidimensional process parameters include: the conformational entropy change rate of inhibitor molecules, the heterogeneity index of the electric double-layer capacitance field, the concentration of organic matter pyrolysis products, the asymmetry of the acoustic field spectrum, the local electromagnetic field leakage gradient, and the residual stress tensor of the wafer substrate.

[0054] The conformational entropy change rate of inhibitor molecules refers to the rate at which the three-dimensional coil conformation of the inhibitor macromolecule changes when it is adsorbed onto the cathode surface; its role is to serve as a core microscopic parameter, directly indicating the degree of evolution of the adsorption-lock-conformation collapse contradiction; in this embodiment, the data is obtained by online real-time monitoring of the electroplating solution using spectroscopy or electrochemical impedance spectroscopy.

[0055] The electric double-layer capacitance field heterogeneity index is a dimensionless index used to measure the degree of difference in electric double-layer capacitance in different regions of the wafer surface. Its role is to serve as a key mesoscopic parameter, reflecting the inhomogeneity of local interface energy. These inhomogeneous regions are potential inducing points for abnormal dendrite growth. In this embodiment, the data is obtained by scanning the wafer surface in real time and calculating it using surface potential scanning technology.

[0056] The concentration of organic matter pyrolysis products refers to the concentration of small molecule fragments generated by the pyrolysis of additives in the plating solution under the action of an electric field; its function is to serve as a macroscopic cumulative effect parameter, reflecting the health status or aging degree of the plating solution. High concentrations of pyrolysis products will deteriorate the electroplating performance; in this embodiment, the data is obtained by sampling and analyzing the plating solution circulation loop using an online chromatographic analyzer.

[0057] The asymmetry of the sound field spectrum refers to the degree of asymmetry in the spectral energy distribution of the fluid sound field generated by equipment such as circulating pumps; its role is as a hidden physical field influencing factor, and its specific frequency will resonate with the inhibitor molecules, inducing a nonlinear change in their conformational entropy rate, which is one of the external disturbance sources that trigger instability; in this embodiment, its data is obtained by collecting acoustic sensors deployed on the plating solution circulation pipeline and performing Fourier transform analysis.

[0058] The local electromagnetic field leakage gradient refers to the spatial variation rate of the non-uniform electromagnetic field around the electroplating equipment, especially near the wafer area. Its role is as another implicit physical field influencing factor. This gradient affects the migration and distribution of charged organic additive micelles, exacerbating local concentration unevenness. In this embodiment, the data is obtained by real-time measurement using a high-precision triaxial electromagnetic field probe array deployed near the wafer fixture.

[0059] The residual stress tensor of a wafer substrate refers to the distribution of residual stress within the wafer substrate material. As another implicit physical field influencing factor, residual stress can change the surface energy of the local lattice, forming hot or cold spots for competitive adsorption of additives, and affecting the uniformity of current distribution. In this embodiment, the data is obtained by offline detection of the wafer before electroplating using Raman spectroscopy or X-ray diffraction techniques, or by integrating an online stress monitoring module.

[0060] By clearly defining and monitoring the above six multidimensional process parameters covering the microscopic, mesoscopic, macroscopic, and multi-physics coupling levels, this system has established an unprecedented high-dimensional and high-fidelity electroplating process information space. This enables the instability risk quantification model to capture deep coupling contradictions that traditional technologies cannot observe, greatly improving the accuracy and sensitivity of risk prediction and laying a solid data foundation for achieving truly intelligent scheduling.

[0061] The specific operation of the instability risk quantification unit is as follows:

[0062] The heterogeneity index of the electric double-layer capacitor field, the concentration of organic matter pyrolysis products, and the asymmetry of the acoustic field spectrum are obtained from the multidimensional process parameters and coupled to determine the first risk component.

[0063] The conformational entropy change rate of the inhibitor molecule is obtained from the multidimensional process parameters and compared with the preset conformational collapse critical entropy change rate to determine the second risk component.

