betaFeedback

WorksPhysics and Astronomy

Isolating the natural edges of bilayer graphene in gate-defined mesoscopic devices

Abstract

We introduce a graphite-gated architecture for bilayer graphene devices in which the active device is completely isolated from the natural graphene edges. Using a single patterned graphite-gate layer, we realize a fully electrostatically defined Hall-bar. Longitudinal and Hall measurements reveal mesoscopic transport features, including Hall-effect quenching and magnetoresistance peaks associated with boundary scattering. The dependence of the mesoscopic features on the carrierdensity shows that the effective channel width increases with the Fermi level and the electrostatic confinement at the gate-defined boundaries, and indicates that the carriers scatter at the electrostatic boundary. Raman spectroscopy and Kelvin probe force microscopy suggest that this boundary is disordered due to the used fabrication methods. Comparably, the quantum mobility in a fieldeffect transistor fabricated with the same architecture is not limited by boundary scattering and the visibility of quantum oscillations down to 4 mT suggests a record value of 2.5 x 10^6 cm2/Vs.

Introduction

In the field of Van-der Waals materials, device reproducibility and yield impose major limitations to the exploration of the underlying physics and to their technological relevance. The complex assembly and fabrication schemes give rise to an immense number of parameters that have to be controlled at the same time. In bilayer graphene (BLG) devices, the mastery of this parameter space is, to some extent, reflected by the opening of a uniform band-gap under the application of an electric field perpendicular to the BLG sheet. This gap is theoretically expected [1] and routinely achieved in experiments [2, 3], but its cleanliness as measured by the in-gap resistance is of varying quality and strongly depends on the fabrication details [4]. In the presence of a clean band-gap, the resistance increases beyond the Teraohm regime when the Fermi energy is tuned into the gap. In a device, conducting regions and insulating regions can be separated with patterned gate electrodes which tune the Fermi level into the band or the band gap, respectively. This allows to electrostatically define nanostructured devices. These structures have replaced etch-defined BLG architectures [5], where the device was defined by etching away the surrounding material and allowed for measurements of clean quantum point contacts [6] and quantum dots [7] [8]. However, even in these devices, the natural flake edge may still provide conducting paths, giving rise to a large measured conduction that is independent of the nanostructured device. Furthermore, the effect of the electrostatic edges on the transport properties has remained largely unexplored.

We have conceived a new architecture of dual-gated BLG devices facilitating the design of a large variety of them. The devices are fully surrounded by electrostatic edges, preventing any transport connection between ohmic contacts and natural flake edges, thus forcing the current to pass the device under study. Further, the active device area remains unaffected by exposure to polymers or high-voltage electron beams used in typical lithography processes. Taking into consideration the concerns about Cr and Ti sticking layers [9] and following recent trends in the field [10, 11], we move away from metal as top-gate electrode, but instead rely on graphite layers.

After introducing the design architecture, we first demonstrate the cleanliness of the achieved band-gaps and the high mobility of the BLG charge carriers by performing magnetotransport measurements on a simple field-effect transistor (FET) device geometry. We then introduce a gate-defined Hall-bar, with which we characterize in more detail the remaining effects of the electrostatic edges in gate-defined devices.

Device architecture

Our device design relies on the band-gap that opens in BLG upon application of a perpendicular displacement field. In contrast to the successful demonstrations of gate-defined quantum point contacts and quantum dots in BLG [6, 7] [8, 12], we don’t rely on the band gap only in close vicinity of the active area. Instead, we tune the whole area of the BLG flake outside our active device into the resistive regime provided by the band-gap. Our device architecture is shown in Figure 1(b,Figure 1c) in the form of a FET and a Hall-bar device. The BLG flake is encapsulated between two hexagonal boron nitride (h-BN) flakes and, in an additional step, a few-layer graphite flake is stacked onto the top h-BN flake. The device geometry is then carved out of this graphite flake, separating the stack into the active device region and a region outside, which will be tuned into the band-gap using electrostatic gating. The latter effectively creates an electrostatic mesa. One more h-BN flake finally covers the graphite flake, allowing to contact all the graphite regions individually via small etched holes and a metal fan-out. Details on the stacking and device fabrication are described in Section VII and S1. The whole stack is visible in Figure 1(a).

Measured devices

Figure 1. Measured devices. a) Optical microscope image of the finished devices and schematics of the setup. b) Atomic force micrograph of the Hall-bar after the patterning of the graphite top gate, in yellow the position of the ohmic contacts, in blue the contacts to the Hall-bar gate, and in light blue the cut in the BLG to facilitate equilibration. The Hall-bar gate is separated from the FET gate by a 90 nm-wide gap. c) Atomic force micrograph of the 800 nm-wide FET, in yellow the position of the ohmic contacts.

The described architecture brings the advantage that conduction only occurs within the electrostatically defined mesa and, in particular, does not rely on the electrostatic opening of a clean band-gap along the flake edges. In contrast to previous BLG quantum dot designs [7] [8], this prevents potential parallel conduction paths created by imperfect edges. For example, due to the way such edges are created by exfoliation, they are not necessarily uniform, and may partially consist of monolayer graphene, where no band-gap opens with a displacement field. Here, no conduction path connects our ohmic contact to the natural flake edges, thus overcoming this problem and increasing the device yield.

Field-effect transistor

We first characterize our new architecture using a simple device design, namely a FET as shown in Figure 1(c), which still allows us to study the magnetotransport properties of the BLG [13, 14]. In our case, the FET consists of two ohmic contacts separated by a graphite region which covers the whole surrounding. This excludes potential current paths along natural flake edges whenever the gated region is tuned into the band-gap, yielding the configurations of large resistance shown in Figure 2(a). There, the back-gate voltage VBGV_{\mathrm{BG}} and the graphite top-gate voltage VFETV_{\mathrm{FET}} are swept while a constant voltage VSD=750 μVV_{\mathrm{SD}} = 750\ \mu\mathrm{V} is applied between the source and drain contacts. The resistance is large at the back-gate voltage VBG,0V_{\mathrm{BG,0}} and the top-gate voltage VTG,0V_{\mathrm{TG,0}} where the BLG is charge-neutral. The resistance reaches the TΩ\mathrm{T} \Omega regime along the diagonal line, as the Fermi level lies in the band-gap opened by the perpendicular displacement field D=Ct/2ϵ0(VTG−VTG,0)−Cb/2ϵ0(VBG−VBG,0)D = C_{\mathrm{t}}/2\epsilon_{0}(V_{\mathrm{TG}} - V_{\mathrm{TG,0}}) - C_{\mathrm{b}}/2\epsilon_{0}(V_{\mathrm{BG}} - V_{\mathrm{BG,0}}). Here, Ci=ϵ0ϵi/diC_i = \epsilon_{0}\epsilon_i/d_i, with did_i and ϵi\epsilon_i the h-BN thickness and relative dielectric constant of the top (i=ti = \mathrm{t}) or bottom (i=bi = \mathrm{b}) flake and ϵ0\epsilon_0 the vacuum dielectric constant, respectively.

Band-gap measurements of the FET

Figure 2. Band-gap measurements of the FET. a) Two-terminal resistance as a function of VFETV_{\mathrm{FET}} and VBGV_{\mathrm{BG}}. The diagonal, highly insulating line appears when the Fermi level lies in the gap opened by the displacement field. Inset: measured current as a function of VBGV_{\mathrm{BG}} at a fixed top gate voltage as indicated by the red line. b) Finite bias spectroscopy at fixed displacement field as indicated by the dotted line in a).

To test the uniformity and homogeneity of the band gap, we extract the resistance at constant displacement field, while tuning the Fermi energy through the band-gap by changing the density n=Ct/e(VTG−VTG,0)+Cb/e(VBG−VBG,0)n = C_{\mathrm{t}}/e(V_{\mathrm{TG}} - V_{\mathrm{TG,0}}) + C_{\mathrm{b}}/e(V_{\mathrm{BG}} - V_{\mathrm{BG,0}}) from positive (hole charge carriers) to negative (electron charge carriers). Meanwhile, we step VSDV_{\mathrm{SD}} in order to measure the size of the band-gap [4], as shown in Figure 2(b). The uniform diamond with its sharp transition from a high in-gap resistance to low-resistive conduction outside the gap suggests a complete suppression of transport channels in the insulating regime. These results demonstrate the absence of common fabrication-related issues named above, which typically lead to impurities and defects in the measured region of the flake.

We now characterize the magnetotransport of the FET in the conducting regime, when the Fermi level lies in the valence or conduction band. In the following measurements, we either work at constant displacement field or constant density around operation points VBG∗V_{\mathrm{BG}}^{*} and VTG∗V_{\mathrm{TG}}^{*}. For this, the top-gate and back-gate have to be moved simultaneously following the relations

VTG−VTG∗=−CbCt(VBG−VBG∗)(for n=const)(1)V_{\mathrm{TG}}-V_{\mathrm{TG}}^{*}=-\frac{C_b}{C_t}\left(V_{\mathrm{BG}}-V_{\mathrm{BG}}^{*}\right) \qquad(\text{for } n=\text{const}) \tag*{(1)}
VTG−VTG∗=CbCt(VBG−VBG∗)(for D=const)(2)V_{\mathrm{TG}}-V_{\mathrm{TG}}^{*}=\frac{C_b}{C_t}\left(V_{\mathrm{BG}}-V_{\mathrm{BG}}^{*}\right) \qquad(\text{for } D=\text{const}) \tag*{(2)}

, with the capacitance ratio extracted from the measurements in Figure 2(a). We again apply a constant voltage VSD=125 μVV_{\mathrm{SD}}=125\,\mu\mathrm{V} and measure the current flowing between the contact pair in a two-terminal setup. With the chosen FET design, as soon as the states in the gated regime are quantized into Landau levels, edge channels form and surround the holes in the graphite layer defining the contact area. Thus, at constant filling factor, the device becomes insulating. However, current flows when the Fermi energy in the gated region is close to a Landau level, within its broadening. The measured resistance is equivalent to the longitudinal resistance expected in a Corbino geometry. The etched regions without top-gate, which directly surround the contacts, remain conductive at all magnetic field values, as etch inhomogeneities S2 prevent Landau quantization.

