agilent designing and calibrating rf fixtures for surface...

AgilentDesigning and Calibrating RFFixtures for Surface-MountDevices

1996 Device Test Seminar

Effective Test Methodsfor Today's RF Devices

Abstract

Many devices in today's communications systems arecontained in leaded or leadless surface-mount packages,without coaxial connectors. Measuring these devicesrequires a test fixture in order to interface to coaxial-basedtest equipment. This paper covers techniques to designhigh-quality RF fixtures and the calibration standardsneeded for accurate measurements. These techniques areuseful for both leaded and leadless components, fromsimple two-port devices such as filters, up to more complexmultiport RFICs. Calibration alternatives will be discussedand methods will be presented to verify the raw andcorrected performance of the fixtures. Included is a casestudy documenting the fixture-design process for a 947MHz leadless dielectric-resonant filter.

Slide #1 Slide #2

Due to the rapid growth of consumer RF productssuch as cellular and cordless telephones, afundamental shift has occurred in the way RF devicesand components are made and used. Surface-mounttechnology (SMT) is pervasive in RF design andmanufacturing, requiring new strategies formeasurement of SMT components. These componentsrange from simple two-port devices, such as filters andamplifiers, to more complex multiport RFICs. They allshare the need to be accurately characterized andverified in R&D to help develop accurate models, andtuned and tested during manufacturing to ensure thatperformance specifications are met.It wasn't that many years ago that most RF systemsconsisted of connectorized components, both passiveand active, that were bolted together to form the finalsystem. When printed-circuit boards (PCBs) wereused, they consisted mostly of discrete componentssuch as resistors, inductors, capacitors, transistors anddiodes, with RF connectors for input and output.Today, size, weight and cost constraints along withhigher operating frequencies and advances intechnology are driving the use of much smaller andmore integrated packaged parts at the PCB level. Andunlike the old days when there were just a fewstandard transistor packages to worry about, nowthere are many non-standard SMT packages to fit amultitude of RF applications. The physical dimensionsof these parts vary greatly, due to differingtechnologies, power handling requirements,environmental conditions, and so forth. But the needfor RF fixtures to accurately measure all these devicesis greater than ever.

4 - 3

Designing and Calibrating RF Fixturesfor Surface-Mount Devices

Designing and Calibrating RF Fixtures for Surface-Mount Devices

1996 Device Test Seminar

Effective Test Methodsfor Today's RF Devices 1996 Device Test Seminar:

Test Methods for Today's RF Devices

RF Design: Old and New

Traditionally, RF systems used many connectorized parts

Modern designs are highly integrated and use SMT parts with a large variety of package styles and sizes

Slide #3

Making quality RF measurements on devices withstandard coaxial connectors is relatively easy. Veryaccurate measurements can be made usingcommercial calibration kits and standarderror-correction routines found in most networkanalyzers. Performing accurate measurements ondevices with non-standard connectors is a little harder,requiring adapters and often custom calibrationstandards. Devices without connectors are the hardestto measure, since some sort of test fixture is requiredto provide an electrical and mechanical connectionbetween the device under test (DUT) andcoaxial-connector-based test equipment. In addition,in-fixture calibration standards are often required toachieve the level of measurement accuracy that manyof today's devices demand.The goal of this paper is to offer guidelines fordesigning effective RF fixtures and calibrationstandards that will yield repeatable and accuratemeasurements of surface-mount devices. We willdiscuss many concerns and issues that must beaddressed as part of the fixture-design process. Wewill focus on applications below 3 GHz, though muchof what will be covered is can be extended to higherfrequencies. We will not attempt to offer acomprehensive design "cookbook", as mostsurface-mount parts have unique sets of attributesrequiring custom fixture designs. A solid backgroundin RF theory and techniques is very helpful fordesigning high-quality test fixtures. Access to moderncomputer-aided engineering (CAE) tools formechanical and electrical design is also very valuable.

Slide #4

This paper features a case study of a 947 MHzdielectric-resonant filter (DRF) intended for GSMapplications, and contained in a leadlesssurface-mount package. This particular part waschosen as a typical representative of currentapplications, and will be used as an examplethroughout the paper. Two different fixtures weredesigned to test this part. One is an internal Agilent designdone expressly for this paper, and the other wasdesigned by Inter-Continental Microwave (ICM) ofSanta Clara, California, a commercial supplier ofstandard and custom RF and microwave fixtures. Bothfixtures had to address the same set of problems, andas we will see, they exhibit both similarities anddifferences. As we explore the different aspects offixture design, we will show how the designapproaches of our sample DRF fixtures attempted tosolve the various problems.

4 - 4


1996 Device Test Seminar: Effective Test Methods for Today's RF Devices

950 MHz GSM Dielectric-Resonant Filter

947

CD - 67A

Center Freq. 947.5 MHzBandwidth 25 MHzInsertion Loss 2.6 dBRipple 1 dBVSWR (RL) 2:1 (9.5 dB)Stopband 12 dB

(+/- 32.5 MHz)

Ground

RF RF

Ground

Ground1996 Device Test Seminar: Effective Test Methods for Today's RF Devices

What You Will Learn

Design guidelines fixturescalibration standardsverification standards

Goal:test leaded and leadless RF SMT devicesmake accurate and repeatable measurements

Focus on RF (< 3 GHz)Not a "cookbook"Helpful:

Solid background in RF theory and techniquesCAE tools for mechanical and electrical design

Slide #5

We will start out discussing fixturing basics, includingthe goal of fixtures and what separates an ideal fixturefrom one that can be achieved in the real world. Wewill cover both electrical and mechanical challenges ofdesigning fixtures, as well as other considerations thatmust be taken into account before finally deciding onthe optimum design approach.Next, we will cover various aspects of fixture designsuch as part insertion, alignment, and clamping,making good transitions between differenttransmission-line types and sizes, minimizingdiscontinuities, making electrical contact, and ifnecessary, providing matching elements and bias.Following that will be a discussion of the variouscalibration strategies available, and then a section ondesigning in-fixture calibration standards to achievethe highest levels of accuracy. Finally, we will discussways to verify corrected performance and estimateerror residuals, which can then be used to computemeasurement uncertainty.

Slide #6

The goal of any RF fixture is to provide some or all ofthe following objectives. First of all, the fixture mustprovide coaxial connectors to allow interfacing to testcables. For RF applications, SMA or Type-Nconnectors and cables seem to be the most commontype. Often what follows is some sort of transition orlaunch from the coaxial connector to a noncoaxialtransmission line such as microstrip. Along the way,there may be other transitions before the signalconductor arrives at the DUT. The fixture must thenmake electrical contact with the DUT. If theimpedance of the DUT is not the same as the systemreference impedance, then matching elements may beneeded for proper measurements. Finally, active partsneed bias current and voltages supplied to the properpins or ports.

4 - 5



Typical Fixture

Coaxial connectors

Launches / Transitions

Contact to DUT


Agenda

Fixture BasicsIdeal vs. Real-world CharacteristicsElectrical and Mechanical ChallengesOther Considerations

Fixture DesignCalibration TechniquesDesigning SOLT Cal Standards

Performance Verification

Slide #7

Let's examine for a moment the behavior of an idealfixture. Simply put, it would provide a transparentconnection between the test instrument and the devicebeing tested. This would allow direct measurement ofthe DUT, without imposition of the fixture'scharacteristics. In parametric terms, this would meanthe fixture would have no loss, a flat frequencyresponse with linear phase, no mismatches, be aprecisely known electrical length, and have infiniteisolation between input and output (zero crosstalk). Ifwe could achieve this, calibration would be easy.There would be no need to calibrate the fixture itself,and the overall system calibration could be done bycalibrating at the end of the test port cables andapplying mathematical port extensions to account forthe electrical length of the fixture.Since it is impossible to make an ideal fixture in thereal world, we can only hope to approximate the idealcase as best as possible. We can do this by optimizingthe performance of the test fixture relative to theperformance of the DUT. We can try to make the lossof the fixture smaller than the specified gain orinsertion-loss uncertainty of the DUT. The bandwidthof the fixture only needs to be large compared to thedesired measurement bandwidth of the DUT.Mismatch can be minimized with good design andeffective measurement tools such as time-domainreflectometry (TDR). The electrical length of thefixture can be measured. Fixture crosstalk need onlybe better than the isolation of the DUT. Since we canonly approximate a perfect fixture, the type ofcalibration required for any particular application willdepend solely on how stringent the DUT specificationsare.

