Pulse electroplating monitoring method and monitoring system

Abstract

The invention belongs to the technical field of electroplating, and discloses a pulse electroplating monitoring system and method.The pulse electroplating monitoring system comprises an electroplating information construction module, a pulse electroplating monitoring module and a pulse electroplating monitoring module, the electroplating information construction module is used for extracting first associated information from current information and plating information, and a first associated information base is constructed; the concentration sensing module is used for extracting second associated information from the current information and the concentration information and constructing a second associated information base; the fusion module is used for mutually fusing the first association information base and the second association information base to generate a ternary information base; and the information output module retrieves the current information from the ternary information base according to the input plating information, and controls the current parameters of electroplating based on the current information. According to the technical scheme, the first associated information represents the mapping relation between the current parameter and the plating quality characteristic, and the second associated information reveals a feedback mechanism of the dynamic change of the electrolyte concentration in the plating information on current regulation and control.

Description

Technical Field

[0001] The present application relates to the field of electroplating monitoring technology, and in particular to a pulse electroplating monitoring method and monitoring system. Background Art

[0002] Pulse plating is an electroplating technique that controls the metal deposition process by periodically switching on and off, or modulating, the current (or voltage) waveform. Its core principle is to utilize the intermittent on (Ton) and off (Toff) characteristics of the pulse current to achieve precise control of the mass transfer efficiency, reaction kinetics, and microstructure of the coating.

[0003] In traditional DC electroplating processes, continuous current application leads to excessively rapid metal deposition, with the nucleus growth rate significantly exceeding the nucleation rate, easily resulting in the formation of a coarse grain structure. This grain coarsening can lead to defects such as increased porosity and reduced hardness in the coating. However, pulse plating technology, through the high current density in the Ton phase, can instantly promote the formation of a large number of nuclei, reducing the grain diameter to the nanometer level, thereby significantly improving the density, hardness, and wear resistance of the coating.

[0004] However, the pulse electroplating process is difficult to control. During the electrolysis process, as metal ions continue to deposit on the workpiece surface, the conductivity of the electrolyte will fluctuate dynamically due to changes in the metal ion concentration. This change will directly affect the stability of the coating information. Especially in high-precision workpiece processing scenarios, the surface texture of the coating is required to maintain a high degree of consistency, but existing pulse electroplating solutions generally adopt a fixed parameter control mode, that is, after generating fixed parameters according to the preset electroplating requirements, the parameters are kept unchanged throughout the electroplating process. This static control method cannot respond to changes in the electrolyte state in real time, resulting in the coating thickness uniformity being difficult to meet the needs of high-end applications. Summary of the Invention

[0005] The content of this application is used to briefly introduce concepts that will be described in detail in the detailed description section below. The content of this application is not intended to identify key features or essential features of the technical solution for which protection is sought, nor is it intended to limit the scope of the technical solution for which protection is sought.

[0006] As a first aspect of the present application, in order to solve the technical problems mentioned in the above background technology section, some embodiments of the present application provide a pulse electroplating monitoring system, including:

[0007] an electroplating information construction module, extracting first associated information from the current information and the plating layer information, and constructing a first associated information database;

[0008] The concentration sensing module extracts second correlation information from the current information and the concentration information and constructs a second correlation information database;

[0009] A fusion module, fusing the first associated information database and the second associated information database to generate a ternary information database;

[0010] The information output module retrieves the current information from the ternary information library according to the input coating information, and controls the current parameters of the electroplating based on the current information.

[0011] In the technical solution of the present application, the first correlation information characterizes the mapping relationship between current parameters and coating quality characteristics, and the second correlation information reveals the feedback mechanism of the dynamic change of electrolyte concentration in the coating information on current regulation. By coupling and modeling the above two types of correlation information, a three-dimensional correlation database containing current information, coating characteristic information and concentration information is constructed, and a quantitative correspondence model between electroplating process parameters and coating quality indicators can be established. This model realizes closed-loop control of coating quality based on current signals by real-time analysis of the intrinsic relationship between current parameters and coating microstructure and composition distribution, combined with the dynamic influence of electrolyte concentration changes on current efficiency, effectively ensuring the performance consistency of electroplating products.