[0064] The local electromagnetic field leakage gradient is obtained from the multidimensional process parameters and identified as the third risk component;

[0065] By combining the first risk component, the second risk component, and the third risk component, and performing a weighted summation based on preset weighting coefficients, a coating growth instability risk index is generated.

[0066] Based on the technical solution of Example 1, this example provides a detailed description of the specific operation of the instability risk quantification unit; the core of this operation is to construct and solve the coating growth instability risk index Φ, which can quantify the risk of the system entering the self-organized critical state;

[0067] To dynamically quantify the potential risks of electroplating system instability, this embodiment introduces a coating growth instability risk index. The calculation method is as follows:

[0068]

[0069] in, : Coating growth instability risk index, dimensionless, calculated by this unit, and used as input for subsequent risk level judgment;

[0070] : Weighting coefficients, dimensionless, obtained by training and optimizing a dataset containing thousands of historical production batches, including normal and failure cases, using support vector machine regression, aiming to maximize the model's prediction accuracy for abnormal coating growth events;

[0071] This dataset contains parameter records for at least 5,000 production batches, of which 1,000 are failure cases of abnormal coating growth. We use a support vector machine regression model with the surface roughness and convex coplanarity of the coating as the regression targets, and optimize the weighting coefficients through methods such as grid search to minimize the root mean square error of the model.

[0072] The heterogeneity index of the electric double-layer capacitor field is dimensionless and is acquired by a multi-parameter real-time monitoring unit through surface potential scanning technology.

[0073] Concentration of organic matter pyrolysis products, unit: mol / L, collected by a multi-parameter real-time monitoring unit via an online chromatographic analyzer;

[0074] : Baseline pyrolysis product concentration, unit: mol / L. This parameter is a preset constant, derived from statistical analysis of a large amount of historical data, representing the critical concentration value that indicates the beginning of significant deterioration in the performance of the plating solution. It is used to... Perform dimensionless processing;

[0075] : Sound field-conformation resonance coupling coefficient, dimensionless, obtained together with the weighting coefficient through machine learning methods, used to characterize the sensitivity of sound field spectral asymmetry to the influence of inhibitor conformational entropy change rate;

[0076] The determination of this coefficient is achieved by analyzing a large amount of experimental data. For example, under different sound field frequencies and amplitudes, the change in the conformational entropy rate of the inhibitor molecule is monitored, and its nonlinear relationship is fitted using a polynomial regression model or a Gaussian process regression model in order to obtain the sensitivity coefficient that best characterizes the resonance effect.

[0077] The sound field spectrum is asymmetric and dimensionless, and is collected by a multi-parameter real-time monitoring unit through acoustic sensors.

[0078] : Conformational entropy change rate of inhibitor molecules, unit: J / (K·mol·s), acquired by a multi-parameter real-time monitoring unit through spectral or electrochemical impedance spectroscopy.

[0079] : Critical entropy change rate of conformational collapse, unit: J / (K·mol·s). This is a physical constant threshold preset according to the specific characteristics of the inhibitor molecule, determined by differential scanning calorimetry or molecular dynamics simulation. It represents the critical rate at which the inhibitor conformation undergoes collective and irreversible collapse.

[0080] Local electromagnetic field leakage gradient, unit: T / m, collected by the multi-parameter real-time monitoring unit through the electromagnetic field probe;

[0081] Reference electromagnetic field leakage gradient, unit: T / m. This parameter is a preset constant, derived from electromagnetic field calibration data of the equipment operating under ideal conditions, and is used to... Dimensionless processing is performed; e: natural constant;

[0082] The instability risk quantification model is constructed through the following steps: Molecular dynamics simulations and finite element analysis software, such as COMSOL, are used to simulate the conformational changes of inhibitor molecules under different acoustic and electromagnetic field conditions, generating a dataset containing all parameters. This dataset is then used to train a deep neural network model to predict the stability of coating growth. The trained deep neural network undergoes interpretability analysis, and its core nonlinear relationships and parameter weights are extracted and simplified into the final formula presented in this paper. This formula uses an exponential term... The nonlinear resonance effect of the sound field on the inhibitor conformation was captured, and the result was obtained through the product term. The synergistic amplification effect of interface inhomogeneity and plating solution aging on system risk was simulated; a deep neural network model with at least three hidden layers was trained using this dataset to predict the stability of plating growth; the trained deep neural network was subjected to interpretability analysis, for example, using SHAP or LIME methods.

