combined capacitive and piezoelectric ... - Semantic Scholar
COMBINED CAPACITIVE AND PIEZOELECTRIC TRANSDUCTION FOR HIGH PERFORMANCE SILICON MICRORESONATORS A.K. Samarao and F. Ayazi Georgia Institute of Technology, Atlanta, Georgia, USA (TCF). The native TCF of approximately –30 ppm/˚C in silicon is significantly larger in magnitude than that of the worst AT-cut quartz resonator [8]. We reported earlier on a capacitively transduced ‘concave’ silicon microresonator geometry that offers an intrinsic TCF compensation unlike conventional rectangular or circular geometries [9, 10]. In this work, we report on similar concave geometries to further compare and study the TCF behavior under piezoelectric-only and combined capacitive and piezoelectric transduction.
ABSTRACT This paper introduces the Aluminum Nitride – High Aspect-Ratio Polysilicon and Single-crystal Silicon (AlN – HARPSS) process technology that for the first time enables combined capacitive (via air-gaps) and piezoelectric (via Mo/AlN/Mo piezo-stack) transduction in silicon micromechanical resonators. Lateral air-gaps as small as 150 nm have been realized for a 20 µm thick microresonator (air-gap aspect-ratio = 133:1) while simultaneously improving the c-axis orientation of aluminum nitride sputtered on its top surface. Such a combined transduction has been demonstrated to efficiently harvest the individual advantages of both the technologies. A 100 MHz silicon microresonator under combined capacitive and piezoelectric transduction measures a ~25 dB reduction in feedthrough compared to a capacitive-only transduction while measuring a 106% improvement in quality factor (Q), 10 dB reduction in insertion loss (I.L.) and a substantial suppression of spurious modes compared to a piezoelectric-only transduction.
FABRICATION In the AlN-HARPSS process, the sputtered Mo/AlN/Mo on a SOI substrate (Figure 1(a) & 1(b)) followed by the steps involved in the conventional HARPSS process [2]. The top Mo defines the piezoelectric drive/sense electrodes while the bottom Mo can be optional and is used to improve the interface between AlN and the underlying high-resistivity (> 1000 Ω-cm) silicon. Using a PECVD oxide mask, trenches are etched using the Bosch DRIE process to define the lateral dimensions of the resonating element (Figure 1(c)).
INTRODUCTION After four decades of relentless research efforts, silicon micromechanical resonator/oscillator technology has finally matured into a viable alternative for quartz in timing and frequency control applications [1]. Among the various possible transduction mechanisms for silicon microresonators, capacitive [2] and piezoelectric [3] technologies have been extensively explored owing to their specific individual advantages. While the former offers high quality factor (Q) and the best possible spectral purity [4], the latter offers low insertion loss (I.L.), lower feedthrough levels and ease of fabrication [5]. The need for achieving all the aforementioned advantages from a single resonator unit is becoming increasingly important as we move onto chip-scale spectrum analysis and decades-wide carrier frequency generation for software-defined and cognitive radio applications [6]. We address the need by introducing a novel AlN-HARPSS fabrication process technology that enables a combined capacitive and piezoelectric transduction in silicon microresonators to efficiently harvest the advantages of both these technologies. Over the years, the conventional HARPSS process [2] has evolved to enable ultra uniform and smooth sub-µm air-gaps with air-gap aspect-ratios > 400:1 [7]. This work extends the HARPSS process to accommodate Mo/AlN/Mo for simultaneously effecting piezoelectric transduction in parallel with capacitive transduction. Although the silicon micromechanical resonators offer the many advantages discussed above, the most significant disadvantage is their temperature coefficient of frequency
