Method and apparatus for analyzing plating solutions, and method and management system for managing plating baths.

Abstract

To provide an analysis method and an analysis device for plating solution that can process data associated with a precipitation pattern obtained from a plating solution to analyze a state of the plating solution quantitatively under identical standards, and a management method and a management system for plating bath that uses the analysis method.SOLUTION: An analysis method for plating solution includes the steps of: inputting data associated with a precipitation pattern obtained from a plating solution; and using an analysis model machine-learnt using, as teacher data, the data or processing data obtained by processing the data so as to obtain an analysis result of the precipitation pattern or the plating solution on the basis of the data or the processing data.SELECTED DRAWING: Figure 7

Description

Technical Field

[0001] The present invention relates to a method for capturing data on a deposition pattern obtained from a plating solution and analyzing the data, or a data processing means obtained by a machine learning method using AI, and an application of the obtained data to plating bath management.

Background Art

[0002] Conventionally, for the management of a plating bath, tests using deposition patterns such as Hull cell tests, uniform electrodeposition tests (Haring cell tests), and vent cathode tests have been used. For example, the evaluation items of the deposition pattern of a Hull cell cathode plate showing the results of a Hull cell test include a gloss range, a semi-gloss range, distortion, cloudiness, pits, cracks, and the like. The purpose of the Hull cell test is to read from a Hull cell cathode plate "changes in plating solution composition" due to continuing the plating operation and "deterioration of the plating solution" due to accumulation of impurities, and to correct the plating solution and plating conditions to an appropriate range.

[0003] Conventionally, the reading (evaluation) of the above deposition pattern has been performed by the human eye. In this case, in order to perform accurate reading, it is necessary to be a skilled technician who has been managing a plating solution for many years. Non-Patent Document 1 shows that evaluation is performed visually, and also shows the results of visual determination and measurement by a photometer, and it has been pointed out as a problem that visual determination is difficult. Further, Patent Document 1 discloses a method for measuring the surface roughness of a metal film deposited on a cathode surface and managing trace components in an electrolytic solution based on this measured value and the current density.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Non-Patent Documents

[0005]

Non-Patent Document 1

[0006] When evaluating the above deposition patterns by human eye, there are problems such as the difficulty in establishing evaluation criteria and the resulting variability among evaluators. Furthermore, there are limits to the sensitivity and resolution of the human eye, making it difficult to increase sensitivity and resolution, and there are individual differences. In addition, the certification criteria for various evaluation items of the deposition pattern (gloss range, semi-gloss range, distortion, cloudiness, pits, cracks, etc.) vary from person to person, and the relationship between various evaluation items and the composition and impurity concentration of the plating solution is complex. As a result, evaluation criteria, judgment criteria, and the criteria for optimizing the preparation of the plating bath and the plating process also vary, making quantitative evaluation difficult. Finally, because the above deposition patterns are read manually, and the reading and evaluation are not digitized, it is difficult to share evaluation results, which leads to delays in feedback to the preparation of the plating bath and the plating process.

[0007] Therefore, the present invention addresses at least one of the above problems, and its objective is to provide an analysis method and analysis apparatus that can quantitatively analyze the state of a plating solution using a unified standard by processing data relating to deposition patterns obtained from the plating solution, as well as a plating bath management method and management system using the analysis method. [Means for solving the problem]

[0008] To solve the above problems, the method for analyzing a plating solution according to the present invention is characterized by acquiring data relating to the deposition pattern obtained from the plating solution, and using an analysis model trained with machine learning using the data or the processed data obtained by processing the data as training data, thereby obtaining an analysis result of the deposition pattern or the plating solution based on the data or the processed data.

[0009] This allows for quantitative analysis of data regarding the deposition patterns of the plating solution, thereby suppressing variations in human evaluation of the plating solution. In other words, by using a machine learning-based analytical model during evaluation, analysis can be performed using a unified standard, allowing for objective and stable understanding of the deposition patterns and the state of the plating solution. Furthermore, since the analysis results are derived from a data-driven analytical model, input and output information can be digitized.

[0010] Here, the analysis results of the deposition pattern include information such as the surface characteristics of the deposition pattern, for example, gloss, matte (cloudy), semi-gloss, (vertical) stripes (streaks), pits (holes), roughness (small protrusions), blistering (peeling), charring (burning), resinous plating, cracks (burrs), and unplated areas. Information such as surface roughness, surface color (hue and saturation), coverage, and adhesion may also be included. Furthermore, the state of the deposition pattern may include information such as the distribution of each surface condition, as well as the arrangement and size (area) of each area.

[0011] Furthermore, the analysis results of the plating solution include information regarding the composition of the plating solution, such as the type and amount of metal ion source, the type and amount of anodic dissolution accelerator, the type and amount of pH buffer, the type and amount of additives in the plating solution (such as primary brighteners and secondary brighteners), and the type and amount of impurities.

[0012] In the present invention, it is preferable that the analytical model derives the type or amount of components contained in the plating solution from the data or processing data. Here, the above components include a metal ion source, an anodic dissolution accelerator, a pH buffer, and additives. Among these, it is preferable to derive the type or amount of additives contained in the plating solution, and it is even more preferable to derive the amount of additives. Examples of additives include primary brighteners and secondary brighteners. Here, primary brighteners include sulfur compounds such as saccharin and sodium naphthalenedisulfonate. Secondary brighteners include unsaturated alcohols such as butynediol, propanegyl alcohol, and coumarin.

[0013] In the present invention, it is preferable that the analytical model derives the type or amount of impurities contained in the plating solution from the data or processing data. In particular, it is preferable to derive the amount of impurities contained in the plating solution. Examples of impurities generally include metals such as Cu, Zn, Fe, Pb, Al, Cr, and Ca eluted from the material to be plated introduced into the plating bath, as well as ions such as sodium, potassium, cyanide, nitric acid, phosphoric acid, and ammonium, peroxides, and organic impurities.

[0014] In the present invention, it is preferable that the analysis model is composed of a neural network trained by deep learning based on the data or the processed data. In particular, if the data or the processed data is image data, it is preferable that the analysis model is a convolutional neural network.

[0015] In the present invention, it is preferable that the data is image data of the surface configuration of the deposition pattern. This allows for the analysis of the deposition pattern or plating solution based on the surface configuration of the deposition pattern, thereby enabling the acquisition of more objective and highly accurate analysis results by utilizing conventional knowledge regarding deposition patterns.

[0016] In the present invention, the image data is preferably generated from the deposition pattern with a resolution of 600 dpi or higher. In particular, a resolution of 800 dpi or higher is desirable. This allows analysis to be performed based on image data generated with sensitivity and resolution exceeding that of the human eye, thereby enabling highly accurate and reliable evaluation of the deposition pattern. The above data usually represents values ​​of brightness (light reflectance) and color (hue, lightness, and saturation). In this case, although the resolution of the human eye is said to be 0.1-0.2 mm, image data with a resolution of one order of magnitude higher (50 μm or less, preferably 35 μm or less) can be obtained. There is no particular upper limit to the resolution, but it is usually preferable to have a resolution of 60,000 dpi or less. This is because within this range, even the fine surface roughness of the deposition pattern can be recognized. However, in order to reduce the number of data and reduce the burden of image processing, it is preferable to have a resolution of 6,000 dpi or less, and preferably 1,200 dpi or less.

