Spice modeling from an EM simulation environment: the use of full-wave electromagnetic modeling can simulate the behavior of a high-speed differential backplane channel and advance the system-level design process.

The operating frequency of high-speed copper backplane serial links is expected to reach 10 Gbps in the next few years. At a 10 Gbps data rate, the clock frequency is 5 GHz, equating to a period of 200 ps, which results in a signal rise time in the range of 30 to 50 ps. This rise time will influence the analog bandwidth and the highest significant frequency component both for the measurement bandwidth and the bandwidth of the channel model. To effectively design a serial link (channel) to operate effectively at this bandwidth, accurate signal integrity modeling is required.

A full wave electromagnetic (EM) simulator is used to create models of the various components (transmission lines, connectors, vias) on a high-speed serial channel. These models are then verified with laboratory measurements on a test vehicle containing backplane to daughter-card links. The link contains two high-speed digital connectors, two modules and a backplane. The frequency of operation is 0 to 10 GHz. Three-dimensional EM models are created for all the link components. The EM simulator provides S-parameters and time domain data, which can later be compared to laboratory measurements.

[FIGURE 1 OMITTED]

Time Domain Reflectometry (TDR) is used to measure the channel, and a software tool provided by the TDR manufacturer is used. This tool enhances the TDR capabilities, improves measurement accuracy and creates S-parameter data from time domain measurements. This tool also allows a designer to create SPICE models of the link components, based on the measured data. The SPICE model results can then be verified by comparing it to the measured data, and the final model assembly can be used to predict the overall system response in the circuit simulation environment.

In this paper, we will present the major stages for modeling the gigabit backplane structure used in the system-level design process. We will start from the description of the interconnect structure followed by the full-wave EM simulation of the design. The results of this simulation will be compared with the measurement data of the prototype. Finally, a topological circuit model created from the measurements will be used to predict data transmission through the backplane in terms of an eye diagram.

[FIGURE 2 OMITTED]

Electromagnetic Modeling of the Channel

The channel chosen for this analysis consists of two test modules, a backplane and a high-speed connector. The connector is the 10-row FCI AirMax connector. The commercial CST Studio Suite 2006B[TM] software is used for the simulations of the channel, at a frequency range of 0 to 10 GHz. CST Microwave Studio, a full-wave 3D EM modeler, which uses the Finite Integration Technique (FIT), is used to model the channel. The AirMax connector model in MWS is shown in FIGURE 1; the 20 conductors on the inner two columns are simulated, and waveguide ports are set up on both sides of the connectors, resulting in a 40 port simulation. The conductors on the outer two columns are terminated in 50 ohms.

[FIGURE 3 OMITTED]

[FIGURE 4 OMITTED]

A broadband multiple port simulation is best suited for the CST Microwave Studio (CST MWS) time domain solver, which is based on a hexahedral Cartesian mesh and Perfect Boundary Approximation (PBA) technique. The resultant Touchstone file has been verified against empirical data to ensure the model accuracy.

The modules consist of four layers, and the high-speed signals are located on the top and bottom layers. The high-speed traces are 12 mils wide and 1.8 in. long, with a characteristic impedance of 50 ohms. The dielectric material on the modules is N4000-6 (FR4), with electrical properties of r=4.3 and loss tangent=0.019. The backplane is 20 layers, 230 mils thick. The channel routes are all on stripline layers, and the backplane contains several layers of back drilling to minimize the via stub effect. The channel traces on the backplane are 9 mils wide, and Tx module to Rx module distance is 10 in. The dielectric material on the backplane is N400013, with r=3.9 and loss tangent=0.012.

A CST MWS Frequency Domain (FD) simulation is used to simulate the transmission lines on the modules and backplane. The CST MWS FD solver employs a tetrahedral mesh and accurately simulates dielectric high-frequency losses. The EM model results of the members of the link (two modules, two connectors and backplane), are combined in the CST Design Studio tool. Design Studio is a circuit simulator that can extract S-parameters and time domain data of individual concatenated blocks. See FIGURE 2 for a view of the Design Studio schematic for this channel.

[FIGURE 5 OMITTED]

An S-parameter simulation is performed on the entire channel at 0 to 10 GHz. In FIGURE 3, the resultant S31 (through) of the channel is illustrated and compares this to the measured data. A good correlation can be observed across the entire frequency range.

Measurement Methodology

Once the prototype is manufactured, its performance must be verified as a part of the overall system, and this can be efficiently done in the SPICE circuit simulation environment. For this purpose, the measurement-based topological modeling methodology is the most feasible candidate. Once the model is generated, a designer can look at the different parts of the circuit model and see how they will affect the final data transmission. In the following subsection, we will briefly describe the measurement-based topological modeling methodology applied for the same gigabit backplane structure.

In general, measurement-based modeling strategy follows four major steps: measure, model, verify and simulate. Although the measurements can be done in time or frequency domains (converted to time), the preference is given to the time-domain measurements because they allow for better resolution for the same measurement bandwidth. The modeling is then performed on time domain voltage or impedance data by applying different strategies for various features of the gigabit backplane. For example, relatively short and lossless discontinuities, such as connectors and vias, can be easily modeled using lumped circuit elements or adding short pieces to the ideal transmission lines, whereas long traces can be modeled using a uniform lossy transmission line model. After the model is generated, it is imperative to compare it with the measurements and adjust its components to obtain the best possible correlation. The final step is to simulate different data patterns to see if the interconnect will perform as expected.

