Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                

Quantitative measurement of viscosity in two-dimensional electron fluids

Yihang Zeng1,5    Haoyu Guo2,4    Olivia M. Ghosh1    Kenji Watanabe3    Takashi Taniguchi3    Leonid S. Levitov4    Cory R. Dean1 1Department of Physics, Columbia University, New York, NY, USA 2Laboratory of Atomic and Solid State Physics, Cornell University, Ithaca, NY, USA 3National Institute for Materials Science, Tsukuba, Japan 4Department of Physics, Massachusetts Institute of Technology, Cambridge, Massachusetts 02139, USA 5Present address: Department of Physics, Purdue University, West Lafayette, Indiana, USA
(July 6, 2024)

Electron hydrodynamics is an emerging framework that describes dynamics of interacting electron systems as conventional fluids. While evidence for hydrodynamic-like transport is reported in a variety of two-dimensional materials, precise quantitative measurement of the core parameter, electron viscosity, remains challenging. In this work, we demonstrate that magnetoresistance in Corbino-shaped graphene devices offers a simultaneous Ohmmeter/viscosometer, allowing us to disentangle the individual Ohmic and viscous contributions to the transport response, even in the mixed flow regime. Most surprising, we find that in both monolayer and bilayer graphene, the effective electron-electron scattering rate scales linearly with temperature, at odds with the expected T𝑇Titalic_T-squared dependence expected from conventional Fermi liquid theory, but consistent with a recently identified tomographic flow regime, which was theoretically conjectured to be generic for two-dimensional charged fluids.

Refer to caption
Figure 1: Magnetoresistance of graphene in the corbino disk geometry. a,Schematic of the corbino disk geometry and measurement configuration. A small opening of width ra/4absentsubscript𝑟𝑎4\approx r_{a}/4≈ italic_r start_POSTSUBSCRIPT italic_a end_POSTSUBSCRIPT / 4 is left out on the top hBN layer at the outer circumference in order for the inner electrode to pass through. Bottom schematics depict current flow lines under small perpendicular magnetic field while in the quasi ballistic (left) and hydrodynamic (right) transport regime. b Zero-magnetic-field resistance R(B=0)𝑅𝐵0R(B=0)italic_R ( italic_B = 0 ) and external resistance RExtsuperscript𝑅𝐸𝑥𝑡R^{Ext}italic_R start_POSTSUPERSCRIPT italic_E italic_x italic_t end_POSTSUPERSCRIPT as a function of temperature. The external resistance consists of both the metal-graphene contact resistance and line resistance. c,Representative magnetoresistance and its derivative against B2superscript𝐵2B^{2}italic_B start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT at T=𝑇absentT=italic_T = 130 K. The derivative is obtained by fitting a quadratic function to symmetrized magnetoresistance in a sliding window of 12 mT wide. The plateau at low field and high field show two quadratic regimes with curvature α𝛼\alphaitalic_α and αohmsuperscript𝛼𝑜𝑚\alpha^{ohm}italic_α start_POSTSUPERSCRIPT italic_o italic_h italic_m end_POSTSUPERSCRIPT respectively. Rosubscript𝑅𝑜R_{o}italic_R start_POSTSUBSCRIPT italic_o end_POSTSUBSCRIPT denotes the intercept of the high-field quadratic function. The determination of fitting range is discussed in SI. d, α𝛼\alphaitalic_α and αohmsuperscript𝛼𝑜𝑚\alpha^{ohm}italic_α start_POSTSUPERSCRIPT italic_o italic_h italic_m end_POSTSUPERSCRIPT versus temperature. Results in b-d are taken at n=0.2×1012𝑛0.2superscript1012n=0.2\times 10^{12}italic_n = 0.2 × 10 start_POSTSUPERSCRIPT 12 end_POSTSUPERSCRIPTcm-2 in a device with ra=1μsubscript𝑟𝑎1𝜇r_{a}=1\muitalic_r start_POSTSUBSCRIPT italic_a end_POSTSUBSCRIPT = 1 italic_μm, rb=4μsubscript𝑟𝑏4𝜇r_{b}=4\muitalic_r start_POSTSUBSCRIPT italic_b end_POSTSUBSCRIPT = 4 italic_μm. e,Sample resistance as a function of temperature. Black circles mark total sample resistance. Red circles show resistance from momentum-relaxing scattering. While blue circles represent the difference between black and red circles, reflect the resistance contributed by momentum-conserving scattering.

Strongly correlated electron systems in which many-body interactions drive novel ground states remain one of the most fascinating, yet difficult to understand, problems in physicsAlexandradinata et al. (2022). A particular challenge is relating the macroscale behaviour typically measured in experiment, such as transport response, with the underlying microscopic interactions and scattering processes relevant to theory. Electron hydrodynamics has emerged as a framework to to fill this gap by identifying that strong interparticle scattering can cause electron systems to behave like a conventional viscous fluidde Jong and Molenkamp (1995); Damle and Sachdev (1997); Davison et al. (2014); Forcella et al. (2014); Levitov and Falkovich (2016). The viscous flow response is parameterized by the electron viscosity and can therefore, in principle, be used as a measure of electron-electron scatteringGurzhi (1963, 1968); Müller et al. (2009). Quantifying this parameter in experiment is especially relevant in two-dimensional (2D) electron systems where theory predicts different families of excitations distinguished by quasiparticle lifetimeLedwith et al. (2019a, b); Hofmann and Gran (2023); Hofmann and Das Sarma (2022); Kryhin et al. (2023); Nilsson et al. (2024); Hong et al. (2020). These long-lived excitations are hard to detect by ordinary spectroscopy techniquesEisenstein et al. (2007); Murphy et al. (1995), however they are expected to dominate the hydrodynamic responseLedwith et al. (2019a); Kryhin et al. (2023).

Evidence for electron hydrodynamics has been identified in a variety of 2D electron systemsLucas and Fong (2018); Levchenko and Schmalian (2020). However, precise quantitative comparison between experiment and theory remains limited. This is the result of two challenges. First, electron-electron (e𝑒eitalic_e-e𝑒eitalic_e) scattering is a momentum conserving interaction that does not directly yield a measurable dissipation. Instead energy is lost through momentum-relaxing boundary scattering. Dissipation in the hydrodynamic regime is therefore determined by device geometry and boundary conditions, and only indirectly related to e𝑒eitalic_e-e𝑒eitalic_e scattering (i.e.viscosity). Second, for a material to demonstrate purely hydrodynamic flow requires a precise combination of properties, namely long mean free path for momentum-relaxing scattering (including electron-impurity, electron-phonon and Umklapp scattering) leelmrmuch-less-thansubscript𝑙𝑒𝑒subscript𝑙𝑚𝑟l_{ee}\ll l_{mr}italic_l start_POSTSUBSCRIPT italic_e italic_e end_POSTSUBSCRIPT ≪ italic_l start_POSTSUBSCRIPT italic_m italic_r end_POSTSUBSCRIPT, and device dimensions larger than leesubscript𝑙𝑒𝑒l_{ee}italic_l start_POSTSUBSCRIPT italic_e italic_e end_POSTSUBSCRIPT but smaller than lmrsubscript𝑙𝑚𝑟l_{mr}italic_l start_POSTSUBSCRIPT italic_m italic_r end_POSTSUBSCRIPT. In real devices this limit is typically only approximately reached so that both hydrodynamic and Ohmic processes contribute to the observed response, and disambiguating their relative contributions is difficult and often imprecise. Experimental progress has been realized through device-size scaling measurements Molenkamp and de Jong (1994); de Jong and Molenkamp (1995); Moll et al. (2016); Krishna Kumar et al. (2017); Ginzburg et al. (2023), non-local transportBandurin et al. (2016); Berdyugin et al. (2019); Kim et al. (2020), thermal transportTalanov et al. (2024), and direct imaging of flow profilesKu et al. (2020); Kumar et al. (2022); Sulpizio et al. (2019); Jenkins et al. (2022); Aharon-Steinberg et al. (2022); Vool et al. (2021); Braem et al. (2018). In most cases hydrodynamic parameters are deduced only in the regime where it is presumed that Ohmic scattering is minimal, restricting the available phase space and limiting the ability to deduce scaling behaviours (e.g. density and temperature dependence).

In this work we demonstrate a new approach to measuring electron viscosity in 2D systems based on magnetotransport in a Corbino geometry. The Corbino structure measures current flow betweeen concentric rings and therefore has no boundary except at the contacts. Recent theoreticalLevchenko et al. (2022); Levchenko and Schmalian (2020); Shavit et al. (2019); Holder et al. (2019) and experimentalKumar et al. (2022) work has demonstrated that the potential distribution and current-flow profiles through the Corbino disc are dramatically different in the Ohmic versus viscous regimes. We show that under mangetic field, the magnetoresistances associated with each regime (Fig. 1a) can be utilized to quantify the resistivity and viscosity, respectivelyShavit et al. (2019). Moreover, we find that it is possible to disentangle the two contributions in the mixed flow regime, enabling a powerful way to isolate the viscous parameters over a wider phase space than previously possible.

