NANO LETTERS
Electromechanical Carbon Nanotube Switches for High-Frequency Applications
2006 Vol. 6, No. 5 942-947
Anupama B. Kaul,* Eric W. Wong, Larry Epp, and Brian D. Hunt Jet Propulsion Laboratory, California Institute of Technology, Pasadena, California 91109 Received December 27, 2005; Revised Manuscript Received March 2, 2006
ABSTRACT We describe the fabrication and characterization of a nanoelectromechanical (NEM) switch based on carbon nanotubes. Our NEM structure consists of single-walled nanotubes (SWNTs) suspended over shallow trenches in a SiO2 layer, with a Nb pull electrode beneath. The nanotube growth is done on-chip using a patterned Fe catalyst and a methane chemical vapor deposition (CVD) process at 850 °C. Electrical measurements of these devices show well-defined ON and OFF states as a dc bias up to a few volts is applied between the CNT and the Nb pull electrode. The CNT switches were measured to have speeds that are 3 orders of magnitude higher than MEMS-based electrostatically driven switches, with switching times down to a few nanoseconds, while at the same time requiring pull voltages less than 5 V.
The remarkable mechanical and electrical properties of carbon nanotubes (CNTs) make them excellent candidates for the design of nanoelectromechanical systems (NEMS). They exhibit an extremely high directional stiffness,1 accommodate very large mechanical strains,2 and depending on diameter and helicity, act as 1D metals or semiconductors.3 Because of these unique properties, coupled with their low mass and chemical inertness, CNTs represent ideal materials for enabling many high-performance NEMS devices. Nanotube-based NEMS have been demonstrated already in applications involving nanotweezers,4 memory devices,5 supersensitive sensors,6 and tunable oscillators.7 Nanorelays8,9 are another promising application of nanotubes that offer the potential for high-performance switching, with high-speed operation at low actuation voltages and power. Although theoretical studies on CNT switches have been reported,10 results from fabricated devices that characterize nanoscale electromechanical switching have been limited. Electromechanical switching in CNTs was first observed by Rueckes5 et al. where single-walled nanotubes (SWNTs) were mechanically manipulated to form crossed nanotubes with an air gap. Lee11 et al. have demonstrated switching in deposited multiwalled nanotube (MWNT) cantilever structures, which were fabricated using an AC electrophoresis technique.12 Cha13 et al. have also observed switching in devices using deposited MWNTs, where the individual tubes were located by SEM for subsequent e-beam and thin-film * E-mail:
[email protected]; tel: (818) 393-7186. 10.1021/nl052552r CCC: $33.50 Published on Web 04/05/2006
© 2006 American Chemical Society
processing. Most recently, Dujardin14 et al. demonstrated switching in deposited MWNTs cantilever devices using a technique that allows the air gap to be controlled to within 1 nm precision. To date, switching in both SWNTs5 and MWNTs11,13,14 has been reported for the case of deposited tubes. Here we demonstrate switching results for SWNT devices, where the tubes are grown on-chip with patterned catalysts using materials that are compatible with the hightemperature CVD synthesis of SWNTs. The tubes bridge prefabricated trenches on a Si wafer over a refractory metal electrode. Important for assessing the potential of CNTs for high-speed applications, we also present switching speed measurements in these devices. Nanotubes have previously been grown15,16 or deposited17 across trenches on a Si wafer. In these studies, the AFM tip has been used to mechanically strain the suspended tubes to understand how structural deformation affects transport properties15,16 and to study the elastic strain in SWNT ropes.17 Unlike these structures where the tubes were suspended directly over SiO2/Si substrates, we focus on device applications of structures where a refractory metal electrode made from niobium (Nb) is placed directly beneath the suspended tube. This serves as the pull electrode for transiently charging the tubes to induce electromechanical switching. A schematic of our device is shown in Figure 1 where the various multilayers are depicted. Our starting substrate was a thermally oxidized 〈100〉 Si wafer, on which a dc magnetron sputtered Nb film was deposited. We chose Nb as a pull electrode material because of its refractory nature
Figure 1. Schematic shows a nanotube switch composed of (a) a 200 nm Nb film deposited using dc magnetron sputtering onto a thermally oxidized Si wafer followed by (b) a 200 nm PECVD SiO2 layer. (c) Active device regions are etched in the PECVD SiO2 layer to a thickness of ∼20 nm, the height of the air gap for the switch. (d) Nanotrenches are defined using e-beam and etched to the Nb layer. (e) SWNTs are grown from patterned Fe in a CVD furnace using CH4/H2 (1500 sccm/50 sccm) at 850 °C. (f) Electrodes are deposited using e-beam evaporated Au/Ti (220 nm/12 nm), which are patterned using a bilayer resist/PMMA process for easy lift-off.
