A high-frequency interconnection structure electrical performance prediction method and system based on auxiliary theory quantity joint supervision and online consistency scoring

Abstract

This invention belongs to the interdisciplinary fields of advanced electronic packaging, high-frequency interconnect modeling, and electronic design automation. Specifically, it relates to a method and system for predicting the electrical performance of high-frequency interconnect structures based on joint supervision of auxiliary theoretical quantities and online consistency scoring. The aim is to address the problems of existing high-frequency interconnect structure electrical performance prediction methods, which only directly regress target parameters and lack theoretical supervision and online reliability verification mechanisms. The method includes: determining the target electrical performance parameters and at least two types of auxiliary theoretical quantities whose theoretical values ​​can be directly calculated from the input design parameters; constructing a ternary sample dataset of input design parameters—target electrical performance parameters—auxiliary theoretical quantities; establishing a dual-branch joint supervision model including a shared feature encoding unit, a main prediction branch, an auxiliary theoretical branch, and a physical consistency scoring module; constructing a joint loss function for model training; and outputting the target electrical performance prediction value, the auxiliary theoretical quantity prediction value, and the consistency score, confidence level, or risk warning during the online inference stage. The system includes a data preprocessing module, an auxiliary theoretical quantity calculation module, a joint supervision model construction module, a model training module, an online inference module, a physical consistency scoring module, a result interface module, and a model management and update module. This invention can improve the stability, physical consistency and engineering reliability of prediction results while maintaining the ability to quickly predict the electrical performance of high-frequency interconnect structures. It is applicable to the rapid evaluation and auxiliary design of advanced packaging, packaging substrates, printed circuit board transmission lines and on-chip or inter-chip high-frequency interconnect structures.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary fields of advanced electronic packaging, high-frequency interconnect modeling, and electronic design automation. Specifically, it relates to a method and system for predicting the electrical performance of high-frequency interconnect structures based on joint supervision of auxiliary theoretical quantities and online consistency scoring. This method is applicable to the rapid evaluation, design assistance, and engineering screening of the electrical performance of silicon interposer redistribution layers and through-silicon via interconnect structures, high-speed traces on packaging substrates, printed circuit board transmission lines, and on-chip or inter-chip high-frequency interconnect structures. Background Technology

[0002] Existing methods for obtaining the electrical performance of high-frequency interconnect structures mainly include full-wave electromagnetic simulation, analytical equivalent circuit modeling, and data-driven proxy models. Full-wave electromagnetic simulation has high accuracy, but its computational cost is high in scenarios involving multi-parameter scanning, parameter optimization, and multiple design iterations, making it difficult to meet the needs of rapid evaluation and real-time assisted design of high-frequency interconnect structures. Analytical models have faster computation speed, but their applicability and accuracy are limited for complex three-dimensional interconnect structures, scenarios with significant high-frequency loss effects, and multi-parameter strongly coupled problems.

[0003] In recent years, neural network surrogate models have been used for predicting the electrical performance of high-frequency interconnect structures due to their high inference efficiency. However, most existing methods only perform direct regression on the target electrical performance parameters and lack a joint supervision mechanism for auxiliary theoretical quantities directly calculated from the input design parameters. This makes the models prone to producing prediction results that are inconsistent with the basic electromagnetic mechanism in sparse sample regions, parameter boundary regions, or local extrapolation scenarios, thus affecting the reliability of engineering applications.

[0004] Furthermore, even when physical prior information is introduced during the training phase, existing technologies typically lack mechanisms to continue using theoretically computable quantities to perform consistency scoring, reliability assessment, and risk alerts on the model output during the online inference phase. Existing methods often only output numerical prediction results and cannot provide further auxiliary judgment information such as whether the results meet theoretical constraints or require verification, thus hindering their direct use in solution selection and engineering decision-making within electronic design automation processes.

[0005] Therefore, there is an urgent need for a prediction method and system that can maintain the ability to quickly predict the electrical performance of high-frequency interconnect structures, and can improve the physical reliability and engineering usability of the predictions through joint supervision of auxiliary theoretical quantities and online consistency scoring. Summary of the Invention

[0006] One of the objectives of this invention is to provide a method for predicting the electrical performance of high-frequency interconnect structures based on joint supervision of at least two types of directly computable auxiliary theoretical quantities, so as to improve the stability, physical consistency and engineering reliability of the prediction results of target electrical performance parameters while maintaining fast reasoning capabilities.