We apply this technique to the study of hydrodynamic flow in monolayer (MLG) and bilayer (BLG) graphene, and make the surprising discovery that in both cases, in the Fermi liquid regime, the electron kinetic viscosity scales with 1/T1𝑇1/T1 / italic_T and is density-independent. This violates conventional Fermi liquid behaviour but finds consistency with a newly-identified tomographic flow regime, theoretically predicted for 2D materialsLedwith et al. (2019a); Hofmann and Das Sarma (2022); Kryhin et al. (2023). We also demonstrate continuous measurement of viscosity through the crossover from the Fermi liquid regime at high density to Dirac fluid regime at low density. We show for the first time that the MLG and BLG viscosity transitions to a T2superscript𝑇2T^{2}italic_T start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT, and T1superscript𝑇1T^{1}italic_T start_POSTSUPERSCRIPT 1 end_POSTSUPERSCRIPT dependence, respectively, in the Dirac fluid regime, consistent with theoretical predicitonsMüller et al. (2009); Yudhistira et al. (2023), and confirm that in the crossover the system behaves according to a two-fluid model with e𝑒eitalic_e-e𝑒eitalic_e and e𝑒eitalic_e-hhitalic_h contributing to hydrodynamic electrical transport in two parallel channelsYudhistira et al. (2023).

Magnetoresistance in a corbino disk
Fig. 1a shows a cartoon schematic of our device structure. Graphene is encapsulated between boron nitride layers. A corbino geometry with concentric inner and outer ring electrodes is then defined lithographically using an edge contact geometryZeng et al. (2019). A local graphite bottom gate is used to tune the carrier density (see methods for more details). We investigate MLG and BLG devices, with outer radius, rbsubscript𝑟𝑏r_{b}italic_r start_POSTSUBSCRIPT italic_b end_POSTSUBSCRIPT, ranging in size from 1.75 μ𝜇\muitalic_μm to 4.5 μ𝜇\muitalic_μm, and inner radius ra=rb/4subscript𝑟𝑎subscript𝑟𝑏4r_{a}=r_{b}/4italic_r start_POSTSUBSCRIPT italic_a end_POSTSUBSCRIPT = italic_r start_POSTSUBSCRIPT italic_b end_POSTSUBSCRIPT / 4, maintaining a constant aspect ratio across all devices.

Fig. 1b plots the two-terminal resistance as a function of temperature for a device with rb=4μsubscript𝑟𝑏4𝜇r_{b}=4\muitalic_r start_POSTSUBSCRIPT italic_b end_POSTSUBSCRIPT = 4 italic_μm, measured at electron density n=0.2×1012𝑛0.2superscript1012n=0.2\times 10^{12}italic_n = 0.2 × 10 start_POSTSUPERSCRIPT 12 end_POSTSUPERSCRIPTcm-2. Interpreting the Corbino response is challenging since our measurement spans a mixed flow regime that includes both viscous and Ohmic contriubtions in addition to contact resistance, and external line resistance, all of which may vary with temperature and density. In the following we demonstrate how analysis of the magnetoresitance allows us to disentangle these individual contributions.

The Corbino magnetoresistance includes a mixture of longitudinal and Hall components giving a quadratic field dependence, R(B)=Ro+αB2𝑅𝐵subscript𝑅𝑜𝛼superscript𝐵2R(B)=R_{o}+\alpha B^{2}italic_R ( italic_B ) = italic_R start_POSTSUBSCRIPT italic_o end_POSTSUBSCRIPT + italic_α italic_B start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT. This holds true in the presence of both momentum-relaxing and momentum-conserving scatteringWeiss and Welker. (1954); Blood and Tree (1971); Wieder (1969); Shavit et al. (2019); Levchenko and Schmalian (2020), so long as the individual scattering terms and contact resistances have a negligible field dependence. Rosubscript𝑅𝑜R_{o}italic_R start_POSTSUBSCRIPT italic_o end_POSTSUBSCRIPT is the zero-field resistance, including all dissipative contributions. The quadratic coefficient, α𝛼\alphaitalic_α, arises from a combination of the Hall voltage, scattering mechanisms, and boundary conditions, and does not include external resistance contributions, allowing us to isolate dissipation across the channel only Weiss and Welker. (1954); Wieder (1969); Blood and Tree (1971); Shavit et al. (2019); Levchenko and Schmalian (2020). The coefficient can be solved for in the mixed flow regime using a combined Stokes-Ohm equation in magnetic field (see methods) to give an analytic expression that depends on both the ohmic resistivity, ρosubscript𝜌𝑜\rho_{o}italic_ρ start_POSTSUBSCRIPT italic_o end_POSTSUBSCRIPT, and shear viscosity, η𝜂\etaitalic_ηLevchenko and Schmalian (2020):

α=lnγ2πρo(ne)2[1f(γ,rb/Dν)]𝛼𝛾2𝜋subscript𝜌𝑜superscript𝑛𝑒2delimited-[]1𝑓𝛾subscript𝑟𝑏subscript𝐷𝜈\alpha=\frac{\ln\gamma}{2\pi\rho_{o}(ne)^{2}}[1-f(\gamma,r_{b}/D_{\nu})]italic_α = divide start_ARG roman_ln italic_γ end_ARG start_ARG 2 italic_π italic_ρ start_POSTSUBSCRIPT italic_o end_POSTSUBSCRIPT ( italic_n italic_e ) start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG [ 1 - italic_f ( italic_γ , italic_r start_POSTSUBSCRIPT italic_b end_POSTSUBSCRIPT / italic_D start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ) ] (1)

where f(γ,rb/Dν)𝑓𝛾subscript𝑟𝑏subscript𝐷𝜈f(\gamma,r_{b}/D_{\nu})italic_f ( italic_γ , italic_r start_POSTSUBSCRIPT italic_b end_POSTSUBSCRIPT / italic_D start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ) is a dimensionless function defined in terms of modified Bessel functions (see methods), γ=rb/ra𝛾subscript𝑟𝑏subscript𝑟𝑎\gamma=r_{b}/r_{a}italic_γ = italic_r start_POSTSUBSCRIPT italic_b end_POSTSUBSCRIPT / italic_r start_POSTSUBSCRIPT italic_a end_POSTSUBSCRIPT is the device aspect ratio, and Dν=η/ρo(ne)2subscript𝐷𝜈𝜂subscript𝜌𝑜superscript𝑛𝑒2D_{\nu}=\sqrt{{\eta}/{\rho_{o}}(ne)^{2}}italic_D start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT = square-root start_ARG italic_η / italic_ρ start_POSTSUBSCRIPT italic_o end_POSTSUBSCRIPT ( italic_n italic_e ) start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG is the Gurzhi length. We note that when Dνsubscript𝐷𝜈D_{\nu}italic_D start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT is much larger than the device size, Dνrbmuch-greater-thansubscript𝐷𝜈subscript𝑟𝑏D_{\nu}\gg r_{b}italic_D start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ≫ italic_r start_POSTSUBSCRIPT italic_b end_POSTSUBSCRIPT, then Eq. 1 reduces to a purely hydrodynamic form that describes viscous flowShavit et al. (2019)

αvisc.=rb216πη[11γ24ln2γ(γ21)].superscript𝛼𝑣𝑖𝑠𝑐superscriptsubscript𝑟𝑏216𝜋𝜂delimited-[]11superscript𝛾24superscript2𝛾superscript𝛾21\alpha^{visc.}=\frac{r_{b}^{2}}{16\pi\eta}[1-\frac{1}{\gamma^{2}}-\frac{4\ln^{% 2}\gamma}{(\gamma^{2}-1)}].italic_α start_POSTSUPERSCRIPT italic_v italic_i italic_s italic_c . end_POSTSUPERSCRIPT = divide start_ARG italic_r start_POSTSUBSCRIPT italic_b end_POSTSUBSCRIPT start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG start_ARG 16 italic_π italic_η end_ARG [ 1 - divide start_ARG 1 end_ARG start_ARG italic_γ start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG - divide start_ARG 4 roman_ln start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT italic_γ end_ARG start_ARG ( italic_γ start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT - 1 ) end_ARG ] . (2)

In the opposite limit, Dνramuch-less-thansubscript𝐷𝜈subscript𝑟𝑎D_{\nu}\ll r_{a}italic_D start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ≪ italic_r start_POSTSUBSCRIPT italic_a end_POSTSUBSCRIPT, Eq. 1 reduces to a purely Ohmic formWeiss and Welker. (1954),

αOhm=lnγ2πρo(ne)2.superscript𝛼𝑂𝑚𝛾2𝜋subscript𝜌𝑜superscript𝑛𝑒2\alpha^{Ohm}=\frac{\ln{\gamma}}{2\pi\rho_{o}(ne)^{2}}.italic_α start_POSTSUPERSCRIPT italic_O italic_h italic_m end_POSTSUPERSCRIPT = divide start_ARG roman_ln italic_γ end_ARG start_ARG 2 italic_π italic_ρ start_POSTSUBSCRIPT italic_o end_POSTSUBSCRIPT ( italic_n italic_e ) start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG . (3)

Fig. 1c upper panel shows the magnetoresistance measured at T=130𝑇130T=130italic_T = 130 K for the same carrier density as in Fig. 1b. The resistance shows the expected B2superscript𝐵2B^{2}italic_B start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT dependence, but with two apparent regimes. This is seen more clearly by plotting the dR/d(B2)𝑑𝑅𝑑superscript𝐵2{dR}/{d(B^{2})}italic_d italic_R / italic_d ( italic_B start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT ) derivative of the magnetoresistance (Fig. 1c, lower panel), which shows two distinct plateaus. By fitting two separate parabolas to the magnetoresistance we extract two curvatures corresponding to the “low” and “high” field regimes, respectively (black and red curves in Fig. 1b – see the determination of fitting range in SI).

