Application Note

AA Hybrid Measurement and Field SolverHybrid Measurement and Field SolverApproach for the Design of High-Approach for the Design of High-Performance SMAPerformance SMA TTransitions ransitions

This application note was presented at DesignCon2004, Santa Clara, CA

AbstractAs the cost of bringing complex high-performance sys-tems to market increases and the time to marketdecreases, new methods must be used to insure first-pass design success. This paper describes a verifi-able design method that establishes excellent signalintegrity and enables leading-edge performance withcommon off-the-shelf materials.

By integrating field solver and measurement-basedmodeling methods, the net process is verifiable. Theoptimum low-cost production design can be attained inthe shortest time. This process yields unmatchedtime-domain and frequency-domain performance usingoff-the-shelf, generic, non-esoteric, low-cost materials.

IntroductionAs the cost of bringing complex high-performance sys-tems to market increases and the time to marketdecreases, new methods must be used to insure firstpass design success. In the area of high-performanceinterconnects, with speeds from 3.125 Gbps to greaterthan 12 Gbps, any one tool alone practically guaran-tees costly pre-production iteration cycles. To improvetime to market, and guarantee first pass success, amore sophisticated approach is needed.

This paper describes such an approach. It offers adesign method that establishes excellent signal integri-ty at the lowest reasonable cost and that is verifiable.By approaching the design with a coordinatedapproach of field solver and measurement-basedmodeling methods, real and theoretical results coordi-nate, yielding the highest possible performance in theshortest time. This process yields unmatched qualityof results, using ordinary materials and processes,without the risk of costly pre-production iterations, orproduction failures.

A verifiable design method is presented that, in sym-phony, integrates model development from both circuitphysical parameters and material properties (fieldsolvers) and measurement-based methods (VNA,TDR, TDT, etc.). This paper shows that eachapproach complements the net process. For example,

models developed using field solver approaches pro-vide an initial starting point for a new design.Measurement-based methods, using time or frequencydomain techniques, can be used to develop a topologi-cal or behavioral model that can be used for fieldsolver verification and correction. Finally, the confi-dence gained through verified full-wave and quasi-stat-ic computational results allows a designer to improvenot only on the structural design parameters of acircuit but also the accuracy of the measurement tech-niques used for future verification. This hybridMeasure-Model-Simulate-Verify approach to high-performance interconnect design maintains the integri-ty of the process throughout.

A simple interconnect problem will be used in thepaper to investigate the launch characteristics of testfixtures. Examples of these applications include ATEfixtures, evaluation and characterization fixtures forsemiconductor companies, backplane designs andgeneral instrumentation.

We show excellent correspondence in the frequencydomain using VNA, time domain techniques such asTDR, and computational numeric simulations for thedesign method. Models are created using both topo-logical structure and behavior to extract multi-pole/zeroSPICE compatible models that can run on genericBerkeley Spice 3 and Synopsys HSPICE with no exter-nal libraries. An important part of the verificationprocess is to generate consistent models in both timeand frequency domain using both time and frequencydomain instruments.

Using relatively new tools, it will also be shown that itis straightforward to predict eye diagrams of basebandNRZ formatted signals. An attendant benefit of thisdesign method, and some new software tools, is thatperformance extrapolations can be made and keyperformance hot spots in the system can easily beidentified. By improving the interconnect design, asevident in the simulated and measured eye diagram, itis also possible to predict high frequency operation ofa backplane or a board design for a future generationproduct.

Copyright © 2004 TDA Systems, Inc. and Teraspeed Consulting Group LLC. All Rights Reserved

1

Field Solvers and MeasurementWhen data signaling rates rise to above 2.5 Gbps, andedges transition in less than 150 ps, it is not "simple"to measure something as simple as a stripline trace.We take impedance measurements for granted on ourPCBs these days. However, to extend the usefulnessof these measurements, it is necessary to not onlyknow the true impedance profile of a trace, but to alsoprecisely know signal delay and the impact of features(vias, pads, connectors, packages, etc.) upon thepropagation of our signals.