[0007] The second objective of this invention is to provide a dual-branch prediction structure, wherein the main branch outputs the predicted value of the target electrical performance parameter, the auxiliary branch outputs the predicted value of the auxiliary theoretical quantity, and the auxiliary theoretical quantity prediction value is compared with the theoretical value directly calculated from the input design parameters through a consistency scoring and alarm mechanism, thereby generating a sample-level or scheme-level consistency score, confidence level or risk warning information.

[0008] The third objective of this invention is to provide a high-frequency interconnect structure electrical performance prediction system that can be deployed in an electronic design automation environment, enabling the system to synchronously return target electrical performance prediction results, auxiliary theoretical quantity prediction results, physical consistency evaluation results, and necessary alarm or verification prompts to designers.

[0009] To achieve the above objectives, the present invention adopts the following technical solution:

[0010] A method for predicting the electrical performance of high-frequency interconnect structures based on joint supervision of auxiliary theoretical quantities and online consistency scoring includes the following steps:

[0011] Step S1: Input design parameters, including geometric parameters, material parameters, and frequency parameters.

[0012] Step S2: Through the computable auxiliary theoretical quantity generation module, at least two types of auxiliary theoretical quantities that can be directly calculated from the input design parameters are generated based on the input design parameters. The first type of auxiliary theoretical quantity is used to characterize the high-frequency loss mechanism of the conductor, preferably at least one of skin depth, AC resistance per unit length, or equivalent loss factor. The second type of auxiliary theoretical quantity is used to characterize the impedance or propagation characteristic mechanism, preferably at least one of characteristic impedance, effective dielectric constant, real part of propagation constant, imaginary part of propagation constant, or group delay.

[0013] Step S3: The prediction model is trained through a joint supervised training module. The prediction model includes a shared feature extraction layer, an electrical performance prediction branch, and an auxiliary theoretical quantity branch. The training samples are a three-element sample dataset consisting of input design parameters, target electrical performance parameters, and auxiliary theoretical quantities. The three-element sample dataset is constructed through electromagnetic simulation, test measurement, or a combination of simulation and measurement.

[0014] Step S4: Output the target electrical performance prediction value and the auxiliary theoretical quantity prediction value through the dual-branch prediction module.

[0015] Step S5: The consistency scoring and alarm module compares the predicted value of the auxiliary theoretical quantity with the theoretical calculation result, generates a consistency score, confidence level or alarm information, and outputs the target electrical performance prediction result and the auxiliary theoretical quantity result by the prediction result output module.

[0016] Furthermore, the joint supervised training module constructs a joint loss function through a joint loss construction module. The joint loss function includes at least a main task loss term and an auxiliary theoretical quantity loss term, and preferably includes a consistency constraint term. The main task loss term is used to constrain the deviation between the predicted value and the true value of the target electrical performance parameter, the auxiliary theoretical quantity loss term is used to constrain the deviation between the predicted value and the theoretical value of the auxiliary theoretical quantity, and the consistency constraint term is used to suppress the directional or trend deviation of the auxiliary theoretical quantity branch output relative to the theoretical calculation result.

[0017] Furthermore, the joint loss function satisfies:

[0018] L=L main +λ1L aux +λ2L cons Among them, L main Indicates the loss of the main task, L aux L represents the auxiliary theoretical quantity loss. cons λ1 and λ2 represent the consistency constraint term and the weight parameters, respectively. The weight parameters can be set to fixed values ​​or dynamically adjusted according to the training rounds.

[0019] Furthermore, the consistency scoring and alarm module can generate a consistency score based on at least one of the following: the absolute deviation, relative deviation, consistency of change direction between the predicted value and the theoretical value of the auxiliary theoretical quantity, and the weighted summary result of multiple auxiliary theoretical quantities.

