betaFeedback

WorksPhysics and Astronomy

Nanoscale Magnonic Neurons

Abstract

We use micromagnetic simulations to demonstrate neuron functionality of two-dimensional (2D) chiral magnonic resonators. Our design exploits nonlinear resonant scattering of spin waves propagating in a YIG medium from an edge mode of a permalloy nano-element. The reduced frequency and volume of the edge mode facilitate matching it to the YIG modes and give rise to their wide-angle scattering. As the amplitudes of the incident spin waves increase, the edge mode exhibits a positive nonlinear frequency shift. This shift leads to a complex frequency-dependent nonlinear variation of the amplitude and phase of spin waves scattered in different directions. We show that the scattered waves are strong enough to activate secondary neurons. This provides the connectivity required for combining our proposed neurons into 2D magnonic neural networks.

The rising significance of research in machine learning and artificial neural networks for modern society was recently recognized by the award of the 2024 Nobel Prize in Physics. Yet, the advent of the artificial intelligence (AI) brings about not only benefits but also the challenge of mounting energy costs of implementing AI with conventional, CMOS-based computers. Magnonics [1] might offer a viable alternative hardware platform for energy-efficient neuromorphic and, more generally, unconventional computing. This would use spin waves and their quanta (magnons) [2, 3]) as information carriers that feature strong nonlinearity, modest propagation loss, and sub-micrometre wavelengths at GHz frequencies [1, 2, 3, 4, 5]. These properties have been used to build complex nonlinear reservoirs underpinning various forms of magnonic reservoir computing [4, 5, 6, 7, 8], showing different levels of trainability and success in common benchmark tasks.

An alternative, bottom-up approach involves devising a magnonic neuron first and then using it as a building block of more complex artificial neural networks. In Ref. [9], we explored nonlinear properties of chiral magnonic resonators [10, 11, 12] in view of their potential applications as artificial “neurons”. However, only a one-dimensional (1D) system was considered under assumption of the incident and transmitted spin wave signals propagating along the same direction. This would limit options for connecting the neuron to those in subsequent layers. Moreover, the need to match the frequency of the propagating modes to those confined in the resonator led us to using the same material (Permalloy) for both the resonator and the magnonic medium. Yet, construction of large neural networks with a competitive computing efficiency dictates the use of low-damping materials, e.g. yttrium-iron garnet (YIG) [13, 14], for the magnonic medium [15, 16].

In this Letter, we use micromagnetic simulations to demonstrate neuron functionality of two-dimensional (2D) chiral magnonic resonators. Such neurons exploit nonlinear resonant scattering of spin waves propagating in a YIG medium from an edge mode of a permalloy disk nano-element. The reduced frequency and volume of the edge mode [17, 18] facilitate matching it to the YIG modes and give rise to their wide-angle scattering. As the amplitudes of the incident spin waves increase, the edge mode exhibits a positive nonlinear frequency shift. This shift leads to a complex frequency-dependent nonlinear variation of the amplitude and phase of spin waves scattered in different directions. We show that the scattered waves have amplitude sufficient to “activate” secondary neurons. This provides the connectivity required for combining our proposed neurons into 2D magnonic neural networks [5, 12].

We envisage implementation of a magnonic neuromorphic device architecture based on a 'unit cell' building block shown in Figure 1(a). Three inputs I1I_1, I2I_2, and I3I_3 emit spin waves that are incident upon a permalloy disk that serves as a magnonic neuron. The neuron receives, nonlinearly transforms, and re-emits spin waves propagating linearly in a low-damping magnonic medium, such as YIG film. To show that such a system is suitable for performing neuromorphic computations, we characterise its response using finite-difference time-domain micromagnetic simulations with the help of the MuMax3 software [21]. The system is discretised using cuboidal cells with edge length of 5×5×105 \times5 \times10 nm3^3 on a grid of 2048×1024×102048 \times1024 \times10 cells (x×y×zx \times y \times z). ‘Edge-smoothing’ of order 4 is adopted to approximately correct the discretisation error at the edges. The mode spectrum of magnetic nanoelements (and especially the edge mode’s frequency) is very sensitive both to the edge quality experimentally [22, 23, 24] and to the discretisation cell size numerically [25, 26]. So, no attempt is made to elaborate our micromagnetic model beyond a qualitative agreement with existing experimental results [17, 22, 23, 24]. To approximate an infinite extent of the film, periodic boundary conditions are applied in the in-plane dimensions, alone with a gradual increase of the Gilbert damping in the ±x\pm x directions. The corresponding procedures are described in the Supplementary material [27]. The boundaries between the regions of increased and uniform damping lie approximately 2 µm from the edges of the simulation grid along xx. The system is relaxed to its static configuration in the bias magnetic field of 12 mT before applying a suitable dynamic magnetic field to study its linear and nonlinear dynamics.