We assume ρosubscript𝜌𝑜\rho_{o}italic_ρ start_POSTSUBSCRIPT italic_o end_POSTSUBSCRIPT is constant in magnetic field, however the viscosity is modified with B𝐵Bitalic_B according to η(B)=η/(1+4lee2/rc2)𝜂𝐵𝜂14superscriptsubscript𝑙𝑒𝑒2superscriptsubscript𝑟𝑐2\eta(B)={\eta}/({1+4l_{ee}^{2}/r_{c}^{2}})italic_η ( italic_B ) = italic_η / ( 1 + 4 italic_l start_POSTSUBSCRIPT italic_e italic_e end_POSTSUBSCRIPT start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT / italic_r start_POSTSUBSCRIPT italic_c end_POSTSUBSCRIPT start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT ), where rc=πn/eBsubscript𝑟𝑐Planck-constant-over-2-pi𝜋𝑛𝑒𝐵r_{c}=\hbar\sqrt{\pi n}/eBitalic_r start_POSTSUBSCRIPT italic_c end_POSTSUBSCRIPT = roman_ℏ square-root start_ARG italic_π italic_n end_ARG / italic_e italic_B is the cyclotron radiusAlekseev (2016); Holder et al. (2019). As B𝐵Bitalic_B increases, rcsubscript𝑟𝑐r_{c}italic_r start_POSTSUBSCRIPT italic_c end_POSTSUBSCRIPT decreases and eventually η0𝜂0\eta\rightarrow 0italic_η → 0, fully suppressing the viscous contribution and yielding a predominantly Ohmic response. We therefore interpret the two curvatures as follows: the low field curvature, labelled α𝛼\alphaitalic_α, describes the mixed flow regime and thus depends on both η𝜂\etaitalic_η and ρosubscript𝜌𝑜\rho_{o}italic_ρ start_POSTSUBSCRIPT italic_o end_POSTSUBSCRIPT (Eq. 1); the high field curvature, labelled αOhmsuperscript𝛼𝑂𝑚\alpha^{Ohm}italic_α start_POSTSUPERSCRIPT italic_O italic_h italic_m end_POSTSUPERSCRIPT reflects the transition to purely Ohmic response, and thus depends on ρosubscript𝜌𝑜\rho_{o}italic_ρ start_POSTSUBSCRIPT italic_o end_POSTSUBSCRIPT only (Eq. 3).

Fig. 1d shows the temperature dependence of the two quadratic coefficients. The high field curvature, αOhmsuperscript𝛼𝑂𝑚\alpha^{Ohm}italic_α start_POSTSUPERSCRIPT italic_O italic_h italic_m end_POSTSUPERSCRIPT, decreases monotonically, as expected from Eq. 3 and assuming dρo/dT>0𝑑subscript𝜌𝑜𝑑𝑇0d\rho_{o}/dT>0italic_d italic_ρ start_POSTSUBSCRIPT italic_o end_POSTSUBSCRIPT / italic_d italic_T > 0 (typical for graphene at these densities). The low field curvature, shows a non-monotonic trend, first increasing then decreasing. This is consistent with the transport response transitioning from being more viscous-like at low temperature to Ohmic-like at high temperature. The crossover results from the fact that while αvisc1/ηsimilar-tosuperscript𝛼𝑣𝑖𝑠𝑐1𝜂\alpha^{visc}\sim 1/\etaitalic_α start_POSTSUPERSCRIPT italic_v italic_i italic_s italic_c end_POSTSUPERSCRIPT ∼ 1 / italic_η (Eq. 2) and αOhm1/ρosimilar-tosuperscript𝛼𝑂𝑚1subscript𝜌𝑜\alpha^{Ohm}\sim 1/\rho_{o}italic_α start_POSTSUPERSCRIPT italic_O italic_h italic_m end_POSTSUPERSCRIPT ∼ 1 / italic_ρ start_POSTSUBSCRIPT italic_o end_POSTSUBSCRIPT (Eq. 3), η𝜂\etaitalic_η decreases with T𝑇Titalic_T whereas ρosubscript𝜌𝑜\rho_{o}italic_ρ start_POSTSUBSCRIPT italic_o end_POSTSUBSCRIPT increases with T𝑇Titalic_T.

The ability to isolate the Ohmic magnetoresistance at high field provides a means to measure the Ohmic scattering term, independent of any zero-field contact or hydrodynamic contributions and without any other free parameters. Once we determine ρosubscript𝜌𝑜\rho_{o}italic_ρ start_POSTSUBSCRIPT italic_o end_POSTSUBSCRIPT, we can then use our measurement of low-field α𝛼\alphaitalic_α along with Eq. 1 to extract η𝜂\etaitalic_η. This same process can be performed at any density and temperature, allowing us to quantify the individual Ohmic and viscous flow-related parameters throughout the mixed-flow regime.

Refer to caption
Figure 2: Scale-dependent magnetotransport and viscous-ohmic crossover a,b,α𝛼\alphaitalic_α vs T𝑇Titalic_T in five MLG corbino devices (a) and four BLG corbino devices (b) with the same aspect ratio γ=4𝛾4\gamma=4italic_γ = 4 but different radii. c,d,The scaling power β𝛽\betaitalic_β which is extracted by fitting α𝛼\alphaitalic_α vs rbsubscript𝑟𝑏r_{b}italic_r start_POSTSUBSCRIPT italic_b end_POSTSUBSCRIPT for MLG (c) and BLG (d) devices as a function of n𝑛nitalic_n and T𝑇Titalic_T (see SI for examples on fitting). Contours are marked by dashed lines. e,f, Measured Gurzhi length Dνsubscript𝐷𝜈D_{\nu}italic_D start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT versus n𝑛nitalic_n and T𝑇Titalic_T for MLG (e) and BLG (f).

First, we focus on the Ohmic resistance. The red data points in Fig. 1e show a plot of lnγ2πρo(T)𝛾2𝜋subscript𝜌𝑜𝑇\frac{\ln\gamma}{2\pi}\rho_{o}(T)divide start_ARG roman_ln italic_γ end_ARG start_ARG 2 italic_π end_ARG italic_ρ start_POSTSUBSCRIPT italic_o end_POSTSUBSCRIPT ( italic_T ) measured at the same density as Fig. 1b,d,e. The temperature dependence of ρo(T)subscript𝜌𝑜𝑇\rho_{o}(T)italic_ρ start_POSTSUBSCRIPT italic_o end_POSTSUBSCRIPT ( italic_T ) closely resembles results from four-terminal Hall bar measurement (see SI) where the linear and exponential terms arise from acoustic and optical phonon scattering, respectively, and any impurity scattering contribution remains temperature-independent over this rangeDean et al. (2010); Chen et al. (2008). Measurement over a wide density range for both MLG and BLG show similar results (see SI). Fitting the linear-T regime gives an acoustic-phonon deformation potential DA=18±1subscript𝐷𝐴plus-or-minus181D_{A}=18\pm 1italic_D start_POSTSUBSCRIPT italic_A end_POSTSUBSCRIPT = 18 ± 1 eV in quantitative agreement with prior measurements, which we take as confirmation that we are faithfully determining ρosubscript𝜌𝑜\rho_{o}italic_ρ start_POSTSUBSCRIPT italic_o end_POSTSUBSCRIPT with our analysis.

The B=0𝐵0B=0italic_B = 0 intercept of the parabolic fit to the αOhmsuperscript𝛼𝑂𝑚\alpha^{Ohm}italic_α start_POSTSUPERSCRIPT italic_O italic_h italic_m end_POSTSUPERSCRIPT regime, Rosubscript𝑅𝑜R_{o}italic_R start_POSTSUBSCRIPT italic_o end_POSTSUBSCRIPT, includes the zero field Ohmic and external resistances, Ro=lnγ2πρo+RExtsubscript𝑅𝑜𝛾2𝜋subscript𝜌𝑜superscript𝑅𝐸𝑥𝑡R_{o}=\frac{\ln\gamma}{2\pi}\rho_{o}+R^{Ext}italic_R start_POSTSUBSCRIPT italic_o end_POSTSUBSCRIPT = divide start_ARG roman_ln italic_γ end_ARG start_ARG 2 italic_π end_ARG italic_ρ start_POSTSUBSCRIPT italic_o end_POSTSUBSCRIPT + italic_R start_POSTSUPERSCRIPT italic_E italic_x italic_t end_POSTSUPERSCRIPT. Using the value of ρosubscript𝜌𝑜\rho_{o}italic_ρ start_POSTSUBSCRIPT italic_o end_POSTSUBSCRIPT determined from the curvature together with Rosubscript𝑅𝑜R_{o}italic_R start_POSTSUBSCRIPT italic_o end_POSTSUBSCRIPT therefore allows us to obtain a value for Rextsuperscript𝑅𝑒𝑥𝑡R^{ext}italic_R start_POSTSUPERSCRIPT italic_e italic_x italic_t end_POSTSUPERSCRIPT. Fig. 1b shows a plot of Rext(T)superscript𝑅𝑒𝑥𝑡𝑇R^{ext}(T)italic_R start_POSTSUPERSCRIPT italic_e italic_x italic_t end_POSTSUPERSCRIPT ( italic_T ) (grey circles) compared with the total two-terminal resistance, Rtot(T)superscript𝑅𝑡𝑜𝑡𝑇R^{tot}(T)italic_R start_POSTSUPERSCRIPT italic_t italic_o italic_t end_POSTSUPERSCRIPT ( italic_T ) (black circles). The two curves show a similar trend and magnitude, indicating a significant proportion of the overall response is dominated by the external resistances. Nonetheless, the sample-only resistance, Rsamplesuperscript𝑅𝑠𝑎𝑚𝑝𝑙𝑒R^{sample}italic_R start_POSTSUPERSCRIPT italic_s italic_a italic_m italic_p italic_l italic_e end_POSTSUPERSCRIPT can be obtained by subtracting these two curves Rsample=RtotRextsuperscript𝑅𝑠𝑎𝑚𝑝𝑙𝑒superscript𝑅𝑡𝑜𝑡superscript𝑅𝑒𝑥𝑡R^{sample}=R^{tot}-R^{ext}italic_R start_POSTSUPERSCRIPT italic_s italic_a italic_m italic_p italic_l italic_e end_POSTSUPERSCRIPT = italic_R start_POSTSUPERSCRIPT italic_t italic_o italic_t end_POSTSUPERSCRIPT - italic_R start_POSTSUPERSCRIPT italic_e italic_x italic_t end_POSTSUPERSCRIPT. The black circles in Fig. 1e shows a plot of Rsample(T)superscript𝑅𝑠𝑎𝑚𝑝𝑙𝑒𝑇R^{sample}(T)italic_R start_POSTSUPERSCRIPT italic_s italic_a italic_m italic_p italic_l italic_e end_POSTSUPERSCRIPT ( italic_T ). We observe a non-monontonic dependence, first decreasing and then increasing with temperature. This behaviour is a manifestation of the Gurzhi effectGurzhi (1963, 1968), which predicts the observed dR/dT𝑑𝑅𝑑𝑇dR/dTitalic_d italic_R / italic_d italic_T sign change as the system evolves from momentum-conserving (viscous flow) to momentum-relaxing (Ohmic flow) dominated scattering with increasing temperatureMolenkamp and de Jong (1994); de Jong and Molenkamp (1995); Moll et al. (2016). Finally, we isolate the viscous-flow contribution to the sample resistance according to the Matthiessen’s rule, Rvisc=RsampleROhmicsuperscript𝑅𝑣𝑖𝑠𝑐superscript𝑅𝑠𝑎𝑚𝑝𝑙𝑒superscript𝑅𝑂𝑚𝑖𝑐R^{visc}=R^{sample}-R^{Ohmic}italic_R start_POSTSUPERSCRIPT italic_v italic_i italic_s italic_c end_POSTSUPERSCRIPT = italic_R start_POSTSUPERSCRIPT italic_s italic_a italic_m italic_p italic_l italic_e end_POSTSUPERSCRIPT - italic_R start_POSTSUPERSCRIPT italic_O italic_h italic_m italic_i italic_c end_POSTSUPERSCRIPT, shown as blue circles in Fig. 1e. Rviscsuperscript𝑅𝑣𝑖𝑠𝑐R^{visc}italic_R start_POSTSUPERSCRIPT italic_v italic_i italic_s italic_c end_POSTSUPERSCRIPT decreases monotonically with temperature, as expected for hydrodynamic transport Shavit et al. (2019); Kumar et al. (2022).

