Construction Method and System of CCD Detection Model for Metal Coating Based on GAN

Abstract

The present invention discloses a method and system for constructing a CCD detection model for metal coatings based on GAN, belonging to the technical field of metal coating detection of semiconductor electronic components. The GAN network constructed in the present invention outputs image data processing results within the threshold range as "qualified products"; while the output image data processing results beyond the threshold range are "defective products", and the high-quality requirements of electronic electroplating of semiconductor electronic components are strictly managed according to this standard. In addition, a total mass threshold discrimination module for the plated parts is used to simulate and monitor the appearance, shape, color and overall morphological characteristics of semiconductor devices after electronic electroplating. By meeting the standard with a minimum threshold of -5% of the coating specification value and a maximum threshold of +5% of the coating specification value, the accuracy of GAN network training is improved, and the training distribution of the generator and discriminator is quickly realized, so as to obtain the preferred standard of the CCD detection mechanism model, achieving the detection effect of accurately and quickly identifying qualified products or defective products.

Description

Technical Field

[0001] The present invention relates to a method and system for constructing a CCD detection model for metal coatings based on GAN, and belongs to the technical field of metal coating detection of semiconductor electronic components. Background Art

[0002] The deposition of metal coatings on semiconductor electronic components is a unique characterization technology that is sensitive to the local electrochemical-physical environment structure of metal ions during the electroplating process, and has extensive applications in the fields of physics, electrochemistry, and high-end electroplating manufacturing. During the operation of electroplating production equipment, there are numerous and complex factors affecting the deposition of metal coatings under the electrochemical-physical environment conditions of metal ions. Therefore, due to the interference of various influencing factors, the actual deposited metal coatings will have various differences. In the light case, the coating area is smaller than the required range or exceeds the range, or the coating area is misaligned. In the severe case, the coating area is missing plating or even the product is deformed.

[0003] Currently, using CCD detection equipment and technologies provided by market suppliers, it is possible to detect in real time the errors in the coating area range and the errors in product deformation in continuous electroplating production equipment. However, it is difficult to accurately detect the missing plating in the coating area. Therefore, exploring and creating a detection mechanism method for CCD metal coatings with online real-time accuracy has become a very important topic. Summary of the Invention

[0004] In order to improve the CCD detection accuracy of semiconductor metal coatings, the present invention provides a method and system for constructing a CCD detection model for metal coatings based on GAN, and the technical solutions are as follows:

[0005] The present invention provides a method for constructing a CCD detection model for metal coatings, including:

[0006] Step 1: Obtain the size data and coating area of the semiconductor workpiece to be processed;

[0007] Step 2: Obtain the size data set of the coating area of the semiconductor workpiece to be processed and the metal coating information to be electroplated, and calculate the total mass of the workpiece modeling m 建模总质量 , and the calculation method is:

[0008] m 建模总质量 = m 镀件质量 + m 镀层1 + m 镀层2 + ...... + m 镀层n

[0009] Step 3: Randomly extract the coating quality standard conditions, and use the electronic electroplating module to perform actual electroplating operations on the semiconductor workpiece to be processed. After electroplating, use a precision balance to measure the total actual mass of the workpiece. m 实测总质量 ;

[0010] Step 4: Compare the total modeled mass m 建模总质量 of the workpiece with the total actual mass m 实测总质量 of the workpiece. When the following relationship is satisfied:

[0011] m 实测总质量 (100% - 5%) ≤ m 建模总质量 ≤ m 实测总质量 (100% + 5%)

[0012] Then save this set of data for training the GAN neural network;

[0013] If the above relationship is not satisfied, correct the coating quality standard conditions based on the total actual mass m 实测总质量 of the measured workpiece, calculate the corrected coating quality standard conditions, return to Step 3 to re-measure the total mass of the workpiece m 实测总质量 , and re-determine whether the total modeled mass m 建模总质量 of the corrected workpiece meets the threshold range until the total modeled mass m 建模值 of the modeled workpiece meets the above relationship, and then this set of data can be used for training the GAN neural network;

[0014] Step 5: Import the total modeled mass m 建模总质量 of the workpiece and the total actual mass m 实测总质量 into the GAN neural network, and perform learning and adversarial training processing on the GAN game system for the total modeled mass m 建模总质量 of the workpiece and the total actual mass m 实测总质量 of the workpiece. The trained GAN neural network classifies the electronic electroplating products into good products and defective products according to the input total modeled mass m 建模总质量 of the workpiece and the total actual mass m 实测总质量 of the workpiece.

[0015] Optionally, the model operation system of the GAN neural network is composed of sensitive condition parameter modeling, and the sensitive condition parameters include: types of electroplating solutions, temperature of electroplating solutions, specific gravity of electroplating solutions, diameter of pipelines for transporting electroplating solutions, types of electroplating power supplies, current intensity, electroplating time, cathode current efficiency, surface area of plating areas, metal density of electroplating layers, current density, and flow rate of electroplating solutions.

[0016] Optionally, the semiconductor workpieces to be processed include: precision terminal types, lead frame types, wafer chip types, and precision ornaments.

[0017] Optionally, the metal plating layers include:

[0018] Monomer plating layers of Au, Ag, Ni, Sn, Cu, Pd, Rh, and Pt;

[0019] Alloy plating layers of Pd-Ni, Ni-P, W-Ni, Ag-Sn, Rh-Ru, and Pt-Rh.