978-1-4244-9634-1/11/$26.00 ©2011 IEEE
Figure 1: The AlN-HARPSS process flow. Such anisotropic etching of silicon leaves teflon (i.e.., C4F8) residues on the sidewalls that are typically removed with piranha (i.e.., sulfuric acid @ 120 °C + hydrogen peroxide). However, AlN is found to be instantaneously etched in piranha which requires the latter to be replaced by a commercially available post-etch residue remover like EKC-265TM. All subsequent steps involve exposure to high temperatures in the furnaces. Although the
169
MEMS 2011, Cancun, MEXICO, January 23-27, 2011
melting point of Mo is ~2600 ºC, it was observed to oxidize in O2 atmosphere at ~700 ºC while its quality severely degrades on exposure to N2 ambient at high temperatures. Hence, it is critical to retain the PECVD oxide for protecting the Mo as well as to avoid possible metal contamination in the furnaces (Figure 1(d)). A sacrificial LPCVD oxide layer deposited at 850 ºC defines the lateral capacitive air-gaps (150 nm in this work, Figure 1(e)). LPCVD polysilicon (at 588 ºC) is boron-doped at 1050 ºC, annealed at 1100 ºC and patterned to form the capacitive drive/sense electrodes (Figure 1(f) & 1(g)). As a final step, the device is released in hydrofluoric acid (HF) (Figure 1(h)). It is to be noted that the buffering agent (NH4F) in BOE was found to attack AlN requiring the devices to be released only in 49% HF. Interestingly, the exposure to very high furnace temperatures (~1100 ºC) anneals and improves the c-axis orientation of AlN as evident from the rocking curve measurements reported in Figure 2. The improvement in FWHM from 1.864 to 1.673 degrees translates to better transduction efficiency from the AlN layer that could result in lower insertion loss and higher-Q.
thick silicon. It is worth mentioning that the air-gap of 150 nm achieved in this work demonstrates an air-gap aspect-ratio of 133:1 in the 20 µm thick C–P–CBARs.
Figure 3: (a) Capacitively transduced Concave silicon Bulk Acoustic Resonator (CBAR) [9]; (b) Piezoelectrically transduced CBAR (P-CBAR); (c) Combined Capacitive and Piezoelectric transduction for CBAR (C-P-CBAR).
Figure 2: The Rocking Curve for the [0 0 2] c-axis orientation of AlN showing the FWHM before and after the high temperature process of AlN-HARPSS. Using AlN-HARPSS, the capacitive-only concave silicon bulk acoustic resonator (or CBAR [9], Figure 3(a)) and piezoelectric-only CBAR (or P–CBAR, Figure 3(b)) can be realized as a single device for combined capacitive and piezoelectric transduction (i.e., C–P–CBAR, Figure 3(c)). The DC polarization voltage (Vp) required for capacitive actuation is applied to the bottom Mo (at the ‘Vp Pad’, Figure 3(c)) and it also acts as the AC ground for piezoelectric actuation. The capacitive and piezoelectric drive/sense electrodes positioned around the center of the concave geometry (Figure 4(a)) actuate the targeted width-extensional-mode (WEM) of resonance as illustrated in Figure 5. The dimensions of the resonator are chosen to set the resonance frequency of the WEM at ~100 MHz. To study the effect of the combined transduction scheme for varying thickness of the piezo-stack and the silicon resonator, two types of C–P–CBARs were fabricated (Figure 4(b)). Type-I consists of a relatively thin piezo-stack (i.e., 500 nm AlN with 50 nm Mo) on a 20 µm thick silicon resonator while type-II consists of a comparable thickness of piezo-stack (i.e., 1 µm AlN with 100 nm Mo) on 3 µm
Figure 4: SEM images (a) showing close-up of the combined capacitive and piezoelectric transduction, and (b) illustrating the thickness of the piezo-stack for the 20 µm (Type-I) and 3 µm (Type-II) thick C-P-CBAR.
Figure 5: ANSYS simulation illustrating the targeted widthextensional mode (WEM) of resonance in the concave silicon resonator geometry. The undeformed resonator structure along with its dimensions is also shown.