[0017] Here, the above image data can be generated by imaging means such as a scanner or camera. Generally, the image data consists of pixels with a predetermined resolution, each possessing a brightness value and color. Here, instead of the brightness and color data, data on glossiness and reflectivity for each pixel may be used. Furthermore, it is preferable to use a lightbox consisting of a reflector or a diffuser for imaging to prevent reflections. Moreover, by generating the image data from the deposition pattern under constant (unified) optical conditions (resolution, illumination conditions, and detection sensitivity), stable analysis becomes possible. In this respect, generating image data using a scanner is preferable because it can easily achieve constant optical conditions. The pupil diameter of a human is said to be 7-8 mm in the dark, with individual differences of about 1 mm, and differences in brightness for the viewer lead to errors, so the evaluation results are likely to change if the person in charge changes. In the present invention, the evaluation results can be stabilized by acquiring image data under constant optical conditions. Here, it is preferable that the image data is processed by a noise reduction filter such as a median filter or a Gaussian filter. In the deposition pattern of the plating solution, it is desirable to remove in advance any surface features that include fine noise caused by factors other than the plating solution, such as dust or sharp scratches.

[0018] In the present invention, the deposition pattern may be any deposition pattern obtained from the plating solution, including, for example, a deposition pattern consisting of a simple test plated piece, a (pair) deposition pattern formed in a Hull cell, or a deposition pattern from a vent cathode test. However, it is particularly preferable that the deposition pattern be obtained by a Hull cell test. If the deposition pattern is obtained by a Hull cell test, it is possible to obtain information on the fluctuation distribution of the deposition state when the current density is changed within a predetermined range, making it possible to analyze in more detail information about the plating solution from which the deposition pattern was obtained.

[0019] In the present invention, it is preferable that the analysis model performs analysis based on the processed data formed by dividing the image data into a plurality of regions and labeling each divided region to one of a plurality of types of surface configurations of the deposition pattern. Here, the labeling of each divided region (labeling process) may be performed by an operator, but generally, from the viewpoint of efficiency, it is preferable that it be performed automatically by image processing of the divided regions. In this case, the labeling by image processing of the divided regions may be performed according to a predetermined set of multiple surface configuration criteria, or it may be performed by a preprocessing model learned from training data labeled for each divided region.

[0020] In the present invention, when the deposition pattern is obtained by a Hull cell test, the analysis model preferably performs the analysis based on the processing data that shows the change in the surface configuration of the deposition pattern along a first direction corresponding to the change in current density of the image data. In this case, it is even more desirable that the processing data is differential (difference) processing data obtained by differentiating the image data along the first direction. Here, it is even more desirable that the processing data is obtained by differentiating the image data twice (second derivative). This makes it possible to clarify the change in the surface configuration along the first direction without increasing noise. Furthermore, it is also desirable that the processing data is one-dimensional data obtained by arranging representative values ​​(preferably average values) obtained in the second direction of the image data along the first direction. This makes it possible to reliably extract the change in the surface configuration in the first direction. In these cases, it is even more desirable that the processing data is based on data from the central region in the width direction, excluding data near both ends in the second direction orthogonal to the first direction of the image data. By removing both ends of the second direction, the effects of fluctuations in the plating solution surface, the presence of the bottom surface, or agitation can be eliminated.

[0021] In the present invention, the analysis model is machine-learned using a plurality of types (two or more types) of the data as teacher data, and preferably outputs an analysis result with the plurality of types of data as input. In particular, it is an analysis model machine-learned using, as teacher data, information other than the image data or the information based on the processing data related to the plating solution together with the image data or the processing data, and preferably obtains an analysis result of the state of the deposition pattern or the state of the plating solution based on the image data or the processing data and the other information. Here, examples of the other information include film thickness data (representative value or data of film thickness distribution) of the deposition pattern, surface roughness data of the deposition pattern (measured by a surface roughness meter, representative value or data of surface roughness distribution, or magnified surface image by a microscope, SEM, etc.), alloy composition data of the deposition pattern (in the case of alloy plating such as solder), the temperature at which the deposition pattern is obtained, and the plating treatment amount of the plating solution (for example, the product A·T of the current density A and the plating treatment time T of the plating bath from which the plating solution for forming the deposition pattern is extracted). Also, two or more image-based data may be used as input to the analysis model.

[0022] Next, the plating solution analysis apparatus according to the present invention includes data acquisition means for acquiring or receiving data related to the deposition pattern obtained from the plating solution, an analysis model that is machine-learned using the data or the processing data obtained by processing the data as teacher data, and outputs an analysis result of the deposition pattern or the plating solution based on the data or the processing data, and information presentation means for presenting, outputting, or transmitting the analysis result or information indicating the situation of the plating solution obtained from the analysis result. Thereby, by quantitatively analyzing the data related to the deposition pattern of the plating solution, it is possible to suppress variations in the evaluation of the plating solution by humans. Note that acquiring the data related to the deposition pattern includes not only generating the data related to the deposition pattern but also receiving the already generated data.

[0023] Next, the plating bath management method according to the present invention comprises: a first step of acquiring or receiving data relating to the deposition pattern obtained from the plating solution removed from the plating bath; a second step of acquiring the data and obtaining analysis results of the deposition pattern or the plating solution based on the data or the processed data by using an analysis model trained on machine learning with the data or the processed data obtained by processing the data as training data; and a third step of presenting, outputting, or transmitting information indicating the status of the plating bath from the analysis results. Here, in the third step, it is preferable to present, outputting, or transmitting information indicating the status of the plating bath, or, if such work as preparation of the plating bath is necessary, information indicating the method of such preparation or work, along with the information indicating the status of the plating bath. Furthermore, in the third step, it is preferable to present, outputting, or transmitting information indicating the status of the plating solution and the method of such preparation or work, using other data relating to the plating solution along with the data or the processed data. Furthermore, in the second step, the analysis model may be configured to obtain analysis results using other data relating to the plating solution along with the data or processing data.

[0024] Next, the plating bath management system according to the present invention includes data acquisition means for acquiring or receiving data regarding the deposition pattern obtained from the plating solution taken out from the plating bath, and an analysis model that is machine-learned using the data or the processed data obtained by processing the data as teacher data, and outputs an analysis result of the deposition pattern or the plating solution based on the data or the processed data. And information presentation means for presenting, outputting, or transmitting information indicating the status of the plating bath from the analysis result. Here, the information presentation means preferably presents, outputs, or transmits information indicating the method of preparation or work for the plating bath when such work as preparation for the plating bath is necessary, together with or instead of the information indicating the status of the plating bath. Further, the information presentation means preferably presents, outputs, or transmits information indicating the status of the plating solution and the method of adjustment or work using other data regarding the plating solution together with the data or the processed data.

[0025] Furthermore, the analysis model may be configured to obtain an analysis result using other data regarding the plating solution together with the data or the processed data. Here, it is preferable to further include data reception means for receiving the data or the processed data included in the data acquisition means, and information transmission means for transmitting the information presented by the information presentation means. In this case, it is desirable to provide a transmission / reception site further including data transmission means configured to be able to transmit the data or the processed data that can be received by the data reception means, and information reception means configured to be able to receive the information transmitted by the information transmission means. This transmission / reception site preferably further includes information presentation means for presenting, outputting, or transmitting the information. It is desirable that the above transmission / reception site has the plating bath, a control unit of the plating bath, and test equipment for obtaining a deposition pattern from the plating solution.