[0020] The present invention provides a construction system for a CCD detection model of a metal plating layer, including:

[0021] A memory 510 for storing data;

[0022] A CCD detection module 320 for obtaining CCD morphology parameters of semiconductor workpieces to be processed;

[0023] An electroplating module 310 for randomly extracting coating quality standard conditions and performing actual electroplating operations on the semiconductor workpieces to be processed;

[0024] A workpiece + coating quality calculation module 330 for calculating the total mass of workpiece modeling m 建模总质量 , and the calculation method is:

[0025] m 建模总质量 = m 镀件质量 + m 镀层1 + m 镀层2 + ...... + m 镀层n

[0026] A total mass threshold discrimination module for the workpiece for comparing the total mass of workpiece modeling m 建模总质量 and the actually measured total mass of the workpiece m 实测总质量 When the following relationship is satisfied between the two:

[0027] m 实测总质量 (100% - 5%) ≤ m 建模总质量 ≤ m 实测总质量 (100% + 5%)

[0028] Then save this set of data for training the GAN neural network;

[0029] If the above relationship is not satisfied, correct the plating quality standard conditions according to the measured total mass of the plated parts m 实测总质量 and calculate the corrected plating quality standard conditions, then return to step 3 to re-measure the total mass of the plated parts m 实测总质量 , and re-determine the total mass of the modeled plated parts after correction m 建模总质量 to see if it meets the threshold range until the total mass of the modeled plated parts m 建模值 meets the above relationship, then this set of data can be used for training the GAN neural network;

[0030] The GAN neural network generation module 200, based on the total mass of the modeled plated parts m 建模总质量 and the measured total mass of the plated parts m 实测总质量 performs learning and adversarial training processing on the GAN game system. The trained GAN neural network divides the electroplated products into qualified products and defective products according to the input total mass of the modeled plated parts m 建模总质量 and the measured total mass of the plated parts m 实测总质量 If applicable, the GAN neural network generation module 200 includes:

[0031] The training module 210, which is used to train the adversarial training of the total mass of the plated parts and the measured total mass of the plated parts;

[0032] The stability generation module 220, which is used to process the ordered arrangement and trend of the measured total mass;

[0033] The plating potential operation module 230, which is used to perform plating simulation operations based on the measured total mass;

[0034] The model morphology discrimination module 240, which is used to discriminate based on the simulated plating result and the modeled morphology of the plated parts.

[0035]

[0036] ​The present invention provides a CCD detection method for semiconductor metal coatings based on a GAN neural network. The method uses the CCD detection model constructed by the method described in any one of the above to detect the metal coatings of semiconductor workpieces, including:

[0037] Step 1: Obtain the actual mass of the semiconductor workpiece to be detected using a precision balance m 实测值 ;

[0038] Step 2: Obtain the CCD morphology parameters of the semiconductor workpiece to be detected;

[0039] Step 3: Calculate the simulated total mass of the semiconductor workpiece based on the CCD morphology parameters M CCD , and the calculation method is:

[0040] M CCD = m 镀件质量 + m 镀层1 + m 镀层2 + ...... + m 镀层n

[0041] m 镀件质量 = ρ v

[0042] m 镀层1 = ρ 1 v 1

[0043] m 镀层2 = ρ 2 v 2 ......

[0044] m 镀层n = ρ n v n

[0045] Wherein, ρ is the metal density of the semiconductor workpiece; v is the metal volume of the semiconductor workpiece; ρ 1 is the metal density of the first coating of the semiconductor workpiece; v 1 is the volume of the first coating; ρ 2 is the metal density of the second coating of the semiconductor workpiece; v2 is the volume of the second plating layer; and so on, ρ n is the metal density of the nth plating layer of the semiconductor plating part; v n is the volume of the nth plating layer;

[0046] Step 4: For the actual mass of the semiconductor plating part m 实测值 and the simulated total mass value M CCD make a judgment:

[0047] m 实测值 (100% - 5%) ≤ M CCD ≤ m 实测值 (100% + 5%)

[0048] If the above relationship is not satisfied, it is determined that the semiconductor plating part is a defective product; if the above relationship is satisfied, the next judgment of the GAN neural network is carried out;

[0049] Step 5: Input the actual mass of the semiconductor plating part m 实测值 and the simulated total mass value M CCD into the trained GAN neural network. The results of the adversarial processing of the GAN neural network are divided into two categories:

[0050] Good product: The processing result of the image data of the electroplated electronic product meets the morphology threshold range;

[0051] Defective product: The processing result of the image data of the electroplated electronic product does not meet the morphology threshold range.

[0052] The beneficial effects of the present invention are:

[0053] The present invention improves and constructs a GAN neural network operation model through a deep learning simulation model for training. The GAN network constructed outputs the processing result of the image data that meets the threshold range as a "good product"; while the processing result of the image data that exceeds the threshold range is output as a "defective product", and strictly manages the high - end quality requirements of the electroplating of semiconductor electronic components according to this standard.

[0054] In addition, a total mass threshold discrimination module for plating parts is used to simulate and monitor the appearance, shape, color and overall morphology characteristics of semiconductor devices after electroplating. Further, by meeting the standard with a minimum threshold of the plating specification value - 5% and a maximum threshold of the plating specification value + 5%, the accuracy of the GAN network training is improved, and the training distribution of the generator and discriminator is quickly realized, so as to obtain the optimal standard of the CCD detection mechanism model, achieving the detection effect of accurately and quickly identifying good products or defective products.

[0055] Based on the detection results of the present invention, in actual engineering, the optimal CCD detection mechanism for various semiconductor devices with different structures can be quickly established for the electroplating production conditions of electroplated semiconductor workpieces, and then the optimal local electroplating control method for semiconductor workpieces can be obtained. It can not only achieve the accuracy of predicting the local coating thickness of semiconductor devices, greatly reduce the error between the coating thickness of the actual electroplated product and the predicted value, but also further improve the electroplating device, promote the improvement of the uniformity of the local coating thickness of semiconductor devices, and meet the requirements of high-quality semiconductor devices. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0057] Figure 1 It is a schematic diagram of the structure of the metal coating CCD detection model based on GAN of the present invention.

[0058] Figure 2 It is a flowchart of the construction method and system of the metal coating CCD detection model based on GAN of the present invention.

[0059] Figure 3 It is a schematic diagram of the experimental wafer of the semiconductor workpiece to be processed in the third embodiment of the present invention.

[0060] Figure 4 It is a schematic diagram of the copper plating area of each unit in the experimental wafer of the semiconductor workpiece to be processed in the third embodiment of the present invention.

[0061] Figure 5 It is a schematic diagram of the silver plating area of each unit in the experimental wafer of the semiconductor workpiece to be processed in the third embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0062] To make the objectives, technical solutions, and advantages of the present invention clearer, the following will further describe the embodiments of the present invention in detail with reference to the drawings.