This effect is important in our measurements, as it allows us to probe the Landau levels in the gated region. We set the displacement field to D=0D=0 and measure a Landau fan as shown in Figure 3. The breaking of the fourfold symmetry due to spin and valley is clearly seen up to filling factor ν=12\nu=12 at magnetic fields B≤0.5 TB\leq0.5\ \mathrm{T}. This indicates long quantum lifetimes τq\tau_q causing the broadening ΓLL∼ℏ/τq\Gamma_{\mathrm{LL}}\sim\hbar/\tau_q of Landau levels to be much lower than their separation even at these relatively low fields. The separation of orbital levels is determined by the cyclotron frequency to ℏωc\hbar\omega_c, while the Zeeman energy giμBBg_i\mu_BB sets the separation of spin- or valley-split levels, with gig_i the spin- or valley g-factor, respectively. The spin-splitting between the first and second Landau level is visible down to B=200 mTB=200\,\mathrm{mT}, and we resolve Shubnikov–de Haas (SdH) oscillations down to B=4 mTB=4\,\mathrm{mT} for filling factor ν=−8\nu=-8, already identified as the most robust filling factor at low magnetic fields in previous studies [15]. Following these observations, we estimate a quantum lifetime of τq∼47 ps\tau_q\sim47\,\mathrm{ps} and a quantum mobility of μq∼2 500 000 cm2/Vs\mu_q\sim2\,500\,000\ \mathrm{cm^2/Vs}.

Magnetotransport measurement of the FET at $D=0$ V/nm

Figure 3. Magnetotransport measurement of the FET at D=0D=0 V/nm. a) Landau fan from 0 T to 2 T, with the visible filling factors labeled in black. b) Zoom-in of the low field and low density region.

Hall-bar

We make use of the flexibility in device design provided by our new architecture to produce a fully gate-defined Hall-bar, visible in Figure 1(b). The displacement field and the density in the Hall-bar are tuned by the voltage VHallV_{\mathrm{Hall}} applied to the central graphite gate. The outside region of the Hall-bar is the same graphite region as the FET gate, hence it is tuned by VFETV_{\mathrm{FET}}. Using the results of the FET measurements, we keep the Fermi energy in that outside region in the band gap. In this setting, we sweep VBGV_{\mathrm{BG}} versus VHallV_{\mathrm{Hall}} in order to determine the capacitance of the Hall-bar gate on the BLG underneath. We measure resistances in the GΩ\Omega regime, again indicating a high device quality with a clean band-gap, and also implying that no transport occurs underneath the ungated 90 nm gap between the Hall-bar gate and the FET-gate.

We apply a magnetic field perpendicular to the BLG flake to measure the transverse and longitudinal resistances RxyR_{xy} and RxxR_{xx} shown in Figure 4(a) and (b), respectively. The typical signatures of the integer quantum Hall effect and SdH oscillations are visible at magnetic fields above about 0.5 T. The low-field data shown in Figure 4(c) and (d) displays an increase of ρxx\rho_{xx} with magnetic field up to a peak, followed by a decreases, while RxyR_{xy} exhibits a decreased slope around zero field. The flattening of the transverse resistance occurs due to the quenching of the Hall effect in the ballistic regime, when the Lorentz force does not bend the trajectories of the injected carriers enough for them to be collected by the neighboring voltage probes [16, 17].

Magnetotransport measurements of the Hall-bar

Figure 4. Magnetotransport measuremenet of the Hall-bar at D=0.1 V/nmD = 0.1\ \mathrm{V}/\mathrm{nm}. a) Landau fan in RxxR_{xx}. b) Landau fan in RxyR_{xy}. c) Low-field trace of RxxR_{xx} taken at nHall=1.23×1012 cm−2n_{\mathrm{Hall}} = 1.23 \times10^{12}\ \mathrm{cm}^{-2}, showing the mesoscopic magnetoresistance peaks. d) Low-field trace of RxyR_{xy} taken at nHall=1.23×1012 cm−2n_{\mathrm{Hall}} = 1.23 \times10^{12}\ \mathrm{cm}^{-2}, showing the quenching of the Hall resistance.

We associate the increase in the longitudinal resistance to the narrow channel compared to the mean free path as estimated further below. In this mesoscopic regime, specular edge scattering in combination with cyclotronic motion deflect the charge carriers back to the source contact [17, 18]. Only at larger fields, when the cyclotron orbit becomes smaller than the width of the Hall-bar, the carriers are forward scattered and the magnetoresistance decreases again. The thin metal theory [18–20] yields the value of the magnetic field Bmax⁡B_{\max} at which the longitudinal resistance peaks as

Bmax⁡=0.55ℏkFeW(3)B_{\max}=0.55\frac{\hbar k_{\mathrm{F}}}{eW} \tag*{(3)}

where ℏ\hbar is the reduced Planck constant, ee is the electron charge, WW is the channel width, and kF=4gπnk_{\mathrm{F}}=\sqrt{\frac{4}{g}\pi n} is the Fermi wave vector. In BLG, the fourfold spin- and valley degeneracy leads to kF=πnk_{\mathrm{F}}=\sqrt{\pi n}. From this equation, we determine the effective width WeffW_{\mathrm{eff}} of our Hall-bar channel as a function of density and displacement field, as shown in Figure 5(a). The data follows a power-law relation, Weff∝nβW_{\mathrm{eff}}\propto n^\beta, consistent with a near-parabolic confinement potential, with β\beta being wavelength dependent on the displacement field (see Supplementary Information). The low-field behavior of the longitudinal resistivity ρxx\rho_{xx}, which we now extract from RxxR_{xx} using the effective width, is dominated by edge scattering. In particular, the mean of ρxx\rho_{xx} measured in our devices at low fields deviates from its zero-field value, a behavior that differs from the characteristic response observed in large devices. This allows us to further analyze the edge scattering. We extract ρ0,avg\rho_{0,\mathrm{avg}} from the mean of the signal measured at fields above 0.5 T but below Landau quantization and compare it with the measured value ρ0,meas\rho_{0,\mathrm{meas}} measured at zero field. The measured deviation (see Supplementary Information, S6) can be parametrized by a scattering parameter pp [17, 18] via

Hall-bar analysis showing effective width and scattering parameter versus Hall density for different displacement fields

Figure 5. Hall-bar analysis. a) Effective width Weff(n)W_{\mathrm{eff}}(n) extracted as a function of density and plotted for different displacement fields. b) Scattering parameter pi(n)p_i(n) calculated for each density and plotted for different displacement fields.

ρeff=ρ0+ρ0l0(1−p)W=ρ0+ℏkFne2(1−p)W(4)\rho_{\mathrm{eff}}=\rho_0+\rho_0l_0\frac{(1-p)}{W}=\rho_0+\frac{\hbar k_{\mathrm{F}}}{ne^2}\frac{(1-p)}{W} \tag*{(4)}
⇔p=1−e2ℏπ(ρ0,meas−ρ0,avg)Weff2nL(5)\Leftrightarrow p=1-\frac{e^2}{\hbar\sqrt{\pi}}\frac{(\rho_{0,\mathrm{meas}}-\rho_{0,\mathrm{avg}})W_{\mathrm{eff}}^2\sqrt{n}}{L} \tag*{(5)}

Here, ρeff\rho_{\mathrm{eff}} is the effective resistivity of the channel, ρ0l0=ℏkF/ne2\rho_0l_0=\hbar k_{\mathrm{F}}/ne^2, LL is the distance between the Hall voltage probes, and in the last line, we replaced the sample width WW with the density-dependent width WeffW_{\mathrm{eff}}. The scattering parameter takes values between 0 and 1, indicating completely diffusive scattering, and 1 denoting exclusive specular scattering. The data in Figure 5(b) shows a trend from predominant specular scattering at low densities to slightly increased diffusive scattering at high density. To gain some insight, let us first comment on the mesoscopic regime we are working in. Even though the Hall-bar width is of the same order of magnitude as the Fermi wavelength, we do not observe quantization effects typical for a quantum point contacts. We suspect that only a few modes are occupied, which are mixed by scattering and thus cannot be resolved individually. Nevertheless, we rationalize the observed evolution of pp by means of the wave-functions expected in a one-dimensional near-parabolic confinement potential, as shown in Figure 6. Using this model, we note that the probability amplitude of the wavefunctions at the lithographic edge of the Hall-bar is close to zero for the ground and first excited states, consistent with the extracted pp close to 1 at low densities. With increasing mode number, the wavefunction amplitudes start to be finite at the lithographic Hall-bar edge, in agreement with pp starting to decrease at higher density. This suggests that the mixing of the subbands is caused predominantly at the edges, while the bulk supports long mean free paths in accordance with the FET measurements.

Transverse confinement potential and quantized modes

Figure 6. Transverse confinement potential and quantized modes. The vertical dashed lines indicate the geometric barrier edges while the horizontal dotted line marks the barrier height VbarV_{\mathrm{bar}}. The calculated transverse eigenenergies EnE_n are shown as horizontal levels for each mode number n=0,4,8,12,16n = 0, 4, 8, 12, 16. The corresponding probability density ∣ψn(y)∣2|\psi_n(y)|^2, scaled for visualization, is plotted with a vertical offset set by its eigenenergy EnE_n, with a local probability scale shown alongside each mode (orange). The dash-dotted horizontal line indicates the Fermi energy EFE_{\mathrm{F}}. Modes approaching the barrier energy exhibit an increasingly extended transverse probability distribution and enhanced penetration into the classically forbidden region.

Using the insights obtained so far, we now estimate the carrier mobility using the resistivity ρ0,avg\rho_{0,\mathrm{avg}}. The Hall mobility extracted at densities above 2×1011 cm−22 \times10^{11}\ \mathrm{cm}^{-2} amounts to 200 000 cm2/Vs\mathrm{cm}^{2}/\mathrm{Vs}, comparable with state-of-the-art devices [4] [12, 21]. The corresponding mean free path of about 1 μm\mu\mathrm{m} to 3 μm\mu\mathrm{m} exceeds the Hall-bar length, suggesting that the diffusive model used to extract transport mobilities is not suitable for describing our device. Nevertheless, the temperature-dependent SdH oscillations (see Supplementary Information, S7) allow us to extract a value for the quantum mobility of roughly 30 000 cm2/Vs\mathrm{cm}^{2}/\mathrm{Vs}, much lower than the one found for the FET, indicating that the edges in the Hall-bar induce small-angle scattering.