[0083] Technological Motivation and Computational Logic:

[0084] The design motivation for this formula stems from a deep understanding of the nonlinear stochastic resonance mechanism under the coupling of the sound field, electromagnetic field, and stress field; it is not a simple linear superposition, but rather achieves this through an exponential term. Sum of product terms To characterize the nonlinear synergistic amplification effect among multiple factors;

[0085] To achieve the above calculations, the underlying logic of the instability risk quantification unit lies in:

[0086] Determining the first risk component: Obtaining the heterogeneity index of the electric double-layer capacitor field from multidimensional process parameters. Concentration of organic matter decomposition products and the asymmetry of the sound field spectrum And perform coupling processing to calculate As the first risk component, this component mainly reflects the basic risks caused by the combined effects of interface inhomogeneity, plating solution aging, and external acoustic field disturbances.

[0087] Determining the second risk component: Obtaining the conformational entropy change rate of the inhibitor molecule from multidimensional process parameters. And compared with the preset critical entropy change rate of conformational collapse Perform comparison processing and calculate As the second risk component; this component is the core risk indicator, directly quantifying the urgency of the inhibitor molecule conformation collapsing.

[0088] Determining the third risk component: Obtaining the local electromagnetic field leakage gradient from multidimensional process parameters. Then, after making it dimensionless, multiply it by the weight. This component was identified as the third risk component; this component characterizes the contribution of electromagnetic field disturbances to system stability.

[0089] Weighted summation to generate the risk index: The first, second, and third risk components mentioned above are weighted and summed to generate the final coating growth instability risk index. ;

[0090] Through the aforementioned nonlinear coupling modeling and calculation, this system can integrate multiple seemingly isolated physicochemical parameters into a single index that can accurately and sensitively predict the system's instability trend. This quantitative method shifts from passively responding to observations to actively predicting system states, making early intervention possible and providing a solid decision-making foundation for avoiding catastrophic coating growth failures. It is particularly important to note that the coating growth instability risk index... The real-time calculation formula does not directly include the wafer substrate residual stress tensor term; this is because this parameter, as a relatively static and far-reaching background factor, mainly affects other dynamic parameters indirectly. Specifically, the residual stress distribution of the wafer substrate directly affects the local energy state of the cathode surface, thereby leading to the heterogeneity index of the electric double-layer capacitor field. The generation and changes of these factors; therefore, in constructing a model, offline training of weighted coefficients is crucial. At that time, the residual stress tensor of the wafer substrate is one of the key input features, and its influence has been coupled into the weights of the model by machine learning algorithms; in real-time monitoring, the system tracks its direct consequences—that is... The changes in this value indirectly reflect the impact of residual stress on the wafer substrate, thereby simplifying the complexity of real-time calculations while retaining the ability to perceive the root cause.

[0091] Although the model simplifies multiple complex physical fields into quantifiable indicators, in practical applications, the system continuously optimizes the model online through machine learning algorithms to adapt to dynamic changes and parameter drift during the production process. In addition, the system has a built-in outlier detection module for input parameters. When any parameter exceeds the preset range, additional safety protocols will be triggered, such as suspending production or switching to manual control, to ensure the stability of the system under extreme conditions.

[0092] Example 3

[0093] The specific operation of the risk level determination unit is as follows:

[0094] When the coating growth instability risk index is less than or equal to the preset risk index threshold, a low-risk signal is generated.

