Introduction

Two-dimensional electronic systems exhibit a rich variety of quantum phenomena1,2. The advent of graphene has not only provided a way to study these phenomena in the quasi-relativistic spectrum, but has also extended their experimental range3,4, made some observations much clearer5,6,7,8 and, of course, revealed many more effects9,10,11,12. These advances are mostly due to graphene’s intrinsically high carrier mobility that is preserved by state-of-the-art heterostructure engineering in which graphene is encapsulated between hexagonal boron nitride layers13,14 and electrically tuned with atomically smooth metallic gates8,15. Nonetheless, one of the first discoveries in quantum transport, well known for over 50 years16,17, has remained conspicuously absent in graphene–magnetophonon oscillations18,19.

In the presence of an applied magnetic field (B), electrons in pristine crystals become localised in closed orbits and their spectra take the form of quantised Landau levels (LLs) separated by energy gaps. However, an electrical current can still flow in the bulk due to carriers resonantly scattering between neighbouring orbits by the absorption or emission of phonons with energies equal to the LL spacing19 (Fig. 1a). In a semiclassical model, the resonant transitions occur between orbits which just touch in real space and induce figure of eight trajectories20 (Fig. 1b), corresponding quantum mechanically to strong overlap of the tails of their wave functions in the vicinity of their classical turning point. This effect, known as magnetophonon resonance (MPR) causes magnetoresistance oscillations that are periodic in inverse magnetic field19,21. Whereas magnetophonon oscillations have been used extensively for studying carrier–phonon interactions in bulk Si and Ge22, semiconducting alloys18 and heterostructures23,24,25, there have been no reported observations in any van der Waals crystal, not even graphene, despite its exceptional electronic quality.

Fig. 1
figure 1

Size-dependent magnetoresistance oscillations in mesoscopic graphene devices. a Landau level spectra of graphene. The diagram illustrates a carrier with momentum k1 (black sphere) making a transition between Landau levels (blue and red rings) by resonant absorption of a phonon (brown arrow) with momentum q = |k1k2| and energy ħωq. Solid black arrows represent the magnitudes of wave vectors k1, k2 and q. b The semiclassical motion of a carrier (black sphere) in real space for the resonance condition sketched in a. The red and blue circles which touch at a tangent point represent the initial and final semiclassical cyclotron orbits of two Landau-quantised states between which an electron is scattered by a phonon. During resonant scattering, the carriers follow a path that resembles the number 8 (figure of eight trajectory). The red and blue arrows show the motion of the charge carriers along this trajectory. The green arrow indicates direction of the applied current I. c Optical image of a graphene device with W = 15 μm. The edges of the mesa are indicated by the white solid line. d Open circles plot experimentally determined Drude mean free path LMFP as a function of W for two T and fixed n = −2 × 1012 cm−2. Black dashed line plots the equation LMFP = W. Solid green line marks the phonon-limited mean free path (Le–ph) at 50 K. Our measurements focussed on the valence band because our wide devices exhibited higher electronic quality for hole doping (e), longitudinal magnetoresistance data Rxx (B) for fixed n = −3.3 × 1012 cm−2 measured in our wide device (c) at two different T. The red arrows indicate peaks that are caused by magnetophonon resonance (MPR). The 50 K curve is off-set vertically for clarity. Inset: Hall resistance Rxy(B) measured simultaneously as Rxx. The solid blue and dashed red lines are data measured at 5 and 50 K respectively. f Rxx/Rxx (B = 0T) measured at fixed n and T in several devices of different W. The shaded area close to B= 0 contains semiclassical effects4

In this article, we consider a subtle yet crucial aspect concerning the design of electronic devices based on graphene, namely the lateral size of the conducting channel. It has so far remained small, only a few micrometres in most quantum transport experiments. Our measurements using graphene Hall bars of different widths show that wider samples start exhibiting pronounced magnetophonon oscillations.

