A SIP multi-cavity coupled-substrate resonance suppression structure and its design method

Abstract

This invention discloses a design method for a SiP (Single-Cavity In-Package) multi-cavity coupling resonance suppression structure with a substrate, belonging to the field of electromagnetic wave suppression technology. The EBG (Electromagnetic Embedded Gauge) structure embedded in the substrate stack can effectively suppress coupling resonance between the cavities and the substrate. Based on this, and using an equivalent transmission line model of the embedded EBG structure, a method is proposed to construct the inverse relationship between the stopband characteristics and structural parameters of the EBG structure using a neural network. This method can accurately predict the required EBG structure parameters while meeting the target stopband requirements. Simulation results show that this method can significantly improve the design efficiency of the coupling resonance suppression structure between the SiP multi-cavity in-pack and the ceramic substrate, providing strong support for the electromagnetic compatibility design of SiP.

Description

Technical Field

[0001] This invention relates to the field of electromagnetic wave suppression technology, and in particular to a method for calculating electromagnetic bandgap structural parameters. Background Technology

[0002] With increasing operating frequencies and integration density, electromagnetic interference (EMI) between multiple chips within a system-in-package (SIP) is becoming increasingly serious. To effectively suppress EMI, metal multi-cavity shielding is commonly used in microwave and radio frequency applications to isolate multiple chips. These shielded multi-cavities are primarily composed of metal wiring layers on the interconnect substrate of the metal package cover. However, because the metal interconnects in the substrate wiring layers are not complete, the shielded multi-cavities of the SIP are not completely sealed, allowing interfering electromagnetic waves generated by radiation sources to leak into the substrate through substrate gaps. The dielectric layer of the substrate, together with the upper and lower metal wiring layers, forms a parallel-plate waveguide-like structure, which not only supports the propagation of TEM mode waves but may also induce parallel-plate resonance. The small size of the ceramic substrate, the high dielectric constant of the dielectric material, and edge effects make the internal resonance characteristics of the substrate extremely complex and difficult to predict accurately. Existing research typically uses methods for calculating the resonant frequency of the substrate analogous to those used for metal resonant cavities, but this analogy cannot accurately reflect the actual resonance characteristics of the substrate. When electromagnetic waves generate a high-intensity electromagnetic field at the resonant frequency of the substrate, these electromagnetic waves can further couple into the shielded multi-cavity of the SIP through the gaps in the substrate's metal wiring layer, thereby interfering with the sensitive components inside the cavity.

[0003] Current research embeds electromagnetic bandgap (EBG) structures into the laminated structure of printed circuit boards (PCBs) to suppress switching noise in PCB power distribution networks by utilizing the unique stopband characteristics of EBG structures. However, there is limited research on embedding EBG structures into SiP ceramic substrate structures to suppress radiated interference electromagnetic waves leaking through substrate gaps. Furthermore, the stopband characteristics of EBG structures embedded in substrate laminates are closely related to factors such as the size of the metal patch, the thickness of the laminate, and the dielectric constant of the dielectric material. Therefore, existing research typically relies on multi-objective optimization algorithms when designing EBG structures to achieve specific stopband characteristics. However, these optimization algorithms suffer from several drawbacks in practical applications, including high computational resource consumption, time-consuming iteration processes, slow convergence speeds, and difficulty in finding the global optimum within a limited time. Moreover, when the design model or objective parameters change, the algorithm needs to iterate again, lacking flexibility and adaptability. In addition, the fitness function for constructing the optimization algorithm depends on finite element parameterization calculations, which undoubtedly increases the difficulty of rapid EBG structure design. Therefore, how to overcome the limitations of current design methods while ensuring design efficiency, and propose a new method for rapidly designing SIP multi-cavity coupling resonance suppression structures with substrates, has become a key problem that urgently needs to be solved. Summary of the Invention