170
RESULTS The results from four different transduction schemes are presented in this work (Figure 6). The combined capacitive (C) and piezoelectric (P) drive (D) and sense (S) is affected by shorting the capacitive and piezoelectric electrodes together for both drive and sense ([CP]D – [CP]S, Figure 6(iii)). At zero Vp, no capacitive actuation is initiated thereby reducing the transduction to piezoelectric-only drive and sense ([P]D – [P]S, Figure 6(i)). C–P–CBARs also offer the possibility to decouple the drive and sense signal paths to a greater extent by using a capacitive-drive piezoelectric-sense scheme ([C]D – [P]S, Figure 6(ii)) which could offer a larger reduction in feedthrough. To facilitate a fair comparison of the aforementioned transduction schemes with capacitive-only transduction, the same C–P–CBAR was converted to a CBAR (Figure 6(iv)) by removing the piezo-stack using piranha. All the reported results are measured in air, as-is without any parasitic de-embedding and at an input power level of 0 dBm.
Figure 7: Achieved spectral purity with increasing Vp using [CP] D – [CP] S plotted across a measured span of (a) 25 MHz (12801 points) for the 20 µm thick and (b) 50 MHz (16001 points) for the 3 µm thick C–P–CBAR. Quality Factor and Insertion Loss Although limited by support-loss, the additional influx of acoustic energy from capacitive actuation in the [CP]D – [CP]S offers an improvement in Q compared to [P]D – [P]S. A 106% improvement in the Q (14,000 vs. 6,800) is measured in the 20 µm thick C–P–CBAR with a Vp of 20 V (Figure 8(a)). A reduction in the resonance frequency with increasing Vp is attributed to the electrical spring softening known to occur in capacitively actuated resonators [2]. The overall motional impedance of the combined transduction (RmCP) is lower than that of a piezoelectric-only transduction (RmP) due to the additional capacitive motional impedance (RmC) in parallel. This reflects as a 10 dB improvement in insertion loss (I.L.) along with the improvement in Q. Further reduction in the air-gaps holds the potential to offer a Q closer to the maximum possible value of 26,000, as measured from the equivalent single-crystal silicon CBAR (Figure 6(iv)).
Figure 6 (i, ii, iii): Possible combinations of transduction in a C–P–CBAR. Figure 6(iv) shows the same C–P–CBAR converted to a CBAR (Figure 3(a)) by removing piezo-stack. Spectral Purity The piezo-stack, in addition to exciting the targeted WEM of resonance (Figure 5), is known to actuate other out-of-plane spurious modes in the silicon resonator [3]. To enhance the former while simultaneously suppressing the latter, a lateral capacitive transduction is effected in parallel in this work (Figure 4(a)). Figure 7 demonstrates the progressive suppression of spurious modes with increasing Vp under the combined transduction scheme (Figure 6(iii)). One of the spurs near the targeted main peak is found to remain unsuppressed and is understood to result from an in-plane WEM-like mode of resonance similar that of the main peak. Interestingly, some spurious modes in the 3 µm thick C–P–CBAR are found to exist only at a Vp of 15V and are thought to result from the composite nature of the thin resonator loaded with a relatively thick piezo-stack (Figure 7(b)). A Vp of 20 V is found to sufficiently suppress the only dominant spur in the 3 µm thick resonators to make the main peak at ~100 MHz dominant by ~30 dB in a span of 50 MHz.
171
The TCF corresponding to the high-resistivity silicon substrate used in this work is ~ –30 ppm/ºC. However, an inherent low-TCF (~ –20 ppm/ºC) resonance mode was reported to occur in the CBAR structure [9, 10]. Such a low-TCF mode occurs as a spurious mode in the 3 µm C-P-CBAR across all transduction schemes (Figure 9(ii)) and is absent in the [CP]D – [CP]S and [C]D – [P]S schemes of the 20 µm C-P-CBAR (Figure 9(i)). Interestingly, the unsuppressed spur reported in Figure 7 also measures a low-TCF of –20 ppm/ºC indicating the presence of another CBAR-like resonance mode. Further engineering of the resonator geometry to target such low-TCF spurious modes could potentially add TCF compensation to the aforementioned list of advantages of the combined capacitive and piezoelectric transduction.
