A signal quality prediction method and device based on the generative adversarial network (GAN)

Abstract

The present application discloses a signal quality prediction method and device based on a generative adversarial network (GAN). By inputting a picture containing a PCB trace to be measured and the corresponding trace setting parameters into the generator G of the trained GAN, an evaluation PCB trace picture with a trace signal quality image is generated; the evaluation PCB trace picture is input into the discriminator D of the trained GAN, and according to the loss degree of the evaluation PCB trace picture, it is determined whether the trace signal quality image in the evaluation PCB trace picture is qualified. It realizes the prediction of the signal quality of PCB traces without a simulation software, improves the efficiency of PCB trace performance evaluation, and reduces the cost.

Description

Technical Field

[0001] The present application relates to the technical field of trace signal detection, and particularly to a signal quality prediction method and device based on the generative adversarial network GAN. Background Art

[0002] In the initial stage of circuit board design, during the circuit PCB layout and wiring stage, it is often necessary to evaluate whether the impedance of high-speed traces such as microstrip lines and stripline meets the design requirements, and it is necessary to judge whether the layout of the PCB traces is reasonable according to the signal quality of the PCB traces.

[0003] In the related art, the eye diagram and bit error rate are usually obtained by simulating the PCB traces through simulation software, so as to determine the signal quality of the PCB traces. However, the simulation modeling has a high operation threshold, and the modeling quality will directly affect the accuracy of the simulation results. Sometimes the time consumed by the simulation is as high as dozens of hours, resulting in low efficiency in obtaining the signal quality of the PCB traces, and the simulation effect is unstable. In addition, the simulation software used for simulation is expensive, resulting in high costs for obtaining the signal quality of the PCB traces.

[0004] Therefore, how to improve the acquisition efficiency of the signal quality of PCB traces and reduce costs is a technical problem that needs to be solved urgently. Summary of the Invention

[0005] The main purpose of the present application is to provide a signal quality prediction method based on the generative adversarial network GAN, aiming to solve the technical problems of low acquisition efficiency and high cost in obtaining the signal quality of PCB traces through simulation.

[0006] In a first aspect, the present application provides a signal quality prediction method based on the generative adversarial network GAN, and the method includes the following steps:

[0007] Input the image containing the PCB traces to be measured and the corresponding trace setting parameters into the generator G of the trained generative adversarial network GAN to generate an evaluation PCB trace image with a trace signal quality image;

[0008] Input the evaluation PCB trace image into the discriminator D of the trained GAN, and determine whether the trace signal quality image in the evaluation PCB trace image is qualified according to the loss degree of the evaluation PCB trace image.

[0009] In some embodiments, the method further includes:

[0010] Obtain the image containing the PCB traces to be measured and the corresponding trace setting parameters, which includes:

[0011] Obtain the picture containing the PCB trace to be measured by taking a screenshot from the electronic design automation (EDA) software, and obtain the corresponding trace setting parameters;

[0012] Clear the traces in the picture containing the PCB trace to be measured that are at a distance greater than a preset multiple of the line pitch from the PCB trace to be measured;

[0013] Set the trace setting parameters as label data, and establish a mapping relationship between the label data and the picture containing the PCB trace to be measured.

[0014] In some embodiments, after obtaining the picture containing the PCB trace to be measured and the corresponding trace setting parameters, it further includes:

[0015] Establish a 3D trace model according to the trace setting parameters;

[0016] Perform three-dimensional verification on the picture containing the PCB trace to be measured according to the 3D trace model.

[0017] In some embodiments, when inputting the evaluation PCB trace picture into the discriminator D of the trained GAN, and determining whether the trace signal quality image in the evaluation PCB trace picture is qualified according to the loss degree of the evaluation PCB trace picture, it includes:

[0018] Input the evaluation PCB trace picture into the trained discriminator D, and determine whether the loss of the evaluation PCB trace picture is less than a preset loss threshold through the trained discriminator D;

[0019] If so, determine the signal quality of the PCB trace to be measured according to the trace signal quality image in the evaluation PCB trace picture;

[0020] Otherwise, segment the PCB trace to be measured, input the picture of each corresponding segment of the PCB trace to be measured into the trained generator G respectively, and determine the signal quality of each segment of the PCB trace to be measured according to the trace signal quality image in the evaluation PCB trace picture generated by the trained generator G.

