© Elsevier accepted for publication in Sensors and Actuators A 14 MHz Micromechanical Oscillator T. Mattila1, O. Jaakkola1, J. Kiihamäki2, J. Karttunen2, T. Lamminmäki3, P. Rantakari3, A. Oja1, H. Seppä1, H. Kattelus2 and I. Tittonen3 1 VTT Automation, P. O. Box 1304, FIN-02044, Finland 2 3 VTT Electronics, P. O. Box 1101, FIN-02044, Finland Helsinki University of Technology, Metrology Research Institute, P. O. Box 3000, FIN-02015, Finland SUMMARY Operation of a 14 MHz micromechanical oscillator is demonstrated and analyzed. Single-crystal silicon microbridge with submicron electrode gaps is utilized as the resonator element. The oscillator shows a noise floor of -120 dBc/Hz and a nearcarrier noise of -105 dBc/Hz at 1 kHz offset. The oscillator noise is dominated by amplifier noise. By comparing the oscillator performance with conventional quartz oscillators, advantages and limitations of the micromechanical components in RFtechnology are discussed. Keywords: micromechanical oscillator, oscillator noise © Elsevier accepted for publication in Sensors and Actuators A INTRODUCTION As the number of wireless communication devices increases, the small size and integration possibilities of micromechanical rf-components become significant advantages. Since the characteristic frequency scales roughly inversely proportional to the size of the component, recent developments in fabrication technology have made possible the realization of microresonators operating in the VHF-to-UHF range (10-100 MHz) miniaturized components [1-5]. However, the small size of the brings numerous challenges to be overcome: in general, the gap between vibrational and thermal energies becomes narrower [6] making a high signal-to-noise ratio harder to obtain. Also, the impedance levels of rf-microcomponents are typically high, making the noise-matching of electromechanical coupling interface problematic. In the present paper we address the above issues by analysing a 14 MHz oscillator based on a micromechanical Si resonator. Although showing significantly improved noise performance compared with the existing work [1,2,4], the present oscillator is found to exhibit a modest performance when compared with macroscopic quartz oscillators [6]. The central performance limitations are discussed. RESONATOR FABRICATION Fig. 1 shows the silicon micromechanical clamped-clamped (bridge) resonator utilized in the present study. The width (length) of the beam was 4 (44) µm, respectively. The component was manufactured using deep reactive ion etching (DRIE) of silicon-on-insulator (SOI) wafer with 8 µm thick device layer. The resonating beam was released from the oxide layer by wet-etching with HF. Coupling to the mechanical vibration of the bridge was made capacitively across the © Elsevier accepted for publication in Sensors and Actuators A gap d = 0.5 µm using two electrodes with the length Le = 34 µm as shown in Fig. 1. Electrical contacts to the structure were made using wire bonding to 100 µm x 100 µm contact pads. Each pad resulted into a parasitic capacitance of Cpad ~ 0.3 pF to the subsrate through the 1 µm thick SiO2 layer. The substrate was grounded to remove the feed-through via the pad capacitances. RESONATOR MEASUREMENT The bridge resonator was driven by one of the electrodes while the other was used for the detection of vibration (Fig. 2). UDC=100 V bias voltage was applied for the capacitive coupling. The resonator was placed in a vacuum chamber with pressure p < 10-2 mbar causing negligible air damping. The resonance current was detected with an amplifier block with a low-noise JFET (Philips BF545B) as the first stage. The amplifier was run in a common-source configuration with drain resistor Rd ~ 1 kΩ corresponding to a 12 dB voltage gain. The input capacitance was estimated as Cin ~ 5 pF from measurements using known capacitor values in place on the resonator. Due to the capacitive detection of the resonance currect, loading of the resonator Q remained insignificant. The JFET was followed by additional amplifier stages forming the output buffer to 50 Ω impedance. The transmission data were recorded with a HP4396B network analyser. Fig. 3 