Stacked chip vertical interconnection coupling electromagnetic field extraction method and computer equipment

Abstract

The invention discloses a stacked chip vertical interconnection coupling electromagnetic field extraction method and computer equipment, and the method comprises the steps: determining a target position of an electromagnetic field generated by TSV vertical interconnection coupling in stacked chips, carrying out simulation to obtain the electromagnetic field intensity of the target position, and storing TSV coupling structure parameters and the extracted electromagnetic field intensity as a data set; loading data from the data set and carrying out feature extraction, constructing a feature matrix by utilizing the extracted features, and inputting the data set into each regression model for training by taking the electromagnetic field intensity as a target variable to obtain an optimal model of the electromagnetic field intensity; comparing a model prediction result with a simulation result, calculating whether an error meets a requirement or not, and if the error cannot meet the requirement, repeating feature extraction and model training; and inputting the TSV coupling structure parameters of the stacked chip to be tested into the prediction model to obtain the electromagnetic field intensity of the target position. According to the method, the electromagnetic field interference of the internal circuit of the core particle caused by TSV vertical interconnection electromagnetic coupling can be quickly predicted.

Description

Technical Field

[0001] The present invention belongs to the field of chip electromagnetic interference evaluation, and particularly relates to a method for extracting the coupled electromagnetic field of vertical interconnection of stacked chips and a computer device. Background Art

[0002] The electromagnetic noise sources of Chiplet stacked chips are wide, and they are greatly affected by stacking process dimensions, operating frequencies, electromagnetic source spacing, shielding protection measures, etc. Stacked chips achieve vertical interconnection through TSV (Through-Silicon Via), but there is a large size span between TSV and CMOS advanced processes. For the heterogeneous integration requirements of radio frequency devices under different processes, the TSV and RDL (Redistribution Layer) paths inside the silicon substrate will form external electromagnetic interference to circuit units and components. For example, the electromagnetic field coupled between the high-speed interconnections of TSVs in the transmission signal lines interferes with the input and output of other die and the internal signal timing.

[0003] Regarding the problem of internal electromagnetic interference evaluation in Chiplet stacked chips, the existing technologies focus on analytical algorithms and experimental tests. Among them, the analytical method extracts the high-frequency parasitic parameters of TSVs and calculates the conductive electromagnetic field and the radiated electromagnetic field. However, due to the complex variables and large analytical workload of Chiplet stacked chips, it is difficult to construct an accurate analytical model. Experimental testing is the standard method for solving the electromagnetic field between board-level chips, but EMC (Electro Magnetic Compatibility) testing requires devices such as test probes. Due to the size limitations of Chiplet stacked chips, there is currently no corresponding solution to conduct in-depth testing between DIE (die or bare chip), resulting in difficult EMC testing. Summary of the Invention

[0004] The purpose of the present invention is to provide a method for extracting the coupled electromagnetic field of vertical interconnection of stacked chips, a computer device, a computer-readable storage medium, and a computer program product, which can quickly predict the electromagnetic field interference of the internal circuits of die caused by the electromagnetic coupling of TSV vertical interconnection.

[0005] To achieve the above purpose, one aspect of the present invention provides a method for extracting the coupled electromagnetic field of vertical interconnection of stacked chips, including:

[0006] Step S1, establishing a three-dimensional model of the stacked chip through finite element simulation, determining the target position of the electromagnetic field generated by the vertical interconnection coupling of TSVs in the stacked chip, running the simulation to obtain the electric field strength and magnetic field strength at the target position, and storing the TSV coupling structure parameters and the extracted electric field strength and magnetic field strength as a data set;

[0007] Step S2, load data from the dataset and perform feature extraction, construct a feature matrix using the extracted features, select multiple different types of regression models, use the grid search method to tune the hyperparameters of the selected regression models, determine the optimal hyperparameters for each regression model, use the electric field strength and magnetic field strength as target variables, input the dataset into each regression model for training, and obtain the best model for the electric field strength and the best model for the magnetic field strength according to the comprehensive scores of the R 2 score and MAPE;

[0008] Step S3, use the best model for the electric field strength and the best model for the magnetic field strength to predict the electric field strength and magnetic field strength respectively, compare the prediction results with the electric field strength and magnetic field strength at the target position obtained by simulation, calculate whether the error meets the requirements, if not, return to Step S2 until the error requirement is met;

[0009] Step S4, input the TSV coupling structure parameters of the stacked chips to be measured, and use the best model for the electric field strength and the best model for the magnetic field strength to output the electric field strength and magnetic field strength at the target position.