Origin of the disorder

The FET measurements revealed record high values of the quantum mobility, indicating exceptional cleanliness of our flakes. Compared to that, the Hall-bar showed significant edge scattering, demonstrating the importance of understanding the electrostatic edges of gated device designs. In our device, the edges are defined by the small gap etched into the graphite top gate, separating the Hall-bar gate from the FET-gate. We, therefore, set out to analyze in more detail the effect of the applied etching process. We test the transport properties of the BLG in regions where the top graphite gate has been etched (see Supplementary Information). Magnetotransport measurements show that resistance peaks corresponding to Landau level gaps are minimally pronounced, indicating the presence of disorder in these regions.

Further, we perform Raman and Kelvin probe force microscopy (KPFM) measurements to analyze the crystal quality and the local surface potential of h-BN and graphite flakes treated with plasma etching. For this, h-BN and graphite flakes are exfoliated on SiO2_2 substrates and treated with a plasma process essentially equal to the one used for our device fabrication. It consists of two (instead of four used in the device) cycles of 20 s of N2_2 plasma at 100 W and 40 s of O2_2 plasma at 60 W with a 10 s CHF3_3/O2_2 plasma step at 60 W.

The Raman measurements performed on the graphite sample are shown in Figure 7(a). The orange solid line represents the Raman spectrum of the graphite flake before being exposed to the plasma treatment. The measured intensity exhibits only one narrow peak at around 1582 cm−1^{-1} attributed to the G-band that arises from in-plane motion of the carbon atoms in bulk graphite. All the spectra are normalized to the intensity measured at the respective G-peak. After the plasma treatment, two more peaks appear in the Raman spectrum, as seen in the pink solid line in Figure 7(a): the D-band and D’-band, around 1350 cm−1^{-1} and 1620 cm−1^{-1} respectively. Both peaks are attributed to defects in the graphite [22], and their intensity relative to the G-band is commonly used to evaluate the defect density in graphitic materials [23, 24]. In our case, we attribute their appearance to defects, edge disorder, and contaminants introduced by the plasma treatment.

Effect of the plasma treatment on a thin graphite and h-BN flake

Figure 7. Effect of the plasma treatment on a thin graphite and h-BN flake, respectively. a) Raman spectra of as-exfoliated (orange), plasma-treated (pink) and annealed (purple) graphite flake on top of SiO2_2/Si. b) Intensity of the D and D′ peaks relative to the G-peak, ID_\mathrm{D}/IG_\mathrm{G} and ID′_\mathrm{D'}/IG_\mathrm{G}, respectively, used to quantify the disorder after plasma treatment and annealing. c) Raman spectra of as-exfoliated (orange) and plasma-treated (pink) h-BN flake on top of SiO2_2/Si. d) Peak position and e) full width at half maximum of the E2g_{2g} mode before and after plasma treatment.

We employ an annealing treatment, 5 h 350 ∘^\circC in a N2_2/H2_2 atmosphere under a pressure of 60 mTorr, to to test whether the observed changes are reversible. The solid purple line in Figure 7(a) shows the Raman spectrum acquired after the annealing procedure. To facilitate the comparison of the D and D’ peaks before and after annealing, we plot their intensity relative to the G-peak in Figure 7(b). Clearly, the relative intensity of both peaks decreases after annealing, indicating a partial removal of the plasma-induced disorder localized at the sample surface. The reduction in disorder is also inferred from the evolution of the G-band linewidth, which increases after the plasma exposure and subsequently decreases after the annealing treatment, as qualitatively seen in Figure 7(a) and quantitatively analyzed in Figure S15 of the Supplementary Information. However, even after annealing, the D and D’ peaks persist and the G-band’s linewidth remains somewhat larger, suggesting that the plasma treatment also introduces permanent structural defects that remain after the used annealing treatment. In addition, the ratio of the D versus D’ peak gives insights into the defect type [24]. The measured ratio of around 3 before and after annealing points to boundary-like defects, such as, for example, grain boundaries [23].

Evolution of the Raman peak position and FWHM

Figure S15. (a) Evolution of the Raman peak position and (b) FWHM before plasma treatment, after (N2+O2)2+CHF3/O2(\mathrm{N}_2 + \mathrm{O}_2)_2 + \mathrm{CHF}_3/\mathrm{O}_2 plasma exposure, and after annealing of the measured graphite flake.

Also the h-BN flake was first measured as exfoliated and then after the plasma treatment, with Raman spectra shown in Figure 7(c). The observed peak around 1366 cm−1^{-1} is characteristic for h-BN and corresponds to the E2g_{2g} phonon mode [25, 26]. After the plasma treatment, the E2g_{2g} peak exhibits a red shift, as shown in Figure 7(d), which can be attributed to a modification of the h-BN lattice [27, 28]. In our case, the plasma treatment might introduce strain, defects, and a reduction of the h-BN thickness. The reduction of thickness is further supported by the change in the optical contrast of the flake, as shown in Figure S16 of the Supplementary Information. In addition to the red shift, the E2gE_{2g} peak becomes wider, as shown in Figure 7(e). This indicates a reduction of the phonon lifetime due to additional scattering channels arising, for example, from the defects. Also here, we attempt at reversing plasma-induced defects with an annealing treatment. However, the observed signatures point towards increased contamination, possibly due to redistributing residues from the exfoliation tape (see Supplementary Information).

Raman spectra and optical images of an h-BN flake

Figure S16. a) Raman spectra of as-exfoliated (orange), plasma-treated (pink) and annealed (purple) h-BN flake on top of SiO2/Si\mathrm{SiO}_2/\mathrm{Si}. b) Optical images of the measured h-BN flake before and after the (N2+O2)2+CHF3/O2(\mathrm{N}_2+\mathrm{O}_2)_2+\mathrm{CHF}_3/\mathrm{O}_2 plasma treatment.

To get insights into whether the induced disorder arises from the oxygen and fluorinated plasmas [27–31], we perform similar measurements using a different etching process. Namely, we replace the final CHF3_3/O2_2 step with a 10 s N2_2 plasma treatment at 100 W, which should leave a clean and inert surface of the h-BN flake [32]. The Raman measurements did not show significant differences, as presented in Figure S19 of the Supplementary Information.

Raman spectra, peak position and FWHM, and optical images of h-BN flake

Figure S19. a) Raman spectra of as-exfoliated (red), plasma-treated (blue) and annealed (green) h-BN flake on SiO2_2/Si. b) Peak position and FWHM of the E2g_{2g} mode before and after plasma treatment and subsequent annealing. c) Optical images of the measured h-BN flake before and after the (N2_2 + O2_2)2_2 + N2_2 plasma treatment.

In addition to the Raman measurements, KPFM measurements provide insight into the work function differences of the substrates under study. We compare two samples with h-BN flakes exfoliated on SiO2_2 chips, one of them exposed to the CHF3_3/O2_2 plasma process, and one to the N2_2 process. In the former, we observe a broader and stronger surface-potential perturbation around the flakes, indicating larger work function differences between the exposed h-BN and SiO2_2. Comparably, the second recipe results in a more homogeneous potential landscape. These observations are in agreement with literature [27, 30, 31]. The measurements do not allow to differentiate whether the two recipes attack the h-BN flake in distinct ways, or whether the observed differences are confined to the boundaries of the flake, which are known to be affected by plasma processes in general [28, 29].

In view of the above findings and considering that the patterned graphite is completely removed, as well as a few layers of the h-BN, there are two different but potentially coexisting explanations for the disorder observed in the transport measurements. Possibly, the h-BN gets disordered and structurally damaged by the etching process, inducing a disordered potential landscape in the BLG. Alternatively, the disorder in the potential landscape originates from the contaminated and potentially functionalized edges of the etched graphite gates.

Conclusion

We introduce a new architecture for graphite-gated BLG devices that enables the definition of active regions completely isolated from any natural flake edge. This architecture yields a highly resistive and clean band gap, together with a large quantum mobility. It also enables new device geometries, and we demonstrate the straightforward realization of a gate-defined Hall-bar, a geometry that has so far received limited attention in graphite-gate-defined BLG devices.

In contrast to a FET, the Hall-bar geometry allows for the measurement of both the longitudinal and transverse resistances, yielding complementary information. In our device, the Hall-bar operates in the mesoscopic regime, with the mean free path comparable to the device dimensions. We are therefore sensitive to transport phenomena such as Hall-effect quenching and boundary scattering. This allows us to investigate the role of electrostatically defined edges, an aspect of gate-defined devices which has not been addressed so far.

We find that the mobility in the Hall-bar measurements is limited by scattering associated with disorder in the electrostatic potential. Raman spectroscopy and KPFM measurements indicate that this originates from defects and disorder introduced by the fabrication process. Meanwhile, measurements of the FET demonstrate the extraordinary quality of the bulk, with a quantum mobility approaching 2.5×106 cm2/Vs2.5 \times10^6\ \mathrm{cm^2/Vs}.

Methods

Devices fabrication. The devices were fabricated using a stack comprising a graphite back gate, h-BN, bilayer graphene, six-layer graphite, and h-BN. The stack was assembled with a standard dry transfer technique. The top-gate patterning involved standard lithographic techniques followed by an reactive-ion etching (RIE) step. The etching process consisted of four cycles of 20 s N2_2 plasma at 100 W and 40 s O2_2 plasma at 10 W, followed by 10 s of CHF3_3/O2_2 plasma at 60 W. Edge ohmic contacts and Hall-bar gate contacts were realized in the same fabrication step, in which a lithographic mask was used to selectively etch the h-BN by RIE with SF6_6 plasma at 6 W, followed by 15 s of CHF3_3/O2_2 plasma at 60 W to etch the unaffected bilayer graphene and expose its edges. The RIE step was followed by metal evaporation of a Cr/Pd/Au stack with thicknesses of 10 nm, 30 nm, and 40 nm, respectively.

Measurements. All the measurements were performed in a BlueFors dilution cryostat at a base temperature of 10 mK. Current-biased measurements were performed by applying a voltage and placing a 200 MΩ resistor in series with the device. The current was measured using a current-to-voltage converter connected to a lock-in amplifier. Lock-in measurements were performed using three synchronized Zurich Instruments lock-in amplifiers to measure VxxV_{xx}, VxyV_{xy} and the current. The voltages VxxV_{xx} and VxyV_{xy} were amplified using differential voltage amplifiers before being fed into the lock-in amplifiers.