[0021] In some embodiments, segmenting the PCB trace to be measured includes:

[0022] Perform segmentation at positions where the local impedance in the PCB trace to be measured is continuous and the trace bending angle is less than a preset angle.

[0023] In some embodiments, before inputting the picture containing the PCB trace to be measured and the corresponding trace setting parameters into the generator G of the trained generative adversarial network (GAN), it further includes:

[0024] Fix the parameters of the original generator G in the established original GAN, input random noise into the original generator G, and obtain the sample pictures generated by the original generator G;

[0025] Input the sample pictures and the training data in the preset training data set into the original discriminator D of the original GAN, and optimize the original discriminator D with the maximization of the discriminator D objective function as the goal according to the discrimination result of the original discriminator D;

[0026] Fix the parameters of the optimized discriminator D, and optimize the original generator G with the minimization of the generator G objective function as the goal;

[0027] Loop to optimize the generator G and the discriminator D until the sample pictures generated by the generator G according to the random noise are input into the discriminator D, and when the discriminator D determines that the loss of the generated sample pictures is less than the preset loss threshold, obtain the trained GAN.

[0028] In some embodiments, the training data set includes multiple verified pictures containing PCB traces, corresponding trace signal quality images, and corresponding trace setting parameters of the PCB traces.

[0029] In some embodiments, the method further includes:

[0030] Verify the evaluation PCB trace pictures and the corresponding trace signal quality images, and put the evaluation PCB trace pictures, the corresponding trace signal quality images, and the corresponding trace setting parameters into the training data set after passing the verification.

[0031] In some embodiments, if the PCB trace to be measured is a microstrip line or a strip line, the trace setting parameters include adjacent layer stack structure, adjacent layer dielectric thickness, conductivity, line width, trace copper thickness, board dielectric constant, and signal rate;

[0032] If the PCB trace to be measured is a via, the trace setting parameters include the via aperture, the outer diameter of the pad, the length of the via, the dielectric, and the dielectric constant.

[0033] In a second aspect, the present application further provides a signal quality prediction device based on a generative adversarial network GAN. The device includes:

[0034] A generation module, which is used to input pictures containing PCB traces to be measured and corresponding trace setting parameters into the generator G of the trained generative adversarial network GAN to generate evaluation PCB trace pictures with trace signal quality images;

[0035] A determination module, which is configured to input the evaluated PCB trace picture into the discriminator D of the trained GAN, and determine whether the trace signal quality image in the evaluated PCB trace picture is qualified according to the loss degree of the evaluated PCB trace picture.

[0036] This application provides a signal quality prediction method and device based on a generative adversarial network (GAN). By inputting a picture containing a PCB trace to be measured and the corresponding trace setting parameters into the generator G of the trained GAN, an evaluated PCB trace picture with a trace signal quality image is generated; the evaluated PCB trace picture is input into the discriminator D of the trained GAN, and according to the loss degree of the evaluated PCB trace picture, it is determined whether the trace signal quality image in the evaluated PCB trace picture is qualified. It realizes the prediction of the signal quality of PCB traces without a simulation software, improves the efficiency of PCB trace performance evaluation, and reduces the cost. Description of the Drawings

[0037] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings required for the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are some embodiments of this application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0038] Figure 1 It is a schematic flowchart of a signal quality prediction method based on a generative adversarial network (GAN) provided by an embodiment of this application;

[0039] Figure 2 It is a schematic flowchart of GAN training;

[0040] Figure 3 It is a picture of a PCB trace to be measured;

[0041] Figure 4 It is a schematic diagram of a 3D trace model;

[0042] Figure 5 It is an evaluated PCB trace picture with a trace signal quality image;

[0043] Figure 6 It is a schematic diagram of the segmented PCB trace to be measured;

[0044] Figure 7 It is a schematic flowchart of the specific process of the signal quality prediction method based on a generative adversarial network (GAN);

[0045] Figure 8 It is a picture of a via PCB;

[0046] Figure 9 Schematic diagram of a via 3D model.