We can further confirm our interpretation of the viscous-Ohmic crossover with increasing temperature by looking at how the magnetoresistance scales with device size. In the purely viscous flow regime, α𝛼\alphaitalic_α varies with the outer radius squared (Eq. 2), whereas in the purely Ohmic limit α𝛼\alphaitalic_α depends on the aspect ratio only (Eq. 3), independent of the size. Fig. 2a (2b) shows α𝛼\alphaitalic_α versus temperature measured in MLG (BLG) devices with size ranging from rb=1.75μsubscript𝑟𝑏1.75𝜇r_{b}=1.75\muitalic_r start_POSTSUBSCRIPT italic_b end_POSTSUBSCRIPT = 1.75 italic_μm to 4μ4𝜇4\mu4 italic_μm (rb=1.75μsubscript𝑟𝑏1.75𝜇r_{b}=1.75\muitalic_r start_POSTSUBSCRIPT italic_b end_POSTSUBSCRIPT = 1.75 italic_μm to 3.5μ3.5𝜇3.5\mu3.5 italic_μm), but fixed γ=4𝛾4\gamma=4italic_γ = 4. The different radius Corbino devices where fabricated from a single heterostructure to ensure identical scale-independent properties (see SI). To quantify the size dependence we calculate the scaling power by assuming αrbβsimilar-to𝛼superscriptsubscript𝑟𝑏𝛽\alpha\sim r_{b}^{\beta}italic_α ∼ italic_r start_POSTSUBSCRIPT italic_b end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_β end_POSTSUPERSCRIPT and then determine β𝛽\betaitalic_β from a linear fit on a log-log plot (see SI for examples of this fitting). Fig. 2b,c plots β𝛽\betaitalic_β versus density and temperature for MLG and BLG, respectively. In both cases, β2similar-to𝛽2\beta\sim 2italic_β ∼ 2 at low T𝑇Titalic_T and then decreases monotonically with increasing T𝑇Titalic_T. This is consisent with a predominantly hydrodynamic flow at low T𝑇Titalic_T, but with Ohmic contributions increasingly participating with increasing T𝑇Titalic_T. β𝛽\betaitalic_β never reaches zero, suggesting that in the density range considered, non-negligible viscous contributions remain present up to at least T300𝑇300T\approx 300italic_T ≈ 300 K. Additionally we observe that β𝛽\betaitalic_β contour lines are overall density indpendent for MLG, but density dependent in BLG reflecting differences in the η(n)𝜂𝑛\eta(n)italic_η ( italic_n ) and ρo(n)subscript𝜌𝑜𝑛\rho_{o}(n)italic_ρ start_POSTSUBSCRIPT italic_o end_POSTSUBSCRIPT ( italic_n ) for the two systems. In Fig. 2e,f we plot the Gurzhi length versus T𝑇Titalic_T and n𝑛nitalic_n, using the magnetoresistance curvature to determine ρosubscript𝜌𝑜\rho_{o}italic_ρ start_POSTSUBSCRIPT italic_o end_POSTSUBSCRIPT and η𝜂\etaitalic_η, as described above. The countour lines of Dνsubscript𝐷𝜈D_{\nu}italic_D start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT exhibit a similar trend as the corresponding β𝛽\betaitalic_β, confirming that the viscous-Ohmic crossover coincides with a reduction in Dνsubscript𝐷𝜈D_{\nu}italic_D start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT, as expected.

Refer to caption
Figure 3: Kinematic viscosity and effective electron-electron scattering rate in doped graphene. a,b Kinematic viscosity ν𝜈\nuitalic_ν versus carrier density at different temperatures for MLG (a) and BLG (b). c,d,Kinematic viscosity versus temperature for heavily doped MLG (c) and BLG (d). Black solid line denotes fit to T1superscript𝑇1T^{-1}italic_T start_POSTSUPERSCRIPT - 1 end_POSTSUPERSCRIPT. The dependence T2superscript𝑇2T^{-2}italic_T start_POSTSUPERSCRIPT - 2 end_POSTSUPERSCRIPT is shown in grey as a comparison. e,f,Electron-electron scattering rate calculated from ν𝜈\nuitalic_ν in (c),(d),respectively, using ν=vF2τee/4𝜈superscriptsubscript𝑣𝐹2subscript𝜏𝑒𝑒4\nu=v_{F}^{2}\tau_{ee}/4italic_ν = italic_v start_POSTSUBSCRIPT italic_F end_POSTSUBSCRIPT start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT italic_τ start_POSTSUBSCRIPT italic_e italic_e end_POSTSUBSCRIPT / 4. Data present in this figure is taken where EF/kBT>2.5subscript𝐸𝐹subscript𝑘𝐵𝑇2.5E_{F}/k_{B}T>2.5italic_E start_POSTSUBSCRIPT italic_F end_POSTSUBSCRIPT / italic_k start_POSTSUBSCRIPT italic_B end_POSTSUBSCRIPT italic_T > 2.5.
Refer to caption
Figure 4: Crossover between Fermi liquid and electron-hole plasma. a,Shear viscosity η𝜂\etaitalic_η versus temperature at different carrier densities in MLG. Black dashed line denotes the ηT2similar-to𝜂superscript𝑇2\eta\sim T^{2}italic_η ∼ italic_T start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT dependence. Solid lines are fitting to data using Eq. 4 (see SI for fitting parameters). b,η𝜂\etaitalic_η versus temperature at different carrier densities in BLG. Black (gray) dashed line denotes the ηTsimilar-to𝜂𝑇\eta\sim Titalic_η ∼ italic_T (T2similar-toabsentsuperscript𝑇2\sim T^{2}∼ italic_T start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT) dependence. Solid lines are fits to data. c,d Percentage of viscosity from electron-hole plasma in total viscosity weh=ηeh/ηsubscript𝑤𝑒subscript𝜂𝑒𝜂w_{eh}=\eta_{eh}/\etaitalic_w start_POSTSUBSCRIPT italic_e italic_h end_POSTSUBSCRIPT = italic_η start_POSTSUBSCRIPT italic_e italic_h end_POSTSUBSCRIPT / italic_η for MLG (c) and BLG (d), respectively. weh=ηeh/ηsubscript𝑤𝑒subscript𝜂𝑒𝜂w_{eh}=\eta_{eh}/\etaitalic_w start_POSTSUBSCRIPT italic_e italic_h end_POSTSUBSCRIPT = italic_η start_POSTSUBSCRIPT italic_e italic_h end_POSTSUBSCRIPT / italic_η is reconstructed from fitted functions (see SI) and Eq. 4. Scattered data points mark the crossover temperature defined as where η𝜂\etaitalic_η is minimized.

