UC Berkeley

Abstract

A Micromechanical RF Channelizer

by

Mehmet Akgul

Doctor of Philosophy in Engineering - Electrical Engineering and Computer Sciences

University of California, Berkeley

Professor Clark T.-C. Nguyen, Chair

The power consumption of a radio generally goes as the number and strength of the RF signals it must process. In particular, a radio receiver would consume much less power if the signal presented to its electronics contained only the desired signal in a tiny percent bandwidth frequency channel, rather than the typical mix of signals containing unwanted energy outside the desired channel. Unfortunately, a lack of filters capable of selecting single channel bandwidths at RF forces the front-ends of contemporary receivers to accept unwanted signals, and thus, to operate with sub-optimal efficiency.

This dissertation focuses on the degree to which capacitive-gap transduced micromechanical resonators can achieve the aforementioned RF channel-selecting filters. It aims to first show theoretically that with appropriate scaling capacitive-gap transducers are strong enough to meet the needed coupling requirements; and second, to fully detail an architecture and design procedure needed to realize said filters. Finally, this dissertation provides an actual experimentally demonstrated RF channel-select filter designed using the developed procedures and confirming theoretical predictions.

Specifically, this dissertation introduces four methods that make possible the design and fabrication of RF channel-select filters. The first of these introduces a small-signal equivalent circuit for parallel-plate capacitive-gap transduced micromechanical resonators that employs negative capacitance to model the dependence of resonance frequency on electrical stiffness in a way that facilitates the analysis of micromechanical circuits loaded with arbitrary electrical impedances. The new circuit model not only correctly predicts the dependence of electrical stiffness on the impedances loading the input and output electrodes of parallel-plate capacitive-gap transduced micromechanical device, but does so in a visually intuitive way that identifies current drive as most appropriate for applications that must be stable against environmental perturbations, such as acceleration or power supply variations. Measurements on fabricated devices in fact confirm predictions by the new model of up to 4× improvement in frequency stability against DC-bias voltage variations for contour-mode disk resonators as the resistance loading their ports increases. By enhancing circuit visualization, this circuit model makes more obvious the circuit design procedures and topologies most beneficial for certain mechanical circuits, e.g., filters and oscillators.

The second method enables simultaneous low motional resistance (Rx < 130 Ω) and high Q (>70,000) at 61 MHz using an improved ALD-partial electrode-to-resonator gap filling technique that reduces the Q-limiting surface losses of previous renditions by adding an alumina pre-coating before ALD of the gap-filling high-k dielectric. This effort increases the Q over the ~10,000 of previous renditions by more than 6× towards demonstration of the first VHF micromechanical resonators in any material, piezoelectric or not, to meet the simultaneous high Q (>50,000) and low motional resistance Rx (< 200Ω) specs highly desired for front-end frequency channelizer requirements in cognitive and software-defined radio architectures. The methods presented in this chapter finally overcome the high impedance bottleneck that has plagued capacitively transduced micro-mechanical resonators over the past decade.

The third method introduces a capacitively transduced micromechanical resonator constructed in hot filament CVD boron-doped microcrystalline diamond (MCD) structural material that posts a measured Q of 146,580 at 232.441 kHz, which is 3× higher than the previous high for conductive polydiamond. Moreover, radial-contour mode disk resonators fabricated in the same MCD film and using material mismatched stems exhibit a Q of 71,400 at 299.86 MHz. The material used here further exhibits an acoustic velocity of 18,516 m/s, which is now the highest to date among available surface micromachinable materials. For many potential applications, the hot filament CVD method demonstrated in this work is quite enabling, since it provides a much less expensive method than microwave CVD based alternatives for depositing doped CVD diamond over large wafers (e.g., 8") for batch fabrication.

The first three methods described so far focus on a single vibrating disk resonator and improve its electrical equivalent modeling, Cx/Co, and Q. Once we craft the resonator that meets the challenging design requirements of RF channel-select filters, the last method presents a design hierarchy that achieves the desired filter response with a specific center frequency, bandwidth, and filter termination resistance. The design procedure culminates in specific values for all mechanical geometry variables necessary for the filter layout, such as disk radii, and beam widths; and process design variables such as resonator material thickness and capacitive actuation gap spacing.

Finally, the experimental results introduce a 39nm-gap capacitive transducer, voltage-controlled frequency tuning, and a stress relieving coupled array design that enable a 0.09% bandwidth 223.4 MHz channel-select filter with only 2.7dB of in-band insertion loss and 50dB rejection of out-of-band interferers. This amount of rejection is more than 23dB better than previous capacitive-gap transduced filter designs that did not benefit from sub-50nm gaps. It also comes in tandem with a 20dB shape factor of 2.7 realized by a hierarchical mechanical circuit design utilizing 206 micromechanical circuit elements, all contained in an area footprint of only 600μm×420μm. The key to such low insertion loss for this tiny percent bandwidth is Q's>8,800 supplied by polysilicon disk resonators employing for the first time capacitive transducer gaps small enough to generate coupling strengths of Cx/Co ~0.1%, which is a 6.1× improvement over previous efforts. The filter structure utilizes electrical tuning to correct frequency mismatches due to process variations, where a dc tuning voltage of 12.1 V improves the filter insertion loss by 1.8 dB and yields the desired equiripple passband shape. An electrical equivalent circuit is presented that captures not only the ideal filter response, but also parasitic non-idealities that create electrical feed-through, where simulation of the derived equivalent circuit matches the measured filter spectrum closely both in-band and out-of-band.

The combined 2.7dB passband insertion loss and 50dB stopband rejection of the demonstrated 206-element 0.09% bandwidth 223.4-MHz differential micromechanical disk filter represents a landmark for capacitive-gap transduced micromechanical resonator technology. This demonstration proves that the mere introduction of small gaps, on the order of 39 nm, goes a long way towards moving this technology from a research curiosity to practical performance specs commensurate with the needs of actual RF channel-selecting receiver front-ends. It also emphasizes the need for tuning and defensive stress-relieving structural design when percent bandwidths and gaps shrink, all demonstrated by the work herein.

Perhaps most encouraging is that the models presented in dissertation used to design the filter and predict its behavior seem to be all be spot on. This means that predictions using these models foretelling 1-GHz filters with sub-200Ω impedances enabled by 20nm-gaps might soon come true, bringing this technology ever closer to someday realizing the ultra-low power channel-selecting communication front-ends targeted for autonomous set-and-forget sensor networks. Work towards these goals continues.