Method for predicting signal transmission capability of microsystem interconnection product in radiation environment

Abstract

The invention discloses a method for predicting the signal transmission capability of a microsystem interconnection product in a radiation environment, and relates to the technical field of electronic engineering, radiation influence analysis and machine learning. According to the invention, the problem that the signal integrity of the micro-system device is affected in a radiation environment is solved, and the performance change of various micro-system devices under the radiation condition is effectively evaluated and predicted by systematically analyzing and modeling the response of different devices under the radiation effect. Therefore, the reliability of the microsystem device in a radiation environment is improved, and the normal operation of the microsystem device in key applications is ensured.

Description

Technical Field

[0001] The present invention relates to the fields of electronic engineering, radiation impact analysis and machine learning technology, and in particular to a method for predicting the signal transmission capability of a microsystem interconnect product under a radiation environment. Background Art

[0002] With the continuous development of aerospace technology, 3D micro-system devices are widely used in various space exploration, satellite communication and navigation systems. These devices have the characteristics of small size, high functional integration and superior performance, and can meet the needs of complex space missions. However, the application of 3D micro-system devices in space environment faces many challenges, especially the influence of radiation.

[0003] There are a large number of high-energy particles in the space environment, including cosmic rays and high-energy particles from the sun. These radiations pose a serious threat to the performance of microsystem devices. Specifically, microsystem devices may experience a variety of damage effects under radiation, such as total dose effect (TID), single event upset (SEU) and dielectric breakdown (DIELECTRIC BREAKDOWN). These radiation damage effects can cause material defects (such as vacancies, dislocations and oxygen defects) and impurities in the transmission structure. These defects and impurities affect the material properties, which are mainly manifested in the conductivity of the transmission line and the relative dielectric constant of the substrate and oxide layer, which in turn affects the signal transmission capability of the entire structure.

[0004] The attenuation of signal transmission capability may lead to a series of serious consequences. First, the reduction of signals may lead to delays or errors in information transmission, affecting the integrity of data and the real-time response capability of the system, which is unacceptable in space missions. Second, signal distortion during transmission may lead to misoperation of equipment, thus affecting the safety and reliability of the entire space system. In addition, long-term signal attenuation may also lead to malfunction or failure of microsystems, increasing the risk of mission failure, resulting in huge economic losses and waste of resources.

[0005] Therefore, studying the signal transmission capability of microsystem interconnect structures is not only to ensure the reliability and stability of devices in radiation environments, but also the key to improving the performance and application effects of microsystems under extreme conditions. Understanding the attenuation of signal transmission capability and its relationship with material defects can provide important theoretical support for designing interconnect structures with better radiation resistance. This will help promote the application of microsystem devices in the aerospace field and ensure that they can effectively perform tasks in future complex space missions.

[0006] In summary, in-depth research on the attenuation mechanism and influencing factors of signal transmission capability can not only improve the performance of microsystem devices, but also lay a solid foundation for the future development of aerospace technology and ensure efficient operation in extreme environments.

[0007] At present, there is still a lack of research on the damage effects that three-dimensional microsystem devices may suffer in the space radiation environment, and there is a lack of systematic evaluation and design methods. Traditional experimental methods often need to be carried out in a specific radiation environment, which is costly and difficult to implement. In addition, existing radiation testing methods usually rely on expensive equipment and complex experimental configurations, and cannot flexibly adapt to the evaluation needs of different types of microsystem devices.

[0008] Therefore, an effective design method and evaluation model is urgently needed to better understand the radiation effects on 3D microsystem devices in the space environment, so as to improve their reliability and performance in aerospace applications. This will provide important technical support for the design and optimization of aerospace devices to ensure normal operation under extreme conditions.

[0009] The existing technologies currently have problems such as high testing costs for the impact of radiation on the signal transmission performance of microsystem interconnect structures, difficulty in obtaining measuring equipment, the need to customize special probes, large errors in the calibration of experimental equipment, difficulty in conducting irradiation experiments, and unclear effects of radiation damage effects on the parasitic parameters of interconnect structures. Summary of the invention

[0010] The technical problem to be solved by the present invention is to overcome the shortcomings of the prior art and provide a method for predicting the signal transmission capability of microsystem interconnect products under a radiation environment. The present invention realizes rapid modeling of the influence of radiation damage on the transmission performance of the interconnect structure, outputs results and joint simulation, and uses big data to train a neural network model. The trained model can timely and reliably predict the scattering parameters of the device based on a given input.