[0012] In the electroplating process, the coating performance indicators are affected by the coupling of electrolyte concentration parameters and current parameters. During the electroplating operation, the concentration of electrolyte components will change dynamically due to the continuous consumption of metal ions. This concentration change will directly change the current field distribution characteristics and the kinetic process of the electrodeposition reaction. When establishing the first correlation model between current parameters and coating quality, the traditional technical solution failed to effectively decouple the superimposed effects of electrolyte concentration changes, resulting in systematic deviations in the correlation model. In order to resolve this technical contradiction, the present application proposes the following technical solutions:

[0013] Furthermore, the electroplating information building module includes:

[0014] The original relationship construction unit generates preliminary correlation information based on the current information and the coating information;

[0015] The implicit weight updating unit dynamically updates the preliminary correlation information based on the temporal change of the coating information to generate the first correlation information.

[0016] The technical solution provided by this application updates the first correlation information in two steps. The initial correlation information is the direct correlation between the current information and the coating information. Then, implicit information is extracted from the temporal changes in the coating information, and the initial correlation information is dynamically updated. The resulting first correlation information accurately describes the internal correlation between the current information and the coating information.

[0017] Furthermore, the preliminary correlation information is generated in the following manner:

[0018] S1: Preprocess the current information to extract the current density, duty cycle and current frequency;

[0019] S2: Obtaining coating information corresponding to the current information, the coating information including coating thickness and coating roughness;

[0020] S3: Use the LSTM model to establish the correlation between current information and coating information, and generate a collection of correlation weights;

[0021] S4: Obtain all association weight sets, map the association weight sets to a low-dimensional space, and generate preliminary association information.

[0022] In the technical solution of the present application, the current characteristic parameters select three key electrical parameters, namely current density, duty cycle and pulse frequency, as modeling inputs. This parameter combination has been experimentally verified to be the core variable that determines the evolution of the microstructure of the coating in the pulse electroplating process. By constructing a deep learning model based on LSTM (long short-term memory network), the temporal coupling relationship between the current parameter sequence and the coating performance index can be effectively analyzed. Its gated cyclic unit structure is particularly suitable for capturing the dynamic mapping law of current excitation and grain growth behavior during the electroplating process. Compared with the traditional static model, the forget gate mechanism of the LSTM network can realize dynamic adaptive adjustment of the model parameters. Under the working condition of continuous change of the composition of the electroplating solution, it can still maintain the prediction accuracy and control stability of the current-coating correlation model, significantly improving the process adaptability in complex electroplating scenarios.

[0023] When generating association weights, since many data in the database are highly similar and difficult to distinguish carefully, directly generating association weights will have a large error rate in practice. To address this problem, this application provides the following technical solutions:

[0024] Furthermore, step S3 includes:

[0025] S31: extracting a set of association weights from the sample database through an LSTM network, where the association weight set includes electroplating information, current information, and the association probability between the two;

[0026] S32: Calculate the association weight probability of the association weight set, and filter the association weight set according to the association weight probability.

[0027] This application achieves dual optimization through the synergy of the joint extraction model and the BERT model: first, the joint extraction model is used to complete the preliminary association weight extraction, and then the BERT model is used to perform secondary verification and correction of the association degree in combination with the associated context.

[0028] The electroplating process is affected by a variety of factors. Simply corresponding current information with coating information is easily affected by unknown factors (temperature, electrolyte concentration, workpiece surface roughness) in practice, and thus cannot accurately describe the actual connection between current information and coating information.

