Reduction in the adhesion energy of contacting metal electrode surfaces in nano-electro-mechanical switches is crucial for operation with low hysteresis voltage. We demonstrate that by forming thin layers of metal-oxides on metals such as Ru and W, the adhesion energy can be reduced by up to a factor of ten. We employ a low-energy ion-beam synthesis technique and subsequent thermal annealing to form very thin layers (∼2 nm) of metal-oxides (such as RuO2 and WOx) on Ru and W metal surfaces and quantify the adhesion energy using an atomic force microscope with microspherical tips.
Nano-electro-mechanical (NEM) switches (relays) have attracted significant attention in recent years due to their potential to overcome the energy-efficiency limit of complementary metal-oxide semiconductor (CMOS) digital logic circuits and technology.1–4 Since relays operate with zero off-state leakage current and abrupt switching characteristics, overcoming the limitation of 60 mV/decade room temperature subthreshold swing for CMOS transistors, they can be used to dramatically lower the energy consumption of digital electronic circuits.3,4 Several theoretical analyses and experimental works have been performed to design and demonstrate relays driven by piezoelectric,2 electrostatic,5 or magnetic forces6 and to achieve lower operating voltage.7 As all relays operate on the basic principle of formation and breaking of mechanical contacts, contact adhesion energy plays an important role in determining the performance and energy efficiency of a relay.8,9 Studies also have been performed to understand the effects of contact area scaling and the dynamics of contact opening/closing.9–11 A general understanding gained from such analyses7,11 is that the minimum energy needed to switch a relay is proportional to the contact adhesive force (Fad.). Therefore, reduction of the adhesion energy (EAd) is necessary to improve the energy efficiency of a relay.
Adhesion is dependent on many parameters, including the contacting electrode material(s). Process limitations as well as device performance requirements impose additional constraints on the choice of the material(s). For high endurance and good performance, we have used hard metals with high electrical conductivity, such as W and Ru, as contacting electrode materials in logic relays.7 With a motivation to reduce the adhesion energy per unit area without dramatically reducing the relay on-state conductance, we show herein that ultra-thin metal-oxides (such as RuO2 and WOx) on Ru and W metal surfaces can significantly decrease adhesion energy, by up to an order of magnitude.
The Ru and W thin films (approximately 30 nm in thickness) were deposited onto 100 nm of oxidized silicon wafer substrates at room temperature, using the reactive rf-sputtering technique inside a high vacuum chamber (3 × 10−6 Torr). To promote adhesion, a thin (1.5 nm thick) layer of TiO2 was deposited onto the SiO2 surface by atomic layer deposition (ALD) prior to Ru deposition. No adhesive layer was used for W, since it adheres well to SiO2. To form oxide on a metal surface, low-energy oxygen-ion implantation followed by thermal annealing at 450 °C for 30 min in forming gas (95% N2 and 5% H2) was employed. The ion implantations were performed with 7° tilt angle and acceleration energies of 3 keV for Ru and 3.5 keV for W samples, respectively. These energies were selected based on Stopping and Range of Ions in Matter (SRIM) simulations,12 to achieve a projected implant range of ∼2–3 nm. (See Fig. S1 in the supplementary material for SRIM-simulated range vs. energy, and an example of oxygen depth distribution in Ru. It should be noted that for such low acceleration energies, the longitudinal and lateral straggles are comparable in magnitude to the projected range.) Two values of O2 implant dose were investigated for each metal: 4.5 × 1015 cm−2 and 7.3 × 1015 cm−2 for Ru; 4.2 × 1015 cm−2 and 7.0 × 1015 cm−2 for W. As an additional experimental split condition, the substrate temperature was either cooled or not cooled during implantation. Atomic force microscopy (AFM) was used to measure adhesive force between contacting materials (metal/metal-oxide and AFM tip) in the force spectroscopic mode.13 The measured adhesive force and the RMS surface roughness were used to calculate the adhesion energy. Material properties of the oxides, such as their composition and thicknesses, were determined by X-ray photoelectron spectroscopy (XPS).
Microspherical SiO2 tips (inset in Fig. 1(a)) with 2 tip diameter were used for the AFM measurements. The bare SiO2 tips were later coated with Ru or W using the same recipe as used to deposit these metals onto oxidized silicon wafer substrates, for measurements of adhesive force between related materials (for example, Ru-Ru contact or RuO2-Ru contact). The relatively large tip diameter provides for consistent measurements of contact adhesive force, in contrast to more conventional sharp AFM tips with smaller diameters (less than 10 nm) that are used for tapping-mode measurements of surface topology. Using a small-diameter tip, we observed significant variations in measured adhesive force from area to area, and even within the same area of the sample. (See Fig. S2 in the supplementary material for details.)