Anomalous Fermi liquid behaviour
Next we examine how the kinematic viscosity ν=η/nm𝜈𝜂𝑛𝑚\nu=\eta/nmitalic_ν = italic_η / italic_n italic_m evolves with temperature and carrier density, as ν𝜈\nuitalic_ν directly captures momentum-diffusion in the equations of motionYudhistira et al. (2023).For MLG, we use the mass definition m=πn/vF𝑚Planck-constant-over-2-pi𝜋𝑛subscript𝑣𝐹m=\hbar\sqrt{\pi n}/v_{F}italic_m = roman_ℏ square-root start_ARG italic_π italic_n end_ARG / italic_v start_POSTSUBSCRIPT italic_F end_POSTSUBSCRIPT where vF=106subscript𝑣𝐹superscript106v_{F}=10^{6}italic_v start_POSTSUBSCRIPT italic_F end_POSTSUBSCRIPT = 10 start_POSTSUPERSCRIPT 6 end_POSTSUPERSCRIPTm/s, and for BLG we use m=0.03me𝑚0.03subscript𝑚𝑒m=0.03m_{e}italic_m = 0.03 italic_m start_POSTSUBSCRIPT italic_e end_POSTSUBSCRIPT Li et al. (2016). Fig. 3a-d plots the density and temperature dependence of ν𝜈\nuitalic_ν measured for both MLG and BLG. In MLG, ν𝜈\nuitalic_ν follows a T1superscript𝑇1T^{-1}italic_T start_POSTSUPERSCRIPT - 1 end_POSTSUPERSCRIPT dependence for nearly a decade of response, and shows no density dependence, within the measurement uncertainty (Fig. 3a,c). The effective electron-electron scattering rate, which is given by τee1=vF2/4νsuperscriptsubscript𝜏𝑒𝑒1superscriptsubscript𝑣𝐹24𝜈\tau_{ee}^{-1}=v_{F}^{2}/4\nuitalic_τ start_POSTSUBSCRIPT italic_e italic_e end_POSTSUBSCRIPT start_POSTSUPERSCRIPT - 1 end_POSTSUPERSCRIPT = italic_v start_POSTSUBSCRIPT italic_F end_POSTSUBSCRIPT start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT / 4 italic_ν, consequently has a T𝑇Titalic_T-linear dependence and, since vFsubscript𝑣𝐹v_{F}italic_v start_POSTSUBSCRIPT italic_F end_POSTSUBSCRIPT is constant for MLG, is density-independent (Fig. 3e). In BLG, ν𝜈\nuitalic_ν again follows T1superscript𝑇1T^{-1}italic_T start_POSTSUPERSCRIPT - 1 end_POSTSUPERSCRIPT behaviour, but in this case varies linearly with n𝑛nitalic_n (Fig. 3b,d). For the massive BLG band, vF=2πn/msubscript𝑣𝐹Planck-constant-over-2-pi2𝜋𝑛𝑚v_{F}=\hbar\sqrt{2\pi n}/mitalic_v start_POSTSUBSCRIPT italic_F end_POSTSUBSCRIPT = roman_ℏ square-root start_ARG 2 italic_π italic_n end_ARG / italic_m so that νn/Tsimilar-to𝜈𝑛𝑇\nu\sim n/Titalic_ν ∼ italic_n / italic_T translates again to an overall T𝑇Titalic_T-linear and density-independent scattering rate (Fig. 3f).

The observation that τee1Tsimilar-tosuperscriptsubscript𝜏𝑒𝑒1𝑇\tau_{ee}^{-1}\sim Titalic_τ start_POSTSUBSCRIPT italic_e italic_e end_POSTSUBSCRIPT start_POSTSUPERSCRIPT - 1 end_POSTSUPERSCRIPT ∼ italic_T and is density independent for both massless electrons in MLG and massive electrons in BLG suggests that this scaling is independent of the band structure. Evidence for similar scaling can be seen in the data from previous graphene studies using different techniques, but was not explicitly discussed Bandurin et al. (2016); Berdyugin et al. (2019); Krishna Kumar et al. (2017). According to Fermi liquid theory, the ee𝑒𝑒e-eitalic_e - italic_e scattering rate is expected to scale as τee1T2/EFsimilar-tosuperscriptsubscript𝜏𝑒𝑒1superscript𝑇2subscript𝐸𝐹\tau_{ee}^{-1}\sim T^{2}/E_{F}italic_τ start_POSTSUBSCRIPT italic_e italic_e end_POSTSUBSCRIPT start_POSTSUPERSCRIPT - 1 end_POSTSUPERSCRIPT ∼ italic_T start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT / italic_E start_POSTSUBSCRIPT italic_F end_POSTSUBSCRIPT in two dimensions Principi et al. (2016). The discrepancy between the temperature and density dependences measured in both MLG and BLG from conventional Fermi liquid theory is surprising and may urge careful reexamination of the standard Fermi liquid description. Recent theoretical work has identified that interacting 2D charged fluids may exhibit an exotic new regime between ballistic transport and hydrodynamic flow. Named “tomographic transport”, this new regime is characterized by a viscous type flow but where the lifetime of odd and even harmonic modes contribute differently to viscosity Ledwith et al. (2019a); Hofmann and Das Sarma (2022); Hofmann and Gran (2023); Kryhin et al. (2023) Within the tomographic flow regime, the effective scattering rate is predicted to go as τee1T/n1/6similar-tosuperscriptsubscript𝜏𝑒𝑒1𝑇superscript𝑛16\tau_{ee}^{-1}\sim T/n^{1/6}italic_τ start_POSTSUBSCRIPT italic_e italic_e end_POSTSUBSCRIPT start_POSTSUPERSCRIPT - 1 end_POSTSUPERSCRIPT ∼ italic_T / italic_n start_POSTSUPERSCRIPT 1 / 6 end_POSTSUPERSCRIPT down to the lowest temperatures Kryhin et al. (2023). The tomographic theory therefore aligns exactly with our observed τee1Tsimilar-tosuperscriptsubscript𝜏𝑒𝑒1𝑇\tau_{ee}^{-1}\sim Titalic_τ start_POSTSUBSCRIPT italic_e italic_e end_POSTSUBSCRIPT start_POSTSUPERSCRIPT - 1 end_POSTSUPERSCRIPT ∼ italic_T behavior, and predicts a n𝑛nitalic_n dependence much closer to the experimental observation (see SI for derivation and estimate of viscosity using tomographic transport theory).

Crossover betwen Fermi liquid and Dirac fluid
So far we have examined charge densities sufficiently far from the charge neutrality point (CNP) to be considered a homogeneous Fermi liquid, and have considered a mixed flow regime involving both viscous and Ohmic componentsLevchenko and Schmalian (2020); Shavit et al. (2019). At low densities close to the CNP (EFkBTmuch-less-thansubscript𝐸𝐹subscript𝑘𝐵𝑇E_{F}\ll k_{B}Titalic_E start_POSTSUBSCRIPT italic_F end_POSTSUBSCRIPT ≪ italic_k start_POSTSUBSCRIPT italic_B end_POSTSUBSCRIPT italic_T), graphene resembles instead a relativistic electron-hole plasma known as a Dirac fluid. In this limit, transport is described by relativsitic hydrodynamics, governed by a combination of both e𝑒eitalic_e-e𝑒eitalic_e and e𝑒eitalic_e-hhitalic_h interactions. It was proposed that at the CNP, graphene demonstrates a nearly “perfect fluid” where the ratio of shear viscosity, η𝜂\etaitalic_η, to entropy density, s𝑠sitalic_s, approaches a universal quantum limit η/s/4πkB𝜂𝑠Planck-constant-over-2-pi4𝜋subscript𝑘𝐵\eta/s\geq\hbar/4\pi k_{B}italic_η / italic_s ≥ roman_ℏ / 4 italic_π italic_k start_POSTSUBSCRIPT italic_B end_POSTSUBSCRIPTKovtun et al. (2005); Son (2007); Son and Starinets (2007); Karsch et al. (2008); Sachdev and Müller (2009); Müller et al. (2009). Prior experimental studies identified evidence of hydrodynamic behaviour near the CNPCrossno et al. (2016); Ku et al. (2020), and with estimates of η/s𝜂𝑠\eta/sitalic_η / italic_s at room temperature reaching 0.3/kBsimilar-toabsent0.3Planck-constant-over-2-pisubscript𝑘𝐵\sim 0.3\hbar/k_{B}∼ 0.3 roman_ℏ / italic_k start_POSTSUBSCRIPT italic_B end_POSTSUBSCRIPT, close to the theoretical value 0.2/kB0.2Planck-constant-over-2-pisubscript𝑘𝐵0.2\hbar/k_{B}0.2 roman_ℏ / italic_k start_POSTSUBSCRIPT italic_B end_POSTSUBSCRIPT predicted for MLG at 300 KMüller et al. (2009); Ku et al. (2020). In the e𝑒eitalic_e-hhitalic_h plasma regime, η(T)𝜂𝑇\eta(T)italic_η ( italic_T ) is expected to show opposite behaviour to the Fermi liquid regime i.e. increasing, instead of decreasing, with temperature - providing an unambiguous and hall mark signature of the Dirac fluid. However, the temperature dependence of η𝜂\etaitalic_η in the graphene electron-hole plasma regime has not previously been reported. Moreover, the crossover regime, EFkBTsimilar-tosubscript𝐸𝐹subscript𝑘𝐵𝑇E_{F}\sim k_{B}Titalic_E start_POSTSUBSCRIPT italic_F end_POSTSUBSCRIPT ∼ italic_k start_POSTSUBSCRIPT italic_B end_POSTSUBSCRIPT italic_T, where gate-induced carriers (Fermi liquid) and thermally excited carriers (Dirac fluid) coexist remains unexplored.