Advantages of the Invention

[0026] According to this invention, it is possible to provide an analysis method and analysis apparatus that can quantitatively analyze the state of a plating solution using a unified standard by processing data relating to deposition patterns obtained from the plating solution, as well as a plating bath management method and management system using the analysis method. [Brief explanation of the drawing]

[0027] [Figure 1] Diagrams (a) and (b) illustrating the Hull cell test, and a diagram (c) illustrating the deposition pattern of the Hull cell cathode plate 13, along with a photograph (d) showing an example of image data of this deposition pattern. [Figure 2] The images shown are (a) image data of the deposition pattern, (b) processed data after noise reduction processing of the image data, (c) and (d) explanatory diagrams showing how the image data is divided and labeled, and (e) image data and (f) processed data obtained by differentiating this image data twice in the first direction (direction of change in current density) X. [Figure 3] Figure (a) shows image data of the deposition pattern; Figure (b) shows data of the widthwise central region obtained by removing both sides of the second direction (direction perpendicular to the first direction X) Y from the image data; Figure (c) shows representative values ​​of the widthwise central region data in the second direction Y arranged along the first direction X; Figure (d) shows the relationship with the axis corresponding to the reference current density in the deposition pattern; Figure (e) shows data of a rectangular central region centered on the axis; and Figure (f) shows data of a detection region for determining the type and amount of impurities in the plating solution. [Figure 4] This diagram schematically shows the structure of a CNN as an example of an analysis model that takes image data or processed image data as input and outputs the analysis results of a plating solution. [Figure 5] This diagram schematically shows the structure of a neural network (NN) that takes one-dimensional data in the first direction X as input and outputs the analysis results of the plating solution. [Figure 6]This diagram schematically illustrates a configuration in which each segmented region, obtained by dividing image data or processed image data, is labeled using a preprocessing model, and this labeled data is then analyzed using an analysis model. [Figure 7] This is a schematic block diagram illustrating the configuration of the analysis method and analysis apparatus according to the embodiment. [Figure 8] This is a schematic block diagram illustrating the configuration of an analysis method and analysis apparatus according to another embodiment. [Figure 9] This is a schematic block diagram illustrating the configuration of an analysis method and analysis apparatus in another embodiment. [Figure 10] This is a schematic block diagram illustrating the configuration of an analysis method and analysis apparatus in yet another embodiment. [Figure 11] This is a schematic block diagram showing a management system for managing the plating bath in the embodiment. [Figure 12] Image (a) shows the state after a predetermined image region has been extracted from the image data, and image (b) shows the state after noise has been removed from the image region. [Figure 13] This image shows the difference in deposition patterns in a nickel bath (Watt bath) depending on the presence or absence of a brightener. [Figure 14] This image shows a comparison of the deposition patterns when only a primary brightener is added to the plating solution and when only a secondary brightener is added. [Figure 15] The original image (a) shows the deposition pattern when 2.0 / L of primary brightener is added to the plating solution, and the image (b) shows the pattern after performing two differential operations in the first direction X. [Figure 16] The original image (a) shows the deposition pattern when 0.6 / L of primary brightener is added to the plating solution, and the image (b) shows the pattern after performing two differential operations in the first direction X. [Figure 17] This is an explanatory diagram for calculating the average value of the 100 data points in the center of the second direction Y. [Figure 18] Figure 17 is a graph showing the average value along the first direction X. [Figure 19]These are the film thickness data (a) along the first direction of the deposition pattern and the film thickness data (b) along the second direction. [Figure 20] Magnified photographs (a) and (b) show the difference in surface condition of the deposition pattern depending on the amount of primary brightener. [Figure 21] Images (a), (b), and (c) show the effect of the type and amount of impurities on the precipitation pattern. [Figure 22] These are photographs (a) and (c) and scanner images (b) and (d) of two deposition patterns in materials 1 and 2. [Figure 23] Figure (a) shows the measurement locations for surface roughness in the deposition pattern of the standard plating solution, and Figure (b) shows the measurement locations for surface roughness in the deposition pattern of the limit plating solution. [Figure 24] These are enlarged views (a) and (b) showing the measurement data of surface roughness in specific regions for each deposition pattern of the standard plating solution and the limiting plating solution. [Modes for carrying out the invention]

[0028] Next, embodiments of the present invention will be described in detail with reference to the attached drawings. First, embodiments of the plating solution analysis method and analysis apparatus according to the present invention will be described. In this embodiment, an example will be described in which the deposition pattern of a Hull cell cathode plate obtained by a Hull cell test is used.

[0029] Figure 1 is a diagram illustrating the deposition pattern used in this embodiment. In this embodiment, as shown in Figure 1(a), an anode 12 and a cathode 13 made of a metal plate or the like are placed in a Hull cell tester 11, and a plating solution 10 is placed inside the Hull cell tester 11. As shown in Figure 1(b), the Hull cell tester 11 is configured such that the cathode 13 facing the anode 12 is in an inclined position. By applying a voltage between the anode 12 and the cathode 13 and passing a current through the plating solution 10, a deposition pattern is formed on the cathode 13. At this time, due to the shape of the Hull cell tester 11, the current density is greatest in the region of the cathode 13 close to the anode 12 (the upper end of the cathode 13 shown in Figure 1(b)), gradually decreases towards the lower part of the diagram, and is smallest in the region farther from the anode 12 (the lower end of the cathode 13 shown in Figure 1(b)).

[0030] Figure 1(c) is a schematic diagram illustrating the deposition pattern formed on the surface of the cathode 13. As shown by the dotted hatching in this figure, the deposition pattern A on the surface of the cathode 13 is rectangular in shape, with sides (longer sides in the illustrated example) along a first direction X, which is the direction in which the current density changes, and sides (short sides in the illustrated example) along a second direction Y, which is perpendicular to the first direction X. An exposed area is provided on the upper part of the cathode 13 that is exposed above the liquid surface LS of the plating solution 10 and is not covered by the deposition pattern A. This deposition pattern A is surrounded by a high-current section HA, a liquid surface position LS, a low-current section LA, and a bottom surface position BT. Figure 1(d) shows a photograph of a typical surface configuration of the deposition pattern A (when a good plating solution is provided).

[0031] The surface characteristics of deposition pattern A described above are first digitized by generating image data using a camera, scanner, or the like. In this embodiment, a high-sensitivity, high-resolution scanner with the same configuration is used as the device for digitizing the data, and the data is digitized under the same optical conditions (configuration conditions of the imaging means (resolution and sensitivity) and illumination conditions). In this digitization process, by "maintaining a constant image acquisition environment," the acquired image data can be objectively compared, enabling analysis and judgment based on a stable standard. In particular, the pupil diameter of a human in dark places is 7-8 mm, with a variation of about 1 mm, so the brightness of deposition pattern A will change depending on the observer. This change in brightness has a significant impact on the analysis and judgment of the surface characteristics. In this embodiment, which performs digitization, such an impact can be eliminated.

[0032] Furthermore, in the above-mentioned digitization process, image data Ga of deposition pattern A is generated with a resolution of 600 dpi (dots per inch) or higher. Normally, the resolution of the human eye is about 0.1-0.2 mm, but by using a resolution of 600 dpi or higher, a pixel size of 50 μm or less on each side can be obtained. In particular, by using a resolution of 800 dpi or higher, a pixel size of 35 μm or less on each side can be obtained. In this way, by "acquiring images with sensitivity and resolution exceeding that of the human eye," accurate analysis and judgment become possible that surpasses the unobjective judgment criteria of experts. There is no particular upper limit to the resolution, but it is generally preferable to have a resolution of 60,000 dpi or lower. This is because within this range, even the fine surface roughness of the deposition pattern can be recognized. For example, the difference in surface characteristics of the SEM images shown in Figures 20(a) and (b) can be distinguished. However, in order to reduce the number of data and reduce the burden of image processing, it is preferable to have a resolution of 6,000 dpi or lower, and it is desirable to have a resolution of 1,200 dpi or lower. In these cases, it becomes difficult to distinguish subtle differences in surface characteristics. However, this is not a problem if the goal is to obtain analytical results that are less affected by these differences. Furthermore, analytical accuracy can be ensured by using other image data or non-image data such as surface roughness data in conjunction with the analysis.

[0033] In this embodiment, in addition to illumination conditions such as incident light wavelength and illumination light intensity, it is important to unify the optical conditions of the image acquisition environment, namely the configuration conditions of the imaging means, i.e., the detection sensitivity and resolution are constant, so that the digitization is always performed under consistent conditions. By using the same scanner, the configuration conditions of the imaging means can be kept constant, and the illumination conditions can also be set by the scanner, so the optical conditions can be easily kept constant. Furthermore, from the viewpoint of improving analysis accuracy, as will be described later, it is also effective to use multiple types of image data with deliberately different optical conditions in combination.

