On-Wafer Vector Network Analyzer Calibration and Measurements

The Vector Network Analyzer or VNA has become the workhorse of most network measurements above 1 GHz. Getting the best on-wafer measurement results requires a solid understanding of measurement system components and their interaction. This application note is intended to introduce on-wafer vector measure- Port-l b_ ) Port-2 ments and provide references for fur- +Cable y3 DUT _ ther study. a2 VNA Calibration Fundamentals The representative VNA block diagram (see Figure 1) shows the key elements of the analyzer. The Figure 1. Main hardware blocks in a Vector Network Analyzer (VNA). RF source at the top provides the system ends and the DUT begins. erence plane. If we ignore any non- device-under-test (DUT) stimulus. In on-wafer VNA measurements idealities of the cables, couplers, A forward/reverse switch directs the this boundary is known as the “ref- mixers, and so forth, then the ratios RF energy to either DUT port. The erence plane” of the measurement of wave amplitudes (a’s and b’s) test set uses directional couplers or and will often be located at the inside of the machine correspond bridges to pick off the forward and probe tips. A typical wafer probing directly to the DUT's S-parameters. reverse waves traveling to and from VNA setup is shown in Figures 2 For example: each port, and down-converts these and 3. signals to four IF sections. Each IF b,/a, = S, 1 of the DUT section filters, amplifies, and digi- Ideally the system will measure tizes the signals for further digital the characteristics of whatever is VNA calibration is the process of processing and display. connected to the measurement ref- measuring devices with known or A VNA measures vector ratios of reflected or transmitted energy to energy incident upon the DUT. AS a stimulus-response measurement, a VNA measurement determines the properties of devices rather than the properties of signals. Signals would be measured by instruments such as oscilloscopes or spectrum analyzers.

A significant challenge in stimulus-response measurements is defin- ing exactly where the measurement

Figure 2. A typical RF device characterization setup is shown above. On the left is a vector network analyzer, in this case an Agilent 8510 system. Cables from the test set ports route over the ® probe station to a pair of microwave probes which are aligned through the microscope or camera CASCADE system onto calibration substrates and test devices. The right side of this picture shows the PC that controls the prober and runs the calibration and measurement software. 1 Figure 3. Close-up of microwave probes and cables on positioners providing probe planarity adjust- ment and cable

Figure 5. The Smith Chart is a polar graph of reflection coefficient with grid lines showing normalized impedance values.

partly known characteristics and tors or at the ends of cables. 1 +l- = r+ix using these measurements to estab- Calibration functions allow the zL =zo - 1 -I- lish the measurement reference user to store measurements of stan- [ 1 planes. Calibration also corrects for dards, compute the error models, the imperfections of the measure- and automatically apply corrections V reflected zL - zll ment system. These imperfections to DUT measurements. = r= Z not only include the non-ideal V incident L + z. nature of cables and probes, but Where the ohmmeter’s calibration also the internal characteristics of simply used a short circuit (con- where Z,= the reference imped- the VNA itself necting the test leads together) to ance of the measurement system determine the extra resistance term The VNA is calibrated in much in the error model, a VNA uses The Smith chart (see Figure 5) is the same manner that the “zero” multiple calibration standards - a polar graph of the reflection coef- function on your ohmmeter sub- typically open circuits, short cir- ficient with grid lines showing the tracts out the resistance of the test cuits, loads, and through (thru) normalized impedance values. All leads. On an ohmmeter, when you connections. passive impedances are represented activate the “zero” function, it in a compact unit circle reflection stores a resistance measurement Measuring Impedance format. Typically the vna auto- which is then subtracted from all mates the display of a wide variety future measurements. The obvious The measured ratios of the ai and of formats: Smith Chart, polar, and error model is simply a resistance in bi wave amplitudes are called linear and log magnitude formats. series with the test port. Corrected S-parameters. S-parameters are just VNA measurements are referred to one type of network representation Calibrated VNA Accuracy as “deembedded.” used for linear, small-signal, a.c. analysis. As with any corrected measure- A VNA does the same thing as ment, the absolute accuracy of a the ohmmeter but uses a more A reflection ratio at a device port, calibrated VNA measurement is complex error model with several with all other ports terminated, is determined by the techniques and terms for each frequency point. The also known as a reflection coeffi- completeness of the error model measurement system is described as cient. The load impedance, Z,, is used, the accuracy of the descrip- an ideal VNA with an error adapter related to the reflection coefficient, tion of the reference devices (cali- network that models all of the sys- I?, by the bilinear transform pair as bration standards), and the repeata- tem’s non-idealities: directivity of follows: bility of the measurement system. the couplers, imperfect match looking back into the reflectometer (test set ports), imperfect frequency P 1 response of the reflectometers and Forward the transmission between ports, and Error DUT the crosstalk between ports. This Adapter Pl error model is shown in Figure 4.