2-D field solvers, such as Ansoft Maxell 2-D, are capa-ble of extracting highly accurate models for the charac-terization of trace impedance, delay and loss modelingand characterization. But a 2-D solver cannot accountfor non-uniform two-dimensional structures, like diverg-ing diagonal differential traces at a connector, or three-dimensional structures such as packages, balls, con-nectors and vias. For these more complex structures,3-D field full-wave field solvers, such as CSTMicroWave Studio, are absolutely necessary. Thesesolvers are capable of extracting highly accuratemodels for these difficult two- and three-dimensionalstructures.

If we assume that we accurately know and understandour PCB manufacturing process, the dimensions of allinterconnect features, and characteristics and proper-ties of all materials, then we can use these advancedsolvers for precise modeling and optimization of ourcircuits for highest performance, enhanced designmargin and lower cost. However, there are uncertain-ties.

It is an assumption that we "know" the characteristicsof our materials and fully understand how our PCBconductors and dielectrics are fabricated. It is also anassumption that the tools we use in our modeling andsimulation process are "correct." Many a designer hasbeen fooled by invalid assumptions. And many adesign has been fabricated just to find that theseassumptions produced an incorrect result, withreduced interconnect performance, increased crosstalkand noise, large impedance mismatches, and otherunaccounted for impairments. Measurement-basedmodeling tests these assumptions and both insuresthe integrity of each design, and refines modeling forfuture designs.

It is equally an assumption that we can precisely andaccurately measure any given real system. At somepoint, we will need to connect our test equipment (aTDR in this case) to our circuit under test. A suitabletool, providing ability to de-embed up to the structurewe need to test, along with ability to address modelingissues using the collected data is necessary. Modelsderived from carefully collected data can reveal issueswith the launch or transition to a signal on a PCB.

Strengths of measurement-based modeling methodsinclude the ability to quickly assess the differencesbetween sample launches.

The following is a list of "launch pathologies" and howthey can be addressed by both measurement-basedand 3-D solver techniques:

1. Ringing (resonance) occurs - Ringing distorts theimpedance measurement. The effect on short tracesand small features may totally obscure the behavior ofthe interconnect under test. Accurate comparison ofsolver techniques and time-domain modeling requiresa true impedance profile. Ringing phenomena must beseparated from multiple reflections that may appear invery high-speed systems.

2. Loss - Available measurement bandwidth isreduced. In terms of frequency, a bad launch reducesthe maximum frequency range that can be character-ized. This may render a specific test impossible toaccomplish, such as the evaluation of the frequencyresponse of a cable being tested. Isolation andlaunch modeling separate risetime degradation due tothe launch from losses due to signal traces.3. Rise time degradation - In the time domain, edgerates are degraded. Reduced risetime of a TDR pulsesignificantly limits the ability to accurately discern andmeasure small features, such as vias, pads, stubs,component packages, connectors and even repetitivefeatures in cables.4. Measurement opacity - A high-speed test systemshould be transparent. For this to occur, the bandwidth(or risetime) of the test equipment, cables and PCBlaunches should be significantly greater than that ofthe bandwidth (or risetime) to be measured.

Measurement-based methods validate the assump-tions and simplifications of electromagnetic field solverbased modeling and simulation. Material characteris-tics are not always homogenous, and the complexshapes of even simple SMAs require some level ofapproximation. Board fabrication and assembly alsocreate a new set of variables that require verificationusing measurement-based techniques.Correspondence of the two approaches enables "tun-ing" for the field solver parameters. Correspondenceof the technique also garners confidence in the designprocess.

There's No Such Thing As A Free LaunchIf you have infinite patience, time and resources, youcan place an SMA connector on a board and confirmwith measurements that you have a "bad" launch.Then, based upon experience, hunches or careful dis-cernment, "corrections" can be made to the layout forthe next go-around. (Microwave and RF engineershave used this technique for years. A microstripdesign is "tweaked" by careful application of the

2

Dremel tool.) After several iterations, you may have abetter launch or a worse launch, but probably not theoptimal launch that money can buy.

