NEMS With Broken T Symmetry: Graphene Based Unidirectional Acoustic Transmission Lines

In this work we discuss the idea of one-way acoustic signal isolation in low dimensional nanoelectromechanical oscillators. We report a theoretical study showing that one-way conversion between in-phase and anti-phase vibrational modes of a double layer graphene nanoribbon is achieved by introducing spatio-temporal modulation of system properties. The required modulation length in order to reach full conversion between the two modes is subsequently calculated. Generalization of the method beyond graphene nanoribbons and realization of a NEMS signal isolator are also discussed.

Owing to their outstanding mechanical and electrical properties, these materials have been utilized as electromechanical oscillators in nanoscale memory cells and nanoelectromechanical switches [20][21][22][23] and resonators [24][25][26] .The study of mechanical waves 27 and the ability to manipulate, control, and detect vibrational motion in such nanoelectromechanical systems (NEMS) [28][29][30][31] provides unprecedented opportunities to employ them in fluidic 32 , electronic 33 , and optical networks 34 . NEMS based oscillators as mechanical sensors and actuators are also used in applications such as ultrasensitive force 35 and displacement detection 36,37 , scanning probe microscopy 38 and resonant mass sensing of chemical and biological species 39,40 where an important functionality is to have one-way communication channels that transmit desired signals only in one direction.
Acoustic rectifiers and diodes [41][42][43][44][45] as well as one-way acoustic isolators 46 have been recently studied. The acoustic diode concept is based on a nonlinear frequency conversion mechanism 44,47,48 . Other studies have focused on nonreciprocity in acoustic circulators and acoustic metamaterials 49,50 as well as optomechanically induced nonreciprocity in resonators 51,52 . More recently, the idea of one-way phonon isolation 46 , motivated by optical equivalents 53,54 , was studied based on creating spatio-temporal modulation of mechanical properties. This leads to a one-way conversion between the guided modes, therefore breaking the symmetry of wave propagation in a waveguide in forward and backward directions.
In this letter we study the idea of one-way signal isolation in low dimensional nanoelectromechanical oscillators where the symmetry of the system under time reversal transformation, also known as the T-symmetry, is broken. To explain the method, we consider a system of graphene nanoribbons (GNRs) on an elastic substrate and demonstrate that the symmetry of wave propagation may be broken by introducing spatial and temporal modulation of elastic properties of the system. We show, both analytically and numerically, that in one of the propagation directions conversion between the modes occurs, whereas in the other direction the signal is transmitted without any perturbations. We also discuss the extension of this method beyond graphene nanoribbons and mention its possible implementation for designing a phonon isolator in nanoelectromechanical oscillators.
In Fig. 1(a), a double-layer graphene nanoribbon is shown. The system consists of two graphene nanoribbons with width b, each of which is perfectly adhered to an elastic substrate. We use nonlocal elasticity theory to study wave propagation along the nanoribbons 55,56 . In this model, shown in Fig. 2, each substrate is treated as a linear elastic medium with stiffness k w 56 , and the nanoribbons themselves interact via van der Waals forces that are also modeled as linear springs with stiffness c 56 . The governing equation for wave propagation in this system is driven from the nonlocal Euler-Bernoulli beam model 57,58 ,  Here w 1 and w 2 are flexural displacements of nanoribbons 1 and 2 in the y direction, A is the cross sectional area of each GNR, ρ is the density, I is the moment of inertia, E is the Young's modulus, a is the C-C bond length and e 0 is a parameter representing nonlocal elastic effects in the GNR 56,57 . Without loss of generality, we consider a simplified case with e a 0 0 = and k c w = 56 , and use the parameters obtained in Ref 56 . The two governing equations (Eqs. 1 and 2) can then be written in the form L u 0 → = where u w w , and L is a linear operator.