VNA's rely on calibration for Port 2 accuracy even in measurements 3 with the reference plane defined at the instrument front panel connec- Figure 4. Error model for a typical network analyzer. All errors are modeled by an error adapter in front of an otherwise perfect system. 2 If performed improperly, calibra- VNA error models are based VNA Calibration Options and tion can introduce errors. The upon the use of S-parameter repre- Standards impedances used in the calibration sentations of network properties. S- The VNA will measure the must be accurately known and parameters are signal flow and uncorrected S-parameters if not cal- entered. Assumption of ideal transmission line based. Only a sin- ibrated, though not very accurately. behavior of standards is a recipe for gle mode of propagation at device Uncorrected measurements are errors, but worse things can hap- terminals is assumed. Situations rarely used (see Figure 7). pen. For example, improperly that violate this assumption such as entering a description of a short using waveguides that can propa- circuit or open circuit standard will gate multiple modes, radiation, or lead to useless results that may not parasitic coupling between net- always be obviously wrong. The works other than at the device ter- VNA will believe everything that minals are not properly handled.

you tell it. Accurate descriptions of l Convenient the electrical behavior of the vari- If a second mode, radiation, or l Generally not accurate ous calibration standards must be extra parasitic doesn’t change for l No errors removed supplied to the VNA. any and all devices being measured then it will drop out with the cali- Figure 7. Uncorrected VNA. Corrected measurements rely on bration. Usually, however, these A “response” calibration is simply the repeatability of the measure- extra modes have behavior that is a vector magnitude and phase nor- ment system. Just as an ohmmeter DUT dependent and VNA calibra- malization of a transmission or cannot correct for a changing resis- tion will not account for the reflection measurement, used only tance of its test leads or drift in the effects. A clean, well-designed ohmmeter circuitry, a VNA cannot probing system (including die pad at low frequencies (See Figure 8). correct for random errors such as pattern) with good quality trans- 0 0 noise or dynamic range, cable mission-like interconnections will thru repeatability, or instrument drift. minimize these errors as much as Any change in the measurement possible (see Figure 6). due to, for instance, drift of the VNA test set, thermally induced An extensive discussion of VNA cable length changes, or even the calibration and accuracy is included l Use when highest effects of noise due to VNA in Calibration and Accuracy accuracy is not require’ dynamic range can invalidate the Factors Summit High-Frequency l Removes frequency response error correction. Sensitivity of cable elec- Probe Station Reference Manual. l trical performance (such as phase Figure 8. Response calibration. delay) to environmental changes is a significant element of quality and Typically a full calibration of all the suitability for VNA use. error parameters is used as a reference or to assure the highest accuracy (see Figure 9). 2 ground

contact pitch 100 - 150 pm

signal DUT I-0 4I I Highest accuracy for P-port 100 - 150 pm devices Removes these errors - directivity - source match - load match ground - reflection tracking - transmission tracking - crosstalk

Figure 6. Recommended microwave pad pattern for each measurement port. Figure 9. Full 2-port calibration. On-wafer standards, fabricated on the same wafer as the DUT, are sometimes desirable since the probe-to-standard transition can be designed to be very similar to the transition to the DUT. This is helpful at frequencies above 20 GHz since a primitive transition may introduce extra parasitics or modes which, since they are occurring at the probe-tip reference plane, may not be corrected by calibration. In many cases these transition errors may be deembedded using simple lumped element equivalent circuit models.3

The use of off-wafer standards on Figure 10. The LRM Impedance Standard Substrate (ISS) has multiple repeated patterns of an Impedance Standard Substrate Short-Load-Thru Standards and alignment marks for setting proper probe separation. A set of is very practical and Cascade transmission lines are also provided for TRL Calibration. Microtech has demonstrated calibration results comparable to Calibration Standards pitch used. The calibration data is results obtained by using the spe- therefore, supplied with the probe, cial on-wafer standards methods On-wafer calibration standards rather than being a single value for recommended by the United States most often are precision thin-film a standard as would be the case for National Institute of Standards and resistors, short-circuit connections, coaxial standards where there is no Technology. In cases where the and 50 ohm transmission lines fab- ambiguity about what connections transition on-wafer is dramatically ricated either on the wafer contain- can be made to it. Calibration data different than that of the ISS, such ing the DUT or on a separate is normally specified for a specific as on silicon substrates, the use of Impedance Standard Substrate probe spacing used with a particu- a set of dummy pads on-wafer (ISS). Cascade Microtech trims lar impedance standard substrate, allows the removal of additional high-performance ISS load resis- and is only valid for a particular pad parasitics. Pad parasitic tors to within 0.3% of their probe-ISS combination. The con- removal algorithms are discussed in desired d.c. value (usually tact pattern, e.g., ground-signal- On the Characterization and 50 ohms). Open circuit standards ground or ground-signal, (see Optimization of High-Speed Silicon are normally implemented by rais- Figure 11) will also impact calibra- Bipolar Transistors.4 ing the probe in air above the tion data (see Table 2). wafer by 250 pm or more. A very useful feature to look for on probe stations is an auxiliary A true 'thru' standard does not chuck or independent vacuum exist for on-wafer measurements hole pattern specifically for hold- since probes cannot directly con- ing small substrates such as an ISS. nect to each other and must The ability to have both the test instead use a very short transmis- wafer and the calibration standards sion line as a thru standard. A thru immediately accessible eliminates line standard may be referred to frequent swapping of the wafer either with the label of thru or and the ISS. In Cascade Microtech line. For the Cascade Microtech thermal chuck systems the auxil- LRM ISS (see Figure 10) the thru iary chucks are insulated from the lines have 1 ps delay for the ver- thermal system, allowing better sion suitable for probes with con- control of the standards. The thin tact pitch (center-to-center spac- film load resistor has a temperature ing) of 250 pm or less, and 4 ps dependence, making the auxiliary delay for the wide-pitch version chuck system a must for obtaining used for probes with more than the most accurate results. 250 pm pitch. Figure 11. Microwave probe contact patterns. The Ground-Signal-Ground (GSG) pat- It is important to note that the tern has higher performance than the specific electrical behaviors of the Ground-Signal (GS/SG) configuration. standards depend upon the probe Transistor Characterization Measurements In characterizing devices on- wafer, the goal is to obtain the electrical behavior of the intrinsic device. This is the transistor without the parasitics associated with GSG Test Device Open Pads & Metal Shorted Metal bond pads and interconnections. (Remove Y pad) (Remove Zmetal)

