Scanning capacitance imaging of compressible and incompressible quantum Hall effect edge strips

The quantum Hall effect (QHE) [1] is a macroscopic quantum phenomenon in a two-dimensional electron gas (2DEG) where the formation of discrete Landau levels (LLs) in a large perpendicular magnetic field, B, leads to field intervals with a quantized Hall resistance and a vanishing longitudinal resistance. Although a theoretical understanding was developed shortly after its discovery [2], the experimental investigation of the microscopic origin using local probe techniques only became feasible in recent years. The challenges of such experiments are the required cryogenic temperatures, high magnetic fields and that most 2DEGs are buried below a dielectric layer.

The most intuitive picture of the QHE is that of edge states [3, 4]. If locally an integer number of LLs is occupied (integer filling factor), the density of states, D, at the Fermi energy is zero. A local electric potential raises the energy of these states, leading to states along equipotential lines at the Fermi energy. These states can either be localized around potential variations in the bulk of a 2DEG or delocalized along the sample edge. The self-consistent spatial rearrangement of these states (edge reconstruction) results in the formation of compressible strips, separated by incompressible strips [5, 6].

Edge states are crucial for the formation of the zero resistance states and quantum Hall plateaus by separating states carrying opposite momentum at opposite edges [6]. Delocalized low-energy excitations are possible only in the compressible regions at the sample edge, which, for example, renders electrically controlled interference experiments with electrons possible [7–9]. Edge states are also relevant for heat transport in QHE samples [10], topologically protected states in the fractional QHE [11] or in novel topological states at zero magnetic field [12, 13]. Between integer filling factors the strips widen and form a percolating network across the sample, which leads to enhanced backscattering and a finite longitudinal resistance and to the transition regions between QHE plateaus in the Hall resistance [14, 15].

Edge states localised at potential fluctuations in the bulk of a macroscopic 2DEG have been imaged using various scanning probe techniques [15–18]. Scanning gate measurements provide the link between these microscopic results and actual transport experiments in real devices, e.g. by demonstrating an enhanced sensitivity of the device resistances to local variations of the potential landscape [14, 19, 20]. The position of the innermost incompressible strip and the question of where the excess current is carried was investigated by locally measuring the electrostatic force gradient due to the Hall potential over a biased sample [21].

Here we present dynamic scanning capacitance microscopy (DSCM) [22] images recorded at the edge of a 2DEG in the QHE regime at zero bias as well as in the QHE breakdown regime at finite currents. A related method was used recently to image square areas of a 2DEG around the average bulk filling factor ν ≈ 2 [23]. However, the DSCM technique is less invasive and allows us to investigate the same 2DEG area many times, so that we can provide a complete set of measurements of the same scan area and over several QHE periods. The DSCM signals are directly related to the complex local sample conductance, which comprises the local quantum capacitance and the local conductivity of the 2DEG. Our instrument has a large scan area and a relatively high scan rate which allows us to image edge states over large distances in two dimensions, leading to very clear and intuitive illustrations of QHE edge states. In addition, we demonstrate for the first time how the compressible and incompressible regions disappear sharply at the critical current at which the QHE breaks down.

We use a home-built dynamic scanning capacitance microscope [22] (DSCM) that operates between 1.9 K and room temperature and magnetic fields perpendicular to the sample surface of up to 12 T. The microscope is based on a non-contact atomic force microscope (AFM) in a low pressure (2 mbar) He atmosphere. A quartz tuning fork sensor and a phase-locked loop allow non-optical sensor excitation and readout of the topography and the average force. An amplitude feedback loop maintains a constant oscillation amplitude and allows us to measure the mechanical sensor dissipation [24]. The AFM tip is fabricated from an electrochemically sharpened PtIr wire and an electrical connection enables simultaneous AFM/DSCM measurements and the compensation of the work function difference between the tip and the sample material.

