Capacitively Transduced Micromechanical Resonators with Simultaneous Low Motional Resistance and Q > 70,000

Capacitively transduced micromechanical disk resonators that exhibit simultaneous low motional resistance (< 130 Ω) and high Q (>70,000) at 61 MHz are demonstrated using an improved ALDpartial 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 work 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. This work finally overcomes the high impedance bottleneck that has plagued capacitively transduced micromechanical resonators over the past decade. INTRODUCTION Capacitively transduced micromechanical resonators have historically achieved the highest Q’s among micro-scale counterparts using other technologies [1][2][3], reaching Q’s past 200,000 at VHF and 14,600 at 1.2 GHz [4]. Such Q’s are highly desired for a myriad of applications, from very narrow band low insertion loss filters for co-site interference suppression [5]; to low phase noise, low power oscillators for radar and communications [6]; to frequency gating spectrum analyzer functions for true softwaredefined cognitive radio [5]. Additional advantages, like zero DC power consumption and the ability to self-switch, further bolster the argument for using these devices [7]. Still, adaptation of capacitive micromechanical resonators in these applications has so far been slowed by their higher than conventional impedances, which makes them difficult to interface with widely used 50Ω RF components. Theoretically, there is actually no good reason why the impedances of such resonators must be high. In particular, just a quick glance at Eq. (1) governing the motional resistance Rx of the disk array-composite resonator shown in Fig. 1 reveals that there are many knobs to turn that allow a designer to achieve a very wide range of motional resistances. Fig. 2(a) presents curves of Rx versus gap spacing and dc-bias that illustrate the wide range of values attainable by capacitively transduced resonators. Although all the variables in (1) can (and have been) used as knobs to tailor Rx, the 4 th power dependence on electrodeto-resonator gap spacing dgap makes it the most effective of the bunch [8]. Furthermore, as detailed in [8], reducing gap spacing is the most effective knob by which to maximize the FOM =1/(RQCo) figure of merit that governs the efficacy of a given resonator for filter construction. And contrary to popular belief, shrinking dgap does not necessarily reduce resonator linearity, as clearly shown by the IIP3 versus gap spacing and dc-bias curves of Fig. 2(b), generated using the theory of [9]. Here, for a dc-bias voltage of 20V, the IIP3 remains above 25 dBm at a gap spacing of 30 nm. For lower dc-bias voltages, e.g., 5V, the IIP3 remains high for gap spacings well below 30 nm. Of the approaches to reducing gap spacing towards lower impedance [8][10][11], the partial atomic layer deposition (ALD)filled gap spacing method described in [8] combines the advantages of better yield relative to brute force gap-release approaches [10][12], and demonstrated application to a lateral disk resonator, to which the silicide gap method of [11] has not yet been applied. Unfortunately, the work of [8] lowered the Q of resonator devices from starting values over 150,000 without partial ALD-gap filling down to only ~10,000 with ALD, supposedly due to surface losses introduced by the HfO2 high-k dielectric film used to partially fill the gap and coat the resonator. Pursuant to restoring Q’s to over 50,000 while simultaneously achieving impedances less than 200, this work employs an alumina (Al2O3) pre-coating layer to improve the material quality of partial gap-filling ALD dielectric material, thereby greatly reducing surface losses towards a more than 6 increase in Q. The measured simultaneous Q of 73,173 and motional resistance of Fig. 1: Schematic of an array-composite disk resonator, showing a one-port excitation/readout scheme, identifying important variables, and providing the expression for the effective motional resistance Rx of the device. Fig. 2: Simulated plots of (a) motional resistance and (b) IIP3 versus dc-bias voltage and electrode-to-resonator gap spacing for a 61-MHz wineglass disk resonator with Q = 70,000. M. Akgul, B. Kim, Z. Ren, and C. T.-C. Nguyen, “Capacitively transduced micromechanical resonators with simultaneous low motional resistance and Q >70,000,” Tech. Digest, 2008 Solid-State Sensor, Actuator, and Microsystems Workshop, Hilton Head, South Carolina, June 6-10, 2010, pp. 467-470. 