SUPPLEMENTARY INFORMATION

Transcription

1 SUPPLEMENTARY INFORMATION DOI: /NNANO A Sub-1V Nanoelectromechanical Switching Device Jeong Oen Lee 1, Yong-Ha Song 1,Min-Wu Kim 1,Min-Ho Kang 2,Jae-Sup Oh 2,Hyun-Ho Yang 1,and Jun-Bo Yoon 1 1 Department of Electrical Engineering, Korea Advanced Institute of Science and Technology (KAIST), 291 Daehak-ro, Yuseong-gu, Daejeon, , Republic of Korea 2 Korea National NanoFab Center (NNFC), 291 Daehak-ro, Yuseong-gu, Daejeon, , Republic of Korea jbyoon@ee.kaist.ac.kr A. Theoretical model for the pipe clip beam structure Throughout this paper we have discussed the effects of the pipe clip structure in reducing the operating voltage of the NEM switch. To theoretically investigate whether the pipe clip structure can effectively reduce the operating voltage, we simplified the pipe clip NEM switch into a single-degreeof-freedom lumped-model, adopting the equivalent spring and capacitor structure shown in Figure S1. The spring can be modeled from a fixed-fixed beam approximation with an effective spring constant k S1 when we assume that the complex bending behavior in the bent region is negligible. This approximation seems to be acceptable because the displacement ( g) of the suspended top electrode is far smaller than the total length of the beam, and consequently the bending effect of the bent region would be ignorable in the structure. In contrast, in deriving the capacitor model, the electric field under the bent region of the suspended beam, where the extremely small air-gap is placed, plays a significant role in determining the operating voltage. Since the actual structure of the clamped-clamped pipe clip beam is too complicated to calculate the operating voltage, we proposed a simplified lumped-model of the capacitor adopting the following assumptions: NATURE NANOTECHNOLOGY 1

2 1. The separation distance of the smaller air-gap (g s ) region (the minimum gap between the edge region of the bottom electrode and the bent region of the top electrode) is homogeneous. 2. The direction of the electrostatic force in the small-gap region is purely vertical which is normal to the bottom electrode. 3. The length of the actuating electrode (L f ) is sufficiently larger than the size of the gap (g f or g s ) where the fringe effect of the electrodes is negligible. With these considerations, the proposed lumped-model employs a dual air-gap structure : g s is the size of the smaller gap, which is located on the edge region of the bottom electrode, and g f is the size of the larger gap placed on the flat region of the actuating electrode which occupies almost the entire area of the electrode. Equivalent spring g Simplified lumped model: Top electrode g s Original position g f Bottom electrode L s L f : flat region L s : minimum gap region Bent region Top electrode Pipe clip structure: Bottom electrode Edge region Initial gap thickness g f g s Figure S1. Simplified lumped model for the pipe clip structure. g f is the larger gap between a bottom electrode and a suspended electrode. g s is the minimum gap between the edge region of the bottom electrode and the bent region of the top electrode and g is the displacement of the top electrode induced by an electrostatic force. 2

3 When a voltage (V) is applied between the top and bottom electrodes, the mechanical system is governed by an electrostatic interaction between the electrodes and a restoring force corresponding to the displacement g induced by flexure of the beam structure, which yields the following equations: 1 ϵ 0 W L f V 2 2 (g f g) ϵ 0 W (2L s )V 2 (g s g) 2 = k g (1) V 2 2k g = ϵ 0 W L f (2) (g f + ϵ 0W 2L s g) 2 (g s g) 2 where W is the width of the beam, L s and L f are the length of the minimum gap region and the flat region respectively, and ϵ s is the permittivity of air. Typically, an attractive electrostatic force between two electrodes is quadratically increased as the distance between two electrodes is decreased, while the restoring spring force increases only linearly. Therefore, there is a critical pull-in position in the electrostatic actuator where the electrostatic force is always larger than the restoring force, leading to sudden collapse of the suspended electrode. Owing to the quadratic behavior, it is important to note that the electrostatic force in the extremely small air-gap region dominates the electrostatic force in the flat region. We provided a 3-D finite element simulation result with the theoretically obtained result in Fig. 2 in the main manuscript. 3