Or with good measurement equipment and softwaretools such as TDA Systems IConnect®, and anadvanced 3-D full wave time domain field solver andsimulator such as CST MicroWave Studio, the iterate-measure-iterate-measure-iterate … infinite loop can beshort circuited into a few quality cycles.

With advanced 3-D parametric modeling, something as"simple" as an SMA connector mounted to a PCB canbe fully characterized, modified and optimized for bestperformance, given the current known assumptionsabout the PCB and manufacturing. After fabrication,measurements can be used to test, verify, and eitherconfirm or modify the working assumptions. Withmodern software tools and measurement techniques,overall measurement accuracy and bandwidth can bebetter than doubled. With these techniques, sub-10psTDR measurements can be routinely made, for accu-rate interconnect characterization.

Field Solver OptimizationWith advanced 3-D full wave field modeling andsimulation tools, it is possible to accurately model, sim-ulate and extract S-parameter models for 3-Dstructures, such as an SMA launch. In our case, wehave optimized a transition from layer 1 to inner layerstripline for a specially designed top-launch SMA.Figure 1 shows the particular Molex SMA chosen foroptimization, part number 73251-1850.

Figure 1 - Molex SMA.

Several features of this SMA are desirable for test pur-poses. First, contact to the PCB is made with com-pression fit contacts that have been machined to con-tact an upper ground ring and signal pad with enoughforce to make a reliable contact for test purposes, andallowing a minimal pad size and via. (In this case, a32-mil pad on top of a 12-mil via.) Second, the signalcontact is integrated into the connector as one piecefor ease of assembly and minimization of toleranceissues. Third, the flange of the connector contains 0-80 UNF tapped screw holes for ease of mounting on a

PCB. Rather than populate large boards with manyhundreds of expensive SMA connectors for testing, afew connectors can be mounted and moved from loca-tion to location, thereby reducing overall test cost, yetmaintaining test launch quality.

To use this SMA, it was necessary to create a reliableprocedure to ensure that the launch was of the highestquality and bandwidth possible. In doing so, there areseveral degrees of freedom available for optimization.First, the layer that is launched into can be chosen forbest performance. If the SMA is mounted on the toplayer of a PCB and launched into the lowest striplinelayer, this configuration reduces the via stub length toits minimum, which increase bandwidth, as is a well-known observation. Conversely, in order to accuratelylaunch into the highest stripline layer, the SMA may beplaced on the bottom layer of the PCB. Second, thesize of the via clearance holes (antipads) on eachlayer can be modified to adjust the capacitance andimpedance of the structure. Finally, the position of aground via ring around the transition via can be adjust-ed to provide 50 ohm impedance for the overall struc-ture.

On our first attempt at optimization, Ansoft 3D, a quasi-static field solver, was used to analyze the structure aswe modified antipad and ground via sizing. Becausethe solver uses the quasi-static approximation, no full-wave electromagnetic field effects were analyzed. Thisproved to be a major problem with this methodology.Optimization consisted of altering the structure untiloverall impedance from the top layer through to thestripline trace was calculated to be as close to 50ohms as possible. Figure 2 shows the result of thisoptimization. A fairly large diameter via separation ofapproximately 185 mils was found to be optimal forthese simulations.

Figure 2 - Ansoft 3D optimized launch pattern.

3

TDR measurement of this structure was made using aTektronix TDS8000/80E04 TDR with TDA SystemsIConnect® and is shown in Figure 3. Several thingsmay be determined from this plot. The impedance ofthe board is measured to be 48.3 ohms. The worst-case launch disruption due to the SMA, vias andantipads is between 56.4 ohms and 48.3 ohms. Thelaunch is primarily "inductive"; that is, it is higherimpedance than the nominal 50 ohms. And finally,there is some ringing that can be seen in the higherresolution plot shown later in Figure 11.