Solutions of the form u D Clearly, the dispersion curves plotted in Fig. 1(b) are symmetric with respect to wavenumber β, implying that the wave propagation in such a waveguide is reciprocal, i.e. waves traveling in forward and backward directions have the same properties. In order to break this symmetry in the wave propagation phenomenon we follow the technique suggested in Refs 46,53 . In particular, we consider wave propagation with a spatio-temporally modulated elastic matrix. In this case we expect that for an appropriately chosen modulation, one-way conversion between the guided modes may be induced, i.e. interaction between the guided waves is possible in one propagation direction only.
We assume spatio-temporal modulation of the elastic matrix constant so that where k w0 is the original stiffness constant and k w δ is the modulation depth. In order to maximize the coupling between modes, we modulate only the upper elastic matrix, as shown in Fig. 2. To solve the governing equations for the modulated system, we assume a general solution as the superposition of the guided modes: , where 1 2 β , and 1 2 ω , are the wavevectors and frequencies of in-phase and anti-phase modes, and a x 1 ( ) and a x 2 ( ) are their slowly varying spatial amplitudes. Next, using the standard techniques of the perturbation theory, two ordinary differential equations are obtained.
Based on the incoming signal frequency 1 ω and the available modulation frequency Ω, a full mode conversion from mode 1 to mode 2 occurs when 2 1 ω ω = + Ω and when the B parameter of the modulation is chosen such that B In this case, under phase-matching conditions, the differential equations can be simplified to . The system shows one-way behavior because the modulation does not convert the mode at β ω ( − , ) 1 1 to any other modes. The resulting end point of the intended transition in the negative β region does not lie on the anti-phase branch of the dispersion curve ( Fig. 1(b)). This one-way behavior arises because the modulation breaks both time-reversal and spatial-inversion symmetry.
If we consider the propagation of the in-phase mode in the system, the initial conditions will be a 0 1 1 ( ) = and a 0 0 2 ( ) = . The solution to equation 4 is then written as a x cos x 1 ξ ( ) = ( ) and a x sin x 2 ξ ( ) = ( ), where ξ is the conversion wavevector. As described in Fig. 3(a), complete transition from in-phase to anti-phase mode is observed for the wave propagating in one direction (solid lines), while in the opposite direction the in-phase mode will not be influenced (dashed lines). The general solution of the governing equations is then proportional to the modulation depth. This means that with stronger modulation, the mode conversion requires a smaller modulation domain. To analyze the conversion length further, we investigate its dependence on the modulation parameters. The intent of adding a modulation domain is to convert an incoming in-phase mode signal with frequency 1 ω and wavenumber 1 β to the anti-phase mode using the available modulation frequency Ω and tunable modulation wavenumber B. Therefore, we plot the dependence of the conversion length on wavenumber 1 β for different values of Ω. As shown in Fig. 3(b), the required conversion length is larger for higher wavenumbers, as l c varies proportional to 1 2 It is worth mentioning that the choice of Ω and 1 β is not arbitrary and should satisfy two criteria. First, the corresponding mode 1 frequency, 1 ω , needs to be below the cut-off frequency. Second, the resulting frequency value 1 Ω ω + needs to lie on the dispersion curve of mode 2 in the forward propagation direction (positive β side) and lie on a band gap region in the backward propagation direction (negative β side) consistent with Fig. 1(b).
As an example, we choose a value slightly below the cut-off frequency to guarantee that at frequency  . It is important to emphasize that the parameter values chosen are arbitrary and longer wavelength signals corresponding to smaller choices of 1 ω can also be used to observe the same isolation effect as long as the required conditions explained earlier are satisfied. Additionally, lower values of the modulation depth also result in a mode conversion, but require a longer modulation domain. Furthermore, the operational bandwidth is also important in evaluating the performance of the isolator. In order to estimate the bandwidth of the system operation depending on the modulation length, we have calculated the transmission characteristics of our system analytically. Typically, the outgoing signal can be a combination of mode 1 and mode 2 signals as described by the spatially varying mode amplitudes, a x 1 ( ) and a x 2 ( ), shown in Fig. 3(a). In the general case where a phase-mismatch could be present, the spatially varying amplitudes are given as , and with different modulation depths of k k 0 15 , and k k 0 05 w w δ = .