Design of the magnetic neuron and its response

Figure 1. (a) Design of the magnetic neuron. Spin waves from sources I1I_1, I2I_2, and I3I_3 are scattered by a chiral magnonic resonator (the magnonic neuron) – a 10 nm thick permalloy (Py) [19] disk of 60 nm diameter spaced by 40 nm from a 50 nm thick YIG film [20]. A uniform bias magnetic field of 12 mT is applied in the +y+y direction. (b) The spectra of the precessional response of the disk to a uniform linear excitation (solid black curve) and the normalised scattering directivity (dashed red curve) are shown for frequencies in the vicinity of the disk’s lowest frequency resonance (edge mode). The inset shows the mode’s profile, with the brightness and colour representing the Fourier amplitude and phase calculated from the out-of-plane component of the dynamic magnetisation. (c) The normalised directivity of the resonant scattering of spin waves at the frequency of the edge mode (1.935 GHz) is shown in the reciprocal space (red curve) and in the real space (blue curve), with the scattering angle of 0° corresponding to the +x+x direction.

The linear spectrum of the system is shown in Figure 1 (b). The spectrum is obtained by exciting the region of the disk (but not the YIG film) by a weak, spatially asymmetric broadband field pulse and then summing the Fourier amplitudes calculated cell-wise from the disk’s magnetization [17, 18, 22, 23, 24, 25, 26]. The lowest frequency resonance at 1.935 GHz lies well within the dipolar spin-wave continuum in YIG and is also the strongest and well separated from higher order modes. It corresponds to the disk’s edge mode, whose frequency is reduced (compared to the Kittel frequency of a plane film made of the same material) by the demagnetising field due to the magnetic charges formed at the disk edges [17, 18].

The dynamic dipolar coupling of the spin waves propagating in YIG to the edge mode (the ‘local’ mode) leads to their resonant scattering. To obtain its directivity shown in Figure 1 (c), a 1.935 GHz frequency plane spin wave is launched in the +x+x direction (0°), after which the scattered wave’s Fourier amplitudes are calculated from the YIG film’s magnetisation for different reciprocal space directions [28]. The directivity reveals that the resonant scattering is chiral, i.e. the disk acts as a chiral magnonic resonator with minimal backscattering [12]. The strong forward-backward asymmetry facilitates implementing feed-forward networks of such neurons. In the reciprocal space, the scattering occurs into a wide angle of about ±60∘\pm60^\circ. However, the anisotropic magnonic dispersion of the in-plane magnetised YIG film [29, 30] means that the scattering angle shrinks by about a half in the real space, i.e. when considered in terms of the directions of the group velocity. This scattering angle determines the number of other neurons with which the given one would be able to communicate (i.e. receive from or feed to) the signal.

Before we present our results of nonlinear spin-wave scattering from the nano-disk, we first characterise the nonlinearity of its edge mode resonance. We drive the nano-disk placed above the YIG film (but not the film itself) by a monochromatic spatially-uniform circularly polarised magnetic field of the form

Bex(t)=B0(x^sin⁡(2πft)+z^cos⁡(2πft)),(1)B_{\mathrm{ex}}(t)=B_0\left(\hat{x}\sin(2\pi f t)+\hat{z}\cos(2\pi f t)\right), \tag*{(1)}

where B0B_0 and ff are the driving field’s amplitude and frequency, respectively. The circular polarisation mimics that of the dynamic stray field of propagating plane spin waves with a positive wave number [31]. The disk’s response to this field therefore provides a reference when considering nonlinear spin-wave scattering later. Given that the spin-wave wavelength λ\lambda is at least an order of magnitude longer than the diameter of the disk, equation (1) provides a good approximation of the spatial variation of the dynamic stray field of incident spin waves in the region of the disk. To speed up the simulations, the grid size is reduced to 320×320×10320 \times320 \times10 cells in the (x×y×z)(x \times y \times z) directions with the cell size of 5×5×10 nm35 \times5 \times10\ \mathrm{nm}^{3}, while increasing the number of grid images in the periodic boundary conditions and adding absorbing boundary conditions in the ±y\pm y directions [27]. When the excitation strength is small, e.g. for B0=0.005 mT≡BlinearB_0 = 0.005\ \mathrm{mT} \equiv B_{\mathrm{linear}}, the response of the edge mode is linear and can be modelled as a Lorentzian in terms of the amplitude ∣φ∣|\varphi| of the local mode