[0011] The present invention adopts the following technical solutions to solve the above technical problems:

[0012] The method for predicting the signal transmission capability of a microsystem interconnect product under a radiation environment proposed by the present invention comprises:

[0013] Step 1: According to the structural parameters and material parameters of the microsystem interconnection product, a geometric model corresponding to the microsystem interconnection product is built in the electromagnetic simulation software to obtain the first scattering parameter describing the signal transmission characteristics;

[0014] Step 2: Based on the structural characteristics of the microsystem interconnect product, a vector network analyzer is used to measure the microsystem interconnect product under different radiation dose conditions at the same operating frequency to measure the second scattering parameter;

[0015] Step 3, comparing and analyzing the first scattering parameter and the second scattering parameter, and using an adaptive method to adjust and correct the material parameters so that the coincidence degree between the first scattering parameter and the second scattering parameter is more than 90%, thereby obtaining a verified geometric model;

[0016] Based on the verified geometric model, the material parameters of the geometric model under different radiation dose conditions are obtained through simulation;

[0017] Step 4: Divide the microsystem interconnect product into multiple sub-parts according to the structural characteristics of the microsystem interconnect product; use the hierarchical equivalent method to calculate the equivalent parameters of each sub-part based on the structural parameters and material parameters of the verified geometric model, the equivalent parameters include resistance, inductance and capacitance, and obtain the equivalent circuit of each sub-part according to the equivalent parameters of each sub-part;

[0018] In the circuit simulation software, the equivalent circuits of each sub-part are combined to build an overall equivalent circuit model, and the same operating frequency is set to carry out simulation operations to obtain the third scattering parameters of the circuit simulation;

[0019] Step 5: Compare the second scattering parameter with the third scattering parameter to lock the radiation sensitive area of ​​the microsystem interconnection product, use the adaptive strategy to adjust the equivalent parameter RLC of the radiation sensitive area, and continue to optimize until the overlap between the second scattering parameter and the third scattering parameter reaches more than 90%, thereby obtaining a verified equivalent circuit;

[0020] Based on the verified equivalent circuit, the equivalent parameters of the equivalent circuit under different radiation dose conditions are obtained through simulation. The equivalent parameters include equivalent resistance, equivalent inductance, and equivalent capacitance.

[0021] Step 6: Build a neural network model and input the database as training data into the neural network model. The data in the database include radiation dose, operating frequency, material parameters in step 3, and equivalent parameters obtained in step 5. Based on the back propagation method, the data in the database are processed layer by layer and feature extracted through the input layer and hidden layer in the neural network model. Finally, the output layer in the neural network model generates scattering parameters for characterizing the signal transmission capability of microsystem interconnect products.

[0022] As a further optimization scheme of the method for predicting the signal transmission capability of microsystem interconnect products under a radiation environment described in the present invention, a geometric model and an equivalent circuit are used to explain the signal transmission capability of microsystem interconnect products under different radiation conditions from a three-dimensional physical perspective and a two-dimensional electrical perspective.

[0023] As a further optimization scheme of the method for predicting the signal transmission capability of microsystem interconnect products under a radiation environment described in the present invention, a layered equivalence method is constructed as follows: the microsystem interconnect products are divided into three sub-parts according to their structure: redistribution layer RDL, through silicon via TSV and bump BUMP.

[0024] Compared with the prior art, the present invention adopts the above technical solution and has the following technical effects:

[0025] (1) The present invention solves the problem of signal integrity being affected in a radiation environment for microsystem devices. By systematically analyzing and modeling the responses of different devices under radiation, the present invention can effectively evaluate and predict the performance changes of various microsystem devices under radiation conditions. This will help improve the reliability of microsystem devices in radiation environments and ensure their normal operation in critical applications.