4 B. Fabrication process In order for nanoelectromechanical (NEM) devices to be integrated with circuits in low power electronics, their fabrication method should be a CMOS-compatible process. Here we present a fully CMOS-compatible fabrication method for realizing a NEM switch with a pipe clip structure. This structure simultaneously provides an effective pathway to accomplish low voltage operation and a simple fabrication process. The detailed fabrication steps are shown in Figure S2. The fabrication started with a silicon substrate having a thermal oxide layer (10-nm-thick SiO 2 ) on the surface. A low pressure chemical vapor deposition (LPCVD) silicon nitride layer (50-nm-thick SiN) was deposited on top of the thermal oxide. A 20 nm of tungsten (W) layer was then deposited for the bottom electrode by a sputter deposition system because sputtered tungsten provides high resistance to physical wear and an excellent etching selectivity with respect to silicon dioxide (SiO 2 ), which is employed as A. Bottom electrode 20 nm-thick tungsten (W) D. Top electrode 40 nm-thick Titanium Tungsten(TiW ) 50 nm-thick SiN 10 nm-thick SiO 2 B. Insulator etch SiN trench E. Release 50 nm-thick SiN C. Sacrificial layer HDP-CVD SiO2 Top electrode Air-gap Bottom electrode Insulator trench Figure S3. Fabrication process for the NEM switch with a pipe clip beam structure. 4

5 a sacrificial layer. The tungsten bottom electrode was patterned by a scanner lithography tool with the minimum feature size of 180 nm. Using the self-aligned technique, the bottom electrode was overetched to obtain a sloped trench structure. A sacrificial SiO 2 was then deposited with a high density plasma chemical vapor deposition (HDP-CVD) equipment. In order to achieve high yield, the thickness of the sacrificial layer (SiO 2 ) was set to be 50 nm in the first experiment, where the conventional plane structure and the proposed pipe clip structure are compared. In contrast, low operating voltage (sub-1 V) was achieved from the reduced thickness of the sacrificial oxide (i.e. 30 nm). A movable top electrode, which is a clamped-clamped pipe clip structure, was formed with titanium-tungsten (TiW) of 40 nm thickness. Although we fabricated the top electrode with tungsten (W) as well, no significant electrical difference was found compared with titanium-tungsten (TiW). However, it was observed that the suspended beam structure of TiW was relatively less sensitive to fabrication issues including film-stress-induced bending. Thus, the fabrication and measurement data reported here are from NEM switches adopting TiW as the top electrode. As the final step, the sacrificial SiO 2 layer was removed by a buffered oxide etchant (BOE). To avoid a stiction phenomenon S2, we dried the device in critical point drying (CPD) equipment or in liquid isopropyl alcohol (IPA). 5

6 C. Simulation model development for the deposition process The profile of the air-gap structure is highly influenced by the morphology of the spacer structure which is defined by the deposition process of the sacrificial layer. Therefore, it is necessary to identify the key parameters determining the profile of the sacrificial layer from which the optimized fabrication process should be suggested to achieve an extremely small air-gap. Thanks to the advanced CMOScompatible process, we could perform a process simulation first to predict the spacer profile of the air-gap structure prior to the device fabrication. However, since the standard process model provided by the commercial simulator (Silvaco/ATHENA S3 ) that we used did not show sufficient agreement with the actual fabrication results, we modified the process code (ATHENA uses code-based model for the unit process) until the simulation results sufficiently converged to the experimentally obtained results. A transmission electron microscopy (TEM) image of the reference structure obtained from the experiments using the PECVD method is shown in Figure S3a. The bumpy profile(a bread-loafing effect) of the sacrificial layer seen at the top corner of the bottom electrode is consistent with the simulation results shown in Figure S3c. In contrast, the spacer structure obtained from the HDP- CVD deposition method (Figure S3b) offers a smooth edge without the bread-loafing effect. It should be noted that the HDP-CVD process is a special form of the PECVD that employs an inductively coupled plasma (ICP) source, providing a highly conformal profile without bread-loafing at the top corners which is especially effective with respect to attaining void-free trench gap-filling. This ability is attributed to the unique features of the HDP-CVD, which entails a simultaneous deposition and etching process. This key feature was mimicked in the coded-model by iterating sub-units of the deposition and etch codes in a single procedure of the HDP-CVD process. Figure S3d illustrates the simulated profile evolution of the HDP-CVD deposition process, which results in good agreement with the experimentally obtained data. All simulation results in our paper are based on these codedmodels. 6

