Effects of Release Hole Size on Microscale Phononic Crystals
Effects of Release Hole Size on Microscale Phononic Crystals
Y. M. Soliman1, D. F. Goettler1, Z. C. Leseman1, I. El-Kady1,2, R. H. Olsson* III2 1 University of New Mexico, Albuquerque, NM 2 Sandia National Laboratories, Albuquerque, NM Corresponding Author: 1 University if New Mexico, MSC01 1150, Albuquerque, NM 87131: [email protected] ABSTRACT We have experimentally investigated the effects of release hole size on acoustic transmission through micromachined phononic band gap crystals. The results confirm previous theoretical studies and determine the range of release hole sizes for which the band gap is minimally compromised. The experiments were performed on micro fabricated 2D phononic crystal plates comprised of high acoustic impedance tungsten rods arranged in a square lattice inside a low acoustic impedance silicon dioxide host medium. The lattice constant and diameter of the W rods in this experiment were 45 µm and 28.8 µm corresponding to a phononic band-gap centered at 67 MHz. Phononic crystals were characterized with 15, 12.5, 10, and 7.5 µm diameter release holes, required to undercut the crystal and suspend it from the substrate, placed in the center of the W rods. As the air hole diameter decreases, the band gap frequency width increases. However, as the diameter of the release holes approach 5 µm, the fabrication yield significantly decreases. From these experiments an optimum release hole diameter of 7.5 µm was found that maximizes phononic band gap performance and manufacturability. 1. INTRODUCTION The propagation of acoustic waves in phononic crystals has recently been given great research interest. Phononic crystals, or acoustic band gap crystals (ABGC), are periodic composite structures comprised of two elastic materials which differ in their acoustic impedances [1]. The increasing interest in ABGCs stems from the existence of complete acoustic band gaps (ABGs), which are ranges of frequencies over which acoustic waves are forbidden to propagate in any direction or polarization inside the crystal. The existence of ABGs suggests possible applications [1] such as acoustic filters, acoustic isolators and perfect phonon-filters, which selectively reflect phonons over a desired frequency range, allowing for perfect heat insulators [2]. Macro-scale ABGCs were the first to be constructed, limiting the operating frequency of the devices to the subMHz range, with the matrix material either being epoxy or water [3-4]. After a number of theoretical [5-11] and experimental [12-14] studies were done proving the existence of ABGs in ABGCs, micron-level devices were fabricated utilizing techniques from the mature microelectronics fabrication industry. Such small devices have many advantages. First, the operating frequency is tremendously higher in the case of micron-level devices because of their smaller feature sizes. In addition, such miniature ABGCs have applications in microelectronics and micro systems, especially in physically isolating devices from the vibration of other devices on the same substrate, or from the vibration of the substrate itself. Micro-machined ABGCs utilize the well-established microelectronics fabrication industry, leading to flexibility in rapidly realizing new research ideas and precision in fabrication. For the same reason, micro-machined ABGCs have higher yield when compared to hand-assembled macro-scale ABGCs. Surface micro-machined devices are fabricated through the deposition and etching of different layers of thin films on top of the substrate. After the process is complete, the last step is usually etching one of the layers deposited earlier in the process, called a release or sacrificial layer, to suspend the membrane containing the device to isolate it from the substrate. In many cases, wet etchants that are used to release the sacrificial layer can cause the suspended membrane to be attracted to the substrate by capillary forces, leading to stiction failure of the device. Another option is dry etchants, which are used in place of the wet etchant thereby removing the liquids
which generate the capillary forces, mainly responsible for stiction [15]. Also, wet etchants for Si generally etch slower and are less controlled than dry etchants [16]. In either case, wet or dry etchant, the release process requires a long time to laterally undercut large suspended membranes, and in the process may cause damage to other sections of the device. Thus periodic air holes, a.k.a. release holes, are strategically placed in the device region to allow the releasing of the device. The size of the release holes has a very pronounced effect on the width of the ABG, which is the main function of an ABGC. In this paper, the effect of the release hole size on the ABGs transmission is investigated. After discussing the fabrication process of the 67 MHz ABGC, experimental results are shown and discussed, after which some conclusions are drawn. 2. FABRICATION The micro-machined ABGC process, shown in figure 1, begins (a) with an oxide deposition of 0.6 µm followed by the deposition of a 2 µm undoped poly-Si release layer. The poly-Si is then removed around each device, limiting the area to which the release gas, SF6, has access to prevent suspension from the substrate in non-device areas of the wafer. The poly-Si patterning is followed by a deposition of a conformal oxide layer. By polishing this oxide layer to the level of the deposited poly-Si, mechanical support for the suspended membrane is secured and higher efficiency of SF6 etching is achieved, as less poly-Si is available. Next a 0.4 µm Al bottom interconnect layer is sputter deposited and patterned. This Al also serves to protect the bottom of the W inclusions during release from the SF6. The ABGC is formed through the deposition of 4 µm of SiO2 which is subsequently polished to remove the topography created by the Al. (b) Trenches in the oxide are etched followed by conformal deposition of a 1.2 µm thick layer of W which forms contacts to the bottom electrode of the AlN couplers and parts of the high density W inclusions. The W layer is polished until it remains only where the trenches were etched in the SiO2 producing W rings. (c) The formation of the W cylindrical scattering centers is completed by filling the rings using an additional oxide etch, W deposition and polish. Next a Ti/TiN/Al bottom electrode is deposited followed by the sputter deposition of 0.75 µm of AlN. The AlN film is highly c-axis oriented with an x-ray diffraction rocking curve full width half maximum of 1.5° which results in strong electro-acoustic coupling. The AlN is patterned and a 0.4 µm Al top electrode is deposited. This Al layer protects the top of the W inclusions during release from the SF6. (d) Finally, release holes are etched through the SiO2 in the center of the W inclusions, down to the poly-Si layer. The device is then released using SF6. The first two thin film depositions were added to the process to avoid etch loading effects during release, which may prevent the successful release of large suspended membranes through small release holes.