Figure 8: Measured variation in Q and I.L. of the targeted main peak with increasing Vp using [CP] D – [CP] S in (a) a 20 µm and (b) 3 µm thick C–P–CBAR. Among the 3 µm C–P–CBARs, although a 71% improvement in Q (6,500 vs. 3,800) is measured (maximum possible Q = 12,000), the relatively large RmC (due to smaller area for capacitive transduction) in parallel with RmP offers only a slight reduction in RmCP that reflects as a ~2 dB improvement in I.L. (Figure 8(b)).
CONCLUSION A novel AlN-HAPRSS process has been developed to facilitate a combined capacitive and piezoelectric transduction in silicon micromechanical resonators. Collective improvement in spectral purity, quality factor, insertion loss and feedthrough has been achieved compared to either capacitive-only or piezoelectric-only transduction.
Feedthrough and TCF [CP]D – [CP]S is measured to offer a significant improvement in feedthrough (~25 dB) compared to the capacitive-only transduction (red vs. pink line). However, the excellent isolation between the drive and the sense signal paths in [C]D – [P]S offers the lowest feedthrough levels (~ 100 dB, black line) among all transduction schemes (Figure 9).
ACKNOWLEDGEMENTS This work was supported by Integrated Device Technology (IDT), Incorporated. The authors would like to thank the staff at the Nanotechnology Research Center (NRC) at the Georgia Institute of Technology for assistance with fabrication.
REFERENCES [1]
F. Ayazi, “MEMS for Integrated Timing and Spectral Processing,” Invited Paper, Proc. IEEE Custom Integrated Circuits Conference (CICC), pp. 65-72, 2009. [2] S. Pourkamali, G. K. Ho and F. Ayazi, “Low-impedance VHF and UHF capacitive SiBARs – Part 1: concept and fabrication & Part II: measurement and characterization,” IEEE Transaction on Electron Devices, vol.54, no.8, pp. 2017-2030, 2007. [3] R. Abdolvand, H. M. Lavasani and F. Ayazi, “Thin-Film Piezoelectric-onSilicon Resonators for High-Frequency Reference Oscillator Applications,” IEEE Trans. on Ultrasonics, Ferroelectrics and Frequency Control, vol. 55, no. 12, pp. 2596 – 2606, 2008. [4] A. K. Samarao and F. Ayazi, "Quality Factor Sensitivity to Crystallographic Misalignments in Silicon Micromechanical Resonators," Solid-State Sensors, Actuators, and Microsystems Workshop (Hilton Head 2010), pp. 479-482, 2010. [5] W. Pan and F. Ayazi, “Thin-Film Piezoelectric-on-Substrate Resonators with Q Enhancement and TCF Reduction,” Proc. IEEE Micro Electro Mechanical Systems, pp. 104-107, 2010. [6] B. Razavi, “Cognitive Radio Design Challenges and Techniques,” IEEE Journal of Solid-State Circuits, vol. 45, pp. 1542-1553, 2010. [7] A. K. Samarao and F. Ayazi, “Self-Polarized Capacitive Silicon Micromechanical Resonators via Charge Trapping,” to be presented at the IEEE International Electron Devices Meeting (IEDM), 2010. [8] W.-T. Hsu, et al., “Stiffness-Compensated Temperature-Insensitive Micromechanical Resonators,” Proc. IEEE Micro Electro Mechanical Systems (MEMS), pp. 731–734, 2002. [9] A. K. Samarao, G. Casinovi and F. Ayazi, “Passive TCF Compensation in High Q Silicon Micromechanical Resonators,” Proc. IEEE Micro Electro Mechanical Systems (MEMS), pp. 116-119, 2010. [10] A. K. Samarao and F. Ayazi, “Intrinsic Temperature Compensation of Highly Resistive High-Q Silicon Microresonaotors via Charge Carrier Depletion,” Proc. IEEE Intl. Freq. Ctrl. Sym. (IFCS), pp. 334-339, 2010.