[0020] Furthermore, when the consistency score is lower than the preset threshold, an alarm is triggered and the corresponding result is marked as a result to be reviewed. If necessary, the secondary simulation module, high-precision alternative model, or result correction module can be called to re-verify or correct the result to be reviewed.

[0021] The present invention also provides a high-frequency interconnect structure electrical performance prediction system for implementing the above method, including a data preprocessing module, an auxiliary theoretical quantity calculation module, a model calling module, a result post-processing and consistency judgment module, and an output module. Preferably, the system further includes a model management and update module for recording model version information and performing model updates, incremental training, or model fine-tuning when new data arrives. Attached Figure Description

[0022] The present invention will be further described below with reference to the accompanying drawings and related design flowcharts.

[0023] Figure 1 This is a flowchart of the overall process for predicting the electrical performance of high-frequency interconnect structures based on joint supervision of auxiliary theoretical quantities and online consistency scoring, as described in this invention.

[0024] Figure 2This is a schematic diagram of the dual-branch prediction and joint loss training structure described in this invention;

[0025] Figure 3 This is a logic diagram of the joint loss construction and online consistency evaluation described in this invention;

[0026] Figure 4 This is a modular architecture diagram of the high-frequency interconnect structure electrical performance prediction system described in this invention.

[0027] Figure 1 The module includes: 1. Input design parameters; 2. A module for generating auxiliary theoretical quantities; 3. Joint supervised training module; 4. Two-branch prediction module; 5. Consistency scoring and alarm module; 6. Prediction result output module.

[0028] Figure 2 In the model: 7. Input vector X; 8. Shared feature extraction layer; 9. Electrical performance prediction branch; 10. Auxiliary theoretical quantity branch; 11. Joint loss construction module; 12. Model parameter update module.

[0029] Figure 3 In the middle: 13. Main task true value input unit; 14. Auxiliary theoretical quantity theoretical value input unit; 15. Model forward calculation unit; 16. Main task loss unit; 17. Auxiliary theoretical quantity loss unit; 18. Joint loss formation unit; 19. Online consistency scoring unit; 20. Prediction result / alarm output unit.

[0030] Figure 4 The module consists of: 21. Data preprocessing module; 22. Auxiliary theoretical quantity calculation module; 23. Model calling module; 24. Result postprocessing and consistency judgment module; and 25. Output module. Detailed Implementation

[0031] The present invention will be further described below with reference to the accompanying drawings. It should be understood that the following embodiments are for illustrative purposes only and are not intended to limit the scope of protection of the present invention.

[0032] Example 1

[0033] Taking the advanced packaged RDL-TSV high-frequency interconnect structure as an example, the input design parameters include geometric parameters, material parameters, and operating frequency. The geometric parameters may include RDL linewidth, line spacing, line thickness, dielectric layer thickness, TSV radius, and TSV depth. The material parameters may include conductor conductivity and dielectric relative permittivity. The target output may be at least one of the S-parameters, preferably at least one of S21, S11, and S22. The auxiliary theoretical quantities may be selected as skin depth δ, characteristic impedance Z0, or other theoretical quantities that can be directly calculated from the input design parameters.

[0034] like Figure 1As shown, the overall process of the method of the present invention includes: first, inputting geometric, material, and frequency information by input design parameter 1; then, calculating auxiliary theoretical quantities by the computable auxiliary theoretical quantity generation module 2 based on the input design parameters; next, performing joint training by the joint supervised training module 3 based on the target electrical performance parameters and the auxiliary theoretical quantities; after training, outputting the target electrical performance prediction value and the auxiliary theoretical quantity prediction value by the dual-branch prediction module 4; then, performing consistency evaluation on the prediction results by the consistency scoring and alarm module 5; finally, outputting the S-parameter prediction result, the auxiliary quantity prediction result, and the corresponding consistency prompts or alarm information by the prediction result output module 6.

[0035] like Figure 2 As shown, the input vector X is input to the shared feature extraction layer 8 by the input vector X unit 7. The shared feature extraction layer 8 is used to extract common features from the input samples. The output of the shared feature extraction layer 8 is input to the electrical performance prediction branch 9 and the auxiliary theoretical quantity branch 10, respectively. The electrical performance prediction branch 9 outputs S21, S11 or other target indicators, and the auxiliary theoretical quantity branch 10 outputs δ, Z0 or other computable quantities. The outputs of the electrical performance prediction branch 9 and the auxiliary theoretical quantity branch 10 are input to the joint loss construction module 11 to construct the joint loss function. The output of the joint loss construction module 11 is used to update the model parameters module 12 to complete the model training.

