A method and device for predicting microwave transmission characteristics based on single gold wire bonding

Abstract

The application discloses a microwave transmission characteristic prediction method and device based on single gold wire bonding, and the method comprises the following steps: acquiring bonding gold wire parameter data; inputting the bonding gold wire parameter data into a pre-built artificial neural network model to obtain a loss function value of the bonding gold wire parameter; and comparing the loss function value of the bonding gold wire parameter with a pre-set threshold value to determine whether the network prediction is accurate. The application predicts the corresponding relationship between the arch height and span of the bonding gold wire and the insertion loss, return loss and standing wave ratio by means of the self-built artificial neural network model. The method can quickly and accurately infer whether the microwave characteristics of the bonding product are qualified, greatly reduces the time and economic cost of the engineering personnel, and improves the work efficiency.

Description

TECHNICAL FIELD

[0001] The present application relates to a single gold wire bonding-based microwave transmission characteristic prediction method and device, and belongs to the technical field of gold wire bonding parameter prediction. BACKGROUND

[0002] With the rapid development of science and technology, gold wire bonding is a special welding technology, which mainly uses gold wire to connect the same or different metals, semiconductors, plastics and ceramics, and is widely used in semiconductor devices and integrated circuit packaging, and is the most common and simplest effective method. Gold wire has the advantages of high electrical conductivity, corrosion resistance and good toughness, and is therefore widely used in small power device and integrated circuit packaging compression welding process. In the microwave chip circuit, gold wire bonding technology is often used to realize the interconnection between the microstrip transmission line, monolithic microwave integrated circuit and lumped component. Unlike the interconnection line in the digital circuit, the number, length, arch height, span, and solder joint position of the bonding gold wire will have a serious impact on the microwave transmission characteristics. Whether the span or the arch height increases will result in an increase in loss, an increase in the standing wave ratio, a decrease in the forward transmission coefficient and a decrease in transmission performance. SUMMARY

[0003] The purpose of the present application is to overcome the shortcomings of the prior art, provide a single gold wire bonding-based microwave transmission characteristic prediction method and device, which predicts the corresponding relationship between the arch height and span of the bonding gold wire and the insertion loss, return loss and standing wave ratio through a self-built artificial neural network model. This method can quickly and accurately determine whether the microwave characteristics of the bonding product are qualified, greatly reducing the time and economic cost of the engineering personnel and improving the work efficiency.

[0004] To achieve the above purpose, the present application adopts the following technical scheme:

[0005] In a first aspect, the present application provides a single gold wire bonding-based microwave transmission characteristic prediction method, comprising:

[0006] Obtaining bonding gold wire parameter data;

[0007] Inputting the bonding gold wire parameter data into a pre-built artificial neural network model to obtain the loss function value of the bonding gold wire parameter;

[0008] Comparing the loss function value of the bonding gold wire parameter with a pre-set threshold value to determine whether the network prediction is accurate.

[0009] Further, a Dropout mechanism is added to the artificial neural network model.

[0010] Further, the bonding gold wire parameter data includes the arch height, span and microstrip line width of the bonding gold wire.

[0011] Further, the mean square error (MSE) is selected as the loss function, and the formula is as follows:

[0012]

[0013] In formula (1), m is the number of samples, y i and are the true value and the predicted value of the i-th sample, respectively.

[0014] Further, the loss function value of the bonding wire parameter includes the insertion loss, the return loss, and the standing wave ratio.

[0015] Further, the training and testing method of the artificial neural network model comprises:

[0016] obtaining a training set and a test set;

[0017] inputting the training set into the artificial neural network model for training, and when the training of the artificial neural network model reaches a set degree, inputting the test set into the artificial neural network model for testing;

[0018] During the testing, the change of the loss function value is used for judgment, and when the loss function value reaches a set convergence condition, the training is terminated and the network model is saved.

[0019] Further, the initial value of the weight parameter of the artificial neural network model is assigned before the artificial neural network model starts training.