[(ω−ω0)2+Γtot2]⋅∣φ∣2=Alinear2.(2)\left[(\omega-\omega_0)^2+\Gamma_{\mathrm{tot}}^2\right]\cdot|\varphi|^2=A_{\mathrm{linear}}^2. \tag*{(2)}

where ω0\omega_0 is the mode’s angular frequency, Γtot\Gamma_{\mathrm{tot}} is the resonance linewidth that includes contributions from both Gilbert damping and radiative damping [11], and AlinearA_{\mathrm{linear}} is a product of BlinearB_{\mathrm{linear}} and a coefficient describing the mode’s coupling strength to the field given by Eq. (1). We fit Eq. (2) to the edge mode’s response obtained from micromagnetic simulations as

∣φdata(B0,f)∣=2NFFT⋅∑i=1N∣F{Mx(i)}∣,(3)|\varphi_{\mathrm{data}}(B_0,f)|=\frac{2}{N_{\mathrm{FFT}}}\cdot\sum_{i=1}^{N}\left|\mathcal{F}\{M_x^{(i)}\}\right|, \tag*{(3)}

where F{Mx(i)}\mathcal{F}\{M_x^{(i)}\} is the fast Fourier transform in time of the xx component of the dynamic magnetisation for the ithi^{\mathrm{th}} cell, with index ii enumerating all cells in the disk with non-zero magnetisation. The pre-factor of 2NFFT\frac{2}{N_{\mathrm{FFT}}} (where NFFTN_{\mathrm{FFT}} is the length of the data in the time domain) ensures normalisation of the Fourier transformed real-valued data. So, ∣φdata(B0,f)∣|\varphi_{\mathrm{data}}(B_0,f)| has the meaning of the MxM_x oscillation amplitude averaged over the disk’s volume. The fit yields ω0=2π×1.935 GHz\omega_0=2\pi\times1.935\ \mathrm{GHz}, Alinear=2π×401.21 A m−1 GHzA_{\mathrm{linear}}=2\pi\times401.21\ \mathrm{A\,m^{-1}\,GHz}, and Γtot=2π×0.077 GHz\Gamma_{\mathrm{tot}}=2\pi\times0.077\ \mathrm{GHz}, which we use below.

To characterise the nonlinearity of the nano-disk, we excite it with microwave fields of increased amplitude B0B_0. The nonlinearity parameter λ~\tilde{\lambda} is then extracted by fitting the simulated results to Eq. (2) with a quadratic frequency shift λ~∣φ∣2\tilde{\lambda}|\varphi|^2 added [11], i.e.

[(ω−ω0+λ~∣φ∣2)2+Γtot2]⋅∣φ∣2=(B0Blinear)2Alinear2.(4)\left[(\omega-\omega_0+\tilde{\lambda}|\varphi|^2)^2+\Gamma_{\mathrm{tot}}^2\right]\cdot|\varphi|^2=\left(\frac{B_0}{B_{\mathrm{linear}}}\right)^2 A_{\mathrm{linear}}^2. \tag*{(4)}

The fit yields λ~=−7.229×10−13 GHz A−2 m2\tilde{\lambda} = -7.229 \times10^{-13}\ \mathrm{GHz}\ \mathrm{A}^{-2}\ \mathrm{m}^{2}, with the fitted curves shown in Figure 2 (a) and the magnitude of nonlinear shift of the edge mode’s resonance frequency shown as a function of B0B_0 in Figure 2 (b).

Fitted local mode amplitude and nonlinear frequency shift

Figure 2. (a) The local mode amplitude ∣φ∣|\varphi| is shown as a function of the frequency detuning, f−f0f-f_0. The coloured solid lines show the data obtained from micromagnetic simulations for different values of the amplitude B0B_0 of the driving field. The black dashed lines show their fits to Eq. (4). (b) The nonlinear frequency shift λ~∣φ∣2/2π\tilde{\lambda}|\varphi|^2/2\pi obtained from micromagnetic simulations (black circles) is shown as a function of B0B_0. The blue dashed line is a cubic spline fit to the data.