[0026] (2) The present invention is suitable for solving the high requirements of the irradiation experiment environment and the extremely long time spent on measuring high-dose points, while avoiding the increase in the human error rate caused by long-term experiments, thereby greatly reducing the cost of radiation experiments and improving experimental safety. At the same time, the model of this scheme has strong scalability. When the radiation dose changes, the modeling method of the present invention can safely, efficiently and quickly derive the scattering parameters at different dose points, while also avoiding the complicated calibration process required for scattering parameter measurement. The present invention can provide technical support for the application of three-dimensional interconnected structures in irradiation environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 is the geometric model of the interconnection structure of the present invention;

[0028] Figure 2 It is a general implementation flow chart of the present invention;

[0029] Figure 3 is an equivalent circuit model of the interconnect structure of the present invention; wherein (a) is an RDL equivalent circuit, (b) is a TSV and BUMP equivalent circuit, and (c) is a simplified overall circuit diagram;

[0030] Figure 4 is a comparison diagram of experimental and simulated S21 when the interconnect structure of the present invention is not radiated;

[0031] Figure 5 This is the L2 regularization method + BP neural network model training and prediction process of the present invention;

[0032] Figure 6 is a neural network structure diagram of the present invention;

[0033] Figure 71 is a training effect diagram of the scattering parameter S21 of the neural network training set sample of the present invention; wherein, (a) is a training data fitting diagram when R=0.99979, (b) is a verification data fitting diagram when R=0.99987, (c) is a test data fitting diagram when R=0.99984, and (d) is a full data fitting diagram when R=0.99981;

[0034] Figure 8 is a prediction effect diagram of the neural network prediction scattering parameter S21 of the present invention; wherein (a) is R 2 = 0.9534 and L = 0.01451. (b) is the comparison of the training set prediction results when R 2 Comparison of test set prediction results when L = 0.9054 and L = 0.089096. DETAILED DESCRIPTION

[0035] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0036] The present invention relates to multiple fields such as electronic engineering, radiation impact analysis and machine learning, and is specifically aimed at a vertical interconnect structure containing RDL, TSV and BUMP layer structures, and is applied to changes in the signal transmission capability of the structure under a radiation environment.

[0037] With the development of software simulation technology, the full-wave three-dimensional electromagnetic simulation software HFSS is used to model and simulate the geometric structure of the device, and the scattering parameter curve under the specific structure of the device can be intuitively obtained. In addition, the EDA tool software ADS can be used to perform circuit-level design on the geometric model, and the scattering parameter curve under the equivalent circuit can also be obtained. Moreover, nowadays, machine learning is in rapid development, which can not only efficiently process and analyze large data sets, but also make predictions based on historical data.

[0038] The present invention combines HFSS, ADS and neural network to predict the influence of different radiation doses on the signal transmission capability of vertical interconnection structure. First, the scattering parameters of the interconnection structure under different radiation doses are actually measured by using a vector network analyzer. Subsequently, a geometric model of the interconnection structure is established in the HFSS software, and its geometric parameters and material parameters are optimized to ensure that the scattering parameter curve obtained by simulation is highly consistent with the experimental data. Then, with the help of the obtained geometric model, it is equivalent to a circuit model through a formula. By deeply studying the influence of radiation on sensitive areas in the structure, the effect of radiation on the interconnection structure is converted into changes in specific material parameters and equivalent circuit parameters. A large number of data sets are generated, which will serve as the basis for machine learning. On this basis, a BP neural network model is constructed to form a mapping relationship between input and output. After sufficient training, the model can provide reliable results under specific input conditions. Through this method, not only an in-depth analysis of the radiation influence on the three-dimensional interconnection structure of the microsystem is achieved, but also the prediction accuracy is improved while reducing costs. With the continuous accumulation of experimental and simulation data, the model can continue to learn, thereby further improving the accuracy of the prediction results.

[0039] Multi-dimensional collaborative simulation of microsystem interconnect products under radiation environment and quantitative prediction method of signal transmission capability, the operation process is as follows:

[0040] (1) Based on the actual structure of the microsystem interconnect sample and the properties of its constituent materials, a geometric model is constructed in electromagnetic simulation software and simulation is performed to obtain its scattering parameters that describe the signal transmission characteristics.

[0041] (2) The probe head is customized according to the structure of the microsystem interconnection sample, and the scattering parameters of the sample under different radiation doses are measured using a vector network analyzer.