[0047] Figure 10 PCB trace picture of a via with a trace signal quality image

[0048] Figure 11 Schematic block diagram of a signal quality prediction device based on a generative adversarial network (GAN) provided by an embodiment of the present application.

[0049] The realization of the purpose of the present application, functional features and advantages will be further described with reference to the embodiments and the accompanying drawings. Specific embodiments

[0050] Next, the technical solutions in the embodiments of the present application will be clearly and completely described with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are some, but not all, of the embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts shall fall within the protection scope of the present application.

[0051] The flowcharts shown in the accompanying drawings are only illustrative, and do not necessarily include all the content and operations / steps, nor do they necessarily need to be executed in the described order. For example, some operations / steps can be decomposed, combined or partially merged, so the actual execution order may be changed according to the actual situation.

[0052] An embodiment of the present application provides a signal quality prediction method and device based on a generative adversarial network (GAN). Among them, the signal quality prediction method based on the generative adversarial network (GAN) can be applied to a computer device, and the computer device can be an electronic device such as a laptop computer or a desktop computer.

[0053] Next, some embodiments of the present application will be described in detail with reference to the accompanying drawings. Without conflict, the following embodiments and the features in the embodiments can be combined with each other.

[0054] Please refer to Figure 1 , Figure 1 Flowchart of a signal quality prediction method based on a generative adversarial network (GAN) provided by an embodiment of the present application.

[0055] As Figure 1 shown, the method includes steps S1 to S2.

[0056] Step S1: Input a picture containing the PCB trace to be measured and the corresponding trace setting parameters into the generator G of the trained generative adversarial network (GAN) to generate an evaluated PCB trace picture with a trace signal quality image.

[0057] Step S2: Input the evaluated PCB trace image into the discriminator D of the trained GAN, and determine whether the trace signal quality image in the evaluated PCB trace image is qualified according to the loss degree of the evaluated PCB trace image.

[0058] It should be noted that the signal quality prediction method based on the generative adversarial network GAN provided in this embodiment is applicable to high-speed traces such as microstrip lines, stripline, and vias. If the PCB trace to be measured is a microstrip line or a stripline, the trace setting parameters include the adjacent layer stack structure, the adjacent layer dielectric thickness, the conductivity, the line width, the trace copper thickness, the board dielectric constant, and the signal rate. If the PCB trace to be measured is a via, the trace setting parameters include the via aperture, the outer diameter of the pad, the via length, the dielectric, and the dielectric constant.

[0059] In this embodiment, the case where the PCB trace to be measured is a microstrip line or a stripline is taken as an example for specific description.

[0060] Preferably, the trace signal quality image in this embodiment can be an eye diagram, and the eye diagram reflects the signal quality of a signal at a certain set rate on the trace.

[0061] It should be noted that before inputting the image containing the PCB trace to be measured and the corresponding trace setting parameters into the generator G of the trained generative adversarial network GAN, it also includes establishing a GAN and training the GAN to obtain the trained GAN.

[0062] Before training the GAN, it is necessary to prepare a training dataset. The training dataset includes multiple verified images containing PCB traces, the corresponding trace signal quality images, and the trace setting parameters of the corresponding PCB traces. Among them, the trace signal quality image corresponding to the verified PCB trace image can be a verified eye diagram. The eye diagram corresponding to the verified PCB trace image in the training dataset can be tested by an eye diagram instrument or the result obtained by simulating with a standard simulation software. Each training data in the training dataset includes a verified image containing a PCB trace, the corresponding eye diagram, and the adjacent layer stack structure, the adjacent layer dielectric thickness, the conductivity, the line width, the trace copper thickness, the board dielectric constant, and the signal rate. This training dataset is used as the real data in the GAN generative adversarial network.