[0004] This invention addresses the aforementioned problem by analyzing the coupling resonance phenomenon between the SIP multi-cavity structure and the substrate, and proposes to suppress this resonance phenomenon by embedding an EBG structure within the substrate. Based on this, an equivalent circuit model of the embedded EBG structure is used to obtain a large number of correspondences between the EBG stopband characteristics and their structural parameters. These data are then used as the input and output of a neural network algorithm. Through the algorithm's learning methods, a surrogate model between the EBG stopband characteristics and their structural parameters is established, and this model is applied to achieve rapid design for any target stopband requirement. Finally, application examples demonstrate that the method of this invention can significantly improve the design efficiency of coupling resonance suppression structures between the SIP multi-cavity structure and the ceramic substrate, providing strong support for the electromagnetic compatibility design of SIPs.

[0005] The technical solution of this invention is: a SIP multi-cavity coupling resonance suppression structure, which is an EBG structure. The EBG structure is embedded between two substrate layers and consists of periodically arranged square metal patches. The center of each square metal patch is connected to the bottom ceramic substrate via a metal pillar. The space between the square metal patches and the upper substrate is filled with a dielectric material ε. r1 The square metal patch is filled with dielectric material ε between itself and the underlying substrate. r2 ; Dielectric material ε r1 With dielectric material ε r2 The dielectric constant of all of them is that of ceramic.

[0006] A design method for a SiP (Solid In-Package) multi-cavity substrate-coupled resonance suppression structure is disclosed. In this method, the metal patch in the EBG (Electronic Embedded Frame) structure is equivalent to a transmission line with a unit size of 1 / 2. The equivalent circuit of the EBG includes: a first transmission line, a second transmission line, a third transmission line, a fourth transmission line, a first inductor, a second inductor, a third inductor, a first capacitor, and a second capacitor. The first transmission line is sequentially connected to the first inductor, the second inductor, and the second transmission line. The connection point of the first inductor and the second inductor is connected to one end of the first capacitor. The other end of the first capacitor is connected to one end of both the third inductor and the second capacitor. The other ends of the third inductor and the second capacitor are shared, and their common nodes are connected to one end of the third transmission line and the fourth transmission line. The lengths of the first, second, third, and fourth transmission lines are l / 2, where l is the side length of the square metal patch. The sizes of the first and second inductors are L, and the sizes of the first and second capacitors are respectively... , The magnitude of the third inductor is ;

[0007] Step 1: Calculate the equivalent parameters of the EBG structure using the following method.

[0008] ;

[0009] ;

[0010] ;

[0011] ;

[0012] ;

[0013] Where, ε r1 and ε r2 These are the dielectric constants of the substrate dielectric material, v and v. d h1 and h2 are the diameters of the connecting column, respectively; and μ0 and ε0 are the permeability and dielectric constant of air, respectively.

[0014] Step 2: Set the parameters of the EBG structure to h1, h2, ε r1 ε r2 , l, v d Train a generalized regression neural network to establish the mapping relationship between the EBG stopband characteristics and the design parameters, that is, to inversely deduce the corresponding EBG structure design parameters by using the known EBG stopband characteristics.

[0015] Step 3: Design the EBG structure using a trained generalized regression neural network.

[0016] Furthermore, the specific method for step 2 is as follows:

[0017] Step 2.1: Generate input parameters (h1, h2, ε) using the Sobol sequence. r1 , ε r2 , l, v d (sample);

[0018] Step 2.2: Calculate the S-parameters of the sample using the equivalent circuit. 21 Frequency response and stopband range are extracted; the extracted stopband characteristics are correlated with sample parameters, and the data is normalized. The normalized data is then divided into training set, validation set and test set.

[0019] Step 2.3: Use the stopband characteristics as input data for the neural network and the EBG parameters as output results for the generalized regression neural network to train the generalized regression neural network.