[0020] In a second aspect, the present application provides a device for predicting the microwave transmission characteristics based on a single gold wire bonding, comprising:

[0021] a data acquisition module for acquiring bonding wire parameter data;

[0022] a prediction module for inputting the bonding wire parameter data into a pre-built artificial neural network model to obtain a loss function value of the bonding wire parameter;

[0023] a judgment module for comparing the loss function value of the bonding wire parameter with a pre-set threshold value to determine whether the network prediction is accurate.

[0024] In a third aspect, the present application provides an electronic device comprising a processor and a storage medium.

[0025] The storage medium is used for storing instructions.

[0026] The processor is used for operating according to the instructions to perform the steps of the method according to any one of the preceding aspects.

[0027] In a fourth aspect, the present application provides a computer readable storage medium, which stores a computer program, and the program is executed by a processor to implement the steps of the method of any one of the preceding aspects.

[0028] Compared with the prior art, the present application has the following beneficial effects:

[0029] The present application provides a microwave transmission characteristic prediction method and device based on single gold wire bonding, which predicts the corresponding relationship between the arch height and span of the bonded gold wire and the insertion loss, return loss and VSWR through an independently built artificial neural network model. The method can quickly and accurately determine whether the microwave characteristics of the bonded product are qualified, greatly reducing the time and economic cost of engineers and improving work efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0030] Figure 1 The figure is a structural model diagram of the gold wire bonding interconnection line of the present application;

[0031] Figure 2 The figure is a flow diagram of the artificial neural network model of the present application;

[0032] Figure 3 The figure is a model flow diagram of the present application;

[0033] Figure 4 The figure is a return loss neural network prediction result diagram of the present application;

[0034] Figure 5 The figure is an insertion loss neural network prediction result diagram of the present application;

[0035] Figure 6 The figure is a VSWR neural network prediction result diagram of the present application. DETAILED DESCRIPTION

[0036] The present application will be further described below in conjunction with the drawings. The following examples are only used to more clearly illustrate the technical solutions of the present application, and cannot be used to limit the protection scope of the present application.

[0037] Example 1

[0038] This example introduces a microwave transmission characteristic prediction method based on single gold wire bonding, which includes:

[0039] Obtain the bonding gold wire parameter data;

[0040] Input the bonding gold wire parameter data into the pre-built artificial neural network model to obtain the loss function value of the bonding gold wire parameter;

[0041] Compare the loss function value of the bonding gold wire parameter with the pre-set threshold value to determine whether the network prediction is accurate.

[0042] In a further embodiment, a Dropout mechanism is added in the artificial neural network model.

[0043] In a further embodiment, the bonding wire parameter data includes arch height, span and microstrip line width of the bonding wire.

[0044] In a further embodiment, mean square error (MSE) is selected as the loss function, and the formula is as follows:

[0045]

[0046] In formula (1), m is the number of samples, y i and are the true value and the predicted value of the i-th sample, respectively.

[0047] In a further embodiment, the loss function value of the bonding wire parameter includes insertion loss, return loss and VSWR.

[0048] In a further embodiment, the training and testing method of the artificial neural network model comprises:

[0049] obtaining a training set and a test set;

[0050] inputting the training set into the artificial neural network model for training, and inputting the test set into the artificial neural network model for testing when the training of the artificial neural network model reaches a set degree;

[0051] During testing, the change of the loss function value is used for judgment, and the training is terminated and the network model is saved when the loss function value reaches a set convergence condition.

[0052] In a further embodiment, the initial value of the weight parameter of the artificial neural network model is assigned before the artificial neural network model starts training.

