Surface acoustic wave coupling between micromechanical resonators

The coupling of micro- or nanomechanical resonators via a shared substrate is intensively exploited to built systems for fundamental studies and practical applications. So far, the focus has been on devices operating in the kHz regime with a spring-like coupling. At resonance frequencies above several 10 MHz, wave propagation in the solid substrate becomes relevant. The resonators act as sources for surface acoustic waves (SAWs), and it is unknown how this affects the coupling between them. Here, we present a model for MHz frequency resonators interacting by SAWs, which agrees well with finite element method simulations and recent experiments of coupled micro-pillars. In contrast to the well-known strain-induced spring-like coupling, the coupling via SAWs is not only dispersive but also dissipative. This can be exploited to realize high quality phonon cavities, an alternative to acoustic radiation shielding by, e.g. phononic crystals. As the size of mechanical resonators is decreased, their resonance frequencies increase and acoustic wave-like propagation in their substrate becomes important. Here, the coupling of two micromechanical resonators by surface acoustic waves is modelled analytically and numerically, showing dispersive and dissipative coupling of the resonators.

T wo mechanical resonators mounted on the same substrate are considered coupled if the motion of one of the resonators affects the motion of the other resonator and vice versa. Such systems of coupled micro-or nanomechanical resonators are widely used in fundamental studies and practical applications. They are utilized to build highly precise mass sensors [1][2][3] , allow for the study of quantum-coherent coupling and entanglement between two distinct macroscopic mechanical objects 4,5 , and enable the investigation of collective dynamics 6 . In all of these cases, the considered resonance frequencies are in the kHz or low MHz regime and the coupling of the resonators is well understood. This is not the case at higher resonance frequencies in the order of several 10 MHz and above. At these frequencies, the wavelength of surface acoustic waves (SAWs) becomes much smaller than the usual dimensions of the resonators' substrates, assuming a typical SAW velocity 7,8 of around 3000 m/s. This allows coupling of the resonators by SAWs. An example of SAWcoupled resonators is illustrated in Fig. 1 by a pair of short micropillar resonators vibrating in the first bulk mode 9 , also known as compression mode 10,11 . Typical resonance frequencies of such pillar resonators are above 50 MHz. Short micro-pillar resonators are compatible with SAW devices [11][12][13] , and pillar-based phononic crystals and metamaterials are utilized to manipulate the propagation of SAWs 10,14 , including applications such as acoustic superlenses 15 , waveguides 16,17 , vibration attenuation 13,18 , mass sensing 19 and phononic graphene 20 .
In a lumped-element model, a substrate-mediated coupling is usually represented by a spring connecting two spring-mass systems 9,21 . In doing so, the interaction between the resonators is assumed to be instantaneous. This assumption is not valid if the resonators interact with SAWs. In this case, the coupling between the resonators has a delay. This is the propagation time of a SAW, which is created by one resonator 22,23 and travels to the other. Delay-coupled systems have been studied before: two resonators mounted on a string 24 , two resonators coupled by a rod 25 , or the interaction of air bubbles in water [26][27][28][29] , for example. However, a radiative coupling by SAWs between three-dimensional micro-or nanomechanical resonators mounted on a semi-infinite substrate has not yet been investigated. A coupling function, analogous to the spring coupling model, is unknown and would be helpful to a deeper understanding of recent experiments of coupled micropillar resonators 30 .
Here, we propose a coupling function for SAW-coupled microor nanomechanical resonators. First, we consider a pair of identical and SAW-coupled pillar resonators. We analytically derive the eigenfrequencies and quality factors of the symmetric and antisymmetric modes of the pillars as a function of their distance, compare the results to finite element method (FEM) simulations and discuss the strength of the SAW coupling. For the FEM simulations, we considered the first bulk mode of the pillars, since it represents the most simple case. Second, we investigate the frequency response of two non-identical and SAW-coupled pillar resonators vibrating in the first bending mode and compare the results with recent experiments of coupled micro-pillar resonators 30 .