[0010] Preferably, the TSV vertical interconnection coupling includes the TSV coupling between dielets on the same layer, the TSV coupling between dielets on different layers, the coupling between TSV and RDL, and the coupling between the metal connections of TSV and RDL.

[0011] Preferably, the features include original features, interaction features, and squared features, where the interaction features are features generated by pairwise combination of the original features, and the squared features are the squares of the original features.

[0012] Preferably, the original features include the number of coupling paths, distance, frequency, TSV diameter, dielet thickness, RDL width, and length.

[0013] Preferably, the regression models include Ridge regression, Lasso regression, random forest regression, gradient boosting regression, and support vector machine.

[0014] Preferably, in Step S2, the dataset is divided into 5 test sets, the model is repeatedly trained and verified under different test sets, the negative MAPE is used as the evaluation index for each training and verification, the average value of 5 times is used as the final evaluation index of the model, and the hyperparameters of the model with the smallest MAPE index are used as the optimal hyperparameters.

[0015] Preferably, in Step S3, the key features that affect the electromagnetic field change among all features are used as input variables, and other non-key features are set to fixed values for fixed dataset setting to reduce the model overhead.

[0016] Another aspect of the present invention provides a computer device, including a memory, a processor, and a computer program stored on the memory, where the processor executes the computer program to implement the steps of the above method.

[0017] Another aspect of the present invention provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the above method.

[0018] Another aspect of the present invention provides a computer program product, including a computer program, and when the computer program is executed by a processor, it implements the steps of the above method.

[0019] According to the stacked chip vertical interconnection coupling electromagnetic field extraction method, computer device, computer-readable storage medium, and computer program product of the above aspects of the present invention, it is possible to quickly predict the electromagnetic field interference inside the die caused by the TSV vertical interconnection electromagnetic coupling. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention, and for those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings:

[0021] Figure 1 is a flowchart of the stacked chip vertical interconnection coupling electromagnetic field extraction method according to an embodiment of the present invention;

[0022] Figure 2 is a schematic diagram of a three-dimensional model of a stacked chip according to an embodiment of the present invention;

[0023] Figure 3 is a schematic diagram of a TSV coupling structure to be measured according to an embodiment of the present invention, where (a) is a sectional view and (b) is a top view;

[0024] Figure 4 is a structural diagram of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0025] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the drawings. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0026] An embodiment of the present invention provides a method for extracting the electromagnetic field of vertical interconnection coupling of stacked chips, as Figure 1 shown. The method for extracting the electromagnetic field of vertical interconnection coupling of stacked chips in the embodiment of the present invention includes steps S1 to S4.

[0027] In step S1, electromagnetic field data generated by TSV vertical interconnection coupling in the stacked chips is extracted through finite element simulation. Specifically, in finite element simulation software such as HFSS and CST, a three-dimensional model of the stacked chips as Figure 2 shown is established, and the target position of the electromagnetic field generated by TSV vertical interconnection coupling in the stacked chips is determined. As Figure 2 shown, the coupling objects include: TSV coupling between dielets on the same layer, TSV coupling between dielets on different layers, coupling between TSV and RDL, and coupling between the metal connection of TSV and RDL. The media between the coupling objects include in-substrate coupling, in-cavity coupling, and multi-media coupling through the interlayer insulation material. The target position points where the electromagnetic field intensity needs to be extracted are determined, the simulation is run, the electric field intensity (E) and magnetic field intensity (H) at the target position are extracted, and the physical parameters of the TSV coupling structure of the stacked chips and the simulation result data of the extracted electromagnetic field intensity are stored as a dataset file in CSV format.

[0028] In step S2, an electromagnetic field intensity prediction model is trained based on a machine learning method, including steps S21 to S24.