[0034] The image data Ga of deposition pattern A comprises pixels Gx of a number and size corresponding to the above resolution. If this image data Ga contains noise S caused by dust adhering to the pattern surface or scratches formed on the surface, which are unrelated to the state of the plating solution, as schematically shown in Figure 2(a), it is preferable to perform noise reduction processing using a noise reduction filter or the like to obtain image data Ga' from which the noise S has been removed, as schematically shown in Figure 2(b). The image processing at this time is preferably carried out according to the nature of the noise S to be removed. Generally, this noise reduction processing is performed by filtering using a median filter or a Gaussian filter. Examples of image data before and after noise processing are shown in Figure 12(a) - before processing and Figure 12(b) - after processing, respectively. In this example, a median filter suitable for removing fine lines is used.

[0035] As a preprocessing step for the above image data Ga, as schematically shown in Figure 2(c), multiple divided regions Gv are set in the image data Ga of deposition pattern A, and a labeling process (labeling) is performed on each of these divided regions Gv, setting a label Lv according to multiple types of surface characteristics (for example, those corresponding to the evaluation items mentioned above). Then, as schematically shown in Figure 2(d), a label planar array data Hv corresponding to the label Lv is generated for each divided region Gv of the image data Ga. In this labeling process, labels corresponding to the brightness distribution and color distribution of the image within a predetermined divided region Gv are set, and the corresponding label for each divided region Gv is determined by image processing. This makes it possible to clearly define the characteristics of the surface characteristics of deposition pattern A by the label Lv for each divided region Gv, thereby reducing the amount of data to be processed compared to the image data Ga and the image data Ga' after noise processing. In addition, the content and accuracy of analysis and judgment can be easily set by setting the labeling in the labeling process. Note that the above labeling process can also be performed using a preprocessing model that has been machine-learned using predetermined training data, as will be described later.

[0036] For the image data Ga and Ga' schematically shown in Figure 2(e), by performing differential processing twice along the first direction X, the differentiated data Gw schematically shown in Figure 2(f) is obtained, and the characteristics of the change in the surface appearance of deposition pattern A along the first direction X are clarified. In the illustrated example, this processing is performed by applying a horizontal differential filter (horizontal differential (difference) filter) twice. Here, an example of a horizontal differential filter is to set each element of a 3x3 matrix to -1 for the 2nd row 1st column, 0 for the 2nd row 2nd column, and +1 for the 2nd row 3rd column, and set all other elements (each column of the 1st row and each column of the 3rd row) to 0. The type of filter is not limited to this, and the horizontal coefficient matrix of a Sobel filter or Rrewitt filter may also be used. The processed data Gw obtained by this second differential analysis (second-order differential analysis) shows the fluctuations in the rate of change of brightness and color in the first direction X of the image data Ga and Ga' by multiple lines that are approximately aligned with the second direction Y. Therefore, changes in the surface configuration of the deposition pattern A in the first direction X become clearly and easily recognizable.

[0037] Image (a) in Figures 15 and 16 is an example of the original image data Ga, and image (b) in Figures 15 and 16 is an example of the processed data Gw after double differentiation. The differentiated data may be data obtained by taking the first derivative (difference) of the image data Ga and Ga' only once in the first direction, or by taking three or more steps of differentiation. However, by performing double differentiation (second derivative), the change in surface characteristics along the first direction X can be made clearer, and less extraneous noise is generated. In addition, in each differentiation process, for example, a threshold value of a predetermined ratio to the maximum value of the derivative after processing can be set, and values ​​below this threshold can be removed as noise.

[0038] The image data Ga,Ga′ schematically shown in Figure 3(a) are arranged in a planar range extending in a first direction X and a second direction Y perpendicular to it. Here, as mentioned above, the first direction X is the direction in which the current density changes in the Hull cell test, and along this direction, the change in the surface configuration of the deposition pattern due to the change in current density is reflected. Therefore, in order to grasp this change, it is necessary to accurately extract the change in the surface configuration along the first direction X. In this embodiment, as shown by the dotted line in Figure 3(a), data near both ends of the second direction Y of the image data Ga,Ga′ is removed, and the widthwise central region data Gb of the second direction Y, corresponding to the widthwise central region B shown in Figure 1(c), is extracted as shown in Figure 3(b). Then, the average value of the data in each column along the second direction Y is calculated from this widthwise central region data Gb, and one-dimensional data Hx consisting of one row of data columns of this average value is obtained.

[0039] This one-dimensional data Hx is generated by pre-removing data from both ends of the second direction Y, thereby excluding the influence of the liquid surface LS (influence of liquid surface LS on the deposition state) and the influence of the bottom surface BT (influence of the container bottom surface and stirring on the deposition state) during the Hull cell test. Furthermore, by determining the representative value of the second direction Y of the remaining widthwise central region data Gb (in the following example, it is the mean value, but a representative value such as the median may also be used), the influence along the second direction Y is reduced, so that changes in the surface appearance of deposition pattern A along the first direction X can be reliably and accurately extracted. In other words, in deposition pattern A, by focusing on the data near the center, which is a stable plating adhesion area, and taking its mean value or other representative value, the two-dimensional data is converted to one-dimensional data, making machine learning easier. Instead of image data Ga, Ga', the widthwise central data Gb may be used as image data without one-dimensionalization. Alternatively, similar to the widthwise central region data Gb, the aforementioned differential processing data Gw may be formed in a range excluding both ends of the second direction Y, and this can be used for analysis.

[0040] Figure 17 is an explanatory diagram showing how 100 pixel data points arranged along the second direction Y are retained as width-central region data Gb from the actual image data Ga, Ga', and the average value of these 100 data points in the second direction Y is calculated. Figure 18 is a graph showing the one-dimensional data Hx, which is the average value arranged along the first direction X. The increase in light intensity on the right side of the graph is thought to be due to diffuse reflection occurring and the resulting increase in the reflectance intensity for light, as the surface of the low-current region LA of the Hull cell cathode plate is rougher compared to the surface of the high-current region HA. This is due to the difference in current density during the Hull cell test.

[0041] As shown in Figure 3(d), the reference current density (for example, 4 A / dm²) can be determined from the image data Ga and Ga' of deposition pattern A. 2Centered on the axis Cx indicating the corresponding position of ), the rectangular central region data Gc corresponding to the rectangular central region C of deposition pattern A shown in Figure 1(c) is extracted as shown in Figure 3(e). This rectangular central region data Gc is suitable image data for deriving how much appropriate deposition region exists around the center when viewed from the axis Cx which indicates the reference current density when actually operating the plating bath. Preferably, for the second direction Y, this rectangular central region data Gc is the region excluding both ends. Also, for the first direction X, preferably, it is the region excluding at least a part of the high current section HA and the low current section LA. The reason for these is to exclude the influence of boundary positions such as the liquid surface LS and bottom surface BT of the plating solution. By using rectangular central region data Gc, it is possible to understand the optimal range of surface characteristics of the deposition pattern centered on the axis Cx (length and range in the first direction X). This makes it easier to understand the range that can withstand changes in the current density of the plating solution, and the stability of the plating bath and the margin for error in changing conditions can be determined using the analysis model AM described later. To obtain such analysis results and output values, using the limited rectangular central region data Gc as described above makes it easy to improve the accuracy of the analysis results showing the stability and margin of error. However, it is also acceptable to obtain the above analysis results using image data Ga and Ga'.