Results SAW coupling model. The coupling by SAWs is schematically depicted in Fig. 1. Each of the resonators emits a SAW, which exerts an effective force F SAW on the other resonator. The origin of F SAW is the mechanical vibration of the resonators and the resulting forces exerted on the substrate surface. In a lumpedelement model, the effective force exerted by the vibration of a single resonator with an effective mass m and a displacement z that drives the SAW mode of the substrate is given by based on the principle of action and reaction. The dimensionless parameter g represents the coupling strength between the resonator's vibrational mode and the SAW mode. We consider the displacement of the resonator in z-direction, since the pillars depicted in Fig. 1 vibrate in the first bulk mode, whose vibration direction is the z-direction. For example, if a bending mode in x-direction is considered, z is replaced by x, the displacement in x-direction. It is not the force F s that directly acts on another resonator but the SAW created by F s . Due to the propagation of the SAW, the force F SAW exerted by the SAW has a smaller amplitude and is phase-shifted compared to F s . The change in the amplitude of F SAW corresponds to the change of the SAW amplitude and the phase shift is the difference in phase between the resonator and its emitted SAW at the site of the other resonator. Consequently, the force exerted on the other resonator by the emitted SAW is given by with Δϕðd; ωÞ ¼ ϕ R ðω; tÞ À ϕ SAW ðd; ω; tÞ where A SAW (d) and ϕ SAW are the normalized (A SAW (0) = 1) amplitude and phase of the emitted SAW at the site of the other resonator, ϕ R and ω are the phase and vibration frequency of the SAW-emitting resonator, t represents the time, c SAW is the velocity of the emitted SAW, and θ represents additional phase changes of the emitted SAW during its formation (see Supplementary Note 1 for details). In Eq. (2), we do not give an explicit expression for the SAW amplitude, since the amplitude of a SAW emitted by a threedimensional resonator mounted on a semi-infinite substrate strongly depends on the materials. Some substrate materials are anisotropic, such as single crystalline silicon or lithium niobate, so the intensity of the SAWs and, accordingly 31 , also their amplitude are a function of the propagation direction. This effect is known as phonon focusing 32,33 .
It is important to note that the presented model assumes linear elasticity and does not take into account non-linear effects. The model assumes further that the dimension of the resonator in the propagation direction of the SAW is small in comparison to the SAW wavelength λ SAW . Beyond that, the model does not consider the propagation loss of the SAWs. Popular SAW substrates, such as lithium niobate or quartz, feature a propagation loss 7,34 of less than 0.0031 dB/ λ SAW at frequencies below 1 GHz. Such a loss is small in comparison to intensity changes of the SAW due to propagation in two dimensions.
Eigenfrequencies and quality factors of two identical and SAW-coupled resonators. In the following, we consider the symmetric and antisymmetric modes of two identical SAWcoupled resonators to investigate the effect of the SAW coupling on the modes' eigenfrequencies and quality factors. Two SAWcoupled resonators exert the force F SAW on each other. For weak damping, this results in the following equations of motion where indices 1 and 2 give the number of the resonator and ω 0 = 2π f 0 and Q 0 are the eigenfrequency and quality factor of a single resonator, respectively. If the two resonators vibrate in a symmetric (+) or antisymmetric (−) mode, we can drop the indices, since the modes feature € z 1 ¼ ± € z 2 . The equations of motion of both resonators are then given by Using the ansatz z = z 0 e i ω t for the steady-state solution with complex amplitude z 0 results in and By comparing Eq. (7) with the case of a single resonator, it becomes clear that ω ± and Q ± are the eigenfrequencies and quality factors of the symmetric (+) and antisymmetric (−) modes of the SAW-coupled resonators. In contrast to the springlike coupling, the coupling by SAWs not only modulates the resonators' eigenfrequencies (dispersive coupling) but also modulates their damping (dissipative coupling). Due to the dependency of Δϕ on the resonators' vibration frequency, ω ± is a function of itself. In the case of small modulations of the eigenfrequencies ω ± ≈ ω 0 , the eigenfrequencies ω ± can be approximated by using Δϕ(d, ω ± ) ≈ Δϕ(d, ω 0 ). Furthermore, it is important to note that the SAWs are a part of the resonators' radiation losses into the substrate, which are represented by Q rad . Hence, the coupling force F SAW can only modulate Q rad and not other damping mechanisms included in Q 0 . Consequently, the product Q 0 g in Eq. (9) must be proportional to Q 0 /Q rad , which gives g ∝ 1/Q rad . This matches with the definition of g as the coupling strength between the resonators' vibrational mode and the SAW mode.  8) and (9) together with the simulated eigenfrequencies and quality factors. For the fitting and plotting, we exploited the small modulation of the eigenfrequencies, as discussed above. In addition to the normalized SAW amplitude A SAW , we also determined the phase difference Δϕ by a FEM simulation of a single pillar resonator. Along a line in y-direction on the surface, we determined the displacement in z-direction u z of an SAW emitted by the single pillar and calculated the absolute value |u z | ∝ A SAW and the phase ϕ uz ¼ ϕ SAW . We chose the displacement in z-direction, since the pillars vibrate in z-direction. Based on ϕ SAW and Eq. results of the single pillar FEM simulation are given in Supplementary Note 2.