[0063] Embodiment 1:

[0064] This embodiment provides a method for constructing a metal coating CCD detection model. The detection model is based on the Generative Adversarial Network (GAN). The construction process mainly includes two parts, namely, the construction of the data set and the training of the GAN network.

[0065] First, the construction of the dataset is based on the semiconductor workpieces to be processed. By collecting data such as the dimensions and plating areas of these semiconductor workpieces, the specific steps are as follows:

[0066] Step 1: Obtain the dimensional data and plating areas of the semiconductor workpieces to be processed;

[0067] Step 2: Obtain the dimensional dataset of the plating areas of the semiconductor workpieces to be processed and the information on the metal plating layers to be electroplated, and calculate the total mass of the workpiece modeling m 建模总质量 , and the calculation method is:

[0068] m 建模总质量 = m 镀件质量 + m 镀层1 + m 镀层2 + ...... + m 镀层n (1)

[0069] m 镀件质量 = ρ v

[0070] m 镀层1 = ρ 1 v 1

[0071] m 镀层2 = ρ 2 v 2 ......

[0072] m 镀层n = ρ n v n

[0073] Among them, ρ is the metal density of the semiconductor workpiece; v is the metal volume of the semiconductor workpiece; ρ 1 is the metal density of the first plating layer of the semiconductor workpiece; v 1 is the volume of the first plating layer; ρ 2 is the metal density of the second plating layer of the semiconductor workpiece; v 2 is the volume of the second plating layer; and so on, ρ n is the metal density of the nth plating layer of the semiconductor workpiece; vn is the volume of the nth coating.

[0074] Step 3: Randomly select the coating quality standard conditions, and use the electronic electroplating module to perform actual electroplating operations on the semiconductor workpiece to be processed. After electroplating, use an ultra-precision analytical balance to measure the total measured mass of the workpiece. m 实测总质量 ;

[0075] Step 4: For the total modeled mass of the workpiece m 建模总质量 and the total measured mass of the workpiece m 实测总质量 are compared. When the two satisfy the following relationship:

[0076] m 实测总质量 (100% - 5%) ≤ m 建模总质量 ≤ m 实测总质量 (100% + 5%) (2)

[0077] Then save this set of data for training the GAN neural network.

[0078] If the above relationship is not satisfied, this set of data cannot be used for training the GAN neural network, and the coating quality standard conditions are corrected based on the total measured mass of the workpiece m 实测总质量 and calculate the corrected coating quality standard conditions, return to Step 3 to re-measure the total mass of the workpiece m 实测总质量 , and re-determine whether the total modeled mass of the corrected workpiece m 建模总质量 meets the threshold range until the total modeled mass of the workpiece m 建模值 meets the above relationship, and this set of data can be used for training the GAN neural network.

[0079] After completing the construction of the dataset, use the dataset to train the GAN neural network.

[0080] Step 5: Import the total modeled mass of the workpiece m 建模总质量 and the total measured mass of the workpiece m 实测总质量 into the GAN neural network, and perform learning and adversarial training processing on the GAN game system for the total modeled mass of the workpiece m 建模总质量 and the total measured mass of the workpiece m 实测总质量 The trained GAN neural network will, according to the input total modeled mass of the workpiece m 建模总质量And the measured total mass of the electroplated parts m 实测总质量 The electronic electroplating products are divided into qualified products and defective products.

[0081] The GAN neural network model operation system is composed of the modeling of sensitive condition parameters. The sensitive condition parameters include the type of electroplating solution; the temperature of the electroplating solution; the specific gravity of the electroplating solution; the diameter of the pipeline for transporting the electroplating solution; the type of electroplating power supply; the current intensity; the electroplating time; the cathode current efficiency; the surface area of the plating area; the metal density of the electroplating layer; the current density; the flow rate of the electroplating solution.

[0082] In this embodiment, the Nash equilibrium principle is applied, and the operation processing of the total mass data set of the modeling semiconductor and the CCD detection visual images and data are represented by differentiable functions D and G represent the discriminator and the generator respectively; the generator deep-learns the real data and distribution, while the discriminator correctly discriminates whether the deep-learning operation result can be used in the GAN adversarial network model according to the threshold standard, and constructs an accurate CCD detection method for the metal plating layer.

[0083] The basic principle of the metal plating layer CCD detection mechanism based on the deep learning simulation model in this embodiment is that any differentiable function can be used to represent the generator and discriminator of the GAN neural network, and its differentiable function D and G The input of is the real data x and the random variable z . G ( z ) is generated by G Samples that satisfy the minimum threshold and maximum threshold range and approach the real data distribution. If the input of the discriminator comes from the real data detected by the CCD, it is labeled as 1; if the input is the P data of the random modeling variable z then it is labeled as 0. Here G ( z ) the goal of is to achieve a two-class discrimination of the data source, that is, the distribution of the real data D that can be used for modeling or the non-modeling data x that fails to meet the minimum threshold and maximum threshold range and originates from the generator G ( z ) and G The goal of is to make the non-modeling data G ( z ) generated by itself perform D on D ( G ( z )) and the real data that can be used for modelingx The performance on D is consistent. These two processes that oppose and iteratively optimize each other enable D ( x ) to continuously improve their performance. When the discriminative ability of D and G finally improves to a certain extent and it can no longer correctly distinguish the data source, the skill deems that this generator D has learned the distribution of the modeled data. G In the CCD detection system of the metal coating, given the generator

[0084] , consider optimizing the discriminator G D . Similar to the training of a general binary classification model based on Sigmoid, training the discriminator D is also a process of minimizing cross-entropy. The loss function equation (3) is:

[0085] (3)

[0086] wherein, x is sampled from the CCD precise detection data distribution P data ( x ), z is sampled from the modeled random prior distribution P z (z), and E(·) represents the expected value. In actual training, different from the binary classification model, the training data set of the discriminator comes from the CCD image data set distribution P data ( x ), marked as 1, and the modeled data distribution of the generator P g ( x ), marked as 0. Given the generator G , it is necessary to obtain the minimum solution of the minimization equation (4) in the continuous space.

