Construction method for predicting double-sided local coating film thickness based on machine learning simulation model

Abstract

The present invention relates to a construction method for predicting the film thickness of double-sided local plating based on a machine learning simulation model, belonging to the field of electroplating technology. The present invention improves and constructs an LR logistic regression model through a machine learning simulation model, and uses the constructed logistic regression model to train the LR logistic regression model, so as to quickly establish the optimal production conditions for the electroplating of various semiconductor devices with different structures, and then obtain the optimal control method for double-sided local electroplating of the plated parts. It can not only achieve the accuracy of predicting the film thickness of the double-sided local plating of semiconductor devices, greatly reduce the error between the film thickness of the actual electroplated product and the predicted value, but also further improve the electroplating module, promote the improvement of the uniformity of the film thickness of the double-sided local plating of semiconductor devices, and meet the requirements of high quality of semiconductor devices.

Description

Technical Field

[0001] The present invention relates to a construction method for predicting the film thickness of a double-sided local coating based on a machine learning simulation model, and belongs to the field of electroplating technology. Background Art

[0002] Metal coating deposition is a unique characterization technology that is sensitive to the local electrochemical-physical environment structure of metal ions during the electroplating process of electronics, and has extensive applications in the fields of physics, electrochemistry, and high-end manufacturing of electroplating. In the electroplating system, under the electrochemical-physical environment, the metal ions are subjected to different magnitudes of constant current, and different coating thicknesses are deposited on the plating area of the product.

[0003] However, the mechanism of the metal coating deposition depends on the type of electroplating system in the electrochemical-physical environment. For the coating thickness deposited in the immersion electroplating system, under the same constant current, a longer electroplating time is required; while for the jet electroplating system, the factors affecting the metal coating deposition are more and more complex; therefore, due to the interference of various influencing factors, there is an obvious difference between the actually measured coating thickness and the predicted result, making it difficult to be applied to the accurate prediction and confirmation of the coating thickness in electroplating production.

[0004] Currently, in the industry, there is already a cross-section test method using a scanning electron microscope (SEM), or a dual beam focused ion beam (Dual Beam FIB) can directly measure the coating thickness by observing the cross-section of the sample with an electron beam while cutting the sample with an ion beam, so as to achieve accurate confirmation of the coating thickness. However, the equipment required for accurate coating thickness testing is expensive and the analysis cost is very high, making it difficult to be widely used in the daily accurate prediction and management of coating thickness in various electroplating systems. Therefore, exploring and creating an accurate, fast, and easy-to-popularize method for calculating the coating thickness has become a very important topic.

[0005] In the prior art, there is only a multi-layer feedback metal coating control method shown in, for example, CN115747750A. This method can find important process parameters during the coating process through an algorithm model, extract these process parameters for optimization to obtain the target film thickness, and thus find the optimal process parameters. However, this metal coating method using plasma vacuum technology neither gives the theoretical basis for predicting and controlling multi-layer metal coating from the metal coating theory, nor describes the actual operation mode of establishing the model. In particular, the important core content of how the multi-layer metal coating theory is organically combined with the established model is not elaborated. Summary of the Invention

[0006] To solve the above problems, the present invention provides a construction method for predicting the thickness of a double-sided local plating film based on a machine learning simulation model. By importing the experimental data of the plating film thickness generated during the electroplating process of an electroplating module under different electroplating conditions, the machine learning simulation model is improved through a system for predicting the thickness of a double-sided local plating film, and an LR logistic regression model is constructed. According to the results of continuous training of the LR logistic regression model, a set of plating film thickness prediction data conditions is obtained, which specifically includes the following steps:[[ID=]1]

[0007] S1: Construct a database based on electroplating conditions, and construct an initial machine learning simulation model linked to the database; set the database in a memory;

[0008] S2: Extract at least two sets of electroplating conditions from the database, and predict the thickness of the double-sided local plating film through a prediction calculation module in the system for predicting the thickness of a double-sided local plating film; use an electroplating module to perform actual electroplating operations on the semiconductor device according to the extracted electroplating conditions, and import the obtained experimental data of the plating film thickness into a plating film thickness threshold discrimination module;

[0009] S3: Compare the predicted thickness of the double-sided local plating film d 预测值 with a target threshold: When the predicted thickness of the double-sided local plating film d 预测值 meets formula D, it enters the LR logistic regression model; while the predicted plating film thickness that does not meet the target threshold d 预测值 cannot enter the training module, and the corresponding electroplating conditions extracted from the database in S2 are corrected according to the measured plating film thickness value d 实测值 to calculate the corrected electroplating conditions, and return to S2 to measure the plating film thickness again d 实测值 and re-discriminate whether the predicted value of the thickness of the double-sided local plating film after the corrected electroplating meets the threshold range until the predicted thickness of the double-sided local plating film d 预测值 d meets formula D before entering the LR logistic regression model;

[0010] Use all the predicted thicknesses of the double-sided local plating film that meet formula D d 预测值 to construct an LR logistic regression model;

[0011] The formula D associated with the target threshold is:

[0012] d 实测值 (100% - 5%) ≤ d ​预测值 ≤ d 实测值 (100% + 5%)

[0013] (Formula D)

[0014] S4: Input the predicted local film thickness of the double-sided coating d 预测值 and the measured value of the electroplating module d 实测值 into the training module in the LR logistic regression operation system, and perform system training through the LR logistic regression operation system. The LR logistic regression operation system includes a training module, a stability generation module, and a coating film thickness calculation module;

[0015] S5: Determine the model morphology based on the coating film thickness result obtained from the operation in S4 and compare it with the target threshold. If the target threshold is reached, it is determined that the electroplating conditions are optimized, and the optimized electroplating conditions are applied to the production debugging of the electroplating device; if the target value is not reached, discard this set of training electroplating conditions and return to S4 for retraining.

[0016] The initial machine learning simulation model includes: a processor, a memory, an electroplating module, a prediction calculation module connected to the electroplating module, a coating film thickness threshold discrimination module, a set partitioning module, and a training module, a stability generation module, a coating thickness calculation module, and a model morphology discrimination module connected in sequence.

[0017] Furthermore, in S1 - S2, during the process of constructing the initial machine learning simulation model linked to the database, it includes using COMSOL Multiphysics finite element simulation software in the prediction calculation module and the electroplating module, and based on Formula C - 2, simulating the local double-sided electroplating process of semiconductor devices according to the prediction conditions obtained from the calculation of Formula C - 2; The Formula C - 2 is:

[0018] d = C I t η k / S ρ Where A (Ampere / dm 2 ) = I / S

[0019] A = ρ d / t C η k

[0020] d = A hh = t C η k / ρ

[0021] (Formula C - 2)

[0022] In Formula C-2, d is the electroplated layer thickness; C is the electrochemical equivalent; I is the current intensity; t is the electroplating time; η k is the cathode current efficiency; S is the surface area of the plating area; ρ is the metal density of the electroplated layer; A is the current density.