[0036] like Figure 3 As shown, during the training phase, the main task true value input unit 13 and the auxiliary theoretical quantity theoretical value input unit 14 are jointly input into the model forward calculation unit 15. The model forward calculation unit 15 outputs the main task predicted value and the auxiliary theoretical quantity predicted value. The main task predicted value and the main task true value together form the main task loss unit 16, and the auxiliary theoretical quantity predicted value and the auxiliary theoretical quantity theoretical value together form the auxiliary theoretical quantity loss unit 17. The main task loss unit 16 and the auxiliary theoretical quantity loss unit 17 are further input into the joint loss forming unit 18 to form the joint loss function. During the deployment phase, the model output after the joint loss forming unit 18 has been trained enters the online consistency scoring unit 19. The online consistency scoring unit 19 generates a consistency evaluation result based on the difference between the auxiliary theoretical quantity predicted value and the theoretical calculation result, and the prediction result / alarm output unit 20 outputs the prediction result, risk warning, or alarm information.

[0037] like Figure 4As shown, the high-frequency interconnect structure electrical performance prediction system of the present invention includes a data preprocessing module 21, an auxiliary theoretical quantity calculation module 22, a model calling module 23, a result post-processing and consistency judgment module 24, and an output module 25. The data preprocessing module 21 is used to perform normalization, standardization, outlier removal, or missing value processing on the input design parameters; the auxiliary theoretical quantity calculation module 22 is used to directly calculate the theoretical values ​​of the auxiliary theoretical quantities based on the input design parameters; the model calling module 23 is used to call the trained bi-branch prediction model to generate the target electrical performance prediction value and the auxiliary theoretical quantity prediction value; the result post-processing and consistency judgment module 24 is used to compare the differences between the auxiliary theoretical quantity prediction values ​​and the auxiliary theoretical quantity theoretical values ​​and perform a consistency judgment; the output module 25 is used to output the prediction value, the auxiliary quantity estimate value, and consistency prompt information.

[0038] Embodiments of the present invention will now be described in detail. Example figures of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote elements with the same or similar functions. The description of these embodiments with reference to the accompanying drawings is exemplary and intended to explain the invention, not to limit it.

[0039] In a preferred embodiment, the shared feature extraction layer 8 may include a fully connected structure, a normalization operation, an activation function, or an attention weighting mechanism; the joint loss function in the joint loss construction module 11 can be written as L=L main +λ1L aux +λ2L cons Among them, L main Indicates the loss of the main task, L aux L represents the auxiliary theoretical quantity loss. cons The above settings represent consistency constraints, with λ1 and λ2 representing the corresponding weight parameters. By using these settings, the accuracy of target electrical performance prediction can be improved, while simultaneously enhancing the physical consistency and engineering reliability of the model output.

[0040] Example 2

[0041] For PCB microstrip lines, striplines, high-speed transmission lines on packaged substrates, or other high-frequency interconnect structures, the framework of this invention remains unchanged. Only the auxiliary theoretical quantities are replaced with two or more of the following, depending on the specific structural characteristics: AC resistance per unit length, effective dielectric constant, real part of the propagation constant, imaginary part of the propagation constant, or group delay. That is, the key to this invention is not limited to a specific interconnect structure, but rather lies in the joint supervised prediction and online consistency scoring output mechanism based on at least two directly computable auxiliary theoretical quantities.