Figure 1. Micro-ABGC process flow, where d is the diameter of the W rods
(a) (b) Figure 2. a) Micro-ABGC device with integrated AlN electro-acoustic couplers – The ABGC is realized by including W scatterers in an oxide matrix. Acoustic frequencies within the gap can not propagate between the AlN couplers. Inset shows close up image of W scatterers and release holes. The lattice constant, a, is 45 µm and the inclusion diameter, d, is 28.8 µm, yielding a d/a ratio of 0.64. The release holes in the center of the W inclusions, in this experiment, vary in size, as discussed in the theoretical and experimental results sections below. b) A matrix corresponding to the crystal in (a). A 67 MHz micro-machined ABGC is shown in figure 2. Cylindrical Tungsten (W) scattering centers are periodically placed in square lattice in a SiO2 host material, with a periodicity of a. By placing tapered Aluminum Nitride (AlN) couplers on both sides of the ABGC, acoustic energy is coupled into, and out of, the ABGC in the form of longitudinal acoustic waves. Due to the appropriate mass density, acoustic velocity and acoustic impedance mismatches between the inclusion and the host materials, an ABG is formed. The tapered couplers are designed to provide wide drive and sense bandwidths, required to include the frequency range over which an ABG appears. A longitudinal wave is launched at the AlN coupler on one end of the ABGC and sensed at the AlN coupler at the other end. The transmission through the structure is then recorded as an S21 measurement. The effect of the periodic W scatterers, i.e. an ABGC, on acoustic transmission is shown in figure 5, and is obtained by normalizing the transmission through an ABGC, shown in figure 2a, to the transmission through an ABG matrix, shown in figure 2b. A detailed discussion of the device’s operation can be found in [13]. 3. THEORETICAL RESULTS To simulate ABGCs Finite Difference Time Domain (FDTD) was utilized. A detailed discussion of the FDTD method can be found in [17], in which space gridding allows for tracking the wave propagation in time while recording displacements caused by the propagating wave. Displacements are then transformed into a frequency spectrum and a transmission vs. frequency plot is generated. Figure 3 shows the FDTD transmission results for ABGCs with release holes of diameter 0, 5, 7.5, 10, and 12.5 µm. A band gap appears between 32 and 75 MHz for ABGCs with no air holes, as shown in figure 3a. Two conclusions can be drawn from figure 3. First, is placing air holes inside the periodic W scatterers in an ABGC induces propagating modes in the band gap. The second is that the mode shifts in frequency as the air hole size gets larger. There are many modes appearing because of the air holes, an example of which is a mode around 60 MHz for the device with 7.5 µm air holes. FDTD simulation for acoustic transmission through an ABGC is compared to experimental data in figure 4 for a 7.5 µm air hole device. Although the FDTD data differs greatly from the experimental data at frequencies below 60 MHz, it predicts the experimental data very accurately between 60 and 90 MHz, as can be seen clearly in
(a)
(b)
Figure 3. FDTD transmission results for ABGCs with release holes of diameter 0, 5, 7.5, 10, and 12.5 µm – The plots show a band gap between 32 and 75MHz with the release hole modes (circled) appearing inside the band gap and shifting to lower frequency for larger release hole size. The mode splits the band gap into two gaps, the sizes of which change according to the location of the release hole mode. figure 4. The reason for the discrepancy at lower frequencies is that the AlN electro-acoustic couplers were designed for optimum operation between 60 and 80 MHz, and therefore the transmitter and the sensor’s capability to launch and detect acoustic waves at lower frequencies is limited, thus providing maximum drive and sense only between 60 and 80 MHz. In comparing theoretical and experimental data in figure 4, a different yaxis was used for the FDTD (right) because FDTD is known to predict band gap depths much deeper than what can be physically realized since FDTD does not account for material loss mechanisms. Theoretical and experimental data in figure 4 show a good agreement in the location of the air hole mode inside the band gap, with the theoretical band gap shifted slightly higher in frequency. Although only one air hole mode appears in the experimental data, many others appear in FDTD as can be seen in figure 3.
Figure 4. Normalized transmission of acoustic waves in an ABGC with 7.5 µm diameter release holes in the center of the W rods – Theoretical data doesn’t agree with experimental data at lower frequency because of the limited bandwidth of the AlN coupler, but good agreement is seen between 60 and 80 MHz. The FDTD data is plotted on an independent y-axis (right) to account for the difference in the depth of the band gap between the FDTD and experimental data.