[0023] In step S3, the corrected electroplating conditions are predicted and calculated through Formula C-2.

[0024] Furthermore, the predicted conditions include the current density and the electroplating time.

[0025] Furthermore, in step S3, meeting the target threshold means that the predicted double-sided local plating layer thickness d 预测值 is between the minimum threshold and the maximum threshold. The minimum threshold is the plating layer specification value - 5%, and the maximum threshold is the plating layer specification value + 5%.

[0026] Furthermore, the LR logistic regression model is constructed by sensitive condition parameters. The sensitive condition parameters include the type of electroplating solution; the temperature of the electroplating solution; the specific gravity of the electroplating solution; the diameter of the pipeline for transporting the electroplating solution; the type of electroplating power supply; the current intensity; the electroplating time; the cathode current efficiency; the surface area of the plating area; the metal density of the electroplated layer; the current density; the flow rate of the electroplating solution.

[0027] In the process of constructing the prediction model for the double-sided local plating layer thickness of the present invention, when using the COMSOL Multiphysics finite element simulation software in the electroplating module, the influencing factors of the machine learning simulation model for predicting the double-sided local plating layer thickness are composed of sensitive parameters and insensitive parameters. The insensitive parameters include the size of the electroplating tank in the electroplating module, the power of the pump for transporting the electroplating solution, the power of the electroplating power supply, the size of the electroplating mold, the type and size of the electrolytic anode.

[0028] Furthermore, in step S4, the further training of the initial LR logistic regression model is carried out by constructing a system of equations operation using the logistic regression algorithm. The specific steps for constructing the logistic regression model are as follows:

[0029] Set a continuous random variable X Subordinating to the logistic distribution means X having the following distribution function and density function:

[0030]

[0031] (Equation 1)

[0032]

[0033] (Equation 2)

[0034] In Equation 1 and Equation 2, μ is the position parameter, γ is the shape parameter;

[0035] When the position parameter μ tends to zero, γ = 1, the common Sigmoid curve is obtained:

[0036]

[0037] (Equation 3)

[0038] Introducing the binomial logistic regression model, which is a classification model represented by the conditional probability P ( Y | X ), in the form of a parameterized logistic distribution; at this time, the random variable X is taken as a real number, the random variable Y takes values of 1 or 0, and the model parameters are estimated by the supervised learning method;

[0039] Furthermore, the binomial logistic regression model is defined as the following conditional probability distribution:

[0040]

[0041] (Equation 4)

[0042]

[0043] (Equation 5)

[0044] In Equation 4 and Equation 5, x ∈R n is the input, y ∈ {0, 1}, w ∈R n is and b ∈R are parameters, w is called the weight vector, b is called the bias, w·x is w and b 's inner product;

[0045] Expand the weight vector and the input vector, still denoted as w , x , that is:

[0046]

[0047]

[0048] Subsequently, the logistic regression model is expressed as:

[0049]

[0050] (Equation 6)

[0051]

[0052] (Equation 7)

[0053] Next, the odds of an event at a given time is defined as the ratio of the probability of the event occurring to the probability of the event not occurring. If the probability of the event occurring is p , then the odds of the event is:

[0054]

[0055] Then the log odds or log(log odds) = logit of the event is expressed as a function:

[0056]

[0057] For logistic regression, from the above equation P ( Y = 1 | x ) and P ( Y = 0 | x ) we can obtain:

[0058]

[0059] (Equation 8)

[0060] From Equation 8, in the logistic regression model, the log odds of the output Y = 1 is a linear function of the input x , or expressed as, the log odds of the output Y = 1 is a model represented by a linear function of the input x , that is, a preliminary logistic regression model is formed;

[0061] Consider the linear function x · w · x for classifying the input w · x ∈ R, and x ∈ R n+1 , w ∈ R n+1; The linear function is transformed into a probability through the logistic regression definition, expressed as: w · x Converted to probability, expressed as:

[0062]

[0063] (Equation 9)

[0064] When the value of the linear function z = w · x → + ∞, then P ( Y = 1| x ) → 1; When the value of the linear function z = w · x → -∞, then P ( Y = 1| x ) → 0; The obtained graph is the logistic regression model.

[0065] The training module in S4 uses the operation of predicting the coating film thickness parameters that meet the threshold, specifically including the following steps:

[0066] In S3, for the given training set data T = {( x 1, y 1), ( x 2, y ) ···, ( x N , y N )}, where

[0067]

[0068] y i ∈ {0, 1}, the maximum likelihood estimation method is applied to predict the model parameters, so as to obtain the iterative logistic regression model. Another assumption

[0069]

[0070] The likelihood function is expressed as:

[0071]

[0072] The log-likelihood function is expressed as:

[0073]

[0074] (Equation 10)

[0075] Calculate in Equation 10L ( w ) maximum value, thus obtaining w predicted value of w The predicted value of represents the coating film thickness.

[0076] For Equation 10, using the gradient descent method and quasi - Newton method, setting w The maximum likelihood estimate value of is w ∧ , then the learned logistic regression model is:

[0077]

[0078] (Equation 11)

[0079]

[0080] (Equation 12)

[0081] However, the binomial logistic regression model is only introduced for binary classification problems; when used for multi - classification problems, the multinomial logistic regression model needs to be introduced. The specific steps are as follows:

[0082] Set the discrete random variable Y The value set of is {1, 2, ···, K}, that is, there are K classes of class labels. By analogy, the obtained multinomial logistic regression model is:

[0083]

[0084] (Equation 13)

[0085]

[0086] (Equation 14)

[0087] Whether it is binomial logistic regression or multinomial logistic regression, it is a widely used classification algorithm. Its principle is based on statistics and probability theory. By modeling the prior combined probability of input features, and then using the logistic function to map the linear combination to a probability value between 0 and 1, so as to represent the probability of belonging to a sample and which class.

[0088] The machine learning (ML) method proposed in the present invention focuses on using data and algorithms to enable it to imitate the human learning method and gradually improve accuracy. The machine learning method is becoming a new tool in many fields of chemistry and physics. Based on a large amount of data, it learns the relationship between variables. In an electroplating system, the structure of the electroplating device for the local electrochemical-physical environment of metal ions and the coating film thickness are a pair of strongly correlated variables. In order to use the machine learning method for prediction and classification, it is necessary to perform feature-labeled data on the structure of the electroplating device for the local electrochemical-physical environment of metal ions, and apply the formed machine learning algorithm to the prediction or classification algorithm according to some possibly labeled or unlabeled input data. The expansion of the relevant error function generated will be used to compare with the known samples and the simulation algorithm to evaluate the accuracy of the method; the algorithm fits the data points in the training set and adjusts the weights to reduce the difference between the known experimental samples and the algorithm evaluation results. For the machine learning algorithm, repeat this iterative "evaluation and optimization" process to autonomously update and improve the weights, and construct an accurate application of the machine learning algorithm for prediction or classification; the machine learning method has been widely used in the fields of physics and chemistry.