[0029] Furthermore, S31 includes the following steps:

[0030] S311: Input sample database AD, AD={A1, A2...A i …A N}, A i represents the i-th sample in the sample database, N represents the total number of samples, and each sample includes corresponding current information and coating information;

[0031] S312: A i Input to LSTM network to generate association weight group T, T={(e h 、r、e t )}, where e h Indicates current information, e t represents the coating information, and r is the association probability.

[0032] In the technical solution provided in the present application, an LSTM network is used to obtain the correspondence between the current information and the coating information in the sample database, and the current information and the coating information are bound according to the correspondence, and the current information and the coating information are bound using probability information. Therefore, the problem that the implicit conditions in the coating information cannot express the invisible influence of electroplating can be solved. By using the association probability, the impact caused by the inability to correlate the current information and the coating information can be effectively solved.

[0033] Furthermore, step 32 includes the following steps:

[0034] S321: Input the sample database AD and the associated weight group T;

[0035] S322: Get e h and e t The number in the sample database AD generates a data set E;

[0036] S323: Calculate the probability of the associated weight;

[0037] P(r|e h , e t , E)=Softmax(W*h [CLS] +b);

[0038] Among them, W represents the classification weight matrix, b represents the bias vector, P(r|e h , e t, E) indicates that the current information is e h Or the coating information is e h When , the probability of the associated weight is r, and Softmax represents the probability conversion function;

[0039] S324: Preset a weight threshold, obtain the association probability whose association weight is higher than the weight threshold, and output the corresponding association weight set T.

[0040] Furthermore, S4 includes the following steps:

[0041] S41: Reorganize the related rights into T={(e h 、r、e t )} is mapped to a low-dimensional space to generate a vector representation of each associated weight group;

[0042] S42: Generate preliminary association information based on the vector representation of all association weight groups T.

[0043] The technical solution provided in this application not only eliminates redundant entities, but also eliminates redundant association weights, thereby reducing the difficulty of building a knowledge base and reducing the information dimension of the knowledge base.

[0044] Furthermore, the method for generating the first association information includes the following steps:

[0045] Z1: obtain preliminary association information G0;

[0046] G0=(e0, r0, T0), where e0, r0, and T0 represent coating information, associated weight, and associated weight group, respectively.

[0047] Z2: Calculate the implicit correlation weight score of each current information and coating information;

[0048] Score(h,r,t)=γ-||h+rt||2; where h represents the vector of coating information, r represents the vector of association weights, t represents the vector of current information, γ represents the margin parameter, and ||h+rt||2 represents the norm range;

[0049] Z3: The implicit correlation weights exceeding the preset value are added to the correlation weights reorganization;

[0050] Z4: Obtain all related rights reorganizations, and update the first related information according to the related rights reorganizations.

[0051] Furthermore, the concentration perception module constructs the second correlation information using the following scheme:

[0052] Step 1: Generate concentration correlation information using the scheme of preliminary correlation information;

[0053] Step 2: Extract implicit features from concentration correlation information;

[0054] Step 3: Construct the second association information based on the implicit features.

[0055] In the technical solution provided in this application, after constructing the association weight knowledge base, the weight changes between the association weight knowledge bases are further analyzed to generate new link association weights. These link association weights are the more representative association weight information in the association weight knowledge base.

[0056] As a second aspect of the present application, a pulse electroplating monitoring method is provided, which uses the aforementioned pulse electroplating monitoring system to monitor the electroplating process. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] The drawings constituting a part of this application are used to provide a further understanding of this application and make other features, purposes and advantages of this application more apparent. The drawings and descriptions of the exemplary embodiments of this application are used to explain this application and do not constitute an improper limitation on this application.

[0058] In addition, throughout the drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and that the elements and components are not necessarily drawn to scale.