Results

Phonon scattering in wide graphene channels

Our experiments involved magnetotransport measurements on graphene Hall bars encapsulated by hexagonal boron nitride, with particular attention paid to ‘wide’ devices with channel widths W > 10 μm. An optical image of one of our widest devices is shown in Fig. 1c (see Supplementary Note 1 for details of device fabrication). Because the electron–phonon coupling is so weak in graphene26, charge carriers scatter more frequently at the device edges of micron-sized samples rather than with phonons in the bulk, especially at low temperature27 (T). This is evident when comparing the Drude mean free path (LMFP) for devices of different W and a fixed carrier density (n) of holes (Fig. 1d). At 5 K, all devices exhibit size-limited mobility (LMFP > W) because carriers propagate ballistically until they collide with the edges of the conducting channel. Even at 50 K, scattering is still dominated by the edges in most of our devices and LMFP increases linearly with W. However, at these higher temperatures we find that LMFP saturates around 8 μm (green line in Fig. 1d) and does not increase upon further widening of the device channel. This saturating behaviour tells us that LMFP is no longer dependent on the device width and carriers scatter mostly with phonons in the bulk (LMFP < W). In effect, widening the channel makes our measurement more sensitive to bulk phenomena rather than edge effects.

Width-dependent magnetoresistance oscillations

The main observation of our work is presented in Fig. 1e, which plots the longitudinal resistance (Rxx) of a 15 μm wide Hall bar (Fig. 1c) as a function of B, at two T and fixed n. At 5 K we observe two distinct oscillatory features. The first, at relatively low B < 0.2 T, are the well-established semiclassical geometrical resonances that occur due to magnetic focussing of carriers between current and voltage probes4 (see Supplementary Note 2, Supplementary Fig. 2). At higher B (1T), quantised cyclotron orbits are formed and we observe 1/B-periodic Shubnikov de Haas (SdH) oscillations. Their origin is confirmed by noting that the charge carrier density n = 4e/(hΔ(B−1)) extracted from the SdH period (Δ(B−1)) agrees with that determined by Hall effect measurements (Fig. 1e, inset). At 50 K, the low-field geometric oscillations remain visible although their relative amplitudes are suppressed due to the reduced carrier mean free path. However, at higher |B| > 0.2T, an additional set of oscillations appears with five clear maxima (indicated by red arrows in Fig. 1e). These high T oscillations are also periodic in 1/B but are distinguished by their markedly slower period. In contrast to Rxx, the Hall resistance, Rxy, shows no oscillatory features and has the same value at both T (Fig. 1e, inset), confirming that n does not change upon warming the sample.

The observation of the high T oscillations depends critically on the sample width. This is shown in Fig. 1f which plots the normalised magnetoresistance, Rxx/Rxx (B = 0T), for devices with different W at fixed T and n. We note that the bulk channels in all our devices are intrinsically clean and free from defects (probed by ballistic transport experiments in Supplementary Note 3, Supplementary Fig. 3). Nonetheless, whereas these oscillations are well developed in the widest devices (resonances marked by purple arrows), they are poorly resolved for devices with W < 8 μm and completely absent in the narrowest one (W = 1.5 μm). As described below, we identify these size dependent, high T oscillations with MPR

A defining feature of magnetophonon oscillations is their unique non-monotonic temperature dependence, in which their amplitude first increases with T and then decays25. Figure 2a shows the temperature dependence of Rxx (B) for fixed n between 5 and 100 K (5 K steps) for another wide Hall bar device (W = 15 μm). In this sample, weak magnetophonon oscillations already appear at 5 K in the field range between the geometric and the SdH oscillations. The resonances are labelled p = 1 to 5, where the integer p refers to the number of LL spacings that are crossed during the transition; p = 1 corresponds to scattering between LLs adjacent in energy (Fig. 1a). With increasing T, the magnetophonon oscillations become more pronounced as more phonons are thermally activated, while the SdH oscillations are strongly suppressed. Although both phenomena require carriers that exhibit coherent cyclotron orbits (μB > 1, where μ is the carrier mobility), MPR is not obscured by smearing of the Fermi–Dirac distribution across Landau gaps25; rather it is enhanced due to an increased number of unoccupied states into which carriers can scatter. Hence magnetophonon oscillations persist to higher T than SdH oscillations. However, they are eventually damped at high enough T (Fig. 2b) when LLs become broadened by additional scattering (μB 1). This non-monotonic behaviour is better visualised in Fig. 2c which plots the oscillatory amplitudes (ΔRxx) as a function of T. Notably, the amplitude of all resonances peak at T below 60 K, corresponding to a thermal energy of a few meV.