A typical force-vs.-displacement plot obtained with a microspherical SiO2 tip and Ru film is presented in Fig. 1(a), wherein the trace and retrace curves correspond to the approach and retraction of the tip to/from the Ru surface, respectively. For each of the adhesion measurements, the maximum applied force was 100 nN. The rationale for this is discussed below. The adhesive force was determined from the retrace curve as shown in Fig. 1(a). To check for reproducibility, 10 measurements were performed at each of 5 randomly chosen spots on each sample. The results for Ru(film)-SiO2(tip) contact and W(film)-SiO2(tip) contact are presented in Fig. 1(b). For Ru, the measurements are very consistent at 386 nN on average, with a standard deviation of 5 nN. For W, there is a greater variation, with a standard deviation of 16 nN for an average adhesive force of 316 nN. This is attributed to the larger RMS surface roughness of the W film (0.45 nm) as compared against the Ru film (0.21 nm), as deduced from AFM measurements (e.g., Fig. 1(d)), since larger surface roughness results in larger differences in the contact environment from measurement to measurement. Nevertheless, the variation in measured values of adhesive force is relatively small (less than 5%), for all of the samples studied in this work.
Adhesive force between materials has several origins including (1) van der Waals force, (2) chemical interaction forces due to bonding between the atoms of the materials, (3) electrostatic interactions, and (4) capillary force due to external contaminants such as water and polymers on the sample surfaces.14 While the first three are intrinsic, the fourth can be mitigated with proper care. Hence the adhesion measurements were performed in a dry-N2 environment. (Prior to measurement, the sample was in situ heated to 140 °C for several hours and then cooled to room temperature under dry-N2 flow.) The capillary force was found to contribute less than 12% to the total adhesive force (see Fig. S3 in the supplementary material).
The adhesion energy per unit area, which is an absolute gauge of the adhesion property of a material, was calculated using the Derjaguin-Muller-Toporov (DMT) continuum mechanics model.14,15 The effect of surface roughness was accounted for using the modified-Rumpf model16,17 since it is best suited for solid materials with surface roughness values less than 2 nm.17 In this model, the roughness is approximated as a single hemispherical asperity on a flat surface; the radius of the asperity depends only on the RMS roughness of the surface.15,17 The equation combining the DMT and modified-Rumpf model is13
where WAd. is the adhesion energy per unit area, FAd. is the adhesive force determined from AFM measurements, Rtip is the radius of the AFM tip, Ratomic and Rq are the RMS roughness values of the sample surface and the tip, respectively, and z0 is the equilibrium separation distance between the tip and the sample when they are in contact. A value of 0.32 nm was used for z0 for all contact cases in this work, which represents the c-axis lattice parameter18 of RuO2. For the bare SiO2 tip, a roughness value of 0.3 nm was used. A sensitivity test of the SiO2 RMS roughness was performed, which showed that the percentage adhesion energy reduction by oxidation of the metal is insensitive to the tip roughness, as expected. However, the adhesion energy values of the individual sample-tip contacts depend on the tip roughness value assumed. For the Ru- or W-coated tip, the roughness value of the metal thin film was used. The RMS roughness of the implanted and annealed RuO2 sample was measured to be 0.55 nm, while the WOx film has an RMS roughness of 1.51 nm.
Calculated values of adhesion energy per unit area for contacts between metal or metal-oxide thin films and AFM tips (bare SiO2, or coated with Ru or W) are presented in Fig. 2. The adhesion energy between Ru and SiO2 is 92 mJ/m2, whereas the adhesion energy between W and SiO2 is about 3× lower, at 30 mJ/m2. Fig. 2 also shows that the adhesion energies are ∼3× and ∼5× smaller for RuO2 and WOx in contact with SiO2 (27 mJ/m2 for RuO2-SiO2 and 5 mJ/m2 for WOx-SiO2). Clearly, oxide formation on metal surfaces reduces adhesion energy. The results for RuOx samples formed with different implantation conditions are very similar (within the error ranges—see Fig. S4 in the supplementary material). This is not altogether surprising, since the samples were annealed together after O2 implantation so that the chemical compositions of the oxides formed on the surfaces are expected to be very similar. Moreover, the RMS roughness values of the samples were found to be very similar. The adhesion energies for Ru-Ru and W-W contacts (i.e., deduced from measurements using an AFM tip coated with metal) are 27 mJ/m2 and 9 mJ/m2, respectively. Oxidation of the metal thin film surface reduces these to very low values of 11 mJ/m2 and 1 mJ/m2, respectively, similarly as for SiO2 contacts. While the measurements and calculated results clearly show a decrease in adhesion energy of oxidized vs. pristine surfaces, one should not compare measurements made with different tips (e.g., SiO2 tip versus Ru tip), as deposition of coating material on the tip will change the tip properties and alter the nature of tip-sample interaction, and hence influence the obtained adhesion values.