[0063] Furthermore, a GAN is established and trained to obtain a trained GAN. The specific steps include: fixing the parameters of an original generator G in the established original GAN, inputting random noise into the original generator G, and obtaining a sample image generated by the original generator G; inputting the sample image and the training data in a preset training data set into an original discriminator D of the original GAN, and optimizing the original discriminator D with the goal of maximizing the objective function of the discriminator D according to the discrimination result of the original discriminator D; fixing the parameters of the optimized discriminator D, and optimizing the original generator G with the goal of minimizing the objective function of the generator G; and cyclically optimizing the generator G and the discriminator D until a sample image generated by the generator G according to random noise is input into the discriminator D, and the discriminator D determines that the loss of the generated sample image is less than a preset loss threshold, thereby obtaining the trained GAN.

[0064] It is worth noting that if Figure 2 As shown in the figure, when training GAN, the sample pictures generated by the generator G based on random noise are used as fake samples, and the pictures in the training data set are used as true samples and are simultaneously input into the discriminator D. The unadjusted discriminator D will produce a large error when identifying the sample pictures and the pictures in the training data set. The discriminator D can be optimized based on the error so that the optimized discriminator D can identify the sample pictures and the pictures in the training data set. Then the generator G is trained and adjusted so that the sample pictures generated by the generator G can deceive the discriminator D, and the training is repeated in this way.

[0065] For demonstration, first establish the objective function of GAN:

[0066]

[0067] Where D is the generator, G is the discriminator, Pdata(x) is the real data distribution, Pnoise(z) is the distribution of the generator, x is a sample of Pdata(x), z is a sample of Pnoise(z), D(x) indicates that x comes from the distribution of real data, and G(z) indicates that the sample is generated by the generator.

[0068] Ideally, D(x) = 1, logD(x) = 0; if it is not ideal, the smaller the D(x) output, the smaller the logD(x). It should be understood that the sample z of Pnoise(z) includes random noise, and the sample x of Pdata(x) includes the data in the training data set.

[0069] During training, first fix the generator G and do not adjust the parameters of the generator G. According to the objective function of the discriminator D, input random noise into the generator G to obtain the sample pictures generated by G. Then input the generated sample pictures into the discriminator D and adjust the discriminator D according to the discrimination result to maximize the objective function of the discriminator D, that is, obtain the discriminator D. The formula of the objective function of the discriminator D is:

[0070]

[0071] Among them, D is the generator, G is the discriminator, Pdata(x) is the real data distribution, Pnoise(z) is the distribution of the generator, D(x) represents the distribution that x comes from the real data, and G(z) represents the sample generated by the generator.

[0072] Then, without changing the settings of the discriminator D, adjust the generator G to minimize the objective function of the generator G and obtain the generator G. Among them, adjusting the generator G includes methods such as regularizing the input data or modifying the loss function, which will not be elaborated here. The obtained objective function of the generator G is:

[0073]

[0074] Among them, D is the generator, G is the discriminator, Pnoise(z) is the distribution of the generator, and G(z) represents the sample generated by the generator.

[0075] Training once in this way can improve the accuracy of the generator G. Continue to perform multiple cyclic iterative trainings to obtain the optimal generator G for predicting the quality of the routing signal. It can be understood that the training mechanism is that the generator G and the discriminator D are trained separately and alternately: when training the discriminator D, fix the parameters of the generator G. If the result output after x is input into the discriminator D is real, and the random noise z is input into the generator to get G(z), and then the output result after inputting into the discriminator D is false, train the discriminator until convergence. When training the generator G, fix the parameters of the discriminator D. The "false picture" obtained by inputting the random noise into the generator G, and then the result obtained by inputting it into the discriminator is real, train the generator until convergence, and finally obtain the trained GAN.

[0076] In some embodiments, the method further includes obtaining the picture containing the PCB routing to be measured and the corresponding routing setting parameters. Specifically, it includes: taking a screenshot from the electronic design automation (EDA) software to obtain the picture containing the PCB routing to be measured, and obtaining the corresponding routing setting parameters; clearing the routings in the picture containing the PCB routing to be measured that are at a distance greater than a preset multiple of the line pitch from the PCB routing to be measured; setting the routing setting parameters as label data, and establishing a mapping relationship between the data label and the picture containing the PCB routing to be measured.