130Ω achieved by this work at 61 MHz further bolster the notion that there are very few limits to the impedance-Q combinations achievable by capacitively transduced micromechanical resonators. PARTIAL ALD ELECTRODE-TO-RESONATOR GAP FILLING Partial ALD gap-filling, first demonstrated in [8], comprises a clever way to achieve tiny sub-30nm electrode-to-resonator gaps without the need to etch tiny sacrificial spacer layers, and thus, without the associated yield losses. As described in Fig. 3, the method involves first fabricating and releasing a micromechanical resonator, e.g., a disk resonator, with relatively wide initial electrode-to-resonator gap-spacings that permit good device yield; then coating the resonator and its electrodes with a conformal high-k dielectric ALD film. If the dielectric constant of the dielectric material is sufficiently high, i.e., if it is a “high-k” material, the effective electrode-to-resonator gap spacing of the coated device will be much smaller than the original spacing, and very close to the physical spacing between the dielectric surfaces, as shown in Fig. 3. Since ALD [13] provides a very uniform and conformal film deposition with <1nm thickness accuracy controlled simply by the number of reaction half cycles used, one can tune the final gap spacing to any desired value very accurately, provided the initial gap spacing is well known. This equates to an ability to achieve virtually any low motional resistance in accordance with (1), while simultaneously improving ωFOM and avoiding the excessive chip area (for arraying) and high bias voltage that would otherwise be needed if large gaps are used. As mentioned, although the work of [8] successfully reduced the gap spacing of a 61-MHz wineglass mode disk resonator from 94 nm to 32 nm by coating the released resonator with a 30.7 nm HfO2 ALD film, it did so with a penalty amounting to a 20x reduction in resonator Q, from the original uncoated Q of 150,527 to a much lower value of 7,368. Needless to say, if not mitigated in some way, such a reduction in Q renders the described partial-gap filling approach quite ineffectual, since Q’s less than 30,000 are not sufficient for the RF channel-selection or low phase noise applications targeted by MEMS-based vibrating resonators [5]. With all other parameters that influence energy loss (e.g., pressure, temperature, device geometry, anchor size) held constant before and after the HfO2 ALD coating, the most probable explanation for the reduction in resonator Q is added surface loss arising from poor HfO2 ALD film quality, and this was indeed postulated in [8]. If surface losses are the main culprit here, then methods for improving ALD film quality are required. Indeed, the need to mitigate surface-film-based loses becomes even more urgent as devices are scaled towards higher frequency, since scaling raises the surface-to-volume ratio, making surface-derived loss mechanisms even more prominent. IMPROVING ALD FILM QUALITY Pursuant to reducing surface-derived losses, this work attempts to improve the ALD film quality by more properly nucleating the high-k dielectric ALD as it grows. In particular, the key to the order of magnitude Q boost in this work is the introduction of an interfacial 3nm Al2O3 ALD pre-coating before ALD of the much higher-k gap-filling dielectric film—in this case, 25nm of TiO2—shown in Fig. 4(c). The need for such a pre-coating is governed by the ALD deposition chemistry of TiO2, which requires a healthy forest of dangling OH bonds over the starting surface to achieve a conformal and high quality film [14]. Without the precoating of Al2O3, the number of free OH bonds on the silicon surface is comparatively sparse, resulting in nucleation problems for the TiO2 film that in turn lead to Q-reducing surface defects and losses (Fig. 5(a)). Al2O3 is an excellent choice for the pre-coating, because it depends less on a perfectly hydroxylated surface; i.e., it Fig. 5: a) Good quality TiO2 ALD film requires good surface hydroxylation, i.e. a rich and uniform OH (hydroxyl) layer, which silicon lacks, leading to a nucleation problem and resultant surface loss mechanism that reduces the resonator Q. b) Al2O3 grows good quality films on various substrates, even without good hydroxylation, and also provides a dense and uniform forest of AlOH hydroxyl group. This then leads to a uniform subsequent TiO2 ALD, eliminating surface loss mechanisms that otherwise limit the resonator Q. Fig. 3: Schematics summarizing the partial gap-filling gap reduction method of [8], where ALD of a high-k dielectric film effectively reduces the gap distance between the electrode and the disk. Fig. 4: SEM images of a) a polysilicon micromachined wine-glass mode disk resonator before gap tuning and b) cross-section after conformal ALD of 3nm Al2O3 + 25nm TiO2. c) Schematic showing reduction of the resonator-to-electrode gap size from 87nm to 37nm via ALD partial-gap filling. M. Akgul, B. Kim, Z. Ren, and C. T.