[0042] Figure 3(f) is an explanatory diagram showing extracted data for understanding the type and amount of impurities in the plating solution. In this example, boundary region data Gd is extracted, corresponding to the L-shaped boundary region D which includes the band-shaped region along the upper edge of deposition pattern A shown in Figure 1(c) and the low-current region LA of the same deposition pattern A. In this boundary region data Gd, as the impurity content increases, a white region appears near the liquid surface LS, or a shadow is formed in the low-current region LA. Here, impurities are usually metals that dissolve from the object to be plated, such as Cu, Zn, Fe, Pb, Al, Cr, and Ca. Other impurities include ions such as sodium, potassium, cyanide, nitric acid, phosphoric acid, and ammonium, as well as peroxides and organic impurities. Figure 21 shows a comparison of images of deposition pattern A with 10 ppm of Zn mixed in the plating solution (a), with no impurities mixed in (b), and with 10 ppm of Cu mixed in (c). Thus, it can be seen that even in trace amounts, the presence of impurities causes characteristic features to appear near the liquid surface LS in the boundary region D and in the low-current region LA. Furthermore, the accuracy of the analysis results can be easily improved by using the limited boundary region data Gd as described above. However, it is also acceptable to obtain the above analysis results using image data Ga and Ga'.

[0043] Figure 4 is a schematic diagram illustrating the configuration of a CNN (Convolutional Neural Network) as an example of an analysis model used in this embodiment. This CNN-based analysis model AM takes planarly arranged data information (image data) that reflects the surface characteristics of deposition pattern A, such as the image data Ga, Ga', label planar array data Hv, differential processing data Gw, widthwise central region data Gb, rectangular central region data Gc, and boundary region data Gd, as input. The structure of this analysis model AM includes a feature extraction unit composed of a convolutional layer and a pooling layer that take the above data information as input, or two or more repetitions thereof, and a recognition unit that takes the pixel data of the feature map extracted by the feature extraction unit as input values. In the figure, circles virtually represent nodes in each layer, and squares virtually represent operators. The feature extraction unit can efficiently extract features of the array pattern of the above data information, and the feature map is processed in the recognition unit, which is composed of fully connected layers, etc., to form an output corresponding to the analysis result. In the illustrated example, there are three output nodes, and the system is shown analyzing the plating solution using a classification function that categorizes (classifies) the image data into one of three categories. However, the number of classifications can be any number other than three, and the system may be configured to calculate an output value that identifies the amount of a specific component by using the output function OF shown by the dotted line in the illustration from multiple output nodes. Examples of output functions OF include the softmax function, logistic function (sigmoid function), and Gaussian function.

[0044] The above analysis model AM is a pre-trained model that uses pre-prepared image data such as Ga, Ga', Gb, Gc, Gd, Gv, and Gw as input values, and employs machine learning to train the model using information indicating the state (surface configuration) of deposition pattern A or the state of plating solution 10 corresponding to these image data as training data. This machine learning is performed by inputting a large number of training data values ​​into the analysis model AM and comparing the analysis results with the correct training data values. Typically, machine learning (in this case, deep learning) is performed using an error function (loss function, evaluation function) calculated from the analysis results and the correct training data values, for example, by known optimization methods such as gradient descent or backpropagation. For fully connected layers, dropout may be set as appropriate to avoid overfitting.

[0045] Figure 5 schematically shows an example of an analytical model AM with a neural network (NN) structure consisting of fully connected layers. This analytical model AM is also basically a trained model that has been machine-learned using training data that has input values ​​and corresponding output values. This analytical model AM is effective when the input value is meaningful array data as one-dimensional data, such as the one-dimensional data Hx mentioned above, but even data with planar arrays (matrix data), such as the image data Ga mentioned above, can be expanded as one-dimensional data and input. The meaning of the symbols in the network diagram, the output function OF, the machine learning method, and dropout are the same as in Figure 4.

[0046] Figure 6 schematically shows an example in which a pre-processing model PM is used for labeling before the analysis model AM. In this example, the pre-processing model PM determines the label Lv corresponding to each divided region Gv according to the image characteristics of the divided regions Gv (originally part of the surface characteristics of deposition pattern A) obtained by dividing the image data Ga and Ga' shown in Figure 2(c), and forms the label planar array data Hv shown in Figure 2(d). Then, using this label planar array data Hv as an input value, the analysis model AM outputs the analysis result, and the output value is obtained by the output function OF as needed. In this case, the pre-processing model PM may be a trained model as shown in Figures 4 and 5 above, but it may also be a functional implementation means realized by executing a computer program that simply assigns the label Lv corresponding to each divided region Gv as a result of image processing, or it may consist of a dedicated processing circuit.

[0047] Figure 7 is a schematic diagram illustrating the overall process from precipitation pattern A to obtaining analysis results. Image data G0 of precipitation pattern A is acquired, and image data G1 formed by processing this image data G0 as appropriate is input to a machine learning-prepared analysis model AM. Analysis results and output values ​​are obtained via the output function OF as needed. Finally, information indicating the analysis results is displayed on an output device DP, which is an information display means such as a display, printer, or transmitter. Here, the image data G1 may be the aforementioned image data Ga, Ga', or it may be data Gb, Gc, Gd, Hv, etc., which can be displayed as various images (planar data arrays) obtained from a part of precipitation pattern A.

[0048] Next, we will describe a more specific method and apparatus for analyzing the plating solution. Figure 13 shows an example of deposition pattern A using a Watt bath for bright nickel plating. The photograph on the left of Figure 13 shows deposition pattern A obtained from a plating solution without any brightener added, and the photograph on the right shows deposition pattern A obtained from a plating solution with a brightener added. As shown above, using a brightener improves the coverage of the plating and achieves a uniform and glossy surface. This is achieved by adding saccharin sodium dihydrate (hereinafter simply referred to as "saccharin") as the primary brightener and 2-butyne 1,4-diol (hereinafter simply referred to as "butyne diol") as the secondary brightener. The primary brightener mainly suppresses internal stress and softens the plating film, while the secondary brightener mainly provides a smoothing (leveling) effect. By adding appropriate amounts of both, good coverage and gloss of the plating film can be obtained. Figure 14 shows the results when only a primary glossing agent is used and when only a secondary glossing agent is used, but neither surface appearance is as good as when both are used.

[0049] The plating solution used in this embodiment is a Watt bath for bright nickel plating, similar to the one described above. The standard composition consists of 240 g / L of nickel sulfate as a metal ion source, 45 g / L of nickel chloride as an anodic dissolution accelerator, 30 g / L of boric acid as a pH buffer, 2 g / L of the above-mentioned saccharin as a primary brightener, and 0.2 g / L of the above-mentioned butynediol as a secondary brightener. Here, if the basic composition other than the additives is significantly changed, the deposition pattern A may change significantly. As an example, when the amount of boric acid was reduced to 0 g / L, nickel hydroxide was deposited in the high-current area HA, and charring and cracks appeared on the surface. Also, even when the amount of boric acid was 15 g / L, a dull area occurred in a narrow range in the high-current area H. In situations like this, it is possible to output the amount of basic composition such as boric acid as an analysis result using the analysis model AM shown in Figures 4-7, based on image data Ga, Ga' based on precipitation pattern A, and data Gb in the central region in the width direction. Furthermore, in this case, since it is relatively easy to create data Hv with labeling processing as shown in Figures 2(c) and (d), and also in Figure 6, it is effective to use the analysis results of the analysis model AM that uses labeled data Hv as input values.

[0050] The amounts of the basic composition described above cause relatively large fluctuations in deposition pattern A, making it easy to obtain analytical results. Furthermore, such a basic composition can also be achieved through chemical analysis of the plating solution 10. In contrast, the types and amounts of additives are difficult to accurately detect through chemical analysis, and even if they can be detected, their effects are difficult to grasp because they may be affected by other components and usage time. Here, additives that exert an adhesion-inhibiting effect as a deposit mechanism include saccharin, benzothiazole, thiourea, Janus Green B (JGB), benzalacetone, lead, and bismuth; additives that exert an interfacial complex-forming effect include chloride ions, CN, SCN, sulfur compounds (thiourea, 3,3'-dithiobis(1-propanesulfonic acid) disodium (SPS), dimercaptothiadiazole (DMTD), etc.), boric acid, oxalic acid, and malonic acid; and additives that exert a film-forming effect include polyethylene glycol (PEG), nonylphenyleicosaethylene glycol ether (NPEGE), polyvinyl alcohol, and gelatin. In addition, additives that exert an electrolytic diffusion consumption mechanism include unsaturated alcohols (butyldiol, propagyl alcohol, coumarin, etc.), NO3, and Fe 3+ These are some examples.