Slide #8

Now that we know how our real-world fixture shouldbehave, let's consider some of the electrical andmechanical challenges that the fixture must address.We'll start with the electrical challenges. The fixturemust deliver an RF signal to the DUT, which meansconverting from a coaxial to a non-coaxialenvironment at some point. The system referenceimpedance Zo (usually 50 ohms) must be maintainedover multiple transitions to provide a good match. Thefixture must provide good RF grounds and eliminatespurious ground paths. Fixtures designed to testnon-50-ohm or balanced devices may need matchingelements. For example, we may need to transform upto a high-impedance passive device like an IF SAWfilter, or transform down to a low-impedance activepart like a transistor. Bias is required for active parts.Perhaps most importantly, it is essential that we haverepeatable connections to the DUT, often forthousands and thousands of insertions.The mechanical challenges of a good fixture are by nomeans trivial either. The fixture has to allow easyinsertion of the DUT, with proper alignment. Forrepeatability, a clamp is generally desirable. For mostapplications, non-destructive contacting is essential.Thermal concerns must also be considered, such asthe need for adequate heatsinking of power devices. Orperhaps the fixture needs to provide heating or coolingfor environmental testing. In manufacturingapplications where it is necessary to grind theresonator dielectric to tune the filter (ceramicduplexers for example), a vacuum dust-removalsystem may be required as part of the fixture design.Finally, the fixture should be rugged to withstandrepeated use, particularly in manufacturingenvironments.

4 - 6



Electrical and Mechanical ChallengesElectrical

Deliver RF signal to DUTTransform from coaxial to non-coaxial environmentMaintain Zo over multiple transitions (minimize reflections)Repeatable connections

Good RF groundsMatching elements for non-50-ohm or balanced devicesBias for active parts

MechanicalEasy insertionAlignmentClampingThermal issues (heatsinking, temperature testing)Vacuum (dust removal)Ruggedness


Ideal versus Real World Fixture

Real-World FixtureOptimize relative to DUT

Loss << specified uncertaintyBW >> DUTMinimize mismatchesMeasure electrical lengthIsolation >> DUT

CalibrationDepends on how well we approximate ideal fixtureTypically need cal standards

Ideal FixtureTransparent connection

No lossFlat mag., linear phaseNo mismatchesKnown electrical lengthInfinite isolation

Simple Calibration2-port cal at end of cablesPort extensions for fixture

Slide #9

There are additional considerations that must beevaluated before we can begin to develop a fixturedesign strategy. The first step is to define our teststrategy. Are we only going to measure a few parts inR&D or are we expecting to measure many parts inmanufacturing? Are we trying to measure the DUTwith a neutral fixture, or will the fixture simulate theend application? If we want to match the applicationenvironment exactly, the fixture and the endapplication should share the same substrate material,thickness, and pad geometry.The package style of our DUT will also affect ourdesign considerably. Fixtures for leadedsurface-mount parts are often easier to design sincethe leads themselves can absorb the non-flatness ofthe fixture or lead geometry. Leadless (and sometimesleaded) parts require compliant contacts to ensuregood RF signal and ground connections.We must also consider our error-correction andverification strategy before undertaking the design ofthe fixture. If we intend to use thru-reflect-line (TRL)calibration, we will require a separable fixture capableof accepting center sections of different length. Thistype of fixture and the corresponding calibrationstandards are often more expensive to build.Alternately, we may wish to use short-open-load-thru(SOLT) calibration, which generally means simplerand therefore less expensive fixtures and standards.Knowing our cost goals from the beginning is veryimportant. It may be sufficient for some applicationsto have a simple, inexpensive fixture, where othertimes a more flexible and robust (and therefore moreexpensive) fixture is desired. For manufacturingapplications, the cost of replacing all or part of thefixture due to wear must also be considered.

4 - 7



Fixture Design Strategy

Best strategy:Optimize design as much as possibleUse error-correction only as needed

Not a good idea:Build fixture with poor RF performance and rely on error correction to fix everythingRepeatability and accuracy will suffer

high loss gives more trace noiselarge mismatch degrades raw directivity and port match, causing more susceptibility to drift

Forces us to use in-fixture calibration


Additional ChallengesTest Strategy

R&D vs. manufacturingNeutral fixture vs. simulating end applicationIf matching application environment, use same substrate material (FR-4, teflon, ceramic, etc.), thickness, geometry

Package Style of DUTLeaded vs. leadless partsFixed vs. compliant electrical contacts

Error-Correction and Verification StrategyTRL requires movable fixture (more expensive)SOLT fixtures and standards usually simpler, cheaper

Cost GoalsInitial costReplacement cost (for manufacturing applications)

947

CD - 67A

Slide #10

Once we have evaluated all of the previously discussedconsiderations, we can go about designing the fixturefor our particular application. The best strategy is tooptimize the fixture design as much as is practicallypossible, and then, if our application demands it, usesome form of error-correction to remove the remainingerrors. Building a poor fixture and expectingcalibration to fix everything is not a good idea, asmeasurement repeatability and accuracy will suffer.For example, a fixture with a lot of loss causesdegradation of measurement signal-to-noise-ratio,resulting in more trace noise and therefore greatermeasurement uncertainty. Fixtures with excessivemismatch severely degrade the raw directivity andport matches of the overall test system, resulting ingreater susceptibility of measurement uncertainty dueto drift. Furthermore, fixtures with poor RFperformance usually force us to take the time and costpenalty of in-fixture calibration, in order to makedecent measurements.On the other hand, if we design a good fixture withperformance that is significantly better than thespecifications of the DUT, we may be able to get awaywithout any in-fixture error correction, using onlycoaxial calibration at the ends of the test cables andperhaps port extensions. Components with moredemanding specifications will often require moreadvanced error-correction techniques such asde-embedding or calibration using physical standards.

Slide #11

This section of the paper will discuss the variousaspects of fixture design that are needed to buildhigh-quality RF fixtures. We will start with the task ofgetting the DUT physically into the fixture, with properalignment and clamping. We will then cover ways ofgetting the test signal to the DUT while minimizingmismatch, and how to electrically and mechanicallyconnect to the leads or pads of the DUT. We will alsobriefly cover some considerations about providing biasfor active parts and matching for non-50 ohm devices.

4 - 8



Insertion, Alignment, Clamping

InsertionManual okay for R&D or low-volume mfg.Part handlers typical in high-volume mfg.

AlignmentEnsures repeatable electrical connectionsUse pins, edges, or depression

ClampingConsistent pressure also helps ensure repeatabilityVariety of ways: manual, automatic, vacuum, part-handlerMay need an ejector pin to force part out of fixture

DUT


Agenda - Fixture Design

Fixture BasicsFixture Design

Insertion, alignment, and clampingGetting signal to DUTConnection to DUTProviding matching, bias

Calibration TechniquesDesigning SOLT Calibration StandardsPerformance Verification

Fixture DesignCalibration TechniquesDesigning SOLT Cal StandardsVerification of Performance

Slide #12

The first step in measuring a surface-mount device isto physically insert it into the test fixture. The part canbe placed by hand in an R&D or low-volumemanufacturing environment, or be inserted with anautomated part handler. The fixture must providesome way of aligning the part to ensure that electricalcontact is made to the proper pads or leads. Thealignment of the part should be very consistent toensure good measurement repeatability. Small shifts inDUT alignment can cause electrical paths to vary,which can result in significant measurementuncertainty. One way to achieve alignment is to placea pin in the fixture that is keyed to a notch or edge onthe part. Another way is to use a depression in thefixture that matches the dimensions of the DUT.Clamping the part in place is critical becausecontrolled, consistent pressure is key to making good,repeatable electrical measurements. While "fingerpressure" may be okay in the R&D lab, it is generallynot suitable for use in manufacturing applications. Theclamp must provide just the right amount ofmechanical travel to ensure proper electrical contactwithout putting undue flex on leads or contacts, orforcing the part to bottom-out in the fixture.Consistent pressure is important since contactresistance between DUT and fixture can vary withpressure. A clamp might consist of a simple handclamp as our example fixtures use, or a moresophisticated clamp that might allow faster throughputfor high-volume manufacturing. Other options includeusing a part handler to provide the clamping function,or a fixture with a vacuum-based clamp. After the partis tested, some fixtures employ a release mechanismthat pops the part out of the fixture, making for easyretrieval.

4 - 9



Test Cables

Low lossStability

with movementwith temperature

Semi-rigid cables bestUse "loops" to accommodate different fixturesHigh-quality flexible cables also okay

DUT


Resonances and Fringing

Lid may cause cavity resonanceMetal near DUT may affect response (detune filter for example)Terminate unused connections

DUT

Slide #13

Ideally, the mechanical attributes of the fixture willhave no impact on the electrical performance.However, sometimes electrical resonances occur,especially at frequencies above 3 GHz or so. Forexample, a cavity resonance can occur when a lid isplaced over the fixture. This problem can usually besolved by proper placement of a lossy substance, suchas polyiron, or by not using a lid in the first place.Another cause of resonances can be adjacent contactpins or transmission lines. To avoid unwantedreflection or crosstalk, unused connections should beproperly terminated.Another potential problem is excess electrical fringingbetween the DUT and the fixture. This can causeelectrical performance of the DUT to be different inthe fixture than in the end application. The problemcan happen if the DUT will not be near any metal onthe PCB, but metal parts of the fixture get too closeduring test. For example, a metal clamp or alignmentpin could cause filter resonators to be detuned orexhibit lower Q. In general, it is best that anynon-electrical contact to the DUT be made withnonmetallic parts.