[0020] The EBG structure embedded in the substrate stack of this invention can effectively suppress coupling resonance between the cavity and the substrate. Based on this, and using an equivalent transmission line model of the embedded EBG structure, a method is proposed to construct the inverse relationship between the stopband characteristics and structural parameters of the EBG structure using a neural network. This method can accurately predict the required EBG structure parameters while meeting the target stopband requirements. Simulation results show that this method can significantly improve the design efficiency of the coupling resonance suppression structure between the multi-cavity SiP and the ceramic substrate, providing strong support for the electromagnetic compatibility design of SiP. Attached Figure Description

[0021] Figure 1 SIP resonance analysis is performed when the substrate is seamless; where (a) is a single cavity; (b) is a multi-cavity cavity; and (c) is an observation point P within a single cavity. i The electric field intensity versus frequency curve; (d) Observation point P inside the multi-cavity chamber i The curve showing the change of electric field intensity with frequency.

[0022] Figure 2 SIP resonance analysis with gaps in the substrate; (a) Schematic diagram of multi-cavity SIP package structure; (b) Leakage path of radiated electromagnetic field; (c) Observation point P i The curve showing the change of electric field intensity with frequency.

[0023] Figure 3 A schematic diagram of the design of an EBG structure embedded in the first and second layers of a SIP ceramic substrate; (a) EBG periodic arrangement structure; (b) equivalent circuit model of the embedded EBG structure.

[0024] Figure 4 It is a SIP package structure with an embedded EBG structure.

[0025] Figure 5 The curve showing the electric field intensity as a function of frequency at observation point P1 of a SIP package structure with an embedded EBG structure.

[0026] Figure 6 Simulation results of the equivalent circuit model for embedding EBG structure on SIP substrate.

[0027] Figure 7 This is a flowchart of the neural network model training process.

[0028] Figure 8 The determination coefficients R for the two neural network models 2 .

[0029] Figure 9 A comparison of the errors in predicting EBG parameters using GRNN.

[0030] Figure 10Finite element simulation results for designing a SIP with a resonance suppression structure based on the GRNN model. Detailed Implementation

[0031] To investigate the EMI problem caused by intrinsic resonance in SIP structures, a system was established as follows: Figure 1 The analytical models shown in (a) and (b) depict a substrate without wiring gaps with a complete metal layer, interconnected with a SIP metal cap to form a closed metal cavity. A dipole radiation source is placed at P0 to simulate chip radiation, and electromagnetic resonances at the locations of sensitive elements at P1, P2, and P3 are observed. The results are first calculated according to equation (1). Figure 1 (a) The resonant frequency of the large single-cavity chamber is 5.26 GHz. Figure 1 (b) shows that the resonant frequency of the four equally sized cavities is 10.5 GHz, which is consistent with the simulation results. It can be concluded that when the substrate surface has a complete metal layer, the method for calculating the resonant frequency of the SIP enclosed cavity is consistent with the method for calculating the resonant frequency of a metal cavity. Figure 2 (a) and (b) show actual SIP structure diagrams, where substrate wiring creates gaps in the metal layers interconnected with the cavity, allowing radiated electromagnetic waves from interference sources to enter the ceramic substrate through these gaps and resonate. Figure 2 (c) Shows the resonance curves of the cavities containing the three sensitive devices when the substrate gap exists, for comparison. Figure 1 (d) It can be seen that multiple frequencies resonate within the cavity within the observation frequency band, and the resonant frequencies at the observation points are consistent but the electric field intensities are different. Since the resonant frequency of the small cavity is 10.5 GHz, the resonance of the SIP cavity at this time is mainly dominated by the substrate.

[0032] (1)

[0033] Where μ0 and ε0 are the permeability and dielectric constant of air, respectively, m, n, and p are the mode numbers, and a, b, and d are the length, width, and height of the cavity, respectively.