[0059] In the attached figure:

[0060] Figure 1 This is an enlarged view of the coating surface under low-frequency pulse;

[0061] Figure 2 This is an enlarged view of the coating surface under high-frequency pulses;

[0062] Figure 3 Schematic diagram of the structure of the pulse electroplating monitoring system. DETAILED DESCRIPTION

[0063] The following will describe embodiments of the present application in more detail with reference to the accompanying drawings. Although certain embodiments of the present application are shown in the accompanying drawings, it should be understood that the present application can be implemented in various forms and should not be construed as being limited to the embodiments described herein. On the contrary, these embodiments are provided to provide a more thorough and complete understanding of the present application. It should be understood that the drawings and embodiments of the present application are for illustrative purposes only and are not intended to limit the scope of protection of the present application.

[0064] It should also be noted that, for ease of description, only the parts related to the invention are shown in the drawings. In the absence of conflict, the embodiments and features in the embodiments of this application can be combined with each other.

[0065] The present application will be described in detail below with reference to the accompanying drawings and in combination with embodiments.

[0066] Pulse plating can change the thickness and roughness of the coating by changing the current density, current frequency and duty cycle. Figure 1 and Figure 2 It is the surface structure of the coating under different current frequencies. In high-precision machining, it is necessary to ensure that the surface structure of the coating is uniform and the coating thickness is consistent. Although the current control technology is mature and the current control accuracy is high, during the electrolysis process, the consumption rate of the electrolyte is not consistent with the replenishment rate of the electrolyte, and the concentration of the electrolyte in the electrolyte fluctuates. In practice, since it is impossible to accurately measure the concentration of each electrolyte in the electrolyte in real time, it is difficult to ensure the thickness uniformity of the electroplating layer by simply controlling the current information output by the electrode, which will be affected by the change in the electrolyte concentration in the electrolyte. To this end, the present application provides the following technical solutions:

[0067] Reference Figure 3 The pulse electroplating monitoring system includes an electroplating information construction module, a concentration perception module, a fusion module, and an information output module. Among them, the electroplating information construction module is used to extract the first associated information from the current information and the coating information to construct a first associated information library. The concentration perception module is used to extract the second associated information from the current information and the concentration information to construct a second associated information library. The fusion module is used to fuse the first associated information library and the second associated information library to generate a ternary information library. The information output module is used to retrieve the current information from the ternary information library according to the input coating information, and control the current parameters of the electroplating based on the current information.

[0068] The first correlation information refers to the relationship between current information and coating information. In theory, current information can directly control coating information. However, in practice, electroplating is affected by multiple factors, such as the local electrolyte concentration and temperature. Therefore, the first correlation information describes the specific impact of current information on coating information under actual sample conditions.

[0069] The second associated information refers to the implicit information between current information and concentration information. The current information implies the electroplating efficiency information, that is, the current information controls the consumption rate of the electrolyte in the electrolyte. After the electrolyte is consumed, it is necessary to rely on the electrolyte replenishment device for rehydration. Since the concentration of the electrolyte is difficult to measure accurately in real time, when rehydrating, it is inevitable that the concentration of the electrolyte will fluctuate periodically around the average concentration. The higher the electrolysis efficiency, the more frequently the electrolyte needs to be replenished, and the more severe the fluctuation. The second associated information shows the impact of the current information on the fluctuation of the electrolyte concentration.

[0070] The first associated information and the second associated information are combined to generate a ternary information library. The ternary information library contains the correspondence information between current information, concentration information, and coating information. The concentration information corresponds to the model of the electroplating equipment, and the coating information corresponds to the demand. Therefore, when the coating information and concentration information are known, the corresponding current information can be found and used to control the electroplating operation. The information output module retrieves the coating information from the ternary information library and outputs the current information. Therefore, the key to this solution lies in how to construct the first associated information and the second associated information.

[0071] The first association information is generated by an electroplating information construction module. The electroplating information construction module includes an original relationship construction unit and an implicit weight updating unit.

[0072] The original relationship building unit is used to generate preliminary correlation information based on the current information and the coating information. The preliminary correlation information is actually the initial correspondence between the current information and the coating information. This correspondence does not take into account the influence of factors such as changes in electrolyte concentration and electrolyte temperature.