-C. Nguyen, “Capacitively transduced micromechanical resonators with simultaneous low motional resistance and Q >70,000,” Tech. Digest, 2008 Solid-State Sensor, Actuator, and Microsystems Workshop, Hilton Head, South Carolina, June 6-10, 2010, pp. 467-470. ALDs into good quality films on various substrates, including silicon, whether or not the starting OH forest is sparse [15]. Of course, Al2O3’s dielectric constant (k=9) is too low to serve as the primary gap-filling material, so the thickness of the Al2O3 film should be kept small: for this work, on the order of 3 nm. Although very thin, this Al2O3 film still provides the desired OH surface for the much higher-k (k=80) TiO2 film to follow [16], as described in Fig. 5(b). Now, with Al2O3 present, the TiO2 film can be deposited free of nucleation defects, hence, with much better quality and fewer sites for defect-derived energy loss. As a result, this Al2O3-TiO2 bilayer approach achieves an optimal solution that provides simultaneous narrow gaps for low motional resistance and low surface losses for high Q. ALD PARTIAL GAP FILLING PROCESS FLOW Fig. 6 summarizes the process flow for the ALD partial gap filling process, with the new measures described above to pre-coat structure surfaces with an Al2O3 seed layer prior to ALD of the main partial gap filling high-k dielectric TiO2. The process starts with conventional steps [17] to achieve an unreleased 61-MHz wineglass mode disk resonator in polysilicon with 87 nm lateral electrode-to-resonator gaps temporarily filled with sacrificial high temperature oxide (HTO), as shown in Fig. 6(a). The resonator is then dipped in a 49% HF acid bath for 30 minutes to etch the sacrificial oxide and release the structure, yielding the cross-section of Fig. 6(b). Because the gaps are 87 nm, and not smaller, the yield of the release etch can be quite high, in excess of ~95%. Next, the released resonator is coated with a 3nm Al2O3 ALD to achieve the cross-section of Fig. 6(c). This Al2O3 layer is quite conformal over the polysilicon surface, which is necessary if it is to serve as a seed layer providing a uniform and dense AlOH hydroxylated surface for a subsequent ALD of 25 nm thick TiO2. Fig. 6(d) presents the final cross-section of the device, for which the final effective gap spacing is 37 nm. For comparative purposes, versions of the process were also run without the Al2O3 pre-coating and just ALD’ing 25 nm TiO2 directly over the polysilicon resonator surfaces. To allow for electrical connection to bond pads, a final lithography step is performed to cover all released devices and expose their bond pads, after which a very short HF dip removes ALD material over the bondpads. Photoresist is then removed via acetone, and critical point drying applied to clean devices. EXPERIMENTAL RESULTS To verify the efficacy of Al2O3 pre-coating as a method for reducing surface losses, the frequency characteristics of 61-MHz wine-glass disk resonators treated with various ALD-coating recipes were measured under 1 Torr vacuum in direct two-port configurations using an HP 5070B network analyzer. All tested devices were identical prior to ALD-coating, and all were measured under the exact same electrical connection and vacuum conditions. Fig. 7 compares measured frequency characteristics for 61-MHz wine-glass disk resonators with electrode-to-resonator gaps partially filled with 25nm of TiO2 only, without the Al2O3 seed layer; and 3nm Al2O3+25nm TiO2; showing Q’s of 54,158 and 73,173, respectively, which clearly indicates the benefits of Al2O3 precoating. Comparison of these results with data from [8] (included in the plot) also makes clear the importance of the high-k dielectric film type and quality, since the Q of 54,158 achieved via the TiO2only case is still much higher than the HfO2-only case of [8], which used an inferior ALD tool to deposit HfO2. It should be noted that ALD film coating does change the resonance frequency of each disk, most likely due to a combination of mass loading and an equivalent stiffness change caused by ALD film stress. Therefore, to better compare resonator Q’s and peak heights for different ALD coating experiments, the data in Fig. 7 is normalized to a common frequency axis by applying to all data points the expression