[0034] SIP ceramic substrate resonance suppression

[0035] Resonance suppression on the ceramic substrate is achieved by embedding an EBG structure. As the previous analysis showed, the presence of substrate gaps causes coupling resonance between the multi-cavity SiP and the substrate, significantly reducing the shielding effectiveness of the SiP cavities. This is mainly due to the complex parallel-plate resonance effect formed by the first dielectric layer of the SiP ceramic substrate and the upper and lower metal layers. Therefore, resonance suppression is achieved by embedding an EBG structure in the one- or two-layer substrate stack-up structure, such as... Figure 3 As shown. To verify the suppression effect of EBG on the coupling resonance between the SIP multi-cavity and the ceramic substrate, an EBG structure was embedded in the above SIP model for analysis, as shown. Figure 4As shown. Maintain the distance between the metal patch and the ground plane h2 = 0.15 mm, the period length l = 3.5 mm, and the diameter V of the via connecting the metal patch and the ground plane. d =0.3mm. From Figure 5 The results show that the resonance phenomenon disappears within a certain observation frequency range in the cavity containing the sensitive component P1. This demonstrates that the embedded EBG structure has a significant effect on suppressing the coupling resonance between the SIP multi-cavity and the substrate. Furthermore, because the EBG structure can only achieve electromagnetic suppression within a specific frequency range, its stopband range should be matched to the operating frequency of the sensitive component when designing the EBG structure.

[0036] Equivalent circuit model of substrate embedded EBG structure

[0037] Currently, the stopband characteristics of EBG structures are mainly studied using full-wave simulation software such as HFSS. To achieve rapid simulation, an equivalent circuit model was established, and high-precision simulation results were obtained. The periodically arranged metal patch units embedded in the EBG structure are equivalent to transmission lines with lengths of half the unit size at both ends and an inductor L, connected in parallel with a capacitor C1 between the first metal layer and the EBG unit metal patch, a capacitor C2 between the metal patch and the third layer of the substrate, and an inductor L1 connecting the metal patch to the post. Figure 3 As shown in (b). The parameters of the equivalent circuit elements are closely related to the parameters of the embedded EBG structure, and their expressions are shown in equations (2) to (6). The transfer matrix can be obtained from the equivalent circuit model of the embedded EBG unit. Finally, the S of the embedded EBG structure can be obtained by cascading the equivalent circuits of the units. 21 The change of parameters with frequency is used to obtain its stopband characteristics. Figure 6 Given Figure 4 The simulation results of the equivalent circuit of the EBG structure are compared. Figure 5 Finite element simulations show that the difference in the suppression bandwidth obtained by the two simulation methods is small, with the equivalent circuit model showing that S 21 The frequency band below -30dB is defined as the stopband range.

[0038] (2);

[0039] (3);

[0040] (4);

[0041] (5);

[0042] (6);

[0043] Where, ε r1and ε r2 These are the dielectric constants of the substrate dielectric material; l is the size of the periodically arranged unit metal patches; v d h1 is the diameter of the connecting post (via diameter); h1 and h2 are the thicknesses of the two dielectric layers, respectively. From expressions (2) to (6), it can be concluded that the stopband characteristics of the embedded EBG structure are affected by multiple parameters. The optimized combination of these parameters enables the embedded EBG structure to effectively suppress electromagnetic wave propagation within a specific frequency range, meeting the electromagnetic compatibility requirements of different application scenarios.

[0044] EBG structure inverse optimization based on generalized neural network

[0045] Both finite element simulation and equivalent circuit models embedded in EBG structures obtain their stopband properties through EBG structure simulation. For engineering applications, this requires extensive simulation calculations and makes it difficult to obtain accurate structural parameters. Equivalent circuit models, on the other hand, can accurately predict the stopband characteristics of embedded EBG structures and offer faster computation. Therefore, by using the equivalent model to obtain the correspondence between a large number of EBG structure parameters and their stopband characteristics, and then using neural network algorithms to establish an inverse calculation model—that is, inputting the stopband characteristics—the design parameters of the EBG structure can be obtained. Neural networks, as powerful nonlinear modeling tools, perform exceptionally well in high-dimensional complex problems. Among them, Generalized Regressive Neural Networks (GRNN) and Multilayer Perceptrons (MLP) are two commonly used models; this invention will compare the performance of these two methods.