[0042] (3) Compare the scattering parameters obtained by geometric model simulation with the scattering parameters measured by the network analyzer, and use an adaptive algorithm to adjust the material parameters so that the overlap between the simulation data and the measured scattering parameters is more than 90%.

[0043] (4) Construction of hierarchical equivalent circuit model: According to the structural characteristics of the sample, it is divided into multiple sub-parts, and each part is modeled using the hierarchical equivalent method based on the structural parameters of the geometric model. The resistance, inductance, and capacitance values ​​of the interconnect structure corresponding to each sub-part are generated. Subsequently, a complete equivalent circuit model is constructed in the circuit simulation software, and the scattering parameters are obtained through simulation analysis.

[0044] (5) Compare the scattering parameters of the equivalent circuit with those measured by the network analyzer, and adjust the equivalent parameters of the sensitive area to make the simulation match the experimental measurement value. In this way, the changes of the equivalent parameters under different radiation doses are obtained.

[0045] (6) Construction of BP neural network model with L2 regularization: The model adopts back propagation (BP) neural network with a topology of 8-6-6-1. The input layer contains 6 features, namely: radiation dose, operating frequency, three material parameters, and equivalent parameters RLGC (inductance, capacitance, transmission line impedance, etc.). The model is designed with two hidden layers, each containing 6 neurons, to capture the nonlinear characteristics of the data. In order to improve the generalization ability of the model and effectively suppress overfitting, L2 regularization is introduced. L2 regularization helps control the size of network weights by adding weight penalty terms to the loss function, thereby avoiding overfitting of the model to the training data. Finally, the output layer generates the scattering parameters of the interconnected structure to describe the signal transmission characteristics. The design of this model fully considers the expressive power of the multi-layer structure and the stability of the regularization, aiming to achieve more efficient and accurate predictions, and is suitable for complex electromagnetic simulation and circuit design optimization.

[0046] (7) Training the neural network model. Its database comes from geometric model simulation, equivalent circuit level simulation and experimental test. There are 1005 groups in total, of which the first 80% are training sets and the last 20% are test samples. A dynamic learning rate adjustment strategy is adopted, and the maximum number of training rounds is 1000.

[0047] The geometric model is established, and the simulation model is divided into three layers: upper, middle and lower. The order is RDL-BUMP-TSV-RDL-TSV-BUMP-RDL connection. The material of this part is CU. A layer of insulating material SiO2 is wrapped on the outside of TSV and RDL, and the entire middle layer is Si substrate.

[0048] The hierarchical equivalent circuit construction method is characterized in that the overall interconnect structure is divided into three equivalent parts according to its structure, namely, RDL partial equivalent, TSV partial equivalent, and BUMP partial equivalent.

[0049] In the training and prediction of the neural network model, the input data is first standardized during the data processing stage to eliminate the dimensional effects between different features and ensure that the model training process is more stable. In addition, in order to improve the effectiveness of the data, data enhancement technology is used to expand the training set through random perturbations and sample generation to enhance the generalization ability of the model.

[0050] During the training process, a dynamic learning rate adjustment strategy is used to make the model converge quickly in the early stage, and the learning rate is reduced when approaching the optimal solution, thereby improving the accuracy and convergence of the model. At the same time, the L2 regularization method is applied to the first hidden layer to effectively control the complexity of the model and reduce the risk of overfitting, thereby further improving the generalization ability of the model on the test set. In addition, an early stopping mechanism is set to prevent the occurrence of overfitting and ensure the superior performance of the model on the test set.

[0051] In the prediction stage, by applying the trained model to new input data, the scattering parameters of the interconnect structure can be calculated in real time. Combining simulation data and experimental test results, the model can provide accurate predictions, assist in design optimization and decision support, and improve the efficiency and reliability of overall circuit design. This model is not only suitable for this project, but also has the potential for promotion and application to other similar engineering problems.

[0052] like Figure 2 As shown, the present invention creates a method for predicting and analyzing the radiation resistance of interconnected structures. By using a neural network to fuse simulation and measured data, analyze logic, and build a prediction method, the signal and anti-interference performance under different radiation doses can be accurately grasped. The steps are as follows:

[0053] Experiments with different radiation doses were carried out on the experimental samples. The radiation type used in this experiment was gamma radiation, and the doses were: 0Krad(si), 100Krad(si), 500Krad(si), 900Krad(si), and 1500Krad(si); and a vector network analyzer was used to detect the scattering parameter S21 of the samples at different radiation doses.