Figure 12. Pad Parasitic Removal structures. Cascade’s WinCal software supports pad parasitic removal, when in pad measurements are signifi- The General Purpose Membrane measurements of identical pad con- cant. Also, series inductance from ISS (see Figure 15) addresses the figurations without an active device layout inductance and/or probe problem of thru length by provid- are used to deembed intrinsic placement errors can resonate with ing a number of microstrip and device behavior (See Figure 12 & device output capacitance and dis- coplanar waveguide (CPW) lines 13). tort f,. results (See Figure 14).

Magnitude of l-l,, (db) 30 , Calibration Standards for Pyramid ProbeTM Cards Calibration standards for use with Pyramid Probe Cards differ little from those used in standard -20 ------microwave probing. In many cases 1.8GHz 3.3 GHz 0.1 1 10 traditional ISS's may be used. The Frequency (GHz) fixed input to output spacing of Figure 13. Apparent bandwidth improvement probe cards may be inconvenient, due to removal of pad parasitics. however. Conventional separate Figure 15. The General Purpose Membrane probes can be adjusted to land on Impedance Standard Substrate. After a probe-tip calibration, the pairs of standards simultaneously. open-circuit admittance parameters For probe card contact separations with a variety of straight and right of open-circuited pads are subtract- differing from the distance between angle lengths. If no line fits well ed from device measurements to standards on ISS's, separate land- enough, the next larger line can be eliminate shunt capacitances and ings for each port are required. cut to length. Loss and delay data conductances. Also, series resistance Additionally, the thru line may not for the straight CPW lines on the and inductance are eliminated by be the appropriate length to con- GP Membrane ISS are shown in subtraction of short-circuit imped- nect the input and output. Table 1. Loads trimmed to 1% ance parameters obtained from Transmission line standards on the accuracy are provided in openings measurements of shorted pads and ISS may be used as the thru if they in a large field of metal which is interconnections. happen to be the length that also used for the short circuit stan- matches your probe contact separa- dard. Apparent device fr can be signifi- tion. The delay and loss of the thru cantly increased by eliminating pad must be accounted for in your cali- The RFC-CAL is a custom mem- parasitics. But be wary of large bration. brane with specific length thru con- increases in fT when residual errors nections mounted on a glass slide. It is sometimes useful to use a thru