4. EXPERIMENTAL RESULTS AND DISCUSSION T r a n s m is s io n ( d B )
T ra n s m is s io n (d B )
10 0 -10 -20 -30 -40 55
60
65
70
75
80
85
10 0 -10 -20 -30 -40 55
90
60
65
Frequency (MHz)
a)
80
85
90
b)
10
10 T r a n s m is s io n (d B )
T r a n s m is s io n (d B )
70 75 Frequency (MHz)
0 -10 -20 -30
0 -10 -20 -30 -40
-40 55
60
65
70
75
80
85
90
55
60
Frequency (MHz)
c)
65
70 75 Frequency (MHz)
80
85
90
d)
Figure 5. Normalized transmission of acoustic waves in ABGCs fabricated with air hole diameters of a) 7.5 µm, b) 10 µm, c) 12.5 µm, and d) 15 µm. The air hole centered in the W rods has a pronounced effect on the band gap with a mode appearing between 75 and 85 MHz, and shifting to slightly lower frequency as the air hole size increases. ABGCs were fabricated with air hole diameter sizes of 5, 7.5, 10, 12.5 and 15 µm to observe the effect of the increasing air hole size on acoustic transmission through the device. As shown in figure 5, an air hole induced mode appears in the center of the band gap for devices with various air hole sizes. The experimental data in figure 5 shows that the air hole induced mode becomes more pronounced and shifts to a lower frequency as the air hole diameter increases, leading to a split in the ABG, which is detrimental to the operation of the device and therefore must be minimized. Table 1 shows a comparison between theoretical and experimental band gap width for the lower frequency band gap induced by the air hole, specifically between 60 and 80 MHz, the range over which the AlN couplers’ response is optimum. Table 1. Effect of Release Hole Diameter on Band gap Width for a Square Lattice ABGC with a Lattice Constant of 45 µm and a filling fraction, d/a, of 0.64
Release Hole Experimental ABG Diameter (μm) width (MHz)
Theoretical ABG width (MHz)
7.5
60-78
64-78
10
61-74
56-74
12.5
62-72
53-71
15
63-72
49-67
Although figure 5 indicates that the air hole size needs to be minimized for the widest band gap, practical issues arise when minimizing the air hole size, the most important of which is manufacturability. Step (d) in the
fabrication process uses an isotropic SF6 plasma etch to remove the poly-Si under the ABGC through both the release holes and large trenches on either side of the relatively large devices (430x900 µm). The lateral undercut is diffusion limited and requires long release times especially for small release holes. The etch rate of the oxide layer which protects the Si substrate from the SF6 etchant is finite. Due to the long etch durations required to undercut the ABGCs for very small release holes, the oxide directly over the substrate Si was removed in the large trenches before the entire device was undercut. The subsequent exposure of the Si substrate to the SF6 etchant causes substrate etch loading, effectively stopping the lateral undercut of the etch in the ABGC areas and preventing a full release from the substrate. Figure 6 shows the progress of the removal of the oxide film with time, leading to substrate etch loading. Etching the inner oxide protecting the W rods is also a concern. However, this does not take place nearly as fast as the etching of the oxide layer normal to the SF6 plasma, since the SF6 plasma etches the SiO2 much slower on vertical surfaces than it does on horizontal surface due to the low bias plasma characteristics. Therefore, the main failure mechanism is due to etch loading, not the SF6 attacking the W rods, which is clear from figures 2 and 6. Thus, etchant diffusion through the release holes has a great impact on fully releasing the devices. Devices with release holes of diameters 7.5 µm and greater were released within the allocated 15 minutes of release time, while the devices with 5 µm diameter air holes were not fully released. Further exposure to SF6 was ineffective, limited by etch loading as shown in figure 6, and therefore no further undercut etching took place. Devices with 5 µm were not functional because the remnants of the poly-Si layer connected the device to the substrate and altered propagation through the crystal. An alternative to releasing the devices in SF6 plasma is vapor etching using XeF2. In either case fluorine radicals etch Si through a mechanism that involves the production of SiF4, a volatile gas. XeF2 has the advantage of being much more selective between Si and SiO2 and the disadvantage of generally being a much slower etchant of silicon than SF6 [16]. XeF2 etches Si at a rate of 190 nm/min, Ti at a rate of 29 nm/min, W at a rate of 80 nm/min and does not etch Al or SiO2 [16]. Since XeF2 does not etch SiO2, and since there is no plasma involved in the etching process using XeF2, the problems experienced when etching with SF6 are now avoided, and thus a longer etching time in XeF2, required to achieve the same undercutting as with SF6, will not experience the same etch loading effect as those experienced with SF6. The aforementioned XeF2 process offers a solution to the problem encountered when releasing with SF6. However, during polishing of the W layer (processing steps b and c in the fabrication process detailed above), some small air holes were not fully cleared of W. Recesses in the oxide in the center of the W rods allowed for a thin film of W to remain above the oxide layer, which the polishing procedure didn’t clear in the case of the 5 µm diameter devices. During the plasma-enhanced procedure of removing the oxide from the release pit & and center of the W rods, the thin W film was removed, however, due to the presence of the W, the oxide layer was not fully removed down to the poly-Si release layer for a small subset of the 5 µm air hole diameter devices. This caused a problem during release, as XeF2 is incapable of penetrating the oxide film through the release holes and therefore no poly-Si was etched around these specific rods, leading to non-uniformity in etching as seen in the inset in figure 7a. This problem was not seen in the SF6 release because SF6 etches SiO2 at a much higher rate than XeF2. This problem can be avoided utilizing a polishing procedure with lower selectivity of W to SiO2. Another problem that occurred when releasing the 5 µm air hole diameter devices in XeF2 was the formation of an oxide layer on the remnants of the sacrificial Si layer when vacuum was pulled off of the device for testing before the poly-Si layer was fully etched. This is pictured in figure 7b. This thin oxide layer connects the suspended membrane, i.e. the ABGC, to the substrate, leading to failure of the device at higher frequencies. XeF2 etches oxide at an extremely low rate [16]. Thus, the native oxide formed when the wafer was exposed to atmosphere was not etched after placing the wafer back under vacuum for more etching. The measured gap width and center frequency is well predicted by the FDTD models. The depth of the band gap is underestimated in the measured data due to capacitive feed through between the input and output electrodes and is overestimated using FDTD because material losses are not supported in FDTD. Inside the band gap region, electrical feed through is higher in magnitude than the acoustic signal propagating through the phononic crystal and thus limits the maximum measurable rejection through the crystal. Measuring higher amounts of rejection requires either increasing the strength of the acoustic couplers or improving the electrical isolation between the input and output electrodes. A few methods for accomplishing this are larger AlN couplers, increasing the piezoelectric response by improving the AlN film quality, moving to high resistivity Si substrates and placing more physical distance between drive and sense pads.
Figures 4, 6 and 7 ascertain that a release hole of diameter 5 µm will lead to a decrease in device yield, unless further investigation and measures are undertaken. Devices with 7.5 µm release holes were released with no complications in SF6, and as shown in figure 4, had minimal air mode interference when compared to the response of devices with larger air hole sizes. Therefore, it is concluded that an ABGC with a 7.5 µm diameter air hole size maximizes the performance and manufacturability of this ABGC, while more research is needed to obtain a reliable method to release the 5 µm air hole devices, such as observing the effect of depositing a thicker oxide to protect from etch loading effects.