Reduction in adhesion energy due to oxide formation on a metal surface could be due to several factors. We believe that the reduction of chemical interaction forces plays an important role in the reduction mechanism. In addition, van der Waals forces and Casimir forces could also be reduced as a result of oxide formation and the resultant lower electrical conductivity of the surface. The adhesion energy per unit area of an interface between two materials A and B is given by the following equation:19
where and are the surface energies of the materials A and B, respectively, and is the energy of the interface between materials A and B. Most metals have high surface energy and low interface energy due to the presence of dangling bonds on the surface. Electrostatic attraction between conduction electrons of the contacting metals makes metal-metal bonds easy to form across the interface.20 A phenomenological expression for the interaction potential of two similar metallic surfaces is21
where is a characteristic decay length of the metal material, D is the distance between the metal surfaces, and D0 is the equilibrium distance. Under the equilibrium condition, the adhesion energy is W(D) = . As a result, metal contacts generally have large adhesion energies. In contrast, oxides (like most ceramics) usually have nearly saturated surface bonds so that they have low surface energies. Also, oxide-oxide bonds are more covalent in character and are more difficult to form than metallic bonds. Therefore, oxide contacts typically have lower adhesion energies compared to metal contacts. Given this general description, it is not surprising that the formation of oxides on metal surfaces was found in this work to considerably reduce adhesion energies.
To determine the stoichiometry and thicknesses of the ion-beam synthesized oxides on the metal surfaces, we performed detailed XPS analyses to compare the O and Ru depth profiles in a pristine Ru sample vs. an implanted and annealed sample. The results presented in Fig. 3 indicate that the pristine-Ru sample is very pure, with negligible native oxide on the surface; this is consistent with our previous experiments demonstrating that native RuOx formation is a self-limiting process.22 The results for the implanted and annealed sample indicate an oxide thickness of ∼2 nm with a stoichiometry of RuO2 at the surface, confirming the accuracy of the SRIM simulations.
Using the DMT model,14,15 we have also estimated the effects of other parameters such as the contact radius (a), sample deformation (), and contact pressure (P) underneath the tip during measurement. The contact radius (a) for the AFM tip is given by the following expression:15
where Etotal is the reduced elastic modulus, which is a combined elastic modulus that considers both the tip material and the sample material through the relation presented in Eq. (5), where vs, Es, vt, and Et are the Poisson’s ratio and Young’s moduli of the sample and tip, respectively,
F is the applied force, and Rtip and WAd. are the tip radius and the adhesion energy, respectively. We find that the contact radius of the AFM tip is approximately 25 nm, which gives a contact area of ∼2000 nm2 for 100 nN of applied force in a Ru-SiO2 contact. Considering that the AFM tip radius is 1 , only a small portion (about 2.5%) of the tip is actually in contact with the sample surface when the applied force is 100 nN. However, our measurements show that 2000 nm2 of contact area is sufficient for reproducible bond formation and breaking (from measurement to measurement), despite the aforementioned effect of surface roughening to increase variability. The deformation of the sample surface () is given by the expression and is approximately 0.63 nm for our measurements. Such a small deformation ensures that the materials are in the elastic regime, for accurate measurement of adhesion energy. The pressure underneath the tip is calculated to be ∼350 bars during the measurements.
Adhesive force per unit area in a functional NEM relay has been extracted by Yaung et al.8 and Lee et al.23 using indirect methods such as electrical characterization and indirect AFM-based analysis wherein the force is applied on the movable electrode of the relay. While the results extracted from the electrical characterization showed an adhesive force of 0.02 nN/nm2 between the W electrodes, the indirect AFM measurement showed a much lower adhesive force of 1.6 × 10−5 nN/nm2. The direct AFM based experimental analysis performed here between W-W contacts shows that the adhesive force is about 0.17 nN/nm2, which is an order of magnitude higher than the estimates by Yaung et al.9 We believe that this difference is due to the assumptions used in estimating the contact area: While Yaung et al.8 have used the dimple (apparent) contact area for the calculation of force per unit area, here we use the DMT model to estimate the microscopic (real) contact area.