Line # Delay (ps) ; 10 GHz Loss (dB) Effects of increasing L, Cd, 1 3.8 ! 0.025 2 5.8 i 0.044 Ld 3 11.5 I 0.093 4 17.1 014 5 22.8 -tm--PO:,9 6 28.3 0.24 7 34.2 0.28 .~ 8 39.7 0.34 - 9 45.3 0.38 10 52.7 0.44 11 60.4 0.52 12 67.7 0.59 log f Table 1. Properties of CPW thru lines on the Figure 14. Inductance, Ld, due to layout inductance can resonate with device output General Purpose Membrane ISS. capacitance and distort the measured frequency response. 5 structure with an inverted match- be an easy task with only three thru line standard. A 1 ps thru is ing network for calibration of coefficients per probe, but improp- normally used with the standard impedance matching probes. er calkit entry is one of the largest probe pitch ISS versions (PNs sources of error in on-wafer VNA 101-190 and 103-726) while a 4 ps A fully custom Impedance measurements. VNA front panel thru is used with the wide probe Standard Substrate is yet another data entry of calibration coeffi- pitch ISS versions (PNs 106-682 alternative. This ISS can have pat- cients can be tedious and error and 106-683). When using long terns that provide each standard prone. thru lines such as for Pyramid type to all of the microwave ports Probes it is important to use an simultaneously. A custom ISS may Some VNA's do not allow accurate delay value and to provide be particularly valuable when lumped element models for all an accurate loss value. The offset automating calibration. Contact standards. The use of an offset loss term (in GflZ/s at 1 GHz) for Cascade Microtech for details on transmission line based equivalent the HP 85 10 is given by: these options. circuit is required. The lumped element series inductance represented CR d&w 1, GHz c @o by the load (term) or short can be OffsetLoss = S 1GHz 10 log,,,(e)e Short-Open-Load-Thru effectively modeled by a short sec- Calibration tion of high impedance transmis- By far the most commonly used sion line offset. Normally the offset where dBloIJ =measured insertion method, the Short-Open-Load- impedance is set to the maximum loss at 1 GHz, 20 = thru line Thru (SOLT) calibration is avail- value allowed by the VNA (e.g., impedance (normally 50 Sz), and able on every commercially avail- 500 ohms for the HP 8510) and I= physical length of the offset. able VNA. This calibration is the the corresponding delay is deter- combination of two one-port mined using the equation For a standard specified by loss Short-Open-Load calibrations with T, = L/Z,. and delay at a particular frequency, additional measurements of a thru the offset loss term calculation sim- standard to complete the two-port For example, the 4.8 pH short plifies to: calibration. All of the standards circuit inductance of a 150 pm must be fully known and specified. pitch, ground-signal-ground configuration ACP probe would be The SOLT standards are reason- entered into an HP 85 10 calkit as a ably well modeled with simple short standard with a 9.6 fs (4.8 where dBlo, is the measured lumped elements: open-circuit pH/500 0) long 500 0 offset. insertion loss (dB) at frequency f capacitance, short-circuit induc- (GHz) for a line with offset delay = tance, load inductance, and thru Another common source of error Delay (ps) and offset impedance = delay (and loss). Often the open- is to forget to enter the delay of the 2, (usually 50 0). For example, circuit capacitance will have a nega- table 1 shows that the CPW line #6 tive value since the probe lifted in air has less tip loading than when it 25pH 25pH is in contact with a wafer. -r All standards must be accurately 20 fF contacted physically, since the inductance values are very depen- 1 dent on probe placement on the - standard. A 1 mil (25.4 pm) longi- delay = E = I ps tudinal change in the overlap of the probe tip over the standard can result in a significant change in the inductance value (see Figure 16). 15fF Typical values of calibration coefficients for individual microwave - probes (Cascade Microtech ACP delay =1/2 2, c = 0.375 ps style) are shown in Table 2.

Calibration coefficients must be Figure 16. Improper probe placement will impact the electrical behavior of a measured standard. correctly entered into the VNA to Use of alignment marks to set probe separation and reasonable care in placement will provide do any good. This would seem to good results.

6 on the general purpose membrane dards set allows the R standard to cally ‘over-determined set of equa- ISS has 0.24 dB loss at 10 GHz be any reciprocal two-port. No tions will not necessarily have a sin- and 28.3 ps delay so this equation other knowledge of the standard is gle entirely self-consistent solution. gives: required to complete the calibra- The implications of this lack of tion. self-consistency are subtle, numer- (230.3).(50).(0.24) = 30.9m OffsetLoss~, , = ous, and unavoidable. One example (28.3).x6 s The SOLR algorithm is very would be measuring slight differ- powerful when only non-ideal ences between S 12 and S,, for a An offset loss value of 31 GR/s thru standards are available: passive reciprocal structure. The should provide good results for all orthogonal probes, excess thru line SOLR calibration is not over-deter- but the shortest of the CPW lines on length, lossy or mismatched thru mined and is inherently self-consis- the general purpose membrane ISS. line, incompatible port geometries tent. (pitch), etc. The SOLR calibration The thru standard loss parameter algorithm is an advanced calibra- Advanced Calibration Methods entry with the requirement for tion method and is not directly nearly ideal behavior of the thru available on any VNA presently S-parameters require measure- standard is a major limitation of made. SOLR requires special soft- ments with each port of the mea- the SOLT calibration method. Any ware - such as Cascade sured device stimulated in turn. For loss behavior that doesn’t match the Microtech’s WinCal - running on a two-port VNA this means that ideal loss model built into the VNA a PC communicating with the the microwave source power must or extra reflections, such as those VNA via IEEE-488 instrument be switched to each port (as shown generated by a right-angle bend bus commands. Additionally, the in Figure 1). When the switch when connecting orthogonally ori- VNA must be one that supports changes position the error model of ented probes, will introduce signifi- true TRL calibration (a four sam- each port changes due to the non- cant error. Similarly, excess length pler VNA) in order to be able to ideal nature of the microwave trans- of a thru line over the spacing provide enough information for fer switch, the source output between RF contacts on a Pyramid the SOLR algorithm to be calcu- impedance, and the response termi- Probe will add parasitic shunt stubs lated. Calibration of right-angle nation load impedance. The full from the excess line length. These oriented probes using SOLR is two-port error model provides for act as open circuit stubs and will discussed in An SOLR Calibration these switching terms by the use of spoil the accuracy of an SOLT cali- for Accurate Measurement of two different error models - one bration. Orthogonal On- Wafer DUE.6 for forward and one for reverse measurements. The error model An alternative to the SOLT cali- A standard SOLT calibration can shown in Figure 17 is consistent bration is the SOLR5 (Short-Open- present the dilemma of conflicting with the SOLT calibration Load-Reciprocal) advanced calibra- information to the VNA. The method.2 tion. This algorithm is similar to number of measurements and the SOLT calibration for one-port standard definitions exceeds the Normally in on-wafer measure- corrections, but the redundant number of unknowns in the two- ments the isolation terms, E,, are information available in the stan- port error model. This mathemati- neglected since they are relatively