shows the measured resonator transmission data. The resonance was detected at fr ~ 14.3 MHz with a quality factor of Q ~ 1500. The rather low Q-value is set by the low aspect ratio (L/w = 11) leading to energy leakage to the support structures [7]. The feedthrough parasitic capacitance (parallel with the resonator) estimated from the measured data was Cpthru ~ 17 fF. Equivalent electrical circuit values [8] (see Fig. 4) reproducing the measured data were found to © Elsevier accepted for publication in Sensors and Actuators A be Rm ~ 1 MΩ, Cm ~ 8 aF, Lm ~ 15 H. The coupling gap capacitance between the bridge and the fixed electrodes was calculated as Cw ~ 4 fF. OSCILLATOR CIRCUIT DESIGN AND MEASUREMENT The resonator measurement setup was converted to an oscillator closed-loop configuration by connecting the amplifier output to resonator input as shown in Fig. 4. The JFET was followed by additional gain stages [represented by the amplifier block containing (G,φ)] to fulfill the selfoscillation condition: the amplifier gain was set to just compensate for the resonator dissipation Rm at closed-loop phase shift set equal to zero. A high-impedance buffer amplifier connected to the low-impedance (50 Ω) output of the sustaining amplifier allowed the measurement of the oscillation signal without perturbing the oscillator operation. In the resonator output, the effect of parasitic pad capacitance Cp ~ 0.3 pF was small due to the dominance of the amplifier input capacitance Cin ~ 5 pF. In the resonator input, the role of the parasitic pad capacitance (Z(Cp) ~ 45 kΩ) was mainly to increase the oscillator power consumption by drawing a major part of the sustaining amplifier output current. For the oscillator measurement, the setup was further refined by a careful shielding of the fringe fields, reducing the feedthrough capacitance to Cpthru ~ 3.5 fF, and the bias voltage was raised to UDC ~ 130 V. HP4396B was used to measure the oscillator output spectrum. The sharp peak in Fig. 5 demonstrates the succesful oscillator operation. The output carrier level, reduced to the amplifier input, was vs ~ 2.0 mV (rms). This corresponds to a mechanical vibration amplitude of a few percent of the gap: xvib ~ vs Cin η ~ 0.05 d = 0.025 µm , where η = U DC ∂C w is the ∂x electromechanical coupling constant. The oscillation amplitude was limited by the resonator nonlinearity [9]. Resonator transmission data measurements at high drive levels revealed that the vibration was constrained by both mechanical (spring-hardening) and capacitive (spring- © Elsevier accepted for publication in Sensors and Actuators A softening) non-linearities [10]. The cross-over from mechanical to capacitive non-linearity was measured to occur at UDC ~ 105 V bias voltage. The measured open-loop (UDC = 0 V, resonator switched off) noise level at the amplifier input was vn ~ 2.0 nV/ Hz (rms). Thus combined with the signal level of vs ~ 2.0 mV (rms), the expected N/S-ratio should approach -120 dBc/Hz. Direct measurement of such large N/S-ratios using a spectrum analyser is not straightforward due their limited dynamic range. This problem is further emphasized for measurement of noise at a small frequency offset from the carrier (oscillator spectral purity), which is the central quantity of interest here. To facilitate the measurement of near-carrier noise, we have used the standard technique [11] depicted in Fig. 6: a reference oscillator is phase-locked to the micromechanical oscillator. The feedback loop is constructed to keep relative phase between the two oscillators at 90 degress, resulting in zero DCoutput level in the mixer output. By filtering away the up-mixed carrier at 2ωr , the carriers have been removed and it is straighforward to measure the noise at audio frequencies, corresponding to the phase noise (amplitude noise “removed” by the phase lock condition) of the oscillators. Obviously, in order to determine the noise of the micromechanical oscillator, the reference oscillator must have a better noise performance: we have a used a HP E4426B precision signal source which was verified to give a single-side-band noise of Lf ~ -125 dBc/Hz at 1 kHz offset from