[0089] Advantages of the present invention:

[0090] The present invention improves and constructs an LR logistic regression model through a machine learning simulation model, and uses the constructed logistic regression model to train the LR logistic regression model, so as to quickly establish the optimal electroplating production conditions for various semiconductor devices with different structures, and then obtain the optimal control method for double-sided local electroplating of workpieces. It can not only achieve the accuracy of predicting the double-sided local coating film thickness of semiconductor devices, greatly reduce the error between the coating film thickness of the actual electroplated product, but also further improve the electroplating mold, promote the improvement of the uniformity of the double-sided local coating film thickness of semiconductor devices, and meet the requirements of high quality of semiconductor devices. Brief Description of the Drawings

[0091] Figure 1 It is a schematic structural diagram of a semiconductor device to be electroplated in an embodiment of the present invention.

[0092] Figure 2 It is a schematic diagram of test points in the test of the coating film thickness of a semiconductor device to be electroplated in an embodiment of the present invention.

[0093] Figure 3 It is a flowchart of the overall method of the present invention.

[0094] Figure 4For Embodiment 1 and Embodiment 3 of the present invention, a graph showing the variation trend of the deposited coating film thickness with different current densities is derived through in-depth analysis of the coating film thickness calculation module of the machine learning simulation model.

[0095] Figure 5 For Embodiments 4 to 6 of the present invention, a graph showing the variation trend of the deposited coating film thickness with different current densities is derived through in-depth analysis of the coating film thickness calculation module of the machine learning simulation model.

[0096] Figure 6 A scatter plot of the coating film thickness of the present invention and the electroplating time for double-sided local gold plating film thickness.

[0097] Figure 7 A scatter plot of the coating film thickness of the present invention after optimization and improvement and the electroplating time for double-sided local gold plating film thickness.

[0098] In the figure, 100 is the base material; 200 is the semiconductor device; 20 is the electroplating area; 300 is the machine learning simulation model; 301 is the system for predicting the double-sided local coating film thickness; 302 is the LR logistic regression operation system; 310 is the prediction calculation module; 320 is the electroplating module; 330 is the coating film thickness threshold discrimination module; 340 is the training module; 350 is the stability generation module; 360 is the coating thickness calculation module; 370 is the model morphology discrimination module. Detailed implementation manners

[0099] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0100] In the present invention, unless otherwise clearly defined and limited, the terms "connected", "connected", and "fixed" shall be understood in a broad sense. For example, it may be a fixed connection, a detachable connection, or integrated; it may be a mechanical connection or an electrical connection; it may be directly connected or indirectly connected through an intermediate medium, and it may be the internal communication of two components or the interaction relationship between two components. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0101] The prediction conditions studied in the present invention include current density and electroplating time.

[0102] When using COMSOL Multiphysics finite element simulation software in the electronic electroplating module during the construction process of the predicted double-sided local plating film thickness model of the present invention, the influencing factors of the machine learning simulation model for predicting the double-sided local plating film thickness are composed of sensitive parameters and insensitive parameters. The insensitive parameters include the electroplating tank size, the pumping power of the electroplating solution delivery pump, the electroplating power supply power, the electroplating mold size, the type and size of the electrolytic anode in the electronic electroplating module.

[0103] The present invention provides a construction method for predicting the double-sided local plating film thickness based on a machine learning simulation model. By importing the plating film thickness experimental data generated during the electroplating process of the electronic electroplating module under different electronic electroplating conditions, the machine learning simulation model is improved through a system for predicting the double-sided local plating film thickness, and an LR logistic regression model is constructed. According to the results of continuous training of the LR logistic regression model, a set of plating film thickness prediction data conditions is obtained, which specifically includes the following steps:

[0104] S1: Construct a database based on electronic electroplating conditions, and construct an initial machine learning simulation model linked to the database; set the database in a memory; the initial machine learning simulation model includes: a processor, a memory, an electronic electroplating module, a prediction calculation module and a plating film thickness threshold discrimination module connected to the electronic electroplating module, a set partitioning module, and a training module, a stability generation module, a plating thickness calculation module, and a model morphology discrimination module connected in sequence

[0105] S2: Extract at least two sets of electronic electroplating conditions from the database, and predict the double-sided local plating film thickness through the prediction calculation module in the system for predicting the double-sided local plating film thickness; perform actual electroplating operations on the semiconductor device according to the extracted electronic electroplating conditions using the electronic electroplating module, and import the obtained plating film thickness experimental data into the plating film thickness threshold discrimination module;

[0106] In the above S1 - S2, during the process of constructing the initial machine learning simulation model linked to the database, it includes using COMSOL Multiphysics finite element simulation software in the prediction calculation module and the electronic electroplating module, and based on formula C - 2 for calculation, simulating the local double-sided electroplating process of the semiconductor device according to the prediction conditions obtained by calculating formula C - 2; the formula C - 2 is:

[0107] d = C I t η k / S ρ Wherein, A (ampere / dm 2 ) = I / S

[0108] A = ρ d / t C ηk

[0109] d = A hh = t C η k / ρ

[0110] (Formula C-2)

[0111] In Formula C-2, d is the thickness of the electroplated layer; C is the electrochemical equivalent; I is the current intensity; t is the electroplating time; η k is the cathode current efficiency; S is the surface area of the plating area; ρ is the metal density of the electroplated layer; A is the current density;

[0112] S3: Predict the thickness of the double-sided local plating film d 预测值 Compare with the target threshold. Meeting the target threshold means that the predicted thickness of the double-sided local plating film d 预测值 is between the minimum threshold and the maximum threshold. The minimum threshold is the plating specification value - 5%, and the maximum threshold is the plating specification value + 5%: When the predicted thickness of the double-sided local plating film d 预测值 meets Formula D, then enter the LR logistic regression model; while the predicted plating film thickness that does not meet the target threshold d 预测值 cannot enter the training module, and the corresponding electroplating conditions extracted from the database in S2 are corrected according to the measured plating film thickness value d 实测值 to calculate the corrected electroplating conditions through Formula C-2, return to S2 to measure the plating film thickness again d 实测值 and re-determine whether the predicted thickness of the double-sided local plating film after correcting the electroplating d predicted value meets the threshold range until the predicted thickness of the double-sided local plating film d 预测值 meets Formula D before entering the LR logistic regression model;

[0113] Use all the predicted thicknesses of the double-sided local plating film that meet Formula D d 预测值 to construct the LR logistic regression model;

[0114] The formula D associated with the target threshold is:

[0115] d实测值 (100%-5%)≤ d 预测值 ≤ d 实测值 (100%+5%)

[0116] (Formula D)

[0117] S4: Predict the thickness of the double-sided local coating d 预测值 and measured values of electronic plating modules d 实测值 The training module in the LR logistic regression operation system is imported and the system is trained through the LR logistic regression operation system. The LR logistic regression operation system includes a training module, a stability generation module and a coating thickness calculation module. The LR logistic regression model is constructed by modeling sensitive condition parameters, and the sensitive condition parameters include the type of electroplating solution; the temperature of the electroplating solution; the specific gravity of the electroplating solution; the diameter of the electroplating solution pipeline; the type of electroplating power supply; the current intensity; the electroplating time; the cathode current efficiency; the surface area of the plating area; the metal density of the electroplating layer; the current density; and the flow rate of the electroplating solution.