Figure 6. SF6 breaks through the oxide in the release pit areas causing etch loading during the release of an ABGC, shown in stages as etching time progresses in which a) the oxide starts to be etched in only a part of the release pit, b) the entire substrate can be seen through the release pit, but the oxide layer is not fully etched, and c) the entire release pit is etched . This takes place before the poly-Si release layer is fully etched under the ABGC, and therefore the 5 µm devices were not released as can be seen by the remnants of poly-Si.
(a) (b) Figure 7. A a) crystal, b) matrix unreleased after etching in XeF2 – Inset in a) Shows the non-uniformity of the XeF2 etch given the blockage of some 5 µm release holes, leading to poly-Si remaining around only a few W rods not being etched after 10 30-second XeF2 pulses. Inset in b) shows the profile of SiO2 which formed when vacuum was pulled off the etching chamber four times for testing before the entire Si layer was etched after 15 30-second pulses. The etching chamber pressure was at 1.5 Torr. It is worth noting that the air hole size problem exists in devices operating in the 100 MHz range, but as the devices get smaller for higher frequency operation, no release holes are needed since lateral undercutting is sufficient to suspend an entire ABGC above the substrate, and therefore the problem is altogether avoided, as discussed in [3]. 5. CONCLSIONS We have experimentally investigated the effects of release hole size on acoustic transmission through micromachined phononic band gap crystals. The square array of W inclusions in SiO2 2D ABGC were characterized with 15, 12.5, 10 and 7.5 µm diameter release holes placed in the center of the W rods, required to undercut the crystal and suspend it from the substrate. As the air hole diameter decreases, the band gap width was found to increase. However, as the diameter of the release holes approach 5 µm, the fabrication yield significantly decreases. From these experiments an optimum release hole diameter of 7.5 µm was found that maximizes phononic band gap performance and manufacturability for an ABGC with a lattice constant of 45 μm and a filling fraction, d/a, of 0.64. ACKNOWLEDGMENTS The authors would like to acknowledge the Microelectronics Development Laboratory (MDL) staff at Sandia National Laboratories for their efforts in fabricating the micro-machined phononic crystals. This work was supported by the Laboratory Directed Research and Development program at Sandia National Laboratories. Sandia National Laboratories is a multiprogram laboratory operated by the Sandia Corporation, Lockheed Martin Company, for the United States Department of Energy’s National Nuclear Security Administration under contract DE-AC04-94AL85000.
References [1] Olsson III R H et al, Microfabricated phononic crystal devices and applications, Meas. Sci. Technol. 20 012002, 2009 [2] Gorishnyy T, Ullal C K, Maldovan M, Fytas G and Thomas E L, Hypersonic phononic crystals, Phys. Rev. Lett. 94 115501, 2005 [3] Olsson III R H, El-Kady I and Tuck M, Microscale Phononic Band-Gap Crystals and Devices, Eurosensors, pp. 3-8, 2008 [4] Miyashita T, Sonic Crystals and Sonic Wave-Guides, Meas. Sci. Technol., No. 16, pp. R47-R63, 2005 [5] Sigalas M M and García N, Theoretical study of three dimensional elastic band gaps with the finite-difference time-domain method, J. Appl. Phys. 87 6 3122-5, 2000 [6] Kee C-S, Kim J-E, Park H Y, Chang K J and Lim H, Essential role of impedance in the formation of acoustic band gaps, J. Appl. Phys. 87 4 1593-6, 2000 [7] Kafesaki M and Economou E N, Multiple-scattering theory for three-dimensional periodic acoustic composites, Phys. Rev. B 60 17, 1999 [8] Hsu J-C and Wu T-T, Bleustein-Gulyaev-Shimizu surface acoustic waves in two-dimensional piezoelectric phononic crystals, IEEE Trans. Ultrasonics, Ferroelectrics and Freq. Cont. 53 6 1169-76, 2006 [9] Vasseur J O, Deymier P A, Chenni B, Djafari-Rouhani B, Dobrzynski L and Prevost D, Experimental and theoretical evidence for the existence of absolute acoustic band gaps in two-dimensional solid phononic crystals, Phys. Rev. Lett. 86 14 3012-5, 2001 [10] Hsiao F, Khelif A, Moubchir H, Choujaa A, Chen C and Laude V, Complete band gaps and deaf bands of triangular and honeycomb water-steel phononic crystals, J. Appl. Phys., 101 044903, 2007 [11] Yang S, Page J H, Liu Z, Cowan M L, Chan C T and Sheng P, Ultrasound tunneling through 3D phononic crystals, Phy. Rev. Lett. 88 10 104301, 2002 [12] Olsson III R H, Fleming J G, El-Kady I F, Tuck M R and McCormick F B, Micromachined bulk wave acoustic bandgap devices, Tech. Dig. Int. Conf. on Solid-State Sensors, Actuators, and Microsystems (Lyon) 31721, 2007 [13] Olsson III R H, El-Kady I F, Su M F, Tuck M R and Fleming J G, Microfabricated VHF acoustic crystals and Waveguides, Sensors and Actuators A: Physical, vol.145, pp. 87-93, 2008 [14] El-Kady I, Olsson III R H and Fleming J G, Phononic band-gap crystals for RF communications, Appl. Phys. Lett., 92, 233504, 2008 [15] Leseman Z C, Carlson S, and Mackin T J, Experimental measurements of the strain energy release rate for stiction failed microcantilevers using a single cantilever beam peel test, Journal of Microelectromechanical Systems, vol. 16, no. 1, pp. 38-43, 2007 [16] Kirt R. Williams, Kishan Gupta and Matthew Wasilik, Etch Rates for Micromachining Processing—Part II, Journal of Miroelectromechanical Systems, vol. 12, No.6, 2003 [17] P. Hsieh, T. Wu, and J. Sun, Three-Dimensional Phononic Band Gap Calculations Using the FDTD Method and a PC Cluster System, IEEE transactions on Ultrasonics, Ferroelectrics, and Frequency Control, vol. 53, no. 1, 2006