[0073] The implicit weight updating unit is used to dynamically update the preliminary correlation information based on the temporal change of the coating information to generate the first correlation information.

[0074] The preliminary association information is generated in the following way:

[0075] S1: Preprocess the current information to extract the current density, duty cycle and current frequency.

[0076] Among the current parameters, current density, duty cycle, and frequency play a key role in affecting coating thickness and surface structure. Generally speaking, increasing current density increases coating thickness, grain size, and electrolyte consumption. Within a reasonable range, controlling current density can control coating thickness and surface structure.

[0077] As the duty cycle increases, the pulse peak current density decreases, resulting in a smaller grain size in the coating, but the coating density increases. A higher duty cycle results in a relatively smoother coating. Therefore, the duty cycle influences the surface structure of the coating. While the pulse frequency has little effect on coating thickness, a higher pulse frequency may lead to an increase in surface voids. Therefore, electroplating information is essentially about adjusting the current density and duty cycle, adjusting the coating structure accordingly.

[0078] S2: Obtain coating information corresponding to the current information, where the coating information includes coating thickness and coating roughness.

[0079] In the electroplating industry, only two factors are generally considered: coating thickness and coating roughness. Coating thickness is generally determined by the electroplating time, which is the result of multi-layer electroplating. The coating thickness in this scheme belongs to the thickness of single-layer electroplating (affected by the grain diameter), and coating roughness represents the smoothness of the coating surface. Generally speaking, the larger the grain diameter, the thicker the single-layer coating and the higher the coating surface roughness. An increase in surface roughness will result in a low density of the coating. As the grain diameter decreases, the single-layer coating becomes thinner, the surface roughness decreases, and the smoothness increases.

[0080] During the electroplating process, the current density and duty cycle are the main factors affecting the coating, while the electrolyte temperature and electrolyte concentration affect the electrolyte resistance, which in turn affects the current magnitude. Therefore, the coating information is affected by the electrolyte concentration.

[0081] The electrolyte temperature can be kept constant by heat exchange in industrial production. Therefore, this application does not consider the impact of temperature on coating information.

[0082] S3: Using the LSTM model to establish the correlation between the current information and the coating information, and generating a collection of correlation weights. Further, step S3 includes:

[0083] S31: Extract the association weight set from the sample database through the LSTM network. The association weight set includes electroplating information, current information and the association probability between the two.

[0084] There is a corresponding relationship between current information and coating information. When the electrolyte concentration is constant, the current information does not change and the coating information does not change. However, when the current information changes, the coating information will change accordingly. For example, the current density is 1.1A / dm 2 , the duty cycle is 60%, the power frequency is 200Hz, the corresponding Ni coating thickness is 0.7um, and the coating surface is smooth. Without changing the electrolyte concentration, temperature and other production conditions, the current density is 1.1A / dm 2 If electroplating is carried out with current information of a duty cycle of 60% and a power frequency of 200Hz, the surface of the coating must be smooth. Correspondingly, if the current density is changed, the surface of the coating will change synchronously.

[0085] In actual production environments, the factors that affect coating are numerous and complex, and the correspondence between current information and coating information is not 100%. This solution inputs a large amount of data into the LSTM network to calculate the correspondence probability between current information and coating information, thereby determining the correlation information between current information and coating information.

[0086] S31 includes the following steps:

[0087] S311: Input sample database AD, AD={A1, A2...A i …A N}, A i represents the i-th sample in the sample database, N represents the total number of samples, and each sample includes corresponding current information and coating information.

[0088] The samples in the sample database AD need to be collected in advance. This is done by pre-configuring current information, then performing electroplating. After the plating layer is formed, the plating layer information is measured and the current information and the plating layer information are mapped. This creates a sample. Subsequently, multiple tests are performed in the same electrolyte environment to collect sufficient sample data and form a database.