Claims

1. A method for predicting the electrical performance of high-frequency interconnect structures based on joint supervision of auxiliary theoretical quantities and online consistency scoring, characterized in that, Includes the following steps: S1. Input design parameters, including geometric parameters, material parameters, and frequency parameters; S2. Through the computable auxiliary theoretical quantity generation module, at least two types of auxiliary theoretical quantities whose theoretical values ​​can be directly calculated from the input design parameters are generated according to the input design parameters. S3. The prediction model is trained through a joint supervised training module. The prediction model includes a shared feature extraction layer, an electrical performance prediction branch, and an auxiliary theoretical quantity branch. S4. Output the target electrical performance prediction value and the auxiliary theoretical quantity prediction value through the dual-branch prediction module; S5. The consistency scoring and alarm module compares the predicted values ​​of the auxiliary theoretical quantities with the theoretical calculation results, generates a consistency score, confidence level or alarm information, and outputs the target electrical performance prediction results and auxiliary theoretical quantity results by the prediction result output module.

2. The method for predicting the electrical performance of high-frequency interconnect structures according to claim 1, characterized in that, Of the at least two types of auxiliary theoretical quantities, the first type of auxiliary theoretical quantity is used to characterize the high-frequency loss mechanism of the conductor, and is at least one of skin depth, AC resistance per unit length, or equivalent loss factor.

3. The method for predicting the electrical performance of high-frequency interconnect structures according to claim 1, characterized in that, Of the at least two types of auxiliary theoretical quantities, the second type of auxiliary theoretical quantity is used to characterize the impedance or propagation mechanism, and is at least one of characteristic impedance, effective dielectric constant, real part of propagation constant, imaginary part of propagation constant, or group delay.

4. The method for predicting the electrical performance of high-frequency interconnect structures according to claim 1, characterized in that, The target electrical performance prediction value is at least one of the S parameters, preferably at least one of S11, S21 and S22.

5. The method for predicting the electrical performance of high-frequency interconnect structures according to claim 1, characterized in that, The joint supervised training module constructs a joint loss function through a joint loss construction module. The joint loss function includes at least a main task loss term and an auxiliary theoretical quantity loss term.

6. The method for predicting the electrical performance of a high-frequency interconnect structure according to claim 5, characterized in that, The joint loss function also includes a consistency constraint term, which is used to suppress directional or trend deviations of the auxiliary theoretical quantity branch output relative to the theoretical calculation results.

7. The method for predicting the electrical performance of high-frequency interconnect structures according to claim 5, characterized in that, The joint loss function satisfies: L=L main +λ1L aux +λ2L cons Among them, L main Indicates the loss of the main task, L aux L represents the auxiliary theoretical quantity loss. cons λ1 and λ2 represent the consistency constraint term and the weight parameters, respectively.

8. The method for predicting the electrical performance of a high-frequency interconnect structure according to claim 1, characterized in that, The consistency scoring and alarm module generates a consistency score based on at least one of the following: absolute deviation, relative deviation, consistency of change direction, and weighted summation results of multiple auxiliary theoretical quantities between the predicted values ​​of auxiliary theoretical quantities and the theoretical calculation results.

9. The method for predicting the electrical performance of a high-frequency interconnect structure according to claim 1, characterized in that, When the consistency score is lower than a preset threshold, the consistency score and alarm module triggers an alarm and marks the corresponding result as a result to be reviewed.

10. The method for predicting the electrical performance of a high-frequency interconnect structure according to claim 9, characterized in that, When the consistency score is lower than the preset threshold, the secondary simulation module, high-precision alternative model, or result correction module is invoked to re-verify or correct the results to be reviewed.

11. A high-frequency interconnect structure electrical performance prediction system based on joint supervision of auxiliary theoretical quantities and online consistency scoring, characterized in that, include: The data preprocessing module is used to preprocess the input design parameters; The auxiliary theoretical quantity calculation module is used to directly calculate the theoretical values ​​of auxiliary theoretical quantities based on the input design parameters. The model calling module is used to call the trained dual-branch prediction model and output the target electrical performance prediction value and the auxiliary theoretical quantity prediction value. The result post-processing and consistency judgment module is used to compare the difference between the predicted value of the auxiliary theoretical quantity and the theoretical value of the auxiliary theoretical quantity, and to perform consistency scoring, confidence level judgment or alarm judgment. The output module is used to output predicted values, estimated auxiliary quantities, and consistency prompts.

12. The high-frequency interconnect structure electrical performance prediction system according to claim 11, characterized in that, It also includes a model management and update module, which records model version information and performs model updates, incremental training, or model fine-tuning when new data arrives.

Images