Figure 2 shows that the sign of the nonlinearity is negative, so that the edge mode resonance is shifted towards higher frequencies (‘blue shift’). This contrasts with our previous findings for a stripe-shaped chiral magnonic resonator [9], which showed a positive nonlinearity (red shift) for both the fundamental and first dark modes. Using the methodology described above, we obtain λ~=8.653×10−12 GHz A−2 m2\tilde{\lambda} = 8.653 \times10^{-12}\ \mathrm{GHz}\ \mathrm{A}^{-2}\ \mathrm{m}^{2} for the stripe-based system from Ref. [9] (see the Supplementary materials [27]). The origin of the sign difference of the nonlinearity lies in the nature of the edge mode. Its frequency is dominated by the effect of the static (rather than dynamic, as for the stripe’s modes in Ref. [9]) internal field, i.e. the sum of the bias and demagnetising fields. Hence, as the amplitude of precession increases, this static demagnetising field decreases, and so, both the static internal field and the mode frequencies increase. The edge mode is confined in the ’pockets’ of this internal field [17] and is therefore most sensitive to its change. The nonlinear model agrees well with the data up to the excitation amplitude B0=0.6B_0 = 0.6 mT (Figure 2 (a)). At higher values of B0B_0, the slope of the curve decreases and then the sign of the nonlinearity reverses (not shown). This behaviour was not observed in Ref. [9] and is not captured by Eq. (4). Accounting for the presence of additional nonlinearities (likely explaining this behaviour) is beyond the scope of this paper.

The key attribute of the disk’s resonant nonlinear response that makes it a viable magnonic neuron is the strong nonlinear frequency shift achieved at modest excitation field amplitudes (B0≤0.6B_0 \le0.6 mT). Notably, when fields of such a strength are induced by spin waves propagating in the YIG medium (e.g. under the disk), the response of the YIG medium itself is still linear. Not only does this satisfy the assumption of the resonant chiral scattering model [9] but also minimises propagation losses due to nonlinear damping. This contrasts e.g. with Ref. [5] where two orders of magnitude stronger excitation fields were used.

Let us now consider the effect of the nonlinearity of the disk’s edge mode on the scattering of incident spin waves. We adopt the geometry shown in Figure 1 (a). The disk is placed at the location (−0.33-0.33 μ\mum, 0 μ\mum) with the origin defined in the middle of the simulation domain. The spatial dependence of the dynamic field in the three two-dimensional Gaussian spin-wave sources I1I_1, I2I_2, and I3I_3 is given by

In(x,y)=exp⁡(−(x−xn)2+(y−yn)22σ2),(5)I_n(x,y) = \exp\left(-\frac{(x-x_n)^2+(y-y_n)^2}{2\sigma^2}\right), \tag*{(5)}

where n=1,2,3n = 1,2,3, x1=x2=x3=−2.7x_1 = x_2 = x_3 = -2.7 μ\mum, y1=0.6y_1 = 0.6 μ\mum, y2=0y_2 = 0 μ\mum, y3=−0.6y_3 = -0.6 μ\mum, and the value of σ\sigma corresponds to the full width at half maximum (FWHM) of 0.35 μ\mum. Such a small FWHM ensures that spin waves are emitted into a wide angle. Each source is driven as described by Eq. (1) for the same excitation amplitude B0B_0. For each combination of B0B_0 and excitation frequency, we compute patterns of scattered spin waves and their stray dynamic magnetic field at a height of 40 nm above the top surface of the YIG film. The patterns of the xx component of this field are presented in Figure 3. Each pattern is normalised by the field’s strength BIB_I obtained for the location of the disk in a reference simulation run under identical conditions but without the disk. The field patterns from the same reference simulations (Figure 3 (a) and (Figure 3e)) are also used to select the disk’s location so that it corresponds to a region of enhanced dynamic magnetic field. This enhancement results from the constructive interference of spin waves emitted from the three point-like sources and forming caustic beams due to the anisotropy of the magnonic dispersion relation in the in-plane magnetised YIG medium [29, 30].