[0054] According to the experimental sample, its structure consists of RDL, TSV and Bump. The sample adopts a symmetrical structure design to facilitate measurement and equivalence, such as Figure 1 The geometric model of the sample interconnect structure is shown. The HFSS software is used to model it. The scattering parameter S21 is adjusted by optimizing the material parameters to make the simulation fit the measured parameters. The changes in material parameters under different radiation doses are shown in Table 1.

[0055] Table 1 Changes in material parameters under different radiation doses

[0056]

[0057] Due to the actual composition structure of the sample: RDL, TSV, BUMP, an equivalent circuit diagram can be made according to the equivalent method of each part, and according to its signal transmission characteristics, it can be known that the connection method of each part is series. The following is the equivalent method of the RDL part, and its equivalent circuit structure is as follows Figure 3 As shown in (a), the resistance needs to be considered as both DC and AC parts.

[0058] RDL DC resistance:

[0059] RDL AC resistance:

[0060] Skin Depth:

[0061] RDL total resistance:

[0062] RDL Inductor:

[0063] Capacitance between RDL:

[0064] The capacitance of the insulating layer below RDL:

[0065] RDL is equivalent to capacitance in silicon:

[0066] Equivalent conductance of RDL in silicon:

[0067] Among them, l RDL is the length of the RDL layer, w RDL is the width of the RDL layer, t RDL is the thickness of the RDL layer, δ skindepth,RDL is the skin depth of RDL, t RDLox is the insulation layer height, S RDL is the area of ​​RDL, S is the area of ​​the insulating layer, and h eff is the effective height, ρ Cu is the resistivity of copper, f is the frequency, μ RDL is the magnetic permeability of the material used in the RDL layer, σ RDL is the conductivity of the material used for the RDL layer, μ0 is the relative magnetic permeability, ε0 ​​is the relative dielectric constant; ε RDLox1 is the dielectric constant of the insulating layer between RDLs, ε RDLox2 is the dielectric constant of the insulating layer below the RDL, ε eff is the effective dielectric constant, σ eff is the effective conductivity.

[0068] Since the structure of BUMP is not much different from that of TSV, and they are also connected together, their equivalent methods are similar. The following is the equivalent method of Bump and TSV structures. The equivalent circuit schematic diagram is as follows: Figure 3 As shown in (b) in .

[0069] Bump AC resistance:

[0070]

[0071] Bump DC resistance:

[0072] Bump resistor:

[0073] Bump Inductor:

[0074] Bump capacitor:

[0075] Bump and unfilled layer capacitance:

[0076] Among them, h Bump is the height of the bump, d Bump is the diameter of the bump, p TSV is the spacing between TSVs, e Bump is the radius of the bump, r TSV is the radius of TSV, t ox is the oxide layer thickness, h Bumpox is the height of the outer oxide layer of the bump, δ skindepth,Bump is the skin depth of the bump, ρ Bump is the resistivity of the material used for the Bump, μ0 is the vacuum permeability, μ Bump is the magnetic permeability of the material used for the Bump, ε0 is the relative dielectric constant, ε Bumpox is the dielectric constant of the outer insulation material of the bump, ε Underfil is the dielectric constant of the unfilled layer.

[0077] TSV DC resistance:

[0078] TSV AC resistance:

[0079] Skin Depth:

[0080] TSV resistor:

[0081] TSV Inductors:

[0082]

[0083] TSV oxide capacitance:

[0084] Equivalent capacitance of TSV in silicon:

[0085] Equivalent conductance of TSV in silicon:

[0086] Among them, h TSV is the height of TSV, r TSV is the radius of TSV, p TSV is the TSV spacing, d TSV is the diameter of TSV, t ox is the oxide layer thickness, ρ TSV is the resistivity of the material used for TSV, δ skindepth,TSVis the skin depth, μ TSV is the magnetic permeability of the material used for TSV, μ TSV is the conductivity of the material used for TSV, μ0 is the relative magnetic permeability, ε0 ​​is the relative dielectric constant, ε TSVox is the dielectric constant of the TSV outer oxide layer material, ε r,si is the dielectric constant of silicon, σ Si is the electrical conductivity of silicon.