[0053] The method for predicting the microwave transmission characteristics based on single gold wire bonding provided in the embodiment, like Figure 1This is a schematic diagram of a gold wire bonding interconnect structure. The chip and the T-shaped microstrip line are connected by a single gold wire interconnect. The arch height of the bonding wire is h, the span is l, and the microstrip linewidth is w. The substrate at the bottom of the model is composed of double-sided copper-clad laminate (Rogers 4350) with a dielectric constant of 3.66 and a loss tangent of 0.004. The adjacent small cube chip is made of nickel-plated gold-plated material (Mo80Cu20), and the T-shaped microstrip is made of copper. All three materials have copper clad on the bottom, with a thickness of 0.035 mm. The interconnecting gold wire is made of gold. The six geometric parameters we designed are the arch height h, span l, and microstrip linewidth w of the bonding wire. The microstrip linewidths are numbered w1, w2, w3, and w4 from left to right.

[0054] To make the research more feasible, we need to impose boundary constraints on the geometric parameters to provide data support for the subsequent construction of the artificial neural network model. The variation range of each design parameter is shown in Table 1 below:

[0055] Table 1. Distribution of Structural Model Design Parameters for Gold Wire Bonded Interconnects

[0056] Parameter name Minimum value Maximum value Arch height h 0.1 mm 0.2 mm Span 1 0.3 mm 0.6 mm Microstrip line width w1 1.841 mm 2.241 mm Microstrip line width w2 0.192 mm 0.392 mm Microstrip line width w3 1.593 mm 1.993 mm Microstrip line width w4 0.267 mm 0.667 mm

[0057] In the X-band, each design parameter limits the parameter range, and parameter simulation is performed within this range. The simulated S-parameters are sampled at equal intervals of 0.1 GHz. Each data set has 41 sampling points, and the three data sets have a total of 123 sampling points, resulting in a total of 39,375 data pairs. The three data sets are S11, S12, and VSWR, where S11 represents return loss, S12 represents insertion loss, and VSWR represents standing wave ratio.

[0058] Self-built artificial neural network models, such as Figure 2 As shown in Table 2 below, to prevent overfitting, a Dropout mechanism is added to the artificial neural network model to enhance the generalization ability of the neural network and prevent overfitting.

[0059] Table 2 Neural Network Structure Parameters

[0060] Name Number of neurons Activation function BN Dropout Input layer Input 6 ReLU No No First hidden layer h1 20 ReLU No No Second hidden layer h2 50 ReLU No No Third hidden layer h3 100 ReLU Yes No Fourth hidden layer h4 400 ReLU Yes No Fifth hidden layer h5 400 ReLU Yes No Sixth hidden layer h6 200 ReLU Yes Yes Output layer Output 123 No No No

[0061] The bonding wire prediction model used in the design is a classic regression model. The performance of a regression model is mainly evaluated by the similarity between the model output value and the true value. MSE reflects the mathematical expectation of the square of the difference between the model prediction value and the true value. The smaller the MSE value, the closer the model prediction value is to the true value, and the higher the model accuracy.

[0062]

[0063] Formula (1) in which m is the number of samples, y i and are the true value and the predicted value of the i-th sample, respectively, is the average value of the true values of the m samples. We select the mean square error MSE as the loss function to minimize the spatial distance between the model output and the actual label, and then train the neural network. The data set is randomly divided into 1:9, where 1 is the test set and 9 is the training set. The data in the training set will participate in the training of the neural network, so that the neural network can learn the data features through back propagation. The test set data does not participate in the training of the neural network. Before the neural network model starts training, the initial value of the network model weight parameter needs to be assigned. The value is relatively small, so that the starting value of the repeatedly trained network is not much different, and plays a positive role in evaluating different parameters on the network model training. During the training of the neural network model, considering the infinite approximation characteristics of the neural network, the model is not unique. When the network training reaches a certain degree, the test set data is input into the network for model evaluation. When the test set data is good, it is considered that the model has reached the convergence condition, and the training is terminated and the network model is saved. The whole model flow chart is shown in Figure 3 .

[0064] The convergence of the model is judged by the change of the loss value. When the loss curve appears divergence or repeated turbulence, it is determined that the model does not converge, and the model needs to be adjusted.