[0089] S312: A i Input to LSTM network to generate association weight group T, T={(e h 、r、e t )}, where e h Indicates current information, e t represents the coating information, and r is the association probability.

[0090] The LSTM network consists of an input gate, a forget gate, an update unit, and an output gate. The input gate is used to input the entire sample database AD. The forget gate and update unit gate analyze the associations between the sample entities in the sample database AD, and then output the association probabilities between the sample entities through the output gate.

[0091] The LSTM network is an existing long-short memory network, and its specific structure is a prior art, which will not be described in detail in this application. The LSTM network can find corresponding associations from discrete data. In this solution, the samples arranged in sequence are used as column data. The samples include current information and coating information. The samples are a data pair, and the data pairs are arranged in sequence and input into the LSTM network. The association probability between the current information and the coating information is used as a label. In essence, the LSTM network is used to calculate the corresponding probability between each sample entity in the entire sample database AD. The reason for using the LSTM network for calculation is mainly to use the forget gate and update unit to extract information in a longer sequence span, and when the amount of data is large, the relationship between each entity can be analyzed quickly and accurately.

[0092] The entities in this scheme mainly refer to current information or coating information, and the entity relationship is the association probability.

[0093] There are multiple association weight sets T, and each association weight set records an association probability between current information and coating information.

[0094] S32: Calculate the association weight probability of the association weight set, and filter the association weight set according to the association weight probability.

[0095] In S31, the relationships between entities are bound by probability. However, not every sample collected is correct. There may be some special erroneous data that affects the judgment of entity relationships. Therefore, it is necessary to use the association weight probability to filter the association weight collection.

[0096] Step 32 includes the following steps:

[0097] S321: Input the sample database AD and the associated weight group T;

[0098] S322: Get e h and e t The number in the sample database AD generates a data set E;

[0099] S323: Calculate the probability of the associated weight;

[0100] P(r|e h , e t , E)=Softmax(W*h [CLS] +b);

[0101]

[0102] Among them, Transformer is the attention mechanism, [CLS] represents the special classification mark, [SEP] represents the special mark for separating entities, and h [CLS] represents the hidden state of the mark, W represents the classification weight matrix, b represents the bias vector, P(r|e h , e t , E) indicates that the current information is e h Or the coating information is e h When , the probability of the associated weight is r, and Softmax represents the probability conversion function.

[0103] Transformer is a deep learning model architecture based on the self-attention mechanism, which can directly model the global dependencies between elements in a sequence.

[0104] S324: Preset a weight threshold, obtain the association probability whose association weight is higher than the weight threshold, and output the corresponding association weight set T.

[0105] In S323, the probability of the associated weights is calculated, and in S324, the associated weight group T is screened. h , e t, E) is less than the preset weight threshold, it means that the credibility of the associated weight group is not high and the associated weight group needs to be deleted. In this way, the reliability of the selected associated weight group T is higher.

[0106] S4: Obtain all association weight sets, map the association weight sets to a low-dimensional space, and generate preliminary association information.

[0107] S4 includes the following steps:

[0108] S41: Reorganize the related rights into T={(e h 、r、e t )} is mapped to a low-dimensional space to generate a vector representation of each associated weight group;

[0109] S42: Generate preliminary association information based on the vector representation of all association weight groups T.

[0110] Mapping the associated weights group T to a low-dimensional space is an existing technology and will not be described in detail here. h As the horizontal axis, e t As the vertical coordinate and r as the height coordinate, the associated weight group T can be mapped into a three-dimensional space, and then the associated weight group T can be represented by a vector in the three-dimensional space.

[0111] The method for generating the first association information comprises the following steps:

[0112] Z1: Get preliminary association information G0;

[0113] G0=(e0, r0, T0), where e0, r0, and T0 represent coating information, associated weight, and associated weight group, respectively.