Figure 3. The patterns of the stray magnetic field, ∣Bx∣\lvert B_x\rvert, created by the scattered spin waves above the YIG medium and normalised by the reference magnetic field BIB_I are shown for different excitation field amplitudes B0B_0. Panels (a) to (d) correspond to the spin-wave frequency equal to that of the edge mode (1.935 GHz), while panels (e) to (h) correspond to 2.08 GHz, i.e. just above the edge-mode resonance. Panels (a) and (e) exemplify reference patterns from which BIB_I values are extracted for each frequency and excitation amplitude B0B_0. The positions of the spin-wave sources and the disk are indicated by the white and black circles, respectively.

Turning to simulations with the disk present, we look first at the results obtained with the lowest excitation amplitude of 0.18 mT for the spin-wave frequencies of 1.935 GHz (Figure 3 (b)) and 2.08 GHz (Figure 3 (f)). In contrast to the reference simulations, we see a stark difference between the patterns obtained for the two frequencies behind the disk within the 60° cone around the horizontal centre line. At this excitation strength, the spin waves’ stray magnetic field is below 0.33 mT, for which the nonlinear frequency shift of the edge-mode resonance is significant yet smaller than the edge-mode’s half-linewidth (Figure 2). So, at 1.935 GHz (Figure 3 (b)), the incident spin waves are still at resonance with the edge mode, and their scattering is strong and accompanied by a phase shift of around π\pi [9, 11]. This phase shift leads to the destructive interference and associated suppression of the spin waves’ stray field behind the disk. As the frequency increases to 2.08 GHz (Figure 3 (f)), the strength of the resonant scattering is reduced and the phase shift deviates from π\pi, weakening the field suppression.

As we increase the excitation strength to 0.38 mT, the positive shift of the edge mode frequency (Figure 2) tunes it into resonance with the incident spin waves at 2.08 GHz, leading to the suppression of spin-wave transmission observed in Figure 3 (g). The resonance is frustrated again at the excitation amplitude of 0.64 mT, leading to the partial recovery of the transmission (Figure 3 (h)). In comparison, the enhancement of the spin-wave transmission at 1.935 GHz as the excitation amplitude is increased to 0.38 mT (Figure 3 (c)) and then 0.64 mT (Figure 3 (d)) is relatively slow. This is due to the nonlinearity-induced asymmetry of the resonant line shape evident from Figure 2 (a). Nonetheless, the transmission pattern at 0.64 mT (Figure 3 (d)) is very similar to that when the disk is absent altogether (Figure 3 (a)). That is, we reach the effective ‘activation’ state of our ‘neuron’, such that the spin wave ‘signal’ is transmitted almost unperturbed.

In a feed-forward neural network, each magnonic neuron must not only respond nonlinearly to the incident spin waves but must also be able to ‘talk’ to neurons in the network’s next layer. This demands that the amplitude of the stray field induced by the scattered wave be strong enough to activate those secondary neurons. To illustrate how our system satisfies this requirement, we map the data from Figure 3 into the “secondary frequency shift”, i.e. the nonlinear frequency shift induced in a secondary neuron by the stray field produced in its position by the scattered spin waves (Figure 4). This is done by equating the excitation amplitude from Figure 2 (b) to the stray-field amplitude in Figure 3. We see that at the excitation amplitude of 0.18 mT, the secondary frequency shift remains <25 MHz for 1.935 GHz (Figure 4 (a)), while being only marginally greater for 2.08 GHz (Figure 4 (d)). At the medium excitation amplitudes, the secondary frequency shift exceeds 100 MHz (Figure 4 (b) and (e)), being especially strong along the caustic beams, another consequence of the anisotropic spin-wave dispersion [29, 30]. At the highest excitation amplitude (Figure 4 (c) and (f)), the scattered wave is strong enough to produce secondary frequency shifts of 150 MHz, with a wider angle and so, potentially a greater number of secondary neurons covered for the frequency of 1.935 GHz (Figure 4 (c)). At both frequencies, the spatial nonuniformity of the scattered-wave patterns provides a range of locations where placement of secondary disks is possible. By placing disks in particular positions, we can ‘weight’ their responses to the wave scattered from the primary disk. Such a disk array would then act as a multi-level neural network designed to perform a specific function, e.g. a logic gate or a gate array.