Y. M. Soliman1, D. F. Goettler1, Z. C. Leseman1, I. El-Kady1,2, R. H. Olsson* III2 1 University of New Mexico, Albuquerque, NM 2 Sandia National Laboratories, Albuquerque, NM Corresponding Author: 1 University if New Mexico, MSC01 1150, Albuquerque, NM 87131: [email protected] ABSTRACT We have experimentally investigated the effects of release hole size on acoustic transmission through micromachined phononic band gap crystals. The results confirm previous theoretical studies and determine the range of release hole sizes for which the band gap is minimally compromised. The experiments were performed on micro fabricated 2D phononic crystal plates comprised of high acoustic impedance tungsten rods arranged in a square lattice inside a low acoustic impedance silicon dioxide host medium. The lattice constant and diameter of the W rods in this experiment were 45 µm and 28.8 µm corresponding to a phononic band-gap centered at 67 MHz. Phononic crystals were characterized with 15, 12.5, 10, and 7.5 µm diameter release holes, required to undercut the crystal and suspend it from the substrate, placed in the center of the W rods. As the air hole diameter decreases, the band gap frequency width increases. However, as the diameter of the release holes approach 5 µm, the fabrication yield significantly decreases. From these experiments an optimum release hole diameter of 7.5 µm was found that maximizes phononic band gap performance and manufacturability. 1. INTRODUCTION The propagation of acoustic waves in phononic crystals has recently been given great research interest. Phononic crystals, or acoustic band gap crystals (ABGC), are periodic composite structures comprised of two elastic materials which differ in their acoustic impedances [1]. The increasing interest in ABGCs stems from the existence of complete acoustic band gaps (ABGs), which are ranges of frequencies over which acoustic waves are forbidden to propagate in any direction or polarization inside the crystal. The existence of ABGs suggests possible applications [1] such as acoustic filters, acoustic isolators and perfect phonon-filters, which selectively reflect phonons over a desired frequency range, allowing for perfect heat insulators [2]. Macro-scale ABGCs were the first to be constructed, limiting the operating frequency of the devices to the subMHz range, with the matrix material either being epoxy or water [3-4]. After a number of theoretical [5-11] and experimental [12-14] studies were done proving the existence of ABGs in ABGCs, micron-level devices were fabricated utilizing techniques from the mature microelectronics fabrication industry. Such small devices have many advantages. First, the operating frequency is tremendously higher in the case of micron-level devices because of their smaller feature sizes. In addition, such miniature ABGCs have applications in microelectronics and micro systems, especially in physically isolating devices from the vibration of other devices on the same substrate, or from the vibration of the substrate itself. Micro-machined ABGCs utilize the well-established microelectronics fabrication industry, leading to flexibility in rapidly realizing new research ideas and precision in fabrication. For the same reason, micro-machined ABGCs have higher yield when compared to hand-assembled macro-scale ABGCs. Surface micro-machined devices are fabricated through the deposition and etching of different layers of thin films on top of the substrate. After the process is complete, the last step is usually etching one of the layers deposited earlier in the process, called a release or sacrificial layer, to suspend the membrane containing the device to isolate it from the substrate. In many cases, wet etchants that are used to release the sacrificial layer can cause the suspended membrane to be attracted to the substrate by capillary forces, leading to stiction failure of the device. Another option is dry etchants, which are used in place of the wet etchant thereby removing the liquids
which generate the capillary forces, mainly responsible for stiction [15]. Also, wet etchants for Si generally etch slower and are less controlled than dry etchants [16]. In either case, wet or dry etchant, the release process requires a long time to laterally undercut large suspended membranes, and in the process may cause damage to other sections of the device. Thus periodic air holes, a.k.a. release holes, are strategically placed in the device region to allow the releasing of the device. The size of the release holes has a very pronounced effect on the width of the ABG, which is the main function of an ABGC. In this paper, the effect of the release hole size on the ABGs transmission is investigated. After discussing the fabrication process of the 67 MHz ABGC, experimental results are shown and discussed, after which some conclusions are drawn. 2. FABRICATION The micro-machined ABGC process, shown in figure 1, begins (a) with an oxide deposition of 0.6 µm followed by the deposition of a 2 µm undoped poly-Si release layer. The poly-Si is then removed around each device, limiting the area to which the release gas, SF6, has access to prevent suspension from the substrate in non-device areas of the wafer. The poly-Si patterning is followed by a deposition of a conformal oxide layer. By polishing this oxide layer to the level of the deposited poly-Si, mechanical support for the suspended membrane is secured and higher efficiency of SF6 etching is achieved, as less poly-Si is available. Next a 0.4 µm Al bottom interconnect layer is sputter deposited and patterned. This Al also serves to protect the bottom of the W inclusions during release from the SF6. The ABGC is formed through the deposition of 4 µm of SiO2 which is subsequently polished to remove the topography created by the Al. (b) Trenches in the oxide are etched followed by conformal deposition of a 1.2 µm thick layer of W which forms contacts to the bottom electrode of the AlN couplers and parts of the high density W inclusions. The W layer is polished until it remains only where the trenches were etched in the SiO2 producing W rings. (c) The formation of the W cylindrical scattering centers is completed by filling the rings using an additional oxide etch, W deposition and polish. Next a Ti/TiN/Al bottom electrode is deposited followed by the sputter deposition of 0.75 µm of AlN. The AlN film is highly c-axis oriented with an x-ray diffraction rocking curve full width half maximum of 1.5° which results in strong electro-acoustic coupling. The AlN is patterned and a 0.4 µm Al top electrode is deposited. This Al layer protects the top of the W inclusions during release from the SF6. (d) Finally, release holes are etched through the SiO2 in the center of the W inclusions, down to the poly-Si layer. The device is then released using SF6. The first two thin film depositions were added to the process to avoid etch loading effects during release, which may prevent the successful release of large suspended membranes through small release holes.
