Refinements in both the measurement probes and the calibration standard structures used with them makes it possible to perform more accurate microwave VNA S-parameter measurements on differential devices.
Senior Product Manager For Probes and Signals
Cascade Microtech, Inc., 2430 NW 206th Ave., Beaverton, OR 97006;
(503) 601-1000, FAX: (503) 601-1002,
Balanced or differential circuitry is becoming more commonplace in high-frequency designs as a means of reducing susceptibility to radiofrequency interference (RFI) and electromagnetic interference (EMI). Of course, testing and characterization of balanced circuits means the use of a vector network analyzer (VNA) and test set with four ports instead of two, and the study of differential and common-mode responses and mode-conversion terms. Although many test systems feature four-port test sets, single-ended measurement methods are still often used, with measurements made one port at a time and differential and common-mode characteristics calculated from the results. Especially for on-wafer testing, advances in balanced measurement techniques and their accompanying calibrations can improve the accuracy of on-wafer differential measurements.
A variety of challenges exist for onwafer or on-substrate measurements of multiport devices, especially when considering that stand-alone four-port VNA systems typically have upper-frequency limits of about 50 GHz. Some of the issues unique to f-port probebased measurements include lack of a probe-centric RF system calibration, lack of appropriate and well-designed precision probe calibration standards, and inaccuracies due to crosstalk in dual-signal probes.
Since VNA systems are commonly used for RF mixed-mode characterization, it may help to example the basic architectures of two-port (Fig. 1) and four-port (Fig. 2) VNA systems and how the different architectures impact calibration. For standard two-port measurements, the VNA's source signal travels to the port transfer switch for routing and part of the signal is split off to the reference receiver R1. The signal connects to a device under test (DUT) at port 1 with a small amount of the reflected signal coupled and routed to the response receiver (A). Similarly, a sample of the transmitted signal is routed to a response receiver (B). At port 2, the transmitted signal continues past the coupler back to the termination path of the port transfer switch. Because this is an imperfect termination, it generates a reflection back to the DUT; a piece of this switch termination reflection signal is split off to reference receiver B. Knowledge of this switch-reflection coefficient for each choice of excitation that allows the traditional two-port 12-term calibration error model (Fig. 3) (really a 6-term model for each excitation choice) to be reduced to an 8-term error model that is used in advanced calibrations (Fig. 4).1-3
While a four-port VNA is basically an extension of two-port architecture, there is one notable exception that impacts system calibrations. In the four-port case (Fig. 2), the R2 reference receiver in the VNA is isolated and is not able to sample any of the reflected waves. The inaccessibility of the R2 reference receiver has been an obstacle to providing a reduced error term, advanced, "probing- centric" calibration algorithm for four-port systems. Therefore, fourport VNA systems primarily have relied on a more limited resident short-open-load-through (SOLT) calibration approach. A SOLT calibration is designed to provide a coaxial reference plane using a 12-term error model. The approach relies on physical standards on substrates with previously determined, fully known electrical behavior. Probe measurements on these standards are used to extend the measurement reference plane. Unfortunately, a fundamental limitation of the 12-term error model is its sensitivity to variations in probe placement on the physical standards.4 Another limitation of both 12- and 8-term calibrations is that they cannot address errors due to coupling between signal paths. Four-port measurement systems that use adjacent-signal dual probes are especially susceptible.
Areas where a good deal of needed progress has been made in supporting probe-based mixed-mode measurements include reducing signal-tosignal crosstalk; improving four-port calibration structures; and developing new "probing-tolerance" calibrations for four-port systems.
Dual probes are often used for multiport mixed-mode measurements. Dual probes feature two signal lines separated by a ground or with two adjacent signal lines each with an adjacent ground. At the probe tips, the two signal lines are in close proximity and accuracy is affected by coupling between the signal lines (Fig. 5). This general coupling is typified by the simple model of capacitance between signals at the probe tip. Complicating this is that this coupling varies with DUT input impedance. Standard VNA error models cannot model this coupling. This allows for only simple isolation correction that is useful for internal VNA port leakage but does not model coupling that varies with DUT input impedance. The best option is to control crosstalk so that the uncorrected error is small.