The simulated eigenfrequencies f ± and quality factors Q ± of the symmetric and antisymmetric mode of the two pillar resonators are plotted in Fig. 3 as a function of the pillars' separation distance. The distances between the pillars range from 0.3 to 10.5 λ SAW /2. In comparison, coupled beams or cantilevers 3,5,21 in the frequency regime around 100 kHz usually have a separation distance in the order of 10 −3 λ SAW /2, assuming a typical SAW velocity 7,8 of around 3000 m/s. The eigenfrequencies f ± and quality factors Q ± of both modes are periodically modulated with increasing distance d. The modulations of f ± and Q ± are phaseshifted by π/2, which implies that the coupling between the pillars is purely dispersive at the maxima and minima of f ± and purely dissipative at its zeros. The results of the FEM simulations are in excellent agreement with the proposed SAW coupling model, as can be seen in Fig. 3 by the red lines.
With increasing distance d, the amplitude of the modulations of f ± and Q ± decreases, since the amplitude of the emitted SAWs decreases with increasing propagation distance. The huge difference in the modulation strength between Q + and Q − at small distances d becomes clear by considering the limiting case d → 0 and pillarradius → 0, which is discussed in the following. The quality factor is defined as the ratio between the energy stored and energy lost during one cycle at resonance 9 . In the antisymmetric case, the pillars behave as a non-radiating source [39][40][41] . No power is emitted due to destructive interference of the emitted waves, which results in Q − → ∞. In the symmetric case, the two pillars emit four times the power emitted by a single pillar due to constructive interference 31 and store together twice the total energy, which yields Q + /Q 0 = 0.5. The discussed limiting case shows that the dissipative coupling can also be interpreted as a result of wave interference, which explains the difference in the modulation strength of the quality factors and the eigenfrequencies. Only a small fraction of the waves emitted by one resonator reaches the other resonator directly. In comparison, the interference affects all waves emitted by the pillars.
In addition to the presented FEM simulations, we simulated the second bending mode of the discussed pillar pair and a typical thin cantilever geometry vibrating in the first bending mode. The results are given in Supplementary Notes 3 and 4 and are in excellent agreement with the SAW coupling model.
Strength of the SAW coupling. A feature of major interest in the context of coupled micro-and nanomechanical resonators is the strength of the coupling. A distinction is typically made between the weak and the strong coupling regime. In the weak coupling regime the energy stored in one resonator dissipates before it is transferred to the other resonator, whereas in the strong coupling regime the resonators exchange energy and the dynamics of the coupled resonators are governed by the motion of both resonators 42 . For spring-like coupled resonators the strength of the coupling can be estimated by the ratio (f − − f + )/(f 0 /Q 0 ), since f − − f + is proportional to the energy exchange rate between the resonators and f 0 /Q 0 is proportional to the energy dissipation rate of the system 42 . This approach can not be used for SAWcoupled resonators, since the quality factor of the resonators is a function of the time t due to the dissipative part of the coupling. To get an impression of the strength of the coupling between the two pillar resonators, we considered the dynamics of two pillar resonators separated by d = 0.3 λ SAW /2 in comparison with two spring-coupled resonators at the threshold of the strong coupling regime. In detail, we calculated the displacements of the resonators as a function of time by numerically solving the equations of motion of each system for the following complex initial conditions where z 0 is a complex amplitude. At t = 0, both resonators are at rest, but resonator 1 is displaced. We chose _ z 1 ðt ¼ 0Þ such that the displacement and the velocity of resonator 1 are phase-shifted by π/2, and that its maximal kinetic energy equals its maximal potential energy.
The equations of motion of two SAW-coupled resonators are given by Eqs. (4) and (5). For weak damping, the equations of motion of two identical spring-coupled resonators are given by where ω c is the coupling rate. We numerically solved both systems for the following parameter values spring-coupled resonators 6,9 . Further details of the numerical calculations are given in the Methods section.