Figure 1. Micro-ABGC process flow, where d is the diameter of the W rods
(a) (b) Figure 2. a) Micro-ABGC device with integrated AlN electro-acoustic couplers – The ABGC is realized by including W scatterers in an oxide matrix. Acoustic frequencies within the gap can not propagate between the AlN couplers. Inset shows close up image of W scatterers and release holes. The lattice constant, a, is 45 µm and the inclusion diameter, d, is 28.8 µm, yielding a d/a ratio of 0.64. The release holes in the center of the W inclusions, in this experiment, vary in size, as discussed in the theoretical and experimental results sections below. b) A matrix corresponding to the crystal in (a). A 67 MHz micro-machined ABGC is shown in figure 2. Cylindrical Tungsten (W) scattering centers are periodically placed in square lattice in a SiO2 host material, with a periodicity of a. By placing tapered Aluminum Nitride (AlN) couplers on both sides of the ABGC, acoustic energy is coupled into, and out of, the ABGC in the form of longitudinal acoustic waves. Due to the appropriate mass density, acoustic velocity and acoustic impedance mismatches between the inclusion and the host materials, an ABG is formed. The tapered couplers are designed to provide wide drive and sense bandwidths, required to include the frequency range over which an ABG appears. A longitudinal wave is launched at the AlN coupler on one end of the ABGC and sensed at the AlN coupler at the other end. The transmission through the structure is then recorded as an S21 measurement. The effect of the periodic W scatterers, i.e. an ABGC, on acoustic transmission is shown in figure 5, and is obtained by normalizing the transmission through an ABGC, shown in figure 2a, to the transmission through an ABG matrix, shown in figure 2b. A detailed discussion of the device’s operation can be found in [13]. 3. THEORETICAL RESULTS To simulate ABGCs Finite Difference Time Domain (FDTD) was utilized. A detailed discussion of the FDTD method can be found in [17], in which space gridding allows for tracking the wave propagation in time while recording displacements caused by the propagating wave. Displacements are then transformed into a frequency spectrum and a transmission vs. frequency plot is generated. Figure 3 shows the FDTD transmission results for ABGCs with release holes of diameter 0, 5, 7.5, 10, and 12.5 µm. A band gap appears between 32 and 75 MHz for ABGCs with no air holes, as shown in figure 3a. Two conclusions can be drawn from figure 3. First, is placing air holes inside the periodic W scatterers in an ABGC induces propagating modes in the band gap. The second is that the mode shifts in frequency as the air hole size gets larger. There are many modes appearing because of the air holes, an example of which is a mode around 60 MHz for the device with 7.5 µm air holes. FDTD simulation for acoustic transmission through an ABGC is compared to experimental data in figure 4 for a 7.5 µm air hole device. Although the FDTD data differs greatly from the experimental data at frequencies below 60 MHz, it predicts the experimental data very accurately between 60 and 90 MHz, as can be seen clearly in
(a)
(b)
Figure 3. FDTD transmission results for ABGCs with release holes of diameter 0, 5, 7.5, 10, and 12.5 µm – The plots show a band gap between 32 and 75MHz with the release hole modes (circled) appearing inside the band gap and shifting to lower frequency for larger release hole size. The mode splits the band gap into two gaps, the sizes of which change according to the location of the release hole mode. figure 4. The reason for the discrepancy at lower frequencies is that the AlN electro-acoustic couplers were designed for optimum operation between 60 and 80 MHz, and therefore the transmitter and the sensor’s capability to launch and detect acoustic waves at lower frequencies is limited, thus providing maximum drive and sense only between 60 and 80 MHz. In comparing theoretical and experimental data in figure 4, a different yaxis was used for the FDTD (right) because FDTD is known to predict band gap depths much deeper than what can be physically realized since FDTD does not account for material loss mechanisms. Theoretical and experimental data in figure 4 show a good agreement in the location of the air hole mode inside the band gap, with the theoretical band gap shifted slightly higher in frequency. Although only one air hole mode appears in the experimental data, many others appear in FDTD as can be seen in figure 3.
Figure 4. Normalized transmission of acoustic waves in an ABGC with 7.5 µm diameter release holes in the center of the W rods – Theoretical data doesn’t agree with experimental data at lower frequency because of the limited bandwidth of the AlN coupler, but good agreement is seen between 60 and 80 MHz. The FDTD data is plotted on an independent y-axis (right) to account for the difference in the depth of the band gap between the FDTD and experimental data.