The complex tip–sample conductance is measured using a radio frequency (rf) resonator. A small hand-made copper coil L is connected to the AFM tip and forms a resonant circuit with the stray capacitance of the tuning fork C_TF and the lead resistances R see figure 1(a). The transmission of the resonator is perturbed by the tip–sample capacitance C_ts, which is probed by an rf excitation and a pick-up coil. We use high-frequency lock-in amplifiers to measure the resulting voltage on the pick-up coil, U_out, at the resonance frequency of the unperturbed resonator (∼130 MHz). An analysis of the tip-sample interaction in terms of lumped-circuit elements is possible since the wavelength of the excitation is much larger than the relevant geometrical dimensions. One can show [22] that for reasonably sharp tips U_out is proportional to the complex conductance between the tip and ground, G_ts, independent of the details of the tip-sample interaction:

$$\begin{equation} U_{\rm out}\propto G_{\rm ts}. \end{equation}\noindent \tag{ 1 }$$

The setup-specific calibrated proportionality constant includes the excitation voltage. For a homogeneous sample, G_ts is given by the tip-sample capacitance, C_ts, in series with the complex sample conductance from the tip position to ground. The voltage amplitude on the AFM tip we estimate to be 24 mV, a compromise between a sufficient signal-to-noise ratio and useful scan rates and minimal charging of the sample surface.

Zoom In Zoom Out Reset image size

The perturbation of the rf resonator due to the sample is exceedingly small and a direct measurement is very slow and not suitable for scanning [22]. We operate the AFM in the dynamic mode where the tip oscillates perpendicular to the sample surface. The tip motion modulates the tip-sample capacitance and thus the extracted conductance with an amplitude G_ts,ω at the frequency ω of the AFM sensor (and its harmonics). This signal is recorded simultaneously with the AFM data using low-frequency lock-in amplifiers.

The DSCM technique allows us to measure output voltage differences of ∼1 nV at the required scan speeds, which corresponds to a capacitance resolution of ∼0.5 aF. Furthermore, DSCM does not require calibration or subtraction of base-line data. The lateral resolution of ∼100 nm is mainly limited by the point of closest approach to the surface, where the change of the tip-sample capacitance is the largest. This resolution is comparable to other scanning capacitance experiments with the tip in contact to the sample, but without the detrimental mechanical abrasion of the tip apex. Features in the DSCM images with a seemingly better lateral resolution can be traced back to changes in the average tip-sample distance due to topographic features on the surface, which are resolved with typical AFM resolution [25].

DSCM signals can be understood intuitively by considering a purely ohmic sample resistor R_S in series with the tip-sample capacitance, C_ts, as shown in figure 1(a). For highly conductive samples, i.e. ω_rfR_SC_ts ≪ 1, one finds

$$\begin{equation} G_{\rm ts, \omega}\approx \frac{\mathrm{d}G_{\rm ts}}{\mathrm{d}C_{\rm ts}}C_{\rm ts, \omega} = \omega_{\rm rf} C_{\rm ts, \omega} (2\omega R_{\rm S}C_{\rm ts} +i) \end{equation} \tag{ 2 }$$

with the capacitance variation C_ts,ω at the frequency ω of the tuning fork oscillation. In this limit, the imaginary part of the signal is given by the capacitance variation over a tuning fork oscillation cycle. For large sensor oscillation amplitudes this reflects directly the local tip-sample capacitance. The real part is given by the RC-constant of the tip-sample capacitance and the sample resistance to ground and thus contains information on the local sample resistivity and dissipation. For R_S → 0 the real part is zero. In the limit of ω_rfR_SC_ts ≫ 1 and R_S → ∞ both, the real and the imaginary parts become zero because the tip-sample capacitance cannot be charged during an rf cycle. In the QHE regime, both limits are conceivable since the 2DEG can contain locally well conducting and poorly conducting regions and the 2DEG's quantum capacitance can go to zero with diminishing density of states.

We performed DSCM measurements at a temperature of T = 2.1 K at the edge of a 500 μm wide and 5 mm long Hall bar defined by wet-chemical etching in a remotely doped AlGaAs/GaAs heterostructure with a 2DEG 50 nm below the sample surface [26]. The magneto-transport characteristics are shown in figure 1(b) and exhibit well-developed quantum Hall plateaus in the transverse resistance, R_xy, and Shubnikov–de Haas (SdH) oscillations in the longitudinal resistance, R_xx. The Hall slope and the SdH oscillations both yield an electron density of ∼3.9 × 10¹⁵ m⁻² and we find a zero-field Drude mobility of ∼45 m²/Vs. Henceforth, we use the term 'filling factor' with reference to these bulk properties. Unless stated otherwise, the DSCM experiments were performed without passing a current through the sample.