The calculated amplitudes of the resonators and their phase relationships are depicted in Fig. 4. It can be seen that the two systems exhibit different dynamics. The spring-coupled resonators exchange their total energy E tot ∝ |z i | 2 back and forth and vibrate with a phase difference of 90 ∘ , as shown in Fig. 4a, c. In contrast, the pillar resonators vibrate antisymmetrically after a transient time and only transfer energy until both pillars have the same total energy, as shown in Fig. 4b, c. The reason for the different dynamics is the dissipative part of the SAW coupling. In contrast to the spring-coupled resonators, the pillar resonators maximize their quality factor, which is displayed by the exponentially decaying lines in Fig. 4a, b. Based on the FEM simulations discussed above, two pillar resonators, which are separated by d = 0.3 λ SAW /2 and vibrate in the antisymmetric mode, have a quality factor of Q − = 1455 instead of Q 0 = 208. By comparing the amplitudes |z 2 | of both systems, it can be seen that the second pillar resonator has a larger maximum amplitude than the second spring-coupled resonator. Consequently, the ratio of the energy exchange rate to the energy dissipation rate is larger for the SAW-coupled pillar resonators than for the springcoupled resonators. Since the spring-coupled resonators are at the threshold of the strong coupling regime, the pillar resonators are strongly coupled. It is important to note that the pillar resonators' damping is dominated by radiation losses into the substrate. If this were not the case, the coupling would be weaker.
Frequency response of two non-identical and SAW-coupled resonators. The motivation to consider the frequency response of two SAW-coupled resonators is to further test the proposed SAW coupling model by comparison to recent experiments 30 . Raguin et al. drove a pair of coupled micro-pillar resonators by an incident SAW and measured the frequency response of the pillars by laser scanning heterodyne interferometry. A schematic of the sample is shown in Fig. 5a. A SAW is created by an interdigital transducer, travels towards the pillars and drives the first bending mode of the pillars, which is illustrated in Fig. 5b.
In the following, we assume that the pillars bend along the x-direction and consider non-identical resonators since identically fabricated resonators usually are not identical due to fabrication limitations. For example, Raguin et al. 30 give an uncertainty of 3-4% on the dimensions of their fabricated micropillars. Consequently, it is probable that the pillars of a pair have slightly different eigenfrequencies ω 1 ≠ ω 2 and quality factors Q 1 ≠ Q 2 , which result in the following equations of motion where F D is the force applied on the pillars by the incident SAW.
It is important to have in mind that F D ≠ F SAW . F SAW is the force exerted on each pillar by the SAW emitted by the other pillar. Applying the ansatz for the steady-state solution to Eqs. (13) and (14), as discussed above, gives the frequency response of the two with C n ¼ 1 þ ω 2 g A SAW ðdÞ e ÀiΔϕðd;ωÞ δx m≠n 1 À ðω 2 gA SAW ðdÞe ÀiΔϕðd;ωÞ Þ 2 δx 1 δx 2 ð16Þ and the normalized frequency response of a single resonator where the indices n, m ∈ {1, 2} give the number of the resonator. From Eq. (15) it can be seen that the frequency response of the SAW-coupled resonators is given by the frequency response of the single resonators modified by C n . where A SAW (d) and Δϕ(d, ω) were evaluated for the distance d = 5.9 μm. The simulated eigenfrequency and quality factor of the single pillar fit well with the experimental results of Raguin et al. 30 , who measured an eigenfrequency of f 0 = 70.54 MHz and a quality factor of around Q 0 ≈ 34. When we fitted Eq. (15) to the experimentally measured frequency responses, we used A SAW determined by the FEM simulation of a single pillar and made the following assumptions. First, we set m 1 = m 2 , since the difference in mass between the identical fabricated micro-pillars is small. Second, we assume Q 1 = Q 2 = Q 0 , since Benchabane et al. 11 showed that the quality factor of the micro-pillars only changes slightly with small variations in their dimensions. Third, we set Δϕ(d, ω) = 0.61, the value which we determined by the FEM simulation of a single pillar since the SAW wavelength changes less than 10% in the applied frequency range relative to f 0 . The fitting of Eq. (15) to the experimental data of Raguin et al. 30  The calculated quality factor is in perfect agreement with the FEM simulation of the single pillar, which gave the identical quality factor. The measured and calculated frequency responses are shown in Fig. 5c. Each pillar shows two resonances: One resonance around 67 MHz and the other around 71 MHz. These frequencies are close to the eigenfrequencies f 1 and f 2 of the single pillars, which we determined by the fitting of Eq. (15). Hence, the difference in dimension of the pillars is the main reason for the frequency difference of the coupled pillars and not the dispersive coupling. In the case of identical pillars, the given value of g would result in a significant frequency difference of a few MHz of the pillars' symmetric and antisymmetric modes. Beyond that, it can be seen in Fig. 5c that the resonances around 67 MHz and 71 MHz differ in their quality factors. We estimated the quality factor of the main resonance of each pillar by fitting the frequency response of a single resonator, which is given by Eq. (17) multiplied by an amplitude. We determined a quality factor of around 30 for the main peak around 67 MHz and a quality factor of around 110 for the main peak around 71 MHz. Such a difference in the quality factor of more than a factor of 3 is unexpected for the identically fabricated micro-pillars, as discussed above. It is more plausible to assume that the difference in the quality factors has its origin in the coupling of the pillars. However, the usually used spring coupling model can not explain such a modification of the quality factors, since the spring-like coupling is purely dispersive. In contrast, the SAW coupling model is able to explain it and describes the experimental data well. The micro-pillars are separated by d < 0.3 λ SAW /2, which is the minimal distance between the two identical resonators discussed above. Based on Fig. 3, this means that the pillars show a decreased quality factor in case of a symmetric vibration and an increased quality factor in case of an antisymmetric vibration. Raguin et al. 30 showed that the pillars vibrate symmetrically around 67 MHz and antisymmetrically around 71 MHz.