Figure 3 - Original SMA TDR profile.

This is not a "bad" launch at all when compared to oth-ers we have measured and will be shown later in thispaper in Figure 15. But we suspected that there wasmore room for further improvement using a full-waveelectromagnetic approach. Ringing in the connectorlaunch area was the cue that something interestingwas going on in this region. Subsequently, a model ofthe SMA transition was built in CST MicroWave Studio,a Finite Integration Technique 3D time domain fieldsolver and simulation environment. This particularSMA transition was used to benchmark the solveragainst measurements, to develop confidence in thesoftware and in these types of optimization proce-dures.

Figures 4, 5 and 6 show various views of the modeledSMA transition. Included in the modeling are all via,pad, plane, metal and dielectric structures of the boardand SMA, along with all excitation ports. Since a CADmodel of the SMA was not available, measurements ofthe smooth transition area were made and used formodeling purposes. More accurate modeling wouldrequire mechanical CAD models for the male andfemale portion of the SMA. But since we were inter-ested in the performance of PCB transition, this wasnot a major issue.

Figure 4 - Original SMA launch perspective view.

Figure 5 - Original SMA launch side view.

Figure 6 - Original SMA launch bottom view.

Because CST MicroWave Studio is a time domain full-wave field solver, it can easily be used to generateTDR/TDT simulations that can be used for correlationpurposes. For the most accurate of simulations, auser-defined pulse source can be used, which match-es the actual TDR launch into the structure.Measurements of the reference TDR pulse were madewith a Tektronix TDS8000/80E04 TDR, edited andimported into CST MicroWave Studio for an accuraterepresentation of our actual waveform. In Figure 7,the imported excitation waveform used in subsequentsimulations is shown. Figure 8 shows the simulatedTDR profile of the structure. All impedancediscontinuities in the structure, as well as high frequen-cy resonance (ringing), are clearly seen.

4

Figure 7 - Measured TDR waveform used for timedomain simulations.

Figure 8 - CST MicroWave Studio TDR profile fororiginal SMA launch.

Comparing Structures Using Measurement-based Methods

TDA Systems IConnect® can be used to both de-embed the TDR stimulus to the DUT SMA-striplinetrace and generate a true impedance profile from thevoltage profile reported by the TDR instrument. Evenin simple cases, multiple impedances generatereflections that mask the true impedance profile, mak-ing it impossible to model with distributed or lumpedcomponents. Generating a true impedance profile isusually the first step in the measurement-basedmodeling process, since it indicates how the topologyof the model should be constructed. It is also animportant step in evaluating and comparing differentstructures. This can be done even before a suitablemodel is generated and simulated. For example,Figure 9 illustrates a bad launch where the trueimpedance profile, or Zline, varies from 25 ohms to 68

ohms and there are major capacitive and inductiveimpedance discontinuities. The initial capacitivediscontinuity matches the topology of the launch sincea long and wide via was used for the SMA launch.

Figure 9 - Representative of a "Bad" SMAImpedance Profile.

Measurement-based Behavioral Modeling

Behavioral models accurately represent the time andfrequency domain behavior of the interconnects up tothe bandwidth of the acquiring TDR instrument. Figure10 illustrates an example of using MeasureXtractor™for modeling the original SMA launch and the entiresignal path through another board via a connector. Atotal of five measurements were required includingreflected, transmitted and easily acquired referencewaveforms. The extracted SPICE-compatible modelconsists of 100 poles/zeros in this case. We can seeexcellent correspondence between the simulatedresults extracted through MeasureXtractor™and thetransmitted (TDT) and reflected (TDR) data from thefour waveforms in Figure 10.

