Microwave Journal
www.microwavejournal.com/articles/33613-moving-beyond-s-parameter-files-advanced-scalable-and-3d-em-models-for-passive-devices

Moving Beyond S-Parameter Files: Advanced Scalable and 3D EM Models for Passive Devices

March 12, 2020

S-parameter data files remain the most commonly available “model” for representing passive devices in the microwave industry. Physically motivated equivalent circuit models can be set up to scale accurately with part value, substrate properties and other parameters, such as solder pad dimensions. Still, such circuit models cannot generally account for electromagnetic (EM) coupling interactions between microwave components and between components and their surrounding shielding and interconnect environment. Consequently, full-wave EM analysis has become a crucial step at RF to account for such interactions. New technology, recently available in some EM simulators, allows for encrypting geometry and material details to protect vendor manufacturing IP and enable 3D EM models to be shared with a wider design community. These advances help designers reduce design risk and re-work and improve time-to-market for today’s increasingly compact and complex product form factors.

Figure 1

Figure 1 Simulated |S11| of an AVX 0402xU 5.6 pF capacitor mounted on three substrates vs. AVX S2P data. Blue: 10 mil Rogers 4350B substrate; Green: 30 mil Rogers 4350B; Red: 60 mil Rogers 4003C. Dashed: AVX data, measured on 20 mil Rogers 4350B.

For decades, measured S-parameter data files have been the most commonly available “model” for representing passive devices of all kinds in the microwave industry. S-parameter files, while useful, ubiquitous and very portable only represent the way a specific device behaves in the test fixture environment and the test conditions used for characterization. On the other hand, properly developed physically motivated equivalent circuit models can be set up to scale accurately with part value and substrate properties, as well as other parameters like solder pad dimensions.1 This advance is a marked improvement that is used by many designers worldwide today. However, circuit simulation is not always sufficient in terms of pre-build risk management for microwave/mmWave designs that involve compact topologies and dense circuit implementations. Accordingly, full-wave 3D EM analysis has become a crucial step to account for possible EM coupling interactions between microwave components and between components and their surrounding shielding and interconnect environment. This unexpected coupling can result in performance degradation and, in turn, lead to costly and lengthened design cycles. Assembling the necessary geometry to complete full-wave analysis that includes representations of passive elements, such as packaged and surface-mount devices as well as packages and connectors, requires close collaboration between vendors and customers of vendors and model providers. In many cases, sharing of manufacturing geometry and material details is required. New technology, recently available in some simulators like ANSYS HFSS, makes it possible to encrypt manufacturing details to better enable 3D EM models to be shared with a wider design community.2,3

S-PARAMETER FILES

S-parameter files are simply measurements of a device that are very commonly used as device “models” in high-frequency design simulations. S-parameter files are good representations of a device for simulation, provided the S-parameter measurement conditions match the design conditions. The properties of the test fixture and measurement conditions used to measure the S-parameters, such as substrate type, PCB pad dimensions and bias condition should be the same in the design to achieve the most accurate simulation. It is important to note that as frequency increases, parasitic effects become more and more significant. Thus, differences in design properties in comparison to measurement properties can lead to a less accurate simulation. Figure 1 illustrates a capacitor’s variation in S-parameter response solely due to the use of different substrates (part series, capacitance and pad dimensions are the same). While it is common for passive component suppliers to provide S-parameters for their devices, at best these represent only one possible mounting/usage scenario. In addition, measurements are typically performed in a standard coaxial-connectorized test fixture in which an air gap under the device may be present.

To evaluate different part values, pad arrangements or substrates in a design using S-parameter files as simulation models, different measurement files need to be used to represent each part value. As seen in Figure 1, the downloaded S-parameter file that represents device performance on a 20-mil substrate would less accurately predict the design response if a 10-mil substrate was used, especially for an application above 1 GHz. Another limitation to consider when using S-parameter files is that the simulation is only accurate within the measurement frequency range. Outside of the measurement frequency, an S-parameter file is generally invalid and may extrapolate to non-physical values depending on the simulator setup.

S-PARAMETER FILE-BASED MODELS

A collection of S-parameter files can be packaged together into a single “S-parameter model” for added convenience. An S-parameter model represents multiple configuration options for a device or component series. For example, it can represent multiple part values of a passive component series or simulate different bias conditions for an active device, depending on the S-parameter files included. The model includes a parameter selection menu to select the properties of the S-parameter file that should be used for the simulation, all while keeping the same model element in the schematic.