[0029] Step S21: Data loading and feature engineering

[0030] Data is loaded from the dataset and features are extracted. These features are divided into material feature variables, process feature variables, working feature variables, etc. A feature matrix is constructed using the extracted features. The feature matrix can include N original features, N(N - 1) / 2 interaction features generated by pairwise combination of the original features, and square features of the N original features, for a total of (N 2 + 3N) / 2 features. The target variables are E (electric field intensity) and H (magnetic field intensity). Among them, the original features include the number of coupling paths (n), distance (d), frequency (f), TSV process diameter (D), dielet thickness (t), RDL width (w), length (l), etc. Interaction features such as the number of paths * frequency, TSV process diameter * RDL width, etc., and square features are the squares of the original features, such as n 2 、d 2 、f 2 etc.

[0031] Step S22: Model selection and hyperparameter tuning

[0032] Model Selection: Multiple different types of regression models such as Ridge regression, Lasso regression, Random Forest regression, Gradient Boosting regression, and Support Vector Machine (SVR) are used for initial training.

[0033] Hyperparameter Tuning: The GridSearchCV method is used to tune the hyperparameters of the selected regression models (parameters such as the regularization term, number of iterations, and depth of decision trees that directly affect the model performance and cannot be learned from training).

[0034] To determine the optimal hyperparameters for each regression model, the specific approach is as follows: Use 5-fold cross-validation to evaluate the performance of the model based on each hyperparameter on the dataset. Divide the dataset into 5 test sets, and then repeatedly train and validate the model under different test sets. Each time of training and validation uses the negative mean absolute percentage error (MAPE) as the evaluation metric, and the average value of 5 times is used as the final evaluation metric for the model. The hyperparameters of the model with the smallest MAPE metric are used as the optimal hyperparameters.

[0035] Step S23: Selection of the Best Model

[0036] After completing the tuning of the hyperparameters for each regression model, the dataset is input into each regression model for training. According to the comprehensive score of R 2 score and MAPE (R 2 -MAPE) as the final score. The model with the highest score is the best model. Here, the models for the target variables E and H are independent, so a best model for E and a best model for H are obtained.

[0037] Step S24: Model Validation and Visualization

[0038] Analyze the distribution of prediction errors through residual plots, and verify the model fitting degree through scatter plots of predicted values vs actual values to show the performance of the best model on the test set.

[0039] In step S3, the prediction model is optimized and verified by setting a fixed dataset. Key feature parameters that affect the electromagnetic field change, such as the number of coupling paths n, TSV diameter D, and other physical parameter features, are used as input variables, and other non-critical features are set to fixed values for the fixed dataset setting to reduce the model overhead.

[0040] Load a new dataset (including TSV structure parameters not involved in training), obtain the E / H results through the best models of E and H, compare the prediction results with the E / H values at the target position obtained by software simulations such as HFSS, and calculate whether the mean square error or error rate meets the requirements. If the requirements are not met (such as the error rate exceeding the threshold of 5%), return to step S2, select more regression models for combined training, and construct a larger feature matrix for optimization.

[0041] For example, various different types of regression models can be adopted, including Ridge regression, Lasso regression, random forest regression, gradient boosting regression, support vector machine (SVR), etc. One of the regression models can be selected during initial training, and other models can be selected during optimization to meet the error requirements. The feature matrix includes single features and multiple interaction features. N original features can be selected as the matrix during initial training, and interaction features can be added during optimization to form a larger feature matrix.

[0042] Repeat steps S2 - S3 until the error requirements are met.

[0043] In step S4, according to the application requirements, input the coupling structure parameters of the stacked chips to be measured, and output the electromagnetic field strength at the target position through the E / H optimal model as the structural data to be finally predicted.

[0044] The following verifies the effect of the method of the embodiment of the present invention through a specific example.

[0045] Taking the number of paths n, the center horizontal distance d, and the target operating frequency f as variable features, and other parameters such as the TSV diameter as fixed values, predict the magnetic field strength (H) of the n coupling paths with a 2 - layer die substrate in the coupling structure. During the prediction process, a random forest regression model is used for training. The original features are the number of paths n, the distance d, and the frequency f. The commonly used interaction features n*s, n*f, d*f, and the square features n Figure 3 、d 2 、d 2 、f 2 A total of 9 features are used to form the feature matrix. Obtain the E / H value results on the central target die. Through residual analysis and comparison of the predicted value vs the actual value, the H simulation value is basically consistent with the model predicted value, and the error is basically controlled within 0.01, verifying the effectiveness of the method of the embodiment of the present invention.