Spatial distributions of the nonlinear frequency shift induced in a secondary disk

Figure 4. Spatial distributions of the nonlinear frequency shift λ~∣φ∣2\tilde{\lambda}|\varphi|^2 that would be induced in a secondary disk placed in the scattered wave of the primary disk. Panels (a)–(c) and (d)–(f) correspond to the spin-wave frequencies of 1.935 GHz and 2.08 GHz, respectively. In each panel, the position of the primary disk is indicated by a green circle, while the labels indicate the excitation amplitude of the spin-wave sources.

Figure 5 shows the stray magnetic field BTB_{\mathrm{T}} and corresponding secondary frequency shift produced by the scattered spin waves at a specific location of (0.76 μm, ± 0.35 μm). The curves show the familiar behaviour of activation and power-limiting [9] for spin waves at the resonance frequency (Figure 5 (a)) and just above (Figure 5 (b)). In Figure 5 (a), we see a sigmoid-like curve where the secondary frequency shift plateaus around BI=1B_{\mathrm{I}} = 1 mT. In Figure 5 (b), the secondary frequency shift increases and then passes through a point of inflection at around BI=0.5B_{\mathrm{I}} = 0.5 mT, before increasing again. The field reaches 50-60% of the level due to the incident spin waves, and the corresponding secondary frequency shift exceeds 0.1 GHz, suggesting a possibility of activating a hypothetical secondary neuron at the considered location and therefore of building artificial neural networks from magnonic proposed here. Such neural networks and their properties are however beyond the topic of this paper.

Plots of secondary frequency shift and scattered spin-wave stray field amplitude as functions of incident spin-wave amplitude

Figure 5. The secondary frequency shift λ~∣φ∣2\tilde{\lambda}|\varphi|^2 (left axis) and the scattered spin-wave stray field amplitude BTB_{\mathrm{T}} (right axis) at the locations of (0.76 μm, ± 0.35 μm) are shown for the frequencies of (a) 1.935 GHz and (b) 2.08 GHz, as a function of the amplitude BIB_{\mathrm{I}} of the incident spin waves. The solid circles show the data derived from micromagnetic simulations, whilst the solid lines are cubic spline fits of the data.

In summary, we have used micromagnetic simulations to demonstrate a two-dimensional chiral magnonic resonator as a magnonic neuron. Its functionality is based on the nonlinear resonant scattering of propagating spin waves from the edge spin-wave mode confined in a permalloy disk located above a low-damping YIG magnonic medium. Depending on the frequency and amplitude of the incident spin waves, we observe complex patterns of the scattered waves and show that their dynamic dipolar field could be strong enough to activate hypothetical magnonic neurons in the next layer of the network. Our results therefore suggest that the proposed magnonic neurons could serve as viable building blocks of future 2D artificial neural networks.

The research leading to these results has received funding from the UK Research and Innovation (UKRI) under the UK government’s Horizon Europe funding guarantee (Grant No. 10039217) as part of the Horizon Europe (HORIZON-CL4-2021-DIGITAL-EMERGING-01) under Grant Agreement No. 101070347 (MANNGA project). Co-funded by the European Union. Views and opinions expressed are however those of the authors only and do not necessarily reflect those of the European Union or the European Health and Digital Executive Agency (HADEA). Neither the European Union nor the granting authority can be held responsible for them.