[0118] Further training of the initial LR logistic regression model uses a logistic regression algorithm to construct an equation system operation. Constructing the logistic regression model specifically includes the following steps:

[0119] Set continuous random variable X The logistic distribution means X It has the following distribution and density functions:

[0120]

[0121] (Equation 1)

[0122]

[0123] (Equation 2)

[0124] In Equation 1 and Equation 2, μ is a positional parameter, γ is the shape parameter;

[0125] When the positional parameter μ tends to zero, γ = 1, we get the common Sigmoid curve:

[0126]

[0127] (Equation 3)

[0128] The introduction of binomial logistic regression model is a classification model composed of conditional probability P (Y | X ), expressed in the form of a parametric logistic distribution; at this time, the random variable X is taken as a real number, the random variable Y takes values of 1 or 0, and the model parameters are estimated by a supervised learning method;

[0129] Furthermore, the binomial logistic regression model is defined as the following conditional probability distribution:

[0130]

[0131] (Equation 4)

[0132]

[0133] (Equation 5)

[0134] In Equations 4 and 5, x ∈R n is the input, y ∈ {0, 1}, w ∈R n is and b ∈R is a parameter, w is called the weight vector, b is called the bias, w·x is w and b 's inner product;

[0135] The weight vector and the input vector are extended and still denoted as w , x , that is:

[0136]

[0137]

[0138] Subsequently, the logistic regression model is expressed as:

[0139]

[0140] (Equation 6)

[0141]

[0142] (Equation 7)

[0143] Next, the odds of an event at a given time is defined as the ratio of the probability of the event occurring to the probability of the event not occurring. If the probability of the event occurring is p , then the odds of the event is:

[0144]

[0145] Then the log odds or log(log odds) = logit of this event is expressed as a function:

[0146]

[0147] For logistic regression, from the above equation P ( Y = 1| x ) and P ( Y = 0| x ) we can get:

[0148]

[0149] (Equation 8)

[0150] From Equation 8, in the logistic regression model, the log odds of the output Y = 1 is a linear function of the input x , or expressed as, the log odds of the output Y = 1 is a model represented by a linear function of the input x , that is, a preliminary logistic regression model is formed;

[0151] Considering the linear function x · w · x for classifying the input w · x ∈R, and x ∈R n+1 , w ∈R n+1 ; By defining logistic regression, the linear function w · x is transformed into a probability, expressed as:

[0152]

[0153] (Equation 9)

[0154] When the value of the linear function z = w · x → +∞, then P ( Y = 1| x ) → 1; When the value of the linear function z = w · x → -∞, then P ( Y = 1| x) → 0; The resulting graph is the logistic regression model.

[0155] The training module in S4 uses the operation of predicting the coating film thickness parameters that meet the threshold, specifically including the following steps:

[0156] In S3, for the given training set data T = {( x 1, y 1), ( x 2, y 2) ···, ( x N , y N )}, where,

[0157]

[0158] y i ∈ {0, 1}, the maximum likelihood estimation method is applied to predict the model parameters, so as to obtain the iterative logistic regression model. Another assumption

[0159]

[0160] The likelihood function is expressed as:

[0161]

[0162] The log-likelihood function is expressed as:

[0163]

[0164] (Equation 10)

[0165] Calculate the maximum value of L ( w ) in Equation 10, so as to obtain w 's predicted value, w 's predicted value represents the coating film thickness.

[0166] For Equation 10, the gradient descent method and the quasi-Newton method are adopted. Assume that w 's maximum likelihood estimation value is w ∧ , then the learned logistic regression model is:

[0167]

[0168] (Equation 11)

[0169]

[0170] (Equation 12)

[0171] However, the binomial logistic regression model is only introduced for binary classification problems; when used for multi-classification problems, the multinomial logistic regression model needs to be introduced. The specific steps are as follows:

[0172] Set the discrete random variable Y The value set of is {1, 2, ···, K}, that is, there are K classes, and the multinomial logistic regression model obtained by analogy is:

[0173]

[0174] (Equation 13)

[0175]

[0176] (Equation 14)

[0177] Whether it is binomial logistic regression or multinomial logistic regression, it is a widely used classification algorithm. Its principle is based on statistics and probability theory. By modeling the prior combination probability of input features, and then using the logistic function to map the linear combination to a probability value between 0 and 1, so as to represent the probability of belonging to a sample and which class;

[0178] S5: According to the coating film thickness result obtained by the operation in S4, the model morphology is discriminated and compared with the target threshold. If the target threshold is reached, it is determined that the electroplating conditions are optimized, and the optimized electroplating conditions are applied to the production debugging work of the electroplating device; if the target value is not reached, the group of training electroplating conditions is discarded and S4 is returned for retraining.

[0179] Using the above method, according to formula B, substituting the relevant various parameters shown in Table 1, in order to obtain a gold plating film thickness of ≥300 nm, Example 1 is set. In Example 1, the lower limit of the available current density range of the gold plating solution is selected as 7 A / dm 2 , since the area of the double-sided local electroplating region of the semiconductor device is 5481 mm 2 =0.5481 dm 2 , therefore, the gold plating current parameter is 7 A / dm 2 ×0.5481 dm 2 =3.84 A.

[0180] Furthermore, according to the method described in Example 1, Example 2 is set. In Example 2, a higher current density of 15 A / dm 2 is selected, and the required gold plating current parameter is 8.22 A.

[0181] The electroplating time is obtained from formula B to get formula C, η kis the current efficiency of the gold-plated cathode. According to the electroplating manual, the theoretical cathode current efficiency of the potassium gold cyanide electroplating solution is 33%. However, as is well known, due to the numerous and complex factors affecting the cathode current efficiency of the gold-plating electroplating solution, the actual cathode current efficiency < 33%.