Table 2. Air-Coplanar Probe (ACP) calibration coefficients. Coefficients for VNA calibration depend on the style and pitch of the probe, as well as the ISS used. The wide-pitch ISS's (106-682 and 106-683) use a 4 ps long thru line while the standard-pitch ISS's (101-190 and 103-726) have a 1 ps thru line. ly high resistance transmission lines such as are found in thin film circuits. Ex :______-______~__-______; In the TRL calibration (and its variations) the thru line must be S 21 E, : ER fully known to allow the measure- . L L L:, v F ment reference plane to be moved a0 4 to the probe tips from the mid- ED S S ES 11 22 point of the thru line. In some A 1 1 applications using on-wafer stan- b0 1 u 12 dards, the DUT is effectively DUT embedded at the thru mid-point, eliminating the need for this reference plane change.

Figure 17. Signal flow graph of the full two-port VNA error model without switch corrections. The The TRL calibration method is forward stimulation case is shown, a similar error model is used for reverse stimulation. not practical for Pyramid Probes and other probe cards since it is small and are not the same for the phase differences over a wide fre- not possible to measure multiple standards and DUT. quency range. line lengths with fixed spacing probes. Even with individual VNA's that have the ability to The TRL calibration process itself probes, the TRL calibration sample the energy reflected from does not determine the characteris- requires programmable probe posi- the switch termination as well as tic impedance that is the reference tioners for fully automatic calibra- the stimulus, transmissions, and for the resulting S-parameters. tion. An alternative to the TRL is reflections - the four sampler archi- Additional assumptions and mea- the Line-Reflect-Match (LRM) cal- tecture - allow removal of the surements may be used to deter- ibration. The LRM calibration uses switching terms from raw measure- mine the 2, and renormalize the TRL-like mathematics but uses a ments. This results in a reduced S-parameters to 50 ohms (for broadband match standard to set error model shown in Figure 18 example). Often this is not done the system impedance. The match which has error boxes at each port and the frequency dependence and essentially acts as an infinitely long, that contain all of the VNA and imaginary part of Z, due to loss infinitely attenuating transmission interconnection errors. A great deal are disregarded. At low frequencies, line. of study has gone into various where the per-unit-length inductive methods for determining the coeffi- reactance is small compared to the The LRM calibration standards cients of the error boxes from mea- ohmic loss, the imaginary part of are suitable for fixed probe spacing surements of known and partially Z, can be large. This effect can be and can be used with probe cards known structures. present at moderate frequencies or manual positioners. The stan- (perhaps up to 1 GHz) for relative- dards are a subset of those used for The most fundamental of these methods, the Thru-Reflect-Line (TRL) calibration algorithm, uses the characteristic impedance of a transmission line as the reference for calibration.7 Line pairs differing only in length are measured along with a reflect standard that is known only to the sign of the DUT reflection coefficient. The line L ‘S length difference must not be near 21 180 degrees of phase since a half S SI& wavelength transmission line mim- 1 ics a zero length line and provides no additional information. In practice a line pair will only provide useful results over a 20 to 160 degree phase difference range. Figure 18. Signal flow graph of the switch corrected two-port VNA eight-term error model. Wider bandwidths require multiple Since S-parameters are ratios, only seven unknowns must be determined to complete the cali- line lengths to provide suitable bration. 8 SOLT since either the short or only one Match standard any MultiCal allows high-perfor- open can be used as the reflect ambiguity in load inductance mance calibration using on-wafer standard. In on-wafer measure- between ports is eliminated, reduc- standards and computes measurements the open standard is pre- ing the sensitivity to probe placement error bounds from calibra- ferred since it has no probe placement since the actual value is direct- tion comparison. 13 ment issues to effect repeatability. ly extracted. The conveniently available stan- One-Tier vs. Two Tier dards combined with superior per- LRRM provides high-perfor- Calibrations formance make the LRM calibra- mance calibrations with fixed probe tion a natural first step to improved separations, allowing simple, auto- In most on-wafer VNA measure- calibration accuracy over using mated calibrations. ment applications a single set of SOLT. calibration measurements are made A Cascade Microtech Summit to characterize everything at once: The one weakness of the LRM probe station using WinCal soft- the VNA errors, cables, and calibration is that the inductance ware allows highly repeatable, fully probes. There are situations where of the load standards becomes part automated VNA calibrations initi- you may want to make this a two- of the reference impedance of the ated with a single mouse click. See step process. A two-tier calibration measurements. Methods for auto- the Cascade Microtech Technical is obtained by performing one cali- matic determination of the load Brief Technique Verifies LRRM bration - the first tier - and fol- inductance and correction of mea- Calibrations for GaAs lowing it with a second calibration surements are available in Cascade measurements. 10 - the second tier - that uses first Microtech’s WinCal VNA calibra- tier corrected standards measure- tion software.8 These more sophis- ments. Two-tier calibrations nor- ticated versions of the LRM cali- NIST MultiCal mally require special software, such bration are only suitable for rela- Research by the Microwave as Cascade’s WinCal, since these tively small thru line lengths (a few Metrology Group of the U.S. functions are generally not provid- ps) but when applicable can yield National Institute of Standards and ed in the VNA. performance comparable to the Technology (NIST) has advanced most advanced TRL calibrations. the quality of rigorous on-wafer The second tier actually obtains VNA measurements. The focus of error boxes that represent the LRRM Calibration with this work has been the develop- change in the system between two Automatic load Inductance ment and application of a multiline calibrations. If a cable calibration is Extraction TRL calibration algorithm using followed by a second tier probe-tip the full redundancy of the stan- calibration the second tier error A variation of the Line-Reflect- dards. While normal TRL selects box will be the network parameters Match, the Line-Reflect-Reflect- line pairs for applicable frequency of the probe (see Figure 19). Match advanced calibration pro- ranges, the NIST multiline TRL vides calibration accuracy compa- uses a continuously varying opti- Separation of the cable cal from rable to the NIST multiline TRL mally weighted average of all of the the probe details can be useful in algorithm described below without data. 11, 12 complex multiport situations. The the need for changing probe sepa- error box S-parameters of the vari- ration.9 The LRRM calibration The multiline TRL algorithm is ous probe contacts are determined algorithm is available only in available in NIST's MultiCal soft- and the data mathematically cas- Cascade Microtech’s WinCal VNA ware. This HP BASIC program caded with the cable calibration calibration software running on a runs on HP 9000 workstations and error boxes. This technique is used PC. PCs and is compatible with the in some microwave production test HP 8510B, HP 8510C, and systems to reduce the number of The standards used for LRRM HP 8700 and Wiltron 360 series on-wafer standards needed for are identical to those used for VNAs multi-port calibration. SOLT but without the requirement for careful specification of the short and open standard i On-Wafer Reference Planes behavior. In LRRM, only the thru line delay (and loss for longer VNA & Cabling lines) and d.c. resistance of one Errors HDUT HProbe Errors MEEt Cabling / load standard must be specified.