[0046] In summary, the method for extracting the electromagnetic field of vertical interconnection coupling of stacked chips according to the embodiments of the present invention generates a high-precision electromagnetic field database through finite element simulation, and uses a machine learning method to extract the influencing factors of multiple electromagnetic noise sources, and predicts the electromagnetic field interference of the internal circuits of stacked die caused by the vertical interconnection electromagnetic coupling of TSV, so as to solve the problem that it is difficult to evaluate the coupling electromagnetic field interference between the die of stacked chips. Compared with the prior art, the method according to the embodiments of the present invention has the following beneficial effects:

[0047] 1) It can extract the unmeasurable electromagnetic field inside the die, with a faster speed;

[0048] 2) The model can be expanded and is applicable to the interference caused by different high-frequency couplings of TSV, with strong applicability.

[0049] The embodiments of the present invention further provide a computer device, which may be a server, and its internal structure diagram may be as shown in Figure 4 The figure. The computer device includes a processor, a memory, and a network interface connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the operation parameter data of each framework. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the steps of the method according to the embodiments of the present invention are implemented.

[0050] Those skilled in the art can understand that Figure 4 the structure shown in is only a block diagram of some structures related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0051] The embodiments of the present invention further provide a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the method according to the embodiments of the present invention are implemented.

[0052] The embodiments of the present invention further provide a computer program product, including a computer program, and when the computer program is executed by a processor, the steps of the method according to the embodiments of the present invention are implemented.

[0053] Only some exemplary embodiments of the present invention have been described by way of illustration. Without doubt, for those of ordinary skill in the art, the described embodiments can be modified in various different ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and description are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A method for extracting electromagnetic field of vertically interconnected stacked chips, characterized in that: include: Step S1, establishing a three-dimensional model of the stacked chip through finite element simulation, determining the target position of the electromagnetic field generated by the TSV vertical interconnection coupling in the stacked chip, running the simulation to obtain the electric field strength and magnetic field strength at the target position, and storing the TSV coupling structure parameters and the extracted electric field strength and magnetic field strength as a data set; Step S2, load data from the data set and extract features, use the extracted features to build a feature matrix, select multiple different types of regression models, use the grid search method to tune the hyperparameters of the selected regression models, determine the optimal hyperparameters of each regression model, take the electric field strength and magnetic field strength as the target variables, input the data set into each regression model for training, and 2 The score and MAPE were combined to obtain the best model for electric field strength and the best model for magnetic field strength; Step S3, respectively using the best model of electric field strength and the best model of magnetic field strength to predict the electric field strength and the magnetic field strength, comparing the predicted results with the electric field strength and the magnetic field strength of the target position obtained by simulation, and calculating whether the error meets the requirement. If not, return to step S2 until the error requirement is met; Step S4, inputting the TSV coupling structure parameters of the stacked chip to be tested, and using the optimal model of electric field strength and the optimal model of magnetic field strength to output the electric field strength and magnetic field strength at the target position.

2. The method according to claim 1, characterized in that TSV vertical interconnect coupling includes TSV coupling between core particles in the same layer, TSV coupling between core particles in different layers, coupling between TSV and RDL, and coupling between metal connections of TSV and RDL.

3. The method according to claim 1 or 2, characterized in that The features include original features, interactive features and square features, wherein the interactive features are features generated by combining the original features in pairs, and the square features are the squares of the original features.

4. The method according to claim 3, characterized in that The original features include the number of coupling vias, distance, frequency, TSV diameter, die thickness, RDL width and length.

5. The method according to claim 1 or 2, characterized in that: The regression models include Ridge regression, Lasso regression, random forest regression, gradient boosting regression and support vector machine.

6. The method according to claim 1 or 2, characterized in that: In step S2, the data set is divided into 5 test sets, and the model is repeatedly trained and verified under different test sets. Negative MAPE is used as the evaluation indicator for each training and verification, and the average value of 5 times is used as the final evaluation indicator of the model. The hyperparameters of the model with the smallest MAPE indicator are taken as the optimal hyperparameters.

7. The method according to claim 1 or 2, characterized in that: In step S3, the key features that affect the change of the electromagnetic field among all the features are used as input variables, and other non-key features are set to constant values, and a fixed data set is set to reduce the model overhead.

8. A computer device comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images