[0182] t = S ρ d / C I η k

[0183] (Formula C)

[0184] The specific parameters and conditions of Example 1 and Example 2 are shown in Table 1 as follows:

[0185] Table 1

[0186]

[0187] First, according to the various conditions in Table 1, assemble the electronic electroplating module 320, and calculate the corresponding predicted coating film thickness according to the current density and electroplating time of Example 1 and Example 2 shown in Table 2 according to Formula C. Use d 预测值 to represent. When it is necessary to predict that the coating film thickness 309 nm ≥ 300 nm, the electroplating times of Example 1 and Example 2 are shown in Table 2.

[0188] Table 2

[0189]

[0190] Furthermore, according to the current density and predicted electroplating time conditions of Example 1 and Example 2, use the electronic electroplating module 320 to Figure 1 perform gold plating on the semiconductor device 200. The semiconductor device 200 has a total of 58 independent electroplating regions 20, and the electroplating area of each region is the same. The actually electroplated semiconductor device 200 is used for actual testing of the coating film thickness.

[0191] Furthermore, for the method of testing the coating thickness of the electroplated semiconductor device 200, according to Figure 2 mark 5 test points as the central black dots of the electroplating region 20, and use the electronic electroplating device module to perform electroplating treatment on the semiconductor device 200 by spraying to obtain the actual coating film thickness; use d 实测值 to represent, and the results are shown in Table 3.

[0192] Table 3

[0193]

[0194] As can be seen from Table 3, the actual measured coating film thickness values of Example 1 and Example 2, 280.8 nm and 276.4 nm, are lower than the minimum threshold of 285 nm. Therefore, they do not meet the restrictions of the coating film thickness threshold discrimination module 330 of Formula D, and the obtained data cannot be incorporated into the simulation model database.

[0195] Set the measured coating film thickness d 实测值 And the specified film thickness of the semiconductor device ≥ 300 nm, as well as the predicted coating film thickness d 预测值 Must satisfy the following restrictive requirements of the coating film thickness threshold discrimination module 330 in the machine learning simulation model as Figure 3 Shown:

[0196] 300 nm (100% - 5%) ≤ d 预测值 ≤ 300 nm (100% + 5%)

[0197] 285 nm ≤ d 预测值 ≤ 315 nm

[0198] Can be summarized into a general formula as:

[0199] d 实测值 (100% - 5%) ≤ d 预测值 ≤ d 实测值 (100% + 5%)

[0200] (Formula D)

[0201] That is, only the predicted coating film thickness values between the minimum threshold of 285 nm and the maximum threshold of 315 nm described by Formula D can be input into the database for the construction of the coating film thickness threshold discrimination module 330.

[0202] As can be seen from Table 3, the actual measured coating film thickness values of Example 1 and Example 2, 280.8 nm and 276.4 nm, are lower than the minimum threshold of 285 nm. Therefore, they do not meet the restrictions of the coating film thickness threshold discrimination module 330 of Formula D, and the obtained data cannot be incorporated into the simulation model database.

[0203] Furthermore, from Formula C and the electroplating conditions in Table 1, it can be seen that during the electroplating process, the cathodic current efficiency of the electroplating solution η kIt is a concentrated reflection of all influencing factors of the electronic electroplating device module system. Therefore, under the condition that the electroplating time remains unchanged, the actual current efficiency of Example 1 < 30% of the current efficiency used during prediction, and the actual current efficiency of Example 2 < 30% of the current efficiency used during prediction.

[0204] Furthermore, from the actual electroplating results of Example 1 and Example 2, it can be seen that when different current densities are used, their cathode current efficiencies are not exactly the same. Therefore, for the error generated between the predicted coating thickness and the measured coating thickness, it is necessary to increase the electroplating time to increase the coating thickness in order to meet the limiting requirements of the formula D for the predicted coating thickness. d 预测值 of the limiting requirements.

[0205] The error between the predicted coating thickness and the measured coating thickness in Example 1 = 300 nm - 280.8 nm = 19.2 nm;

[0206] Furthermore, using the film thickness difference of 19.2 nm and calculating it again with formula C, the required electroplating time is obtained as 3.1 sec. That is, the corrected electroplating time of Example 1 is: 35 + 3.1 sec = 38.1 sec;

[0207] Similarly, the error of Example 2 = 300 nm - 276.4 nm = 21.6 nm; the required electroplating time is obtained as 1.9 sec. That is, the corrected electroplating time of Example 2 is: 21 + 1.9 sec = 22.9 sec;

[0208] Furthermore, the corrected electroplating time is shown in Table 4.

[0209] Table 4

[0210]

[0211] According to the same conditions and processing methods of the first Example 1 and Example 2, the second measured film thickness data of Example 1 and Example 2 are shown in Table 5.

[0212] Table 5

[0213] [[ID=..]]

[0214] As can be seen from Table 5, by adjusting the electroplating time, for the predicted coating thickness of 300 nm, the actual coating thickness values of 301.2 and 302.2 nm for Example 1 and Example 2 are within the range of the minimum threshold of 285 nm and the maximum threshold of 315 nm described by formula D. Therefore, the electroplating conditions of the corrected Example 1 and Example 2 meet the coating thickness calculation module 360 for constructing the double-sided local prediction coating thickness of machine learning, and the actual experimental data is imported into the memory for storage.

[0215] Further, in order to screen the preferred range of the coating film thickness varying with the current density, Example 3 is set. Example 3 needs to adopt the 38.1 sec prediction of the coating film thickness the same as that in Example 1, and needs to carry out the electroplating experiment in the same electroplating method as that in Example 1. The experimental conditions of Example 3 are shown in Table 6.

[0216] Table 6

[0217]

[0218] Further, according to the experimental conditions in Table 6, for the semiconductor device 200 prepared by electroplating in Example 3, according to the same coating film thickness test method as that in Example 1, the obtained coating film thickness data is shown in Table 7.

[0219] Table 7

[0220]

[0221] As can be seen from Table 6, the predicted coating film thickness of Example 3 is 570 nm. According to formula D, its minimum threshold is 541.5 nm and the maximum threshold is 598.5 nm; further, from Table 7, the average value of the actual film thickness experiment is calculated to be 567.6 nm. Therefore, the error between the predicted value and the test value is only 570 - 567.6 = 2.4 nm, which has excellent accuracy; at the same time, the average value of the actual film thickness experiment of Example 3 is 567.6 nm, which is at a position close to the center between the minimum threshold of 541.5 nm and the maximum threshold of 598.5 nm. The result satisfies the coating thickness calculation module 360 for constructing the double-sided local prediction of the coating film thickness by machine learning, and the actual experimental data is imported into the memory for storage.