Slide #14

Before we begin discussion on the electrical aspects offixture design, it is worthwhile to talk about the testcables used between the network analyzer and thefixture. They should exhibit low loss over thefrequency range which we want to measure. Thestability of these test cables also needs to be good toget repeatable measurements. Inexpensive cables cangive different results even with minimal movement,particularly when doing phase measurements. Thisphenomenon gets worse at higher measurementfrequencies. The test cables also need to be stable overtemperature. Normal temperature fluctuations can bea significant source of random errors if poor testcables are used. The best choice for test cables is semi-rigid coax.Loops can be added to allow for some variation infixture width if multiple fixtures will be used at a giventest station. High-quality flexible test cables, whilemore expensive, also have excellent repeatability.

4 - 10



Compensation and Lead Pitch

Inductive (narrow) trace used to compensate for fringing capacitance(yields 25-30 dB return loss)

Problem:Microstrip line too wide for lead pitchSolution:Use thinner substrate (and trace)

or

Use coplanar lines

Reflections occur here

Launch center pinRibbon


Transitions and Transmission Lines

Characteristic impedance for microstrip transmission lines

(assumes nonmagnetic dielectric)

DUT

Transition to non-coaxial transmission line (microstrip or coplanar)

DUT

Transition to coaxial transmission line

MicrostripCoplanar

h

ww1

w2

Slide #15

One of the most challenging aspects of fixture designis to provide a good RF path between the test cablesand the DUT. The goal is to maintain Zo (usually 50ohms) across different transmission line media, whileminimizing reflections at discontinuities.The first step in the RF path is the coaxial connector.For the RF range, type-N or APC 3.5 connectors arethe best choices (since cal standards are available inthose connector styles), but SMA connectors are alsoacceptable. What follows will be a transition to somesort of transmission line. For leaded parts, non-coaxiallines are usually used as it is easy to make contactbetween the transmission line and the DUT. Forleadless parts (when compliant contacts are needed),coaxial lines often make more sense.Microstrip is the most common non-coaxialtransmission line for RF applications. It is easy tomake a connector-substrate launch, as the center pinor tab of the connector can directly touch themicrostrip line. The bottom of the microstrip substratecan have direct contact to the fixture, providing a solidRF ground. The graph above shows the width/heightratio versus line impedance for different substratedielectric-constants.Another common transmission medium is coplanarline, consisting of a center conductor with groundsurfaces on either side. This is more common formicrowave applications as it is convenient for probing(the probes can use the standard ground-signal-groundlayout). For non-probe applications, it is harder toproperly ground a coplanar substrate to the fixture.Both microstrip and coplanar lines allow easyplacement of discrete components for matching,supply bypassing, etc.

4 - 11



ICM Fixture

DUT

Transition

Midsection

Set screw

Spring

Pin support

Dielectric supportRF pinRF connector

Spring-loaded clamping bar


Agilent Fixture

Cutaway view

Slide #16

Providing a low-reflection transition between thecenterpin of a bulkhead connector and the signal traceof a non-coaxial transmission line can be challenging.Typically the diameter of the centerpin is not the sameas the width of the trace, causing some fringingcapacitance. Agilent solved this problem with its Agilent 83040 series of modular microcircuit packages (now obsolete) by using inductive compensation at thistransition. The center conductor of the microstrip linewas briefly narrowed to look inductive, effectivelycanceling the fringing capacitance. The centerpin wasalso bonded to the microstrip line with a wrap-aroundgold mesh, to ensure consistent electrical connection.Return loss of 25-30 dB or better was obtained in thismanner.When dealing with multi-leaded RFICs, lead pitch canbe a problem. A wide microstrip trace might have tonarrow considerably to match the size of a pad or lead.This discontinuity will cause some reflection to occur.One alternative is to use a thinner substrate, resultingin a microstrip line with a smaller width. A coplanardesign can easily accommodate this problem sincetransmission lines can gradually get narrower and stillmaintain the proper Zo. If discontinuities occur, it isdesirable to place them as far away from the DUT aspossible so time-domain gating can easily separate thereflections from the DUT and the fixture (morediscussion on this later).

Slide #17

An alternative to having a coaxial-connector launch toa microstrip or coplanar line is to use a coaxialtransmission line structure right up the DUT. Thisapproach was taken by both the Agilent and ICM fixturedesigns for the example GSM filter. The Agilent fixtureuses pieces of 0.141" semi-rigid cable with SMAconnectors on one end and flush cuts at the other end.Small pieces of conductive cylindrical elastomer areused for electrical and mechanical contact betweenthe semi-rigid cables and the DUT. A 50-ohmenvironment is maintained around each elastomerpiece. This design provides a low-cost transition withreasonably good RF performance. Our approach fromthe beginning was to design a low-cost fixture formanufacturing use. We felt this approach was a goodway to achieve an economical fixture with an easyway to replace the contacts to the DUT should theywear out. While the use of elastomer contacts to the DUT workswell for RF, it may not be suitable for DUTs requiringhigh-current DC connections, as the elastomercontacts can have excessive DC resistance for thisapplication.

4 - 12



Test Socket Contacts

HP 84000 Contact > 12 GHz

DUT

DUT

Cantilever Contact < 500 MHz

Commercially-Available RFIC Contact < 7 GHz

Contactssmaller contacts mean less inductanceavoid ferromagnetic alloysmake plating thickness 2-3x skin depth

DUT


Connection to DUT

Want non-destructive contactsMust account for non-flatness of contact surfacesMost leaded parts can absorb non-flatnessLeadless parts need compliant contacts

pogo pinsconnector tabselastomer contactsspring-loaded centerpins

pogo-pin SMA connector with microstrip tab

Slide #18

Instead of using semi-rigid cable, the ICM fixture usessuspended coaxial transmission lines. The middle ofthe centerpin is suspended in air. One end is supportedwith a plastic dielectric insert, and the other end isinserted into a flange-mounted SMA connector. Tomaintain a constant impedance over the length of thetransmission line, the ratio of diameters of thecenterpin and the hollow cylinder in which it restsmust remain constant. When the centerpin steps to alarger or smaller size, the outer conductor wall muststep accordingly. If this constraint is not followed,significant reflections would occur at the transitions.The end of the center conductor nearest the DUT isextended slightly beyond the coax structure in order tomake contact with the DUT. The contact end is pushedupward slightly by a spring-loaded dielectric dowel inorder to provide a compliant contact to the DUT.As one can see from the above and previousillustrations, using modern CAE tools (such as Agilent'sME-30) for fixture design is extremely helpful asconsiderable mechanical detail is necessary to achievean effective design.

Slide #19

Providing a good electrical contact to the DUT is oneof the most important aspects of fixture design. Formanufacturing (and often for R&D), non-destructivepressure connections are desirable, so many parts canbe quickly and easily measured. Soldering parts intothe fixture is not a viable option for manufacturingtest. Making good pressure connections requires thatnon-uniformity of the DUT contact surface beconsidered. For example, a leadless package cannotsimply rest directly on the fixture substrate, becausesignal or ground pads may not make good contact dueto non-uniform part flatness. Some sort of contactcompliance is needed, supplied either by thespring-force of the DUT leads, or by compliantcontacts in the fixture. Some manufacturers requirecompliant fixture contacts even for devices in leadedpackages, to avoid undue flexing of the leads orbecause lead stiffness is too great to overcome theircontact unevenness.There are several different ways to make compliantelectrical connections. We have seen two approachesalready. Another choice is to use "pogo-pins", whichare spring-loaded metal pins that contact the pad ofthe part. They must be incorporated into atransmission-line structure of some sort. One verysimple approach used by some device manufacturersare flange-mounted SMA connectors with microstriptabs on the launch end of the connector (rather thanthe more-common extended centerpin). The contactpads of the DUT are placed directly on these tabs,which provide the necessary spring force forcompliant connections.

4 - 13



GroundingProviding good RF grounds is asimportant as good signal connectionsDesire:

low inductancerepeatablecompliant grounds (especially for leadless parts)

Avoid multiple ground paths (can cause high-Q dropouts)Challenge for high-power devices

providing ground and adequate heatsinking simultaneouslythermal compound not useful for high-volume manufacturing testthin elastomer sheetsspring-loaded heatsinks

DUTDUT


RepeatabilityShort Term

Measure same part several timesUse normalization to better see variation

Long TermShould not get worse as fixture wears outDesire:

rugged contactsconstant compliant-contact spring forceeasily replaceable contacts

DUT

Direct DUT-trace contact yields limited contact life

Nickel-coated diamond particles provide excellent contact life

Slide #20

For automated semiconductor test systems, thecontact to the DUT is often made with a compliant testsocket integrated onto a larger test board. Standardcommercial test sockets designed for digital test use acantilever technology which is only useful to around500 MHz. For RFIC test, shorter contacts with lowerinductance are essential. Commercial test sockets arereadily available that work well to around 7 GHz. For the Agilent 84000 series of custom RFIC test systems,a very low-inductance, compliant contact wasdesigned with excellent performance to above 12 GHz.The spring force for these contacts is provided byconductive elastomers. Special alloys are used toprovide good electrical conductivity with low frictionbetween moving parts. Individual contacts can easilybe replaced. This technology is only available in theAgilent 84000 test systems, and is not sold separately.Some care must be taken when choosing alloys orplating material for compliant contacts. Materials thatexhibit ferromagnetic properties (such as iron or steel)should be avoided as they can actually increaseintermodulation distortion, especially at high powerlevels. Another concern is that at very-highfrequencies, "skin effect" starts to be significant(current travels near the surface of the contact). Forplated contacts used above 1 GHz or so, the platingthickness should be several times the skin depth. Goldis commonly used to plate contacts as it yieldsexcellent contact life and can be applied over a varietyof metals such as beryllium copper, aluminum, orbrass.