Coaxial Reference Planes i The inductance of the load is extracted from the redundancy of Figure 19. Error boxes for a two-tier calibration. A coaxial first tier is followed by an on-wafer information provided by the extra second tier. standard measurements. By using and provide a good indication con- S,, 8 Ml log MAG firming visual identification of REF 0.0 dB proper alignment and contact. 5.0 dB/ A The load standard shows a significant reduction in the magnitude of the uncorrected reflection coefficient when contact is made. Contact with a 50 ohm load will tend to reduce the Smith chart display toward the ideal center dot. Particularly at higher frequencies the various reflections from adapters, cable connectors, and probes can degrade the overall Start 0.045000000 GHz return-loss of the load standard. Stop 40.000000000 GHz The measured reflections will normally be reduced by more than ‘--*-++I 10 dB when the load is in contact. Good Marginal Poor Smaller reductions are signs of poor system return loss that may require Figure 20. The relative magnitudes of uncorrected measurements of a short and a load provide correction (see Figure 20). a good check on proper system function. The thru line measurement will Another application of second tier The raised probe open measure- exhibit the matching characteristics calibrations is for determination of ment does not have any contact of the load and will also show an error bounds due to measurement issues and can be used as a refer- increase in the raw S,, transmission system repeatability. A first-tier ence. The measurement of a short measurement. A few experiments probe-tip calibration followed by a standard will show an ideally will provide a good idea of what to second tier calibration at the same 180 degree phase rotation of the expect from your fixturing and reference plane will determine error reflection coefficient. This can be VNA measurement system. boxes that represent the change in seen as a rotation of the raw data the system due to drift between the Smith chart constellation or a shift Calibration Verification calibrations. This information can in a phase display occurring when be used to calculate bounds on the the probe makes contact. In prac- Verification of proper calkit entry repeatability errors impacting mea- tice second order terms in the error and calibration can be demonstrat- surements that were made between box will also somewhat impact the ed by measuring a long open circuit the two calibrations, providing a actual change of observed reflection, and/or long transmission line. The useful measure of one of the com- but the shift will be very noticeable reflection coefficient of the long ponents effecting overall measurement accuracy.

Measuring Cal Standards The process of completing the calibration entails placement of probes on standards and directing the VNA measurement system to Open stub Pad acquire data. This process is somewhat error prone since it is not always easy to determine if the standard is properly connected.

While uncalibrated or ‘raw’ S- parameters from a VNA are often far from accurate they can be used to help verify proper contact with Line (delay) the calibration standards. Figure 21. Typical on-wafer verification elements. In many cases measurement of such structures will catch the most common errors and problems.

10 open circuit should exhibit a smooth monotonic inward spiral on the Smith chart. Look for linear phase in the transmission response of the long transmission line verification standard. High Q-factor GSG pads shield like CPW inductors or capacitors can similarly be used for calibration verification. It is a good practice to per- form one or more of these verification methods after completing every calibration. Some common verification standards are shown in GS pads fringe to the ground plane or chuck Figure 2 1.