Figure 10 - Simulated (using Berkeley Spice 3) ver-sus collected data using TDA System’sMeasureXtractor™

5

Measurement-based Behavioral ModelingTopological models differ from behavioral models inseveral aspects. Fundamentally, topological modelsretain the geometries mapped to the model andbehavioral do not. They are, however, often moretime-consuming to generate. Figure 11 illustrates themodel simulation results using IConnect® andTDS8000 mainframe and an 80E04 TDR samplinghead for the SMA structure. There are two plots inFigure 11: the green plot represents the output of thefield solver and the blue plot is the simulation resultsgenerated from IConnect®. The structures peak inthe resonance was accurately modeled along with the43 ps periodicity. The source needed to be de-embed-ded carefully and used a 85052D Agilent CalibrationKit open load source, since the open load of the SMAshort cable resonated more than the SMA launch weintended to model. Typically, this is not required for12GHz bandwidth specified models and below. In thiscase, we were after very high-frequency simulation - a43 ps period is 23 GHz! All modeling using either fre-quency or time-domain are limited by themeasurement capability and associated methods, andnot typically by the software tool itself.

Figure 11 - Result of Field Solver simulationimported into IConnect® and compared againstgenerated time-domain topologically createdmodel.

Measurement-based Behavioral ModelingTDR measurements of the initial SMA design showconfirmation for the simulated field solver results. Butexperience leads us to believe that there is room forimprovement. Figure 11, a high frequency 23 GHzresonance, is clearly seen which, if reduced, could sig-nificantly increase the bandwidth of the structure.Measurement confirmation of the field solver resultshas added confidence to our modeling methodology,allowing us to further explore the SMA launch throughsimulation.

Our intuition (well refined after way too many years oflooking at these sorts of things) tells us that the reso-nance we are seeing is a result of the cavity formed

between the vias in the grounding ring. CSTMicroWave Studio allows us several methods to evalu-ate resonance conditions. Figure 12 shows a plot ofthe S-parameters for the launch. Note the well-definedresonance condition at about 23.056 GHz, in agree-ment with our simulated and measured TDR plots.Now that the resonance is confirmed in the frequencydomain, a second feature of MicroWave Studio can beused to monitor and visualize the fields.

Figure 12 - S-parameter sweep of original SMAtransition.

Figure 13 shows a plot of an E-field monitor at 23.056GHz. The resonant field pattern can be clearly seen,bounded on all sides by the ground vias of the struc-ture. Our intuition was correct. As the signal fieldtransitions from the SMA contact on the top layerthrough the board, it passes down a via, where it turnsthe corner and attaches itself to trace. During this"turn of the corner," there is a break in the signalreturn path due to a transition from coaxial mode tostripline mode. It is this "mode conversion" wheresome energy in the signal is lost and converted to"parallel plate mode" energy, which is able to propa-gate along the planes as a circular wave front. Thiswave front is clearly seen with our field monitor as ithits the boundary of the vias and begins to reflectbackwards.

Ansoft 3D optimization assumed that the excitation fre-quency was below that which causes resonant condi-tions. This is clearly not the case for our SMA launch.But CST MicroWave Studio has the ability to performoptimization of complex structures such as these withall full-wave characteristics. The degrees of freedomfor optimization that we discussed above were thenused as parameters in this structure. Antipad size oneach layer and ground via locations were parameter-ized and allowed to vary during multiple optimizationruns.

6

Figure 13 - S-parameter sweep of original SMAtransition

Figure 14 shows the top view of the original vs. theoptimized launch design. Since we are ultimately con-cerned with reducing reflections off the structure in thetime domain, return loss measurements can be usedto guide optimization. In our case, minimum S11 in thefrequency range from 10 GHz to 30 GHz was specifiedas the criteria used by the MicroWave Studio optimiz-er. The results for the final launch structure are shownin Figure 15.

Figure 14 - Original vs. optimized SMA launchdesign.