Fig. 2
figure 2

Temperature dependence of the magnetophonon effect. a Magnetoresistance Rxx (B) for T between 5 K (blue curve) and 100 K (black curve) in 5 K steps for fixed n measured in another Hall bar with W = 15 μm. b Extended data set of a showing high T behaviour (10 K steps). c Temperature dependence of the amplitude of MPR peaks, ΔRxx (T), indicated in a by colour coded letters, p = 1–5

MPR spectroscopy

For the doping levels and B-fields at which the oscillations occur, the charge carriers occupy high-index LLs (N 20 for p = 1) separated by small energy gaps (5 meV) with a classical cyclotron radius up to RcħkF/eB 300 nm, where kF is the Fermi-wave vector. Resonant inter-LL transitions occur due to inelastic scattering by low-energy acoustic phonons that induce figure of eight trajectories (Fig. 1b). This type of trajectory occurs with high probability because the wave functions of the initial (blue circle in Fig. 1b) and final states (red circle) have a large spatial overlap where they touch in real space24. During figure of eight trajectories, the velocity of the carrier is reversed at the intersection of the two cyclotron orbits (see arrows in Fig. 1b). This process requires a phonon of specific momentum q ≈ 2kF 109 m−1 and energy ħωq (2kF) 5 meV that can back-scatter the carriers during the inter-LL transition. Energy and momentum conservation for such a process requires that EN+p − EN = ħωq (2kF), where EN is the energy of an electron in the Nth LL, so that resonances occur at B values given by

$$B_{p} = \frac{{nhv_{\mathrm{s}}}}{{pev_{\mathrm{F}}}}$$
(1)

(see Supplementary Note 4 for a detailed derivation using the semiclassical model). Here, vF and vs are the Fermi velocity and low-energy acoustic phonon velocity in graphene respectively. This resonant condition is unique to massless Dirac electrons and is strikingly different to the case of massive electrons in a conventional two-dimensional electron gas (2DEG) system24 where Bp scales with n0.5. On resonance, inelastic scattering between neighbouring orbits (Fig. 1b) gives rise to a finite and dissipative current in the bulk. This behaviour causes maxima in ρxx at Bp; the 1/B periodicity results in oscillations described by Δρxx eγ/Bcos(2πBF/B) where BFpBp and the factor γ depends on temperature28. Equation (1) predicts that the position of maxima scales linearly with n. With this in mind, Fig. 3a, b plot maps of Rxx (n, B) for one of our 15 μm devices at 5 K (Fig. 3a) and 50 K (Fig. 3b). In addition to the typical Landau fan structure that is dominant at low T (filling factors, ν, are marked by blue dashed lines), the maps reveal a broader set of fans at lower B that are more prominent at 50 K (Fig. 3b). They are caused by MPR (p values are labelled in red) and demonstrate that their frequency scales linearly with n. Furthermore, the positions of MPR peaks in Fig. 3b can be fitted precisely by Eq. (1) (red dashed lines) with a constant vs/vF = 0.0128. By studying the temperature dependence of SdH oscillations in our graphene devices (Supplementary Note 5, Supplementary Fig. 4), we extract vF and determine vs accordingly. We note that vF shows no significant dependence on n, as expected for graphene devices on hexagonal boron nitride at high doping29, because e–e interactions that cause velocity renormalisation30 are heavily screened. Hence, using the extracted vF = 1.06 + 0.05 × 106 ms−1, we determined a phonon velocity, vs = 13.6 + 0.7 km s−1. This value is close to the speed of transverse acoustic (TA) phonons in graphene (13 km s−1) calculated in numerous theoretical works31,32,33,34,35. Therefore, we infer our oscillations arise from inter-LL scattering by low energy and linearly dispersed TA phonons.