Fig. 4a shows η𝜂\etaitalic_η versus T𝑇Titalic_T measured at different carrier densities in MLG. The high density/low T𝑇Titalic_T behaviour is well fit by ηeeMLGn3/2/Tsimilar-tosuperscriptsubscript𝜂𝑒𝑒𝑀𝐿𝐺superscript𝑛32𝑇\eta_{ee}^{MLG}\sim n^{3/2}/Titalic_η start_POSTSUBSCRIPT italic_e italic_e end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_M italic_L italic_G end_POSTSUPERSCRIPT ∼ italic_n start_POSTSUPERSCRIPT 3 / 2 end_POSTSUPERSCRIPT / italic_T, consistent with the τee1Tsimilar-tosuperscriptsubscript𝜏𝑒𝑒1𝑇\tau_{ee}^{-1}\sim Titalic_τ start_POSTSUBSCRIPT italic_e italic_e end_POSTSUBSCRIPT start_POSTSUPERSCRIPT - 1 end_POSTSUPERSCRIPT ∼ italic_T discussed above. By comparison, the low-n𝑛nitalic_n/high-T𝑇Titalic_T behaviour collapses to a universal, density independent, curve where ηehMLGT2similar-tosuperscriptsubscript𝜂𝑒𝑀𝐿𝐺superscript𝑇2\eta_{eh}^{MLG}\sim T^{2}italic_η start_POSTSUBSCRIPT italic_e italic_h end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_M italic_L italic_G end_POSTSUPERSCRIPT ∼ italic_T start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT in agreement with predictions for the charged Dirac fluid for MLGMüller et al. (2009). Assuming EF=kBTsubscript𝐸𝐹subscript𝑘𝐵𝑇E_{F}=k_{B}Titalic_E start_POSTSUBSCRIPT italic_F end_POSTSUBSCRIPT = italic_k start_POSTSUBSCRIPT italic_B end_POSTSUBSCRIPT italic_T, we estimate ν0.3𝜈0.3\nu\approx 0.3italic_ν ≈ 0.3 m/2{}^{2}/start_FLOATSUPERSCRIPT 2 end_FLOATSUPERSCRIPT /s for CNP at 300 K, consistent with result in recent magnetometry imaging experimentKu et al. (2020). Analogous behaviour is observed in BLG (Fig. 4b) with η(T)𝜂𝑇\eta(T)italic_η ( italic_T ) transitioning from ηeeBLGn2/Tsimilar-tosuperscriptsubscript𝜂𝑒𝑒𝐵𝐿𝐺superscript𝑛2𝑇\eta_{ee}^{BLG}\sim n^{2}/Titalic_η start_POSTSUBSCRIPT italic_e italic_e end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_B italic_L italic_G end_POSTSUPERSCRIPT ∼ italic_n start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT / italic_T dependence at high density/low temperature, to a universal curve at low density/high temperature, ηehBLGTsimilar-tosuperscriptsubscript𝜂𝑒𝐵𝐿𝐺𝑇\eta_{eh}^{BLG}\sim Titalic_η start_POSTSUBSCRIPT italic_e italic_h end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_B italic_L italic_G end_POSTSUPERSCRIPT ∼ italic_T, consistent with τee1Tsimilar-tosuperscriptsubscript𝜏𝑒𝑒1𝑇\tau_{ee}^{-1}\sim Titalic_τ start_POSTSUBSCRIPT italic_e italic_e end_POSTSUBSCRIPT start_POSTSUPERSCRIPT - 1 end_POSTSUPERSCRIPT ∼ italic_T, and Dirac fluid theory for BLGYudhistira et al. (2023), respectively. In both MLG and BLG, the shear viscosity continuously interpolates between the Fermi liquid and Dirac fluid response as we vary n𝑛nitalic_n or T𝑇Titalic_T. The full temperature range can be described by simply adding the contribution from gate-injected and thermally excited carriers, shown as solid curves in Fig. 4a,b (see SI for determination of the fitting coefficients).

η=ηeh+ηee𝜂subscript𝜂𝑒subscript𝜂𝑒𝑒\eta=\eta_{eh}+\eta_{ee}italic_η = italic_η start_POSTSUBSCRIPT italic_e italic_h end_POSTSUBSCRIPT + italic_η start_POSTSUBSCRIPT italic_e italic_e end_POSTSUBSCRIPT (4)

.

This result suggest a two-fluid model where the gate-induced electrons and thermally excited electron-hole plasma coexist and contribute to hydrodynamic electrical transport in two parallel channelsYudhistira et al. (2023). This allows us to construct a phase diagram based on the percentage contribution from electron-hole plasma in total hydrodynamic transport (Fig. 4c,d, plotted in Fermi energy units using the MLG and BLG band dispersion, respectively). The data points label the crossover temperature, identified by the η𝜂\etaitalic_η minimum in Fig. 4a,b. The crossover boundary between the degenerate Fermi liquid and electron-hole plasma regimes follows approximately EF=2.3kBTsubscript𝐸𝐹2.3subscript𝑘𝐵𝑇E_{F}=2.3k_{B}Titalic_E start_POSTSUBSCRIPT italic_F end_POSTSUBSCRIPT = 2.3 italic_k start_POSTSUBSCRIPT italic_B end_POSTSUBSCRIPT italic_T for MLG and EF=1.4kBTsubscript𝐸𝐹1.4subscript𝑘𝐵𝑇E_{F}=1.4k_{B}Titalic_E start_POSTSUBSCRIPT italic_F end_POSTSUBSCRIPT = 1.4 italic_k start_POSTSUBSCRIPT italic_B end_POSTSUBSCRIPT italic_T for BLG.

Finally, from our measurement of η(T)𝜂𝑇\eta(T)italic_η ( italic_T ), and using a calculated value for the entropy density from bandstsructure, s(T)𝑠𝑇s(T)italic_s ( italic_T )Müller et al. (2009), we can determine the ratio η/s𝜂𝑠\eta/sitalic_η / italic_s as a function of temperature. We find that for both BLG and MLG the ratio saturates in the low density high temperature limit, as expected (see SI). For MLG, η/s0.2/kB𝜂𝑠0.2Planck-constant-over-2-pisubscript𝑘𝐵\eta/s\approx 0.2\hbar/k_{B}italic_η / italic_s ≈ 0.2 roman_ℏ / italic_k start_POSTSUBSCRIPT italic_B end_POSTSUBSCRIPT, consistent with theoretical predictions Müller et al. (2009) and previous experimental value estimated from profile imaging at 300 KKu et al. (2020). For BLG the measured ratio is approximately four times as large Yudhistira et al. (2023).

Summary
We demonstrate that the magnetoresistance in a Corbino geometry can be utilized to quantify the electron hydrodynamic viscosity, and therefore interparticle scattering rate in a 2D electron system. Applying this technique to the study of monolayer and bilayer graphene we confirm the persistence of viscous effects over large range of density and temperature, and for the first time demonstrate the ability to quantify the Ohmic and viscous contributions in the crossover regime. Our finding that temperature and density scaling of of viscosity is at odds with Fermi liquid theory provides new insight into the microscopic nature of interparticle scattering in a charged two-dimensional fluid. The Corbino viscometer geometry demonstrated here can be universally applied to any 2d material system where the e𝑒eitalic_e-e𝑒eitalic_e mean free path represents the dominant scattering length scale. A particularly promising application of hydrodynamic theory is understanding transport response in flat band systems with strong correlations. Here strong e𝑒eitalic_e-e𝑒eitalic_e interactions play a dominant role in determining ground state behaviour, but, reduced mobility due to the quenched kinetic energy makes it difficult within experimental parameters to eliminate Ohmic considerations and isolate the hydrodynamic components. The capability to disentangle the resistivity and viscosity parameters could therefore provide significant new opportunity to understanding these materials.

I Method

I.1 Device fabrication

The heterostructure is assembled using standard van der Waals transfer technique. The device area is defined using e-beam lithography and shaped using consecutive SF6/O2 plasma etching. A small opening (width ra/4similar-toabsentsubscript𝑟𝑎4\sim r_{a}/4∼ italic_r start_POSTSUBSCRIPT italic_a end_POSTSUBSCRIPT / 4) is left out on the outer circumference for easy access to the inner electrode. Cr/Pd is deposited using e-beam evaporator to make one-dimensional contactWang et al. (2013). All MLG (BLG) devices shown in the this work are fabricated on the same heterostructure consisting of MLG (BLG) to ensure almost identical scattering length among devices. The carrier density n𝑛nitalic_n at each gate voltage is determined from Shubnikov-de Haas oscillation. The hBN dielectric for both MLG and BLG devices are 50similar-toabsent50\sim 50∼ 50 nm thick.

I.2 Electrical transport measurements

We perform electrical measurement using Stanford Research SR830 lock-in amplifier at a frequency of 77.77 Hz. Current flow through each device is less than 500500500500 nA to ensures a linear IV𝐼𝑉I-Vitalic_I - italic_V responseBandurin et al. (2016). The higher frequency and larger current used here comparing to quantum transport measurement at much lower temperaturesZeng et al. (2019) is to gain a better signal to noise ratio. Our measurement configuration is consistent with that in previous experiments studying hydrodynamics in graphene Bandurin et al. (2016); Kumar et al. (2022); Krishna Kumar et al. (2017); Ku et al. (2020). Since multiples devices are sitting on the same substrate and are electrically isolated from each other, they are probed in parallel. An excitation voltage is applied to the inner electrodes of multiples devices simultaneously using one lock-in amplifier. Current drained out of the outer electrodes for each device is measured using separate lock-in amplifiers.