[0095] When the coating growth instability risk index is greater than the preset risk index threshold, but less than or equal to 1.5 times the preset risk index threshold, a medium risk signal is generated.

[0096] A high-risk signal is generated when the coating growth instability risk index is greater than 1.5 times the preset risk index threshold.

[0097] The closed-loop correction control unit responds to low-risk signals, maintaining routine ampere-hour management and PID control strategies.

[0098] Based on the above technical solution, this embodiment describes the specific operation of the risk level discrimination unit and the response strategy of the closed-loop correction control unit in a low-risk state.

[0099] The purpose of the risk level determination unit is to classify continuous risk indices. The risk level is converted into a discrete level to guide subsequent operations; the judgment logic is as follows:

[0100] The preset risk index threshold refers to the critical value used to distinguish between system stability and instability tendencies; its function is to serve as a benchmark for risk classification; the method for determining the value is: based on statistical analysis of historical production data, select values ​​that can cover more than 99% of normal production batches. The quantiles of the value fluctuation range are set in conjunction with engineering experience to ensure high specificity and avoid false alarms;

[0101] The threshold The determination is based on the coating growth instability risk index of over 10,000 batches of all normal production batches in the past two years. The statistical analysis selected its 99.9th percentile as... The initial value is set to ensure that false alarms are rarely triggered under normal production conditions; for example, if 99.9% of the batches are normal. If all values ​​are less than 3.5, then... Set to 3.5;

[0102] When the instability risk quantification unit calculates the coating growth instability risk index Less than or equal to the preset risk index threshold At that time, the risk level determination unit generates a low-risk signal;

[0103] When the coating growth instability risk index Greater than the preset risk index threshold and less than or equal to When the risk level is 1.5 times the risk level, the risk level discrimination unit generates a medium-risk signal;

[0104] When the coating growth instability risk index Greater than the preset risk index threshold When the risk level is 1.5 times the normal value, the risk level discrimination unit generates a high-risk signal;

[0105] The threshold This was achieved by analyzing data from over 10,000 batches of normal production over the past two years. Statistical analysis was performed on the values, and the 99.9th percentile was selected as the initial value. Subsequently, by artificially introducing small-amplitude perturbations in specific batches and observing the system response, the initial value was fine-tuned to a point that could accurately capture early signs of instability, while ensuring that the false alarm rate was below 1%. Similarly, for the impulse parameters in the first-level correction strategy... and And the supplementary coefficient in the second-level correction strategy The optimization process should also be explained in detail, for example: designing orthogonal experiments under different conditions. and Under the combined effect, the inhibitory effect on the conformational entropy change rate of the inhibitor was evaluated, and the Bayesian optimization algorithm was used to find the optimal parameter combination to minimize the system recovery time after slight perturbation.

[0106] The closed-loop correction control unit executes corresponding strategies upon receiving different risk signals; specifically, as an implementation described above, when the closed-loop correction control unit responds to a low-risk signal, the system maintains conventional ampere-hour management and PID control strategies, and its output is the base DC current. The system only activates the first-level correction strategy and generates the corrected time-varying current according to the formula when a medium-risk signal is received. This means that the system will continue to produce according to the preset process parameters without initiating any additional corrective interventions. Specifically, conventional ampere-hour management is an open-loop management method that controls the average thickness of the coating by accumulating the total amount of electricity, based on Faraday's law of electrolysis. The PID control strategy, on the other hand, is a closed-loop control method that monitors and adjusts the small fluctuations in the base DC current in real time to maintain the stability of the macroscopic parameters of the electroplating process. Under this low-risk condition, the system assumes that these two conventional strategies are sufficient to cope with minor disturbances in the process.

[0107] Through the aforementioned risk classification logic, the system achieves systematic and precise decision-making; maintaining routine control in low-risk states ensures the continuity and economy of production and avoids unnecessary intervention; while the setting of a 1.5 times threshold provides a clear and verified boundary for distinguishing between medium and high risks, enabling subsequent first and second-level correction strategies to be precisely triggered, ensuring the effective use of control resources, and achieving the best balance between risk and cost.