Slide #21

One very important aspect of fixture performance isrepeatability. Inserting and measuring the same partseveral times gives a good indication of how muchmeasurement uncertainty to expect due to partalignment, contact pressure, and grounding variations.The first measurement can be used as a reference fornormalization. Subsequent measurements will thendirectly display measurement uncertainty.Long-term repeatability is a big consideration forfixtures intended for manufacturing applications. Thefixture should continue to give repeatable andaccurate measurements even after repeated insertionsand connections have been made. Spring force shouldremain fairly constant over the life of the contacts. Thecontacts should be easily replaceable as well, so theentire fixture does not have to be discarded after thecontacts wear out. While the elastomer contacts use inthe Agilent fixture are easily replaceable, an evaluationwould have to be undertaken to determine how manyconnections could repeatably be made before newcontacts should be inserted. Fixtures withspring-loaded metal contacts like the one designed byICM have been shown to provide reliable contacts forseveral hundred thousand connections before needingreplacement of the center pin.For fixtures that rely on direct contact to leaded parts(without compliant contacts), one contact technologythat can be used is nickel-plated diamond particles.These particles are bonded directly to the PCB duringthe fabrication process, and provide a rugged contactsurface capable of many 100's of thousands ofcontacts. Using direct contact between the lead of thepart and the PCB is not a good solution formanufacturing test, as contact life is very limited.

4 - 14



Bias

Can use bias-tee if RF and DC share same lineIf separate, fixture needs extra connectors or pinsProper bypassing is important

DC Bias

RF DUT


Non-50-Ohm and Balanced Devices

Matching needed for non-50 ohm devicesBaluns can be used for balanced DUTsConsider matching elements...

part of fixture if trying to characterize DUT by itself (load standard should match DUT)part of DUT if simulating end application (use normal load standard)

Slide #22

Providing low-impedance repeatable RF grounds isjust as important as providing good signal connections.Ground connections that are not part of the expectedgrounding scheme of the DUT should be avoided.Multiple ground paths can cause small, discretediscontinuities (high-Q dropouts) in the measuredresponse, as the phase of the various paths vary versusfrequency. Inductive grounds can also causemeasurement impairment, such as degraded stopbandisolation or poor return loss. As with signal lines,compliant ground contacts can be useful to ensureconsistent ground connections. For example, theAgilent-designed fixture uses small rectangular pieces ofconductive elastomer that sit in recesses in the fixture.These flexible contacts align with the ground pads ofthe filter.One challenge when testing high-power devices isproviding good RF grounding and adequateheatsinking simultaneously. While a thermalcompound can be used during R&D characterizationand during assembly, it is generally not suitable forhigh-volume manufacturing test. A thin elastomersheet with high electrical and thermal conductivity hasbeen successfully used by some transistormanufacturers. Another successful solution used forSOIC devices is to use a spring-loaded heatsink whichcontacts the underside of the DUT.

Slide #23

When testing non-50-ohm devices, sometimes amatching network is needed to make goodmeasurements. Matching elements include discreteinductors and capacitors, baluns and other types oftransformers, and transmission line stubs and steps.Baluns can be used to test balanced parts such asdifferential-input amplifiers or balanced filters. There are two different in-fixture calibration strategiesthat can be used when testing non-50 ohm or balancedparts. If we are trying to measure the performance ofthe DUT by itself, we must consider the matchingelements as part of the fixture. This requires that ourload standard be the same impedance as our DUT. Ifwe are testing a balanced device, our load must alsobe balanced. This approach should be used if we aretrying to determine the optimum matching networkbased on the performance of our DUT.If we already know what our matching circuit will looklike in the end application and we want to characterizethe combined network (DUT plus matching elements),we can consider the matching network part of theDUT. In this case, we should use a 50-ohm loadstandard before the matching circuitry. The fixturedesign must somehow allow us to break theconnection before the matching elements in order toperform a calibration. The matching elements shouldbe located as close to the DUT as possible.

4 - 15



Modeling

DUT

Port Extensions Assume:

no lossflat magnitudelinear phaseconstant impedance

De-embedding External software required

DUT

Two-port calibrationMathematically extend reference plane

Two-port calibration

Accurate S-parameter data (from model or measurement)


Error-Correction Model

Normalization (response cal)

De-embedding

One and Two-port Corrections (SOLT, TRL, LRM...)

Modifying cal kit definitions

Port Extensions

DifficultyEasiest Harder

AccuracyLower Higher

Time-Domain Gating (refl. only)

Mod

elin

gD

irec

t M

easu

rem

ent

Fixture BasicsFixture DesignCalibration TechniquesDesigning SOLT Cal StandardsPerformance Verification

Slide #24

Making in-fixture measurements of active partsrequires that DC bias be supplied along with the RFsignal. Traditionally, when bias was needed for testingtransistors, external bias tees were used in the mainRF signal path. This approach is still valid today.However, most packaged amplifiers and RFICs requirethat DC power be supplied on separate pins. Thismeans the fixture must provide extra connectors orDC-feedthroughs for the necessary bias. Just as with matching elements, discrete elements canbe place directly on the fixture near the DUT toprovided proper RF bypassing and isolation of the DCsupply pins. Good RF bypassing techniques can beessential as some amplifiers will oscillate if RF signalscouple onto the supply lines. As was previouslymentioned, bias connections to the DUT shouldpresent a low DC impedance.

Slide #25

Now that we have looked at the various aspects ofmaking good RF fixtures, let's examine our error-correction choices. The relative performance of ourfixture compared to the specifications of the DUT willdetermine what level of calibration is required to meetthe necessary measurement accuracy. There are two fundamental error-correctiontechniques: modeling and direct measurement. Eachhas relatively simple versions and more complicatedversions that require greater work, but yield moreaccurate measurements.Calibration based on modeling uses mathematicalcorrections derived from an accurate model of thefixture. Often, the fixture is measured as part of theprocess of deriving an accurate model.Direct measurement usually involves measuringphysical calibration standards and calculating errorterms. This method provides accuracy that is based onhow precisely we know the characteristics of ourcalibration standards. The number of error terms thatcan be corrected varies considerably depending on thetype of calibration used. Normalization only removesone error term, while full two-port error correctionaccounts for twelve error terms. Since standard cal-kitdefinitions are based on coaxial standards, modifyingthese definitions for in-fixture calibration is veryimportant for accurate measurements. We will coverthis in more detail later in the paper.Time-domain gating is a special technique that can beused for making error-corrected reflectionmeasurements on broadband devices. We will see laterthat time-domain gating can be a valuable tool to helpus characterize the reflection performance of both ourload and thru load standard.

4 - 16



Thru-Reflect-Line (TRL) Calibration

AdvantagesMicrowave cal standards easy to make (no open or load)Based on transmission line of known length and impedanceDo not need to know characteristics of reflect standard

DisadvantagesFixtures usually more complicated (and expensive)Impractical length of RF transmission lines8:1 BW limitation per transmission line


Direct Measurement

DUT

DUT

Measurement plane (for cal standards and DUT)

Various calibration standards

Reference Measurement

Response Calibration

Two-port cal at ends of cables improves in-fixture response cal

Slide #26

Modeling has the advantage that it does not requiredevelopment of in-fixture calibration standards. Thesimplest form is port extensions, whichmathematically extends the measurement plane to theDUT. This feature is included in the firmware of mostnetwork analyzers. Port extensions assume the fixturelooks like a perfect transmission line: no loss with aflat magnitude, linear phase response, and constantimpedance. Port extensions are usually done after atwo-port calibration has been performed at the end ofthe test cables. If the fixture performance isconsiderably better than the specifications of the DUT,this technique may be sufficient.A more rigorous approach is to use de-embedding.De-embedding requires an accurate linear model of thefixture, or measured S-parameter data of the fixture.Computer-aided design (CAD) tools can help analyzeand optimize the model of the fixture. In-fixturecalibration standards may be used to help measure theperformance of the fixture. External software isneeded to combine the error data from a calibrationdone without the fixture (using coaxial standards)with the modeled fixture error. If the error terms of thefixture are generated solely from a model, the overallmeasurement accuracy depends on how well theactual performance of the fixture matches the modeledperformance. For fixtures that are not based on simpletransmission lines, determining a precise model isusually harder than developing good in-fixturecalibration standards, especially in the RF range.