[0077] Exemplary, such as Figure 3 As shown, a picture containing the PCB trace to be measured can be directly obtained by taking a screenshot from EDA software. At the same time, the setting parameters of the PCB trace to be measured need to be attached, including the adjacent stack-up structure, adjacent layer dielectric thickness, conductivity, line width, trace copper thickness, board dielectric constant, and signal rate. These information form a mapping relationship with the picture in the form of tag data. After obtaining the picture of the PCB trace to be measured, irrelevant traces with a line distance more than three times are cleared, and only the target trace is retained to reduce the interference of irrelevant traces on the prediction result.

[0078] As a preferred implementation manner, after obtaining the picture containing the PCB trace to be measured and the corresponding trace setting parameters, it further includes: establishing a 3D trace model according to the trace setting parameters; performing three-dimensional verification on the picture containing the PCB trace to be measured according to the 3D trace model. Here, the three-dimensional verification means generating a 3D trace model according to parameters such as the stack-up structure, dielectric, conductivity, line width, copper thickness, and dielectric constant. A corresponding 3D trace model can be made using 3D modeling software. The 3D trace model is as Figure 4 shown, and then comparing the S parameters of the picture of the PCB trace to be measured to verify the correctness of the data and avoid the invalid interference of incorrect data on the generator.

[0079] Furthermore, the preprocessed picture containing the PCB trace to be measured and the corresponding trace setting parameters are input into the generator G of the trained generative adversarial network GAN, and an evaluated PCB trace picture with a trace signal quality image generated by the generator G is obtained. The evaluated PCB trace picture is as Figure 5 shown. And the evaluated PCB trace picture is input into the discriminator D of the trained GAN to determine whether the trace signal quality image in the evaluated PCB trace picture generated by the generator G can be used as the predicted signal quality of the PCB trace to be measured.

[0080] Specifically, determining whether the trace signal quality image in the evaluated PCB trace image is qualified according to the loss degree of the evaluated PCB trace image includes: inputting the evaluated PCB trace image into the trained discriminator D, and determining whether the loss of the evaluated PCB trace image is less than a preset loss threshold through the trained discriminator D; if so, determining the signal quality of the to-be-tested PCB trace according to the trace signal quality image in the evaluated PCB trace image, that is, when the loss is extremely small after the image data output by the generator G is calculated by the discriminator D and is determined to be real, the trace signal quality image in the obtained evaluated PCB trace image is the signal quality prediction of the trace. Otherwise, segment the to-be-tested PCB trace, input the to-be-tested PCB trace image corresponding to each segment into the trained generator G respectively, and determine the signal quality of the to-be-tested PCB trace of each segment according to the trace signal quality image in the evaluated PCB trace image generated by the trained generator G for each segment.

[0081] Further, segmenting the to-be-tested PCB trace includes: segmenting at a position where the local impedance in the to-be-tested PCB trace is continuous and the trace bending angle is less than a preset angle.

[0082] It should be noted that due to the complex actual trace length and winding method, according to the characteristics of microstrip lines and stripline characteristic impedance being approximately the same, for the result that the eye diagram generated by the generator G is not ideal, the trace is segmented. The segmentation principle is that the local impedance is continuous and the trace bending is less than a certain acute angle. After separating several segments of the trace and the corresponding trace setting parameters are respectively put into the generator G, the performance evaluation of each segment is obtained, the generation or training difficulty caused by complex traces is reduced, and the accuracy of the trace signal quality image generated by the generator G is improved. The segmented PCB trace is as Figure 6 shown.

[0083] As a preferred implementation manner, verify the evaluated PCB trace image and the corresponding trace signal quality image generated by the generator G, and put the evaluated PCB trace image, the corresponding trace signal quality image and the corresponding trace setting parameters into the training dataset after passing the verification for subsequent training of the GAN.