Figures 2(a)–(c) show the topography of the sample and the simultaneously recorded imaginary and real part of G_ts,ω, respectively, in a scan across the Hall bar edge at B = 0. The white contour lines overlaid on all images are extracted from the topography and indicate the position of the physical edge of the Hall bar. The real part in figure 2(c) is essentially zero and shows no structure. In contrast, the magnitude of the imaginary part in figure 2(b) is significantly larger on the Hall bar than over the etched region. On the 2DEG the signal is homogeneous except for small variations introduced by the topography, as for example highlighted by arrows in figures 2(a) and (b) [25]. The physical edge does not correspond exactly to the onset of the capacitance signal, which we attribute to a depletion of the 2DEG near the edge due to missing dopants and the increased surface potential leading to the 2DEG confinement. We note a small minimum close to the edge of the Hall bar where the signal is reduced below the value far from the 2DEG (see also cross sections in figure 3). Finite-element method (FEM) simulation confirm that this is a pure geometrical effect whereby the capacitance variation over a sensor cycle is enhanced due to an increased coupling of the 2DEG to the side of the AFM tip [27].

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The amplitude of the imaginary part of the signal can be interpreted using the lumped circuit model in the inset of figure 2(c) [28]. The coupling between the tip and the 2DEG can be described by three capacitances in series; the tip–surface capacitance, C_ts(t); the capacitance of the GaAs cap layer, C_cap, and the self or quantum capacitance C_2DEG of the 2DEG. The current path to ground is closed by the capacitance of the substrate, C_sub, in parallel with the resistance through the 2DEG, R. We estimate R from the homogeneous zero-field bulk conductivity and a circular-symmetric current density to the tip [22, 27]. Simple modelling indicates that except for very large R the substrate capacitance can be neglected. Therefore, the impedance G_ts can be reduced to the tip-sample capacitance in series with the sample resistance.

Figure 3 shows DSCM scans in the QHE regime between bulk filling factor ν = 2.08 and ν = 1.58. The left-hand column, images (a)–(e), shows the imaginary part of G_ts,ω; the central column, images (f)–(j), shows the real part of G_ts,ω. Cross-sections perpendicular to the sample edge are plotted in the right-hand column, panels (k)–(o).

At ν = 2.08 the imaginary part of the DSCM signal shows a ∼900 nm wide bright stripe close to the sample edge. The signal level in the interior of the sample is the same as the background over the etched region. Both findings are in stark contrast to the homogeneous signal over the 2DEG observed at zero magnetic field. The maximum signal on the stripe, however, is comparable to the signal at zero magnetic field. The real part of the DSCM signal shows a sharp peak (FWHM ≈ 250 nm) at the boundary between the stripe in the imaginary part and the sample interior. It occurs where the imaginary part decays to approximately half of its maximum value, and it has a width that is comparable to the distance over which the imaginary part drops to the background level in the sample interior. In the interior of the sample the real part is negligible, similar to that at zero magnetic field.

The features at ν = 1.81 are similar in form to those at ν = 2.08. However, the width of the stripe in the imaginary part has increased by approximately 50% compared to ν = 2.08 and the signal does not return exactly to the background level in the sample interior. At ν = 1.72 the stripe structures remain clearly defined in both the real and imaginary part. Both show an increased background in the interior of the sample. This background increases considerably for ν = 1.66, where the stripe structures are difficult to distinguish and both signals are relatively large in the bulk. We note that the stripe maxima in the imaginary part have not changed significantly between the different filling factors. At ν = 1.58, near a maximum of R_xx, the imaginary part image is almost identical to B = 0, see figure 2, e.g. the response is homogeneous across the sample. The real part, though smaller than at ν = 1.66, still exhibits a shallow monotonic increase away from the edge.

Figure 4(a) shows a composite image of numerous scans of Im(G_ts,ω) along ∼100 μm of the sample edge at ν = 1.90. In previous experiments, we found that the stripe width depends on the edge confinement potential [25]. Here we observe no significant variation in the shape or the amplitude of the stripes. This also holds for bent edges at Hall bar contacts, as illustrated in figure 4(b). These measurements suggest that the electron density and the edge confinement do not vary significantly even over macroscopic distances in this sample.

Zoom In Zoom Out Reset image size

The described structures in the DSCM images are repeated 1/B-periodically at lower magnetic fields. As an illustration, figure 5 shows line scans across the edge as a function of magnetic field at a fixed position along the sample. The periodicity is consistent with the electron density extracted from the transport data. For comparison, the longitudinal resistance of the Hall bar is plotted in grey on the same field axis.