a 14 MHz carrier. The phase-lock loop was operated at ~20 Hz bandwidth. The oscillator phase noise, measured using the outlined techique, is shown in Fig. 7. The phase noise level at 1 kHz offset from the carrier is ~ -105 dBc/Hz. At 10 kHz frequency offset the N/Sratio approaches a noise floor of -120 dBc/Hz, expected from the open-loop amplifier noise level. Below 1 kHz offset the phase noise increases rapidly (roughly as 1/f3, dashed line in Fig. 7); we tentatively associate this effect to the mixing of low-frequency noise to near-carrier frequencies © Elsevier accepted for publication in Sensors and Actuators A due to the non-linear operation point of the microresonator (the oscillation amplitude was limited by the resonator non-linearity). In general, we find that the oscillator noise is dominantly set by amplifier noise: this can be seen in detail by calculating the intrinsic resonator noise. The resonator mechanical dissipations are conveniently described by the equivalent circuit resistance Rm (Fig. 4). The associated noise voltage is obtained as v n , Rm = 4kTRm . Since the amplifier input impedance Zin = 1/ωrCin ~ 3 kΩ is much smaller than Rm ~1 MΩ and Cin >> Cpad, the resonator induced noise voltage in the amplifier input (at ωr) can be approximately obtained as v nin, Rm = Z (C in ) v n , Rm ~ 0.4 nV/ Hz . This is Rm thus much smaller than the measured open-loop effective amplifier noise of vn ~ 2.0 nV/ Hz . The domination of amplifier noise demostrates the noise-matching challenge with highimpedance levels in micromechanics. The noise-temperature TA for a typical amplifier seeing the noise-optimum impedance is negligible compared with room temperature; in the present case, an estimate for the optimum source impedance in the case of JFETs is Ropt ~ 1.63 Z(Cg), where Cg is the gate capacitance [12]. Noise matching of the impedance levels, i.e. Rm ~ Ropt, would mean that the resonator, being at T = 300 K, could be read with a negligible contribution of amplifier noise. In the present case, however, the resonator series resistance Rm ~ 1 MΩ is much larger than Ropt of a few kΩ's indicating that the system is far from noise-matched optimum. This is indeed reflected by the dominance of amplifier noise. However, it should be emphasized that the above analysis applies to near-carrier frequencies where the resonator impedance remains nearly constant at Rm; at larger frequency offsets reaching a noise optimum requires additional impedance matching techniques. © Elsevier accepted for publication in Sensors and Actuators A CONCLUSIONS The noise performance of a commercial 10 MHz quartz oscillator is typically better than -150 dBc/Hz at 1kHz offset from the carrier [6]. From the present study it is clear that, athough showing improvement over previous published work [1,2,4], the measured micromechanical oscillator short-term noise performance is currently clearly inferior to quartz oscillators. This is principally due to two facts: 1) The energy reservoir (resonator mechanical vibration energy) is unavoidably reduced as the size of the component is reduced. 2) Optimal electrical coupling (noise-matching) to the mechanical energy reservoir is difficult to accomplish, since the resonator is typically characterized by a high impedance. Several methods to improve the noise performance can be considered: The small size of microcomponents could allow the use of parallel resonators to increase the vibrational energy and lower the impedance. A lower resonator impedance level could be obtained also by a more optimized capacitive coupling (UDC close to pull-in voltage). Also, scaling down of the amplifier in size would improve the noise-matching. This would be feasible using a fabrication process that allows the integration of the resonator and amplifier on the same silicon chip. Using an integrated amplifier would also assist in reducing the parasitic capacitances, leading to a reduction in energy consumption to maintain the oscillations. ACKOWLEDGMENT We thank V. Ermolov, H. Kuisma, and T. Ryhänen for useful discussions. The financial support from Nokia Research Center, VTI Hamlin and Finnish National Technology Agency is acknowledged. © Elsevier accepted for publication in Sensors and Actuators A REFERENCES [1] T. A Roessig, R. T. Howe, A. P. Pisano and J. H. Smith, "Surface-Micromachined 1 MHz Oscillator with Low-Noise Pierce Configuration", Solid-State Sensor and Actuator Workshop, Transducer Res. Found., Cleveland, USA, (1998). [2] C. T.