[0084] Exemplarily, as Figure 7 shown, the signal quality prediction method based on the generative adversarial network GAN specifically includes the following steps:

[0085] Step S201, obtain an image containing the to-be-tested PCB trace.

[0086] Step S202, preprocess the image containing the to-be-tested PCB trace.

[0087] Step S203: Input the image containing the PCB trace to be measured into the GAN generator G to obtain an evaluated PCB trace image.

[0088] Step S204: Input the evaluated PCB trace image into the GAN discriminator D.

[0089] Step S205: The discriminator D determines whether the loss of the evaluated PCB trace image exceeds the loss threshold; if so, go to Step S206, otherwise go to Step S207.

[0090] Step S206: Segment the PCB trace to be measured and go to Step S203.

[0091] Step S207: Use the corresponding eye diagram in the evaluated PCB trace image as the signal quality of the PCB trace to be measured.

[0092] Step S208: Verify the signal quality of the evaluated PCB trace and the corresponding eye diagram, and put them into the training dataset after successful verification.

[0093] It should be noted that the PCB trace to be measured can be a via, and the via structure and signal quality can also be trained and generated through the GAN generation network using the above method. As Figure 8 shown, it is a via PCB image. After preprocessing and translation of the via PCB image, the corresponding Figure 9 3D via model can be obtained. The set parameters of the via include the aperture of the via, the outer diameter of the pad, the length of the via, the medium, and the dielectric constant, etc. After training the generator and discriminator with a large number of verified via PCB images, corresponding eye diagrams, and set parameters, the performance evaluation eye diagram of the via as shown in Figure 10 can be generated according to the new via image and model.

[0094] This embodiment provides a signal quality prediction method and device based on the generative adversarial network GAN. By inputting the image containing the PCB trace to be measured and the corresponding trace setting parameters into the generator G of the trained generative adversarial network GAN, an evaluated PCB trace image with the trace signal quality image is generated; the evaluated PCB trace image is input into the discriminator D of the trained GAN, and according to the loss degree of the evaluated PCB trace image, it is determined whether the trace signal quality image in the evaluated PCB trace image is qualified. It realizes the prediction of the signal quality of the PCB trace without using the license-charging simulation software, improves the efficiency of the PCB trace performance evaluation, and reduces the cost.

[0095] Please refer to Figure 11 Figure 11 ​Schematic block diagram of a signal quality prediction device based on a generative adversarial network (GAN) provided by an embodiment of the present application.

[0096] As Figure 11 shown, the device includes: a generation module configured to input a picture including a PCB trace to be measured and corresponding trace setting parameters into a generator G of a trained generative adversarial network (GAN) to generate an evaluation PCB trace picture with a trace signal quality image;

[0097] a determination module configured to input the evaluation PCB trace picture into a discriminator D of the trained GAN, and determine whether the trace signal quality image in the evaluation PCB trace picture is qualified according to the loss degree of the evaluation PCB trace picture.

[0098] Among them, the device is further configured to:

[0099] obtain the picture including the PCB trace to be measured by taking a screenshot from an electronic design automation (EDA) software, and obtain corresponding trace setting parameters;

[0100] clear traces in the picture including the PCB trace to be measured that are at a distance greater than a preset multiple of the line pitch from the PCB trace to be measured;

[0101] set the trace setting parameters as label data, and establish a mapping relationship between the label data and the picture including the PCB trace to be measured.

[0102] Among them, the device is further configured to:

[0103] establish a 3D trace model according to the trace setting parameters;

[0104] perform three-dimensional verification on the picture including the PCB trace to be measured according to the 3D trace model.

[0105] Among them, the determination module is further configured to:

[0106] input the evaluation PCB trace picture into the trained discriminator D, and determine whether the loss of the evaluation PCB trace picture is less than a preset loss threshold through the trained discriminator D;

[0107] if so, determine the signal quality of the PCB trace to be measured according to the trace signal quality image in the evaluation PCB trace picture;

[0108] otherwise, segment the PCB trace to be measured, input the corresponding picture of each segment of the PCB trace to be measured into the trained generator G respectively, and determine the signal quality of each segment of the PCB trace to be measured according to the trace signal quality image in the evaluation PCB trace picture generated by the trained generator G.