[0087] After the equivalence of each part is completed, the connection order of the structure is: RDL-BUMP-TSV-RDL-TSV-BUMP-RDL; after simplification, the following is obtained Figure 3 The overall circuit diagram shown in (c) .

[0088] After the equivalent circuit is drawn, the simulation frequency of the scattering parameter is set to obtain the scattering curve, and the radiation equivalent parameter RLC is adjusted to obtain the scattering curve matching the experimental different doses. The changes of the equivalent parameters under different doses are shown in Table 2. It can be seen that the sensitive area of ​​the structure under radiation environment is concentrated at the TSV. Figure 4 The non-radiated experiment is compared with the HFSS and ADS simulation curves. In the entire frequency range of 0-30GHz, the experimental data and simulation data are compared and analyzed. The overlap between the simulation and the experiment exceeds 90%, which verifies the reliability of the simulation model.

[0089] Table 2 Changes of equivalent parameters under different radiation doses

[0090]

[0091]

[0092] like Figure 5 As shown, it is a flowchart of the L2 regularization method + BP neural network training and prediction of the present invention, which includes the following steps: initializing network parameters, data preparation, preprocessing data, dividing data sets, importing databases, forward propagation, calculating total loss function, back propagation, updating network parameters (weights and biases), judging convergence, and data prediction.

[0093] like Figure 6As shown, it is a schematic diagram of the neural network structure of the present invention. The input is the radiation dose, the operating frequency, the conductivity of the CU, the relative dielectric constant of Si and SiO2, and the equivalent inductance, resistance and capacitance at the TSV, and the output is the scattering parameter S21 of the interconnection structure. The network consists of two hidden layers, each containing 6 neurons. During the forward propagation of the network, the parameters are passed, and during the back propagation, the parameters are updated. We call the input layer the I layer, the hidden layers are the J layer and the K layer respectively, and the output layer is the O layer. During the training process, the L2 regularization method is applied to the first hidden layer to effectively control the complexity of the model and reduce the risk of overfitting, thereby further improving the generalization ability of the model on the test set. The output of each layer can be expressed according to formulas (1) to (3). Among them, w ij and w jk represents the weights between different layers, b j and b k is the bias parameter of the hidden layer and the output layer, and a i and a k Represents the output value of each layer. The activation function used is the sigmoid function, as shown in formula (4).

[0094] a j =σ(W ij ·a i +b j ) (1)

[0095] a k =σ(W jk ·a j +b k ) (2)

[0096] S 21 =W ko ·a k +b o (3)

[0097]

[0098] The loss function is represented by L, and its specific form is shown in formula (5). Here, N is the number of samples, y n represents the true output, λ is the regularization coefficient, W ij and W jk is the weight parameter.

[0099]

[0100] The network weights and bias parameters are updated through the back propagation algorithm. The derivation process is detailed in formulas (6) to (11). Combined with the gradient descent algorithm, the optimization is iterated continuously until the loss function gradually decreases, thereby obtaining model parameters that meet the requirements. At this point, the model training is completed.

[0101]

[0102]

[0103] Figure 7 This is the training effect diagram of the scattering parameter S21 of the neural network training set sample of the present invention, and the evaluation index adopts the determination coefficient R 2 It is represented as shown in formula (12).

[0104]

[0105] like Figure 7 As shown, the R of the training set samples 2 They are 0.99979, 0.99987, 0.99984 and 0.99981 respectively, which meet the requirements and show excellent fitting effect. Figure 7 It is a training effect diagram of the scattering parameter S21 of the neural network training set sample of the present invention; wherein, Figure 7 (a) is the training data fitting diagram when R=0.99979. Figure 7 (b) is the validation data fitting diagram when R=0.99987. Figure 7 (c) is the test data fitting diagram when R=0.99984. Figure 7 (d) is the fitting diagram of all data when R=0.99981;