[0051] For example, in the standard composition of the above-described example, the amount of primary brightener is 2.0 g / L. However, in the surface characteristics of deposition pattern A in the Hull cell test, the primary brightener is indistinguishable in the range of 1.0 g / L to 2.0 g / L, and in the range of 0.5 g / L to 1.0 g / L, even a skilled craftsman can barely distinguish it. In fact, even when the primary brightener was varied within the range of 0.5 g / L to 2.0 g / L, it was not possible to perceive any difference in the surface characteristics of deposition pattern A with the naked eye. On the other hand, the difference between primary brightener at 0.5 g / L and 0.0 g / L was clearly distinguishable in the gloss state of deposition pattern A. However, the important point for actually knowing the state of the additives in the plating solution 10 and obtaining a stable gloss state of plating is not the presence or absence of primary brightener, but the fact that the amount of primary brightener, which is difficult to distinguish in the appearance of deposition pattern A, can be distinguished within the aforementioned range of 0.5 g / L to 2.0 g / L.

[0052] Therefore, in this example, a standard plating solution with a primary brightener at the standard composition (2.0 g / L) and a limiting plating solution with a primary brightener at 0.6 g / L were used, and Hull cell tests were conducted on each under the same conditions. 500 ml of each plating solution was prepared, approximately 250 ml was taken out for the Hull cell test, the solution was returned to 500 ml after the test, stirred, and then another 250 ml was taken out for the Hull cell test. This process was repeated. Image data Ga was then created from the deposition patterns A obtained in this test. Noise reduction was performed by applying a median filter to prepare image data Ga'.

[0053] A portion (70%) of the above image data Ga' was used as training data, and machine learning was performed on the analysis model AM shown in Figure 4. Here, a 3x3 filter was used in the convolutional layer, a 2x2 pooling layer was used, and the total number of fully connected layers was 3. The dropout rate was set to 0.2. The output layer had two nodes: one for the standard plating solution and one for the limiting plating solution. As a result, the remaining (30%) of the image data Ga' was used as validation data (test data), and a positive correlation was observed between the output values ​​obtained from the two nodes and the correct answer.

[0054] As shown in Figures 15 and 16, image processing involving second differentiation was performed on the image data Ga' of the standard plating solution and the limiting plating solution, respectively, to form differentiated data Gw with a linear pattern. This differentiated data Gw was used as the input value for the analysis model AM shown in Figure 4. The number of output nodes at this time was also two, standard and limiting. Through the image processing involving second differentiation described above, it was possible to process the image so that a clear difference was created between the input image of the standard plating solution and the input image of the limiting plating solution, as shown in the images in Figures 15(b) and 16(b). As a result, after machine learning of the analysis model AM shown in Figure 4, a trained model that can obtain a high accuracy rate was obtained.

[0055] Here, the original images of deposition pattern A shown in Figures 15(a) and 16(a) are indistinguishable to the human eye. However, by processing the scanned data with higher sensitivity and resolution than the human eye, the features can be easily grasped by the human eye. Here, the area emphasized by the second derivative is the location where the change in reflected light intensity was greatest (white line in the figure). This location continues continuously in the second direction Y, clearly showing that the transition of the plating state along the second direction Y is continuous. Comparing Figure 15(b) and Figure 16(b), in the standard image, the emphasized lines are concentrated to the right, and these lines are straight along the second direction Y. In contrast, in the limit image, the emphasized lines extend to the left, and these lines are slanted. In this way, even samples that could not be directly judged by the human eye can have their features judged through image processing. This means that the accuracy of the analysis results can be improved by image processing of deposition pattern image data.

[0056] On the other hand, for the image data Ga′ of the standard plating solution and the limiting plating solution, one-dimensional data Hx shown in Figure 18 was generated using the method shown in Figure 17, and this was used as the input value for the analysis model AM shown in Figure 5. Furthermore, a sigmoid function was used as the output function OF to obtain output values ​​(probability values ​​for the standard and limiting solutions). A portion of the above one-dimensional data Hx was used as training data to train the analysis model AM shown in Figure 5. Subsequently, the output loss function (error function) was calculated and machine learning was advanced to improve the accuracy, and finally, the above analysis model AM was trained to be able to achieve an accuracy of approximately 80-90%.

[0057] Next, with reference to Figure 8, another embodiment of various analysis methods and analysis devices using the analysis model AM, which has been machine-trained using training data in the same manner as described above, will be explained. As an analysis method and analysis device using the deposition pattern A of the plating solution, the analysis model AM, which has been machine-trained using both image data G1 based on the deposition pattern A and non-image data D1 based on data other than the deposition pattern A, as training data, is used. Here, the non-image data D1 includes surface roughness data of the deposition pattern A (especially the surface roughness value limited to a specific area), and the overall light reflectance and scattering coefficient of the deposition pattern A (especially their representative values). As shown in Figure 19, the film thickness distribution along the first direction X of the deposition pattern A of the standard plating solution (solid line) and the limiting plating solution (dashed line) (graph (a) on the left side of the figure) and the axis Cx (current density 4A / dm²) are shown. 2 The film thickness distribution of deposition pattern A, such as the film thickness distribution along the second direction Y (graph (b) on the right side of the figure) above the position corresponding to the above, shows almost no difference between the standard plating solution and the limiting plating solution mentioned above, making it unsuitable for analyzing the amount of primary brightener. However, when analyzing the amount of components that change the current density characteristics of the deposition rate, for example, when analyzing the amount of metal ion supply source in the basic composition, the amount of anodic dissolution accelerator, the amount of pH buffer, etc., the accuracy of the analysis can be improved by adding and inputting the film thickness distribution along the first direction X, as shown in Figure 19(a), to the image system data.

[0058] Figure 20 shows a surface electron microscope (SEM) image of the region between the center of the first direction X in the widthwise center of deposition pattern A and the low-current section LA, in order to observe the difference in surface characteristics between the standard plating solution and the limiting plating solution. As can be seen from these images, although the surface characteristics are actually quite different, the difference in surface characteristics is difficult to discern from the image data G1 described above. Therefore, as non-image data D1, for example, surface roughness data measured with a surface roughness meter (for example, array data of surface roughness along the first direction X) can be used. By using this, the accuracy of the analysis can be further improved. In particular, the accuracy of the analysis can be improved by using data within a specific region corresponding to the analysis result of interest.

[0059] As mentioned above, the method and apparatus using the analytical model AM shown in Figure 8 are one solution to the problem that manual analysis of deposition pattern A is extremely difficult, and even if the analysis itself can be done, the standards vary from person to person, and there are large differences due to experience, making it impossible to obtain a stable and objective evaluation. Non-image data D1 includes, in addition to the film thickness data and surface roughness data mentioned above, the temperature at the time of deposition, chemical analysis results such as basic composition and impurities, data on the use of the plating solution (for example, current value A × deposition time T), and alloy composition data of deposition pattern A (in the case of alloy plating such as solder plating).

[0060] Figure 9 is a schematic diagram illustrating different embodiments of various analysis methods and analysis devices using an analysis model AM that has been machine-learned using training data, similar to those described above. In this embodiment, an analysis model AM is used to analyze the surface characteristics of deposition pattern A and the state of the plating solution, taking two image data sets G1 and G2 as input, as multiple types of data. Furthermore, the machine learning of this analysis model AM is performed using training data of the two image data sets G1 and G2. Note that the number of types of image data to be input is not limited to two, but may be three or more. The reason for using multiple image data sets as input to the analysis model is that, although this embodiment has advantages in terms of objectivity of analysis and accuracy of analysis criteria compared to conventional manual analysis when analyzing the state of the plating solution based on the surface characteristics of deposition pattern A, it remains difficult to improve the accuracy of the analysis.