[0065] The three groups of data in this experiment are trained under the same model. In order to reduce the error, each experiment is iterated 20000 times, and the loss function value is shown in Table 3.

[0066] Table 3 Loss function value change table

[0067] Loss function name Training set loss Test set loss Return loss S11 0.121 0.593 Insertion loss S12 0.001 0.005 VSWR 0.002 0.011

[0068] During the training of the model, the loss function value shows a straight line downward trend, then the speed slows down, and finally the speed tends to be flat, converging around a small value. From the loss function values of training and prediction, the error is small. We set the variance threshold to 1.5. When the variance value is less than 1.5, it is determined that the network prediction is accurate, otherwise it is determined to be inaccurate. We selected 1300 groups of data in the test set, and the data that met the prediction preparation reached 1274 groups, with an accuracy rate of 98%. At this time, we consider that the network training has reached the convergence state, and the network training is terminated and the network model is saved.

[0069] Part of the return loss test set prediction results are as follows Figure 4As shown in the X-band, the center frequency is 10GHz, the actual and predicted results of return loss in (a) are better than 24dB, and the average error between them is within 0.1dB, the actual and predicted results of return loss in (b) are better than 23dB, and the average error between them is within 0.2dB.

[0070] The prediction results of the partial insertion loss test set are as follows Figure 5 As shown in the X-band, the center frequency is 10GHz, the actual and predicted results of return loss in (c) are better than 0.2dB, and the average error between them is within 0.05dB, the actual and predicted results of return loss in (d) are better than 0.16dB, and the average error between them is within 0.1dB.

[0071] The prediction results of the partial standing wave ratio test set are as follows Figure 6 As shown in the X-band, the center frequency is 10GHz, the actual and predicted results of standing wave ratio in (e) are better than 1.1dB, and the average error between them is within 0.05dB, the actual and predicted results of return loss in (f) are better than 1.22dB, and the average error between them is within 0.1dB.

[0072] From the above Figure 4 , Figure 5 , Figure 6 It can be seen that the neural network has learned the microwave transmission characteristics of the single gold wire bonding interconnection line S parameter, and can well predict the S parameter. The neural network model training needs to consume a certain amount of time, and the trained neural network model can predict data in an instant, while the electromagnetic simulation needs a lot of time, and under the same conditions, the simulation results and the prediction results are not much different. Moreover, this model can also study the microwave transmission characteristics of any parameter under this structure, and only needs to input the data into the neural network to obtain the microwave transmission characteristics under the corresponding parameter.

[0073] The beneficial effects of the embodiment are: based on the deep learning method in this paper, an artificial neural network model is built to realize the prediction of the microwave transmission characteristics of the single gold wire bonding interconnection line S parameter. The mean square error of return loss is 0.593, the mean square error of insertion loss is 0.005, and the mean square error of standing wave ratio is 0.011. The prediction error is small and within the acceptable range. It can be seen that the prediction result of the artificial neural network model proposed in this paper has high accuracy, and it is feasible to apply the deep learning method to the prediction of the S parameter of the bonding gold wire.

[0074] Embodiment 2

[0075] The embodiment provides a microwave transmission characteristic prediction device based on single gold wire bonding, comprising:

[0076] A data acquisition module is configured to acquire bonding gold wire parameter data.

[0077] a prediction module configured to input the bonding wire parameter data into a pre-built artificial neural network model to obtain a loss function value of the bonding wire parameter;

[0078] a judgment module configured to compare the loss function value of the bonding wire parameter with a pre-set threshold value to determine whether the network prediction is accurate.

[0079] Embodiment 3

[0080] This embodiment provides an electronic device, comprising a processor and a storage medium;

[0081] The storage medium is configured to store instructions;

[0082] The processor is configured to operate according to the instructions to perform the steps of the method according to any one of embodiments 1.

[0083] Embodiment 4

[0084] This embodiment provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method according to any one of embodiments 1.