[0106] Figure 8 This is a prediction effect diagram of the neural network prediction scattering parameter S21 of the present invention, Figure 8 (a) in the equation is R 2 Comparison of training set prediction results when L = 0.9534 and L = 0.01451, Figure 8 (b) in the equation is R 2 The comparison chart of the prediction results of the test set when L = 0.9054 and L = 0.089096. The comparison chart of the prediction results of the training set shows the prediction results of the training set. The green line in the figure represents the true value, and the red line represents the predicted value. It can be seen from the figure that in most cases, the predicted value is close to the true value, but there is still some deviation. The determination coefficient R is given in the figure 2 =0.9534 and loss value L = 0.01451. The coefficient of determination shows that the model has a good fitting effect on the training set, but there is still room for improvement. The loss value is a key indicator for measuring prediction error. The smaller the value, the smaller the deviation of the prediction from the actual situation. The comparison chart of the prediction results of the test set shows the prediction results of the test set. Compared with the training set, the deviation between the predicted value and the actual value on the test set is relatively large. However, its coefficient of determination R 2=0.9095 and loss value L =0.089096, indicating that although the fitting effect of the model on the test set is slightly inferior to that of the training set, it also has a very high fitting effect.

[0107] The above description is only a specific implementation of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by a technician familiar with the technical field within the technical scope disclosed by the present invention should be covered by the protection scope of the present invention.

Claims

1. A method for predicting the signal transmission capability of a microsystem interconnect product under a radiation environment, characterized in that: include: Step 1: According to the structural parameters and material parameters of the microsystem interconnection product, a geometric model corresponding to the microsystem interconnection product is built in the electromagnetic simulation software to obtain the first scattering parameter describing the signal transmission characteristics; Step 2: Based on the structural characteristics of the microsystem interconnect product, a vector network analyzer is used to measure the microsystem interconnect product under different radiation dose conditions at the same operating frequency to measure the second scattering parameter; Step 3, comparing and analyzing the first scattering parameter and the second scattering parameter, and using an adaptive method to adjust and correct the material parameters so that the coincidence degree between the first scattering parameter and the second scattering parameter is more than 90%, thereby obtaining a verified geometric model; Based on the verified geometric model, the material parameters of the geometric model under different radiation dose conditions are obtained through simulation; Step 4: Divide the microsystem interconnect product into multiple sub-parts according to the structural characteristics of the microsystem interconnect product; use the hierarchical equivalent method to calculate the equivalent parameters of each sub-part based on the structural parameters and material parameters of the verified geometric model, the equivalent parameters include resistance, inductance and capacitance, and obtain the equivalent circuit of each sub-part according to the equivalent parameters of each sub-part; In the circuit simulation software, the equivalent circuits of each sub-part are combined to build an overall equivalent circuit model, and the same operating frequency is set to carry out simulation operations to obtain the third scattering parameters of the circuit simulation; Step 5: Compare the second scattering parameter with the third scattering parameter to lock the radiation sensitive area of ​​the microsystem interconnection product, use the adaptive strategy to adjust the equivalent parameter RLC of the radiation sensitive area, and continue to optimize until the overlap between the second scattering parameter and the third scattering parameter reaches more than 90%, thereby obtaining a verified equivalent circuit; Based on the verified equivalent circuit, the equivalent parameters of the equivalent circuit under different radiation dose conditions are obtained through simulation. The equivalent parameters include equivalent resistance, equivalent inductance, and equivalent capacitance. Step 6: Build a neural network model and input the database as training data into the neural network model. The data in the database include radiation dose, operating frequency, material parameters in step 3, and equivalent parameters obtained in step 5. Based on the back propagation method, the data in the database are processed layer by layer and feature extracted through the input layer and hidden layer in the neural network model. Finally, the output layer in the neural network model generates scattering parameters for characterizing the signal transmission capability of microsystem interconnect products.

2. The method for predicting the signal transmission capability of microsystem interconnect products under radiation environment according to claim 1, characterized in that: Using geometric models and equivalent circuits, the signal transmission capabilities of microsystem interconnect products under different radiation conditions are explained in multiple dimensions from a three-dimensional physical perspective and a two-dimensional electrical perspective.

3. The method for predicting the signal transmission capability of microsystem interconnect products under radiation environment according to claim 1, characterized in that: The construction method of the hierarchical equivalence method is as follows: the microsystem interconnection product is divided into three sub-parts according to the structure: redistribution layer RDL, through silicon via TSV and bump BUMP.

Images