[0222] Further, all the electroplating parameters of Example 1 and Example 3, as well as the corresponding coating film thickness data, after being deeply analyzed by the training module 340, the stability generation module 350, and the coating thickness calculation module 360 of the machine learning simulation model, and after being verified by the accuracy of the model morphology discrimination module 370, if the accuracy verification requirement is satisfied as Yes, all the data is imported into the memory 380 for backup storage; if the accuracy verification requirement is not satisfied as No, it returns to the training module 340 for the next cycle of processing until the accuracy verification requirement is satisfied.

[0223] In short, through the deep analysis and data calculation of the experimental data of Example 1 and Example 3 by the machine learning simulation model, a relatively optimal simulation experimental result of the organic integration of the experimental data and the simulated quantization data is obtained;

[0224] Taking 38.1 seconds as the reference for the electroplating time, the following will specifically detail the simulation quantization experiment results for deriving the simulation data, that is, the process of the scatter trend of the gold plating film thickness distribution on both sides locally.

[0225] First, under the condition of constant electroplating time, formula C is transformed into:

[0226] I / S = ρ d / t C η k I / S = A (ampere / dm 2 )

[0227] A = ρ d / t C η k

[0228] d = A hh = t C η k / ρ

[0229] (Formula C-2)

[0230] In formula C-2, the specific gravity, electroplating time, and C are known constants, and the electroplating efficiency η k is a variable within a known range. Therefore, all constants are represented by h to obtain the simplified formula C-2, that is, the mutual relationship between the current density A (ampere / dm 2 ) and the coating film thickness d .

[0231] By deeply analyzing the experimental data of Example 1 and Example 3 through the coating film thickness calculation module of the machine learning simulation model, the simulation quantization experiment results of the fusion of simulation data and experimental data are derived. The specific parameters are shown in Table 8, and the derived model is as Figure 4 shown.

[0232] Table 8

[0233]

[0234] Through the application machine learning simulation model constructed by the present invention, as Figure 4 shown, under the conditions of electroplating gold for electronic devices set in Example 1 and Example 3, the coating film thickness of the machine learning simulation model is deeply analyzed and calculated to obtain the simulation quantization experiment results, that is, the scatter trend morphology of the coating film thickness distribution effectively plating the metal on both sides locally of the semiconductor device.

[0235] From Figure 4It can be seen that the scatter plot of the coating thickness distribution of the double-sided local coating film thickness electroplating module based on the machine learning simulation model is a parabola that gradually rises upward as the current density increases; in the range of current density of 5-50 A / dm 2 the coating thickness shows a relatively fast upward trend, and it rises slowly after exceeding 50 A / dm 2 and starts to show a downward trend after 60 A / dm 2 This change pattern reflects the actual situation that the cathode current efficiency in formula C-2 η k will decrease as the current density A increases; that is, in the region where the current density is relatively low < 5 A / dm 2 the cathode current efficiency η k 底 tends to approach the 2 average value in the region of current density 5-50 A / dm η k 中 and the average value in the region of 51-70 A / dm 2 is lower than the η k 高 average value; therefore, the average value in the region of current density 5-50 A / dm η k 中 is the electroplating condition of the best region for the efficiency of forming the coating film thickness. 2 η k 中

[0236] Figure 4 From

[0237] it can be further known that the measured values of the current density of 7 A / dm in Example 1 2 and 15 A / dm in Example 3 2 are in good agreement with the simulated quantitative experimental results derived from the in-depth analysis of the machine learning simulation model, which can prove that the machine learning simulation model constructed by the present invention has good accuracy.

[0237] In summary, from Figure 4 it can be seen that through the in-depth analysis of the coating thickness calculation module of the machine learning simulation model, the obtained simulated quantitative experimental results have good accuracy in comparison with the actual test results of the electroplating module in the range of 7 A / dm 2 to 15 A / dm 2 in Example 1 and Example 3.

[0238] Furthermore, to optimize and improve the matching performance in a wider current density region, 20 A / dm used for prediction based on the machine learning simulation model of the present invention2 , 50 A / dm 2 and 70 A / dm 2 Under the current density conditions of 4 - 6, the actual electroplating tests of Examples 4 - 6 were carried out through the electronic electroplating module. The specific conditions are shown in Table 9.

[0239] Table 9

[0240]

[0241] Furthermore, according to the electronic electroplating conditions shown in Table 9, electroplating tests were carried out through the electronic electroplating module. The actual measured film thickness data are shown in Table 10.

[0242] Table 10

[0243]

[0244] From the actual measured coating data of Examples 4 - 6 in Table 10, which are indicated by a dash in Figure 5 , the comparison results with the predicted coating film thickness value ● of the machine learning simulation model of the present invention are shown in Table 11.

[0245] As Figure 5 shown, according to the scatter plot of the coating film thickness distribution of the double - sided local coating film thickness electronic electroplating module of the machine learning simulation model, it is a parabola that gradually rises upward with the increase of the current density; in the current density range of 5 - 50 A / dm 2 , the coating film thickness shows a relatively fast upward trend. After exceeding 50 A / dm 2 , the rise becomes slow. After 60 A / dm 2 , a downward trend begins to appear, which is similar to the overall trend of Figure 4 .

[0246] Furthermore, as can be seen from Figure 5 , for Examples 4 - 6, the current density conditions of 20 A / dm 2 , 50 A / dm 2 and 70 A / dm 2 were extracted from the machine learning simulation model database, and compared with the current density conditions of 7 A / dm Figure 4 , 15 A / dm 2 and 70 A / dm 2 extracted from Examples 1 and 3 in 2 . The selected range of the current density is wider. Therefore, through the in - depth analysis of the coating film thickness calculation module of the machine learning simulation model, the trend of the coating film thickness changing with the current density Figure 5Closer to the measured value obtained by the electroplating module; that is, the more measured data there are, the more accurate the trend of the coating thickness changing with the current density obtained after importing the machine learning simulation model processing system of the present invention. Figure 5 The more accurate.

[0247] See Table 11:

[0248] Table 11

[0249]

[0250] According to the standard discrimination module of the machine learning simulation model of the present invention, the predicted value must satisfy the range of formula D;

[0251] d 实测值 (100% - 5%) ≤ d 预测值 ≤ d 实测值 (100% + 5%)

[0252] (Formula D)

[0253] As can be seen from Table 11, under the conditions of current densities of 20, 50, and 70 A / dm 2 , the predicted values of the double-sided local coating thickness based on the machine learning simulation model not only satisfy the standard discrimination range of formula D, but also the errors between the predicted values and the measured values are only 1.8, -2.8, and 4 nm respectively, fully demonstrating the prediction accuracy of the machine learning simulation model constructed by the present invention.