Conclusions
In conclusion, our results indicate that the coupling between MHz frequency resonators can significantly differ from the usually assumed and purely dispersive spring-like coupling. The coupling via SAWs is not only dispersive but also dissipative. This makes our results important for all fields working with micro-or nanomechanical resonators with resonance frequencies in the MHz regime and above, for instance, NEMS/MEMS and acoustic metamaterials. In MEMS and NEMS, a major approach to increase the devices' sensitivity is to shrink the sizes of the mechanical resonators, which shifts their resonance frequencies from the kHz to the MHz regime [43][44][45] . Furthermore, for quantum applications at room temperature, a high Qf-product is needed 46,47 . The dissipative coupling opens up a new possibility to reduce acoustic radiation losses via the substrate by positioning two or more resonators at certain distances. In comparison to acoustic radiation shielding with phononic crystals [48][49][50] , phonon cavities based on dissipative coupling are accessible by external SAWs and, thereby, allow to perform quantum acoustics experiments between a SAW 51 and coupled mechanical resonators.
Pillar-based metamaterials are built on the local resonances of the pillar resonators 10 . Hence, the exact knowledge of the pillars' resonance frequencies and quality factors is of great importance for the design of metamaterials, e.g., for the effective guidance 16,17 or attenuation 13,18 of acoustic waves with a known frequency. Our results give insights into the coupling of the individual pillar resonators and the resulting impact on their resonance frequencies and quality factors.

Methods
Details of the FEM simulations. The FEM simulations were carried out with COMSOL Multiphysics (Version 5.4) using the Modules AC/DC (Electrostatics) and Structural Mechanics (Solid Mechancis, Piezoelectricity). An unstructured tetrahedral mesh was used for the pillars with a maximum element size of an eighth of the pillars' diameter. A swept mesh was utilized for the PML and an unstructured tetrahedral mesh for the inner part of the substrate. The latter had a maximum element size of an eighth of the SAW wavelength λ SAW up to a distance of λ SAW to the surface, which is approximately the penetration depth of a SAW 7 . For elements deeper in the substrate, the maximum element size was reduced by a factor of two. In Solid Mechanics, quadratic serendipity elements were used as Lagrange elements were used in Electrostatics. We performed an eigenfrequency study to calculate the eigenfrequencies and quality factors of the symmetric and antisymmetric modes for different distances d between the pillars and in the simulation of a single pillar resonator.
To minimize the memory requirements and the solution time, we reduced the simulated domain to half of the considered domain by using the following symmetric boundary condition u Á n ¼ 0; where u is the displacement vector and n is the unit normal vector of the considered sectional plane. In our case, this is the yz-plane, as shown in Fig. 2. The yz-plane lies in a plane of mirror symmetry of the lithium niobate substrate 38 , and a pillar resonator made out of an isotropic material shows no displacement in x-direction at any point in the yz-plane when vibrating in the first bulk mode in the limit of small vibrational amplitudes 9,52 .
Details of the numerical calculations. The numerical calculations were carried out in Matlab 2020a. We converted the second-order differential equations to firstorder differential equations and solved the first-order differential equations by the function ode45 using a stepzise of Δt = 25/f 0 , and a relative and absolute error tolerance of 10 −6 and 10 −8 , respectively.

Data availability
The data that support the findings of this study are available from the corresponding author upon reasonable request.

Code availability
The codes used in this study are available from the corresponding author upon reasonable request.