[0061] For example, even when determining the type and amount of impurities in Figure 21 using the boundary region data Gd, the surface characteristics of the region near the liquid surface LS of deposition pattern A and the low-current region LA (see Figure 1(c)), such as hue and brightness, are subtle, making it difficult to improve discrimination accuracy. In such cases, discrimination accuracy can be improved by inputting other image data, such as rectangular central region data Gc, in addition to the boundary region data Gd. Similarly, as shown in Figure 22, for materials 1 and 2 of deposition pattern A, photographs taken with a camera (upper row) and images input with a scanner (lower row) yield completely different images despite being the same material. Therefore, even with the same image data, different appearances can be seen depending on the input method and optical conditions, and different information can be extracted from these different appearances. Thus, by using multiple image data G1 and G2 in combination, the analysis accuracy of the analysis model AM can be improved.

[0062] Figure 23 shows photographic diagrams (a) and (b) illustrating the measurement range of the surface roughness of deposition pattern A with the standard plating solution and the surface roughness of deposition pattern A with the limiting plating solution. For both deposition patterns, the surface roughness was measured by scanning the surface roughness meter along the first direction X, with the center of the second direction (width direction) Y indicated by the arrow. The surface roughness meter used was the stylus-type step / surface roughness meter "SurfCorder ET3000i" (Kosaka Laboratory). In the overall measurement range 1 of deposition pattern A, as shown in Figure 23, the measurement distance in the first direction X was 96 mm, the vertical and horizontal magnification was 1000 times, the cutoff value was 0.25 mm, and the feed rate was 0.05 mm / s. However, as a specific region within the measurement range 1 described above that is susceptible to the concentration of the glossing agent, a measurement range 2 was set with a center point 10 mm back from the right end of the first direction X in the diagram, and an range of ±0.125 mm in the X direction (totaling 0.25 mm). For this measurement range 2, measurements were taken in more detail with a vertical and horizontal magnification of 10,000 times and a feed rate of 0.02 mm / s.

[0063] In the surface measurement of deposition pattern A described above, both the standard plating solution and the limiting plating solution are affected by the curvature of the metal plate constituting the cathode 13. Therefore, a leveling process was performed to match the values ​​at both ends of the X direction of the data in measurement range 1. After this process, the data in measurement range 2 was further leveled to match the values ​​at both ends of measurement range 2. The surface unevenness data in measurement range 2 after these processes is shown in Figure 24. Here, the data for the deposition pattern of the standard plating solution is shown in Figure 24(a), and the data for the deposition pattern of the limiting plating solution is shown in Figure 24(b). However, in Figure 24(a), one division on the vertical axis is 0.01 μm, and in Figure 24(b), one division on the vertical axis is 0.02 μm. Note that one division on the horizontal axis is 0.02 mm in both cases. The surface roughness of the deposition pattern of the standard plating solution in measurement range 2 in Figure 24(a) was 0.217 μm in RMS (root mean square) value, and the surface roughness of the deposition pattern of the limiting plating solution in measurement range 2 in Figure 24(b) was 0.402 μm in RMS value. This relationship of surface roughness is consistent with the correspondence shown in the SEM images in Figure 20. In this example, RMS values ​​(the square root of the mean of the squares of each data point) were used as surface roughness data, but it is natural that various data (numerical values) such as arithmetic surface roughness Ra, maximum height Rz, and root mean square height (standard deviation, i.e., the square root of the mean of the squares of the differences between each data point and the mean) Rq can be used.

[0064] As the non-image data D1 and D2 mentioned above, surface roughness data (such as the RMS value) of a specific region (measurement range 2) of deposition pattern A can be used. On the other hand, regarding the surface roughness of deposition pattern A, it is not limited to the non-image data mentioned above; for characteristics that depend on the distribution of surface roughness, mesh data of the surface roughness of deposition pattern A can also be used as the image data G1 and G2 mentioned above.

[0065] Figure 10 is a schematic diagram illustrating an embodiment of an analysis method and analysis apparatus using the analysis model AM, which takes two non-image data sets D1 and D2 as input. Any of the aforementioned non-image data sets can be used as multiple types of data. For example, non-image data can include the above-mentioned film thickness data, surface roughness data, temperature during deposition, chemical analysis results such as basic composition and impurities, plating solution usage data (e.g., current value A × deposition time T), and alloy composition data of deposition pattern A (in the case of alloy plating such as solder plating). The combination of multiple non-image data sets is arbitrary, but it is preferable that at least one of them is distribution data relating to at least a part of the deposition pattern (e.g., the above-mentioned one-dimensional data, film thickness data along the first direction X, surface roughness data along the first direction X). Furthermore, it is preferable that the combination of multiple types of data includes the necessary information depending on the analysis result to be obtained. For example, the above-mentioned one-dimensional data and film thickness data, or chemical analysis data of the basic composition and surface roughness data can be used.

[0066] Furthermore, in the analysis method and analysis apparatus using multiple types of data shown in Figures 8 to 10 above, for example, multiple input layers corresponding to each type of data are prepared, and after configuring them to perform data processing for each type, the final output or output value can be obtained by integrating the outputs. For example, two or more image data sets may be input into separate input layers dedicated to each data type, feature extraction may be performed on each, and then the data may be integrated in a fully connected layer, or the outputs may be integrated to obtain the optimal output or output value. In addition, even with non-image data, if it is data with a large number of parameters, such as one-dimensional data, a dedicated input layer corresponding to that data may be provided as described above. However, in the case of data with a small number of parameters, such as temperature or processing time (A·T), it may be used only as correction data to modify the output of the original analysis model. In any case, by using multiple types of input data that are thought to have a correlation with the required analysis results, it is possible to improve the accuracy of the analysis results.

[0067] Figure 11 is a schematic diagram illustrating an embodiment of the plating bath management method and management system according to the present invention. Here, the plating bath 1 is installed in a factory or the like for performing plating treatment, and the temperature state, stirring state, current application state, etc. are controlled by the plating bath control unit 1a in the factory or the like. From the plating bath 1, the plating solution 10 is taken periodically or irregularly and tested using the Hull cell tester 11 described above. If necessary, a uniform electrodeposition test (Hurling cell test) may be performed along with the Hull cell test, or instead, a vent cathode test may be performed to confirm the coating properties. In any case, image data Ga, Ga', other processing data and other data are obtained from the deposition pattern A obtained from each test, and these data are output directly or transmitted via a transmitter by wire or wireless. These data may be image data or non-image data, but in this embodiment, they are image data G0. This image data G0 is subjected to appropriate processing such as noise reduction and data processing as necessary, and is input to the analysis model AM as image data G1. Here, the analysis model AM can be any of the models mentioned above. Furthermore, the analysis results are output via the output function OF as needed, and these results are output to the storage device 1M or the management device 1S.

[0068] The management device 1S manages information such as the basic composition of the plating bath 1, additives such as brighteners, usage history, plating specifications, information on the material to be plated, and chemical analysis results such as the type and concentration of impurities. It stores information and calculation results input from external sources in the storage device 1M and is configured to process information retrieved from the storage device 1M, information input from the plating bath control unit, or analysis results obtained from the analysis model as needed. The management device 1S can be configured by a computer device such as an MPU (microprocessor unit). Physically, the functions of the analysis model AM and output function OF may be incorporated into a part of this management device 1S.