In pursuit of this goal, dual probes and their calibration standards are designed to be as compatible as possible with available error models. To do this, both the probes and the calibration standards must exhibit high isolation. A practical goal has been to try to exceed 30 dB isolation between signals for the dual-signal probe and standard combinations.5 This begins to approach the behavior of state-ofthe- art single probes in the opposing configuration that often achieve 50 dB isolation between signals to 50 GHz. Conventional dual-signal probe tips have contact fingers in air or on a dielectric substrate. Since any signal path (line of sight) from signal one to signal two is blocked by a central ground, the isolation should be good. However, additional coupling possibilities are introduced when the probe is used in contact in its intended environment. The affinity for the electric field to exist in a higher dielectric material means that close to the tips of the wafer probe, the signal path is no longer line of sight but direct and through the dielectric material. Fortunately, using a microstrip architecture with inherently high isolation, it has been possible to develop dual-signal probes with high isolation.
With attention to the design of the transitions, these probes achieve better than 40 dB isolation through 40 GHz for a ground-signal-ground-signalground (GSGSG) configuration and 30 dB isolation through 40 GHz for a ground-signal-signal-ground (GSSG) configuration (Fig. 6).
For dual probes, a dual calibration standard with the appropriate physical dimensions is needed that simultaneously contacts both probe ports. For four-port systems, dual standards present a number of challenges. First, the standard itself can be a source of signal-to-signal coupling. Also, the additional conductors and the increased size of such standards (needed to contact both ports) increases the likelihood of undesired mode conversions.
Through attention to detail, it has been possible to optimize the short, open, load, and through-line (thru) calibration structures used with dual probes. For the short calibration standards, for example, when a simple bar structure is used, it can exhibit inductance in the signal-to-signal and signalto- ground paths. In this case, it is the equivalent of adding inductors to both signal paths as part of the calibration measurements. Fortunately, an improved short structure has been developed that is designed to mitigate the inductance effect (Fig. 7).
An additional problem is the unshielded contact pads on open or load standards. These unshielded pads add capacitance between the signals. An improvement involves pads that are at least partially surrounded by ground. A line-of-sight path from pad to pad that is interrupted by ground significantly reduces the pad-topad capacitance by means of electrostatic shielding.
In the design of dual probes, a coplanar waveguide (CPW) structure will have an imbalance of energy if the length of one slot is longer than another this is compounded with dual signal probes and structures since the outside grounds are that much farther away for a given pitch. The probe grounds are also longer on the outside than in the center. This imbalance can cause mode conversions and potentially resonant behavior may be observed. An improvement has been the loop-under ground structure that connects the grounds to block the conversion of energy from coplanar to slot line mode very close to the probe tip.
Additionally, with the larger dual structures at higher frequencies it is more likely another mode will be excited. For this reason Impedance Standards Substrate holders now use a microwave absorber that has a non-unity relative permeability at microwave frequencies. Energy that is excited in, for instance, the microstrip mode that relies on a reflective, metallic back surface of the calibration substrate is at least partially absorbed. Finally, as calibration standard structures are made larger, resonances occur at lower frequencies. Another mode-dampening technique has been designed in that uses thin film resistors on the edges of the calibration structures to terminate mode or resonant energy.
As the number (N) of ports increases, more calibration standards must be measured. For example, for thru structures, worst-case combinations can be proportional to N2. Each structure requires probes to be landed precisely and repeatably and is inherently an opportunity for a probing error. So, it is desirable to minimize the number of structures used. For the case of a multiport calibration, with ports split evenly on opposing probes, it is possible to perform a calibration using only five structures: parallel straight reference thrus (as ideal as possible or at least with known electrical behavior), shorts on all ports, opens on all ports, loads on all ports, and loop-back thrus.
Over the years, external advanced ??probing tolerant? two-port calibrations have been developed. These calibration algorithms rely less on assumed electrical knowledge of probed standards and on fewer error terms using a reduced eight-term error model. The eight-term error model makes use of knowledge of the VNA switch termination reflection coefficient (or more directly the reflected signal) provided in the twoport VNA to mathematically convert the S-parameters as if the switch termination was perfect. As pointed out earlier, knowledge of the four-port VNA switch termination reflection coefficient is often unavailable.