Slide #27

Direct measurements have the advantage that theprecise characteristics of fixture don't need to beknown beforehand, as they are measured during thecalibration process. Another benefit is that the errorcorrection is done in the network analyzer, without anexternal computer as required for de-embedding. Thesimplest form of direct measurement is a responsecalibration, which is a form of normalization. Areference trace is placed in memory and subsequenttraces are displayed as data divided by memory. Aresponse cal only requires one standard each fortransmission (a thru) and reflection (a short or open). Response calibration has a serious inherent weaknessbecause no correction can be done for errors due tosource and load match. This is especially a problemfor low-loss transmission measurements (such asmeasuring a filter passband or a cable) and forreflection measurements. Using response calibrationfor transmission measurements on low-loss devicescan result in considerable measurement uncertainty inthe form of ripple. Measurement accuracy will dependon the relative mismatch of the test fixture andnetwork analyzer compared to the DUT. Responsecalibration is often acceptable for transmissionmeasurements with significant loss in at least onedirection (an amplifier for example), but is not a goodidea for reflection measurements.When response calibration is used for transmissionmeasurements with fixtures, considerablemeasurement improvement can be made by firstperforming a two-port correction at the ends of thetest cables. This will improve the effective source andload match of the network analyzer, thus helping toreduce the measurement ripple due to reflections fromthe fixture and the analyzer's test ports.

4 - 17



Modifying Cal Kits

Open Short Load Thru/Line ArbitraryImpedance

Capacitance

Inductance

Offset delay

Offset Zo

Offset loss

Min/max frequency

Coax/waveguide

Fixed/sliding/offset

Terminal impedance


Short-Open-Load-Thru (SOLT) Calibration

SOLT Cal is Attractive for RF FixturesSimpler and less-expensive fixtures and standardsEasy to make broadband calibration standardsCal standards must match pad geometry of DUT

SHORT

OPEN

Depends on:temperature driftwear of fixtureDUT specifications

How often should I calibrate?

THRULOAD

Slide #28

More advanced error correction thanresponse/normalization is achieved with moremeasurements and more calibration standards. Forexample, a one-port open-short-load (OSL) calibrationrequires three measurements of three standards, whilea two-port SOLT calibration requires twelvemeasurements on four standards. The most accuratecalibration of a fixture is achieved by using a full set ofcalibration standards measured at the contact plane ofthe DUT. The two most common types of fixturecalibrations based on physical calibration standardsare TRL and SOLT. Each has its unique set ofadvantages and disadvantages, but if properly done,both yield excellent measurements. The main advantage of TRL is that the cal standardsare relatively easy to make and define. This is a bigbenefit for microwave applications, where it is difficultto build good open and load standards that are neededfor an SOLT calibration. TRL uses a transmission lineof known length and impedance as one standard. Theonly restriction is that the line needs to be significantlylonger in electrical length than the thru line, whichtypically is of zero length. The general rule-of-thumbfor the line standard is that it should be between 20and 160 degree in length, which for our GSM filter at947 MHz, would result in a transmission line betweenone and nine centimeters in length. TRL calibrationalso uses a high-reflection standard (usually a short oropen) whose impedance does not have to be wellcharacterized.

Slide #29

The biggest disadvantage of TRL is that the fixture isgenerally more complicated and expensive tomanufacture compared to a fixture designed for SOLTstandards. This is because a TRL-based fixture mustbe capable of being separated in half, to allowinsertion of the long transmission lines. At RF, thetransmission lines can get too long to be practical aswell. Another drawback is that the transmission linescan only be used over an 8:1 bandwidth. For a broadercalibrated bandwidth, additional line standards mustbe made.For RF applications, using SOLT-based errorcorrection for calibrating the fixture is very attractivebecause the standards and fixtures can be simple andinexpensive. SOLT calibration does not require thefixture to move. For electrical contact, the calibrationstandards have to match the pad geometry of the DUT.For mechanical insertion, the standards have to fit inthe fixture with proper clamping, but it is notnecessary that they match the DUT package exactly.Another advantage of SOLT calibration is that thestandards generally work over a very broad bandwidth- calibration from DC to 3 GHz is easy to achieve.How often will we be required to perform acalibration? It depends primarily on three factors. Thefirst is the thermal stability of the test environment. Asthe instrument drifts due to temperature changes,calibration may be necessary depending on themagnitude of the change and the specifications of theDUT. The second factor is how mechanically stablethe fixture is. As the fixture wears out, itscharacteristics might change slightly, requiring a newcalibration. Finally, any changes in test-systemperformance must be compared to the specificationsof the DUT to evaluate severity. In general, each

application is unique and the user must determine the appropriate calibration interval.



Time-Domain GatingTDR and gating can remove undesired reflections (a form of error correction)Only useful for reflection measurements of broadband devices (a load or thru for example)Define gate to only include DUTUse two-port calibration at ends of test cables

CH1 S11&M log MAG 5 dB/ REF 0 dB

START .050 000 000 GHz STOP 20.050 000 000 GHz

Gate

Cor

PRm

1

2

2: -15.78 dB 6.000 GHz

1: -45.113 dB 0.947 GHz CH1 MEM Re 20 mU/ REF 0 U

CH1 START 0 s STOP 1.5 ns

CorPRm

RISE TIME 29.994 ps 8.992 mm

1

2

3

1: 48.729 mU 638 ps

2: 24.961 mU 668 ps

3: -10.891 mU 721 ps

Thru in time domain

Thru in frequency domain, with and without gating


Time-Domain Reflectometry (TDR)

Analyze impedance versus timeDifferentiate inductive and capacitive transitionsHigh-speed oscilloscope:

yields fast update rate200 mV step typical

Network analyzer:broadband frequency sweep (often requires microwave VNA)inverse FFT to compute time-domainresolution inversely proportionate to frequency span

Zo


application is unique and the user must determine the

4 - 18

appropriate calibration interval.

Most high-performance network analyzers allow theuser to modify the standard definitions for thecalibration standards. This is very important forfixture-based measurements, since the in-fixturecalibration standards rarely will have the sameattributes as any of the standard coaxial-based cal kits.For normalized transmission measurements, we canachieve a little better accuracy by modifying thedefinition of the thru to include loss and electricallength. Two-port calibration requires proper definitionof all of the reflection and transmission standards.This includes terms such as offset delay and loss,impedance, and fringing capacitance (for opens). Theresulting definition of the in-fixture calibrationstandards can be stored in the analyzer as a customuser-cal kit. Coming up with the proper definitions ofthe standards is crucial for accurate measurements,and will be covered in more depth later.

Slide #31

Once the fixture has been designed and fabricated, wecan use time-domain reflectometry (TDR) toeffectively evaluate how well we minimizedreflections. As long as individual transitions can bediscerned, TDR can show us which ones need moreoptimization. TDR can also help evaluate and improvethe quality of our load and thru standards.There are two basic ways to perform TDRmeasurements. One way is accomplished bygenerating a high-speed step function (usually a fewhundred millivolts in amplitude) and measuring it witha high-speed oscilloscope. This technique providesmeasurements with a high update rate, which allowsreal-time tweaking. It is very easy to determiningwhich transition is which, as the designer can place aprobe on a transition and look for the spike on theTDR trace. However, in general, oscilloscopes cannotdisplay data in the frequency domain as can a networkanalyzer. The second technique uses a network analyzer makingnormal frequency-domain swept measurements. Theinverse-Fourier transform is used to transform thefrequency-domain data to the time domain, yieldingTDR measurements. While the update rate is muchslower, the advantage is that one instrument can beused to provide both time and frequency-domainmeasurements. When using a network analyzer, thespatial resolution is inversely proportional to thefrequency span of the measurement - i.e., the higherthe stop frequency, the smaller the distance that canbe resolved. For this reason, it is generally necessaryto make microwave measurements on the fixture toget sufficient resolution to analyze the varioustransitions. Providing sufficient spacing between



Short Standard

Ideal: unity reflection with 180 phase shiftSimply short signal conductor to ground (metal slug or bar)If using coplanar lines, short to both ground planesAvoid excess inductance by keeping length short

o


Designing SOLT Calibration StandardsFixture BasicsFixture DesignCalibration TechniquesDesigning SOLT Cal StandardsPerformance Verification

Good fixture important but still may need cal standards to achieve desired accuracy (depends on DUT specs.)Short is easiestOpen and Thru require characterizationLoad is hardest (quality determines corrected directivity)

SHORTOPEN

THRULOAD

Slide #30

transitions may eliminate the need for microwavecharacterization, but can result in very large fixtures.

4 - 19

Gating can be used in conjunction with time-domainmeasurements to isolate the reflections of the DUTfrom those of the fixture. This is a form of errorcorrection. For time-domain gating to work effectively,the time domain responses need to be well-separatedin time (and therefore distance). There are two waysto use time-domain gating. The simpler methodcorrects for mismatch errors, but not for fixture lossor phase shift. The procedure involves calibrating atthe ends of the test cables with a standard cal kit,connecting the fixture containing the DUT, making atime-domain measurement, and defining the gate toexclude reflections occurring before and after theDUT.There is also a more accurate way to use time-domaingating, which can correct for loss and phase shift aswell as mismatch. As before, the reflection of thelaunch and other transitions must be distinguishable inthe time domain from the reflection of an open orshort in the fixture. If the fixture is small, a broadfrequency sweep will be needed to provide thenecessary resolution. To use this method, begin bycalibrating at the ends of the test cables with astandard cal kit. Connect the fixture with a short (oropen) loaded instead of the DUT. Look at thetime-domain response and use gating to remove allexcept the response of the short (or open). Return tothe frequency domain and perform a normalizationwith the time-domain gating still on. Now the DUT canbe inserted and measured in the fixture. Gatingremoves mismatch effects, while normalizationremoves the loss and phase shift of the fixture.