4. EXPERIMENTAL RESULTS AND DISCUSSION T r a n s m is s io n ( d B )
T ra n s m is s io n (d B )
10 0 -10 -20 -30 -40 55
60
65
70
75
80
85
10 0 -10 -20 -30 -40 55
90
60
65
Frequency (MHz)
a)
80
85
90
b)
10
10 T r a n s m is s io n (d B )
T r a n s m is s io n (d B )
70 75 Frequency (MHz)
0 -10 -20 -30
0 -10 -20 -30 -40
-40 55
60
65
70
75
80
85
90
55
60
Frequency (MHz)
c)
65
70 75 Frequency (MHz)
80
85
90
d)
Figure 5. Normalized transmission of acoustic waves in ABGCs fabricated with air hole diameters of a) 7.5 µm, b) 10 µm, c) 12.5 µm, and d) 15 µm. The air hole centered in the W rods has a pronounced effect on the band gap with a mode appearing between 75 and 85 MHz, and shifting to slightly lower frequency as the air hole size increases. ABGCs were fabricated with air hole diameter sizes of 5, 7.5, 10, 12.5 and 15 µm to observe the effect of the increasing air hole size on acoustic transmission through the device. As shown in figure 5, an air hole induced mode appears in the center of the band gap for devices with various air hole sizes. The experimental data in figure 5 shows that the air hole induced mode becomes more pronounced and shifts to a lower frequency as the air hole diameter increases, leading to a split in the ABG, which is detrimental to the operation of the device and therefore must be minimized. Table 1 shows a comparison between theoretical and experimental band gap width for the lower frequency band gap induced by the air hole, specifically between 60 and 80 MHz, the range over which the AlN couplers’ response is optimum. Table 1. Effect of Release Hole Diameter on Band gap Width for a Square Lattice ABGC with a Lattice Constant of 45 µm and a filling fraction, d/a, of 0.64
Release Hole Experimental ABG Diameter (μm) width (MHz)
Theoretical ABG width (MHz)
7.5
60-78
64-78
10
61-74
56-74
12.5
62-72
53-71
15
63-72
49-67
Although figure 5 indicates that the air hole size needs to be minimized for the widest band gap, practical issues arise when minimizing the air hole size, the most important of which is manufacturability. Step (d) in the
fabrication process uses an isotropic SF6 plasma etch to remove the poly-Si under the ABGC through both the release holes and large trenches on either side of the relatively large devices (430x900 µm). The lateral undercut is diffusion limited and requires long release times especially for small release holes. The etch rate of the oxide layer which protects the Si substrate from the SF6 etchant is finite. Due to the long etch durations required to undercut the ABGCs for very small release holes, the oxide directly over the substrate Si was removed in the large trenches before the entire device was undercut. The subsequent exposure of the Si substrate to the SF6 etchant causes substrate etch loading, effectively stopping the lateral undercut of the etch in the ABGC areas and preventing a full release from the substrate. Figure 6 shows the progress of the removal of the oxide film with time, leading to substrate etch loading. Etching the inner oxide protecting the W rods is also a concern. However, this does not take place nearly as fast as the etching of the oxide layer normal to the SF6 plasma, since the SF6 plasma etches the SiO2 much slower on vertical surfaces than it does on horizontal surface due to the low bias plasma characteristics. Therefore, the main failure mechanism is due to etch loading, not the SF6 attacking the W rods, which is clear from figures 2 and 6. Thus, etchant diffusion through the release holes has a great impact on fully releasing the devices. Devices with release holes of diameters 7.5 µm and greater were released within the allocated 15 minutes of release time, while the devices with 5 µm diameter air holes were not fully released. Further exposure to SF6 was ineffective, limited by etch loading as shown in figure 6, and therefore no further undercut etching took place. Devices with 5 µm were not functional because the remnants of the poly-Si layer connected the device to the substrate and altered propagation through the crystal. An alternative to releasing the devices in SF6 plasma is vapor etching using XeF2. In either case fluorine radicals etch Si through a mechanism that involves the production of SiF4, a volatile gas. XeF2 has the advantage of being much more selective between Si and SiO2 and the disadvantage of generally being a much slower etchant of silicon than SF6 [16]. XeF2 etches Si at a rate of 190 nm/min, Ti at a rate of 29 nm/min, W at a rate of 80 nm/min and does not etch Al or SiO2 [16]. Since XeF2 does not etch SiO2, and since there is no plasma involved in the etching process using XeF2, the problems experienced when etching with SF6 are now avoided, and thus a longer etching time in XeF2, required to achieve the same undercutting as with SF6, will not experience the same etch loading effect as those experienced with SF6. The aforementioned XeF2 process offers a solution to the problem encountered when releasing with SF6. However, during polishing of the W layer (processing steps b and c in the fabrication process detailed above), some small air holes were not fully cleared of W. Recesses in the oxide in the center of the W rods allowed for a thin film of W to remain above the oxide layer, which the polishing procedure didn’t clear in the case of the 5 µm diameter devices. During the plasma-enhanced procedure of removing the oxide from the release pit & and center of the W rods, the thin W film was removed, however, due to the presence of the W, the oxide layer was not fully removed down to the poly-Si release layer for a small subset of the 5 µm air hole diameter devices. This caused a problem during release, as XeF2 is incapable of penetrating the oxide film through the release holes and therefore no poly-Si was etched around these specific rods, leading to non-uniformity in etching as seen in the inset in figure 7a. This problem was not seen in the SF6 release because SF6 etches SiO2 at a much higher rate than XeF2. This problem can be avoided utilizing a polishing procedure with lower selectivity of W to SiO2. Another problem that occurred when releasing the 5 µm air hole diameter devices in XeF2 was the formation of an oxide layer on the remnants of the sacrificial Si layer when vacuum was pulled off of the device for testing before the poly-Si layer was fully etched. This is pictured in figure 7b. This thin oxide layer connects the suspended membrane, i.e. the ABGC, to the substrate, leading to failure of the device at higher frequencies. XeF2 etches oxide at an extremely low rate [16]. Thus, the native oxide formed when the wafer was exposed to atmosphere was not etched after placing the wafer back under vacuum for more etching. The measured gap width and center frequency is well predicted by the FDTD models. The depth of the band gap is underestimated in the measured data due to capacitive feed through between the input and output electrodes and is overestimated using FDTD because material losses are not supported in FDTD. Inside the band gap region, electrical feed through is higher in magnitude than the acoustic signal propagating through the phononic crystal and thus limits the maximum measurable rejection through the crystal. Measuring higher amounts of rejection requires either increasing the strength of the acoustic couplers or improving the electrical isolation between the input and output electrodes. A few methods for accomplishing this are larger AlN couplers, increasing the piezoelectric response by improving the AlN film quality, moving to high resistivity Si substrates and placing more physical distance between drive and sense pads.
Figures 4, 6 and 7 ascertain that a release hole of diameter 5 µm will lead to a decrease in device yield, unless further investigation and measures are undertaken. Devices with 7.5 µm release holes were released with no complications in SF6, and as shown in figure 4, had minimal air mode interference when compared to the response of devices with larger air hole sizes. Therefore, it is concluded that an ABGC with a 7.5 µm diameter air hole size maximizes the performance and manufacturability of this ABGC, while more research is needed to obtain a reliable method to release the 5 µm air hole devices, such as observing the effect of depositing a thicker oxide to protect from etch loading effects.