When compared to the previous "good" SMA transitionprofile, several features of the "optimized" transitionare apparent. First, the ground via ring diameter hasbeen reduced significantly. Second, by decreasing thevia ring diameter, the cavity resonant frequency hasbeen increased to just below 40 GHz, almost a 2Xperformance improvement. And finally, insertion lossover the operating SMA connector frequency range of0 Hz to 26.5 GHz has been reduced significantly.

Figure 15 - Optimized SMA launch S-parametersweep.Figure 16 shows a close-up of the IConnect® meas-ured TDR profile for the Teraspeed-optimized SMA-to-stripline transition. On first glance, it seems that theimpedance profile ranges from 48 ohms to 53.5 ohms.However, further investigation shows that the large53.5 ohm impedance discontinuity was internal to theSMA connector. A discontinuity exists at the matingsurface between the male connector on the cable andthe female receptacle on Molex SMA. After taking thisinto account, the optimized SMA transition impedanceprofile is seen to be 48 ohms to 51.5 ohms, or approxi-mately 50 ohms +/- 4%. This SMA to stripline transi-tion actually outperforms the connector itself!

Figure 16 - Measured TDR response of optimizedSMA launch.

7

In Figure 17, TDA Systems IConnect® is used to over-lay measurements from various SMA transitions onvendors' test boards available in our labs. Typicalnon-optimized SMA connector launches generallyshow large high- and low-impedance discontinuities,some as low as 35 ohms and as high as 85 ohms.The difference between these and the Teraspeed-opti-mized launch is dramatic. Note that thesediscontinuities do not end at the connector, but mani-fest themselves as ringing that can last as long as 600ps, making accurate measurements of the real struc-ture under test all but impossible. This "poor" SMAtransition exhibits all the pathologies described above:ringing, high loss, risetime degradation and highmeasurement system opacity.

Figure 16 - Measured comparison of various SMAlaunches. The Teraspeed launch is about 50 Ohms+/- 2 Ohms. On the other hand, the Teraspeed-optimized launchshows minimal ringing and exceptionally highbandwidth. The hybrid approach of using advanced 3Dfield solvers in conjunction with measurement-basedmodeling methods has been proven, using no specialmaterials or exotic connectors.

Closing The Loop And Getting ItRightThe combination of solver-based modeling methodsand measurement-based modeling methods is formi-dable. Modeling, simulation and measurement areeach enhanced when we combine them into anadvanced design methodology. Measurements andmeasured model extraction can be used for themodeling of existing systems, but for future designs,our methods are most efficient when we do not haveto build test prototypes every time.

When we utilize measurements to test our modelingassumptions and the accuracy of our tools, we providea calibration to the design process that is unsur-passed. At the simplest level, a TDR pulse may becompared between simulation and measurement. Butfor more advanced performance studies, a measure-extracted behavioral SPICE model is used insimulation to compare with field solver modeledresults. This provides the feedback to "close the loop"for confirmation of, and correction to, the modelingassumptions. Once we have verified our solver meth-ods with measurement, confidence in our final resultsincreases. Changes can be made to designs, withincreased density, higher data rates and lower cost.Ultimately, design iterations can be reduced and totaldevelopment cycles are decreased.

The debate is over. Combining field solver andmeasurement-based methods push the designprocess into a new realm. This realm does not requiremultiple design and fabrication iterations to obtainspecified performance. The one-time cost of software,test equipment, and process development is easily off-set by the hidden costs of multiple design spins,design failures and increased time to market.

© 2004 TDA Systems, Inc. All Rights Reserved4000 Kruse Way Pl. #2-300, Lake Oswego, OR 97035 USATelephone: (503) 246-2272 Fax: (503) 246-2282E-mail: info@tdasystems.com Web site: www.tdasystems.com

HMFS-0104

Data subject tochange without notice

The Interconnect Analysis Company™

© 2004 Teraspeed Consulting Group LLC. All Rights Reserved2926 SE Yamhill St., Portland, OR 97214 USATelephone: (503) 239-5536 Fax (503) 239-4400http://www.teraspeed.comscott@teraspeed.com

8