Raman spectroscopy. Raman spectra were acquired using a 515 nm COBOLT laser with a power of 1 mW for the graphite samples and 400 μW m for the h-BN samples. The laser was focused onto the samples using a 50x objective with a numerical aperture of 0.60. The Raman spectra were collected with a Horiba T6400 triple monochromator with a spectral resolution of 0.5 cm−1^{-1} and equipped with a N2_2 cooled charge coupled device (CCD) detector. The Raman spectra of the graphite samples were collected with an integration time of 60 s and averaged over 3 scans, whereas the Raman spectra of the h-BN samples were acquired with an integration time of 120 s and averaged over 3 scans.

Acknowledgments

We thank Michael Steinacher for his valuable contributions to the optimization of the measurement setup. We thank Clara Galante, Jessica Richter, Tijl Degroote, Max Ruckriegel, Markus Niese, Marta Perego and Artem Denisov for helpful input and discussions around building first stacks and exploring bilayer graphene physics.

We acknowledge financial support from the Swiss State Secretariat for Education, Research and Innovation, the Quantum Transitional Call, and from the Swiss National Science Foundation through grant number 200021-231373.

Data availability

The data that support the findings of this study are openly available in ZENODO at https://zenodo.org/doi/xxx.

References

Device fabrication

The devices were fabricated using a stack comprising a graphite back gate, hexagonal boron nitride (h-BN), bilayer graphene (BLG), six-layer graphite, and h-BN.

Flake exfoliation and selection

The flakes used for the stack were exfoliated from graphite flakes and h-BN crystals. For exfoliation, Nitto blue tape ELP BT-130E-SL was used. After several tape-folding steps, the exfoliated flakes were transferred onto Si/SiO2_2 (285 nm) chips, which were treated differently depending on the exfoliated material. For graphite flakes, the chips were exposed to oxygen plasma for 2 minutes at 20 W, whereas for h-BN flakes, the chips were heated on a hot plate for 5 minutes at 100°C.

Subsequently, the treated silicon chips were flipped and brought into contact with the tape. For h-BN, the tape was peeled off after a few minutes of contact. For graphite, the chips with the tape on them were heated on a hot plate for 35 s at 75°C before peeling.

After exfoliation, the chips were transferred into a glovebox with a nitrogen atmosphere. To identify suitable candidates for the stack, the chips underwent a screening procedure in which h-BN and graphite flakes were inspected using a fluorescence microscope and evaluated in three different modes: bright field, dark field, and high-contrast mode. In this way, the cleanliness of the flakes and the presence of cracks or folds could be assessed. Suitable flakes were then selected based on their thickness and area. The thickness of h-BN was estimated from the colour, whereas the contrast of the green component of the reflected light was used to infer the number of graphene layers. This was done to identify bilayer graphene as the active material and six-layer graphene for the top gate.

Stack assembly

A standard polymer-based stacking procedure was employed. PDMS stamps covered with a polycarbonate (PC) film were used and operated through a three-axis micromanipulator inside the glovebox. Initially, the first part of the stack was assembled: the BLG was fully encapsulated between two h-BN flakes, terminating with a graphite back gate. The assembled stack was then transferred onto a pre-patterned marker chip by increasing the microscope stage temperature up to 180°C, above the glass-transition temperature of the PC film. At this stage, the stamp was detached from the chip, leaving the stack covered with melted PC.

The chip was then cleaned overnight in chloroform and rinsed with isopropanol (IPA). After inspection under the microscope to ensure that any visible PC had been removed, the chip underwent an annealing step of 5 hours, with a peak temperature of 365°C, in an atmosphere of 95% Ar and 5% H2_2 at a pressure of 60 mbar. After this step, another overnight cleaning step in chloroform was performed, followed by the transfer of the six-layer graphite flake serving as the top gate. The same stacking procedure and cleaning steps used for the first stack were also followed after this flake transfer.

Top-gate patterning

The top-gate patterning involved standard lithographic techniques followed by a reactive-ion etching (RIE) step.

The lithographic PMMA mask was exposed by electron-beam lithography using a RAITH EBPG 5150 and developed in a solution of IPA and deionized water at 0°C for 5 minutes. The same procedure was used for all subsequent lithography steps.

The RIE step involved the alternating use of N2\mathrm{N}_{2} plasma and O2\mathrm{O}_{2} plasma at 100 W and 10 W, respectively, followed by 10 s of CHF3/O2\mathrm{CHF}_{3}/\mathrm{O}_{2} plasma at 60 W. After this step, the mask was stripped in DMSO at 95°C for 1 hour, followed by further cleaning in acetone and IPA. Atomic force microscopy (AFM) images confirmed that the six-layer graphene was fully etched where desired, as evidenced by 7 nm of over-etching into the h-BN layer.

The top gate was etched to pattern the Hall bar and the FET. Circular areas were also etched to host the ohmic contacts.

The choice of this specific plasma sequence for the etching process resulted from several tests inspired by the work of [1]. However, unlike the system used in that study, our process employs reactive-ion etching rather than inductively coupled plasma (ICP). It therefore combines the plasma-chemical effects reported by [1] with the physical effects of ion bombardment arising from the DC self-bias developed in the RIE system. In addition, we demonstrate that, under these conditions, not only h-BN but also graphite is etched, most likely because of the enhanced ion-bombardment contribution. Alternating the O2\mathrm{O}_{2} and N2\mathrm{N}_{2} plasma steps may also help remove carbonaceous or nitrogen-containing by-products, such as graphitic carbon nitride compounds, which could otherwise inhibit the subsequent etching steps.

Subsequently, another h-BN flake was transferred on top of the patterned gate layer. This flake serves as a support to bridge over the top gate, enclosing the region intended for the ohmic contacts and the Hall bar gate.

AFM scan of the Hall bar device after the top gate patterning, profile trace of the etched region, 3D view of the AFM scan, and 3D sketch of the stack

Figure S1. AFM scan of the Hall bar device after the top gate patterning (top left). Profile trace of the etched region across the graphite top gate and the etched h-BN (top right). 3D view of the AFM scan (bottom left). 3D sketch of the stack (bottom right).

Ohmic contacts and fan-out lines

Edge ohmic contacts and Hall bar gate contacts were realized in the same fabrication step. A lithographic mask was used to selectively etch the h-BN by RIE with SF6\mathrm{SF}_{6} plasma at 6 W, followed by 15 s of CHF3/O2\mathrm{CHF}_{3}/\mathrm{O}_{2} plasma at 60 W to etch the unaffected bilayer graphene and expose its edges. The RIE step was followed by metal evaporation of a Cr/Pd/Au stack with thicknesses of 10 nm, 30 nm, and 40 nm, respectively.

Subsequently, another PMMA mask was followed by an etching step used to interrupt the circular path around the ohmic contacts in the Hall bar by etching away the BLG connecting the contact to the top gate.

Finally, a lithographic step followed by Cr/Ti/Au metallization with thicknesses of 2 nm, 3 nm, and 180 nm, respectively, was used to realize the fan-out lines to the bond pads.

Edge channels and equilibration

In the context of the quantum Hall effect, the presence of device regions with different densities—and hence, different filling factors—would limit the contact equilibration of the transport channel of interest or alter the longitudinal transport, leading to characteristic checkerboard patterns. However, in our measurements, we do not observe these features, likely due to the etched top-gate area, which induces a highly disordered potential on the BLG underneath as a consequence of the damaged superficial h-BN layers exposed to the RIE plasma. To validate this hypothesis, we performed a Landau fan two-terminal measurement in a region where the graphite top gate was removed.

Landau fan measured at $D=0\ \mathrm{V}/\mathrm{nm}$ in a two-terminal configuration

Figure S2. Landau fan measured at D=0 V/nmD=0\ \mathrm{V}/\mathrm{nm} in a two-terminal configuration in the area where the top-graphite gate was fully etched.

Quantum mobility extraction

From the onset of Shubnikov–de Haas (SdH) oscillation, we can roughly estimate a lower bound for the quantum mobility following the relation ωcτq∼1\omega_c\tau_q \sim1, where ωc\omega_c is the cyclotron frequency, and τq\tau_q the quantum lifetime. Hence, from the definition ωc=eBm∗\omega_c = \frac{eB}{m^*} and μq=eτqm∗\mu_q = \frac{e\tau_q}{m^*}, where BB, ee, m∗m^*, and, μq\mu_q, are the magnetic field, the electron charge, the effective mass and the quantum mobility, respectively, we obtain μqBonset∼1\mu_q B_{\mathrm{onset}} \sim1. We report our highest obtained value, for Bonset=4mTB_{\mathrm{onset}} = 4\mathrm{mT}, of μq≈2.5×106 cm2/Vs\mu_q \approx2.5 \times10^6\ \mathrm{cm}^2/\mathrm{Vs} for D=0 V/nmD = 0\ \mathrm{V}/\mathrm{nm}, corresponding to filling factor ν=−8\nu= -8. Using the same approach, we extracted the quantum mobility for D=0.3 V/nmD = 0.3\ \mathrm{V}/\mathrm{nm}, where the SdH oscillations are visible down to Bonset=50mTB_{\mathrm{onset}} = 50\mathrm{mT}, resulting in μq≈2×105 cm2/Vs\mu_q \approx2 \times10^5\ \mathrm{cm}^2/\mathrm{Vs}.

Magnetotransport measurement of the FET at $D = 0.3\,\mathrm{V}/\mathrm{nm}$

Figure S3. Magnetotransport measurement of the FET at D=0.3 V/nmD = 0.3\,\mathrm{V}/\mathrm{nm}

Gate dependencies

To define our structures electrostatically, isolating the BLG edges, we have to operate the device-defining surrounding top gate and the back-gate simultaneously, opening a bandgap and tuning the Fermi level within this gap. However, the same backgate also participates in the dual-gate operation of the Hall-bar together with the Hall-bar gate. In this way it is possible to tune the density and displacement field in the Hall bar device. To achieve this functionality, the surrounding top gate must adapt to changes in the back gate, following the fixed-density line in the gate-gate space. Therefore, the back gate and the Hall bar gate can tune the displacement field and density in the Hall bar, while the surrounding gate depends on the chosen working point. As a consequence, the insulating state under the outer gate is maintained while the displacement field and the induced bandgap are affected. Due to this working principle, the devices might lack a clear definition in cases of an insufficient displacement field that limits the formation of a bandgap.