Verification is particularly impor- Figure 22. GS pads fringe to the ground plane or chuck. tant with the SOLT calibration to ensure both proper calkit entry and The thru line standard is a useful GSG vs. GS Pads successful probe calibration stan- verification element for an SOLR calibration. Again the verification After probe positioning and gross dard measurements. Do not remea- cal calibration errors, the most sig- sure a calibration standard as a veri- should reflect expected behavior so this is only useful when the behav- nificant inaccuracy in microwave fication, since the only thing that probing is associated with parasitic will be revealed is the repeatability ior is predictable. The SOLR can calibrate successfully even with a coupling at the probe tip (see Figure of the measurement. If the system 22). is repeatable then the remeasure- highly reactive and complicated ment will simply show you what frequency response of the reciprocal thru which won’t be pre- A ground-signal-ground interface you entered into the calkit for that terminates field lines effectively, but standard. A VNA cannot recognize dictable. However, a good line standard will exhibit attenuation ground-signal pads allow the open the errors caused by using incorrect side fields to terminate on the wafer impedance standards. A load resis- with square-root of frequency dependence consistent with thin- ground plane or prober chuck. The tor of the wrong value will not be degree of this coupling is dependent recognized: it will be accepted as film transmission lines. Excess line length beyond the probe contacts on substrate thickness and pad 50 ohms. The VNA will even spacing. accept interchanged open and short will act like capacitive stubs result- standards without complaint. ing in increased attenuation at higher frequencies. Linear trans- An undesirable mode of propaga- Either mistake can result in signifi- tion where all top side conductors cant measurement error. mission phase would be expected unless the standard has dispersive form a microstrip with the bottom characteristics. of the wafer can occur in these Measurement of calibration stan- structures. The best way to deal dards for the advanced calibrations A golden standard device - a with this problem is usually to min- where standards don’t have to be imize the excitation of that mode. specified is useful for verification. device that has been characterized Examining the open that is used using a (hopefully) known good test system - is another way to vali- The GSG pad configuration for the reflect standard in a TRL or excites the undesired microstrip LRM calibration can identify prob- date a test system. It is difficult to mode less than the GS configura- lems. After calibration, the open validate a calibration and measurement method this way however. tion. A reasonable rule is to avoid may not have perfectly zero return GS pads for good results above loss. The probe in air will actually Slight differences in calibration ref- about 10 GHz. GHz. have slightly less loss then the erence planes may appear as signif- probe in contact with the wafer, so icant error. Differences in circuit in addition to some negative capac- bias or power supplies can also itance a small amount of return appear as errors in the microwave gain from the calibrated open may measurement. A golden standard normally be observed. The specific can, however, be an effective and strength of this effect is very depen- simple test for qualifying a new sys- dent on the probe design and will tem that uses identical equipment vary for different probe types and and methods as a reference mea- thru losses. surement system.

11 GS Probing Parasitics Figure 23 depicts a GS probe tip, two connection pads, and a DUT, along with the main reactive parasitics of the connection. The ground is the back side of the wafer, the wafer chuck, or a high- conductivity substrate such as Silicon.

Figure 24. Calibrated measurement of short circuit standard before and after distressing a cable. A change has occurred in the cable electrical behavior resulting in a 20 GHz resonance.

together at the measurement ports and needle probe system, or a using bias tees. If the performance Pyramid Probe Card is needed. Low of the VNA internal bias tee is impedance is the primary desirable inadequate, or your VNA does not characteristic in a power supply and provide one, then an external bias is a major feature of the Pyramid Figure 23. GS probing parasitics. tee may be preferred. Probe Card with its bypass capacitors very close to the contacts and The desired transmission line uses A bias tee is a three-port network excellent grounding. the inductances in series with the with the properties of a diplexer; DUT and the corresponding bridg- that is, a wide bandwidth port is Needles, either used in a hybrid ing capacitors. The resistor repre- coupled to the bias port via a low- probe card or as additional probes sents polyiron material used on pass filter and to the RF port on separate positioners, have high Cascade Microtech single line through a high-pass filter. inductance since they are not close- probes that acts to absorb energy in Important selection parameters of a ly coupled to a ground return. the undesired mode. This attenu- bias tee are: cross-over frequency, Inductance on supplies can cause ates energy propagating up the bandwidth, current and power rat- power supply ‘bounce’ in digital cir- ground which could resonate with ing, resistance and loss, and con- cuits and instability or gain flatness the probe holder. nector type. problems in analog circuits. Adjacent needles may be inductive- Proper selection and use of a The series and shunt resistance of ly coupled or even by virtue of shar- microwave probe will allow the par- the bias tee will be significant when ing a common ground return path asitic CSProbe and CGProbe to be measuring individual transistors or may create signal paths between small and repeatable. However, dif- other devices where a precise bias various portions of a circuit by way fering pad sizes or conductor point is desired. For moderate to of power connections. geometries on test structures can large currents the series voltage cause CGDur or CSDut to be differ- drop will make a change from the Cable Sensitivity ent from one structure to another, expected bias voltage. Expected or from the calibration standards. device currents will vary from actu- Correcting the network analyzer Only when these values are equal al due to the current path provided measurement through a lossy cable for calibration standards and mea- by shunt resistance. This resistance is similar to the correction we sub- surement devices will these para- is high, but not high by the stan- consciously use to see an image sitics be removed by VNA calibra- dards of parametric measurements through a dirty window. We correct tion. But changes in metal configu- with currents possibly ranging from for the attenuation due to dirt on ration will often occur. For mea- 1 fA to 100 mA. Probes, cables, the window as well as reflections surements above 10 GHz, the GS and probe stations designed specifi- from the window’s surfaces. It is configuration is inadequate and the cally for dc parametric measure- important that the cables be highly superior shielding of GSG is pre- ments are generally preferred for repeatable as well as reasonably low ferred. these cases. loss, so that there is enough dynamic range after looking through and Device Bias Considerations For microwave integrated circuits correcting for the lossy cable. the bias and power lines will often Calibrated measurements of a Power supplies may be supplied use separate pads from the RF con- short circuit standard before and through the VNA bias ports which nections. For these circuits, either a after the cable has been significantly combine the DC and RF signals multi-contact probe, hybrid RF flexed, shown in Figure 24, demon- Symptom Likely Problem Measurements do not Cables or adapters are loose or Calibrate in coax with and without cables and repeat at any freq faulty adapters Analyzer fault Calibrate in coax directly at test set port Probe Contact fault Verify probe contact visually or at dc Probe head fault Calibrate in coax at probe connector; if good, replace probe head Oscillating DUT Use different bias condition to verify; change low-frequency DUT terminations Measurement noisy at all Insufficient averaging on Increase averaging to at least 4 in ramp freqs HP8510 mode, or 128 in step mode