so that it would be chemically and structurally stable at the high growth temperatures (∼850 °C) required for the CVD synthesis of SWNTs. Although other refractory metals could also be considered, some of these materials, such as Ti and Ta, form volatile metal hydrides at high temperatures in a hydrogen ambient, and are preferentially etched.18 Niobium was sputter deposited to a thickness of ∼200 nm. A ∼200 nm layer of PECVD SiO2 was then deposited and served as the intermetal dielectric. In the first masking layer, active device regions in the PECVD SiO2 were patterned to thin down the oxide from 200 nm to ∼20 nm, which was comparable to the air gap of the switch. Nanotrenches that were as small as 130 nm were then defined using e-beam lithography in these regions. The nanotrenches were patterned using CF4/O2 RIE to etch the PECVD SiO2 down to the Nb layer. For nanotube growth, the catalyst was patterned by photolithography and liftoff of 0.5-nm-thick Fe, which was deposited by e-beam evaporation. The sample, with patterned Fe, was then placed in the CVD furnace for nanotube growth at 850 °C for 10 min. using CH4 and H2 at flow rates of 1500 and 50 sccm, respectively, where SWNT growth predominates.19 Prior characterization has revealed tubes with diameters typically between 1 and 3 nm, with TEM studies indicating most of these to be SWNTs.20 After CNT growth, the metal electrodes (Au/Ti) were patterned to contact the CNTs, using a bilayer AZ 5214/ PMMA lift-off process, which results in easy lift-off of metal films because of an undercut in the PMMA layer. The top electrode metals Au/Ti were then deposited in an e-beam evaporator and lifted off in acetone. An SEM micrograph of a finished device is shown in Figure 2a, where the underlying Nb pull electrode in the nanotrench is contacted by the Au/Ti electrodes labeled “pull”. Also shown are the source and drain electrodes, which contact the CNT that bridges the trench. The high-magnification SEM image of Figure 2b depicts a nanotube crossing the trench. An AFM image is shown in Figure 2c, which is Nano Lett., Vol. 6, No. 5, 2006
a cross section of a wide-area 1 µm trench. The nominal trench depth was measured to be ∼20 nm, but because of surface roughness the uncertainty in trench depth was large. Electrical measurements were performed using an HP 4156C parameter analyzer connected to a dc probe station. The conductance between the source and drain electrodes of a typical device (Figure 2d) yields a resistance of ∼200 kΩ, much of which is attributed to contact resistance between the CNT and Au/Ti films. Source-to-drain resistances typically ranged in value from tens of kiloohms to tens of megaohms, where the presence of multiple tubes can also contribute to the differences in resistance. The actuation voltages were measured by applying a dc voltage between the source and pull electrodes. As transient charge develops on the tube with increasing bias voltage, the resulting electrostatic force is sufficient to overcome the elastostatic force and deflects the suspended tube down toward the pull electrode. The current was measured as a function of the dc bias voltage between the source and pull electrodes. Shown in Figure 3a is an I-V characteristic for the device illustrated in Figure 2b. For voltages up to 2.0 V, the currents are very low, a few pA. Then from ∼3.5 V, the current begins to rise rapidly to ∼250 nA at ∼4.5 V. This switching between the low- and high-current states represents more than a ∼4 order of magnitude increase, implying welldefined OFF and ON states, respectively. The currents measured are believed to originate from a tunneling mechanism; as the tube deflects closer to the bottom electrode with increasing bias voltage, the magnitude of the tunnel current increases exponentially. In the rapidly rising current regime, the data is increasingly noisy, reflecting the stochastic nature of the tunneling mechanism. Hysteresis between the increasing and decreasing bias voltage paths was also evident, as illustrated in Figure 3a, and arises from the interaction of the tube with surface van der Waals forces. Lateral leakage currents within the dielectric were extremely small, ∼10 pA at (5 V, as indicated by currents in the absence of tubes. Shown in Figure 3b is an I-V characteristic of another device that was actuated over several cycles. Turn-on occurs at ∼2.4 V in this case, with a slight variation with cycling that is also reported in other CNT11 and MEMS switches. The rapidly rising current regime arises in both the forwardbiased (pull electrode grounded) and reverse-biased (pull electrode positive) cases, as indicated by the inset of Figure 3b, although the exact switching voltages are slightly different in the two cases, ∼2.4 V (forward-biased) and ∼2.2 V (reverse-biased). The differences in turn-on voltage can perhaps arise from the random distribution of metallic and semiconducting tubes observed in current SWNT growth processes, but still suggest that the differences in resistance between the ON and OFF states far outweigh any differences that may arise from contact resistances. As shown by the inset of Figure 3b, this switching behavior is polarityindependent, as would be expected for electrostatic actuation, and rules out field emission as a likely mechanism at these voltages. In general, the magnitude of the switching voltages in these air-bridge devices was a few volts, which is smaller by at 943
Figure 2. (a) Low-magnification SEM image (JEOL 6700) of a CNT switch indicates the various electrodes. (b) High-magnification SEM micrograph shows a single nanotube bridging the 130-nm-wide trench. (c) A detailed view of a 1-µm-wide trench is depicted by the AFM height image (DI Nanoscope III). The nominal air gap was ∼20 nm with a large uncertainty in trench height due surface roughness mainly in the PECVD SiO2 layer. (d) Source-drain conduction measurements for a typical device gives R ≈ 200 kΩ (trench width ∼250 nm, 2 µm × 5 µm catalyst area).