[0109] Among them, the determination module is further configured to:

[0110] Segment at positions where the local impedance in the PCB trace to be measured is continuous and the bending angle of the trace is less than a preset angle.

[0111] Among them, the device is further configured to:

[0112] Fix the parameters of the original generator G in the established original GAN, input random noise into the original generator G, and obtain the sample images generated by the original generator G;

[0113] Input the sample images and the training data in the preset training dataset into the original discriminator D of the original GAN, and optimize the original discriminator D with the goal of maximizing the objective function of the discriminator D according to the discrimination result of the original discriminator D;

[0114] Fix the parameters of the optimized discriminator D, and optimize the original generator G with the goal of minimizing the objective function of the generator G;

[0115] Iteratively optimize the generator G and the discriminator D until, when the sample images generated by the generator G according to random noise are input into the discriminator D and the discriminator D determines that the loss of the generated sample images is less than a preset loss threshold, the trained GAN is obtained.

[0116] Among them, the training dataset includes multiple verified images containing PCB traces, the corresponding trace signal quality images, and the trace setting parameters of the corresponding PCB traces.

[0117] Among them, the device is further configured to:

[0118] Verify the evaluation PCB trace images and the corresponding trace signal quality images, and put the evaluation PCB trace images, the corresponding trace signal quality images, and the corresponding trace setting parameters into the training dataset after passing the verification.

[0119] Among them, if the PCB trace to be measured is a microstrip line or a strip line, the trace setting parameters include the adjacent layer stack structure, the adjacent layer dielectric thickness, the conductivity, the line width, the trace copper thickness, the board dielectric constant, and the signal rate;

[0120] If the PCB trace to be measured is a via, the trace setting parameters include the via aperture, the outer diameter of the pad, the via length, the dielectric, and the dielectric constant.

[0121] It should be noted that those skilled in the art can clearly understand that for the convenience and conciseness of description, the specific working processes of the above-described device and each module and unit can refer to the corresponding processes in the foregoing embodiments, and will not be described herein again.

[0122] It should be noted that in this document, the term "including", "comprising" or any other variant thereof is intended to cover non-exclusive inclusion, such that a process, method, article or system including a series of elements not only includes those elements but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or system. Without further limitation, an element defined by the statement "including a..." does not exclude the existence of additional identical elements in the process, method, article or system including such element.

[0123] The serial numbers of the embodiments of the present application above are only for description and do not represent the superiority or inferiority of the embodiments. The above is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application can easily think of various equivalent modifications or substitutions, and these modifications or substitutions should all be covered within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

Claims

1. A signal quality prediction method based on the Generative Adversarial Network (GAN), characterized in that, Including: Input the image containing the PCB trace to be measured and the corresponding trace setting parameters into the generator G of the trained generative adversarial network GAN to generate an evaluation PCB trace image with a trace signal quality image; Input the evaluation PCB trace image into the discriminator D of the trained GAN, and determine whether the trace signal quality image in the evaluation PCB trace image is qualified according to the loss degree of the evaluation PCB trace image; Among them, the step of inputting the evaluation PCB trace image into the discriminator D of the trained GAN and determining whether the trace signal quality image in the evaluation PCB trace image is qualified according to the loss degree of the evaluation PCB trace image includes: Input the evaluation PCB trace image into the trained discriminator D, and determine whether the loss of the evaluation PCB trace image is less than a preset loss threshold through the trained discriminator D; If so, determine the signal quality of the PCB trace to be measured according to the trace signal quality image in the evaluation PCB trace image; Otherwise, segment the PCB trace to be measured, input the corresponding PCB trace image of each segment into the trained generator G respectively, and determine the signal quality of the PCB trace to be measured in this segment according to the trace signal quality image in the evaluation PCB trace image generated by the trained generator G.