[0108] The primary correction strategy is nonlinear current pulse modulation. The closed-loop correction control unit generates the corrected time-varying current and controls the electroplating power supply accordingly.

[0109] The specific operation of the closed-loop correction control unit in generating the corrected time-varying current is as follows:

[0110] The risk deviation is determined by dividing the difference between the coating growth instability risk index and the preset risk index threshold by the preset risk index threshold.

[0111] Determine the adaptive decay factor based on the risk deviation.

[0112] By combining the adaptive attenuation factor, the preset base DC current, the pulse amplitude modulation depth, and the pulse modulation frequency, a corrected time-varying current is generated.

[0113] Based on the technical solution of Embodiment 1, this embodiment provides a detailed description of the first-level correction strategy and its specific implementation method; this strategy aims to address medium-risk states by perturbing the system in a non-intrusive manner to suppress the further development of instability trends;

[0114] When the closed-loop correction control unit receives a medium-risk signal, that is When the time is right, the first-level correction strategy is nonlinear current pulse modulation; the unit then generates the corrected time-varying current and controls the output of the electroplating power supply accordingly.

[0115] To generate this corrected current, this embodiment introduces an adaptive nonlinear current pulse modulation model, calculated as follows:

[0116]

[0117] in, The corrected time-varying current, in A, is calculated by this unit and output to the electroplating power supply controller.

[0118] Baseline DC current, unit: A, is the baseline production current value set by the conventional PID strategy;

[0119] : Pulse amplitude modulation depth, dimensionless. This is a preset adjustable parameter, and its value is determined through a series of controlled experiments at different... The values ​​are optimized to find the optimal modulation intensity, ensuring that the inhibitor enrichment is effectively disturbed without excessively affecting the total charge; in this embodiment, the range is set to [0.1, 0.3].

[0120] : Pulse modulation frequency, unit: Hz. This is a preset adjustable parameter. The value is determined by spectral analysis of historical coating growth data to identify characteristic frequencies related to the conformational collapse of inhibitor molecules. The frequency can be an integer multiple or a non-integer multiple of this frequency to achieve resonance enhancement suppression or non-resonance suppression; in this embodiment, It is usually set in the range of 10-100Hz;

[0121] Time, measured in seconds, represents the variable in time-varying current;

[0122] and The meaning is the same as above;

[0123] : Adaptive attenuation factor, dimensionless, dynamically calculated by this unit based on risk deviation; sin: sine function;

[0124] Technological Motivation and Computational Logic:

[0125] The technical motivation of this formula is to actively change the local electric field gradient by introducing a high-frequency, variable-amplitude pulse modulation term, thereby effectively disturbing the excessive enrichment of the inhibitor molecule micelles that cause instability at the interface without significantly changing the average current, and avoiding the collective collapse of their conformation.

[0126] To achieve the above modulation, the closed-loop correction control unit operates as follows:

[0127] Determine the risk deviation: Use the currently calculated coating growth instability risk index. With preset risk index threshold The difference, divided by the preset risk index threshold. This allows for the determination of a standardized risk deviation. ;