Fig. 3
figure 3

Density dependence of magnetophonon oscillations. a Longitudinal resistance Rxx (grey scale map) as a function of n and B measured at 5 K (W = 15 μm). Logarithmic grey scale: white: 1 Ω to black: 15 Ω. The blue dashed lines trace Landau gaps corresponding to high filling factors ν = nh/Be. b Same as a measured at 50 K. Logarithmic grey scale: white: 5.5 Ω to black: 18 Ω. The red dashed lines plot Eq. (1) for p = 1 to 4 which corresponds to carriers scattering with transitions across 1 to 4 Landau level spacings. Features appearing for B < 0.2 T are the semiclassical geometrical oscillations4 not relevant in this work (see Supplementary Note 2 for details)

Equation (1) is generic for linearly dispersed phonons in graphene. This motivated us to search for MPR arising from longitudinal acoustic (LA) phonons, which should occur at higher B due to their significantly higher vs34,35. Careful inspection of the data in Figs. 1 and 2 shows that the p= 1 resonance for TA phonons is followed by a weak shoulder-like feature at higher B. We therefore studied a dual-gated graphene device (Supplementary Fig. 1) that permitted measurements at higher n −1 × 1013 cm−2 which, according to Eq. (1), should better separate this feature from the TA resonances. Figure 4a plots Rxx (B) for this device for several n. Measurements at these high n reveals that the shoulder-like feature develops into a well-defined peak (indicated by coloured arrows). Its position (Bp=1) is accurately described by Eq. (1) with a constant vs/vF = 0.0198. Using the experimentally extracted value of vF, we obtained vs = 21.0 + 1.0 km s−1. This value is indeed close to that calculated for LA phonons in graphene31,34,35, and hence we attribute this feature to inter-LL scattering by LA phonons.

Fig. 4
figure 4

Phonon spectroscopy in graphene by measurement of magnetophonon oscillations. a Longitudinal resistance Rxx as a function of B measured for several high n of holes in our wide (W = 13.8 μm) dual-gated graphene Hall bar. The curves have been off-set vertically for clarity. b Red symbols plot the fundamental frequency BFpBp of magnetophonon oscillations as a function of absolute n for three different devices; open circles correspond to the dual-gated device which allowed high doping. The blue symbols mark the positions Bp=1 of the broad peak (indicated by coloured arrows in a) which appears clearly at high n. The red and blue solid lines represents Eq. (1) with vs/vF = 0.0128 and 0.0198 respectively. Knowing vF (Supplementary Note 5) we extract the TA (vTA) and LA (vLA) phonon velocities. Inset: Data in main panel transformed to phonon dispersion curves. Coloured squares−experimental data points (error bars reflect the error in the experimentally extracted vF), solid lines plot the equation ħωq = ħvsq (same vs as in the main panel), purple stars—data taken from ref. 36. c Calculation of the oscillatory part of the resistivity Δρxx(Ω) using the Kubo formula (see Supplementary Note 6 for details)

Further validation of our model is presented in Fig. 4b, which plots the magnetophonon oscillation frequency for TA phonons (BFpBp) as a function of n for several different devices (red symbols). It shows that a linear dependence (red line) fits the data to Eq. (1) for all our measured devices over a range of n spanning an order of magnitude. The weaker LA resonance was also found to occur at the same Bp=1 = BF in different devices (blue symbols). Furthermore, the data in Fig. 4b can be transformed directly into phonon dispersion curves (inset of Fig. 4b) by noting that q ≈ 2kF = 2()0.5 and ħωq ≈ (2eBFvsvFħ)0.5. The extended tunability of the carrier density in our dual-gated devices allows measurement of phonon branches up to wave vectors >109 m−1. Note that these dispersion plots are significantly more precise than those measured by X-ray scattering experiments in graphite36 (purple stars). Studies of magnetophonon oscillations thus enable an all-electrical measurement of the intrinsic phonon dispersion curves in gate-tuneable materials.