[0114] Z2: Calculate the implicit correlation weight score Se of each current information and coating information;

[0115] Se(h,r,t)=γ-||h+rt||2; where h represents the vector of coating information, r represents the vector of association weights, t represents the vector of current information, γ represents the margin parameter, and ||h+rt||2 represents the norm range;

[0116] Because the number of data samples cannot be infinite, the complete correlation between current and coating information cannot be captured from all collected samples. For example, a change in the frequency of the current information has little effect on the coating information. Therefore, there is no strict correspondence between current and coating information, and the correlation between coating and current information cannot be determined from a limited number of samples.

[0117] This solution calculates the implicit correlation weight score Se for each current and coating information. Based on the existing sample database, it analyzes the degree of correlation between similar samples and adds new sets of correlation weights. This approach further explores the implicit relationship between coating and current information without adding new data samples, adding more sets of correlation weights.

[0118] Z3: The implicit correlation weights exceeding the preset value are added to the correlation weights reorganization;

[0119] Z4: Obtain all related rights reorganizations, and update the first related information according to the related rights reorganizations.

[0120] The concentration perception module uses the following scheme to construct the second association information:

[0121] Step 1: Generate concentration correlation information using the preliminary correlation information scheme.

[0122] The first correlation information is essentially a correlation weight group, namely, coating information, current information and correlation ratio.

[0123] Concentration correlation information is also a set of correlation weights, but it correlates current information and concentration change information, as well as the correlation probability between the two.

[0124] When extracting the second correlation information, it is necessary to first extract the concentration correlation information in the same manner as extracting the first correlation information.

[0125] The concentration association information includes several concentration information groups, and the elements in each concentration information group are current information, concentration change information and association probability.

[0126] When obtaining the second associated information, it is necessary to accurately measure the electrolyte concentration under laboratory conditions to obtain an accurate electrolyte concentration change curve and then obtain electrolyte concentration fluctuation information.

[0127] Step 2: Extract implicit features from concentration correlation information.

[0128] The implicit feature in this solution is actually the concentration change information. This means that the concentration change information needs to be corrected into the change in current information. The current information is corrected based on the effect of the concentration change on the electrolyte resistance. This corrected information is the implicit feature. For example, when the electrolyte concentration decreases, the coating thickness decreases. To maintain the original coating thickness, the current density needs to be increased. The increased current density is the implicit feature, that is, the change in current information.

[0129] Step 3: Construct the second association information based on the implicit features.

[0130] The second association information includes several concentration association groups, and each concentration association group includes three elements, namely current information, electrolyte concentration information, and current correction information.

[0131] The fusion module fuses the first associated information database and the second associated information database to generate a ternary information database.

[0132] The ternary information database includes multiple information merging groups, each of which includes current information, electrolyte concentration information, current correction information, coating information and association probability.

[0133] When the information output module performs electroplating control, it first inputs the coating information (the coating information is filtered out from the ternary information library to select all information merging groups including the coating information), and then selects the information merging group with the largest correlation probability, and uses the current information in the information merging group as the initial information at the beginning of electroplating, and then uses the current correction information to correct the current information.

[0134] The above description is only an illustration of some preferred embodiments of the present application and the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present application is not limited to the technical solutions formed by the specific combination of the above-mentioned technical features, but should also cover other technical solutions formed by any combination of the above-mentioned technical features or their equivalent features without departing from the above-mentioned inventive concept. For example, the above-mentioned features and the technical features with similar functions disclosed in the embodiments of the present application (but not limited to) are replaced with each other to form a technical solution.

Claims

1. A pulse plating monitoring system, characterized in that: include: an electroplating information construction module, extracting first associated information from the current information and the plating layer information, and constructing a first associated information database; The concentration sensing module extracts second correlation information from the current information and the concentration information and constructs a second correlation information database; A fusion module, fusing the first associated information database and the second associated information database to generate a ternary information database; The information output module retrieves the current information from the ternary information library according to the input coating information, and controls the current parameters of the electroplating based on the current information.