least an order of magnitude compared to actuation voltages typically observed in MEMS switches.21 In cantilever CNT devices,11 the turn-on voltages were also somewhat higher in the 6-20 V range. The differences in device geometries, such as larger air gaps (∼80 nm) and the use of MWNTs, may be sufficient to explain the larger turn-on voltages required in that case. However, Dujardin14 et al. report low switching voltages of 2.8-3.0 V in their MWNT cantilever structures, a result that can be attributed to the very shallow (4 nm) air gaps. The I-V characteristic in Figure 3c is a representative example of a device in which stiction was observed. In this case, the current rises rapidly at 2.5 V and saturates at the instrumentation compliance, which was set to ∼20 µA. On the decreasing-voltage path, an ohmic resistance was observed, which was typically in the range of a few kiloohms to hundreds of kiloohms. The ohmic behavior persisted upon subsequent cycling and the device appeared stuck. Dujardin14 et al. have noted that a decanethiol self-assembled monolayer 944
(SAM) coating over the Au electrode that the CNT contacts sufficiently prevents stiction in their devices. The operation and design of NEM switches resembles MEM switches, and pull-in voltages can be calculated using continuum beam mechanics.8 When the contribution from van der Waals forces is ignored, the pull-in voltage, VPI, to first order, is calculated using
VPI )
x
8kg3 27owL
(1)
where g is the air gap, o is the effective permittivity, w is the beam width, and L is the length. The spring constant, k, for a doubly clamped beam is given by k)
384EI L3
(2) Nano Lett., Vol. 6, No. 5, 2006
Figure 3. (a) Actuation voltage measurements of the device shown in Figure 2b. From the I-V characteristic, the 4-orders-of-magnitude rise in current is seen at ∼4.5 V, indicating well-defined ON and OFF states. Hyteresis is also seen between the increasing and decreasing voltage paths. (b) I-V characteristic of a device actuated over multiple cycles (250 nm trench width). The inset shows the ON state voltage to be similar in the forward-bias (pull electrode grounded) and reverse-bias (pull electrode positive) regimes, indicating that field emission is an unlikely possibility at these voltages. (c) Example of a device where stiction was observed. In the rising voltage path, the current rises rapidly at ∼2.5 V to the instrumentation compliance current (20 µA); and on the decreasing voltage cycle, ohmic behavior is seen and persists with subsequent cycling and the device appears stuck. (d) Calculation of actuation voltage as a function of tube length at air gaps ranging from 10 to 40 nm with assumed E ≈ 1 TPa and tube diameter do ≈ 2 nm. The measured pull voltages are mapped onto these data, where the tube length was assumed equal to the trench width.