[0087] (4)

[0088] In this embodiment, for any non-zero real numbers m and n , and the real value y ∈[0, 1], the expression - m log( y ) - n log(1 - y) It obtains the minimum value at m / (m + n). Therefore, under the condition of a given generator, the objective function obtains the minimum value at Equation (5), which is the optimal solution of the discriminator. As can be seen from Equation (4), the constructed metal plating model GAN in this embodiment evaluates the ratio of two probability distribution densities.

[0089] (5)

[0090] Furthermore, D ( x )represents x the probability of image data that meets the threshold range or generated data that exceeds the threshold range from the CCD detection device. When the input data is sampled from the image data of the CCD detection device that meets the threshold range x at this time, D the goal is to make the output probability value D ( x )approach 1, while when the input comes from the generated data that exceeds the threshold range from the generator G ( z )at this time, D the goal is to correctly judge the data source, making D (G(z))approach 0, and at the same time G the goal is to make it approach 1.

[0091] The CAN of the metal plating CCD detection mechanism based on the deep learning simulation model in this embodiment is a game system about G and D The loss function of the generator G is Obj G ( θ G ) = -Obj D ( θ D， θ G ) Therefore, the optimized GAN is a minimax problem, and the objective function as shown in Equation (6) is obtained.

[0092] (6)

[0093] All in all, for the learning process of the CAN game system of the metal plating CCD detection mechanism based on the deep learning simulation model in this embodiment, it is necessary to train the model D to maximize the accuracy rate of distinguishing the distribution of image data that meets the threshold range or generated data that exceeds the threshold range from the CCD detection device G ( z ), and at the same time, it is necessary to train the modeling data model Gto minimize log(1 - D ( G (z))). The game system can adopt the method of alternating optimization: that is, first fix the generator G , optimize the discriminator D , so that D 's discrimination accuracy is maximized; then fix the generator D , optimize the discriminator G , so that D 's discrimination accuracy is maximized. When and only when P data = P g , the global optimum is reached and it becomes the preferred standard for the CCD detection mechanism. When training the GAN, in a unified round of parameter updates, generally update the parameters of D K times, and then update the parameters of G once.

[0094] Example 2:

[0095] This example provides a semiconductor metal plating CCD detection method based on a GAN neural network, and uses the detection model constructed by the method described in Example 1 to classify whether the semiconductor workpiece is a good product or a defective product. The specific steps are as follows:

[0096] Step 1: Use a precision balance to obtain the actual mass of the semiconductor workpiece to be detected m 实测值 ;

[0097] Step 2: Obtain the CCD morphology parameters of the semiconductor workpiece to be detected;

[0098] Step 3: Calculate the simulated total mass of the semiconductor workpiece based on the CCD morphology parameters M CCD , and the calculation method is:

[0099] M CCD = m 镀件质量 + m 镀层1 + m 镀层2 + ...... + m 镀层n

[0100] m 镀件质量 = ρ v

[0101] m 镀层1= ρ 1 v 1

[0102] m 镀层2 = ρ 2 v 2 ......

[0103] m 镀层n = ρ n v n

[0104] Among them, ρ is the metal density of the semiconductor plating; v is the metal volume of the semiconductor plating; ρ 1 is the metal density of the first layer of the semiconductor plating; v 1 is the volume of the first layer; ρ 2 is the metal density of the second layer of the semiconductor plating; v 2 is the volume of the second layer; and so on, ρ <![CDATA[ n ]]is the metal density of the nth layer of the semiconductor plating; v <![CDATA[ n ]]is the volume of the nth layer.

[0105] Step 4: Judge the actual mass m <![CDATA[ 实测值 ]]of the semiconductor plating and the simulated total mass value M <![CDATA[ CCD ]]:

[0106] m <![CDATA[ 实测值 ]](100% - 5%) ≤ M <![CDATA[ CCD ]]≤ m <![CDATA[ 实测值 ]](100% + 5%)

[0107] If the above relationship is not satisfied, it is judged that the semiconductor plating is a defective product; if the above relationship is satisfied, the next judgment of the GAN neural network is carried out;

[0108] Step 5: Input the actual mass m <![CDATA[ 实测值 ]]of the semiconductor plating and the simulated total mass value M <![CDATA[ CCD ]]into the GAN neural network trained in Example 1. The results of the adversarial processing of the GAN neural network are divided into two categories:

[0109] Good product; the image data processing result of the electronic electroplating product meets the morphological threshold range;

[0110] Missing product: The image data processing result of the electroplated electronic product does not meet the morphological threshold range.

[0111] In this embodiment, through the constructed GAN confrontation network, accurate detection of qualified products and missing products of electroplated electronic products is realized.

[0112] Embodiment 3:

[0113] In this embodiment, in combination with the actual semiconductor plating parts to be processed, the construction process of the detection model data set of the present invention is described in detail, which specifically includes the following contents.

[0114] The experimental wafer of the semiconductor plating part to be processed in this embodiment is as Figure 3 shown, made of copper alloy material with a size of 60×60mm and a thickness of 0.127mm; there are 16 basic units in total. The size of the copper plating area 10 of each unit is 11.5mm×11.5mm, and its area is 132.25mm 2 ; then the total single-sided area of a semiconductor copper plating is 16×132.25 = 2166mm 2 ; in addition, the size of the silver plating area 20 of each unit is 7.3mm×7.3mm, and its area is 53.29mm 2 ; then the total single-sided area of a semiconductor silver plating is 16×53.29 = 852.64mm 2 .

[0115] The electroplating specifications of the plating area are: the thickness of the Cu coating is 1700nm, and the thickness of the Ag coating is 5000nm.

[0116] The CCD detection center position of the plating area of the semiconductor plating part in this embodiment is the center black dot 21 of the electroplating area of each semiconductor plating part unit as shown in Figure 5 , and its coordinates are: 4(14.6, 7.3), 3(14.6, 21.9), 2(14.6, 36.5), 1(14.6, 51.1) in column I; 4(21.9, 7.3), 3(21.9, 21.9), 2(21.9, 36.5), 1(21.9, 51.1) in column II; 4(36.5, 7.3), 4(36.5, 21.9), 2(36.5, 36.5), 1(36.5, 51.1) in column III, and 4(51.1, 7.3), 3(51.1, 21.9), 2(51.1, 36.5), 1(51.1, 51.1) in column IV.