However, even though many systems do not provide access to direct measurement of the switching terms required to perform the error model reductions, this limitation can be circumvented. An alternate method is to extract the necessary switching terms mathematically from a known good coaxial calibration at the VNA front panel. By using a modern electronic calibration module and a coaxial calibration, the switching terms can be extracted.
Having the switching terms opens up the possibility of using a variation of the line-reflect-reflect-match (LRRM) VNA calibration with automatic load inductance correction that has been an accepted and reliable method for twoport on-wafer probing measurements for more than a decade. For the LRRM case, each straight-thru is switch-term corrected and applied to the LRRM algorithm with the associated opens, shorts, and loads. Including every port in at least one of these port pair combinations for LRRM provides information equivalent to doing SOL on every port except much more accurately since the requirement of known 1-port standard behavior is eliminated.
There is one more hurdle in achieving an advanced multiport calibration, however: Further optimization of the LRRM calibration is needed. This is due to the fact that dual probes use thru standards that loop back on themselves for the two signal lines on each probe. These ??loopback? structures will inherently have undesired behavior as standards due to their unequal length slots??they have unknown behavior in terms of loss, frequencydependent delay.6 So these non-ideal thrus are best treated as unknowns but electrically reciprocal.
A new hybrid calibration has been developed for four-port VNA systems as an enhanced version of LRRM algorithm by combining LRRM with elements of the short-open-line-reflect (SOLR) calibration technique. In the calibration math for SOLR, the loopback thrus can simply be treated as unknown but reciprocal.7,8 The result is a multiport calibration that is relatively insensitive to small errors in probe placement that are inherent in microwave probing and that handles the necessary non-ideal thrus.9
In the last few years there have been several significant advances in probing systems for characterization of RF mixed-mode circuitry. A precision four-port RF measurement system should now offer:
- Low signal-to-signal crosstalk
- Low sensitivity to probe placement on calibration standards
- Reduced dependence on electrical standard definitions
- Reduced undesired modes
- Fixed-probe positioning with automatic wafer moves
- Ability to account for non-ideal four-port standard loop-back thrus
External calibration software such as WinCal XE(TM)10 from Cascade Microtech provides many powerful features for four-port VNA differential measurements including LRRM- SOLR calibration, guidance for four-port setup, and post-processing of S-parameters for single-ended to mixed-mode conversions.
- A. Davidson, K. Jones, and E. Strid, ??LRM and LRRM calibrations with automatic determination of load inductance,? 36th ARFTG Conference Digest, Monterey, CA, Fall 1990, pp. 57-63.
- L. Hayden, ??An enhanced Line-Reflect-Reflect-Match calibration,? 67th ARFTG Conference Digest, San Francisco, CA, Spring 2006, pp. 143-149.
- R. B. Marks, ??Formulations of the basic vector network analyzer error model including switch-terms,? 50th ARFTG Conference Digest, Portland, OR, Fall 1997, pp. 107-114.
- A.M.E. Safwat and L. Hayden, ??Sensitivity analysis of calibration standards for SOLT and LRRM,? 58th ARFTG Conference Digest, San Diego, CA, Fall 2001.
- L. Hayden, ??Calibration errors when neglecting crosstalk,? 66th ARFTG Conference Digest, Fall 2005, pp. 65-68.
- R. Senguttvan, L. Hayden, and A. Weisshaar, ??Mode coupling in coplanar waveguide bends: a simple four-port model,? 33rd European Microwave Conference, Munich, Germany, October 2003, Vol. 2, pp. 643-646.
- A. Ferrero and U. Pisani, ??Two-port network analyzer calibration using an unknown `thru??,? IEEE Microwave and Guided Wave Letters, December 1992, Vol. 2, No. 12, pp. 505-507.
- S. Basu and L. Hayden, ??An SOLR calibration for accurate measurement of orthogonal on-wafer DUTs,? 1997 IEEE MTT-S International Microwave Symposium Digest, Denver, CO, June 1997, Vol. 3, pp.1335-1338
- L. Hayden, ??A hybrid probe-tip calibration for multiport vector network analyzers,? 68th ARFTG Conference Digest, Fall 2006, pp. 176-183.
- WinCal XE, Cascade Microtech, Inc., Beaverton, OR, www.cascademicrotech.com.