, which correspond to modulation lengths of around 24 1 λ , 35 1 λ and 71 1 λ respectively. As shown in Fig. 3(c), the transmission coefficient drops fairly rapidly as the frequency is changed. This drop occurs because, due to the shape of the dispersion curves, the spatial modulation B is no longer commensurate with the difference between the wavenumbers of modes 2 and 1 at 2 ω δω + and 1 ω δω + . The corresponding bandwidth was calculated by considering a frequency span at which the transmission coefficient drops by 50%. For the three cases studied here shown by the red, blue, and green curves, the bandwidth is around 0 014 c ω .
respectively. The results show that in order to have a larger operational bandwidth, a larger modulation depth needs to be provided.
In order to confirm the analytical model, we used numerical simulations of the wave propagation and mode conversion in double-layer GNRs using the finite difference time-domain (FDTD) method to solve the governing equations in the presence of spatio-temporal modulation. For this purpose, we choose a modulation domain with length l c as predicted from the analytical model. The numerical simulation results are shown in Fig. 4. An incoming in-phase wave of wavelength 1 λ will be converted to an anti-phase wave of wavelength 2 λ after passing through the modulation domain. The Fourier spectra of both the incoming and outgoing signals are shown in Fig. 4(b). From the FDTD simulations, the peak corresponding to the frequency of the outgoing signal (red curve) is around 1 08 which shows good agreement with the analytical method. The design of the system provides conversion of a signal of mode 1 to a signal of mode 2 in only one of the propagation directions. Therefore, in the opposite direction where no conversion between the two modes happens, we essentially observe a transmission ratio of 1 for the signal. In the other direction the signal of mode 1 is converted to mode 2 which implies a transmission ratio of close to zero. In fact, qualitatively, this type of isolation results in the following scattering matrix for the two ports: 0 0 1 0 ( ) . More specifically, if we consider the Fourier spectrum of the outgoing signal in Fig. 4(b), a transmission ratio of slightly above zero is observed for the forward propagation direction. However, this value of transmission ratio is influenced by the numerical errors within the framework of the FDTD method and it is important to keep in mind that the analytical method suggests a transmission ratio of zero for this scenario. Experimental realization of spatio-temporal modulation of elastic properties has been investigated recently. Spatial property modulation is possible through periodic arrangement of elastic material similar to the case of phononic crystals 59 . Temporal property modulation, although more difficult, is also possible in practice through applying electric fields, magnetic fields or mechanical strains in a periodic fashion [60][61][62] . In general, spatial modulations with nanometer scale periodicity provide suitable conditions for conversion between the two vibrational modes of double-layer GNRs. Furthermore, THz range frequencies are appropriate for temporal modulations to guarantee the unidirectional nature of such systems. Beyond double-layer GNRs, the aforementioned technique is also applicable to other systems.
The key here is to have a system with two different branches of dispersion curve separated by a cut-off frequency. Using the same method, one-way mode conversion is achievable in systems such as single layer GNRs or double-wall carbon nanotubes where similar types of governing equations describe their behavior 56,[63][64][65][66] . It is also important to mention that for very short wavelength limits the physics of the problem may differ from the predictions of continuum models and therefore atomistic level techniques such as molecular dynamics simulations 67 may be required for the analysis.
In this paper, we explored one-way acoustic signal isolation in graphene nanoribbons. We showed that through spatio-temporal modulation of the system properties, we can convert a signal of one mode to another mode in one direction, while no conversion is observed in the opposite direction. Combined with appropriate mode filters that filter out signals of frequency 2 ω , a signal of frequency 1 ω can be . absorbed or filtered out after conversion to a signal of frequency 2 ω in one propagation direction while it will be transmitted with no disturbance as the same signal with frequency 1 ω in the opposite direction. This method is not limited to graphene nanoribbons and can be used to induce the same type of signal isolation for acoustic wave propagation in other low dimensional oscillators. The realization of a NEMS based signal isolators raises intriguing possibilities for a wide range of applications in scanning probe microscopes, force and displacement detection devices and chemical and biological sensors.