Figure 9: Measured frequency response (25 MHz span) of the (a) 20 µm thick and (b) 3 µm thick C–P–CBAR from the four transduction schemes (Figure 6) along with the TCF values of the main and spurious peaks.
172
ABSTRACT This paper introduces the Aluminum Nitride – High Aspect-Ratio Polysilicon and Single-crystal Silicon (AlN – HARPSS) process technology that for the first time enables combined capacitive (via air-gaps) and piezoelectric (via Mo/AlN/Mo piezo-stack) transduction in silicon micromechanical resonators. Lateral air-gaps as small as 150 nm have been realized for a 20 µm thick microresonator (air-gap aspect-ratio = 133:1) while simultaneously improving the c-axis orientation of aluminum nitride sputtered on its top surface. Such a combined transduction has been demonstrated to efficiently harvest the individual advantages of both the technologies. A 100 MHz silicon microresonator under combined capacitive and piezoelectric transduction measures a ~25 dB reduction in feedthrough compared to a capacitive-only transduction while measuring a 106% improvement in quality factor (Q), 10 dB reduction in insertion loss (I.L.) and a substantial suppression of spurious modes compared to a piezoelectric-only transduction.
FABRICATION In the AlN-HARPSS process, the sputtered Mo/AlN/Mo on a SOI substrate (Figure 1(a) & 1(b)) followed by the steps involved in the conventional HARPSS process [2]. The top Mo defines the piezoelectric drive/sense electrodes while the bottom Mo can be optional and is used to improve the interface between AlN and the underlying high-resistivity (> 1000 Ω-cm) silicon. Using a PECVD oxide mask, trenches are etched using the Bosch DRIE process to define the lateral dimensions of the resonating element (Figure 1(c)).
INTRODUCTION After four decades of relentless research efforts, silicon micromechanical resonator/oscillator technology has finally matured into a viable alternative for quartz in timing and frequency control applications [1]. Among the various possible transduction mechanisms for silicon microresonators, capacitive [2] and piezoelectric [3] technologies have been extensively explored owing to their specific individual advantages. While the former offers high quality factor (Q) and the best possible spectral purity [4], the latter offers low insertion loss (I.L.), lower feedthrough levels and ease of fabrication [5]. The need for achieving all the aforementioned advantages from a single resonator unit is becoming increasingly important as we move onto chip-scale spectrum analysis and decades-wide carrier frequency generation for software-defined and cognitive radio applications [6]. We address the need by introducing a novel AlN-HARPSS fabrication process technology that enables a combined capacitive and piezoelectric transduction in silicon microresonators to efficiently harvest the advantages of both these technologies. Over the years, the conventional HARPSS process [2] has evolved to enable ultra uniform and smooth sub-µm air-gaps with air-gap aspect-ratios > 400:1 [7]. This work extends the HARPSS process to accommodate Mo/AlN/Mo for simultaneously effecting piezoelectric transduction in parallel with capacitive transduction. Although the silicon micromechanical resonators offer the many advantages discussed above, the most significant disadvantage is their temperature coefficient of frequency
978-1-4244-9634-1/11/$26.00 ©2011 IEEE
Figure 1: The AlN-HARPSS process flow. Such anisotropic etching of silicon leaves teflon (i.e.., C4F8) residues on the sidewalls that are typically removed with piranha (i.e.., sulfuric acid @ 120 °C + hydrogen peroxide). However, AlN is found to be instantaneously etched in piranha which requires the latter to be replaced by a commercially available post-etch residue remover like EKC-265TM. All subsequent steps involve exposure to high temperatures in the furnaces. Although the
169
MEMS 2011, Cancun, MEXICO, January 23-27, 2011
melting point of Mo is ~2600 ºC, it was observed to oxidize in O2 atmosphere at ~700 ºC while its quality severely degrades on exposure to N2 ambient at high temperatures. Hence, it is critical to retain the PECVD oxide for protecting the Mo as well as to avoid possible metal contamination in the furnaces (Figure 1(d)). A sacrificial LPCVD oxide layer deposited at 850 ºC defines the lateral capacitive air-gaps (150 nm in this work, Figure 1(e)). LPCVD polysilicon (at 588 ºC) is boron-doped at 1050 ºC, annealed at 1100 ºC and patterned to form the capacitive drive/sense electrodes (Figure 1(f) & 1(g)). As a final step, the device is released in hydrofluoric acid (HF) (Figure 1(h)). It is to be noted that the buffering agent (NH4F) in BOE was found to attack AlN requiring the devices to be released only in 49% HF. Interestingly, the exposure to very high furnace temperatures (~1100 ºC) anneals and improves the c-axis orientation of AlN as evident from the rocking curve measurements reported in Figure 2. The improvement in FWHM from 1.864 to 1.673 degrees translates to better transduction efficiency from the AlN layer that could result in lower insertion loss and higher-Q.