Step 1: Time-Domain Measurements

To measure the backplane structure, a Tektronix CSA8200 sampling oscilloscope with the IConnect software from the same company was used. The time domain sampling oscilloscope generates a high-speed time domain step that propagates through the interconnect structure. The reflected and transmitted step response can be converted to the frequency domain to obtain S-parameter data such as return and insertion losses.

[FIGURE 6 OMITTED]

[FIGURE 8 OMITTED]

The measurement setup used in the analysis is shown in FIGURE 4. It consists of the sampling oscilloscope with two high bandwidth (20 GHz) sampling modules connected via SMA cables to the test cards of the backplane. Time domain reflectometry measurements allow viewing of the interconnect characteristics in time. The measured data can be displayed as impedance, reflection coefficient or voltage waveforms. FIGURE 5 shows the result of these time-domain measurements. Different regions of the backplane structure can be easily identified from the time domain measurements by disconnecting the test cards from the measured backplane structure and observing time domain reflections from the opens at different points.

When a signal enters the backplane structure, it first sees the discontinuity due to the SMA connector of the test card; this is point A in Figure 5. The SMA connector is then followed by the test card's trace labeled as the region B in the figure. After that, the signal sees a test-card-to-backplane connector and the via-to-trace discontinuities are labeled as the region C. Then the signal propagates through differential traces of the backplane structure (region D) finally approaching the receiver's test card side, where regions E and F indicate the via-connector and test-card traces, respectively.

Step 2: Modeling of Reflections and Losses

After identifying the different regions, a designer can start performing modeling. The reflections from the SMAs, high-speed connectors and vias can be modeled using lumped elements or pieces of ideal transmission lines, while the traces have to be modeled with lossy transmission lines. The next figures illustrate the general modeling procedure to model the backplane structure shown in Figure 1. FIGURE 6 shows the impedance profile-based modeling of the reflections from the SMA and high-speed connector regions. To model the structure, the impedance profile is simply subdivided into the different regions and ideal transmission lines are assigned for each region.

[FIGURE 7 OMITTED]

[FIGURE 9 OMITTED]

The test card and backplane traces are relatively long structures that exhibit lossy behavior and have to be modeled with lossy transmission lines. FIGURE 7 and FIGURE 8 show the lossy models (and their correlation with the measurements) created for the test card and backplane traces respectively. Note that the test card model was obtained from TDR data only, while the backplane's losses were obtained from TDR and TDT measurement data.

Step 3: Verification of Model's Performance

When individual components for the backplane model assembly are generated, an engineer can assemble the model and verify its performance with the measurements when the same step is applied to the generated model. When all modeling pieces that correspond to reflections and losses

for the channel are assembled together, each of the individual components can be fine tuned to archive the desired accuracy in terms of reflected and transmitted data. The general model topology and time domain correlation are shown in FIGURE 9.

Step 4: Simulation of Data Transmission

Once an accurate model is generated, it can be used to predict real signal propagation through the interconnect structure. Different data patterns can be easily synthesized with the software and their response can be plotted in a form of an eye diagram. FIGURE 10 shows the channel model's synthesized eye diagram generated for XAUI data rates. It measures 24 ps peak-to-peak jitter and 249 mV noise at 3.125 Gbs and 60 ps (20% to 80%) rise time.

[FIGURE 10 OMITTED]

Summary

We have shown the feasibility of using full-wave electromagnetic modeling to simulate the behavior of a high-speed differential backplane channel. The simulated results match well with empirical data. The measurement technique described consists of taking TDR measurements and using software to create S-parameter data and an accurate model of the channel.

ACKNOWLEDGEMENTS

Many thanks go to Martin Schauer and Antonio Ciccomancini at CST for the valuable feedback provided on the modeling of this channel. We'd also like to acknowledge the assistance of Dana Bergey and Jonathan Buck, with FCI, who provided the test vehicles used in this experiment and the model file for the AirMax connector.

Ed note: This paper was previously presented at PCB Design Conference East in Durham, NC, October 2007.

REFERENCES

(1.) J. S. Pak, J. Kim et al., "Coupling of Through Hole Signal Via to Power/ Ground Resonance and Excitation of Edge Radiation in Multi-layer PCB, IEEE Int. Electromagnetic Compatibility Symposium, Boston, MA, Aug. 2003, pp. 231-235.

(2.) CST Studio Suite 2006, cst.com

(3.) A. Ciccomancini, F. Zanella., "Broadband Signal Integrity Characterization of a High Speed Differential Backplane Pair", IEEE International Symposimum on Electromagnetic Compatibility, Portland, OR, Aug. 2006, pp. 24-28.

(4.) R. Kollipara, B. Chia, F. Lambrecht, C. Yuan et al, "Practical Design Considerations for 10 to 25 Gbs Copper Backplane Serial Links" DesignCon 2006.

EUGENE MAYEVSKIY is an applications engineer at Tektronix; [e-mail]. FABRIZIO ZANELLA is the support manager at CST of America; [e-mail].