7 a b Top electrode Spacer structure Bottom electrode PECVD SiO2 HDP-CVD SiO2 c PECVD SiO2 Bottom electrode Insulator Substrate HDP-CVD SiO2 d Bottom electrode Insulator Substrate Figure S3. Development of simulation model for evaluating the effect of deposition process. (a) TEM image of the spacer structure fabricated by PECVD method. (b) TEM image of the spacer structure fabricated by HDP-CVD method. (c) Simulated profile evolution obtained from the modified PECVD code. (d) Simulated profile evolution obtained from the modified HDP-CVD code. 7

8 D. Optimization of the trench profile The profile of the pre-patterned trench (denoted as self-aligned trench in the main text) is a key factor responsible for tuning the spacer structure which defines the air-gap structure. To suggest the optimized spacer structure, simulations for three different trench profiles were performed with variation of the deposition method. Figure S4a presents the spacer structures generated from the PECVD simulation and the change of the slope of the pre-patterned trench. Similarly, the simulation for the HDP-CVD method was performed as shown in Figure S4b. The profile of the sacrificial layer appears to follow the side wall slope of the pre-patterned trench structure. It is important to note that the bread-loafing effect of the PECVD method, mentioned in section C, is exacerbated on the pre-patterned trench structures. Guided by the simulation results, we constructed a spacer structure with a slope profile similar to that found in Figure S4b- slope A, which is a potential candidate to achieve an extremely small air-gap between the top and bottom electrodes adopting the HDP-CVD technique and a deep trench structure. Although the fabricated spacer structure was consistent with the prediction, as shown in Figure S5a, we observed in the fabricated device that the suspended electrode was taken apart after removing the sacrificial layer when the slope of the trench was too steep as shown in Figure S5b. To address this issue, relatively smooth and shallow trench structure profile was introduced with the HDP-CVD deposition, from which the final spacer structure (shown in the main text, Figure 3) could be achieved. This combination of the self-aligned trench technique, HDP- CVD deposition, and controlled side wall profile is a unique approach to provide a nano-sized air-gap structure with an extremely small air gap at the edge region. We believe that the proposed fabrication technique has strong potential to be applied for various nanotechnologies with high controllability and reproducibility. 8

9 a PECVD method Sacrificial layer Bottom electrode Slope A Slope B Slope C b HDP-CVD method Slope A Slope B Slope C Figure S4. Optimization of the spacer structure. (a) Spacer structures simulated by PECVD method with variation of the slope of the trench side wall (A,B,C). (b) Spacer structures simulated by HDP-CVD method. a b SiO 2 W SiN Broken link 40nm 200nm Figure S5. Fabricated device with a steep trench structure. (a) TEM image of the spacer structure fabricated with a deep and steep trench structure. (b) SEM image of the fabricated device using the spacer structure shown in (a). 9