[0117] As Figure 3 shown, the quality of the plating part before electroplating of a semiconductor plating part m 镀件质量 is further calculated and derived from Formula 2.

[0118] m 实测总质量 = m 镀件质量 + m 镀层1 + m 镀层2 + ...... + m 镀层n (Formula 2)

[0119] Based on the 3D design drawing of the semiconductor plating part as Figure 3 shown, the volume of a semiconductor plating part is v mm 3 , and the specific gravity of the copper alloy used is ρ mg / mm 3 , so its total mass is:

[0120] m 镀件质量 = ρ v

[0121] According to the single-sided copper plating surface area of the above-mentioned semiconductor plating part s is 2166 mm 2 , the thickness of the Cu plating layer h 铜1 is 1700 nm, the single-sided silver plating surface area of a semiconductor plating part s is 852.64 mm 2 , the Ag plating layer h 银2 is 5000 nm, so its total mass is:

[0122] m 镀层总质量 = m 铜镀层1 + m 银镀层2

[0123] = ρ 铜1 v 铜1 + ρ 银2 v 银2

[0124] = ρ 铜1 s 铜1 h 铜1 + ρ 银2 s 银2h 银2

[0125] Step 1: The quality of the semiconductor plating workpiece before electroplating m 镀件质量 After alkali degreasing, acid activation and drying by the electroplating module 310, the quality of the semiconductor plating workpiece before electroplating after pretreatment is tested using an ultra-precision analytical balance m 镀件质量 , and the test results are shown in Table 1

[0126] Table 1 Quality of the semiconductor plating workpiece before electroplating after pretreatment

[0127]

[0128] As can be seen from Table 1, the quality range of the samples 1 - 8 before electroplating is between 2.6757 g and 2.6793 g. Taking the average value of 2.6773 g as the standard, the minimum error is -0.06% and the maximum error is 0.08%. From the perspective of the quality control of the semiconductor plating workpiece before electroplating alone, compared with the overall quality control threshold of the semiconductor plating workpiece of -5.0% to 5.0%, the modeling conditions are met

[0129] Step 2: Electroplating copper treatment is carried out one by one using a dedicated electroplating device

[0130] As is well known in the field of electroplating technology, although under conventional electroplating conditions, a constant current is set for electroplating process treatment, due to the real-time changes of its various components in the electroplating solution, different changes in the surface environment of the plating workpiece will occur during the actual electroplating process, resulting in that the applied constant current cannot be 100% effectively used for the electrolytic deposition of copper. Therefore, the current efficiency is usually used to represent the effective current percentage of electroplating. Its characteristic is that during the electroplating process, the current efficiency fluctuates within a certain range with the change of the electroplating environment. When the electroplating process is completed, after cleaning the semiconductor plating workpiece sample, its quality is tested using an ultra-precision analytical balance, that is, the quality of the semiconductor plating workpiece after copper plating after the above pretreatment m 镀件质量 + m 铜镀层质量 Denoted by mixture 1, the quality of the copper coating m 铜镀层质量 Using the quality of mixture 1 m 镀件质量 + m 铜镀层质量 Subtracting m 镀件质量 , and the results are shown in Table 2

[0131] Table 2 Quality of the copper coating

[0132]

[0133] As can be seen from Table 2, after electroplating copper on Samples 1 to 8, the mass range of the copper plating layer is between 0.0348 and 0.0377 grams. Taking the average value of 0.0360 grams as the standard, its minimum error is -3.4% and the maximum error is 4.6%. From the perspective of the quality control of the semiconductor plating parts after electroplating copper alone, compared with the overall quality control threshold of -5.0% to 5.0% of the semiconductor plating parts, it meets the modeling conditions.

[0134] Step 3: For the copper-plated samples 1 to 8 whose masses have been measured in Table 2, silver plating treatment is carried out one by one using a special electroplating device. After cleaning, their masses are measured using an ultra-precision analytical balance, that is, the mass of the semiconductor plating parts after final silver plating m 镀件质量 + m （铜+银）镀层质量 Denoted by mixture 2, the mass of the silver plating layer m 银镀层质量 With the mass of mixture 2 m 镀件质量 + m （铜+银）镀层质量 Subtract the mass of mixture 1 m 镀件质量 + m 铜镀层质量 , and the results are shown in Table 3.

[0135] Table 3 Mass of the silver plating layer

[0136]

[0137] As can be seen from Table 3, after electroplating silver on Samples 1 to 8, the mass range of the silver plating is between 0.0477 and 0.0514 grams. Taking the average value of 0.0494 grams as the standard, its minimum error is -3.3% and the maximum error is 4.2%. From the perspective of the quality control of the semiconductor plating parts after electroplating silver alone, compared with the overall quality control threshold of -5.0% to 5.0% of the semiconductor plating parts, it meets the modeling conditions.

[0138] Step 4: After the implementation of Step 1 and before the implementation of Step 2, use the CCD detection mechanism to perform morphology tests on Samples 1 to 8 before electroplating.

[0139] For the plating part samples, a dedicated CCD device is used to detect their overall morphology. After the obtained data is processed by the plating part + plating layer mass calculation module 330 and judged as Yes qualified by the plating part total mass threshold discrimination module 340, the test data can be imported into the GAN neural network generation module 200 for in-depth learning. Otherwise, in the case of No, it will be judged as a defective product and cannot enter the next implementation operation.

[0140] The deep learning model includes: a training module 210, a stability generation module 220, a plating potential calculation module 230, and a model morphology discrimination module 240.

[0141] The training module 210 is used to perform adversarial training on the total mass of the plated part and the actually measured total mass of the plated part to obtain a model that can be used for classifying good products and defective products.

[0142] The stability generation module 220 is used to process the ordered arrangement and trend of the actually measured total mass;

[0143] The plating potential calculation module 230 is used to perform plating simulation calculations based on the actually measured total mass;

[0144] The model morphology discrimination module 240 is used to discriminate based on the plating result of the simulation calculation and the modeled morphology of the plated part.