2. The pulse plating monitoring system according to claim 1, characterized in that: The electroplating information building module includes: The original relationship construction unit generates preliminary correlation information based on the current information and the coating information; The implicit weight updating unit dynamically updates the preliminary correlation information based on the temporal change of the coating information to generate the first correlation information.

3. The pulse plating monitoring system according to claim 2, characterized in that: The preliminary association information is generated in the following way: S1: Preprocess the current information to extract the current density, duty cycle and current frequency; S2: Obtaining coating information corresponding to the current information, the coating information including coating thickness and coating roughness; S3: Use the LSTM model to establish the correlation between current information and coating information, and generate a collection of correlation weights; S4: Obtain all association weight sets, map the association weight sets to a low-dimensional space, and generate preliminary association information.

4. The pulse plating monitoring system according to claim 3, characterized in that: The step S3 comprises: S31: extracting a set of association weights from the sample database through an LSTM network, where the association weight set includes electroplating information, current information, and the association probability between the two; S32: Calculate the association weight probability of the association weight set, and filter the association weight set according to the association weight probability.

5. The pulse plating monitoring system according to claim 3, characterized in that: S31 includes the following steps: S311: Input sample database AD, AD={A1, A2...A i …A N }, A i represents the i-th sample in the sample database, N represents the total number of samples, and each sample includes corresponding current information and coating information; S312: A i Input to LSTM network to generate association weight group T, T={(e h 、r、e t )}, where e h Indicates current information, e t represents the coating information, and r is the association probability.

6. The pulse plating monitoring system according to claim 5, characterized in that: Step 32 includes the following steps: S321: Input the sample database AD and the associated weight group T; S322: Get e h and e t The number in the sample database AD generates a data set E; S323: Calculate the probability of the associated weight; P(r|e h ,e t ,E)=Softmax(W*h [CLS] +b); Among them, Transformer is the attention mechanism, [CLS] represents the special classification mark, [SEP] represents the special mark for separating entities, and h [CLS] represents the hidden state of the mark, W represents the classification weight matrix, b represents the bias vector, P(r|e h , e t , E) indicates that the current information is e h Or the coating information is e h When , the probability of the associated weight is r, and Softmax represents the probability conversion function; S324: Preset a weight threshold, obtain the association probability whose association weight is higher than the weight threshold, and output the corresponding association weight set T.

7. The pulse plating monitoring system according to claim 6, characterized in that: S4 includes the following steps: S41: Reorganize the related rights into T={(e h 、r、e t )} is mapped to a low-dimensional space to generate a vector representation of each associated weight group; S42: Generate preliminary association information based on the vector representation of all association weight groups T.

8. The pulse plating monitoring system according to claim 7, characterized in that: The method for generating the first association information comprises the following steps: Z1: obtain preliminary association information G0; G0=(e0, r0, T0), where e0, r0, and T0 represent coating information, associated weight, and associated weight group, respectively. Z2: Calculate the implicit correlation weight score Se of each current information and coating information; Se(h,r,t)=γ-||h+rt||2; where h represents the vector of coating information, r represents the vector of association weights, t represents the vector of current information, γ represents the margin parameter, and ||h+rt||2 represents the norm range; Z3: The implicit correlation weights exceeding the preset value are added to the correlation weights reorganization; Z4: Obtain all related rights reorganizations, and update the first related information according to the related rights reorganizations.

9. The pulse plating monitoring system according to claim 8, characterized in that: The concentration perception module uses the following scheme to construct the second association information: Step 1: Generate concentration correlation information using the scheme of preliminary correlation information; Step 2: Extract implicit features from concentration correlation information; Step 3: Construct the second association information based on the implicit features.

10. A pulse electroplating monitoring method, characterized in that: The pulse electroplating monitoring system according to any one of claims 1 to 9 is used to monitor the electroplating process.

Images