Poor ramp sweep repeatability Use step mode

Measurement noisy at Sweeper band switching Increase averaging, use step mode. Change certain freqs sweep rate Loose connection Tighten connectors Faulty probe resonating Check open stub on ISS - should be smooth inward spiral

Measurement noisy at Poor return loss at high freqs Measure ISS load with corrections off - high freqs only should be at least 10 dB below the open or short-trace back to test set using coaxial load and short

DUT measurements noisy DUT is being biased by rectified Reduce RF power level, add device input port at low freqs only, or RF,or RF is compressing DUT attenuation, use power slope Transistor S 21 gm compresses at low freqs (passive elements ok) Passive element Incorrect cal coefficients Adjust cal coefficients reflection coefficient outside Smith chart Poor probe placement on Use alignment marks to improve placement standards

Table 3. Troubleshooting guide. strates the importance of measure- References 7. C. Hoer, G. Engen, “Calibrating a ment repeatability. The resonance dual six-port or four-port for measuring 1. Calibration and Accuracy Factors at about 20 GHz would signifi- two-ports with any connectors,” IEEE Summit High-Frequency Probe Station MTT-S Digest, pp. 665-668, 1986. cantly distort further DUT mea- Reference Manual. Cascade Microtech, surements. 8. A. Davidson, K. Jones, E. Strid, “LRM part number 103-375-A and LRRM Calibrations with Automatic 2. D. Rytting, “Appendix to an Analysis Determination of Load Inductance,” 36th When troubleshooting the of Vector Measurement Accuracy ARFTG Conference Digest, Nov. 1990. responses of a VNA, it is useful to Enhancement Techniques,” RF & 9. J. Pence, “Verification of LRRM keep in mind that impedances that Microwave Symposium and Exhibition, Calibrations with Load Inductance vary rapidly with frequency cannot Hewlett-Packard Inc, 1986. Compensation for CPW Measurements on be due to errors near the DUT. 3. D.K. Walker, D.F. Williams, GaAs Substrates,” 42nd ARFTG Conference When wafer probing there is insuf- “Compensation for Geometrical Variations Digest, Nov. 1993. ficient electrical length on the in Coplanar Waveguide Probe-Tip 10. Cascade Microtech Technical Brief Calibration, IEEE Microwave and Guided wafer or in the probes to cause res- Technique Verifies LRRM Calibrations for Wave” Letters, Vol.7, No.4, pp. 97-99, GaAs measurements. onances or impedances that change April 1997. rapidly with frequency. 11. R Marks, “A Multi-line Method of 4. Paul J. van Wijnen, On the Network Analyzer Calibration,” IEEE Characterization of. . Cascade Microtech Trans. on MTT, Vol. 39, No. 7, pp. 1205- Troubleshooting 5. A. Ferrero, “Two-Port Network 1215, July 1991. Analyzer Calibration Using an Unknown 12. D. Williams, J. Pence, A. Davidson, Table 3 may be helpful in trou- 'Thru,'" IEEE Microwave and Guided Wave “Comparison of On-Wafer Calibrations," bleshooting VNA measurement Letters, Vol. 2, No. 12, pp. 505-507, Dec. 38th ARFTG Conference Digest, Dec. 1991.1. problems (some details apply only 1992. 13. For more information contact Dylan to Agilent 8510 systems). 6. Hayden and Basu, "An SOLR Williams, NIST 813.06, 325 Broadway, Calibration for Accurate Measurement of Boulder, CO 80303. Orthogonal On-Wafer DUTs," IEEE MTT-S Digest, 1997.