[0145] If the model discrimination result is Yes, it is a good product, and its corresponding data is backed up in the memory 510. If the discrimination is No, it is a defective product, and the measured image data cannot be backed up and stored.

[0146] Quality of semiconductor plated parts before electroplating by deep learning model morphology discrimination based on CCD device detected image data M 镀件质量 , and the results are shown in Table 4.

[0147] Table 4 Quality of semiconductor plated parts detected by CCD

[0148]

[0149] As can be seen from Table 4, the quality range of the electronic electroplated copper samples 1 to 8 before electroplating detected by the CCD device is between 2.6759 and 2.6785 grams. Taking the average value of 2.6770 grams as the standard, its minimum error is -0.04% and the maximum error is 0.06%. From the perspective of only the quality control of the semiconductor plated parts before electroplating, compared with the overall quality control threshold of the semiconductor plated parts of -5.0% to 5.0%, it meets the modeling conditions.

[0150] Step 5: After Step 2 and before Step 3, use the CCD detection mechanism to perform morphology tests on the electronic electroplated copper samples 1 to 8; the test uses a dedicated CCD device to detect its overall morphology. After the data obtained is judged to be qualified by the total mass threshold discrimination module 340 of the plated part, the test data is imported into the GAN neural network generation module 200 for deep learning, and the content of the deep learning is similar to that in Step 4.

[0151] Quality of semiconductor plated parts electroplated with copper by deep learning model morphology discrimination based on CCD device detected image data M 镀件质量 + M铜镀层质量 Represented by mixture 3, the mass of the copper plating M 铜镀层质量 With the mass of mixture 3 M 镀件质量 + M 铜镀层质量 Subtract M 镀件质量 , and the results are shown in Table 5.

[0152] Table 5 Mass of copper plating on semiconductor workpieces detected by CCD

[0153]

[0154] As can be seen from Table 5, based on the CCD device detection, the mass range of the electroplated copper coatings of Samples 1 to 8 is between 0.0343 and 0.0363 grams. Taking the average value of 0.0351 grams as the standard, the minimum error is -2.2% and the maximum error is 3.5%. Simply from the perspective of the quality control of the semiconductor workpieces before electroplating, compared with the overall quality control threshold of the semiconductor workpieces of -5.0% to 5.0%, therefore, these measured data meet the modeling conditions.

[0155] Furthermore, it can be observed that for the same test samples 1 to 8, the mass of the copper plating on the semiconductor workpieces obtained by the metal plating CCD detection method based on the deep learning simulation model of this embodiment M (g) The error result range is -2.2% to 3.5%, which is better than the measured mass of the semiconductor copper plating obtained in Step 1 m (g) The error result range is -3.4% to 4.6%; this result exactly confirms that the GAN neural network adversarial operation model of the present invention can further improve and optimize the measured quality of the semiconductor copper plating m (g) The accuracy of the result.

[0156] Step 6: For the silver-plated samples 1 to 8 after Step 3, use the CCD detection mechanism to perform a morphology test on the electroplated silver samples 1 to 8.

[0157] Use a dedicated CCD device to detect its overall morphology. After the obtained data is judged to be qualified by the workpiece total mass threshold discrimination module 340, the test data is imported into the GAN neural network generation module 200 for deep learning. The content of the deep learning is similar to that in Step 4 and will not be elaborated here.

[0158] The quality of the electroplated silver on the semiconductor workpiece discriminated by the deep learning model of the morphology based on the CCD device detection image data M 镀件质量 + M （铜+银）镀层质量 Represented by mixture 4, the mass of the silver plating M 银镀层质量Using a mixed mass of 4 M 镀件质量 + M （铜+银）镀层质量 Subtracting the mixed mass of 3 M 镀件质量 + M 铜镀层质量 The results are shown in Table 6 below.

[0159] Table 6 Quality of the silver coating on semiconductor plated parts detected by CCD

[0160]

[0161] As can be seen from Table 6, after electroplating silver on samples 1 to 8, the mass range is between 0.0465 and 0.0490 grams. Taking the average value of 0.0475 grams as the standard, the minimum error is -2.1% and the maximum error is 3.2%. From the perspective of the mass control of semiconductor plated parts after electroplating copper alone, compared with the overall mass control threshold of -5.0% to 5.0% for semiconductor plated parts, it meets the modeling conditions.

[0162] Furthermore, it can be observed that for the same test samples 1 to 8, the quality of the silver coating on semiconductor plated parts obtained by the metal coating CCD detection method based on the deep learning simulation model of the present invention M (g) The error result range is -2.1% to 3.2%, which is better than the measured quality of the semiconductor silver coating obtained in step 3 m (g) The error result range is -3.3% to 4.2%; this result further and more precisely confirms that the GAN neural network adversarial operation model of the present invention can effectively improve the accuracy of the measured quality of the semiconductor silver coating m (g) of the result.

[0163] Comparative example

[0164] Based on the use of CCD detection equipment and technology provided by market suppliers, it can only detect the error in the coating area range and the lack of product deformation in real time; it is difficult to accurately detect the lack of missing plating in the coating area.

[0165] At present, the existing CCD detection technology can control the plating area by setting the electroplating area data model, but it can only be used when there is a significant difference in the color of the electroplated deposit layer from the raw material or the bottom plating metal layer. In addition, for the deformation of the plated part, by comparing with the standard image model, it can be used to judge various deformations of the plated part. First, the shape and size test of a single semiconductor plated part is shown in Table 7 below.

[0166] Table 7 Shape and size of a single semiconductor plated part

[0167]

[0168] As can be seen from Table 7, the test result range of the shape and size of the electroplating samples 1 to 8 before pretreatment cleaning is 59.95 to 60.06 mm. Compared with the product design standard value of 60 mm, the average value of 60.005 mm has an average error of 0.01%, a minimum error of -0.08%, and a maximum error of 0.10%. Therefore, the existing CCD can utilize the feature of accurate plating part size testing to control product size abnormalities caused by deformation.

[0169] Secondly, the existing CCD measures the size of local copper plating areas as shown in Table 8.