where E is the elastic modulus, and I is the moment of inertia given by I)
(( ) ( ) )
Di π Do 4 4 2 2
4
(3)
Here Do and Di are the nanotube outer and inner diameter, respectively. In the case of SWNTs if we assume that Di ) 0, then ISWNT > I, yielding kSWNT > k. Hence, if the pull-in voltage for a SWNT device, VPI(SWNT), is calculated, then this pull-in voltage will set an upper bound for the actuation voltage in the case where Di is nonzero. Using eqs 1-3, we have calculated VPI(SWNT) as a function of beam length at air gaps ranging in value from 10 to 40 nm, similar to the range of air gaps for our fabricated devices when surface roughness is also considered. The results are plotted in Figure 3d, where we have assumed that E ≈ 1 TPa and Do ≈ 2 nm. Nano Lett., Vol. 6, No. 5, 2006
The measured voltage data obtained for our fabricated airbridge devices was mapped onto the plot, where the length of the suspended tube was assumed to be the same as the designed trench width. To first order, the pull voltages seem to fall in the window of expected voltages given the uncertainty in all of the parameters involved, such as the actual air gap, elastic modulus, and tube diameter in each device. In addition, because of the random orientation of tubes during growth, the length of the suspended tube is not necessarily equal to the trench width, although recent progress in the electric-field-directed growth of SWNTs appears to result in more controllable nanotube architectures.22 We have also performed switch speed measurements on our CNT air-bridge devices. The measurement setup is shown in Figure 4a. A step function was applied to the device using a pulse generator (Agilent 81101A), and an output voltage 945
Figure 4. Switching speed measurements. (a) Measurement setup. (b) Prescreening measurement that shows the I-V characteristic of a candidate device for the speed measurement. (c) Switching is observed at 5 V indicated by the “switch output” waveform. The delay from the instrumentation (cable lengths, stray capacitances, and inductances both on- and off-chip) was measured and is represented by the “calibration output” waveform. The difference in time required to reach the maximum voltage between the two waveforms results in a switching time of ∼2.8 ns attributed to the device, as indicated.
was measured across a sense resistor (R ≈ 110 kΩ) that was connected in series to ground. Both the input and output pulses were displayed synchronously on two channels of a digital oscilloscope (Tektronix TDS 3054), from which the delay times were determined. A prescreening measurement was first done to select a candidate device for the switching speed measurement. An I-V characteristic of such a candidate device is shown in Figure 4b, where the currents begin to rise rapidly at ∼2 V. To prevent stiction in candidate devices selected for the speed measurement, the voltages used 946
in the prescreening measurement were limited to the point where the first onset of switching was observed. Speed measurements were performed for the device shown in Figure 4b. The voltage applied by the pulse generator was incrementally increased by 200 mV intervals. Up to 4.8 V, no output pulse was detected. Then at ∼5 V, the output voltage increased rapidly as shown in Figure 4c (“switch output” waveform). The “calibration output” waveform refers to the case where the device was removed and the probes were placed on a metal strip on-chip to measure the intrinsic delay associated with the instrumentation, such as that arising from cable lengths, stray capacitances, and inductances both on- and off-chip. The difference in time required to reach the maximum voltage between the two waveforms was determined to be ∼2.8 ns, as indicated in Figure 4c, yielding an upper bound of the intrinsic switching speed of the CNT switch. The oscillations in the calibration and switch waveforms likely arise from resonances associated with the leadline inductances and the capacitances from the instrumentation; we do not have sufficient information available to calculate the nature of the damping mechanisms involved in these cases. In the switching example of Figure 4c, the maximum voltage was attained in one switching event. When some other devices were measured, switching occurred in more than one stage, suggesting the presence of multiple tubes that switch at different voltages. Compared to state-of-the-art MEMS devices, the switching times of our CNT switches are several orders of magnitude smaller. In general, for switches that rely on electromechanical actuation, the switching time is composed of the response time, which is the time required to overcome mechanical inertia, as well as the rise time of the voltage pulse due to charging capacitances. In surface-mount relays, the response time is in the millisecond range and dominates switching speed. While device dimensions and mass in MEMS switches are much smaller, the response time is still a significant fraction of the total switching time. For example, Peroulis21 et al. reported the total switching time for Si MEMS devices to be 52 µs, for which the response time alone was ∼30 µs. The ultralow mass, exceptionally high spring constant, and extremely low capacitance of the CNT all contribute to the small response and rise times in the CNT switch, which lead to the extremely small total switching times, as observed. Jonsson23 et al. provide a detailed theoretical analysis of the switching dynamics in CNT switches with predicted times in the nanosecond range and also discuss the effect of surface dissipative forces. The fastest MEMS switch was developed at MIT Lincoln Labs and is reported to have a switching time of 1 µs; this was realized by decreasing device dimensions, but there is a concomitant increase in the voltage, with 60-70 V needed for actuation.24 These voltages are difficult to obtain in applications where low-voltage power supplies are used, such as hand-held mobile phones and other wireless applications, as well automotive vehicles. The CNT air-bridge switch has the unique advantage of low actuation voltage,