[0254] Optimization of the coating thickness and electroplating time for constructing the machine learning simulation model:

[0255] In the simulation model for predicting the double-sided local coating thickness constructed by the foregoing coating thickness and current density, although the optimization results of the coating thickness and current density of the machine learning simulation model are obtained on the premise of a fixed electroplating time, that is, the constructed model of their mutual relationship already has good accuracy; however, due to the electroplating measured data with an electroplating time of only 38.1 sec, the dependency relationship between the coating thickness and the electroplating time in the database of the machine learning simulation model needs to be further optimized, improved, and a simulation model with more accurate prediction of the double-sided local coating thickness is obtained.

[0256] After the following arrangement of the formula C, the electroplating time t and the current density A of formula C-3 are obtained:

[0257] t = S ρ d / C I η k I / S = A (Ampere / dm 2 )

[0258] t = ρ d / C A η k S / I = 1 / A

[0259] t = d H / AH = ρ / C η k

[0260] (Formula C-3)

[0261] In Formula C-3, the specific gravity of the electroplating solution, the electroplating time, and C are known constants, and the electroplating efficiency η k is a variable within a known range. Therefore, all constants are represented by h, H to obtain the simplified Formula C-3, that is, the electroplating time t , the coating film thickness d and the current density A (amperes / dm 2 ); H is a variable constant as shown in the formula.

[0262] Apply Formula C-3 to construct an optimization method for the coating film thickness calculation film thickness module that changes with the electroplating time in the machine learning simulation model of the present invention.

[0263] First, based on the actual electroplating time of 38.1 sec in the database for predicting the double-sided local coating film thickness of the machine learning simulation model of the present invention, through the following steps as described Figure 4 above, as can be seen above, the measured coating film thickness corresponding to the electroplating time of 38.1 sec meets the requirements of the coating film thickness threshold discrimination module 330, and further through the training module 340, the stability generation module 350, the simulation depth analysis operation of the coating thickness calculation module 360, and after being screened and selected by the model morphology discrimination module 370, the data that meets the model construction is stored in the memory 380; and so on. The simulation operation results of the double-sided local coating film thickness predicted by the model are shown in Table 12 and Figure 6 as shown.

[0264] Table 12

[0265]

[0266] From Figure 6 the double-sided local gold coating film thickness scatter plot of the coating film thickness and the electroplating time as shown, it can be seen that the simulation operation results of the double-sided local coating film thickness predicted by the machine learning simulation model show an arc-shaped upward trend, forming a model morphology with a good distribution.

[0267] To further optimize and improve the simulation results and make them closer to the measured values of the gold plating film thickness, the electroplating time conditions of 20 sec, 50 sec, and 70 sec used for prediction based on the machine learning simulation model are preferably selected, and the actual electroplating tests of Examples 7 to 9 are carried out through the electronic electroplating module 320. The specific electroplating conditions are shown in Table 13.

[0268] Table 13

[0269]

[0270] Furthermore, through the electroplating test implemented by the electronic electroplating module, the actual test film thickness data obtained are shown in Table 14 and Figure 7 as follows.

[0271] Table 14

[0272]

[0273] The comparison results between the simulated operation results of the double-sided local plating film thickness predicted from the model shown in Table 12, that is, the predicted plating film thickness values at electroplating times of 20 sec, 50 sec, and 70 sec and the actual test plating values of Examples 7 to 9, are shown in Table 15.

[0274] Table 15

[0275]

[0276] Note: According to the standard discrimination module of the machine learning simulation model of the present invention, the predicted value must satisfy the range of formula D;

[0277] d 实测值 (100% - 5%) ≤ d 预测值 ≤ d 实测值 (100% + 5%)

[0278] (Formula D)

[0279] As can be seen from Table 15, for Examples 7 to 9, under the selected electroplating time conditions of 20 sec, 50 sec, and 70 sec, the simulated operation results of the double-sided local plating film thickness predicted based on the model are all lower than the experimental test plating values, which are -15.4 nm, -47.2 nm, and -27.2 nm respectively; each predicted value meets the minimum and maximum threshold ranges calculated by formula D and can be applied to predict the double-sided local plating film thickness.

[0280] Further, all the electroplating conditions obtained in Examples 6 to 9 and the measured values of the coating film thickness are imported into the machine learning simulation model. Through the simulation depth analysis operations of the training module 340, the stability generation module 350, and the coating thickness calculation module 360, and after the screening by the model morphology discrimination module 370, the data that conforms to the model construction is stored in the memory 380; the models that do not meet the model morphology are returned to the training module 340 again to repeat the above steps, and so on, to make the model morphology tend to be more perfect and achieve more excellent prediction accuracy of the machine learning simulation model.

[0281] Furthermore, from Figure 7 the simulation operation results of the predicted double-sided local coating film thickness with electroplating times of 10 sec and 60 sec are selected again. The current density is also 30 A / dm 2 . The electroplating test is carried out through the electroplating module 320, and then the electroplating conditions and the corresponding coating film thickness are Figure 3 simulated and calculated according to the process to obtain the construction method of the predicted double-sided local coating film thickness based on the machine learning simulation model of the present invention.

[0282] Generally speaking, according to the above Examples 1-9, the present invention can import the coating film thickness experimental data generated during the electroplating process of the electroplating module under different electroplating conditions, continuously optimize and improve the predicted coating film thickness data extracted from the machine learning simulation model database. The more the coating film thickness experimental data actually measured by the electroplating module, the closer the predicted coating film thickness data obtained after the simulation model is trained and improved is to the measured value; it can quickly establish the optimal electroplating production conditions for various semiconductor devices with different structures, accurately manage the electroplating conditions, improve the electroplating production efficiency, greatly reduce the use amount of precious metals, and effectively improve the uniformity and high-performance quality of the coating film thickness.

[0283] In summary, for the construction method of the predicted double-sided local coating film thickness based on the machine learning simulation model of the present invention, through the predicted double-sided local coating film thickness system and the LR logistic regression operation system, the electroplating conditions associated with the predicted coating film thickness in the model database are extracted, and the electroplating conditions are used for the electroplating module to obtain the measured coating film thickness value. Further, the measured coating film thickness value is imported into the machine learning simulation model. The predicted coating film thickness system and the logistic regression operation system of the simulation model perform training simulation processing on the threshold relationship between the predicted coating film thickness and the measured coating film thickness, so as to obtain the construction method of the predicted double-sided local coating film thickness based on the machine learning simulation model.

[0284] First, for the optimization of the current density and coating film thickness of the machine learning simulation model, by extracting the electroplating conditions corresponding to the predicted film thickness, with the electroplating time fixed, the current density is taken as the only variable, and the electroplating module is used to actually measure the coating film thickness data. The measured film thickness data is processed by the predicted double-sided local coating film thickness system and the LR logistic regression operation system, realizing the optimization model of the current density and coating film thickness. It can be seen from the actual operation results that the more coating film thickness data is tested under the actual current density conditions, the more accurate the optimization model of the current density and coating film thickness of the simulation model is.