[0170] Table 8 Size testing of local copper plating areas by the existing CCD

[0171]

[0172] As can be seen from Table 8, the test results of the copper plating area sizes of the electroplated copper samples 1 to 10 are in the range of 11.47 to 11.56 mm. Compared with the product design standard value of 11.5 mm for the copper plating area, the average value of 11.506 mm has an average error of 0.05%, a minimum error of -0.26%, and a maximum error of 0.52%. Therefore, the existing CCD can utilize the feature of accurate plating part size testing to control product size abnormalities caused by deformation. However, since the color boundaries of the copper plating area cannot be accurately identified, it is impossible to accurately determine whether there is missing plating or plating penetration in the coating.

[0173] Furthermore, the test results of the size of local silver plating areas for the electroplated copper samples 1 to 8 are shown in Table 9.

[0174] Table 9 Test results of the size of local silver plating areas detected by CCD

[0175]

[0176] As can be seen from Table 9, the test results of the copper plating area sizes of the electroplated copper samples 1 to 8 are in the range of 7.27 to 7.35 mm. Compared with the product design standard value of 7.3 mm for the copper plating area, the average value of 7.301 mm has an average error of 0.01%, a minimum error of -0.41%, and a maximum error of 0.68%. Therefore, the existing CCD can utilize the feature of accurate plating part size testing to control product size abnormalities caused by deformation. However, since the color boundaries of the hard plating area cannot be accurately identified, it is impossible to accurately determine whether there is missing plating or plating penetration in the silver coating.

[0177] Since the silver-plated layers of Samples 1 to 8 are silver-white and there is no obvious contrast with the color of the underlying nickel plating layer, the current CCD detection technology cannot detect and distinguish metal plating layers with no obvious color contrast. That is, it cannot accurately detect whether there is missing plating or plating penetration outside the plating area in the electroplating region.

[0178] The current CCD detection technology, for semiconductor samples plated with gold for example, can be applied to accurately detect missing plating or plating penetration in the plating area due to the distinct color difference. Therefore, using the same semiconductor plating parts as shown Figure 3 as the starting raw materials, replace the copper plating in the Figure 4 electroplating area with nickel plating, and the nickel plating film thickness is 2500 nm; further, replace the silver plating in the Figure 5 electroplating area with gold plating, and the gold plating film thickness is 300 nm. Based on the nickel plating and gold plating film thicknesses, prepare Samples 9 to 16. The preparation method is the same as that of Sample 1. Use the CCD detection equipment of the comparative example to test Samples 1 to 8 and Samples 9 to 16, and the current CCD inspection results through color differences are shown in Table 10. The tests of Samples 1 to 8 and Samples 9 to 16 are carried out after the silver plating and gold plating are completed.

[0179] Table 10 CCD Inspection Results with Color Differences

[0180]

[0181] As can be seen from Table 10, for the current CCD detection technology, regarding the detection results of the aforementioned silver-white Samples 1 to 8, although it can monitor the deformation of each area of the product, it cannot confirm the silver-white electroplating. Therefore, it is not applicable to the detection and control of such electroplated products; on the other hand, the detection results of the gold-colored Samples 9 to 16 as described show that it can identify and confirm the plating area, that is, both missing plating within the plating area and plating penetration outside the plating area can be qualitatively detected and controlled.

[0182] Table 11 Comparison Results between the Invention and the Existing CCD Detection Method

[0183]

[0184] As can be seen from the comparison results in Table 11, the existing CCD detection device not only does not have the function of quantitatively detecting all electroplated products in terms of electroplating color confirmation, but also does not have the ability to quantitatively analyze and control various metal plating layers by detecting and controlling the quality of the plating parts, that is, the quality of the raw materials of the plating parts before electroplating and the quality of each plating layer.

[0185] Therefore, from the comparison results in Table 11, it is not difficult to see that the existing CCD detection devices are to a large extent not fully applicable to the detection and quality management functions of electronic electroplating products of semiconductor plating parts; for these technical problems that urgently need to be solved in the field of electronic electroplating, through the construction and application of the metal coating CCD detection method based on the deep learning simulation model, the present invention solves the above-mentioned technical problems existing in the existing CCD detection devices, realizes the quality detection of electronic electroplating semiconductor plating parts, and the simulation model of the CCD detection mechanism for GAN operation and its discriminant management provides an innovative detection and control method for the semiconductor electronic electroplating industry.

[0186] Some steps in the embodiments of the present invention can be implemented by software, and the corresponding software program can be stored in a readable storage medium, such as an optical disc or a hard disk, etc.

[0187] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for constructing a CCD detection model for metal coatings, characterized in that, The method includes: Step 1: Obtain the dimensional data and plating areas of the semiconductor workpieces to be processed. Step 2: Obtain the size data set of the coating area of the semiconductor workpiece to be processed and the information of the metal coating to be electroplated, and calculate the total mass of the workpiece modeling. The calculation method is as follows: m 建模总质量 , and the calculation method is: m 建模总质量 = m 镀件质量 + m 镀层1 + m 镀层2 + ...... + m 镀层n Step 3: Randomly extract the plating quality standard conditions, perform actual electroplating operations on the semiconductor workpiece to be processed using the electronic electroplating module, and after electroplating is completed, measure the total actual mass of the workpiece using a precision balance m 实测总质量 ; Step 4: Model the total mass of the plated part m 建模总质量 and the measured total mass of the plated part m 实测总质量 are compared. When the two satisfy the following relationship: m 实测总质量 （100%-5%）≤ m 建模总质量 ≤ m 实测总质量 （100%+5%） Then save the total mass of the modeled plating parts in this group m 建模总质量 and the total measured mass of the plating parts m 实测总质量 The data is used for the training of the GAN neural network; If the above relationship is not satisfied, the coating quality standard conditions are corrected based on the measured total mass of the coated parts m 实测总质量 and the corrected coating quality standard conditions are calculated, then return to step 3 to re-measure the total mass of the coated parts m 实测总质量 , and re-determine the total mass of the modeled coated parts m 建模总质量 to check if it meets the threshold range. This process continues until the total mass of the modeled coated parts m 建模总质量 meets the above relationship. Only then can the total mass of the modeled coated parts m 建模总质量 and the measured total mass of the coated parts m 实测总质量 be used for the training of the GAN neural network; Step 5: Import the total mass of the modeled plating part m 建模总质量 and the actually measured total mass of the plating part m 实测总质量 into the GAN neural network, and perform learning and adversarial training processing on the GAN game system for the total mass of the modeled plating part m 建模总质量 and the actually measured total mass of the plating part m 实测总质量 After the training, the GAN neural network classifies the electroplated products into qualified products and defective products according to the input total mass of the modeled plating part m 建模总质量 and the actually measured total mass of the plating part m 实测总质量 ​ 2. The method for constructing a metal plating CCD detection model according to claim 1, wherein The model operation system of the GAN neural network is composed of sensitive condition parameter modeling. The sensitive condition parameters include: types of electroplating solutions, temperature of electroplating solutions, specific gravity of electroplating solutions, diameters of pipelines for conveying electroplating solutions, types of electroplating power supplies, current intensity, electroplating time, cathode current efficiency, surface area of plating areas, metal density of electroplating layers, current density, and flow rate of electroplating solutions.