thick silicon. It is worth mentioning that the air-gap of 150 nm achieved in this work demonstrates an air-gap aspect-ratio of 133:1 in the 20 µm thick C–P–CBARs.
Figure 3: (a) Capacitively transduced Concave silicon Bulk Acoustic Resonator (CBAR) [9]; (b) Piezoelectrically transduced CBAR (P-CBAR); (c) Combined Capacitive and Piezoelectric transduction for CBAR (C-P-CBAR).
Figure 2: The Rocking Curve for the [0 0 2] c-axis orientation of AlN showing the FWHM before and after the high temperature process of AlN-HARPSS. Using AlN-HARPSS, the capacitive-only concave silicon bulk acoustic resonator (or CBAR [9], Figure 3(a)) and piezoelectric-only CBAR (or P–CBAR, Figure 3(b)) can be realized as a single device for combined capacitive and piezoelectric transduction (i.e., C–P–CBAR, Figure 3(c)). The DC polarization voltage (Vp) required for capacitive actuation is applied to the bottom Mo (at the ‘Vp Pad’, Figure 3(c)) and it also acts as the AC ground for piezoelectric actuation. The capacitive and piezoelectric drive/sense electrodes positioned around the center of the concave geometry (Figure 4(a)) actuate the targeted width-extensional-mode (WEM) of resonance as illustrated in Figure 5. The dimensions of the resonator are chosen to set the resonance frequency of the WEM at ~100 MHz. To study the effect of the combined transduction scheme for varying thickness of the piezo-stack and the silicon resonator, two types of C–P–CBARs were fabricated (Figure 4(b)). Type-I consists of a relatively thin piezo-stack (i.e., 500 nm AlN with 50 nm Mo) on a 20 µm thick silicon resonator while type-II consists of a comparable thickness of piezo-stack (i.e., 1 µm AlN with 100 nm Mo) on 3 µm
Figure 4: SEM images (a) showing close-up of the combined capacitive and piezoelectric transduction, and (b) illustrating the thickness of the piezo-stack for the 20 µm (Type-I) and 3 µm (Type-II) thick C-P-CBAR.
Figure 5: ANSYS simulation illustrating the targeted widthextensional mode (WEM) of resonance in the concave silicon resonator geometry. The undeformed resonator structure along with its dimensions is also shown.
170
RESULTS The results from four different transduction schemes are presented in this work (Figure 6). The combined capacitive (C) and piezoelectric (P) drive (D) and sense (S) is affected by shorting the capacitive and piezoelectric electrodes together for both drive and sense ([CP]D – [CP]S, Figure 6(iii)). At zero Vp, no capacitive actuation is initiated thereby reducing the transduction to piezoelectric-only drive and sense ([P]D – [P]S, Figure 6(i)). C–P–CBARs also offer the possibility to decouple the drive and sense signal paths to a greater extent by using a capacitive-drive piezoelectric-sense scheme ([C]D – [P]S, Figure 6(ii)) which could offer a larger reduction in feedthrough. To facilitate a fair comparison of the aforementioned transduction schemes with capacitive-only transduction, the same C–P–CBAR was converted to a CBAR (Figure 6(iv)) by removing the piezo-stack using piranha. All the reported results are measured in air, as-is without any parasitic de-embedding and at an input power level of 0 dBm.