Hall/SdH single transport path validation

To further characterize the Hall bar, and to ensure that no parallel transport was happening under the etched top gate (as already suggested by the gate-gate map of the back gate and the Hall bar gate, when the surroundings were insulating), the Hall density was complemented by the SdH extracted density. As a result, the two densities are in agreement with deviation only in the case of poor hall bar definition.

SdH density and Hall density comparison

Figure S4. SdH density and Hall density comparison for D=0.3 V/nmD=0.3\ \mathrm{V}/\mathrm{nm}.

SdH density extracted with FFT only and with FFT powered sinusoidal fitting, and Hall density comparison

Figure S5. SdH density (extracted with FFT only and with FFT powered sinusoidal fitting) and Hall density comparison for D=0.3 V/nmD=0.3\ \mathrm{V}/\mathrm{nm}.

Relative discrepancy for SdH density and Hall density

Figure S6. Realtive discrepancy for SdH density and Hall density for D=0.3D=0.3 V/nm.

Hall density

The Hall density is extracted by taking the anti-symmetrized Hall resistance and calculating the slope, averaging over different smaller ranges, moving away from the extremes. In fact, due to the presence of the Hall resistance quenching around B=0B=0 the slope is extracted for 100mT≤B≤495mT100\mathrm{mT} \le B \le495\mathrm{mT}, reducing the window by 5mT5\mathrm{mT} each time, until a minimal range of 70mT70\mathrm{mT}. However, as a consequence of the high mobility, the onset of the Hall plateau occurs already at small fields, and for this reason, averaging over different ranges helps mitigate the plateau contributions. For each Hall fitting window, the uncertainty is estimated using a bootstrap procedure. The antisymmetrized Hall data points (B,RxyA)(B,R_{xy}^{A}), in a given range, are resampled with replacement Nboot=500N_{\mathrm{boot}}=500 times, and each resampled dataset is fitted with a linear function,

RxyA(B)=sB+b,R_{xy}^{A}(B)=sB+b,

where s=dRxyA/dBs=dR_{xy}^{A}/dB is the Hall slope. Each bootstrap slope sjs_j is converted into a carrier density according to

nH,j=1esj=1e dRxyA/dB,∀j=1,…,Nboot.n_{H,j}=\frac{1}{e s_j}=\frac{1}{e\,dR_{xy}^{A}/dB}, \qquad\forall j=1,\ldots,N_{\mathrm{boot}}.

For a given fitting window ii, the uncertainty is taken as the standard deviation of the resulting bootstrap density distribution,

δnH,iboot=std⁡({nH,j}j=1Nboot).\delta n_{H,i}^{\mathrm{boot}}=\operatorname{std}\left(\left\{n_{H,j}\right\}_{j=1}^{N_{\mathrm{boot}}}\right).

The final Hall density is then obtained by inverse-variance-weighted averaging over all accepted fitting windows,

nˉH=∑iwinH,i∑iwi,wi=1(δnH,iboot)2.\bar{n}_{H}=\frac{\sum_i w_i n_{H,i}}{\sum_i w_i}, \qquad w_i=\frac{1}{\left(\delta n_{H,i}^{\mathrm{boot}}\right)^2}.

The final error bar is computed by adding in quadrature the statistical uncertainty of the weighted mean nˉH\bar{n}_{H} and the systematic error due to the choice of the window,

δnˉH=(1∑iwi)2+[∑iwi(nH,i−nˉH)2∑iwi]2.\delta\bar{n}_{H}=\sqrt{\left(\frac{1}{\sqrt{\sum_i w_i}}\right)^2+\left[\sqrt{\frac{\sum_i w_i\left(n_{H,i}-\bar{n}_{H}\right)^2}{\sum_i w_i}}\right]^2}.

Confinement potential

The density dependence of the effective channel width was analyzed independently for each displacement field and for electrons and holes. The experimental data were fitted using the power-law dependence Weff(n)=Aref(∣n∣nref)βW_{\mathrm{eff}}(n)=A_{\mathrm{ref}}\left(\frac{|n|}{n_{\mathrm{ref}}}\right)^{\beta} where ArefA_{\mathrm{ref}} and β\beta were treated as free parameters and nref=1010 cm−2n_{\mathrm{ref}}=10^{10}\,\mathrm{cm}^{-2} was kept fixed for all datasets. This normalization improves the numerical conditioning of the nonlinear fit. A reweighted least-squares procedure with Tukey bisquare weights was used in the fitting routine to account for potential outliers. However, given the small datasets available for some displacement fields, this analysis points more towards a qualitative than a quantitative result.

Fit results for electrons at different displacement fields

Figure S7. Fit results for electrons at different displacement fields. The left panel shows the fitted data with ArefA_{\mathrm{ref}}, β\beta, and the normalized χ2\chi^{2} reported in the legend box. The central panel shows the residual plots and includes the number of data points that scored a weight below 0.5 in the robust fit procedure. The right plot shows the calculated confinement potential based on the fit results.

The fitted width dependence was subsequently mapped onto an equivalent transverse confinement potential. To convert carrier density into an energy scale, we assume an isotropic two-dimensional parabolic band described by E(k)=ℏ2k22m∗E(k)=\frac{\hbar^2 k^2}{2m^*}, where EedgeE_{\mathrm{edge}} is the corresponding band edge, kk is the magnitude of the in-plane wave vector, and m∗m^* is the effective mass. For a circular two-dimensional Fermi surface with total degeneracy gg, the carrier density is related to the Fermi wave vector kFk_F by n=gkF24πn=\frac{gk_F^2}{4\pi}. Solving for the Fermi wave vector gives kF=4π∣n∣gk_F=\sqrt{\frac{4\pi|n|}{g}}, where ∣n∣|n| is used so that the same expression applies to both electron- and hole-like carrier densities. The Fermi energy relative to the corresponding band edge is then obtained by evaluating the parabolic dispersion at kFk_F, E(kF)=ℏ2kF22m∗E(k_F)=\frac{\hbar^2 k_F^2}{2m^*}. Substituting the density-dependent Fermi wave vector yields 2πℏ2gm∗∣n∣\frac{2\pi\hbar^2}{gm^*}|n|. Therefore, although the Fermi wave vector scales with carrier density as kF∝∣n∣k_F\propto\sqrt{|n|}, the Fermi energy of a two-dimensional parabolic band varies linearly with the carrier-density magnitude, EF(n)−Eedge∝∣n∣E_F(n)-E_{\mathrm{edge}}\propto|n|. This linear density-to-energy relation is subsequently used to convert the experimentally extracted Weff(n)W_{\mathrm{eff}}(n) dependence into an equivalent transverse confinement potential. The lateral classical turning point, denoted by yty_t, is identified with half of the effective channel width, ∣yt∣=Weff2|y_t|=\frac{W_{\mathrm{eff}}}{2}, where yy is the transverse coordinate measured from the center of the channel. The experimentally fitted effective width is described by Weff(n)=Aref(∣n∣nref)βW_{\mathrm{eff}}(n)=A_{\mathrm{ref}}\left(\frac{|n|}{n_{\mathrm{ref}}}\right)^\beta where Aref=Weff(nref)A_{\mathrm{ref}}=W_{\mathrm{eff}}(n_{\mathrm{ref}}) is the effective channel width at the reference density nrefn_{\mathrm{ref}}, and β\beta determines the density dependence of the confinement width. Substituting the fitted width into the turning-point condition gives ∣yt(n)∣=Aref2(∣n∣nref)β|y_t(n)|=\frac{A_{\mathrm{ref}}}{2}\left(\frac{|n|}{n_{\mathrm{ref}}}\right)^\beta. This relation can be inverted to associate a carrier density with a given transverse position yy, ∣n(y)∣=nref(2∣y∣Aref)1/β|n(y)|=n_{\mathrm{ref}}\left(\frac{2|y|}{A_{\mathrm{ref}}}\right)^{1/\beta}. At the classical turning point, the transverse electrostatic potential energy is identified with the Fermi energy corresponding to that carrier density. Assuming a two-dimensional parabolic dispersion, the Fermi energy is EF(n)=2πℏ2gm∗∣n∣E_F(n)=\frac{2\pi\hbar^2}{gm^*}|n| where m∗m^* is the effective mass and gg is the total spin-valley degeneracy. It is convenient to define the energy corresponding to the reference density as Eref=EF(nref)=2πℏ2gm∗nrefE_{\mathrm{ref}}=E_F(n_{\mathrm{ref}})=\frac{2\pi\hbar^2}{gm^*}n_{\mathrm{ref}}. The equivalent transverse confinement potential is then obtained by substituting n(y)n(y) into the density-energy relation, Veq(y)−Veq(0)=Eref(2∣y∣Aref)1/βV_{\mathrm{eq}}(y)-V_{\mathrm{eq}}(0)=E_{\mathrm{ref}}\left(\frac{2|y|}{A_{\mathrm{ref}}}\right)^{1/\beta}. Here, Veq(y)V_{\mathrm{eq}}(y) represents the effective electrostatic potential energy at transverse coordinate yy, while Veq(0)V_{\mathrm{eq}}(0) is the potential energy at the center of the channel. The spatial dependence of the reconstructed confinement potential can therefore be written as Veq(y)−Veq(0)∝∣y∣pV_{\mathrm{eq}}(y)-V_{\mathrm{eq}}(0)\propto|y|^p with p=1βp=\frac{1}{\beta}. Consequently, the experimentally extracted exponent β\beta directly determines the effective shape of the lateral confinement potential within the assumed parabolic-band approximation. In particular, β=1/2\beta= 1/2 corresponds to p=2p = 2. The reconstructed Veq(y)V_{\mathrm{eq}}(y) should be interpreted as an effective confinement potential resulting from the measured Weff(n)W_{\mathrm{eff}}(n) dependence and the assumed parabolic density-to-energy relation. In regimes where the BLG dispersion or density of states departs substantially from the parabolic approximation, the quantitative mapping between carrier density and energy must be modified accordingly.