Zoom In Zoom Out Reset image size

Considering the imaginary part of the signal in figure 5(a), we find that at magnetic fields where R_xx is zero the stripe structure is well defined and the signal in the bulk is at a minimum. With increasing field the stripe widens continuously and at the high-field end of the R_xx = 0 region, the signal becomes homogeneous on the scale of our experiment. At the low-field ends, near integer filling factors, the transition from the strip structure to an almost homogeneous signal with decreasing field occurs on a much smaller field scale. At fields between integer filling factors where R_xx has a maximum, the scans become comparable to the ones at zero field. The real part of the signal, Re(G_ts,ω), follows the inner boundary of the stripe in Im(G_ts,ω) and diverges into the sample interior at the high-field ends of the R_xx = 0 intervals, see figure 5(b). At these fields the width of the Re(G_ts,ω) stripe, measured perpendicular to the sample edge, increases continuously. At the low-field ends, the bulk signal of Re(G_ts,ω) is clearly different from zero, but the stripe is still visible. Around ν = 3, where the spin-degeneracy of the LLs is lifted by the Zeeman energy, the imaginary part is slightly suppressed in the bulk and a bright stripe in the real part occurs.

We consider the lumped circuit representation of the experiment shown in figure 2(c). Neglecting the substrate capacitance, the total complex conductance is given by the tip–surface and the 2DEG capacitance in series with the resistance of the 2DEG from the tip-position to ground and can be described by equation (2) for large enough sample conductances. The capacitance variation due to the tuning fork oscillation is approximately constant throughout a scan, so that the contrast in our images, apart from topographic and geometric artefacts, is determined solely by the local properties of the 2DEG.

In this simplified picture of a large local sample conductivity, variations in Im(G_ts,ω) depend on the quantum capacitance of the 2DEG, which is proportional to the density of states at the Fermi energy D(E_F). Variations in the real part, Re(G_ts,ω), are determined by the effective RC-charging time of the 2DEG at the sensor position. If the AFM-cycle average of C_ts remained constant over a scan, the variations of Re(G_ts,ω) could simply be interpreted as changes in the local sample conductivity. However, in the QHE regime both, R and the 2DEG's capacitance vary with position and their contributions cannot be disentangled without model assumptions [27]. Additional complexity arises from the fact that the AFM tip couples to a relatively large area so that the finite size of the tip and the inhomogeneity of the sample characteristics should be considered.

Based on the above model the bright stripe in Im(G_ts,ω) can be interpreted as the compressible regions at the edge of the sample. The model assumptions are valid over these regions as D(E_F) is large and the resistance to ground relatively small. The signal strength is similar to the zero-field scans, which adds weight to the conclusion that the signal is determined by the capacitive part, while the resistive part is negligible. Theoretical models suggest that the compressible regions at the edge are separated from the sample interior by a large resistance across the innermost incompressible strip [6, 28, 29]. In our experiments Re(G_ts,ω) measures the RC-time constant of the local portion of the 2DEG. It increases at the boundary between the compressible strip and the sample interior and is consistent with the inner-most incompressible strip insulating the sample interior from the contacts. Over an incompressible strip the signal is further suppressed due to D(E_F) tending to zero. Between integer filling factors the edge strips at the physical sample edges and at potential fluctuations in the bulk extend and broaden and form a percolating network of compressible regions [15]. At these fields the coupling between delocalised edge states increases, which leads to the transition between the quantized QHE plateaus [14]. The resulting increase in back-scattering leads to the finite longitudinal resistance. Our experiments therefore illustrate how the QHE edge strips determine the transport characteristics at a given magnetic field: if the strip structure is well-defined the longitudinal resistance is zero and a QHE plateau develops, while the resistance becomes finite when the strips diverge into the bulk of the sample.

The compressible and incompressible strips develop due to the nonlinear screening of the edge potential by the 2DEG in the QHE regime [5]. In our experiments, we find for the width of the incompressible strips ∼250 nm, consistent with other experiments in an unstructured 2DEG [29]. These values are larger than those extracted from experiments on quantum point contacts [30, 31], since the strip width scales inversely with the electron density gradient at the corresponding position [5]. Wider strips have been observed on tip-induced depleted regions of the 2DEG, with a more shallow density gradient [25]. In contrast to theoretical expectations [5], in our experiments the incompressible strips diverge into the bulk at non-integer bulk filling factors, consistent with the high-field end of the zero-resistance intervals of the longitudinal resistance. However, the integer filling factors roughly mark the low-field end of these intervals. This relationship of the strip structure to the bulk filling factor might be related to the nature of the disorder potential that determines the QHE transition [6].