[0128] Determine the adaptive decay factor: Calculate the adaptive decay factor based on the risk deviation. Here, the correct form of the adaptive attenuation factor should be: This ensures that the intensity of the pulse modulation is inversely proportional to the degree of risk deviation. The technical logic is that when the coating growth instability risk index... Just exceeded the preset risk index threshold This indicates that the system is in the early stages of instability. At this point, applying a pulse modulation of maximum intensity with a decay factor close to 1 can most effectively provide external disturbance energy, break the unfavorable enrichment state of inhibitor molecules, and pull the system back from the instability critical point to the stable range. Meanwhile, when the coating growth instability risk index... A sustained increase indicates that the system has entered a state of deep instability, and simple physical field disturbances may no longer be able to reverse the trend. At this point, the exponential term in the formula will... The value increases and then decreases significantly, thereby actively weakening the modulation intensity to avoid ineffective intervention. To achieve dynamic and appropriate intervention, this embodiment introduces the adaptive attenuation factor, which ensures that the intensity of pulse modulation is inversely proportional to the degree of risk deviation. That is, the modulation intensity is the maximum when the risk just exceeds the threshold, and the modulation intensity gradually weakens as the risk index increases, so as to avoid excessive intervention that could cause more complex system oscillations and to prepare for possible secondary strategies.

[0129] This inverse relationship design is based on a deep understanding of nonlinear stochastic resonance mechanisms. When system risk... When the threshold is just exceeded, the coating growth mode has not yet fully entered the avalanche phase transition. At this point, the strongest pulse modulation, with an attenuation factor close to 1, can provide the optimal external disturbance energy to pull the system back from the instability critical point to a stable state. As the risk index continues to rise... Much larger This indicates that the system has entered a state of deep instability. Traditional pulse modulation may not be able to effectively reverse the trend, and may even cause more complex system oscillations or accelerate the formation of defects due to excessive disturbance. Therefore, at this time, appropriately reducing the modulation intensity is to avoid ineffective intervention and prepare for the launch of a more powerful secondary correction strategy, thereby achieving a smooth transition and hierarchical progression of the control strategy.

[0130] The design aims to avoid ineffective, high-intensity physical intervention in a system already in a state of deep instability, and instead prepare for the initiation of a more fundamental secondary chemical correction strategy—intelligent fine-tuning of the plating solution composition—thus achieving a smooth transition and hierarchical progression of the control strategy. This modification makes the logic of the entire control system more self-consistent and rational.

[0131] Generate corrected time-varying current: This involves adjusting the adaptive attenuation factor and the preset base DC current. Pulse amplitude modulation depth and pulse modulation frequency Substituting into the above formula, the corrected time-varying current is generated. ;

[0132] By implementing nonlinear current pulse modulation, this system can perform fine and dynamic micro-perturbation intervention on the electroplating process without interrupting production or significantly changing the total ampere-hours. This strategy can effectively suppress the cascade amplification effect of abnormal coating growth, stabilize the system in a controllable medium-risk range, and prevent it from sliding into an irreversible high-risk state, thereby significantly improving process stability while ensuring production efficiency.

[0133] Example 4

[0134] The secondary correction strategy involves intelligent fine-tuning of the plating solution composition. The closed-loop correction control unit determines the required increase in inhibitor concentration and controls the injection of inhibitor into the plating solution.

[0135] The specific steps for the closed-loop correction control unit to determine the required increase in inhibitor concentration are as follows:

[0136] Obtain the conformational entropy change rate of the inhibitor molecule from the multidimensional process parameters, and divide it by the preset conformational collapse critical entropy change rate to determine the collapse urgency.

[0137] When the collapse urgency is greater than 1, the incremental concentration of inhibitor that needs to be supplemented is determined by combining the collapse urgency with the preset supplementation coefficient.

[0138] Based on the technical solution of Example 1, this example provides a detailed description of the secondary correction strategy and its specific implementation method; this strategy is the ultimate safeguard against high-risk situations, aiming to reverse the trend of system runaway by directly changing the chemical composition of the plating solution;

[0139] When the closed-loop correction control unit receives a high-risk signal, that is When this happens, the immediately activated secondary correction strategy is intelligent fine-tuning of the plating solution composition; the core task of this unit is to determine the required increase in inhibitor concentration and precisely control the automatic addition system to inject the corresponding dose of inhibitor into the plating solution.

[0140] To determine this concentration increment, this embodiment introduces an inhibitor supplementation model based on critical collapse, calculated as follows:

[0141]

[0142] in, The required increase in inhibitor concentration, in mol / L, is calculated by this unit and output as an instruction to the automatic plating solution addition system.