From the fit results we identify some limitations of the model, especially at low densities. From the cases with more data points, we observe a deviation from the expected near-parabolic confinement. This might be caused by a different confinement shape at lower densities, where the channel width does not necessarily extrapolate to zero. Further limitations originate from the effective width approaching the lithographical edges of the device, an eventuality neglected by our model. In addition, the lithographical width and the confinement turning point might largely vary depending on the carrier type. Although we found that in the 90nm gaps no transport occurs, some residual charge might still exist, and tuned by the only effect of the back gate, this can result in the presence of a p-n junction at the edges, another effect that is not included in our model. Yet, further investigation along these lines will require more experiments and simulations. As already mentioned, the limited dataset does not allow us to precisely quantify the parameters at play, and in some cases, for example, for D=0.100V/nmD = 0.100\mathrm{V}/\mathrm{nm}, it is hard to mitigate the effect of outliers and statistical oscillations, see Figure S8.

Fit results for holes at different displacement fields

Figure S8. Fit results for holes at different displacement fields. The left panel shows the fitted data with ArefA_{\mathrm{ref}}, β\beta, and the normalized χ2\chi^2 reported in the legend box. The central panel shows the residual plots and includes the number of data points that scored a weight below 0.5 in the robust fit procedure. The right plot shows the calculated confinement potential based on the fit results.

Calculated wavefunctions and associated probability for modes confined by a potential well

Figure S9. Calculated wavefunctions (left) and associated probability (right) for modes confined by a potential well with β=0.35\beta= 0.35. Titles on the figures on the right panel show the probability of tunneling outside the barrier, hence the increased chance of interaction for the higher modes with the lithographical edges.

IQHE

Rxx and σxy traces at B⊥ = 6T and D = 0.1V/nm vs carrier density

Figure S10. RxxR_{xx} and σxy\sigma_{xy} traces at B⊥=6TB_{\perp}=6T and D=0.1V/nmD=0.1V/nm vs carrier density. Red dashed lines indicate integers of e2/he^2/h.

The Hall conductivity was calculated from the directly measured longitudinal and transverse resistances through tensor inversion. The longitudinal resistivity was obtained by applying the effective Hall-bar geometry,

ρxx=RxxWeffL,(6)\rho_{xx}=R_{xx}\frac{W_{\mathrm{eff}}}{L}, \tag*{(6)}

The Hall conductivity is therefore

σxy=−ρxyρxx2+ρxy2=−ρxy(RxxWeffL)2+ρxy2.(7)\sigma_{xy}=-\frac{\rho_{xy}}{\rho_{xx}^{2}+\rho_{xy}^{2}}=-\frac{\rho_{xy}}{\left(R_{xx}\frac{W_{\mathrm{eff}}}{L}\right)^{2}+\rho_{xy}^{2}}. \tag*{(7)}

SdH density

The SdH density is extracted from the dominant peak in the Fourier spectrum of the oscillatory component of RxxR_{xx} as a function of 1/B1/B. Since the raw data are acquired with approximately uniform steps in BB, they are not uniformly sampled in 1/B1/B. Therefore, before performing the frequency analysis, the data are first transformed to 1/B1/B and interpolated onto a uniform 1/B1/B grid.

To reduce the risk of assigning the SdH frequency to a spurious FFT peak and to test whether the oscillations are consistent with a single dominant frequency, rather than with beating or density inhomogeneity, the oscillations are also fitted directly as a function of 1/B1/B using a damped sinusoidal function. The frequency identified from the FFT is used as the initial guess for the fitting procedure:

ΔRxx(x)=Ae−x/τsin⁡(2πfx+ϕ)+d,x=1B.\Delta R_{xx}(x) = A e^{-x/\tau} \sin(2\pi f x + \phi) + d, \qquad x = \frac{1}{B}.

The fitted frequency ff is then used as a refined estimate of the SdH frequency. The corresponding SdH density is obtained from

∣nSdH∣=gehFSdH,|n_{\mathrm{SdH}}| = g\frac{e}{h}F_{\mathrm{SdH}},

where gg is the degeneracy factor. Since the SdH frequency determines only the magnitude of the carrier density, the sign of nSdHn_{\mathrm{SdH}} is assigned from the Hall density measured at the same gate voltage:

nSdH=sign⁡(nH)∣nSdH∣.n_{\mathrm{SdH}} = \operatorname{sign}(n_{\mathrm{H}})|n_{\mathrm{SdH}}|.

Drude resistivity and cyclotron orbit extraction

To extract the scattering parameter pp or the Hall mobility, the Drude resistivity, calculated from the resistance at B=0B = 0 is needed. However, in our measurements, Rxx(B=0)R_{xx}(B = 0) is affected by the boundary scattering. Hence, a better estimation is obtained by extracting the baseline of the first visible SdH oscillations. At that point, where the cyclotron orbit is smaller than the channel width, the real Drude resistivity can be recovered. To minimize extraction errors, the window used to extract the baseline is dynamically adapted to consider only oscillations around a constant value. In this way, the extracted value is not affected by the tail of the boundary-scattering-induced magnetoresistance, or by the onset of fully quantized states.

Rxx(B) trace showing mesoscopic magnetoresistance and SdH oscillations

Figure S11. Rxx(B)R_{xx}(B) trace at fixed carrier density featuring the mesoscopic magnetoresistance and the SdH oscillations are shown.

From the same curve, by the position of the peak, identified by the maximum of a polynomial fit, the cyclotron orbit and consequently the WeffW_{\mathrm{eff}}, is derived, as discussed in the main text.

Effective mass extraction and quantum mobility

To determine the quantum mobility within the Hall bar, we first extracted the cyclotron effective mass from the temperature dependence of SdH oscillation amplitude. The analysis was performed using the Lifshitz–Kosevich (LK) thermal damping factor,

A(T)=A0Xsinh⁡X,X=2π2kBTℏωc=2π2kBTm∗ℏeB,(8)A(T)=A_0\frac{X}{\sinh X},\qquad X=\frac{2\pi^2 k_{\mathrm{B}}T}{\hbar\omega_c}=\frac{2\pi^2 k_{\mathrm{B}}Tm^*}{\hbar eB}, \tag*{(8)}

where ωc=eB/m∗\omega_c=eB/m^* is the cyclotron frequency and m∗m^* is the cyclotron effective mass. By fitting the temperature dependence of a selected SdH peak at fixed magnetic field to Eq. (8), we extracted m∗m^*. This value was then used as a fixed input parameter in the subsequent Dingle analysis.

Temperature dependence of a selected SdH oscillation amplitude

Figure S12. Temperature dependence of a selected SdH oscillation amplitude. The red curve shows the LK fit used to extract the cyclotron effective mass.

Due to the limited number of visible oscillations, resulting from disorder broadening and the early onset of the fully quantized regime, the extracted effective masses exhibit relatively large uncertainties. Nevertheless, a clear density dependence of the effective mass is observed, while no pronounced dependence on displacement field is resolved within the experimental accuracy.

BLG effective mass extracted for different carrier densities and displacement fields

Figure S13. BLG effective mass extracted for different carrier densities and displacement fields.

The observed density dependence of the effective mass has previously been attributed to electron–electron interactions [2].

After determining m∗m^* from the LK analysis, the quantum mobility was extracted from a final fit of the SdH oscillations as a function of inverse magnetic field, x=1/Bx = 1/B. The fit was performed directly on the measured Rxx(x)R_{xx}(x) trace, with the smooth background included as a polynomial baseline. The fitting function was

Rxx(x)=Pn(x)+A0RT(x)exp⁡(−xτB)sin⁡(2πFx+ϕ),x=1B,(9)R_{xx}(x) = P_n(x) + A_0 R_T(x) \exp\left(-\frac{x}{\tau_B}\right) \sin(2\pi F x + \phi), \qquad x = \frac{1}{B}, \tag*{(9)}

where Pn(x)P_n(x) is a low-order polynomial baseline, A0A_0 is the oscillation amplitude prefactor, FF is the SdH frequency, ϕ\phi is a phase offset, and τB\tau_B is the inverse-field damping scale. The LK thermal damping factor is

RT(x)=Xsinh⁡X,X=2π2kBTm∗ℏex.(10)R_T(x) = \frac{X}{\sinh X}, \qquad X = \frac{2\pi^2 k_B T m^*}{\hbar e}x. \tag*{(10)}

Thus, the SdH frequency, phase, amplitude, polynomial baseline, and Dingle damping scale were determined simultaneously from a global nonlinear fit.

The dashed envelope lines shown in the Dingle plots were reconstructed after the fit from the fitted parameters. The oscillation envelope was defined as

Aenv(x)=A0RT(x)exp⁡(−xτB),(11)A_{\mathrm{env}}(x) = A_0 R_T(x) \exp\left(-\frac{x}{\tau_B}\right), \tag*{(11)}

and the upper and lower dashed curves were plotted as

Renv±(x)=Pn(x)±Aenv(x).(12)R_{\mathrm{env}}^{\pm}(x) = P_n(x) \pm A_{\mathrm{env}}(x). \tag*{(12)}

The factor exp⁡(−x/τB)\exp(-x/\tau_B) in Eq. (9) corresponds to the Dingle damping factor. By comparing it with the conventional Dingle form,

exp⁡(−xτB)=exp⁡[−2π2kBTDm∗ℏex],(13)\exp\left(-\frac{x}{\tau_B}\right) = \exp\left[-\frac{2\pi^2 k_B T_D m^*}{\hbar e}x\right], \tag*{(13)}

the fitted inverse-field damping scale was converted into the Dingle temperature as

TD=ℏe2π2kBm∗τB.(14)T_D = \frac{\hbar e}{2\pi^2 k_B m^* \tau_B}. \tag*{(14)}

The corresponding quantum lifetime was then obtained from

τq=ℏ2πkBTD,(15)\tau_q = \frac{\hbar}{2\pi k_B T_D}, \tag*{(15)}

and the quantum mobility was calculated as

μq=eτqm∗.(16)\mu_q = \frac{e\tau_q}{m^*}. \tag*{(16)}

For the different displacement fields and densities explored, the quantum mobility extracted from the Dingle analysis was found to be of the order of μq∼30×103 cm2V−1s−1\mu_q \sim30 \times10^3\ \mathrm{cm}^2\mathrm{V}^{-1}\mathrm{s}^{-1}.