We do not resolve any signal variations that might indicate the presence of the outer incompressible strips with a smaller local filling factor which one might expect closer to the sample edge. This is in qualitative agreement with theoretical results [6] suggesting that these strips are very narrow (i.e. possibly below the resolution of the DSCM) and that they do not sufficiently insulate the compressible states from the contacts. In our experiments, we observe that when the innermost incompressible strip becomes very narrow the real part of the DSCM signal in the bulk of the sample is enhanced. This suggests that the resistance across an incompressible strip is reduced considerably for smaller widths.

The QHE breaks down for currents larger than a specific critical value [26, 32]. The mechanisms of this breakdown are only poorly understood and no scanning probe data are available that investigate the electronic structure of the 2DEG in this regime. In figures 6(a) and (b), the longitudinal resistance R_xx is plotted as a function of the applied total current I for the filling factors ν = 1.81 and ν = 2.08 (the current is given on the axis of figures 6(e) and (f)). At ν = 1.81 the longitudinal resistance increases sharply above a critical current I_c ≈ 50 μA, from zero to R_xx ≈ 1 kΩ, where it saturates. This resistance is considerably larger than the zero magnetic field value of ∼240 Ω. At ν = 2.08 we observe in figure 6(b) a weak continuous increase of the resistance starting around I = 50 μA to R_xx ≈ 100 Ω, followed by a sharp increase to R_xx ≈ 1 kΩ at I_c ≈ 200 μA.

Zoom In Zoom Out Reset image size

Figures 6(c)–(f) show DSCM line scans as a function of the current I at the two filling factors discussed above; panels (c) and (d) show the imaginary, (e) and (f) the real part of the signal. For all currents smaller than I_c both signals exhibit the same stripe structure as at zero current. At I_c the pattern changes abruptly from the narrow stripe at the edge of the sample to a much more homogeneous signal across the 2DEG. In the imaginary part at ν = 1.81, shown in figure 6(c), the bright stripe does not end abruptly at I_c, but decreases in width when further increasing the current. In contrast, the bulk of the sample switches abruptly from zero to a finite value, which corresponds approximately to half of the zero-field and zero-bias experiment. The stripe observed in the real part immediately disappears at I_c. The behaviour at ν = 2.08 is generally similar to ν = 1.81, except that the critical current is much larger. However, while we observe a finite longitudinal resistance already below I_c, the stripe structure in the DSCM images at these intermediate currents deviates only marginally from the zero-current structure. However, a slight narrowing and a small decrease in amplitude are clearly visible. The images do not change under reversal of the bias current.

At zero-bias the incompressible strip, mainly observed in the real part of the signal, is slightly wider at ν = 1.81 than at ν = 2.08. In addition the imaginary part shows a finite signal in the sample interior, see figure 3. From the latter one can deduce that the density of states at the Fermi energy in the interior is not zero. We adopt the following picture of the QHE breakdown [33]: at a large enough bias the localized compressible strips in the bulk follow the equipotential lines, which become distorted by the Hall potential for finite currents. At the critical current the localized states percolate and potential fluctuations become irrelevant. This transition from inhomogeneous to homogeneous sample properties happens on a very small current interval, as observed in our experiments. Qualitatively this model also explains why the compressible strip can still be discriminated at I > I_c at ν = 1.81, since the Hall potential would affect the electronic structure at the edges of the sample less than in the bulk. The differences found for different filling factors might then be due to the different insulating and screening properties of the compressible and incompressible strips at a given magnetic field. In particular, the narrowing of the compressible strip at ν = 2.08 at intermediate currents might indicate that the Hall potential is screened differently than at low currents.

We have imaged the physical edge of a 2DEG in the QHE regime using DSCM. We interpret our measurements as spatial images of the local complex tip-sample-ground conductance. We find bright stripes in both, the imaginary and the real part of the signal around integer filling factors, whereas at fields where the longitudinal resistance has a maximum, we find essentially homogeneous signals comparable to the scans at zero magnetic fields. We argue that the stripes are directly related to the compressible and incompressible strips forming at the edge of the sample as a consequence of the nonlinear screening of the 2DEG in the QHE regime. We present DSCM images of these stripes over macroscopic distances and over many QHE periods. An investigation of these edge strips as a function of the applied current shows an abrupt transition from the strongly inhomogeneous DSCM signals at low currents to essentially homogeneous signals for currents larger than the critical current obtained for the QHE break-down in transport experiments.

This work was financially supported by the Engineering and Physical Sciences Research Council (UK) through research grant no. EP/D500222/1.