Figure 6. SF6 breaks through the oxide in the release pit areas causing etch loading during the release of an ABGC, shown in stages as etching time progresses in which a) the oxide starts to be etched in only a part of the release pit, b) the entire substrate can be seen through the release pit, but the oxide layer is not fully etched, and c) the entire release pit is etched . This takes place before the poly-Si release layer is fully etched under the ABGC, and therefore the 5 µm devices were not released as can be seen by the remnants of poly-Si.
(a) (b) Figure 7. A a) crystal, b) matrix unreleased after etching in XeF2 – Inset in a) Shows the non-uniformity of the XeF2 etch given the blockage of some 5 µm release holes, leading to poly-Si remaining around only a few W rods not being etched after 10 30-second XeF2 pulses. Inset in b) shows the profile of SiO2 which formed when vacuum was pulled off the etching chamber four times for testing before the entire Si layer was etched after 15 30-second pulses. The etching chamber pressure was at 1.5 Torr. It is worth noting that the air hole size problem exists in devices operating in the 100 MHz range, but as the devices get smaller for higher frequency operation, no release holes are needed since lateral undercutting is sufficient to suspend an entire ABGC above the substrate, and therefore the problem is altogether avoided, as discussed in [3]. 5. CONCLSIONS We have experimentally investigated the effects of release hole size on acoustic transmission through micromachined phononic band gap crystals. The square array of W inclusions in SiO2 2D ABGC were characterized with 15, 12.5, 10 and 7.5 µm diameter release holes placed in the center of the W rods, required to undercut the crystal and suspend it from the substrate. As the air hole diameter decreases, the band gap width was found to increase. However, as the diameter of the release holes approach 5 µm, the fabrication yield significantly decreases. From these experiments an optimum release hole diameter of 7.5 µm was found that maximizes phononic band gap performance and manufacturability for an ABGC with a lattice constant of 45 μm and a filling fraction, d/a, of 0.64. ACKNOWLEDGMENTS The authors would like to acknowledge the Microelectronics Development Laboratory (MDL) staff at Sandia National Laboratories for their efforts in fabricating the micro-machined phononic crystals. This work was supported by the Laboratory Directed Research and Development program at Sandia National Laboratories. Sandia National Laboratories is a multiprogram laboratory operated by the Sandia Corporation, Lockheed Martin Company, for the United States Department of Energy’s National Nuclear Security Administration under contract DE-AC04-94AL85000.
References [1] Olsson III R H et al, Microfabricated phononic crystal devices and applications, Meas. Sci. Technol. 20 012002, 2009 [2] Gorishnyy T, Ullal C K, Maldovan M, Fytas G and Thomas E L, Hypersonic phononic crystals, Phys. Rev. Lett. 94 115501, 2005 [3] Olsson III R H, El-Kady I and Tuck M, Microscale Phononic Band-Gap Crystals and Devices, Eurosensors, pp. 3-8, 2008 [4] Miyashita T, Sonic Crystals and Sonic Wave-Guides, Meas. Sci. Technol., No. 16, pp. R47-R63, 2005 [5] Sigalas M M and García N, Theoretical study of three dimensional elastic band gaps with the finite-difference time-domain method, J. Appl. Phys. 87 6 3122-5, 2000 [6] Kee C-S, Kim J-E, Park H Y, Chang K J and Lim H, Essential role of impedance in the formation of acoustic band gaps, J. Appl. Phys. 87 4 1593-6, 2000 [7] Kafesaki M and Economou E N, Multiple-scattering theory for three-dimensional periodic acoustic composites, Phys. Rev. B 60 17, 1999 [8] Hsu J-C and Wu T-T, Bleustein-Gulyaev-Shimizu surface acoustic waves in two-dimensional piezoelectric phononic crystals, IEEE Trans. Ultrasonics, Ferroelectrics and Freq. Cont. 53 6 1169-76, 2006 [9] Vasseur J O, Deymier P A, Chenni B, Djafari-Rouhani B, Dobrzynski L and Prevost D, Experimental and theoretical evidence for the existence of absolute acoustic band gaps in two-dimensional solid phononic crystals, Phys. Rev. Lett. 86 14 3012-5, 2001 [10] Hsiao F, Khelif A, Moubchir H, Choujaa A, Chen C and Laude V, Complete band gaps and deaf bands of triangular and honeycomb water-steel phononic crystals, J. Appl. Phys., 101 044903, 2007 [11] Yang S, Page J H, Liu Z, Cowan M L, Chan C T and Sheng P, Ultrasound tunneling through 3D phononic crystals, Phy. Rev. Lett. 88 10 104301, 2002 [12] Olsson III R H, Fleming J G, El-Kady I F, Tuck M R and McCormick F B, Micromachined bulk wave acoustic bandgap devices, Tech. Dig. Int. Conf. on Solid-State Sensors, Actuators, and Microsystems (Lyon) 31721, 2007 [13] Olsson III R H, El-Kady I F, Su M F, Tuck M R and Fleming J G, Microfabricated VHF acoustic crystals and Waveguides, Sensors and Actuators A: Physical, vol.145, pp. 87-93, 2008 [14] El-Kady I, Olsson III R H and Fleming J G, Phononic band-gap crystals for RF communications, Appl. Phys. Lett., 92, 233504, 2008 [15] Leseman Z C, Carlson S, and Mackin T J, Experimental measurements of the strain energy release rate for stiction failed microcantilevers using a single cantilever beam peel test, Journal of Microelectromechanical Systems, vol. 16, no. 1, pp. 38-43, 2007 [16] Kirt R. Williams, Kishan Gupta and Matthew Wasilik, Etch Rates for Micromachining Processing—Part II, Journal of Miroelectromechanical Systems, vol. 12, No.6, 2003 [17] P. Hsieh, T. Wu, and J. Sun, Three-Dimensional Phononic Band Gap Calculations Using the FDTD Method and a PC Cluster System, IEEE transactions on Ultrasonics, Ferroelectrics, and Frequency Control, vol. 53, no. 1, 2006