[0069] The control device 1S receives the output of the analysis model AM or the output value of the output function OF, and as an internal processing result, outputs or transmits information to the plating bath control unit 1a using the analysis results output by the analysis model AM. If necessary, in addition to the analysis results, information collected from the plating bath control unit 1a or information read from the storage device 1M may be added to output and transmit the above information. The control device 1S may also consider the history information of the plating bath 1 to determine what information should be output or transmitted and at what timing. This information includes, for example, the analysis results, or the basic composition, amount of additives, amount of impurities obtained from the analysis results and other information, or information about problems with the plating solution obtained from these, and furthermore, information about the processing and operations required for the plating bath obtained from this information. Examples of processing and operations required for the plating bath include adding chemicals that constitute the basic composition, adding additives, weak electrolytic treatment (used to remove Cu and Zn that tend to precipitate in the constant current section LA) and activated carbon treatment (a method to remove organic impurities, with oxidizing agent treatment also used if necessary) which are performed depending on the type and amount of impurities.

[0070] As a management method and management system for the plating bath 1, all the components shown in Figure 11 may be located within the same factory. However, for example, the plating bath 1 and the Hull cell tester 11 may be located at a first location such as a factory, while the management device 1S, recording device 1M, analysis model AM, etc., may be installed at a second location separate from the first location. Here, the first location and the second location may be relatively distant locations such as different offices or different companies. In this case, it is preferable that the plating bath control unit 1a and the management device 1S are each equipped with transceivers and are configured to communicate with each other via wired or wireless communication devices. In this way, deposition pattern A can be prepared at the first location (plating site) such as a distant factory, and various data such as image data related to this deposition pattern A can be transmitted from the plating bath control unit 1a to the management device 1S at the second location, and various information can be transmitted from the management device 1S to the plating bath control unit 1a. This allows the positional relationship between the site and the management device 1S to be arbitrarily set.

[0071] It should be noted that the method and apparatus for analyzing plating solutions, as well as the method and management system for managing plating baths according to the present invention, are not limited to the examples described above, and various modifications can be made without departing from the spirit of the present invention. For example, while second differentiation (second derivative) is most preferable for the differential processing when emphasizing the features of image data, it is not limited to this, and can be done once or three or more times. In addition, the machine learning method (e.g., weighting of data for each pixel) and the effective image range (e.g., white lines adhering near the liquid surface) can be changed to suit the evaluation items for deposition pattern A (gloss, semi-gloss, pits, blemishes, etc.).

[0072] Furthermore, although the above embodiments described plating solutions and plating baths (such as Watt baths, citric acid baths, and sulfamic acid baths) for nickel plating as examples, the present invention can also be used for the deposition patterns of plating solutions and plating baths for copper plating (such as copper sulfate baths and cyanide baths), chromium plating (such as Sargent baths and fluoride baths), zinc plating (such as cyanide baths, zincate baths, and ammonium chloride baths), gold plating (such as alkaline baths and acidic baths), silver cyanide plating, tin sulfate plating, electroless nickel plating, and electroless copper plating.

[0073] Furthermore, by indicating the candidate positions of the evaluation items automatically detected by the computer to the Hull cell evaluator using markers or other indicators on a display image such as an LCD monitor, it is expected that careless errors by the evaluator can be reduced. Also, for example, a chemical manufacturer that sells plating solutions can accurately judge the condition of the customer's plating solution (using only data without moving any physical objects) by having the customer transmit a Hull cell pattern created on-site, and then provide feedback to the customer as management data. [Explanation of Symbols]

[0074] 10…Plating solution, 11…Hull cell tester, 12…Anode, 13…Cathode, A…Deposition pattern, GaGa′…Image data, Gx…Pixel, Gv…Divided region, Gw…Processing data, X…First direction, Y…Second direction, Hx…One-dimensional data, 1…Plating bath, 1a…Plating bath control unit, 1S…Management device, AM…Analysis model, 1M…Recording device, G1,G2…Image data, D1,D2…Non-image data, DP…Output device, PM…Pre-processing model

Claims

1. We import data on the deposition pattern obtained from the plating solution. By using a machine learning-trained analysis model that uses the aforementioned data or the processed data obtained by processing the aforementioned data as training data, A method for analyzing a plating solution, which obtains the deposition pattern or the analysis result of the plating solution based on the aforementioned data or processing data, The data mentioned above is image data of the surface configuration of the deposition pattern. The aforementioned precipitation pattern was obtained by a Hull cell test. The analysis model performs the analysis based on the processing data which shows the change in the surface configuration of the deposition pattern along a first direction corresponding to the change in current density of the image data. The processed data is one-dimensional data obtained by arranging representative values ​​obtained in a second direction orthogonal to the first direction of the image data along the first direction. Method for analyzing plating solutions.

2. We import data on the deposition pattern obtained from the plating solution. By using a machine learning-trained analysis model that uses the aforementioned data or the processed data obtained by processing the aforementioned data as training data, A method for analyzing a plating solution, which obtains the deposition pattern or the analysis result of the plating solution based on the aforementioned data or processing data, The data mentioned above is image data of the surface configuration of the deposition pattern. The analysis model is machine-trained using the processed data formed by dividing the image data into multiple regions and labeling each divided region to one of multiple types of surface configurations of the deposition pattern, and performs analysis based on the processed data. Method for analyzing plating solutions.

3. The analysis model derives the type or amount of components contained in the plating solution from the data or the processing data. The method for analyzing a plating solution according to claim 1 or 2.

4. The aforementioned component is an additive. The method for analyzing a plating solution according to claim 3.

5. The analysis model derives the type or amount of impurities contained in the plating solution from the data or the processing data. A method for analyzing a plating solution according to any one of claims 1 to 4.

6. The aforementioned image data is data generated from the deposition pattern with a resolution of 600 dpi or higher. A method for analyzing a plating solution according to any one of claims 1 to 5.

7. The image data is generated from the aforementioned deposition pattern under constant optical conditions. The method for analyzing a plating solution according to claim 6.

8. The processed data is differential processed data obtained by differentiating the image data along the first direction. The method for analyzing a plating solution according to claim 1.

9. The processed data is second derivative processed data obtained by taking the second derivative of the image data along the first direction. The method for analyzing a plating solution according to claim 8.

10. The aforementioned representative value is the mean value. The method for analyzing a plating solution according to claim 1.

11. The processing data is based on data from the central region in the width direction of the image data, excluding data near both ends in a second direction perpendicular to the first direction. The method for analyzing a plating solution according to claim 1.

12. A data acquisition means for acquiring or receiving data related to the deposition pattern obtained from the plating solution, An analysis model is machine-trained using the aforementioned data or the processed data obtained by processing the aforementioned data as training data, and outputs the analysis results of the deposition pattern or the plating solution based on the aforementioned data or the processed data. Information presentation means for presenting, outputting, or transmitting the aforementioned analysis results or information indicating the status of the plating solution obtained from the aforementioned analysis results, It is equipped with, A plating solution analysis apparatus that performs the plating solution analysis method according to claim 1 or 2 using the aforementioned analysis model.

13. A first step of acquiring or receiving data regarding the deposition pattern obtained from the plating solution removed from the plating bath, A second step involves taking in the aforementioned data and using a machine learning-trained analysis model that uses the aforementioned data or the processed data obtained by processing the aforementioned data as training data to obtain the analysis results of the deposition pattern or the plating solution based on the aforementioned data or the processed data. A third step of presenting, outputting, or transmitting information indicating the status of the plating bath based on the analysis results, or information indicating the preparation and work methods for the plating bath. It is equipped with, A method for managing a plating bath, wherein the second step involves performing the method for analyzing the plating solution according to claim 1 or 2.

14. A data acquisition means for acquiring or receiving data relating to the deposition pattern obtained from the plating solution removed from the plating bath, An analysis model trained using the aforementioned data or processed data obtained by processing the aforementioned data as training data, wherein the analysis model outputs the analysis results of the deposition pattern or the plating solution based on the aforementioned data or the processed data, Information presentation means that presents, outputs, or transmits information indicating the status of the plating bath based on the analysis results, or information indicating the preparation and work methods for the plating bath. It is equipped with, A plating bath management system wherein the method for analyzing a plating solution according to claim 1 or 2 is performed using the aforementioned analysis model.

Images