Discussion

To understand why magnetophonon oscillations are absent in narrow samples, we first note that figure of eight trajectories (Fig. 1b) have a spatial extent 4Rc, which can reach values of several microns for the high-order resonances (p > 3). If the sample is too narrow, so that 4Rc is comparable to W, the carrier trajectories are skewed by elastic scattering at the device edges. In this case, they propagate along the edges of the device in skipping orbits2, effectively short-circuiting the resistive behaviour of the bulk caused by MPR. However, if W > 4Rc, both MPR and skipping orbits contribute to Rxx. We can estimate the width of the device required to observe MPR by comparing the relative contributions of these two processes. Carriers that diffuse in MPR-induced figure of eight trajectories move a distance 2Rc in a characteristic time, τe–ph = Le–ph/vF with a drift velocity vMPR = 2Rc/τeph. This is significantly slower than skipping orbits which can have speeds approaching vF. On the other hand, skipping orbits occupy only a width Rc at each edge, whereas MPR occurs approximately over the full width, W, of the bulk. By comparing these two contributions, we deduce that MPR dominates when WvMPR 2RcvF. This corresponds to the condition WLe–ph, in good agreement with the measured data in Fig. 1d, f.

Our measurements provide an important insight into the intrinsic electron–phonon interaction in graphene: namely, the dominance of carrier scattering by low-energy TA phonons. This is in agreement with several theoretical works35,37,38 and contrasts with a widely held view that deformation potential scattering by LA phonons prevails over TA phonons39. To investigate this point further, we calculated magnetoresistance using the Kubo formula40 (Supplementary Note 6). A typical calculation is shown in Fig. 4c, which plots the contribution (Δρxx) of MPR for TA and LA phonon velocities of vs = 13.6 and 21.4 km s−1, respectively35, and the Fermi velocity41 vF = 1 × 106 m s−1. It accurately describes the oscillatory form of the measured data. Such good agreement is only possible when our calculations include the effect of carrier screening35,38,42,43 which significantly reduces the electron–LA phonon deformation potential coupling. Without screening, LA phonons would dominate the observed MPR (Supplementary Fig. 5). Our results therefore highlight the importance of carrier screening on electron–phonon interactions and thus helps resolve a long-standing discussion of the relative importance of LA39,44 and TA37,38,43 phonon scattering in graphene.

To conclude, we report the observation of pronounced magnetophonon oscillations in graphene, where the Dirac spectrum strongly modifies the resonant condition compared to previously studied electronic systems. Other two-dimensional crystals can also be expected to exhibit this phenomenon. The oscillations enable the study of low-energy acoustic phonon modes that are generally inaccessible by Raman spectroscopy45,46. Our measurements combined with the Kubo calculations provide strong evidence that TA phonons limit temperature-dependent mobility in graphene35,37,38. Most importantly, graphene's transport properties are shown to strongly depend on device size, even for conducting channels as wide as several microns. This should motivate further experiments on graphene and related two-dimensional materials in a macroscopic regime beyond the scope of previous mesoscopic devices.

Methods

Quantum transport measurements

For measuring resistance in our graphene devices, we used standard low-frequency AC measurement techniques with a lock-in amplifier at 10–30 Hz. The measurements of Rxx(Ω) = Vxx/Ixx are obtained by driving a small AC excitation current (Ixx = 0.1–1 μA) down the length of the Hall bar while simultaneously measuring the four probe voltage drop (Vxx) between two side contacts located on the edge of the Hall bar devices (Fig. 1c). We tune the Fermi level in our graphene devices by applying a DC voltage between the silicon substrate and the graphene channel, where the SiO2 and bottom hexagonal boron nitride encapsulation layer serve as the dielectric (see Supplementary Note 1 for details on device fabrication). In our top gated device (see Supplementary Fig. 1), we simultaneously apply a potential to the metal top gate which allowed us to reach higher doping levels (see Fig. 4). All measurements were performed inside a variable temperature inset of a wet helium-4 flow cryostat that allowed us to carry out temperature-dependent magnetotransport measurements using a cold superconducting magnet.