The short is the simplest standard to make, givingideal reflection. The open, while not difficult to "make"(it is usually the fixture containing no part or a dummypart), it is harder to characterize for our cal kitdefinition because we have to account for fringingcapacitance. The load is the hardest standard to makewell. The quality of the load will determine ourcorrected system directivity which in turn determineshow much uncertainty we will have for reflectionmeasurements. For the thru standard, we need toaccurately know the impedance and length of thetransmission line. These values must then beincorporated into the cal-kit definition of the thru.Let's examine each standard in a little more detail.



Determining Open Capacitance

Perform calibration at ends of cablesInsert short in fixture and addport extension until flat 180 phaseInsert open, read capacitance from admittance Smith chartEnter capacitance coefficient(s) in cal-kit definition of open

o

Watch out for "negative" capacitance (due to inductive short)adjust with negative offset-delay in open <or>positive offset-delay in short

CH1 S22 1 U FS


Cor

Del

PRm

1

1: 228.23 uS 1.2453 mS 209.29 fF 947.000 MHz


Open Standard

Ideal: unity reflection with no phase shiftActual model accounts for fringing capacitance (a concern around 300 MHz and above)Open can be empty fixtureIf compliant contacts, depress with "dummy" partInclude pads to account for pads on PCB

Slide #32 Slide #33

standard (without normalization). We see about a 7 dBimprovement in return loss at 947 MHz using time-domain gating, resulting in a return loss of 45 dB.

The plots above show the performance of our thru

4 - 20

Designing a good fixture solves a large portion of ourcomponent-test problem, but we may need a set ofquality in-fixture calibration standards before we canmake accurate enough measurements on our DUT.Fortunately, making good SOLT standards is not adifficult task in the RF range.

Slide #34

The electrical definition of an ideal short is unityreflection with 180 degrees of phase shift. This meansall of the incident energy is reflected back to thesource, perfectly out of phase with the reference. Asimple short-circuit from signal conductor to groundmakes a good short standard. For example, the shortcan be a metal slug or bar, similar in shape to the DUT.If coplanar transmission lines are used, the shortshould go to both ground planes.To reduce the inductance of the short, avoid excesslength. A good RF ground should be nearby the signaltrace to accomplish this. If the short is not exactly atthe contact plane of the DUT, an offset length can beentered as part of the user-cal kit definition (in termsof electrical delay).

Slide #35

The electrical definition of an ideal open is unityreflection with no phase shift. The actual model for theopen, however, does have some phase shift due tofringing capacitance. The open can simply be thefixture with no part in place. When compliant pins areused, it is best to depress the pin with a non-metallic"dummy" part shaped like the DUT, to avoid excess pinlength affecting the fringing capacitance. The part canbe made out of a small piece of PCB, ceramicsubstrate, or plastic. It may be desirable to include asmall pad on the dummy part to account for thecapacitance of the PCB pads used in the actual circuit.Determining the fringing capacitance for our cal-kitdefinition is only worth doing above 300 MHz or so.The fringing capacitance can be measured directly asfollows: first perform a one-port calibration at the endof the test cable using the closest connector typebefore the fixture (e.g., APC 3.5 for an SMA fixture).Next, connect the fixture and insert the short standard.Set the port extension to get a flat 180o phaseresponse. To fine-tune the value of port extension, setthe reference value of the trace to 180o and expand thedegrees-per-division scale. Mismatch and directivityreflections may cause a slight ripple so use your bestjudgment for determining the flattest trace. Now,remove the short, insert the open standard, change themarker display format to an admittance Smith chart.This displays G+jB instead of the more common R+jXof an impedance Smith chart. Admittance must beused because the fringing capacitance is modeled as ashunt element, not a series element. The fringingcapacitance (typically .03 - 0.25 pF) can be directlyread at the frequency of interest using a trace marker.At RF, a single capacitance value (Co) is generallyadequate for the cal-kit definition of the open.

4 - 21



Adjusting Load with Time-Domain Gating

Use time-domain gating to see load reflections independent from fixtureDo compensate for mismatch at or after contact planeDo not compensate for fixture mismatch


START .050 GHz STOP 6.000 GHz

Gate

C

PRm

1

1: -38.805 dB 947 MHz

CH1 S11 Re 100 mU/ REF 0 U

CorPRm

1

2

2: 159.74 mU 749 ps

1: -61.951 mU 707 ps

START .5 ns STOP 1.5 ns

Fixture mismatch (don't compensate)

Load mismatch (do compensate)


Load Standard

Ideal: zero reflection at all frequenciesCan only approximate at bestNeeds to match impedance of line connecting to DUTTwo 100-ohm resistors in parallel better than a single 50

CH1 S11 5 dB/

STOP 6 000.000 000 MHz

Cor

PRm

CH2 MEM log MAG REF 0 dB5 dB/

Cor

PRm

1

1

2

1: -41.908 dB 1 GHz

One 50-ohm SMT resistor

START .300 000 MHz

2

12 2: -32.541 dB 3 GHz

Two 100-ohm SMT resistors

1: -24.229 dB 1 GHz

REF 0 dB

2: -14.792 dB 3 GHz

Slide #36

In some cases, a single capacitance number may notbe adequate, as capacitance can vary with frequency.This is particularly true for measurements that extendwell into the microwave frequency range. Most cal-kitdefinitions allow a third-order polynomial to be usedto describe the fringing capacitance versus frequency.The polynomial is of the form C0 + C1f + C2f

2 + C3f 3.

The user must fit the measured data to this polynomialto determine the correct capacitive coefficients.When measuring the fringing capacitance, a problemcan arise if the short standard is electrically longerthan the open standard. The impedance of the opencircuit will appear to be the result of a negativecapacitor. This is indicated by a trace that rotatesbackwards (counter-clockwise) on the Smith chart.The problem is a result of using the longershort-standard as a 1800 phase reference. Theelectrically-shorter open will then appear to havepositive phase. A remedy for this is to decrease theport extension until the phase is monotonicallynegative. Then the model for the open can have anormal (positive) capacitance value. The value ofnegative offset-delay that needs to be included in theopen-standard definition is simply the amount the portextension was reduced. In effect, we have now set thereference plane at the short. Alternatively, the offsetdelay of the open can be set to zero, and a smallpositive offset delay can be added to the model of theshort standard. This will set the effective referenceplane at the open.

Slide #37

An ideal load reflects none of the incident signal,thereby providing a perfect termination over a broadfrequency range. We can only approximate an idealload with a real termination because some reflectionalways occurs at some frequency, especially withnoncoaxial standards. The load standard should matchthe impedance of the signal line at the contact plane ofthe DUT. Typically this is 50 ohms, but could bedifferent if matching is used.At RF, we can build a good load using standardsurface-mount resistors. Sometimes it is better to usetwo 100 ohm resistors in parallel instead of a single 50ohm resistor, as parasitics are decreased. Forexample, "0805" size SMT resistors have about 1.2 nHseries inductance and 0.2 pF parallel capacitance. Twoparallel 100-ohm 0805 resistors have about a 20 dBbetter match than a single 50 ohm resistor.Time-domain gating can be a very useful tool toevaluate how well our load is performing. We can gateout the response of the fixture and just look atreflections due to the load standard, provided we canget enough spatial resolution (this may require the useof a microwave vector network analyzer).

4 - 22



Thru Length

Offset loss (G-ohm/sec) = Zo * thru loss (dB@ 1 GHz) thru length (sec) * 10 log (e)

SHORT

Mathematically extend reference plane to get flat 180 phase (subtract any offset delay of short definition)

Two-port calibration

o

THRU

Adjust electrical delay until flat phase (value is electrical length of thru)

Apportion loss appropriately between fixture and thru

Reflection measurement:

Transmission measurement:


Thru Standard

Thru usually simple transmission line between contact padsDesire constant impedance and minimal mismatch at endsUse time-domain to measure impedance and tweak mismatchLength must be measured and entered in cal-kit definition

Slide #38

It is possible to adjust our load standard tocompensate for unavoidable parasitics that degradethe reflection response. It is important that we onlycompensate for mismatch at or after the DUT contactplane (e.g., due to an imperfect load or contactmismatch), but not before contact plane (i.e.,reflections from the fixture itself). If we compensateour load to counteract fixture mismatch, we will beadding reflection error during calibration instead ofremoving it. However if the contact area hasmismatch (due to a pad, for example) and it isrepresentative of the contact in the end application,compensating the load to account for mismatch isgood. Now if the device is tuned in the fixture foroptimum return loss, the effect of the pad mismatchwill also be tuned out. Compensation may take theform of a shunt capacitance or series inductance forexample. Techniques for adding or subtracting smallamounts of reactance include adding excess solder,cutting traces or pads, and adding wire mesh oradhesive-backed copper tape.Time-domain gating is an excellent tool for helping todetermine the proper compensation. We can easily seeif mismatch is inductive or capacitive. When switchingfrom a frequency-domain display of reflection (inlog-magnitude format) to a time-domain display, besure to change the data display to real format.In the above plot, we see that our load standard lookssomewhat inductive. This is most likely due toexcessive length of our feed-throughs, which could beshortened if we reduced the thickness of the PCB.Even with the inductive nature of the load, wemeasure a respectable return loss (using time-domaingating) of 39 dB at 947 MHz. The capacitive dip justprior to the load is part of the fixture itself. The effect



Performance Verification

Fixture BasicsFixture DesignCalibration TechniquesDesigning SOLT Cal StandardsPerformance Verification

Verification standards allow us to estimate residual measurement errors (due to imperfect cal standards)Can calculate measurement uncertainty from residual errorsCannot measure corrected performance by remeasuring cal standards (this would only give indication of measurement repeatability)Most versatile verification standard: electrically-long transmission line


Response versus Two-port Calibration


Cor

Del

PRm

CH2 S11&M log MAG REF 0 dB5 dB/

START .922 000 000 GHz STOP .972 000 000 GHz

Cor

Del

PRm

4 - 23

of this dip is removed when using two-port error correction.

