Toward Better Acoustic Resonators

Where to build efficiency into the MEMS design flow.


After some lively conversations with the top researchers in MEMS acoustic resonators during the 2014 Sensors and Actuators Workshop (familiarly known to the MEMS community as “Hilton Head”), we set to work on a simulation solution to better serve researchers and commercial designers.

Acoustic resonators for RF filtering have received a lot of attention in the past few years as the number of filters in mobile handsets has jumped to somewhere around 30. The best-known success story is Avago’s Film Bulk Acoustic Resonator (FBAR), which proved the high performance of bulk-mode devices over conventional SAW filters.

What most surprised me about those conversations at Hilton Head last year was how little emphasis had been placed on fast computational algorithms to serve this community. Also missing were design flow efficiencies such as automatic generation of S-parameters for N-port devices and Butterworth-Van Dyke parameter extraction. For instance, there was talk that frequency sweep simulations could take days and that 3D simulation is so impractically long that only 2D simulation is done. (Simulation in 2D, unfortunately, is not as accurate and leaves it to measurement to determine all the spurious modes in the frequency response). As another example, one researcher found that accurately determining just the peak resonance frequency in 3D took too long. He actually considered writing an optimization algorithm to judiciously choose frequency points to search for the peak in the frequency response just to reduce the number of 3D frequency point simulations. It struck me as wasted effort because their simulation tools could not perform a fast frequency sweep in a reasonable time.

Some might say graduate students are cheap and it’s a good learning experience, but in a commercial setting with time-to-market pressures, this is unacceptable. Even for research groups focused on proving out concepts that are sensitive to design details, it’s unacceptable. This, I believe, is what intrigued Professor Songbin Gong at the University of Illinois when I told him about our proprietary fast frequency-sweep algorithms specifically for MEMS piezoelectric resonators. I claimed we could simulate his 3D designs in minutes where his current tool took hours, and do in hours what previously took days. As you can read below, that has largely proven to be true.

Since the Hilton Head 2014 conference, Professor Gong and his group have worked with us to polish this offering specifically for rapid design of cutting-edge acoustic resonators. We’ve named the offering “FastPZE” and it was part of our CoventorWare 10 release earlier this year. Professor Gong and his students Ruochen Lu and Anming Gao used CoventorWare late last year to carefully design dual resonances of a 2-port resonator to demonstrate parametric resonance for the first time in these type of resonators. Their work is detailed in a paper they presented in April 2015 at the International Frequency Control Symposium (IFCS 2015).

If you are interested in learning more, I gave a talk in the MEMS and NEMS session of the ASME InterPACK 2015 conference in San Francisco. You can also download my white paper entitled “Fast Acoustic Resonator Analysis for the Rapidly Growing Premium RF Filter Market”. Below are my favorite 3 figures from the white paper.

Figure 1: Frequency response of 3D model of laterally vibrating resonator (LVR) similar to design from 2013 MTT paper by Gong and Piazza. A conventional frequency sweep (red circles) does not resolve spurs or the main resonance compared to the FastPZE simulation (black line). Both simulations took the same amount of simulation time. Achieving similar resolution with the conventional algorithm would have required 28X more computation time.

Figure 2: FastPZE simulation results for LVR (blue) and automatically extracted BVD model (red). The BVD parameters and electromechanical coupling, Screen Shot 2016-05-18 at 10.35.00 PM, are inset on the bottom left. Motional RLC for the spurs and other resonances can be computed as well with the upcoming CoventorWare 10.1 release available this fall (or sooner upon request).

Figure 3: Automatically extracted 2-port S-parameters and corresponding exported file for RF circuit simulation.