[0285] Second, for the optimization of the electroplating time and coating film thickness of the machine learning simulation model, by extracting the electroplating conditions corresponding to the predicted film thickness, with the current density fixed, the electroplating time is taken as the only variable, and the electroplating module is also used to actually measure the coating film thickness data. The measured film thickness data is processed by the predicted double-sided local coating film thickness system and the LR logistic regression operation system. Under the condition that the predicted film thickness meets the threshold range, the optimization model of the electroplating time and coating film thickness is realized. Similarly, it can be known that the more electroplating time data is tested by the electroplating module, and the measured coating film thickness data is fed back and applied to the training model operation process, the more accurate the optimization model of the current time and coating film thickness of the obtained simulation model is.

[0286] Furthermore, the insensitive parameters related to the electroplating module of the machine learning simulation model, such as the electroplating mold size, the type and size of the electrolytic anode, can also be optimized for the electroplating mold size, the type and size of the electrolytic anode, and the coating film thickness through the construction method of predicting the double-sided local coating film thickness based on the machine learning simulation model of the present invention, realizing the optimization model of the electroplating mold size, the type and size of the electrolytic anode, and the coating film thickness, thereby improving and perfecting the electroplating mold size, the type and size of the electrolytic anode, and achieving the purpose of obtaining better coating film thickness distribution performance and high-performance quality.

[0287] All in all, the construction method of predicting the double-sided local coating film thickness based on the machine learning simulation model of the present invention can quickly establish the optimal electroplating production conditions for various semiconductor devices with different structures through the comprehensive operation and processing of the predicted double-sided local coating film thickness system and the LR logistic regression operation system, and then obtain the optimal control method for the double-sided local electroplating of the workpiece. It can not only achieve the accuracy of predicting the double-sided local coating film thickness of semiconductor devices, greatly reduce the error between the coating film thickness of the actual electroplated product, but also realize the further improvement of the electroplating mold, promote the improvement of the uniformity of the double-sided local coating film thickness of semiconductor devices, and meet the requirements of high quality of semiconductor devices.

[0288] Although the present invention has been disclosed above in preferred embodiments, it is not intended to limit the present invention. Anyone skilled in this technology can make various modifications and alterations without departing from the spirit and scope of the present invention. Therefore, the protection scope of the present invention should be defined by the claims.

Claims

1. A construction method for predicting the film thickness of double-sided local plating based on a machine learning simulation model, characterized in that, By importing the experimental data of the coating film thickness generated during the electroplating process of the electroplating module under different electroplating conditions, the machine learning simulation model is improved through the prediction system for double-sided local coating film thickness, and the LR logistic regression model is constructed. According to the results of continuous training of the LR logistic regression model, the coating film thickness prediction data conditions are obtained, which specifically include the following steps: S1: Construct a database based on electroplating conditions, and construct an initial machine learning simulation model linked to the database; set the database in the memory; S2: Extract at least two sets of electroplating conditions from the database, and predict the double-sided local coating film thickness through the prediction calculation module in the prediction system for double-sided local coating film thickness; use the electroplating module to perform the actual electroplating operation of the coating film thickness on the semiconductor device according to the extracted electroplating conditions, and import the obtained experimental data of the coating film thickness into the coating film thickness threshold discrimination module; S3: Predict the double-sided local plating film thickness d 预测值 Compare it with the target threshold: When the predicted double-sided local plating film thickness d 预测值 meets formula D, then enter the LR logistic regression model; while the predicted plating film thickness d that does not meet the target threshold 预测值 cannot enter the training module, and based on the measured plating film thickness value d of the corresponding electroplating conditions extracted from the database in S2 实测值 make corrections, calculate the corrected electroplating conditions, return to S2 to measure the plating film thickness d again 实测值 , and re-determine whether the predicted double-sided local plating film thickness d after correcting the electroplating 预测值 meets the threshold range until the predicted double-sided local plating film thickness d 预测值 meets formula D before entering the LR logistic regression model; Utilize all predicted double-sided local coating film thicknesses d that satisfy formula D 预测值 to construct an LR logistic regression model; The formula D associated with the target threshold is: d 实测值 (100% - 5%) ≤ d 预测值 ≤ d 实测值 (100% + 5%) S4: Import the predicted double-sided local coating film thickness d 预测值 and the measured value d of the electroplating module 实测值 into the training module in the LR logistic regression operation system, and perform system training through the LR logistic regression operation system. The LR logistic regression operation system includes a training module, a stability generation module, and a coating film thickness operation module; S5: Judge the model morphology according to the coating film thickness calculated by the coating film thickness operation module in S4 and compare it with the target threshold. If the target threshold is reached, it is determined that the electroplating conditions are optimized, and the optimized electroplating conditions are applied to the production debugging work of the electroplating device; If the target threshold is not reached, discard the coating film thickness and return to S4 for retraining.

2. The construction method for predicting the thickness of a double-sided local plating film based on a machine learning simulation model according to claim 1, wherein, The initial machine learning simulation model includes: a processor, a memory, an electroplating module, a prediction calculation module and a coating film thickness threshold discrimination module connected to the electroplating module, a set partitioning module, and a training module, a stability generation module, a coating thickness calculation module, and a model morphology discrimination module connected in sequence.

3. The construction method for predicting the thickness of double-sided local plating film based on the machine learning simulation model according to claim 2, wherein In S3, meeting the target threshold means that the predicted double-sided local coating film thickness d 预测值 is between the minimum threshold and the maximum threshold. The minimum threshold = coating specification value - 5% of the coating specification value, and the maximum threshold = coating specification value + 5% of the coating specification value.

4. The construction method for predicting the thickness of double-sided local plating film based on the machine learning simulation model according to claim 3, characterized in that, The LR logistic regression model is constructed by sensitive condition parameters, and the sensitive condition parameters include types of electroplating solutions; electroplating solution temperatures; electroplating solution specific gravities; Diameter of the pipeline for transporting electroplating solutions; types of electroplating power supplies; current intensities; electroplating times; Cathode current efficiencies; surface areas of plating areas; metal densities of electroplating layers; current densities; electroplating solution flow rates.

5. The construction method for predicting the film thickness of double-sided local plating based on a machine learning simulation model according to claim 4, wherein When using COMSOL Multiphysics finite element simulation software in the electroplating module, the influencing factors of the machine learning simulation model for predicting double-sided local coating film thickness are composed of sensitive parameters and insensitive parameters. The insensitive parameters include the electroplating tank size, the pumping power of the pipeline for transporting electroplating solutions, the electroplating power supply power, the electroplating mold size, the types and sizes of electrolytic anodes in the electroplating module.

Images