Slide #38

The thru standard is usually a simple transmission linebetween the appropriate input and output pins or padsof the fixture. A good thru should have smallmismatches at the input and output contacts, andmaintain a constant impedance over its length (whichis generally the case). Nominally, the impedance of thethru is the same as the reference impedance of thesystem. Even if the geometry of the thru is slightly off,resulting in an impedance value that is say 51 ohmsinstead of 50, we can correct for this by changing theuser-cal kit definition to the measured value.Time-domain tools are an excellent way to measurethe impedance of the thru standard. As we justdiscussed with the load, compensation should be madefor input and output contact mismatch (at either endof the thru), but not for any fixture mismatches.

Slide #40

We also need to know the electrical length of the thrustandard. Most default cal-kit definitions assume azero-length thru, whereas SOLT-based fixtures requirea finite length thru. We can measure the electricallength of the thru quite easily as follows: first performa two-port calibration at the ends of the test cablesusing standard coaxial standards. Connect the fixtureand insert the short standard. Using a reflectionmeasurement, set the value for port extension toremove delay up to the contact plane (remember tosubtract any offset delay built into the shortdefinition). Repeat for the other side of the fixture.Now the thru can be inserted and measured with atransmission measurement. Measure group delay oradjust electrical delay until the phase response is flat.The value for the delay is the electrical length of thethru, which we can now enter in the cal-kit definitionas offset delay. For short thru-standards, it is usually not necessary toaccount for any loss since it is likely to be very small.However, if desired, a value for the loss of the thru canbe entered in the cal-kit definition as offset loss. Offsetloss accounts for loss due to skin effect, whichincreases as the square-root of frequency. Offset loss(in G-ohms/sec) can be computed from the measuredloss at 1 GHz as follows:

Zo∗loss(dB@1GHz)electrical−length(sec)∗10 log(e)

If offset-loss is accounted for, the overall measuredloss of the fixture and thru must be properly appor-tioned. One easy way to accomplish this is to scalebased on length. For example if the port extension is100 ps on one port, 150 ps on the other, and the lengthof the thru line is 750 ps, then 75% of loss is due to thethru. Use engineering judgment if the thru is signif-icantly more or less lossy than the fixture (e.g., the



Terminated Transmission LineUsed to estimate corrected directivityIf verification load matches calibration load:

peaks at 1/2, valleys at 1/4 wavelengthsdirectivity is 6 dB below value of peak

If verification load is better than calibration load:read reflection trace directly

LOAD

S11 log MAG 5 dB/ REF 0 dB


Cor

Del

PRm

1

22: -13.713 dB 6.000 GHz

1: -31.161 dB 0.947 GHz


Residual Error Terms

Main sources of residual error (after calibration):Five in each direction (forward and reverse)

directivityreflection trackingtransmission trackingsource matchload match

Not considering isolation (adds two more terms)only important for high insertion-loss measurementscan easily be cal'd out if independent of device matchleakage across fixture often is dependent on DUT match (more difficult to remove in this case)

thru is on PCB, and the fixture uses a ceramic substrate).

4 - 24

Slide #41

Now that we have covered the design of both fixturesand calibration standards, let's see how good a job wedid with our fixture design. The above plot shows thepassband performance of the 947 MHz GSM filtermeasured in the Agilent-designed fixture. In one case, afull in-fixture two-port calibration was done, and in theother, an in-fixture response calibration was done. Thetwo sets of data are quite close, indicating that the rawperformance of our fixture is very good, at least atthese frequencies. Since this particular DUT does nothave a very stringent return-loss specification, ourfixture mismatch is small in comparison. If the returnloss of the DUT were better, a response calibrationmay not yield enough measurement accuracy.For this particular DUT, we have achieved our goal ofdesigning a test fixture with good enough performanceto eliminate the need for two-port calibration. A simpleresponse calibration gives good results. However, hadwe not had the calibration standards to do a two-portcalibration, we would not have been able to make thiscomparison. It is a good idea to have calibrationstandards available to characterize the performance ofour fixture, even if they are not used forerror-corrected measurements during the manufactureof our DUT.

Slide #42

Once our fixture and calibration standards have beenfabricated and we have performed a two-portcalibration, we may wish to confirm that we aremaking good measurements. This process is calledverification, and involves further measurements on aset of passive devices called verification standards.Measuring these verification standards will give us agood indication of residual measurement errors, whichare the result of slight imperfections in the calibrationstandards (it is impossible to make perfect calibrationstandards). Once we know the residual error terms, wecan calculate measurement uncertainty for anyparticular measurement of an actual DUT. Theverification standards are specifically designed tohighlight the various effects of residual errors. Wecannot measure our corrected performance simply byremeasuring the calibration standards, as this wouldonly give an indication of our measurementrepeatability.The most versatile verification standard is anelectrically-long length of transmission line. The linemust be long so that systematic error terms becomeout-of-phase with respect to the error terms measuredduring calibration. OSL calibration standards areplaced at the end of the line to highlight variousresidual error terms. Our approach to building longtransmission lines was to use semi-rigid coaxial cablewith an SMA connector on the end to allow differentcoaxial standards to be connected.

4 - 25



Reflection Tracking - 100 ohm ResistorMeasuring known reflection also indicates reflection trackingEffect of source mismatch reducedby reflection of mismatch standardExample: 100 ohm resistor (9.5 dB RL)reduces effect of source match by 9.5 dB

100

log MAG .5 dB/CH1 S22 REF -9.54 dB

START .030 000 MHz STOP 6 000.000 000 MHz

Cor

Hld

PRm

1

1: -9.94 dB 6.00 GHz


Short / Open On Transmission Line

SHORT

OPEN

Ripple combination of directivity, source match, reflection trackingAssume source match is dominant contributorMeasure open, store in memory, then measure shortSource match related to peak-peak ripple as follows:Residual source match (dB) = 20 log (1 - 10 / 1 + 10 )Open/short average indicates reflection tracking

(p-p)/20 (p-p)/20



Cor

Del

PRm 1

2

1: Ref 2: -.797 dB -749.439 MHz

Slide #43

Let's briefly recall the major sources of residualmeasurement error that remain even after a one- ortwo-port calibration has been done. There are fivemain terms in each measurement direction (forwardand reverse). Reflection and transmission trackingindicate how well two receiver channels track overfrequency during ratioed measurements, which isimportant in reflection measurements such as returnloss, VSWR, and impedance, and in transmissionmeasurements such as gain or insertion loss, andisolation. Load and source match are indications ofhow close we have made our test ports to the systemreference impedance (usually 50 ohms). Finally,directivity is the effective leakage term of oursignal-separation devices, which are usually couplersor bridges that are integrated into the networkanalyzer. This term is critical for reflectionmeasurements.One additional source of error that we will not attemptto verify is crosstalk. Crosstalk is only important whenmaking high-insertion loss transmissionmeasurements, such as when measuring the stopbandof high-rejection filters or when measuring amplifierswith extremely good isolation. Crosstalk can beeffectively removed with calibration if it isindependent of device match, for example, if directsource to receiver leakage occurs. If the crosstalk isdependent on the port matches of the DUT, then it ismore difficult to remove. This can occur if there isradiation leakage in the fixture or some alternatepropagation mode, both of which are generallydependent on the match presented by the DUT.Another source of match-dependent crosstalk canoccur if high-level signals in the reflection receiverleak into the transmission receiver.

4 - 26



Summary

Most important goal: repeatable measurementsEssential elements of a good fixture:

place and clamp part repeatablymaintain Zo across transitions to minimize reflectionsprovide good RF signal and ground connectionsuse compliant contacts when appropriate

No one correct solution for any given DUTcan meet design objectives with different approachesdifferent approaches can be evaluated based on measurement accuracy, throughput, ease of replacement, cost, etc.