SdH oscillations measured at $D = 0.3$ V/nm and $n = 1.1 \times 10^{12}$ cm$^{-2}$

Figure S14. SdH oscillations measured at D=0.3D = 0.3 V/nm and n=1.1×1012n = 1.1 \times10^{12} cm−2^{-2}. The red curve shows the combined LK–Dingle fit, while the dashed lines indicate the fitted oscillation envelope. The fit yields nSdH=(1.102±0.001)×1012n_{\mathrm{SdH}} = (1.102 \pm0.001) \times10^{12} cm−2^{-2}, τq=(7.15±0.70)×10−13\tau_q = (7.15 \pm0.70) \times10^{-13} s, and μq=(30.73±1.68)×103\mu_q = (30.73 \pm1.68) \times10^{3} cm2^{2}V−1^{-1}s−1^{-1}.

RAMAN

The Raman spectra of the graphite flakes were fitted with a single Lorentzian function for the as-exfoliated sample and with three Lorentzian functions for the plasma and annealed samples. An additional broad Lorentzian component was used to account for the increase in the background observed after plasma and annealing treatment. For each Raman peak, the intensity, peak position, and FWHM were extracted. The evolution of the peak position and FWHM at each step is shown in Figure S15(a) and (b), respectively.

The Raman spectra of the h-BN flake discussed in Sec. V, including the spectrum acquired after annealing, are shown in Figure S16. The Raman spectra were fitted with a Lorentzian function, and the corresponding fitted curves are shown as solid lines on top of the experimental data. For the annealed spectrum, a parabolic baseline was used for the fitting procedure to account for the luminescence background. The optical images of the measured h-BN flake before and after the (N2+O2)2+CHF3/O2(\mathrm{N}_2+\mathrm{O}_2)_2+\mathrm{CHF}_3/\mathrm{O}_2 plasma treatment are shown in Figure S16. The change in the optical contrast indicates that the plasma etched away some h-BN layers, resulting in a reduction of its thickness.

While measuring the h-BN flake after annealing, we noticed that the luminescence background gradually decreased with continuous laser exposure. To investigate this behavior more in detail, we acquired consecutive Raman spectra every 60 s until the intensity of the background reached a plateau. We initially used a laser power of 100 μ\muW. Once the background reached a plateau, the laser power was increased, and the acquisition sequence was repeated. This procedure was repeated for laser power ranging from 100 μ\muW to 2.5 mW, as shown in Figure S17. For every excitation power, a progressive decrease in luminescence background was observed. This behavior is consistent with a photobleaching effect, in which the laser irradiation modifies the surface contaminants rather than permanently changing the h-BN Raman response.

Raman spectra of an h-BN flake after plasma treatment and subsequent annealing

Figure S17. Raman spectra of an h-BN flake after plasma treatment with (N2+O2)2+CHF3/O2(\mathrm{N}_2 + \mathrm{O}_2)_2 + \mathrm{CHF}_3/\mathrm{O}_2 and subsequent annealing. The Raman spectra are acquired every 60 s for each laser power. The laser power ranges from 100 μW100\,\mu\mathrm{W} to 2.5 mW2.5\,\mathrm{mW}.

We investigated the effect of a second plasma etching treatment using (N2+O2)2+N2(\mathrm{N}_2 + \mathrm{O}_2)_2 + \mathrm{N}_2. Figure S18 shows the Raman spectra of the graphite flake before the plasma treatment, after the plasma treatment, and after the annealing. Similar to the case of CHF3/O2\mathrm{CHF}_3/\mathrm{O}_2 plasma etching, the plasma treatment leads to the appearance of two defect peaks in graphite: the D and D’ bands. The intensities of both defect peaks decrease after the annealing process. Also, the peak positions (Figure S18) and FWHM (Figure S18) follow the same trend observed for the CHF3/O2\mathrm{CHF}_3/\mathrm{O}_2 plasma treatment.

Raman spectra and peak positions of graphite flake

Figure S18. a) Raman spectra of as-exfoliated (red), plasma-treated (blue) and annealed (green) graphite flake on top of SiO2_2/Si. b) Evolution of the Raman peak position and (c) FWHM before plasma treatment, after (N2_2 + O2_2)2_2 + N2_2 plasma exposure, and after annealing.

Fig. S19(a) shows the Raman spectra of the h-BN flake before the plasma treatment, after the plasma treatment, and after subsequent annealing. Similar to the case shown in Figure S16, the annealing results in the appearance of a broad luminescence background. An increase in the FWHM as well as a small red shift of the E2g_{2g} mode is observed after plasma treatment, although it is less pronounced than the one observed in the case of CHF3_3/O2_2 plasma, as shown in Figure S19(b). This smaller shift may be related to the lower etching rate of the N2_2 etching plasma compared with the CHF3_3/O2_2 plasma. Indeed, the optical images in Figure S19(c) show a smaller change in optical contrast compared to Figure S16(b), suggesting a smaller reduction in the h-BN thickness.

Kelvin Probe Force Microscopy of Surface-Treated h-BN and Graphene Flakes

The quality of the band gap discussed in Sec. III of the main text relies on a homogeneous electrostatic environment of the BLG sheet. Two contributions are difficult to control during fabrication: charges trapped in the SiO2_2 of the substrate, and residues or charges introduced by the reactive-ion etching steps used to define the graphite gate geometry. In order to assess how the plasma treatments employed in our process affect the local electrostatic landscape of the constituent flakes, we performed Kelvin probe force microscopy (KPFM) [3, 4] on exfoliated h-BN and graphene/graphite flakes on SiO2_2/Si that were subjected to the same treatments as the flakes used for the devices.

Experimental details

All measurements were carried out with a Nanosurf DriveAFM operated inside a nitrogen glove box (H2_2O < 0.1 ppm, O2_2 < 0.1 ppm). Working in a dry and oxygen-free atmosphere removes two effects that are known to compromise surface potential measurements under ambient conditions. An adsorbed water film screens the electrostatic field between tip and sample and thereby reduces the measured potential contrast, an effect that disappears when imaging in dry nitrogen [5]. In addition, molecular oxygen acts as an acceptor for graphene on SiO2_2, with the resulting hole doping becoming stronger and increasingly irreversible once moisture is present [6]. We used PtIr-coated silicon cantilevers (PPP-NCLPt, Nanosensors) with a first flexural resonance at f1=166.642f_1 = 166.642 kHz. The topography was recorded in amplitude-modulation mode with a free oscillation amplitude of roughly 10 nm and a setpoint of 80 % of the free amplitude.

The cantilever was kept at ground potential, while the sample was connected through a separate coaxial line to an HF2LI lock-in amplifier (Zurich Instruments), which supplies both the DC and the AC bias. Applying the bias exclusively to the sample and keeping the tip grounded suppresses the capacitive cross-talk between the excitation line and the deflection detection down to the noise level of the setup. The surface potential was measured by amplitude-modulated KPFM (AM-KPFM) [3]: the AC bias was tuned to the second flexural resonance of the cantilever at f2=1034.18f_2 = 1034.18 kHz, and the amplitude of the normal-force signal at f2f_2 was demodulated by the lock-in amplifier. After adjusting the demodulation phase, the in-phase component was nulled by a PID controller acting on the DC voltage applied to the sample. The compensating DC voltage is the contact potential difference (CPD) reported below. With the bias applied to the sample, the work function of the sample follows from

Φsample=Φtip+eVCPD,(17)\Phi_{\mathrm{sample}} = \Phi_{\mathrm{tip}} + e V_{\mathrm{CPD}}, \tag*{(17)}

with Φtip≈4.3\Phi_{\mathrm{tip}} \approx4.3 eV for the PtIr-coated tip. We stress that Eq. (17) only yields the true work function of the sample if the measured CPD is exclusively determined by the electronic structure of the surface. This is not the case here: charges located at the interfaces of the flakes or trapped in the underlying SiO2_2 generate an additional long-range electrostatic force. No KPFM detection scheme can separate this contribution from the work-function term, since the feedback loop compensates the total electrostatic signal, so that the trapped charges enter the measured CPD [4]. All h-BN measurements were performed with one and the same cantilever, and likewise all graphene measurements with a single, but different one; with each of the two series major changes of the tip apex were avoided. Nevertheless the absolute CPD values differ by several volts between the individual measurements of one series. This scatter is in itself an indication that the signal is governed by trapped charges rather than by the work functions of the surfaces. In the following we therefore restrict the discussion to the CPD contrast within a single image and to its spatial homogeneity, both of which are independent of the global offset.

Three surface conditions were compared for each material: an untreated reference flake, a flake exposed to an N2_2+O2_2 plasma (four cycles consisting of a 20s N2_2 plasma treatment at 100W and a 40s O2_2 plasma treatment at 10W, followed by 10s N2_2 plasma exposure at 100W), and a flake that received the same N2_2+O2_2 treatment followed by a short CHF3_3/O2_2 reactive-ion etch (two cycles consisting of a 20s N2_2 plasma treatment at 100W and a 40s O2_2 plasma treatment at 10W, followed by 10s CHF3_3/O2_2 plasma exposure at 60W), i.e. the etch chemistry used to pattern the graphite gates and to open the contact vias. Topography and CPD were recorded simultaneously; the slow-scan direction is indicated by the arrow in the bottom right corner of each topography image.

For each treatment several flakes were investigated, and the images shown are representative of the respective sample. We did not observe any systematic influence of the surface treatment on the topography channel: the differences in flake thickness, terracing, step morphology and lateral shape between the individual panels reflect the properties of the particular exfoliated flake rather than the preparation it received. The discussion below is therefore based entirely on the CPD channel, with the topography serving only to locate the flake edges within the same frame.

h-BN flakes

Figure S20 summarises the measurements on h-BN. For all three treatments the h-BN flakes appear darker than the surrounding substrate in the CPD channel. In the following we quote the contrast ΔVCPD=VCPDflake−VCPDsubstrate\Delta V_{\mathrm{CPD}} = V_{\mathrm{CPD}}^{\mathrm{flake}} - V_{\mathrm{CPD}}^{\mathrm{substrate}}, which is negative for all three flakes and lies between −0.4-0.4 V and −1.5-1.5 V depending on the treatment. Within the convention of Eq. (17) this corresponds to a lower apparent work function of the h-BN surface.