I.3 Magnetoresistance derived from the Stokes-Ohm equation

The magnetotransport of electrons in the hydrodynamic-ohmic crossover regime is described by the Stokes-Ohm equation: ne𝑬+𝒋×𝑩=ηΔ𝒋ne+ρ0𝒋ne𝑛𝑒𝑬𝒋𝑩𝜂Δ𝒋𝑛𝑒subscript𝜌0𝒋𝑛𝑒ne\bm{E}+\bm{j}\times\bm{B}=-\eta\Delta\dfrac{\bm{j}}{ne}+\rho_{0}\bm{j}neitalic_n italic_e bold_italic_E + bold_italic_j × bold_italic_B = - italic_η roman_Δ divide start_ARG bold_italic_j end_ARG start_ARG italic_n italic_e end_ARG + italic_ρ start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT bold_italic_j italic_n italic_e. In the coordinate of (r,θ)𝑟𝜃(r,\theta)( italic_r , italic_θ ), it writes:

ne(Er+jθBne)=ηΔjrne+ρojrne𝑛𝑒subscript𝐸𝑟subscript𝑗𝜃𝐵𝑛𝑒𝜂Δsubscript𝑗𝑟𝑛𝑒subscript𝜌𝑜subscript𝑗𝑟𝑛𝑒\displaystyle ne(E_{r}+\dfrac{j_{\theta}B}{ne})=-\eta\Delta\dfrac{j_{r}}{ne}+% \rho_{o}j_{r}neitalic_n italic_e ( italic_E start_POSTSUBSCRIPT italic_r end_POSTSUBSCRIPT + divide start_ARG italic_j start_POSTSUBSCRIPT italic_θ end_POSTSUBSCRIPT italic_B end_ARG start_ARG italic_n italic_e end_ARG ) = - italic_η roman_Δ divide start_ARG italic_j start_POSTSUBSCRIPT italic_r end_POSTSUBSCRIPT end_ARG start_ARG italic_n italic_e end_ARG + italic_ρ start_POSTSUBSCRIPT italic_o end_POSTSUBSCRIPT italic_j start_POSTSUBSCRIPT italic_r end_POSTSUBSCRIPT italic_n italic_e
nejrneB=ηΔjθne+ρojθne𝑛𝑒subscript𝑗𝑟𝑛𝑒𝐵𝜂Δsubscript𝑗𝜃𝑛𝑒subscript𝜌𝑜subscript𝑗𝜃𝑛𝑒\displaystyle-ne\dfrac{j_{r}}{ne}B=-\eta\Delta\dfrac{j_{\theta}}{ne}+\rho_{o}j% _{\theta}ne- italic_n italic_e divide start_ARG italic_j start_POSTSUBSCRIPT italic_r end_POSTSUBSCRIPT end_ARG start_ARG italic_n italic_e end_ARG italic_B = - italic_η roman_Δ divide start_ARG italic_j start_POSTSUBSCRIPT italic_θ end_POSTSUBSCRIPT end_ARG start_ARG italic_n italic_e end_ARG + italic_ρ start_POSTSUBSCRIPT italic_o end_POSTSUBSCRIPT italic_j start_POSTSUBSCRIPT italic_θ end_POSTSUBSCRIPT italic_n italic_e

Where Ersubscript𝐸𝑟E_{r}italic_E start_POSTSUBSCRIPT italic_r end_POSTSUBSCRIPT and Eθsubscript𝐸𝜃E_{\theta}italic_E start_POSTSUBSCRIPT italic_θ end_POSTSUBSCRIPT are the radial and azimuthal electric field, jrsubscript𝑗𝑟j_{r}italic_j start_POSTSUBSCRIPT italic_r end_POSTSUBSCRIPT and jθsubscript𝑗𝜃j_{\theta}italic_j start_POSTSUBSCRIPT italic_θ end_POSTSUBSCRIPT are the radial and azimuthal electrical current. Considering the rotational symmetry of the corbino disk, Ersubscript𝐸𝑟E_{r}italic_E start_POSTSUBSCRIPT italic_r end_POSTSUBSCRIPT, Eθsubscript𝐸𝜃E_{\theta}italic_E start_POSTSUBSCRIPT italic_θ end_POSTSUBSCRIPT, jrsubscript𝑗𝑟j_{r}italic_j start_POSTSUBSCRIPT italic_r end_POSTSUBSCRIPT and jθsubscript𝑗𝜃j_{\theta}italic_j start_POSTSUBSCRIPT italic_θ end_POSTSUBSCRIPT depend on r𝑟ritalic_r but not θ𝜃\thetaitalic_θ. Δ2r2+1rr1r2Δsuperscript2superscript𝑟21𝑟𝑟1superscript𝑟2\Delta\equiv\frac{\partial^{2}}{\partial r^{2}}+\frac{1}{r}\frac{\partial}{% \partial r}-\frac{1}{r^{2}}roman_Δ ≡ divide start_ARG ∂ start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG start_ARG ∂ italic_r start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG + divide start_ARG 1 end_ARG start_ARG italic_r end_ARG divide start_ARG ∂ end_ARG start_ARG ∂ italic_r end_ARG - divide start_ARG 1 end_ARG start_ARG italic_r start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG. In the corbino measurement setup, Eθ=0subscript𝐸𝜃0E_{\theta}=0italic_E start_POSTSUBSCRIPT italic_θ end_POSTSUBSCRIPT = 0. And the radial current is known from the charge conservation law: jr=I2πrsubscript𝑗𝑟𝐼2𝜋𝑟j_{r}=\frac{I}{2\pi r}italic_j start_POSTSUBSCRIPT italic_r end_POSTSUBSCRIPT = divide start_ARG italic_I end_ARG start_ARG 2 italic_π italic_r end_ARG. This gives us the following equation for the azimuthal current:

BI2πr=ηne(d2jθdr2+1rjθr1r2jθ)+ρonejθ𝐵𝐼2𝜋𝑟𝜂𝑛𝑒superscript𝑑2subscript𝑗𝜃𝑑superscript𝑟21𝑟subscript𝑗𝜃𝑟1superscript𝑟2subscript𝑗𝜃subscript𝜌𝑜𝑛𝑒subscript𝑗𝜃\frac{BI}{2\pi r}=\frac{\eta}{ne}(\frac{d^{2}j_{\theta}}{dr^{2}}+\frac{1}{r}% \frac{\partial j_{\theta}}{\partial r}-\frac{1}{r^{2}}j_{\theta})+\rho_{o}nej_% {\theta}divide start_ARG italic_B italic_I end_ARG start_ARG 2 italic_π italic_r end_ARG = divide start_ARG italic_η end_ARG start_ARG italic_n italic_e end_ARG ( divide start_ARG italic_d start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT italic_j start_POSTSUBSCRIPT italic_θ end_POSTSUBSCRIPT end_ARG start_ARG italic_d italic_r start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG + divide start_ARG 1 end_ARG start_ARG italic_r end_ARG divide start_ARG ∂ italic_j start_POSTSUBSCRIPT italic_θ end_POSTSUBSCRIPT end_ARG start_ARG ∂ italic_r end_ARG - divide start_ARG 1 end_ARG start_ARG italic_r start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG italic_j start_POSTSUBSCRIPT italic_θ end_POSTSUBSCRIPT ) + italic_ρ start_POSTSUBSCRIPT italic_o end_POSTSUBSCRIPT italic_n italic_e italic_j start_POSTSUBSCRIPT italic_θ end_POSTSUBSCRIPT

The general solution takes the form

jθ(r)=BI2πρone[AI1(r/Dν)+BK1(r/Dν)1r]subscript𝑗𝜃𝑟𝐵𝐼2𝜋subscript𝜌𝑜𝑛𝑒delimited-[]𝐴subscript𝐼1𝑟subscript𝐷𝜈𝐵subscript𝐾1𝑟subscript𝐷𝜈1𝑟j_{\theta}(r)=\frac{BI}{2\pi\rho_{o}ne}[AI_{1}(r/D_{\nu})+BK_{1}(r/D_{\nu})-% \frac{1}{r}]italic_j start_POSTSUBSCRIPT italic_θ end_POSTSUBSCRIPT ( italic_r ) = divide start_ARG italic_B italic_I end_ARG start_ARG 2 italic_π italic_ρ start_POSTSUBSCRIPT italic_o end_POSTSUBSCRIPT italic_n italic_e end_ARG [ italic_A italic_I start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT ( italic_r / italic_D start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ) + italic_B italic_K start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT ( italic_r / italic_D start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ) - divide start_ARG 1 end_ARG start_ARG italic_r end_ARG ]

I1subscript𝐼1I_{1}italic_I start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT and K1subscript𝐾1K_{1}italic_K start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT are the modified Bessel functions of the second kind. Assuming the no-slip boundary condition, the normal velocity at the edge of corbino equals zero: jθ(ra)=jθ(rb)=0subscript𝑗𝜃subscript𝑟𝑎subscript𝑗𝜃subscript𝑟𝑏0j_{\theta}(r_{a})=j_{\theta}(r_{b})=0italic_j start_POSTSUBSCRIPT italic_θ end_POSTSUBSCRIPT ( italic_r start_POSTSUBSCRIPT italic_a end_POSTSUBSCRIPT ) = italic_j start_POSTSUBSCRIPT italic_θ end_POSTSUBSCRIPT ( italic_r start_POSTSUBSCRIPT italic_b end_POSTSUBSCRIPT ) = 0. One finds the coefficients:

A=1C(1raK1(λ)1rbK1(λ/γ))𝐴1𝐶1subscript𝑟𝑎subscript𝐾1𝜆1subscript𝑟𝑏subscript𝐾1𝜆𝛾\displaystyle A=\frac{1}{C}(\frac{1}{r_{a}}K_{1}(\lambda)-\frac{1}{r_{b}}K_{1}% (\lambda/\gamma))italic_A = divide start_ARG 1 end_ARG start_ARG italic_C end_ARG ( divide start_ARG 1 end_ARG start_ARG italic_r start_POSTSUBSCRIPT italic_a end_POSTSUBSCRIPT end_ARG italic_K start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT ( italic_λ ) - divide start_ARG 1 end_ARG start_ARG italic_r start_POSTSUBSCRIPT italic_b end_POSTSUBSCRIPT end_ARG italic_K start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT ( italic_λ / italic_γ ) )
B=1C(1raI1(λ)1rbI1(λ/γ))𝐵1𝐶1subscript𝑟𝑎subscript𝐼1𝜆1subscript𝑟𝑏subscript𝐼1𝜆𝛾\displaystyle B=-\frac{1}{C}(\frac{1}{r_{a}}I_{1}(\lambda)-\frac{1}{r_{b}}I_{1% }(\lambda/\gamma))italic_B = - divide start_ARG 1 end_ARG start_ARG italic_C end_ARG ( divide start_ARG 1 end_ARG start_ARG italic_r start_POSTSUBSCRIPT italic_a end_POSTSUBSCRIPT end_ARG italic_I start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT ( italic_λ ) - divide start_ARG 1 end_ARG start_ARG italic_r start_POSTSUBSCRIPT italic_b end_POSTSUBSCRIPT end_ARG italic_I start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT ( italic_λ / italic_γ ) )

where C=I1(λ/γ)K1(λ)I1(λ)K1(λ/γ)𝐶subscript𝐼1𝜆𝛾subscript𝐾1𝜆subscript𝐼1𝜆subscript𝐾1𝜆𝛾C=I_{1}(\lambda/\gamma)K_{1}(\lambda)-I_{1}(\lambda)K_{1}(\lambda/\gamma)italic_C = italic_I start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT ( italic_λ / italic_γ ) italic_K start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT ( italic_λ ) - italic_I start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT ( italic_λ ) italic_K start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT ( italic_λ / italic_γ ), rb/Dν=λsubscript𝑟𝑏subscript𝐷𝜈𝜆r_{b}/D_{\nu}=\lambdaitalic_r start_POSTSUBSCRIPT italic_b end_POSTSUBSCRIPT / italic_D start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT = italic_λ. We then obtain the B𝐵Bitalic_B-dependent radial electric field:

Er(B)=B22πρ(ne)2[1rAI1(r/Dν)BK1(r/Dν)]subscript𝐸𝑟𝐵superscript𝐵22𝜋𝜌superscript𝑛𝑒2delimited-[]1𝑟𝐴subscript𝐼1𝑟subscript𝐷𝜈𝐵subscript𝐾1𝑟subscript𝐷𝜈E_{r}(B)=\frac{B^{2}}{2\pi\rho(ne)^{2}}[\frac{1}{r}-AI_{1}(r/D_{\nu})-BK_{1}(r% /D_{\nu})]italic_E start_POSTSUBSCRIPT italic_r end_POSTSUBSCRIPT ( italic_B ) = divide start_ARG italic_B start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG start_ARG 2 italic_π italic_ρ ( italic_n italic_e ) start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG [ divide start_ARG 1 end_ARG start_ARG italic_r end_ARG - italic_A italic_I start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT ( italic_r / italic_D start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ) - italic_B italic_K start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT ( italic_r / italic_D start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ) ]

The potential drop between the inner and outer electrode is evaluated by integrating Ersubscript𝐸𝑟E_{r}italic_E start_POSTSUBSCRIPT italic_r end_POSTSUBSCRIPT. ΔU=rarbEr𝑑rΔ𝑈superscriptsubscriptsubscript𝑟𝑎subscript𝑟𝑏subscript𝐸𝑟differential-d𝑟\Delta U=\int_{r_{a}}^{r_{b}}E_{r}drroman_Δ italic_U = ∫ start_POSTSUBSCRIPT italic_r start_POSTSUBSCRIPT italic_a end_POSTSUBSCRIPT end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_r start_POSTSUBSCRIPT italic_b end_POSTSUBSCRIPT end_POSTSUPERSCRIPT italic_E start_POSTSUBSCRIPT italic_r end_POSTSUBSCRIPT italic_d italic_r. Finally, the magnetoresistance is obtained by dividing ΔUΔ𝑈\Delta Uroman_Δ italic_U with I (Eq. 1 in the main text):

R(B)=B2lnγ2πρo(ne)2[1f(γ,λ)]𝑅𝐵superscript𝐵2𝛾2𝜋subscript𝜌𝑜superscript𝑛𝑒2delimited-[]1𝑓𝛾𝜆R(B)=\frac{B^{2}\ln\gamma}{2\pi\rho_{o}(ne)^{2}}[1-f(\gamma,\lambda)]italic_R ( italic_B ) = divide start_ARG italic_B start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT roman_ln italic_γ end_ARG start_ARG 2 italic_π italic_ρ start_POSTSUBSCRIPT italic_o end_POSTSUBSCRIPT ( italic_n italic_e ) start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG [ 1 - italic_f ( italic_γ , italic_λ ) ]

where f(γ,λ)=𝑓𝛾𝜆absentf(\gamma,\lambda)=italic_f ( italic_γ , italic_λ ) =

1lnγ{[I0(λ)I0(λ/γ)][(γ/λ)K1(λ)(1/λ)K1(λ/γ)]I1(λ/γ)K1(λ)I1(λ)K1(λ/γ)+[K0(λ)K0(λ/γ)][(γ/λ)I1(λ)(1/λ)I1(λ/γ)]I1(λ/γ)K1(λ)I1(λ)K1(λ/γ)}1𝛾delimited-[]subscript𝐼0𝜆subscript𝐼0𝜆𝛾delimited-[]𝛾𝜆subscript𝐾1𝜆1𝜆subscript𝐾1𝜆𝛾subscript𝐼1𝜆𝛾subscript𝐾1𝜆subscript𝐼1𝜆subscript𝐾1𝜆𝛾delimited-[]subscript𝐾0𝜆subscript𝐾0𝜆𝛾delimited-[]𝛾𝜆subscript𝐼1𝜆1𝜆subscript𝐼1𝜆𝛾subscript𝐼1𝜆𝛾subscript𝐾1𝜆subscript𝐼1𝜆subscript𝐾1𝜆𝛾\frac{1}{\ln\gamma}\Bigl{\{}\frac{[I_{0}(\lambda)-I_{0}(\lambda/\gamma)][(% \gamma/\lambda)K_{1}(\lambda)-(1/\lambda)K_{1}(\lambda/\gamma)]}{I_{1}(\lambda% /\gamma)K_{1}(\lambda)-I_{1}(\lambda)K_{1}(\lambda/\gamma)}\\ +\frac{[K_{0}(\lambda)-K_{0}(\lambda/\gamma)][(\gamma/\lambda)I_{1}(\lambda)-(% 1/\lambda)I_{1}(\lambda/\gamma)]}{I_{1}(\lambda/\gamma)K_{1}(\lambda)-I_{1}(% \lambda)K_{1}(\lambda/\gamma)}\Bigr{\}}start_ROW start_CELL divide start_ARG 1 end_ARG start_ARG roman_ln italic_γ end_ARG { divide start_ARG [ italic_I start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT ( italic_λ ) - italic_I start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT ( italic_λ / italic_γ ) ] [ ( italic_γ / italic_λ ) italic_K start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT ( italic_λ ) - ( 1 / italic_λ ) italic_K start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT ( italic_λ / italic_γ ) ] end_ARG start_ARG italic_I start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT ( italic_λ / italic_γ ) italic_K start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT ( italic_λ ) - italic_I start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT ( italic_λ ) italic_K start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT ( italic_λ / italic_γ ) end_ARG end_CELL end_ROW start_ROW start_CELL + divide start_ARG [ italic_K start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT ( italic_λ ) - italic_K start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT ( italic_λ / italic_γ ) ] [ ( italic_γ / italic_λ ) italic_I start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT ( italic_λ ) - ( 1 / italic_λ ) italic_I start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT ( italic_λ / italic_γ ) ] end_ARG start_ARG italic_I start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT ( italic_λ / italic_γ ) italic_K start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT ( italic_λ ) - italic_I start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT ( italic_λ ) italic_K start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT ( italic_λ / italic_γ ) end_ARG } end_CELL end_ROW

I0subscript𝐼0I_{0}italic_I start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT and K0subscript𝐾0K_{0}italic_K start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT are the modified Bessel functions of the first kind.

II acknowledgments

We thank Zhiyuan Sun, Alex Levchenko, Andrey Shytov, Shahal Ilani, Shaffique Adam for helpful discussion. YZ thanks Maelle Kapfer for assistance in data analysis.

This research is primarily supported by the Columbia University Materials Science and Engineering Research Center (MRSEC), through NSF grants DMR-2011738. K.W. and T.T. acknowledge support from the JSPS KAKENHI (Grant Numbers 21H05233 and 23H02052) and World Premier International Research Center Initiative (WPI), MEXT, Japan. The theory part of this work at MIT was supported by the Science and Technology Center for Integrated Quantum Materials, National Science Foundation grant No. DMR1231319; US-Israel Binational Science Foundation Grant No. 2018033.

III Data availability

The data that support the plots within this paper and other findings of this study are available from the corresponding author upon reasonable request.

IV Author contribution

Y.Z. and C.R.D. conceived the experiment. Y.Z.and O.M.G. fabricated the devices. Y.Z. performed the measurement and data analysis. H.G and L.L provides theory model. Y.Z., L.L. and C.R.D. wrote the manuscript with input from all authors.

Competing financial interests

The authors declare no competing financial interests.

References