Figure 7: Achieved spectral purity with increasing Vp using [CP] D – [CP] S plotted across a measured span of (a) 25 MHz (12801 points) for the 20 µm thick and (b) 50 MHz (16001 points) for the 3 µm thick C–P–CBAR. Quality Factor and Insertion Loss Although limited by support-loss, the additional influx of acoustic energy from capacitive actuation in the [CP]D – [CP]S offers an improvement in Q compared to [P]D – [P]S. A 106% improvement in the Q (14,000 vs. 6,800) is measured in the 20 µm thick C–P–CBAR with a Vp of 20 V (Figure 8(a)). A reduction in the resonance frequency with increasing Vp is attributed to the electrical spring softening known to occur in capacitively actuated resonators [2]. The overall motional impedance of the combined transduction (RmCP) is lower than that of a piezoelectric-only transduction (RmP) due to the additional capacitive motional impedance (RmC) in parallel. This reflects as a 10 dB improvement in insertion loss (I.L.) along with the improvement in Q. Further reduction in the air-gaps holds the potential to offer a Q closer to the maximum possible value of 26,000, as measured from the equivalent single-crystal silicon CBAR (Figure 6(iv)).
Figure 6 (i, ii, iii): Possible combinations of transduction in a C–P–CBAR. Figure 6(iv) shows the same C–P–CBAR converted to a CBAR (Figure 3(a)) by removing piezo-stack. Spectral Purity The piezo-stack, in addition to exciting the targeted WEM of resonance (Figure 5), is known to actuate other out-of-plane spurious modes in the silicon resonator [3]. To enhance the former while simultaneously suppressing the latter, a lateral capacitive transduction is effected in parallel in this work (Figure 4(a)). Figure 7 demonstrates the progressive suppression of spurious modes with increasing Vp under the combined transduction scheme (Figure 6(iii)). One of the spurs near the targeted main peak is found to remain unsuppressed and is understood to result from an in-plane WEM-like mode of resonance similar that of the main peak. Interestingly, some spurious modes in the 3 µm thick C–P–CBAR are found to exist only at a Vp of 15V and are thought to result from the composite nature of the thin resonator loaded with a relatively thick piezo-stack (Figure 7(b)). A Vp of 20 V is found to sufficiently suppress the only dominant spur in the 3 µm thick resonators to make the main peak at ~100 MHz dominant by ~30 dB in a span of 50 MHz.
171
The TCF corresponding to the high-resistivity silicon substrate used in this work is ~ –30 ppm/ºC. However, an inherent low-TCF (~ –20 ppm/ºC) resonance mode was reported to occur in the CBAR structure [9, 10]. Such a low-TCF mode occurs as a spurious mode in the 3 µm C-P-CBAR across all transduction schemes (Figure 9(ii)) and is absent in the [CP]D – [CP]S and [C]D – [P]S schemes of the 20 µm C-P-CBAR (Figure 9(i)). Interestingly, the unsuppressed spur reported in Figure 7 also measures a low-TCF of –20 ppm/ºC indicating the presence of another CBAR-like resonance mode. Further engineering of the resonator geometry to target such low-TCF spurious modes could potentially add TCF compensation to the aforementioned list of advantages of the combined capacitive and piezoelectric transduction.