3. The method for constructing a metal coating CCD detection model according to claim 1, wherein The semiconductor workpieces to be processed include: precision terminal types, lead frame types, wafer chip types, and precision ornaments.

4. The method for constructing a metal plating CCD detection model according to claim 1, wherein The metal plating layers include: Monomer plating layers of Au, Ag, Ni, Sn, Cu, Pd, Rh, and Pt; Alloy plating layers of Pd-Ni, Ni-P, W-Ni, Ag-Sn, Rh-Ru, and Pt-Rh.

5. A system for constructing a CCD detection model of a metal coating, characterized in that, The system includes: A memory (510) for storing data; A CCD detection module (320) for obtaining the CCD morphology parameters of the semiconductor workpieces to be processed; An electroplating module (310) for randomly extracting the quality standard conditions of the plating layer and performing actual electroplating operations on the semiconductor workpieces to be processed; Plated part + coating mass calculation module (330), used to calculate the total mass of the plated part modeling m 建模总质量 , and the calculation method is as follows: m 建模总质量 = m 镀件质量 + m 镀层1 + m 镀层2 + ...... + m 镀层n The total mass threshold discrimination module of the plated part is used to model the total mass of the plated part m 建模总质量 and the measured total mass of the plated part m 实测总质量 are compared. When the two satisfy the following relationship: m 实测总质量 （100%-5%）≤ m 建模总质量 ≤ m 实测总质量 （100%+5%） Then save the total mass of the modeled group of plated parts m 建模总质量 and the measured total mass of the plated parts m 实测总质量 data for use in the training of the GAN neural network; If the above relationship is not satisfied, the plating quality standard conditions are corrected based on the measured total mass of the plated parts m 实测总质量 and the corrected plating quality standard conditions are calculated, and the process returns to step 3 to re-measure the total mass of the plated parts m 实测总质量 , and the total mass of the modeled plated parts after correction is re-judged m 建模总质量 to check whether it meets the threshold range. This process continues until the total mass of the modeled plated parts m 建模总质量 meets the above relationship. At this time, the total mass of the modeled plated parts in this group m 建模总质量 and the measured total mass of the plated parts m 实测总质量 data can be used for the training of the GAN neural network; GAN neural network generation module (200), based on the total mass of the electroplated part model m 建模总质量 and the measured total mass of the electroplated part m 实测总质量 Carry out learning and adversarial training processing of the GAN game system. The trained GAN neural network divides the electronic electroplated products into qualified products and defective products according to the input total mass of the electroplated part model m 建模总质量 and the measured total mass of the electroplated part m 实测总质量 ​ 6. The construction system of the metal plating CCD detection model according to claim 5, characterized in that, The GAN neural network generation module (200) includes: A training module (210) for training the adversarial training of the total quality of the workpieces and the actually measured total quality of the workpieces; A stability generation module (220) for processing the ordered arrangement and trend of the actually measured total quality; A plating potential operation module (230) for performing plating simulation operations based on the actually measured total quality; A model morphology discrimination module (240) for discriminating between the plating results of the simulation operations and the modeled morphology of the workpieces.

7. A CCD detection method for semiconductor metal coatings based on a GAN neural network, characterized in that, The method uses a CCD detection model constructed by the method described in any one of claims 1-4 to detect the metal plating layers of semiconductor workpieces, including: Step 1: Use a precision balance to obtain the actual mass of the semiconductor workpiece to be detected m 实测值 ; Step 2: Obtain the CCD morphology parameters of the semiconductor workpieces to be detected. Step 3: Calculate the simulated total mass value of the semiconductor plating part based on the CCD topography parameters M CCD , and the calculation method is as follows: M CCD = m 镀件质量 + m 镀层1 + m 镀层2 + ...... + m 镀层n m 镀件质量 = ρ v m 镀层1 = ρ 1 v 1 m 镀层2 = ρ 2 v 2 ...... m 镀层n = ρ n v n Among them, ρ is the metal density of the semiconductor plating; v is the metal volume of the semiconductor plating; ρ 1 is the metal density of the first plating layer of the semiconductor plating; v 1 is the volume of the first plating layer; ρ 2 is the metal density of the second plating layer of the semiconductor plating; v 2 is the volume of the second plating layer; and so on, ρ n is the metal density of the nth plating layer of the semiconductor plating; v n is the volume of the nth plating layer; Step 4: Judge the actual quality m 实测值 and the simulated total mass M CCD as follows: m 实测值 （100%-5%）≤ M CCD ≤ m 实测值 （100%+5%） If the above relationships are not satisfied, it is determined that the semiconductor workpiece is a defective product; if the above relationships are satisfied, the next judgment of the GAN neural network is performed. Step 5: The actual quality of the semiconductor plating part m 实测值 and the total mass simulation value M CCD are input into the trained GAN neural network. The results of the adversarial processing of the GAN neural network are divided into two categories: Qualified products: The image data processing results of the electroplated products meet the morphology threshold range. Defective products: The image data processing results of the electroplated products do not meet the morphology threshold range.

Images