10 E. Electrical measurement An Agilent 4156C semiconductor parameter analyzer was used for the electrical measurements. All measurements were performed in a vacuum chamber (about 100 mtorr) equipped with an electrical feed-through to prevent environmental interference including moisture. The OFF-state current characteristic was better in the vacuum chamber, but no significant change in the switching characteristic was observed. Typical current vs. voltage (I-V) curves were measured for each device. Figure S6 exhibits representative data sets measured from the fabricated NEM switches (see Figure 2 of the main text). The NEM switch with the plane structure (Figure S6a) shows a relatively high operating voltage. For example, the device with a beam length of 3 µm showed operating voltage of 10 V. a b L c L c Figure S6. I-V characteristic of plane structure and pipe clip structure. (a) schematic illustration of the plane structure and its electrical behavior. (b) schematic illustration of the pipe clip structure and its electrical behavior. The width of the suspended beam and the air gap were 300 nm and 50 nm, respectively. 10

11 Typically, a mechanical motion of the NEM switch causes a sudden change of the conduction state, and thus unavoidable high discharge current induces severe damage on the device. Hence, a current compliance of 100 na was applied throughout the measurement to prevent localized Joule heating S4 and high current related problems such as melting and fuse failure. However, in the real measurement of the NEM switch with the plane structure, upon operation no device returned to its initial state, and the electrodes were permanently fused together and lost their switching characteristics. Furthermore, the plane structure sample, with length of 1500 nm and operated at 40 V, showed an unstable ONstate current and high resistance such that the current could not reach the compliance limitation (100 na). Once the device was damaged by high voltage, the high resistance and unstable ON-sate current sustained and the device could not be returned to the normal state. In contrast, the devices with the pipe clip structure showed similar I-V characteristics to those of the plane structure, but their operating voltage was reduced as predicted (Figure S6b). In Figure 4 of the main text, we report the repeatable switching characteristic from sub-1v operation, while this characteristic could not achieved in the device with the high operating voltage. In this respect, we suggest the possibility that the low voltage suppressed the electrical damage or the high discharge current at the contact spot, resulting in improvement of the contact reliability. F. Sub-pull-in current behavior If we attempt to evaluate a possible current flow caused by the tunneling or field-emission effect in the sub-pull-in region, a parasitic leakage component that may disturb the observation of the tunneling effect should be investigated first. Therefore, we performed an isolation experiment to find out the leakage current components, such as a substrate (or an equipment) leakage current. In Figure S7, the leakage current component in the sub-pull-in region was examined. A solid red line with a solid red circle indicates the current measured from a probing pad pattern without the device structure (upper right SEM image in Figure S7). On the other hand, solid lines with triangles (blue and black) indicate 11

12 Figure S7. Measured I-V curves from a probing pad and a device pattern. An electrical measurement was performed on the various patterns. A red line with a solid circle indicates the I-V curve measured from the probing pad (upper right SEM image). The I-V curve measured from the device pattern (lower right SEM image) presents electromechanical behavior of the nano-gap structure, showing the identical I-V characteristic with the probing pad in the sub-pull-in region. the I-V characteristic measured from the device pattern (lower right SEM image in Figure S7). The probing pad and the device pattern were simultaneously fabricated on the same substrate. A black line with a diamond indicates the noise floor of the measurement equipment when the probes are up in the air, which shows a current level about several hundreds of fa. We observed a pico-ampere (pa) level of the current flow from the probing pad pattern. Therefore, the measured current from the probing pad pattern indicates a parasitic leakage component that is not associated with the nano-gap structure. It is important to note that the current curve of the device pattern is identical with that of the probing pad pattern in the sub-pull-in region (when the switch is turned off). It also implies that the sub-pull-in current of the device pattern is independent on the presence of the nano-gap structure, thus this current component would be induced from a surface leakage current that flows between the probing pads or pad-to-substrate. 12