S-parameter models can also be set up to work with tuning and optimization. Two S-parameter model examples are shown in Figure 2. In Figure 2a, an S-parameter model for the Mini-Circuits EP2K+ splitter illustrates how models can be simulated on two different substrates and in Figure 2b the Gowanda C225FL conical inductor series model shows how multiple part values can be simulated without replacing the model.

Data-based or equivalent circuit models typically available often lack detailed information concerning how the measurements and/or models were developed, as well as usage information. We believe that such information is very important to designers. A model datasheet represents one way that this information can be presented. Such datasheets contain critical information concerning the model, including validation measurements along with details regarding reference planes, application/usage and more. Modelithics provides this type of detailed information for all models distributed and available for use by various circuit and system designers. The information varies based on the type of device or component represented by the datasheet; for example, different information is included in the model datasheets for the scalable equivalent circuit and 3D-geometry-based EM models discussed next.

Figure 2

Figure 2 S-parameter models with dropdown menus to select the substrate (a) or part value (b).

Figure 3

Figure 3 |S11| (a) and |S21| (b) from a scalable equivalent circuit model for an AVX 0402xU 5.6 pF capacitor on three different substrates vs. AVX S2P data. Blue: 10 mil Rogers 4350B; Green: 30 mil Rogers 4350B; Red: 60 mil Rogers 4003C; Dashed: AVX data, measured on 20 mil Rogers 4350B.

SCALABLE EQUIVALENT CIRCUIT “GLOBAL” MODELS

A scalable equivalent circuit model represents a much more efficient and powerful type of component simulation model. These models are designed and validated using S-parameter data, and oftentimes other measurements, to build a comprehensive model for a device series that scales continuously over part value ranges and other design parameters. Modelithics uses the term “Microwave Global Models™” to represent its unique approach to scalable equivalent circuit models for microwave devices. In addition to part value, these scalable global models are set up to scale with pad dimensions, substrate properties (e.g., height and dielectric constant), and if applicable, temperature, bias and other properties. The unique and advanced scaling capabilities give them numerous advantages over previously discussed file-based models.

Compared to file-based S-parameter models that represent only a set number of simulation conditions, scalable equivalent circuit models represent an extremely broad range of possibilities in terms of design properties. These models accurately simulate broadband device and parasitic behavior for many embedding configurations. In general, since the models are based on a physically meaningful equivalent circuit topology, they remain physically valid when extrapolating and interpolating parameters. That is, equivalent circuit models will exhibit physically meaningful and reasonably correct behavior even beyond the measurement frequency range. Figure 3 corresponds to Figure 1, but with the frequency sweep extended to 30 GHz. The simulated S-parameters from the scalable equivalent circuit model reveal a continuation of physical behavior on each substrate for the full frequency sweep (and beyond). The S-parameter file, which has a maximum frequency of 10 GHz, extrapolates as a straight line and is not valid above 10 GHz.

To illustrate the concept of a physically motivated equivalent circuit model, Figure 4 depicts a surface-mount inductor mounted on a microstrip substrate. Also shown is a simple equivalent circuit that can be fit to the measured S-parameters of the part up to and including the first self-resonance. Scalable Microwave Global models can be likened to multi-faceted extensions of this approach in which additional elements are added to the circuit to represent higher-frequency and higher-order-resonant behavior. Complex equation sets are incorporated to add the scalability features discussed. In addition, the resistive loss is typically fit to effective-series-resistance measurements made with separate equipment, such as an impedance analyzer or resonant line.

Compatibility with statistical analysis evaluations is another advantage associated with scalable equivalent circuit models. Component tolerance values can be set, making it possible to perform a full design yield analysis.4 If the yield is unacceptable, one can simply test different part tolerances until the most efficient and acceptable design is achieved. With suitably arranged scalable models, this design “testing” is executed quickly at the simulation stage, thereby saving time and cost incurred with extra board runs and redesigns.

Figure 4

Figure 4 Surface-mount inductor parasitics (a) and first-order, physically motivated equivalent circuit model (b).

Figure 5

Figure 5 Three capacitors in a shunt configuration on a 50 Ω line: 10 mil spacing (a), 42 mil spacing (b) and 74 mil spacing (c) between them.