As the contact area increases with applied force, so should chemical interaction forces. To investigate this effect, we performed an experiment in which the applied force was increased from 100 nN to 1 . The results presented in Fig. 4 show that the RuO2-SiO2 adhesion energy increases with the applied force, by about 25%. Using the DMT model, we estimate that the corresponding increase in the contact area is about 44%. This suggests that the sample is plastically deformed for 1 of applied force. Therefore, an applied force of 100 nN was used to obtain all of the data presented in Fig. 2. (An applied force of 50 nN was found to be too small to yield reproducible measurements.) It should be noted that the contact force in NEM relays is a strong function of the relay operation mode and body-biasing condition and could range from a few micro-newtons to very small values (theoretically approaching zero when the body-bias voltage is close to the release voltage). Thus, the 100 nN applied force in our AFM measurements is within the range relevant for relay operation.
Logic relays should be stable and operate with high endurance, more than 1014 cycles,24 in order to have a broad application. In the previous work,25 we demonstrated that relays with Ru contacts can operate with sufficient on-state conductance for at least 108 cycles, under low vacuum conditions. Stability of contact adhesion is also critical for achieving highly reliable relay operation, since a large adhesive force can cause a relay to be stuck in the on state. We have therefore studied the evolution of adhesive force in a RuO2-SiO2 contact over the course of many operating cycles. Due to the low frequency (0.25 Hz) of the AFM measurement technique, we were able to monitor the contact for only 1000 cycles; nevertheless, the results provide insight into the stability of adhesion energy. The results presented in Fig. 4(b) show that the adhesive force remains stable over the 1000 operating cycles that is expected to result in efficient device performance. Adhesion cycling tests performed on bare-Ru/SiO2 (shown in the supplementary material) indicate an increase in adhesive force by ∼6% over the 1000 contact cycles, probably due to wearing of the tip, highlighting the benefit of oxides in terms of contact stability.
Having discussed the adhesion properties of the metals and metal-oxides, we now comment briefly on their electrical properties. Previous reports22 indicate that RuO2 (the only Ru oxide that is stable in atmospheric ambient) is electrically conductive, since it has a resistivity that is only a factor of seven higher than that of pure Ru, at room temperature. Considering that the ion-beam synthesized RuO2 is only ∼2 nm thick, and that a contact resistance of ∼1 was measured between two RuO2 disks with 1 cm diameter,26 we would not expect it to dramatically increase the contact resistance. Also, oxide formation on Ru electrodes should help stabilize the stiction coefficient as a function of cycles under hot-switching conditions (see supplementary material). In contrast, WO3 is known to be electrically insulating,27 so the stoichiometry and thickness of the WOx layer must be carefully controlled (by adjusting the O2 ion implant dose and acceleration energy) to ensure a sufficiently high on-state electrical conductance.
In conclusion, we have demonstrated that low-energy ion beam synthesis and subsequent thermal annealing result in nanolayers of metal oxides (such as RuO2 and WOx) on Ru and W metal surfaces, and the adhesion energy of the metals can be reduced by up to a factor of ten through the formation of the oxide nanolayers. XPS analysis shows that the oxide nanolayers are ∼2 nm thick and stoichiometric (for RuO2 on the Ru metal surfaces), which is expected to result in low on-state resistance needed for relay operations. The adhesion energy is quantified using an AFM with microspherical tips. Reduction of the adhesion in NEM contact metals through oxide nanolayers is important not just for the relay switch applications, but also for several other types of MEMS devices including RF MEMS switches that rely on the contact between a conductive membrane and an insulating dielectric. Therefore, this technique can be used to improve the energy efficiency of a broad class of MEMS and NEMS devices.
See supplementary material for the details about SRIM simulation results, adhesion measurements with AFM tip of small diameter, quantification of capillary forces, effects of ion-implantation process parameters on adhesion, sensitivity test of SiO2 probe tip roughness, contact cycling results of bare-Ru/SiO2 adhesive force, effects of AFM contact frequency on measurements, possible effects of electrical current and triboelectric effect on contact adhesion, and possible packaging conditions of the NEM relays.
The work was supported by the Center for Energy Efficient Electronics Science (NSF Award No. 0939514). XPS experiments were performed at the Molecular Foundry of the Lawrence Berkeley National Laboratory, which was supported by the Office of Science, Office of Basic Energy Sciences of the U.S. Department of Energy under Contract No. DE-AC02–05CH11231. A. Peschot gratefully acknowledges post-doctoral fellowship support from the DGA. The authors declare no conflicts of interest.