Error-correction depends on quality of fixture and DUT's specs


Transmission ResidualsConnect long transmission line as thruReflection measurement shows residual load match (load match is 6 dB below peak)Transmission measurement shows transmission tracking

CH1 S11 log MAG 5 dB/ REF 0 dB


Cor

PRm

1

2

1: -29.108 dB 1.108 GHz2: -9.715 dB 5.340 GHz

CH1 S 21 log MAG .2 dB/ REF 0 dB

Cor

PRm

1

2

1: Ref

2: -.202 dB -454.914 MHz


The first verification standard we will consider is aterminated transmission line. This will give us a goodestimate of the corrected (residual) directivity of ourtest fixture and network analyzer. The transmissionline should be the same impedance as the systemreference and should be long enough to generate rippleover the frequency range of interest. A minimum of aquarter wavelength at the high end of the frequencyrange is required, but several wavelengths is better asit allows characterization to lower frequencies. Thetermination at the end of the line should match thecalibration standard as closely as possible. We willnow generate a ripple pattern in the reflectionmeasurement with valleys occurring at frequenciesthat are odd numbers of quarter wavelengths of theline, and peaks occurring at integral numbers of halfwavelengths. The value of the peak will be 6 dB abovethe actual return loss of the load. This is whatdetermines our system directivity, as we can nevermeasure better directivity than the return loss of thecalibration termination. The effective systemdirectivity is simply equal to the peak (in dB) minussix.What happens if the load used as the verificationstandard is much better than the one used duringcalibration? Then our residual directivity can be readdirectly from the corrected reflection measurement.An example of this is shown above, where a coaxialload standard was used at the end of a long line. Ourresidual directivity is about 31 dB at 947 MHz,degrading to only 14 dB at 6 GHz. While we haveplenty of directivity at the center frequency of ourfilter to perform accurate reflection measurements, wewould have to improve the bandwidth of our loadstandard to make accurate measurements at 6 GHz.

The next verification standard is a short or open (orboth) placed at the end of a transmission line. Again,the line should be long enough to generate ripples.This time, the reflection ripples will be caused by acombination of the residual directivity, source match,and reflection tracking. A good estimate of the residualsource match can be achieved by assuming the rippleis primarily due to source match, as it is usuallysignificantly worse than residual directivity. Reflectiontracking generally gives an overall slope to the ripplepattern. With this assumption, the residual sourcematch can never be better than the residual directivity,as we can never see the reflection effects of less signalthan that which leaks through due to directivity.The peak to peak ripples can be seen more easily if thewe first measure the open and store the trace inmemory, and then measure the short and display bothtraces. The envelope of the ripples represent thecombined error. The average of the open and shortripples is a good estimation of residual reflectiontracking. Any loss of the verification line, however,will be added (twice) to the residual reflectiontracking. Residual source match can be estimated (worst case)as follows:

Residual source match (dB) = 20 log( 1−10−(p−p)/20

1+10−(p−p)/20 )

For example, the above plot shows a peak-peak rippleof 0.797 dB around 2 GHz. This is equivalent to a worstcase source match of or -26.8 dB.20 log( 1−10(−0.797)/20

1+10 (−0.797)/20 )

4 - 27


Another good way to estimate reflection tracking is tomeasure a known reflection. This is easilyaccomplished by measuring the reflection of a single100-ohm resistor for example, giving a SWR of 2:1 (areturn loss of 9.54 dB). Here again, the ripple on themeasurement is a combination of directivity, sourcematch, and reflection tracking, but the effect of animperfect source match is decreased by the reflectioncoefficient of the mismatch standard (9.54 dB in thiscase, or nearly 70%). For the worst case, we canassume that the trace ripple of this measurement issolely due to reflection tracking.

Now that we have looked at the major sources ofresidual reflection errors, let's look at the maintransmission residuals. The first one we will consideris residual load match. Any mismatch of the thrustandard used during calibration will contribute toresidual load match. To verify this term, simplymeasure the reflection of a long transmission lineconnected as a thru across the fixture. This line shouldbe a different length than the one used for the thrucalibration, and should be greater than a quarterwavelength to generate sufficient ripple. Themeasurement is essentially identical to the onepreviously described for measuring residualdirectivity. In this case, the peak of the ripple is 6 dBabove the residual load match. In our example above,we see a residual load match of about 35 dB (29 + 6)around 1 GHz.Finally, let's check residual transmission tracking. Wecan use the same long transmission line, but wemeasure transmission instead of reflection. Thepeak-peak ripple in this measurement is residualtransmission tracking, and is due to the raw sourcematch times the residual load match plus the raw loadmatch times the residual source match( ).ρs(raw) ∗ ρl(resid) + ρ l(raw) ∗ ρs(resid)

4 - 28


We have covered quite a bit of information ondesigning and calibrating high-quality test fixtures forSMT devices. The most important goal of fixturedesign is to ensure repeatable measurements. To thisend, we must make sure that the part is placed andclamped in the fixture in a repeatable manner. Next,we have to maintain the proper Zo between the coaxialconnectors and the DUT, across any and alltransitions, to minimize reflections. And, we have toprovide good RF signal and ground connections, usingcompliant contacts when appropriate. Even withinthese parameters, we have seen that for any particularDUT, there are different design approaches that can betaken, all of which may yield acceptable results. Thebest approach will depend on the mechanical andelectrical specifications of the part to be measured, theoverall test goals (including the required measurementaccuracy), and the cost to implement the fixture.We discussed several forms of error-correctionincluding time-domain techniques and using physicalcalibration standards. We showed how to make yourown SOLT standards for calibration, and how to verifycalibrated performance with verification standards.The appropriate level of calibration depends on thequality of the fixture and the DUT's specifications.This paper attempted to provide enough tools andguidelines to allow anyone to make effective RFfixtures and calibration standards that will yieldrepeatable and accurate measurements ofsurface-mount devices. If your expertise is not in RFdesign or fixture fabrication, Inter-Continental Microwave may help solve your test-fixturing needs,from simple surface-mount devices, up to more complex RFICs.

4 - 29


References

Specifying Calibration Standards for the Agilent 8510Network Analyzer, Agilent Technologies Application Note 8510-5B, Literature number 5956-4352, April 23, 2004

In-Fixture Microstrip Device MeasurementsUsing TRL* Calibration, Agilent Technologies Product Note 8720-2, Literature number 5091-1943E, August 2000

www.agilent.com

Agilent Technologies’ Test and Measurement Support, Services, and AssistanceAgilent Technologies aims to maximize the value you receive, while minimizing yourrisk and problems. We strive to ensure that you get the test and measurementcapabilities you paid for and obtain the support you need. Our extensive supportresources and services can help you choose the right Agilent products for yourapplications and apply them successfully. Every instrument and system we sell has a global warranty. Two concepts underlie Agilent’s overall support policy: “OurPromise” and “Your Advantage.”

Our PromiseOur Promise means your Agilent test and measurement equipment will meet its advertised performance and functionality. When you are choosing newequipment, we will help you with product information, including realistic perfor-mance specifications and practical recommendations from experienced test engi-neers. When you receive your new Agilent equipment, we can help verify that itworks properly and help with initial product operation.

Your AdvantageYour Advantage means that Agilent offers a wide range of additional expert testand measurement services, which you can purchase according to your unique tech-nical and business needs. Solve problems efficiently and gain a competitive edge by contracting with us for calibration, extra-cost upgrades, out-of-warranty repairs,and onsite education and training, as well as design, system integration, projectmanagement, and other professional engineering services. Experienced Agilentengineers and technicians worldwide can help you maximize your productivity, opti-mize the return on investment of your Agilent instruments and systems, and obtaindependable measurement accuracy for the life of those products.

United States: Korea:(tel) 800 829 4444 (tel) (080) 769 0800(fax) 800 829 4433 (fax) (080)769 0900Canada: Latin America:(tel) 877 894 4414 (tel) (305) 269 7500(fax) 800 746 4866 Taiwan:China: (tel) 0800 047 866(tel) 800 810 0189 (fax) 0800 286 331(fax) 800 820 2816 Other Asia PacificEurope: Countries:(tel) 31 20 547 2111 (tel) (65) 6375 8100Japan: (fax) (65) 6755 0042(tel) (81) 426 56 7832 Email: [email protected](fax) (81) 426 56 7840 Contacts revised: 05/27/05

For more information on Agilent Technologies’ products, applications or services,please contact your local Agilent office. The complete list is available at:

www.agilent.com/find/contactus

Product specifications and descriptions in this document subject to change without notice.

© Agilent Technologies, Inc. 1996, 2005Printed in USA, July 11, 20055989-3247EN


Third-Party Companies

Fixtures:Inter-Continental Microwave1515 Wyatt DriveSanta Clara, California 95054-1586USAPhone: (408) 727-1596Fax: (408) 727-0105

Test Sockets:Johnstech International1210 New Brighton BoulevardMinneapolis, MN 55413-1641USAPhone: (612) 378-2020Fax: (612) 378-2030

Conductive Elastomer:Chomerics North America77 Dragon CourtWoburn, MA 01888Phone: (781) 935-4850Fax: (781) 933-4318

www.agilent.com/find/emailupdatesGet the latest information on the products and applications you select.

Agilent Email Updates

agilent designing and calibrating rf fixtures for surface...

Documents