[0143] Supplementary coefficient, unit: mol / L. This is a preset adjustable parameter, and its value is determined based on a series of gradient experimental data. The goal of setting this coefficient is to ensure that the conformational entropy change rate of the inhibitor molecule is reduced in the shortest possible time. Pulled back to well below the critical value Within the safe range, while avoiding new problems such as excessive coating stress caused by excessive replenishment;

[0144] and The meaning is the same as above;

[0145] Technological Motivation and Computational Logic:

[0146] The technical motivation for this formula stems from the understanding that inhibitor conformational collapse is the core mutation mechanism leading to uncontrolled coating growth. When the system enters a high-risk state, it indicates that the conformational entropy change rate of the inhibitor molecules has approached or exceeded the critical collapse threshold. At this point, the problem must be solved chemically. By directly supplementing fresh, highly active inhibitors, their concentration on the cathode surface can be effectively increased, thereby regaining an advantage in competitive adsorption with accelerators and reversing the trend of uncontrolled growth.

[0147] Although the calculation of the second-order correction strategy is based solely on the conformational entropy change rate of the inhibitor molecule. This parameter, but its logic is closely related to the entire system; in high-risk situations, The system determines that it is no longer within the range that can be effectively controlled by physical field perturbations, such as current pulse modulation. Instead, the core chemical mechanism—namely, inhibitor conformational collapse—has become the dominant contradiction. Therefore, the technical logic of the secondary correction strategy lies in the fact that when the coating growth instability risk index... When the high-risk threshold is exceeded, the system determines that the failure mode has shifted from a critical state that can be suppressed by physical field perturbations to a deep instability state dominated by core chemical mechanisms, on the verge of an avalanche-like phase transition. At this stage, the most direct and effective intervention is no longer adjusting electrical parameters, but rather addressing the problem at its chemical root. Therefore, this strategy focuses on the multi-dimensional parameter coating growth instability risk index. The shift to the core, the direct cause of instability By precisely replenishing inhibitors, the chemical stability of the plating solution is restored, thereby reversing the trend of uncontrolled coating growth.

[0148] To achieve the above fine-tuning, the closed-loop correction control unit operates as follows:

[0149] Determine the urgency of the collapse: Obtain the latest inhibitor molecule conformational entropy change rate from the multi-parameter real-time monitoring unit. and divide it by the preset critical entropy change rate of conformational collapse. This allows for the determination of the dimensionless key indicator—the urgency of collapse. The collapse urgency refers to the ratio of the current inhibitor conformational entropy change rate to the critical collapse threshold, which directly quantifies the degree of danger of the system from collapsing from its core mechanism.

[0150] Triggering Supplementary Calculation: When the system determines that the collapse urgency is greater than 1, it indicates that the inhibitor conformation has reached or exceeded the critical collapse point, and immediate intervention is necessary. At this time, the collapse urgency is compared with a preset supplementary coefficient. Substitute into the above formula to calculate the required increase in inhibitor concentration. If the collapse urgency is no greater than 1, then the replenishment amount is zero.

[0151] By implementing intelligent fine-tuning of the plating solution composition, this system possesses the capability for final melting and emergency repair under high-risk conditions. In critical moments when the electroplating process is about to completely fail, it can rapidly restore the chemical stability of the plating solution by intelligently calculating and injecting precise doses of inhibitors, regaining control over the microscopic growth pattern and effectively preventing the formation of catastrophic defects such as millimeter-level macroscopic protrusions. This strategy, a rapid intervention at the chemical composition level, is the last and most powerful line of defense to ensure production safety and product yield under extreme conditions.