3D EM MODELS

Typically, for lower-frequency designs in which surface-mount passive components and discrete or MMIC active devices are not packed into a compact layout, it is sufficient to employ circuit simulation for equivalent circuit models combined with planar EM simulation of the layout. Further facilitating these types of simulations is the availability of useful simulation features in the equivalent circuit models, such as advanced pad treatment that allows for the de-embedding of solder pads. Thanks to such features, the result is greater ease of use and accuracy when executing EM co-simulations.

Figure 6

Figure 6 Circuit (a) and 3D EM (b) simulations vs. measured data for the capacitor arrays of Figure 5. Red: 10 mils; Blue: 42 mils; Green: 74 mils.

However, once components are placed in close proximity to each other and design frequencies are extended higher, this approach no longer captures all of the interactions present after the design is fabricated. This necessitates the use of 3D EM simulation to capture the coupling between components and their environment. Compounding this, the source of this coupling can be difficult to pinpoint without using 3D EM tools to visualize fields, whether it is component-to-component coupling or components coupling to adjacent lines or shielding. Fortunately, full-wave 3D EM analysis is a proven method with years of development leading to excellent prediction of measurement results, as illustrated in Figures 5, 6 and 7. In Figure 5, three capacitors are shown placed in a shunt configuration on a 50-ohm line. The capacitor model is a 12-pF Johanson R14S in an 0603 case size (CAP-JOH-0603-101). The substrate is 16-mil Rogers 4003C. Three configurations are illustrated with 10, 42 and 74 mil spacings between capacitors.

Figure 7

Figure 7 E-field visualization of the closest spacing shunt capacitors showing component-to-component coupling.

3D SIMULATION: COMMON ISSUES

Obtaining Device Information

Figure 8

Figure 8 3D simulation of a 10 nH Coilcraft IND-CLC-0603-101 inductor mounted on 10 mil Rogers 5880, comparing lumped and wave ports.

One of the main problems associated with using 3D EM simulations in a design flow lies in obtaining the necessary physical and material properties to enable analysis of the complete circuit. This information is typically proprietary manufacturer intellectual property. Even with all the physical parameters and material properties available to the designer, initial 3D model results may not be good when compared to measured data. Therefore, tuning of the model is often required. Model-to-measurement discrepancies are likely due to fabrication tolerances in the component or layout or uncertainty in material properties. For this reason, measurement validation of EM models and EM simulation expertise is crucial for the designer. Using a library of validated 3D component models, such as Modelithics’ COMPLETE+3D Library, is one way to mitigate some of the issues mentioned. These sorts of libraries are currently limited in terms of total model count but are continuing to grow.

Port Selection and Setup

Assuming the 3D component models are accurate, designers still need to integrate their layouts into the 3D environment and find the most appropriate way to excite the circuit with ports. While many port types exist in HFSS, lumped or wave ports generally provide similar results and are most appropriate for exciting a PCB-based 3D structure (see Figure 8). Different port setups have their own advantages, depending on the layout itself and whether the ability to shift reference planes is needed. If the reference planes of the simulation are internal to the model and do not need to be shifted, lumped ports are simple and quick to set up. If the ports are external to the model or if the reference planes need to shift, wave ports may be a more appropriate choice (see Figure 9).

Solution Setup

Figure 9

Figure 9 Simulation geometries of the 10 nH Coilcraft IND-CLC-0603-101 inductor mounted on 10 mil Rogers 5880 using lumped (a) and wave (b) ports.

Figure 10

Figure 10 MMIC amplifier mounted in a 4 mm QFN package used to illustrate 3D co-simulation.

Figure 11

Figure 11 E-field visualization of the package and bond wire on an alumina motherboard, showing coupling between the bond wire and the adjacent package pin.

HFSS offers guidance in their help documentation on selecting an appropriate solution frequency. For example, in a resonant structure, the solution frequency should be set to the design frequency for that structure. The reason to choose the resonant frequency as the solution frequency is that the adaptive meshing process used by HFSS analyzes the mesh for large field differences between adjacent mesh elements. If the fields or transmission for a structure are very low at the chosen solution frequency, there may not be much difference between adjacent elements and the mesh may converge prematurely. At a resonant frequency, the fields are strongest, so the adaptive mesh process is likely to converge on an accurate mesh for that structure.