References

  1. [1]V. V. Kruglyak, S. O. Demokritov, and D. Grundler, Magnonics, J. Phys. D: Appl. Phys. 43, 264001 (2010).
  2. [2]A. G. Gurevich and G. A. Melkov, Magnetization Oscillations and Waves (Chemical Rubber Corporation, New York, 1996).
  3. [3]A. Prabhakar and D. D. Stancil, Spin Waves (Springer New York, 2009).
  4. [4]A. V. Chumak, P. Kabos, M. Wu, C. Abert, C. Adelmann, A. O. Adeyeye, J. Åkerman, F. G. Aliev, A. Anane, A. Awad, C. H. Back, A. Barman, G. E. W. Bauer, M. Becherer, E. N. Beginin, V. A. S. V. Bittencourt, Y. M. Blanter, P. Bortolotti, I. Boventer, D. A. Bozhko, S. A. Bunyaev, J. J. Carmiggelt, R. R. Cheenikundil, F. Ciubotaru, S. Cotofana, G. Csaba, O. V. Dobrovolskiy, C. Dubs, M. Elyasi, K. G. Fripp, H. Fulara, I. A. Golovchanskiy, C. Gonzalez-Ballestero, P. Graczyk, D. Grundler, P. Gruszecki, G. Gubbiotti, K. Guslienko, A. Haldar, S. Hamdioui, R. Hertel, B. Hillebrands, T. Hioki, A. Houshang, C.-M. Hu, H. Huebl, M. Huth, E. Iacocca, M. B. Jungfleisch, G. N. Kakazei, A. Khitun, R. Khymyn, T. Kikkawa, M. Kläui, O. Klein, J. W. Kłos, S. Knauer, S. Koral­tan, M. Kostylev, M. Krawczyk, I. N. Krivorotov, V. V. Kruglyak, D. Lachance-Quirion, S. Ladak, R. Lebrun, Y. Li, M. Lindner, R. Macêdo, S. Mayr, G. A. Melkov, S. Mieszczak, Y. Nakamura, H. T. Nembach, A. A. Nikitin, S. A. Nikitov, V. Novosad, J. A. Otálora, Y. Otani, A. Papp, B. Pigeau, P. Pirro, W. Porod, F. Porrati, H. Qin, B. Rana, T. Reimann, F. Riente, O. Romero-Isart, A. Ross, A. V. Sadovnikov, A. R. Safin, E. Saitoh, G. Schmidt, H. Schultheiss, K. Schultheiss, A. A. Serga, S. Sharma, J. M. Shaw, D. Suess, O. Surzhenko, K. Szulc, T. Taniguchi, M. Urbánek, K. Usami, A. B. Ustinov, T. van der Sar, S. van Dijken, V. I. Vasyuchka, R. Verba, S. Viola Kusminskiy, Q. Wang, M. Weides, M. Weiler, S. Wintz, S. P. Wolski, and X. Zhang, Roadmap on spin-wave computing, IEEE Trans. Magn. 58, 0800172 (2022).DOI
  5. [5]Á. Papp, W. Porod, and G. Csaba, Nanoscale neural network using non-linear spin-wave interference, Nature Commun. 12, 6422 (2021).DOI
  6. [6]S. Watt, M. Kostylev, A. B. Ustinov, and B. A. Kalinikos, Implementing a magnonic reservoir computer model based on time-delay multiplexing, Phys. Rev. Appl. 15, 064060 (2021).DOI
  7. [7]L. Körber, C. Heins, T. Hula, J.-V. Kim, S. Thlang, H. Schultheiss, J. Fassbender, and K. Schultheiss, Pattern recognition in reciprocal space with a magnon-scattering reservoir, Nature Commun. 14, 3954 (2023).DOI
  8. [8]N. Zenbaa, C. Abert, F. Majcen, M. Kerber, R. O. Serha, S. Knauer, Q. Wang, T. Schrefl, D. Suess, and A. V. Chumak, A universal inverse-design magnonic device, Nature Electron. 8, 106 (2025).DOI
  9. [9]K. G. Fripp, Y. Au, A. V. Shytov, and V. V. Kruglyak, Nonlinear chiral magnonic resonators: Toward magnonic neurons, Appl. Phys. Lett. 122, 172403 (2023).DOI
  10. [10]Y. Au, M. Dvornik, O. Dmytriev, and V. V. Kruglyak, Nanoscale spin wave valve and phase shifter, Appl. Phys. Lett. 100, 172408 (2012).DOI
  11. [11]K. G. Fripp, A. V. Shytov, and V. V. Kruglyak, Spin-wave control using dark modes in chiral magnonic resonators, Phys. Rev. B 104, 054437 (2021).DOI
  12. [12]V. V. Kruglyak, Chiral magnonic resonators: Rediscovering the basic magnetic chirality in magnonics, Appl. Phys. Lett. 119, 200502 (2021).DOI
  13. [13]S. Geller and M. A. Gilleo, Structure and ferrimagnetism of yttrium and rare-earth iron garnets, Acta Cryst. 10, 239 (1957).DOI
  14. [14]V. Cherepanov, I. Kolokolov, and V. Lvov, The saga of YIG: spectra, thermodynamics, interaction and relaxation of magnons in a complex magnet, Phys. Reps. 229, 81 (1993).DOI