[0152] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A smart scheduling system for electroplating production based on big data analysis, characterized in that, It includes a multi-parameter real-time monitoring unit, an instability risk quantification unit, a risk level discrimination unit, and a closed-loop correction and control unit; A multi-parameter real-time monitoring unit is used to collect multi-dimensional process parameters during the electroplating process in real time. The instability risk quantification unit is used to calculate the coating growth instability risk index based on multidimensional process parameters; The risk level discrimination unit is used to compare and analyze the coating growth instability risk index with the preset risk index threshold to generate low risk signal, medium risk signal or high risk signal. The closed-loop correction control unit is used to activate the first-level correction strategy in response to a medium-risk signal and the second-level correction strategy in response to a high-risk signal, so as to dynamically adjust the electroplating process. The specific operation of the instability risk quantification unit is as follows: The heterogeneity index of the electric double-layer capacitor field, the concentration of organic matter pyrolysis products, and the asymmetry of the acoustic field spectrum are obtained from the multidimensional process parameters and coupled to determine the first risk component. The conformational entropy change rate of the inhibitor molecule is obtained from the multidimensional process parameters and compared with the preset conformational collapse critical entropy change rate to determine the second risk component. The local electromagnetic field leakage gradient is obtained from the multidimensional process parameters and identified as the third risk component; By combining the first risk component, the second risk component, and the third risk component, and performing a weighted summation based on preset weighting coefficients, a coating growth instability risk index is generated.

2. The intelligent scheduling system for electroplating production based on big data analysis according to claim 1, characterized in that, The multidimensional process parameters include: the conformational entropy change rate of inhibitor molecules, the heterogeneity index of the electric double-layer capacitance field, the concentration of organic matter pyrolysis products, the asymmetry of the acoustic field spectrum, the local electromagnetic field leakage gradient, and the residual stress tensor of the wafer substrate.

3. The intelligent scheduling system for electroplating production based on big data analysis according to claim 1, characterized in that, The specific operation of the risk level determination unit is as follows: When the coating growth instability risk index is less than or equal to the preset risk index threshold, a low-risk signal is generated. When the coating growth instability risk index is greater than the preset risk index threshold, but less than or equal to 1.5 times the preset risk index threshold, a medium risk signal is generated. A high-risk signal is generated when the coating growth instability risk index is greater than 1.5 times the preset risk index threshold.

4. The intelligent scheduling system for electroplating production based on big data analysis according to claim 3, characterized in that, The closed-loop correction control unit responds to low-risk signals and maintains the normal ampere-hour management and PID control strategy.

5. The intelligent scheduling system for electroplating production based on big data analysis according to claim 1, characterized in that, The primary correction strategy is nonlinear current pulse modulation. The closed-loop correction control unit generates the corrected time-varying current and controls the electroplating power supply accordingly.

6. The intelligent scheduling system for electroplating production based on big data analysis according to claim 5, characterized in that, The specific operation of the closed-loop correction control unit in generating the corrected time-varying current is as follows: The risk deviation is determined by dividing the difference between the coating growth instability risk index and the preset risk index threshold by the preset risk index threshold. Determine the adaptive decay factor based on the risk deviation. By combining the adaptive attenuation factor, the preset base DC current, the pulse amplitude modulation depth, and the pulse modulation frequency, a corrected time-varying current is generated.

7. The intelligent scheduling system for electroplating production based on big data analysis according to claim 1, characterized in that, The secondary correction strategy involves intelligent fine-tuning of the plating solution composition. The closed-loop correction control unit determines the required increase in inhibitor concentration and controls the injection of inhibitor into the plating solution.

8. The intelligent scheduling system for electroplating production based on big data analysis according to claim 7, characterized in that, The specific operation by which the closed-loop correction control unit determines the required increase in inhibitor concentration is as follows: Obtain the conformational entropy change rate of the inhibitor molecule from the multidimensional process parameters, and divide it by the preset conformational collapse critical entropy change rate to determine the collapse urgency. When the collapse urgency is greater than 1, the incremental concentration of inhibitor that needs to be supplemented is determined by combining the collapse urgency with the preset supplementation coefficient.

Images