13 G. Transition current behavior Since a high electric field is involved in the proposed structure, there may be a field emission or tunneling current occurred when the switching current changes from OFF to ON state. In order to verify the field emission effect, if any, in the transition region of the I-V curve obtained from the fabricated device, a Fowler-Nordheim (F-N) plot analysis was performed. Fig. S8(a) shows the I-V curve shown in Fig. 2c in the manuscript. Fig. S8(b) shows a F-N plot (ln(i/v 2 ) against 1/V) corresponding to the measured I-V data shown in the dashed region of Fig. S8(a). It is observed that the value of ln(i/v 2 ) decreases linearly as a function of 1/V with a slope S of (extracted from the linear fit). We modeled the contacting region with a metalinsulator(air)-metal (MIM) structure as shown in the inset of Fig. S8(b). According to the F-N model, the relationship between the emission current (I) and the applied electric potential (V) can be written Figure S8. F-N plot analysis (a) The measured I-V curves from the NEM switches. (b) A Fowler- Nordheim plot (ln(i/v 2 ) vs. 1/V) obtained from the data points in the dashed region. Red circles correspond to the measured data and a solid line corresponds to the linear fit through the data where an extracted slope parameter is

14 as I = Aaβ 2 E 2 exp ( ) bψ 3/2 0 βe ( = Aa β2 V 2 exp t 2 4 ) 2m (qψ 0 ) 3/2 t 3q βv (3) Where I is the tunneling current, A is an emission area, a and b are constants (a = AV 2 ev, b = V m 1 ev 3/2 ), t is a distance of the air gap separation (10 nm), V is the applied voltage and ψ is the work function of metal electrode (tungsten, 4.5 ev). β is the enhancement factor of electric field, which is usually evaluated from measuring the slope of the F-N plot. A relationship between the slope parameter S and the enhancement factor β can be expressed as S = dln(i/v 2 ) d(1/v ) = 4 2m (qψ 0 ) 3/2 t 3q β (4) The field enhancement factor (β) of 0.51 were obtained by substituting the measured slope parameter S in equation (4). Now substituting the extracted parameters into the equation (3), the field emission model for the simplified band structure and its I-V characteristic can be obtained, in which the model has the same slope parameter S as one from the measured data. The F-N model which is obtained by above method is named P 10nm and the parameters are shown in the table of Fig. S9(a). The overall trend of the predicted I-V curve P 10nm (black line with solid squares) is far different from the measured curve M (red line), exhibiting significantly low current level at the given applied voltage. It implies that the assumed MIM tunneling structure cannot provide a suitable current corresponding to the measured one, even if the parameter slope of P 10nm (S 10nm =-1279) can be consistent with the measured one from the measured curve M (S M =-1288). It also implies that, when the device is in the initial state with the 10 nm air gap separation, the tunneling current in the sub-pull-in region is insignificant because high voltage (over 40V) is required to observe the measurable tunneling effect. 14

15 If we assume the thickness of the air gap is reduced to 3 nm or 2 nm, we can obtain curves P 3nm andp 2nm respectively. The curves P 3nm and P 2nm show relatively similar current level with the I-V behavior from the measured curve (M). The curve P 2nm implies that the air gap should be reduced to at least 2 nm to observe the tunneling current in the given bias voltage region. However, the predicted slope parameter (S 3nm = -195 ands 2nm = -130) is still far lower than the measured slope S M. Consequently, the F-N analysis indicates a rare possibility of the tunneling effect in the sub-pull-in region with the given bias voltage. We also verified that the stiff slope from the measured I-V curve cannot be explained by the field emission or tunneling phenomenon without considering the changes in the air gap thickness. Figure S9. A MIM band structure and the predicted I-V curves. (a) An energy-band diagram of the metal-insulator-metal structure corresponding to the contact region of the NEM switch for analyzing the tunneling current that occurs before making a contact. (b)p 10nm is the theoretically obtained curve with the parameters shown in (a).p 3nm and P 2nm are obtained by assuming the air gap thickness of 3nm and 2nm respectively. 15

16 References S1. Pamidighantam, S. et al. Pull-in voltage analysis of electrostatically actuated beam structures with fixed-fixed and fixed-free end conditions. J. Micromech. Microeng (2002). S2. Tas, N. et al. Stiction in surface micromachining. J. Micromech. Microeng (1996). S3. http : // s imulation/athena.html S4. Dong. L. et al. Effects of local Joule heating on the reduction of contact resistance between carbon nanotubes and metal electrodes. J. Appl. Phys. 101, (2007). 16