[0046] The goal of neural network training is to establish a mapping relationship between the stopband characteristics of EBG and design parameters, that is, to inversely deduce the corresponding EBG structure design parameters using known EBG stopband characteristics. In model training, input parameters (h1, h2, ε) are first generated using a Sobol sequence. r1 , ε r2 , l, v d The Sobol sequence is a method for generating efficient, uniformly distributed numerical sequences. Then, the S-parameters of the sample are calculated using an equivalent circuit. 21Frequency response and stopband range extraction; the extracted stopband characteristics are mapped to sample parameters, and the data is normalized. The normalized data is divided into training, validation, and test sets to evaluate the model's generalization ability; the stopband characteristics are used as input data for the neural network, and the EBG design parameters are used as the output results; Spread is an important parameter in GRNN that controls the kernel function width, determining the sensitivity of GRNN to input samples. Therefore, the spread parameter is optimized on the validation set to select the optimal spread value that minimizes the mean squared error (MSE). MLP affects model performance through the number of hidden layers; changing the number of neurons in the hidden layers can adjust the model's accuracy; the trained surrogate model is used to predict the test set data to obtain the design parameters; finally, performance metrics for evaluating model accuracy and generalization ability, including MSE and the coefficient of determination (R²), are used. The training process is as follows: Figure 7 As shown.

[0047] To verify the effectiveness of the above method, the extension... Figure 4 The EBG structural parameters in the SIP model mainly include six design parameters embedded in the EBG: h1, h2, ε. r1 ε r2 , l, v d Importantly, smaller patch sizes correspond to higher upper cutoff frequencies in the stopband. This means that smaller patch sizes are needed to achieve the stopband when interfering electromagnetic waves are in the high-frequency region, while larger patch sizes correspond to lower upper cutoff frequencies. This study covers a frequency range of 1-10 GHz. Therefore, the substrate dielectric layer thickness h is set to 0.15-0.5 mm; the relative permittivity of the ceramic dielectric varies from 5-10; the size of the metal patch is defined as 1.5-5 mm; and the grounding hole diameter is set to 0.1-0.3 mm, ensuring a minimum stopband width of 1 GHz. Subsequently, a Sobol sequence generator was used to generate N=10000 sample points for neural network training. Table 1 and... Figure 8 The performance metrics of the two models are compared, and the results show that GRNN's MSE and R... 2 All of these are superior to MLP. Considering prediction accuracy, generalization ability, and model complexity, GRNN, which has a more stable performance, was ultimately chosen as the prediction model.

[0048] Table 1. Mean Square Error (MSE) of the Two Neural Network Models

[0049]

[0050] To verify the accuracy and efficiency of the model, new, realistic EBG structure parameters are first given, and the stopband range is calculated using an equivalent circuit model. Then, the stopband range is input into a pre-trained GRNN model. The GRNN model predicts and outputs the EBG structure parameters. The predicted parameters are compared with the actual given parameters. The comparison results are as follows: Figure 9 As shown, the differences between each parameter are relatively small.

[0051] In practical SIP structures, the dielectric constant of the dielectric material is generally fixed, for example... Figure 4 The substrate of the SIP in this example uses a ceramic material with a dielectric constant of 9.8. Therefore, the parameter ε is fixed. r1 ε r2 The DRNN model is obtained by training other parameters. Figure 4 The operating frequencies of all components within the demonstrated SIP are between 4-6 GHz. Therefore, the target stopband range of 4-6 GHz was input into a pre-trained GRNN. The EBG structure predicted by the GRNN was embedded into the SIP substrate, and a corresponding model was built and analyzed using HFSS software. The results are as follows. Figure 10 As shown in the figure, the predicted structure can meet the target stopband requirements.