2. The signal quality prediction method based on the generative adversarial network GAN according to claim 1, wherein It also includes obtaining the image containing the PCB trace to be measured and the corresponding trace setting parameters, which includes: Obtain the image containing the PCB trace to be measured by taking a screenshot from the electronic design automation (EDA) software, and obtain the corresponding trace setting parameters; Clear the traces in the image containing the PCB trace to be measured that are at a distance greater than a preset multiple of the line pitch from the PCB trace to be measured; Set the trace setting parameters as label data, and establish a mapping relationship between the label data and the image containing the PCB trace to be measured.

3. The signal quality prediction method based on the generative adversarial network GAN according to claim 2, characterized in that, After obtaining the image containing the PCB trace to be measured and the corresponding trace setting parameters, it also includes: Establish a 3D trace model according to the trace setting parameters; Perform three-dimensional verification on the image containing the PCB trace to be measured according to the 3D trace model.

4. The signal quality prediction method based on the generative adversarial network GAN according to claim 1, characterized in that, Segmenting the PCB trace to be measured includes: Segment at the position where the local impedance of the PCB trace to be measured is continuous and the trace bending angle is less than a preset angle.

5. The signal quality prediction method based on the generative adversarial network GAN according to claim 1, characterized in that Before inputting the image containing the PCB trace to be measured and the corresponding trace setting parameters into the generator G of the trained generative adversarial network GAN, it also includes: Fix the parameters of the original generator G in the established original GAN, input random noise into the original generator G, and obtain the sample image generated by the original generator G; Input the sample image and the training data in the preset training dataset into the original discriminator D of the original GAN, and optimize the original discriminator D with the maximization of the discriminator D objective function as the goal according to the discrimination result of the original discriminator D. Fix the parameters of the optimized discriminator D, and optimize the original generator G with the goal of minimizing the objective function of the generator G; Iteratively optimize the generator G and the discriminator D until, when the sample image generated by the generator G based on random noise is input into the discriminator D and the discriminator D determines that the loss of the generated sample image is less than a preset loss threshold, the trained GAN is obtained.

6. The signal quality prediction method based on the generative adversarial network GAN according to claim 5, wherein The training dataset includes multiple verified pictures containing PCB traces, corresponding trace signal quality images, and trace setting parameters of the corresponding PCB traces.

7. The signal quality prediction method based on the generative adversarial network GAN according to claim 5, wherein It further includes: Verify the evaluation PCB trace pictures and the corresponding trace signal quality images, and after passing the verification, put the evaluation PCB trace pictures, the corresponding trace signal quality images, and the corresponding trace setting parameters into the training dataset.

8. The signal quality prediction method based on the generative adversarial network GAN according to claim 1, characterized in that: If the PCB trace to be measured is a microstrip line or a strip line, the trace setting parameters include the adjacent layer stack structure, the adjacent layer dielectric thickness, the conductivity, the line width, the trace copper thickness, the board dielectric constant, and the signal rate; If the PCB trace to be measured is a via, the trace setting parameters include the via aperture, the outer diameter of the pad, the via length, the dielectric, and the dielectric constant.

9. A signal quality prediction device based on the generative adversarial network GAN, characterized in that, It includes: A generation module, which is used to input the picture containing the PCB trace to be measured and the corresponding trace setting parameters into the generator G of the trained generative adversarial network GAN to generate an evaluation PCB trace picture with a trace signal quality image; A determination module, which is used to input the evaluation PCB trace picture into the discriminator D of the trained GAN, and determine whether the trace signal quality image in the evaluation PCB trace picture is qualified according to the loss degree of the evaluation PCB trace picture; Among them, the determination module is further used for: Input the evaluation PCB trace picture into the trained discriminator D, and determine whether the loss of the evaluation PCB trace picture is less than a preset loss threshold through the trained discriminator D; If so, determine the signal quality of the PCB trace to be measured according to the trace signal quality image in the evaluation PCB trace picture; Otherwise, segment the PCB trace to be measured, input the pictures of each corresponding PCB trace to be measured into the trained generator G respectively, and determine the signal quality of each segment of the PCB trace to be measured according to the trace signal quality images in the evaluation PCB trace pictures generated by the trained generator G.

Images