-C. Nguyen abd R. T. Howe, "An Integrated CMOS Micromechanical Resonator High-Q Oscillator", IEEE J. of Solid-State Circuits 34, 440 (1999). [3] K. Wang, A.-C. Wong, and C. T.-C. Nguyen, "VHF Free-Free Beam High-Q Micromechanical Resonators", J. Microelectromech. Syst. 9, 347 (2000). [4] S. Lee, M. U. Demirci, C. T.-C. Nguyen, "A 10-MHz Micromechanical Resonator Pierce Reference Oscillator for Communications", Digest of Technical Papers, Transducers '01 (Munich, Germany, 2001), p. 1094. [5] D. W. Carr, S. Evoy, L. Sekaric, H. G. Craighead and J.M. Parpia, "Measurement of mechanical resonance and losses in nanometer scale silicon wires", Appl. Phys. Lett. 75, 920 (1999). [6] J. R. Vig and Y. Kim, "Noise in Microelectromechanical System Resonators", IEEE Trans. Ultrason. Ferroelect. Freq. Contr. 46, 1558 (1999). [7] K. Y. Yasumura, T. D. Stowe, E. M. Chow, T. Pfafman, T. W. Kenny, B. C. Stipe and D. Rugar, "Quality Factors in Micron- and Submicron-Thick Cantilevers", J. Microelectromech. Syst. 9, 117 (2000). [8] T. Mattila, P. Häkkinen, O. Jaakkola, J. Kiihamäki, J. Kyynäräinen, A. Oja, H. Seppä, P. Seppälä and T. Sillanpää, "Air damping in resonant micromechanical capacitive sensors", in Proc. of 14th European Conference on Solid-State Transducers, Eurosensors XIV, (Copenhagen, Denmark, 2000) p. 221. © Elsevier accepted for publication in Sensors and Actuators A [9] C. Gui, R. Legtenberg, H. A. C. Tilmans, J. H. J. Fluitman, and M. Elwenspoek, "Nonlinearity and Hysteresis of Resonant Strain Gauges", J. Microelectromech. Systems 7, 122 (1998). [10] T. Veijola and T. Mattila, "Modeling of Nonlinear Micromechanical Resonators and Their Simulation with Harmonic-Balance Method", accepted for publication in RF and Microwave Computer-Aided Engineering. [11] see e.g. D. Allan, in ”Time and Frequency: Theory and Fundamentals”, National Bureau of Standards, USA, 1974, p.171. [12] M. J. Buckingham, Noise in Electronic Devices and Systems, Halsted, New York, 1983. © Elsevier accepted for publication in Sensors and Actuators A Figure captions: Fig. 1: SEM (scanning electron microscope) picture of the 14 MHz micromechanical bridge resonator. Bridge supports and outer parts of the electrodes are metallized. Fig. 2: Resonator measurement setup. Fig. 3: Amplitude g = Uout/Uin and phase of the resonator transmission response. Fig. 4: Oscillator setup. Fig. 5: Oscillator output spectrum in a 100 kHz range around the 14.3 MHz carrier. The spectrum analyser resolution bandwidth was 300 Hz Fig. 6: Oscillator phase noise measurement setup. Fig. 7: Oscillator single-sideband phase noise. The noise decay slope for 1/f3 is illustrated by dashed line. © Elsevier accepted for publication in Sensors and Actuators A VTT re so na to r el ec tro de el ec tro d e L = 44 um d = 0.5 um w = 4 um Mattila et al: Figure 1. © Elsevier accepted for publication in Sensors and Actuators A UDC G Uin ~ Mattila et al: Figure 2. network analyzer Uout © Elsevier accepted for publication in Sensors and Actuators A -44 100 |g| (dB) 80 -48 -50 60 -52 -54 14.24 10 kHz 14.26 14.28 14.30 Frequency (MHz) Mattila et al: Figure 3. 14.32 40 14.34 Phase (deg) -46 © Elsevier accepted for publication in Sensors and Actuators A Rd = 1 kΩ Ω Rm Cm Lm Cthru Cpad JFET BF545B Cpad Cin 0.3 pF 5 pF G,φ oscillator output Mattila et al.: Figure 4. buffer amplifier © Elsevier accepted for publication in Sensors and Actuators A Output spectrum (dBV) 0 -20 -40 -60 -80 -100 -50000 -25000 0 Frequency offset (Hz) Mattila et al.: Figure 5. 25000 50000 © Elsevier accepted for publication in Sensors and Actuators A SA ~ MEMS OSC LPF + ~ REF VCO Mattila et al.: Figure 6 PLL © Elsevier accepted for publication in Sensors and Actuators A SSB phase noise (dBc/Hz) -60 -70 -80 3 1/f -90 -100 -110 -120 -130 100 1000 Frequency offset (Hz) Mattila et al.: Figure 7. 10000