KPFM of h-BN flakes on SiO2 for three different surface treatments

Figure S20. KPFM of h-BN flakes on SiO2_2 for three different surface treatments. Top row: untreated flake. Middle row: after 4×4\times N2_2+O2_2 plasma. Bottom row: after 2×2\times N2_2+O2_2 plasma followed by a 10 s CHF3_3/O2_2 etch. (a) Topography, (b) simultaneously recorded CPD, and (c) line profiles of the topography (top) and of the CPD (bottom) extracted along the line marked in (a) and (b). The arrow in the bottom right corner of (a) denotes the slow-scan direction.

For the untreated flake (top row) the main, flat part of the flake shows ΔVCPD≈−1.3\Delta V_{\mathrm{CPD}} \approx-1.3 V. Two further features are resolved in the CPD map. First, part of the flake is folded up along its left side, clearly visible as the elevated ridge in the topography of (a). Over the entire folded region the CPD is higher than over the part of the flake that lies directly on the SiO2_2, as is also apparent from the profile in (c). Second, all edges of the flake are outlined by a rim of increased CPD. This rim lies entirely within the outline of the flake and extends over a considerably wider range than the topographic edge itself. Its smooth, gradient-like shape is suggestive of a species that has diffused inwards from the flake edges, for which residual water, intercalated either between the h-BN layers or at the flake/substrate interface, would be an obvious candidate. We emphasise that our data do not allow us to identify the chemical nature of this contribution. What can be stated is that the rim is characteristic of the untreated surface: it is absent on both plasma-treated flakes discussed below.

After the N2_2+O2_2 plasma treatment (middle row) the situation changes qualitatively. The CPD map reproduces the outline of the flake with sharp edges, the plateaus on the flake (≈−1.20\approx-1.20 V) and on the substrate (≈−0.80\approx-0.80 V) are flat to within the noise of the measurement, and no long-range features are observed on the surrounding oxide. The reduced contrast of ΔVCPD≈−0.4\Delta V_{\mathrm{CPD}} \approx-0.4 V together with the absence of an uncorrelated background indicates that the measured signal is dominated by the material contrast rather than by trapped charges. Repeated scans of the same area reproduced the CPD map without measurable drift, which makes this treatment the most stable of the three.

The flake that received the additional CHF3_3/O2_2 etch (bottom row) shows the opposite behaviour. Its CPD map is dominated by a broad, smooth depression that extends far beyond the flake and reaches ΔVCPD≈−1.5\Delta V_{\mathrm{CPD}} \approx-1.5 V. The transition between substrate and flake in the CPD profile is smeared out over several micrometres, in stark contrast to the abrupt topographic step at the same position. Such a decoupling of the two channels is the signature of a charge distribution located below the surface, i.e. in the SiO2_2, rather than of a work-function contrast at the surface. We conclude that the fluorine-based etch, even when applied for only 10 s, leaves the oxide in a considerably more charged state.

Graphene and graphite flakes

The corresponding measurements on graphene and graphite flakes are shown in Figure S21. Here an additional effect is observed for all three treatments: the CPD maps contain regions whose boundaries are aligned with the fast-scan axis and which are therefore not features of the sample but events that occurred during imaging. They are caused by a substantial contact between the oscillating tip and the flake, which injects charge into the flake. The injected charge shifts the compensating voltage of the whole affected area, so that the CPD map changes abruptly at the scan line at which the event occurred; the boundaries are consequently parallel to the fast-scan axis, and the affected part of the image is the one that follows the event along the slow-scan direction indicated by the arrow in (a). Tip-induced charging of graphene on SiO2_2 has been reported before [7], where the injected charge was found to decay exponentially with time constants of the order of tens of minutes under ambient humidity. In our glove-box environment the discharge channel provided by an adsorbed water film is suppressed and the charged state is markedly more persistent: it survives for at least two hours, i.e. far beyond the duration of a single scan, and we found no significant change of the CPD maps over the course of a day. The charge state could only be altered deliberately, either by further tip contacts or by applying a suitable DC voltage to the tip, which allowed us to charge and discharge the flakes at will.

KPFM maps and line profiles of graphene/graphite flakes on SiO2 for three surface treatments

Figure S21. KPFM of graphene/graphite flakes on SiO2_2 for three different surface treatments. Top row: untreated flake. Middle row: after N2_2+O2_2 plasma. Bottom row: after N2_2+O2_2 plasma followed by a CHF3_3/O2_2 etch. (a) Topography, (b) simultaneously recorded CPD, and (c) line profiles of the topography (top) and of the CPD (bottom) extracted along the line marked in (a) and (b). The arrow in the bottom right corner of (a) denotes the slow-scan direction. The abrupt, scan-line-aligned steps in the CPD maps mark charging events caused by intermittent tip–sample contact during scanning.

The untreated flake (top row) shows the most pronounced example: a rectangular region of elevated CPD, reaching up to ≈2.7\approx2.7 V, is bounded by two horizontal scan lines, and is surrounded by a halo that decays smoothly over more than 22 µm into the surrounding oxide. The line profile in (c) illustrates this asymmetry: the CPD rises gradually towards the charged region but drops in abrupt steps at the scan lines where the charge state changed. Neither the halo nor the steps correlate with the topographic step at the flake edge. On this sample the injected charge is therefore not confined to the flake, but ends up in the surrounding oxide as well.

For the N2_2+O2_2-treated flake (middle row) the CPD contrast between flake and substrate is ΔVCPD≈−80\Delta V_{\mathrm{CPD}} \approx-80 mV, more than an order of magnitude smaller than for the untreated sample, and both plateaus are flat across the entire field of view. A single charging event is visible as a bright line along the upper flake edge, i.e. in the region scanned last, but it does not develop into an extended halo. The consequence of this event is seen in the subsequent scan of the same area, shown in the top row of Figure S22: the flake as a whole now carries a charge, and the contrast has changed sign to ΔVCPD≈0.22\Delta V_{\mathrm{CPD}} \approx0.22 V, i.e. the potential of the flake has shifted by about 0.3 V relative to the substrate. Two observations are worth emphasising. First, the shift is homogeneous over the entire flake, apart from faint stripes along the fast-scan direction, and it terminates abruptly at the flake edges without any gradient on the surrounding oxide, whose CPD level is the same as before the charging event. The injected charge therefore resides on the flake, where it spreads freely over the conducting graphite and remains confined by the flake boundary — in clear contrast to the untreated sample of the top row, whose halo extends micrometres into the substrate. Second, the charged state was reached at the end of one scan and is still fully present in the following one, which illustrates the persistence discussed above. The small residual contrast in the uncharged state, together with the absence of long-range gradients in both states, indicates that the density of trapped charges in the oxide is lowest for this treatment.

Charged states of the two plasma-treated graphene samples

Figure S22. Charged states of the two plasma-treated graphene samples, imaged after the charging events visible in Figure S21. Top row: N2_2+O2_2-treated flake of the middle row of Figure S21, recorded directly afterwards along the same line and with the opposite slow-scan direction. Bottom row: N2_2+O2_2+CHF3_3/O2_2-treated sample. (a) Topography, (b) CPD, and (c) line profiles of the topography (top) and of the CPD (bottom). In both cases the flake carries a charge over its full area, but the magnitude and the spatial confinement differ strongly; note the different scan sizes and colour scales of the two rows.

The sample that received the additional CHF3_3/O2_2 etch (bottom row) behaves qualitatively like the N2_2+O2_2 sample, but on a scale that is larger by more than an order of magnitude. The field of view contains several graphene and graphite flakes, and the CPD is constant over extended areas whose boundaries follow the flake outlines, with levels that differ by more than 1 V between neighbouring flakes: along the line profile the CPD drops from 3.4 V to 2.2 V at the position of a flake edge. The individual flakes are thus in markedly different charge states already before any interaction with the tip. Up to the lower part of the image the scan proceeds without charging events. There, however, a major event occurs and the CPD rises steeply by nearly 7 V to about 9 V. Since the slow-scan direction is downwards for this measurement, this part of the frame is the one recorded last, so that the charged state it reveals persists until the end of the scan.

The following scan of this sample, shown in the bottom row of Figure S22, confirms this interpretation and reveals one further difference. The flake is charged homogeneously over its full area and reaches a CPD of about 9 V, i.e. some 8 V above the surrounding oxide, which is larger than the shift observed on the N2_2+O2_2 sample by more than an order of magnitude. In addition, the potential does not terminate at the flake edge: it decays smoothly over roughly 1 μm into the substrate, whereas the topographic edge at the same position is sharp. Part of this broadening is expected on instrumental reasons, since the long-range electrostatic force sensed by the tip cone and the cantilever averages over an area that grows with the potential difference involved, and the two rows of Figure S22 are compared at 8 V and 0.3 V, respectively. The observation is nevertheless consistent with the h-BN measurements, where the etched sample was the only one showing a potential landscape extending far beyond the flake, and suggests that on the fluorinated oxide the injected charge is not confined to the flake alone.

References

  1. [1]Z. Ma, C. Prawoto, Z. Ahmed, Y. Xiao, L. Zhang, C. Zhou, and M. Chan, J. Mater. Chem. C 7, 6273 (2019).
  2. [2]K. Zou, X. Hong, and J. Zhu, Phys. Rev. B 84, 085408 (2011).
  3. [3]S. Sadewasser and T. Glatzel, eds., Kelvin Probe Force Microscopy: Measuring and Compensating Electrostatic Forces, Springer Series in Surface Sciences, Vol. 48 (Springer, Berlin, Heidelberg, 2012).DOI
  4. [4]S. Sadewasser and T. Glatzel, eds., Kelvin Probe Force Microscopy: From Single Charge Detection to Device Characterization, Springer Series in Surface Sciences, Vol. 65 (Springer, Cham, 2018).DOI
  5. [5]H. Sugimura, Y. Ishida, K. Hayashi, O. Takai, and N. Nakagiri, Appl. Phys. Lett. 80, 1459 (2002).
  6. [6]S. Ryu, L. Liu, S. Berchaud, Y.-J. Yu, H. Liu, P. Kim, G. W. Flynn, and L. E. Brus, Nano Lett. 10, 4944 (2010).
  7. [7]A. Verdaguer, M. Cardellach, J. J. Segura, G. M. Sacha, J. Moser, M. Zdrojek, A. Bachtold, and J. Fraxedas, Appl. Phys. Lett. 94, 233105 (2009).

Paper details

vlnc:W_zsqken5z
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 1

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