For broadband structures or structures in which the resonant frequencies are unknown, selecting the solution frequency is less straightforward. The user could hypothetically choose any frequency in the band of interest. The approximate frequency of first resonance could be a good initial selection if this value is known. The user should investigate the effect of selecting different solution frequencies on the results, as well as the maximum number of passes and maximum delta, to converge on an appropriate mesh and ensure a good result.

Imported Geometry Issues

If the designer is bringing the layout or a specific component model into the HFSS 3D environment from another tool, initial difficulties can arise when integrating these elements due to geometry issues. For example, non-manifold geometry errors that manifest once meshing is attempted are commonly encountered with imported computer-aided design (CAD) files. Other examples of commonly encountered errors are non-manifold vertices and edges. While the “Heal” function in HFSS may be able to resolve some of these issues, re-drawing the geometry within HFSS may still be required.

Tolerances in Model Geometry and Material Properties

Manufacturing tolerances on physical dimensions or material properties can cause discrepancies in simulation results, especially when comparing against measurement data. If a fabricated version of the structure is available, a detailed inspection of the actual dimensions can be useful to help identify possible causes for the differences between simulation and measurement. Unfortunately, material property variation can be harder to isolate and nearly impossible to identify by physical inspection. In such a case, the 3D model can be used to study the effect of material property variation. Tolerances can be added to the 3D model and a parametric analysis (or analytical derivatives) can be executed to study changes in performance.

An example of combining 3D EM analysis with S-parameter file data for completing a simulation is illustrated in Figures 10 and 11. Shown is a MMIC amplifier mounted in a 4-mm QFN package. Package, bond wires and alumina motherboard were simulated in HFSS with ports added to enable co-simulation. Measured S-parameters of MMIC amplifier were connected to the ports in EM simulation. 3D co-simulation of the structure was performed using both S-parameter data and 3D EM simulation data yielding good agreement with measurement, as shown in Figure 12. Also shown for reference is non-packaged MMIC performance available from a measured S2P file.

CONCLUSION

When moving beyond traditional S-parameter file representations for passive components, designers are provided with many advantages and conveniences. A next step involves well-documented S-parameter file-based models. Such models feature part value pull-down menus, for example. A physically motivated and scalable equivalent circuit model can provide for validity under extrapolation, as well as interpolation between measured part values or test conditions. Such models can also enable a more accurate representation of loss and can enable tuning, optimization and tolerance analyses. 3D EM component models address coupling and interactions and can be used in the final stages of the design process to verify that EM interactions will not cause a failure when fabrication is complete. A well-balanced design flow takes advantage of the different strengths of circuit simulation scalable equivalent circuit models and EM analysis, provided the necessary circuit and 3D EM models are available and use S-parameter file-based models and S2P files when that is the best available representation.

ACKNOWLEDGMENTS

We would like to thank Laura Levesque for her contributions. In addition, the very helpful technical collaborations with our electronic design automation (EDA) software partners in supporting Modelithics advancement and distribution of a wide range of passive model types is gratefully acknowledged. For this work, we thank Keysight Technologies and ANSYS Corporation for providing software and technical support.

Figure 12

Figure 12 3D co-simulation vs. measured performance of the MMIC amplifier mounted in an RJR 4 mm QFN package on an alumina motherboard and connected with bond wires: |S11| (a) and |S21| (b). Red: co-simulation; Blue: measurement; Green: measurement of the MMIC mounted on a carrier with no package or bond wire effects.

References

  1. L. Dunleavy, “Understanding S-Parameter vs. Equivalent Circuit-Based Models for Surface Mount RLC Devices,” Institution of Engineering and Technology (IET) 3rd Annual Seminar on Passive RF & Components, March 2012.
  2. M. Commons, “Introducing HFSS 3D Components” IEEE International Microwave Symposium Micro Applications, May 2015.
  3. W. Sun, “Accurate EM Simulation of SMT Components in RF Designs,” IEEE Radio Frequency Integrated Circuits Symposium (RFIC), June 2017, pp. 140143.
  4. L. Dunleavy and L. van der Klooster, “Improve Microwave Circuit Design Flow Through Passive Model Yield and Sensitivity Analysis,” IEEE International Microwave Symposium Micro Applications, June 2012.