[0052] Furthermore, in practical SiP applications, the dielectric thickness of the ceramic substrate is typically determined by the lamination process, and the dielectric constant of the material is always 9.8. Therefore, when using the GRNN model, only the dimensions of the metal sheet and grounding via need to be output. The GRNN model is then retrained using the given data. For example... Figure 4 As shown, the operating frequencies of all components within the SiP are between 4 and 6 GHz. Therefore, the target stopband range of 4-6 GHz is input into a pre-trained GRNN, and the EBG structure predicted by the GRNN is embedded into the SiP substrate. Then, a corresponding model is built and analyzed in HFSS software. The results are as follows. Figure 10 As shown in the figure, the predicted EBG structure achieved the desired target stopband.

[0053] The main results are as follows:

[0054] (1) Analysis revealed that the presence of wiring gaps on the substrate caused coupling resonance between the SIP multi-cavity and the substrate, which significantly reduced the shielding effectiveness of the multi-cavity.

[0055] (2) Based on the equivalent transmission line model of the embedded EBG structure, a method is proposed to establish the inverse relationship between the stopband characteristics of EBG and the structural parameters through neural network learning;

[0056] (3) Application examples have verified the effectiveness of the method, proving that it can quickly design a structure that effectively suppresses the coupling resonance between the SIP multi-cavity and the substrate.

Claims

1. A SiP (Solid-In-Package) multi-cavity coupling resonance suppression structure, wherein the structure is an EBG (Embedded EBG) structure, which is used to suppress the resonance phenomenon of the SiP multi-cavity. The EBG structure is disposed between two substrate layers and consists of periodically arranged square metal patches. The center of each square metal patch is connected to the bottom ceramic substrate through a metal pillar. The space between the square metal patches and the upper substrate is filled with a dielectric material ε. r1 The square metal patch is filled with dielectric material ε between itself and the underlying substrate. r2 ; Dielectric material ε r1 With dielectric material ε r2 The dielectric constant of all of them is that of ceramics; The design method for the SIP multi-cavity and substrate coupling resonance suppression structure is as follows: In this method, the metal patch in the EBG structure is equivalent to a transmission line of 1 / 2 unit size. The equivalent circuit of the EBG includes: A first transmission line, a second transmission line, a third transmission line, a fourth transmission line, a first inductor, a second inductor, a third inductor, a first capacitor, and a second capacitor are connected sequentially. The first transmission line connects to the first inductor, the second inductor, and the second transmission line. The connection point of the first inductor and the second inductor is connected to one end of the first capacitor. The other end of the first capacitor is connected to one end of both the third inductor and the second capacitor. The other ends of the third inductor and the second capacitor are shared, and their common connection point is connected to one end of both the third and fourth transmission lines. The length of the first, second, third, and fourth transmission lines is l / 2, where l is the side length of the square metal patch. The size of the first and second inductors is L, and the sizes of the first and second capacitors are respectively... , The magnitude of the third inductor is ; Step 1: Calculate the equivalent parameters of the EBG structure using the following method. ； ； ； ； ； Where, ε r1 and ε r2 These are the dielectric constants of the substrate dielectric material, v and v. d h1 and h2 are the diameters of the connecting column, respectively; and μ0 and ε0 are the permeability and dielectric constant of air, respectively. Step 2: Set the parameters of the EBG structure to h1, h2, ε r1 ε r2 , l, v d A generalized regression neural network is trained to establish the mapping relationship between the EBG stopband characteristics and the design parameters, that is, to inversely deduce the corresponding EBG structure design parameters through the known EBG stopband characteristics. Step 3: Design the EBG structure using a trained generalized regression neural network.

2. The SIP multi-cavity and substrate coupling resonance suppression structure as described in claim 1, characterized in that, The specific method for step 2 is as follows: Step 2.1: Generate input parameters (h1, h2, ε) using the Sobol sequence. r1 ,ε r2 ,l,v d (sample); Step 2.2: Calculate the S-parameters of the sample using the equivalent circuit. 21 Frequency response and stopband range are extracted; the extracted stopband characteristics are correlated with sample parameters, and the data is normalized. The normalized data is then divided into training set, validation set and test set. Step 2.3: Use the stopband characteristics as input data for the neural network and the EBG parameters as output results for the generalized regression neural network to train the generalized regression neural network.

Images