Figure 8: Measured variation in Q and I.L. of the targeted main peak with increasing Vp using [CP] D – [CP] S in (a) a 20 µm and (b) 3 µm thick C–P–CBAR. Among the 3 µm C–P–CBARs, although a 71% improvement in Q (6,500 vs. 3,800) is measured (maximum possible Q = 12,000), the relatively large RmC (due to smaller area for capacitive transduction) in parallel with RmP offers only a slight reduction in RmCP that reflects as a ~2 dB improvement in I.L. (Figure 8(b)).
CONCLUSION A novel AlN-HAPRSS process has been developed to facilitate a combined capacitive and piezoelectric transduction in silicon micromechanical resonators. Collective improvement in spectral purity, quality factor, insertion loss and feedthrough has been achieved compared to either capacitive-only or piezoelectric-only transduction.
Feedthrough and TCF [CP]D – [CP]S is measured to offer a significant improvement in feedthrough (~25 dB) compared to the capacitive-only transduction (red vs. pink line). However, the excellent isolation between the drive and the sense signal paths in [C]D – [P]S offers the lowest feedthrough levels (~ 100 dB, black line) among all transduction schemes (Figure 9).
ACKNOWLEDGEMENTS This work was supported by Integrated Device Technology (IDT), Incorporated. The authors would like to thank the staff at the Nanotechnology Research Center (NRC) at the Georgia Institute of Technology for assistance with fabrication.
REFERENCES [1]
F. Ayazi, “MEMS for Integrated Timing and Spectral Processing,” Invited Paper, Proc. IEEE Custom Integrated Circuits Conference (CICC), pp. 65-72, 2009. [2] S. Pourkamali, G. K. Ho and F. Ayazi, “Low-impedance VHF and UHF capacitive SiBARs – Part 1: concept and fabrication & Part II: measurement and characterization,” IEEE Transaction on Electron Devices, vol.54, no.8, pp. 2017-2030, 2007. [3] R. Abdolvand, H. M. Lavasani and F. Ayazi, “Thin-Film Piezoelectric-onSilicon Resonators for High-Frequency Reference Oscillator Applications,” IEEE Trans. on Ultrasonics, Ferroelectrics and Frequency Control, vol. 55, no. 12, pp. 2596 – 2606, 2008. [4] A. K. Samarao and F. Ayazi, "Quality Factor Sensitivity to Crystallographic Misalignments in Silicon Micromechanical Resonators," Solid-State Sensors, Actuators, and Microsystems Workshop (Hilton Head 2010), pp. 479-482, 2010. [5] W. Pan and F. Ayazi, “Thin-Film Piezoelectric-on-Substrate Resonators with Q Enhancement and TCF Reduction,” Proc. IEEE Micro Electro Mechanical Systems, pp. 104-107, 2010. [6] B. Razavi, “Cognitive Radio Design Challenges and Techniques,” IEEE Journal of Solid-State Circuits, vol. 45, pp. 1542-1553, 2010. [7] A. K. Samarao and F. Ayazi, “Self-Polarized Capacitive Silicon Micromechanical Resonators via Charge Trapping,” to be presented at the IEEE International Electron Devices Meeting (IEDM), 2010. [8] W.-T. Hsu, et al., “Stiffness-Compensated Temperature-Insensitive Micromechanical Resonators,” Proc. IEEE Micro Electro Mechanical Systems (MEMS), pp. 731–734, 2002. [9] A. K. Samarao, G. Casinovi and F. Ayazi, “Passive TCF Compensation in High Q Silicon Micromechanical Resonators,” Proc. IEEE Micro Electro Mechanical Systems (MEMS), pp. 116-119, 2010. [10] A. K. Samarao and F. Ayazi, “Intrinsic Temperature Compensation of Highly Resistive High-Q Silicon Microresonaotors via Charge Carrier Depletion,” Proc. IEEE Intl. Freq. Ctrl. Sym. (IFCS), pp. 334-339, 2010.
Figure 9: Measured frequency response (25 MHz span) of the (a) 20 µm thick and (b) 3 µm thick C–P–CBAR from the four transduction schemes (Figure 6) along with the TCF values of the main and spurious peaks.
172
