A panel discussion at the 2010 Compound Semiconductor IC Symposium (CSICS) in Monterey, CA found hopeful anticipation for the future of nonlinear measurement-based models. The panel was entitled “X-Parameters and Measurement-Based Behavioral Models: Real Models or Just Marketing?” It was moderated by Joe Gering (RFMD) and Seyed Tabatabaei (EndWave Corporation). The members of the panel were David Root (Agilent Technologies), Paul Tasker (Cardiff University), Mike Golio (Golio Consulting), Yusuke Tajima (Auriga Microwave), Gayle Collins (Freescale Semiconductor), and Ed Anthony (Skyworks Solutions).
Agilent’s X-parameters and the Cardiff Model from Mesuro/Cardiff University represent two paths to the same model end-goal and were respectively presented by David Root and Paul Tasker. Both Agilent and Mesuro provide complete measurement systems capable of generating simulation-ready models directly and instantaneously from measurements of amplifiers and other nonlinear devices, and these models include effects over power, frequency, and termination impedance. The two approaches depart somewhat. Mesuro’s “Waveform Engineering” emphasizes creating load impedance conditions during the measurement around the optimum for a design and simulating where these measurements have been taken. Agilent offers X-parameters as capable of extrapolating in load impedance from a nominal measurement during design. In other parameters (for example power and bias), both models perform lookup-table interpolation, although the Cardiff Model has more elaborate means of data representation and ultimately may have a more compact data set. As defended by David Root, both models have a fundamental mathematical basis that makes them suitable to properly represent the nonlinear behavior of devices and circuits. Also, current research is finding ways to handle higher order effects such as memory effects, which are not in current model formulations.
Mike Golio offered some historical perspective. Looking back at the invention, availability, and adoption of S-parameters, Mike reported that it took several decades to go from the introduction of S-parameters to wide-scale access in measurement equipment. It then took another decade until they were being regularly reported in the literature and provided by manufacturers as the de facto model for linear components. Mike suggested that since we are well ahead of this schedule for S-parameters’ nonlinear counterparts, it’s only a matter of time until industry understands their benefits and limitations and begins to use them. However, he did caution that the rate at which these nonlinear data-based models are adopted will be directly proportional to their usefulness in making design work easier, faster, and ultimately more cost effective. He had a concluding observation that as S-parameter use grew, the use of equivalent circuit models also grew, because these two representations (data versus equivalent circuit) ultimately supported each other. The same should be true with nonlinear models.
Much of the work with measurement-based models such as the Cardiff Model and X-paramters has been aimed at characterizing finished amplifiers. Yusuke Tajima did offer a cautionary note. In modeling today’s leading edge transistors, one often finds anomalous behavior for example due to traps or thermal effects. Neither the Cardiff Model nor X-parameters currently handle these memory effects. To be useful at the device level and not just the system level, the extensions hinted at by David Root will need to be brought to fruition.
From the end user’s perspective, both good and bad experiences were shared by industry representatives Skyworks and Freescale. Skyworks has been studying X-parameters as a means to provide their mobile device customers with comprehensive component models, including control-type operation like power-conserving “turn-down” modes. For power amplifiers with well-defined input signals, Skyworks has found that X-parameters appear to satisfy the need for accurate and capable models for their customers’ simulation and design requirements. Again, areas where memory effects (such as bias network bandwidth or thermal or electrical transients) come into play are not modeled well, and simulations involving these effects are regrettably not presently possible. Freescale has embarked on a similar evaluation for its LDMOS devices and is finding X-parameters less inviting. In comparison to the well-defined application space for Skyworks’ amplifiers, Freescale’s LDMOS customers could be using their devices with radically different design conditions. Furthermore, the difference between matched and unmatched devices as well as the low output impedance of an LDMOS transistor makes the use and interpretation of a measurement-based model difficult. A related concern for LDMOS characterization is the viability of measuring very high power levels with network analyzers or digital sampling oscilloscopes while keeping the instrument safe and achieving sufficient dynamic range. As a counterpoint to Freescale’s experience, Paul Tasker discussed their successes with measurement-based transistor models in power amplifier design. Taking precise measurements with well-defined power, frequency, and impedance and using the same input signal in measurement that would be used in simulation, the Cardiff Model allows for first-pass success using a developing MMIC process for which compact device models may not be immediately available.
While there were no “winners” or “losers” in this panel discussion, the audience was left with a good sense for the current capabilities and weaknesses and the future potential for these measurement-based models. Most assuredly, this panel session will not be the final word on the subject and more expostulations and debates will happen through publications and presentations. For example, in April 2011, IEEE Microwave Magazine will have a special issue on nonlinear vector measurements and related behavioral models. At the International Microwave Symposium in June 2011, there will be a workshop on compact versus table-based models and a MicroApp Expert Forum on nonlinear device characterization. Also, interested readers are encouraged to look into the spring and fall Automatic RF Techniques Group (ARFTG) Microwave Measurements Conference, which has become a home for nonlinear measurements, and, of course, the CSICS. To be sure, the debate and proliferation of measurement-based behavioral models will continue.
Michael Heimlich, Macquarie University, NSW 2109 Australia Joseph Gering, RFMD, 7628 Thorndike Rd., Greensboro, NC 27409, USA
Click here for details on the Nonlinear Characterization Expert Forum at this year's IMS 2011 MicroApps session.