[0085] The above description is only the preferred embodiments of the present application, and it should be pointed out that for those skilled in the art, without departing from the technical principles of the present application, a number of improvements and modifications can be made, and these improvements and modifications should also be considered as the protection scope of the present application.

Claims

1. A method for predicting microwave transmission characteristics based on a single gold wire bonding, characterized by, The method comprises the following steps: Obtaining bonding wire parameter data; the bonding wire parameter data comprises arch height, span and microstrip line width of the bonding wire; the arch height h ranges from 0.1 mm to 0.2 mm, the span l ranges from 0.3 mm to 0.6 mm, the microstrip line width has four values, w1, w2, w3 and w4, w1 ranges from 1.841 mm to 2.241 mm, w2 ranges from 0.192 mm to 0.392 mm, w3 ranges from 1.593 mm to 1.993 mm, and w4 ranges from 0.267 mm to 0.667 mm; parameter simulation is performed within the above parameter range, the S parameters obtained through simulation are sampled at equal intervals, the interval is 0.1 GHz, each group of data has 41 sampling points, three groups of data have a total of 123 sampling points, and a total of 39375 pairs of data are obtained, and the three groups of data are S11, S12 and VSWR, wherein S11 represents return loss, S12 represents insertion loss, and VSWR represents standing wave ratio; The bonding wire parameter data is input into a pre-established artificial neural network model with a Dropout mechanism to obtain a loss function value of the bonding wire parameter; the structure parameters of the artificial neural network model include an input layer, a first hidden layer, a second hidden layer, a third hidden layer, a fourth hidden layer, a fifth hidden layer and a sixth hidden layer, and the Dropout mechanism is added in the sixth hidden layer; The loss function value of the bonding wire parameter comprises insertion loss, return loss and standing wave ratio; wherein the return loss of the training set loss is 0.121, the return loss of the test set loss is 0.593, the insertion loss of the training set loss is 0.001, the insertion loss of the test set loss is 0.005, the standing wave ratio of the training set loss is 0.002, and the standing wave ratio of the test set loss is 0.011; The loss function value of the bonding wire parameter is compared with a pre-set threshold value to determine whether the network prediction is accurate. The training and testing method of the artificial neural network model comprises the following steps: Obtaining a training set and a test set; The training set is input into the artificial neural network model for training, and when the artificial neural network model training reaches a set degree, the test set is input into the artificial neural network model for testing; During testing, the change of the loss function value is used for judgment, and when the loss function value reaches a set convergence condition, the training is terminated and the network model is saved.

2. The single-wire gold wire bonding based microwave transmission characteristics prediction method of claim 1, wherein, The mean square error (MSE) is selected as the loss function, and the formula is as follows: （1）； In equation (1), m is the number of samples, and are the true and predicted values of the ith sample, respectively.

3. The single-wire gold bond-based microwave transmission characteristic prediction method of claim 1, wherein, Before the artificial neural network model starts training, the initial value of the weight parameter of the artificial neural network model is assigned.

4. An apparatus for predicting the microwave transmission characteristics of a single gold wire bond, for implementing the method for predicting the microwave transmission characteristics of a single gold wire bond according to any one of claims 1 to 3, characterized in that, The device comprises: A data acquisition module for acquiring bonding wire parameter data; A prediction module for inputting the bonding wire parameter data into a pre-established artificial neural network model to obtain a loss function value of the bonding wire parameter; A judgment module for comparing the loss function value of the bonding wire parameter with a pre-set threshold value to determine whether the network prediction is accurate.

5. An electronic device, comprising: The device comprises a processor and a storage medium; The storage medium is used to store instructions; The processor is used to operate according to the instructions to perform the steps of the method according to any one of claims 1-3.

6. A computer readable storage medium having stored thereon a computer program, characterized in that: The program, when executed by the processor, implements the steps of the method of any one of claims 1-3.

Images