  15. [15]H. Qin, R. B. Holländer, L. Flajšman, F. Hermann, R. Dreyer, G. Woltersdorf, and S. van Dijken, Nanoscale magnonic Fabry-Pérot resonator for low-loss spin-wave manipulation, Nature Commun. 12, 2293 (2021).DOI
  16. [16]O. A. Santos and B. J. van Wees, Magnon Confinement in an All-on-Chip YIG Cavity Resonator Using Hybrid YIG / Py Magnon Barriers, Nano Lett. 23, 9303 (2023).
  17. [17]V. V. Kruglyak, A. Barman, R. J. Hicken, J. R. Childress, and J. A. Katine, Picosecond magnetization dynamics in nanomagnets: Crossover to nonuniform precession, Phys. Rev. B 71, 220409 (2005).DOI
  18. [18]R. Rückriem, P. Krone, T. Schrefl, and M. Albrecht, Spin dynamics of magnetic nanostructures investigated by micromagnetic simulations, Appl. Phys. Lett. 100, 242402 (2012).DOI
  19. [19]Assumed magnetic parameters of permalloy: the saturation magnetisation of Mₛ = 800 kA / m, the exchange constant of Aₑₓ = 13 pJ / m, the Gilbert damping constant of 0.005, and zero anisotropy.
  20. [20]Assumed magnetic parameters of YIG: the saturation magnetisation of Mₛ = 140 kA / m, the exchange constant of Aₑₓ = 3.5 pJ / m, the Gilbert damping constant of 0.0005, and zero anisotropy.
  21. [21]A. Vansteenkiste, J. Leliaert, M. Dvornik, M. Helsen, F. Garcia-Sanchez, and B. van Waeyenbergе, The design and verification of MuMax3, AIP Adv. 4, 107133 (2014).DOI
  22. [22]P. S. Keatley, V. V. Kruglyak, A. Neudert, E. A. Galaktionov, R. J. Hicken, J. R. Childress, and J. A. Katine, Time-resolved investigation of magnetization dynamics of arrays of nonellipsoidal nanomagnets with nonuniform ground states, Phys. Rev. B 78, 214412 (2008).DOI
  23. [23]J. M. Shaw, T. J. Silva, M. L. Schneider, and R. D. McMichael, Spin dynamics and mode structure in nanomagnet arrays: Effects of size and thickness on linewidth and damping, Phys. Rev. B 79, 184404 (2009).DOI
  24. [24]H. T. Nembach, J. M. Shaw, T. J. Silva, W. L. Johnson, S. A. Kim, R. D. McMichael, and P. Kabos, Effects of shape distortions and imperfections on mode frequencies and collective linewidths in nanomagnets, Phys. Rev. B 83, 094427 (2011).
  25. [25]R. E. Camley, B. V. McGrath, Y. Khivintsev, Z. Celinski, R. Adam, C. M. Schneider, and M. Grim्सditch, Effect of cell size in calculating frequencies of magnetic modes using micromagnetics: Special role of the uniform mode, Phys. Rev. B 78, 024425 (2008).
  26. [26]O. Dmytriev, M. Dvornik, R. V. Mikhaylovskiy, M. Franchin, H. Fanghor, L. Giovannini, F. Montoncello, D. V. Berkov, E. K. Semenova, N. L. Gorn, A. Prabhakar, and V. V. Kruglyak, Calculation of high-frequency permeability of magnonic metamaterials beyond the macrospin approximation, Phys. Rev. B 86, 104405 (2012).
  27. [27]See Supplementary materials for the description of the ways the periodic and absorbing boundary conditions were implemented in different situations, and for the results of calculation of the nonlinearity parameter for the stripe-based chiral magnonic resonator from Ref. [9].
  28. [28]N. J. Whitehead, S. A. R. Horsley, T. G. Philbin, and V. V. Kruglyak, A Luneburg lens for spin waves, Appl. Phys. Lett. 113, 212404 (2018).DOI
  29. [29]V. Veerakumar and R. E. Camley, Magnon focusing in thin ferromagnetic films, Phys. Rev. B 74, 214401 (2006).DOI
  30. [30]E. H. Lock, The properties of isofrequency dependences and the laws of geometrical optics, Phys.-Usp. 51, 375 (2008).DOI
  31. [31]T. Yu and G. E. W. Bauer, Chiral coupling to magnetodipolar radiation, Top. Appl. Phys. 138, 1 (2021).arxiv.org/abs/2001.06821

Paper details

vlnc:W_d7msx9kb
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 2

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents