betaFeedback

WorksPhysics and Astronomy

Zero-field plasmon splitter: Controlled interaction between quantum anomalous Hall edge states

Abstract

The quantum anomalous Hall (QAH) effect provides chiral edge states without the need for an external magnetic field, but unlike in conventional quantum Hall systems, their trajectories cannot be readily controlled electrostatically. Here, we engineer Coulomb interactions between two spatially separated, counter-propagating QAH edge channels using a floating metallic island. Microwave measurements reveal frequency-dependent redistribution of injected edge plasmons between two output channels, demonstrating interaction-induced hybridization of the collective modes. A long-range interaction model quantitatively reproduces the response and reveals a very low plasmon velocity in the coupling region, strongly reduced by electrostatic screening. Our results establish engineered Coulomb coupling as a route to control collective excitations in QAH edge states and provide a building block for chiral plasmonic circuits and electron quantum optics in magnetic topological insulators.

Introduction

The quantum anomalous Hall (QAH) effect in magnetic topological insulators supports chiral edge transport without the need for an external magnetic field [1, 2, 3]. These edge states also support collective charge excitations, or edge plasmons, which can be excited and detected at microwave frequencies [4, 5, 6]. Their propagation direction is set by the magnetization of the material and can be reversed by magnetic-field or current-induced switching [7, 8, 9]. This makes QAH systems interesting for applications ranging from chiral electronics and quantum metrology to quantum-information concepts [10, 11, 12, 13, 14]. Controlling how two QAH edge states couple is an important step toward using them in more complex devices. In one-dimensional electronic systems, Coulomb interactions lead to collective charge modes whose propagation depends strongly on the surrounding electrostatic environment [15]. In conventional quantum Hall systems, capacitive coupling between nearby edge channels has been used to study plasmon propagation, mode coupling, and charge fractionalization [16, 17, 18, 19, 20]. Importantly, such coupling allows the charge dynamics of one edge channel to influence another without requiring electrons to tunnel between them.

Quantum Hall systems also offer direct control over chiral edge trajectories through quantum point contacts. These can partition electronic excitations between different paths and have been used to realize electronic beam splitters and interferometers [21, 22, 23, 24]. In a QAH insulator, however, there is only a single chiral edge state, and an equivalent electrostatic control of its path is not readily available. An alternative approach is therefore required to manipulate QAH edge excitations without relying on electron tunneling.

Coulomb interactions provide such an alternative. When two spatially separated edge states are brought close together, their collective charge excitations can interact through their electrostatic environment even in the absence of particle exchange. Such interaction-mediated coupling has been extensively studied in conventional quantum Hall edge channels [18, 19, 25, 26, 27], where it can lead to hybridization of co- or counter-propagating plasmons and to frequency-dependent redistribution of an injected excitation between different channels. Implementing and controlling this mechanism in the QAH regime would provide a new route toward manipulating chiral collective excitations at zero magnetic field.

Here, we engineer such an interaction between two spatially separated, counter-propagating QAH edge states using a floating metallic island. The island provides a common electrostatic environment for the two edges and thereby mediates a nonlocal Coulomb coupling. We realize this geometry as a plasmon splitter in which an excitation injected into one QAH edge can emerge in either of two spatially separated output channels. Microwave transmission measurements reveal frequency-dependent oscillations in the relative transmission between the two output paths, demonstrating interaction-induced hybridization and mode conversion of the coupled edge plasmons. The coupling region strongly modifies the plasmon propagation through electrostatic screening. The metallic island screens the electric field associated with charge excitations along the edge, thereby increasing the effective capacitance per unit length and equivalently reducing the plasmon velocity. By comparing devices with different interaction-region lengths and with a reference device without the metallic island, we extract the propagation time and the effective plasmon velocity in the coupled region. We find a velocity of vs=(12±4) km s−1v_{\mathrm{s}} = (12 \pm4)\ \mathrm{km}\ \mathrm{s}^{-1}, more than an order of magnitude smaller than the velocities vs=(300−500) km s−1v_{\mathrm{s}} = (300 - 500)\ \mathrm{km}\ \mathrm{s}^{-1} reported for isolated QAH edge states [4, 5, 28, 29]. The resulting shorter plasmon wavelength brings the interaction-induced mode conversion into the experimentally accessible microwave-frequency range. Our results establish nonlocal Coulomb coupling as a means of engineering interactions between QAH edge states without modifying their topology or introducing electron tunneling. More generally, the demonstrated control of collective excitations provides a route toward interaction-engineered chiral plasmonic circuits and interferometric devices operating at zero magnetic field.

Results

Principles of the experiment

We study the Coulomb interaction between chiral edge states of the QAH insulator V-doped (Bi1−xSbx)2Te3(\mathrm{Bi}_{1-x}\mathrm{Sb}_{x})_{2}\mathrm{Te}_{3} (V-BST) [1, 2] by fabricating a plasmon splitter device, as shown in Figure 1.

Plasmon splitter device

Figure 1. Plasmon splitter device. (A) Full microscope image of a splitter device. Ohmic contacts (highlighted in light blue) allow for Hall measurements and coplanar waveguides transmit microwave signals to the four ports used for microwave experiments. (B) Zoomed-in image of a device. The two V-BST bars (light brown) are physically separated by a thin trench in the center. The edge states of each bar (highlighted by the red and blue lines) are capacitively coupled via the metallic island covering the trench. Port 1 is used for the microwave excitation, and the other three for detection of the transmitted signal, yielding the scattering parameters S21S_{21}, S31S_{31} and S41S_{41}.

The plasmon splitter device comprises a coupling region in which two counter-propagating edge channels coexist in close proximity. This is achieved by etching a thin trench into the V-BST layer, separating it into two disconnected bars. A floating metallic island is placed above the trench by capacitively coupling it via a 14 nm-thick Al2O3\mathrm{Al}_{2}\mathrm{O}_{3} layer serving as a dielectric. The metallic island strongly enhances the inter-edge coupling within the coupling region. Narrow microwave finger gates, connected to coplanar waveguides (labeled Port 1-4), excite and detect edge plasmons via a four-port vector network analyzer (VNA). Port 1 injects the microwave signal, while the remaining three ports measure the transmission (S21S_{21}, S31S_{31}, S41S_{41}) through the edge channels. Ohmic contacts (located away from the island) allow for the dc device characterization via standard low-frequency lock-in measurements, confirming the QAH state. Device A exhibits a Hall resistance within 2% of RKR_{\mathrm{K}} with small longitudinal resistance as shown in Figure 2. The dc characteristics of all devices are summarized in Table 2. Using the ohmic contacts, we verify electrically that the island is capacitively coupled to the V-BST bars and is not ohmically connected.

Figure 2. Magnetic field dependence of dc and microwave response. (A) Magnetic field sweeps of dc longitudinal and transverse resistances RxxR_{xx} (light green lines) and RyxR_{yx} (violet lines). The black dashed lines correspond to R=±RKR = \pm R_{\mathrm{K}}, with RK=h/e2R_{\mathrm{K}} = h/e^2. (B) Magnetic field sweep of corrected transmission amplitude of S21S_{21} (blue lines), S31S_{31} (beige lines) and S41S_{41} (red lines) for f=1.5f = 1.5 GHz and an excitation power of P≈−51P \approx-51 dBm (VNA output power of −10-10 dBm). Both measurements are taken on device A at T≈20T \approx20 mK.

SampleRyx/RKR_{yx}/R_K [%]RxxR_{xx} [Ω]Coupling
A (Dec1A)99.3 ± 0.2170 ± 206 μm island
B (Dec17B)98.7 ± 0.2200 ± 3012 μm island
C (Mar6B)129 ± 41200 ± 50no island
D (Jul9A)96.9 ± 0.3600 ± 706 μm island
E (Mar6B)129 ± 41200 ± 502 μm island
F (Oct9A)99.3 ± 0.223 ± 10no island

Table 2. Overview of samples. Hall resistance RyxR_{yx}, longitudinal resistance RxxR_{xx}, and the coupling of all devices. All measurements are taken for μ0H=2\mu_0 H = 2 T and T≈20T \approx20 mK.

We investigated six splitter devices with different dimensions of the metallic island. Two devices contain no metallic island to quantify the residual inter-edge coupling. The remaining four devices incorporate metallic islands with trench lengths LL of 2 μm, 6 μm (two devices), and 12 μm, enabling a systematic study of the length dependence. We focus on three devices in the main text. The lengths of the metallic islands on devices A and B are 6 μm and 12 μm, respectively (see Figure 1). Device C has no metallic island and a trench length of 2 μm. We discuss the other devices in Sec. S7 of the Supplementary Material.

Evidence for inter-edge coupling

We first establish inter-edge coupling and show that the metallic island strongly enhances the interaction between the two edge channels. Figure 2B displays the magnetic-field sweeps of the scattering-parameter amplitudes for each detection port, recorded at the frequency f=1.5f = 1.5 GHz. In the present geometry, a negative magnetization (at - 2 T) results in clockwise chirality, as highlighted in Figure 1B. In that case, plasmons injected in port 1 are directed towards the coupling region and are detected at ports 2 and 4, yielding a finite transmission S21S_{21} and S41S_{41}. A positive magnetization of the sample (at +2 T) corresponds to a counterclockwise chirality of the edge channel, resulting in no transmission (see Materials and Methods). In analogy to the dc characterization, the microwave signals measured at ports 2 and 4 exhibit a pronounced hysteresis upon reversal of the magnetization. The hysteretic microwave response confirms that the detected signals originate from chiral QAH edge transport [4]. For clockwise chirality, the simultaneous finite transmission to ports 2 and 4 indicates redistribution of the injected excitation within the coupling region. By contrast, the corrected signal at port 3, upstream, remains small, consistent with the expected propagation direction in this geometry. A weak hysteretic contribution is nevertheless observed at port 3 for some devices.

We now turn to analyzing the frequency-dependent amplitude and phase of both transmission coefficients S21S_{21} and S41S_{41}. The amplitude and phase of the spectra of samples A and B are shown in Figure 3B and C, respectively. Interestingly, at low frequencies, the signal transmitted to the opposite side of the plasmon splitter S21S_{21} exceeds the signal remaining on the same side S41S_{41}. This behavior reverses at higher frequencies (around 1.2 GHz for device A and around 1.6 GHz for device B), such that S41S_{41} becomes the dominant transmission channel. Both transmission amplitudes exhibit oscillations before approaching the same value at higher frequencies. The oscillations are more pronounced for Device A (6 µm-long splitter) than for Device B (12 µm-long splitter), which we attribute to stronger dissipation in the longer island, leading to equal amplitudes in S21S_{21} and S41S_{41} already above 1.6 GHz. These oscillations are absent in reference device C without a metallic island, as shown in Figure 3A. Here, most of the signal is transmitted via S41S_{41}, while S21S_{21} increases slowly with frequency. This demonstrates that, in the absence of a metallic island, a weak residual coupling remains, whereas the metallic island mediates a much stronger coupling between the edge states, leading to an oscillatory behavior, previously observed for co-propagating quantum Hall edge channels [18].

Figure 3. Microwave transmission S21S_{21} and S41S_{41}. Amplitude and phase of S21S_{21} and S41S_{41} for (A) Device C without a metallic island, (B) Device A with a 6 μ\mum-long metallic island, and (C) Device B with a 12 μ\mum-long metallic island. The solid lines show fits using the complete transmission model of Eq. (3). The interaction parameters τp,s\tau_{p,s} and τr,s\tau_{r,s} are fixed to the values extracted from the ratio analysis in Figure 5, while the remaining parameters τp,e\tau_{p,e} and τr,e\tau_{r,e} are determined in a second step, as detailed in the main text. The fit parameters are shown in Table 1. Measurements are taken for μ0H=−2\mu_0 H = -2 T and T≈20T \approx20 mK.

The coupling between the two edge channels depends on the phase, which the injected plasmon accumulates while traversing the coupling region. For an accumulated phase ϕ=2πn\phi= 2\pi n, the positive and negative charge oscillations cancel over the length of the coupling region, leaving the metallic island uncharged and resulting in only weak capacitive coupling between the edge channels and maximal transmission in S41S_{41}. In contrast, for ϕ=(2n+1)π\phi= (2n + 1)\pi, the oscillations add constructively, leading to maximal charging of the floating island and therefore to the strongest coupling between the counter-propagating edge channels. As the microwave frequency increases, the accumulated phase changes continuously, giving rise to the oscillatory splitting observed experimentally. Electrostatic screening by the metallic island strongly reduces the plasmon velocity and therefore shortens the plasmon wavelength compared to the bare edge, bringing the oscillation to an observable range f<4f < 4 GHz. Based on this physical picture, we develop a quantitative model of the plasmon splitter in the following.

Long-range interaction model

To quantitatively describe the measured microwave transmission, we model the device using a distributed-circuit approach that has previously been shown to accurately capture the excitation and propagation of edge plasmons in both quantum Hall and QAH systems [4, 18, 25, 26, 30]. Here, we extend this framework to include the capacitive coupling between two counter-propagating edge states mediated by the floating metallic island. As illustrated in Figure 4A, the device is described by three scattering coefficients corresponding to (i) microwave excitation SegS_{\mathrm{eg}} and detection SgeS_{\mathrm{ge}}, (ii) propagation along the bare edge channels SeS_{\mathrm{e}}, and (iii) the plasmon splitter Ss,iS_{\mathrm{s},i}. Microwave

Figure 4. Model of the plasmon splitter. (A) Equivalent circuit describing the microwave transmission. The measured scattering parameters are decomposed into three contributions: microwave excitation SegS_{\mathrm{eg}} and detection SgeS_{\mathrm{ge}}, propagation along the bare edge channels SeS_{\mathrm{e}}, and propagation through the plasmon splitter Ss,2S_{\mathrm{s},2} and Ss,4S_{\mathrm{s},4}. The bare edge is characterized by the propagation and relaxation times τp,e\tau_{\mathrm{p,e}}, τr,e\tau_{\mathrm{r,e}}, while the splitter is described by τp,s\tau_{\mathrm{p,s}}, τr,s\tau_{\mathrm{r,s}}, τc,s\tau_{\mathrm{c,s}}. (B) Cross-sectional schematic of the plasmon splitter. Two counter-propagating edge channels form collective plasmon modes mediated by a floating metallic island connected via a capacitance CiC_{\mathrm{i}}, corresponding to the charging time τc,s=RKCi\tau_{\mathrm{c,s}} = R_{\mathrm{K}}C_{\mathrm{i}}.

excitation and detection occur capacitively through the gate capacitance CgC_{\mathrm{g}} and are modeled following Ref. [30]. Propagation along the bare edge channels follows the distributed transmission-line model developed in Ref. [4] and takes charge puddles screening into account. We introduce a propagation time τp,e\tau_{\mathrm{p,e}} and a relaxation time τr,e\tau_{\mathrm{r,e}}, yielding the complex wavevector ke(ω)k_{\mathrm{e}}(\omega) such that:

ked=ωτp,e1+iωτr,e(1)k_{\mathrm{e}}d = \frac{\omega\tau_{\mathrm{p,e}}}{1 + i\omega\tau_{\mathrm{r,e}}} \tag*{(1)}

Here, τp,e=d/ve\tau_{p,e}=d/v_e describes the propagation time through a section of length dd at a velocity vev_e, while τr,e\tau_{r,e} phenomenologically accounts for frequency-dependent dissipation. Throughout this work, Eq. (1) is also used to describe propagation in the coupling region, with different values of τp,s\tau_{p,s} and τr,s\tau_{r,s} reflecting the different electrostatic environments. For the bare edge channels, the propagation is described by the scattering coefficient Se=Aexp⁡(−iked)S_e=A\exp(-\mathrm{i}k_e d), where kek_e follows Eq. (1) with the parameters τp,e\tau_{p,e} and τr,e\tau_{r,e} and AA is a real number that accounts for losses in dc.

The coupling region requires an additional description because the two counter-propagating edge states become capacitively coupled underneath the metallic island. Two limiting cases have previously been proposed [18]. The zero-range model assumes local capacitive coupling along the propagation direction in a Luttinger liquid framework [31, 32], whereas the long-range model describes the interaction through the common electrostatic potential of a floating conductor represented by a lumped capacitance CiC_i. In our device, the metallic island remains at an equipotential because its internal RC time is much shorter than the plasmon propagation times of several hundred picoseconds. The long-range model is therefore expected to apply. A plasmon propagating along either edge therefore modifies the electrostatic potential experienced by both edge channels over the entire length of the island, mediating a nonlocal Coulomb interaction between them. The effective interaction strength is consequently set by the island length LL, rather than by the distance of the edge channels. Solving the coupled equations for the collective plasmon modes (see § 13 of the Supplementary Material for details) yields:

Ss,2=1−e−iksL2(1+iωτc,s(1−e−iksL)),Ss,4=1−Ss,2,(2)S_{s,2}=\frac{1-e^{-\mathrm{i}k_sL}}{2\left(1+\frac{\mathrm{i}}{\omega\tau_{c,s}}\left(1-e^{-\mathrm{i}k_sL}\right)\right)},\qquad S_{s,4}=1-S_{s,2}, \tag*{(2)}

where τc,s=RKCi\tau_{c,s}=R_KC_i is the charging time of the floating metallic island acting as a splitter. The propagation inside the coupling region is described by the same dispersion relation, Eq. (1), using different parameters τp,s\tau_{p,s}, τr,s\tau_{r,s} and replacing the bare edge length dd with the length LL of the coupling region.

To reduce the number of fit parameters, we consider the appropriate limiting case for each device. For devices containing a metallic island, the investigated frequencies satisfy ωτc,s≫1\omega\tau_{c,s}\gg1 and the charging dynamics of the floating island can be neglected such that the response depends only on the two parameters τp,s\tau_{p,s} and τr,s\tau_{r,s}. For the reference device without a metallic island, we instead set τr,s=0\tau_{r,s}=0, corresponding to the absence of additional dissipation in the coupling region, while retaining CiC_i to account for the residual inter-edge coupling capacitance Cs=Ci/2C_s=C_i/2. Finally, owing to the chiral nature of the edge channels and the absence of reflections, the total transmission is obtained as

Si1=SgeSeSs,iSeSeg,i=2,4.(3)S_{i1}=S_{ge}S_eS_{s,i}S_eS_{eg},\qquad i=2,4. \tag*{(3)}

Although the model contains only a small number of physically meaningful parameters, several of them contribute simultaneously to the measured transmission spectra. To reduce parameter correlations, we first determine the splitter parameters τp,s\tau_{p,s} and τr,s\tau_{r,s} from the ratio η=S21/S41\eta= S_{21}/S_{41} (denoted hereafter as the splitting ratio). The scattering coefficient for the common excitation port SegS_{eg} and bare-edge propagation SeS_e are expected to cancel out for S21S_{21} and S41S_{41}, thanks to the symmetric device geometry. The parameters τp,s\tau_{p,s} and τr,s\tau_{r,s} are subsequently fixed when fitting the complete transmission according to Eq. (3). A comparison with the zero-range interaction model is presented in § 14 of the Supplementary Material.

Comparison between model and data

The solid lines in Figure 5A,B show fits using the simplified long-range interaction model in Eq. (2) describing the amplitude and phase of the splitting ratio η\eta. Despite containing only the two free parameters τp,s\tau_{p,s} and τr,s\tau_{r,s}, the model reproduces the oscillation period and the overall frequency dependence over the complete experimental bandwidth. The only noticeable discrepancy is a systematic overestimation of the phase. As shown in § 15 of the Supplementary Material, this originates from unequal gate-edge coupling SgeS_{ge} of the two detection gates.

Figure 5. Frequency dependence of the splitting ratio S21/S41S_{21}/S_{41}. (A) Amplitude and (B) phase of the ratio as a function of the frequency for Devices A (6 μ\mum metallic island, shown as green line) and C (no metallic island, shown as violet line). The solid lines represent the fit with Eq. (2). We find: τp,s=(630±10)\tau_{p,s} = (630 \pm10) ps, τr,s=(31±2)\tau_{r,s} = (31 \pm2) ps for Device A and τp,s=(144±5)\tau_{p,s} = (144 \pm5) ps, Ci=(3.0±0.1)C_i = (3.0 \pm0.1) fF for Device C. (C) The fit parameter τp,s\tau_{p,s} is shown as a function of the island length LL. Light green points correspond to Devices A, B, D, E, and light blue points to the reference Devices C and F, which are plotted at L=0L = 0 to account for the residual inter-edge coupling without a metallic island. The fit parameter τp,s\tau_{p,s} stems from fits of the ratio S21/S41S_{21}/S_{41}, as shown in (A) and (B). The error bars show the fit uncertainties. The black line corresponds to a linear fit, which yields a velocity of vs=(12±4)v_s = (12 \pm4) km s−1^{-1}. All measurements are taken for μ0H=−2\mu_0 H = -2 T and T≈20T \approx20 mK.

For Device A (6 μ\mum-long metallic island), the fit yields a propagation time of τp,s=(630±10)\tau_{\mathrm{p,s}} = (630 \pm10) ps and a relaxation time of τr,s=(31±2)\tau_{\mathrm{r,s}} = (31 \pm2) ps. Repeating the analysis for different island lengths yields the propagation times summarized in Figure 5C, including the propagation times for samples without metallic islands shown for L=0L = 0. Their roughly linear dependence on the island length LL demonstrates that the measured propagation times accumulate within the coupling region. A linear fit gives a propagation velocity of vs=(12±4) km s−1v_{\mathrm{s}} = (12 \pm4)\ \mathrm{km\ s}^{-1} in the coupling region, about 20 to 50 times smaller than in bare edge states [4]. The pronounced reduction of vsv_{\mathrm{s}} is an expected consequence of the electrostatic screening introduced by the floating metallic island, which increases the effective capacitance and thereby slows the collective plasmon mode. This relation can be quantified using the dispersion relation of Eq. (1). In the experimentally relevant regime ωτr,s<1\omega\tau_{\mathrm{r,s}} < 1, the dispersion can be approximated as ksL≈ωτp,sk_{\mathrm{s}}L \approx\omega\tau_{\mathrm{p,s}}, yielding a group velocity vs=∂ω/∂ks=L/τp,sv_{\mathrm{s}} = \partial\omega/\partial k_{\mathrm{s}} = L/\tau_{\mathrm{p,s}}. Within the distributed-capacitance description, the propagation time is related to the distributed capacitance CeC_{\mathrm{e}} of the edge channel by τp,s=RKCeL\tau_{\mathrm{p,s}} = R_{\mathrm{K}}C_{\mathrm{e}}L. The line capacitance CeC_{\mathrm{e}} accounts for the electrostatic environment of the edge channel, including bare edge interactions and coupling to charge puddles and the metallic gates [4, 33]. As observed on gated devices [33], the capacitance to the gate is the relevant quantity, and hence we assume that the geometric capacitance of the metallic island is the dominant contribution to CeC_{\mathrm{e}}. Therefore, we estimate CeC_{\mathrm{e}} geometrically as Ce=cgeowC_{\mathrm{e}} = c_{\mathrm{geo}}w. Here, cgeoc_{\mathrm{geo}} is the capacitance per unit area between the metallic island and the edge channel, and ww is the effective overlap width, which we approximate as w=1 μmw = 1\ \mu\mathrm{m}. The experimentally determined τp,s\tau_{\mathrm{p,s}} then corresponds to cgeo≈3 mF m−2c_{\mathrm{geo}} \approx3\ \mathrm{mF\ m}^{-2}, in reasonable agreement with the geometrical estimate cgeo=ε0εr/t≈4.4 mF m−2c_{\mathrm{geo}} = \varepsilon_0\varepsilon_{\mathrm{r}}/t \approx4.4\ \mathrm{mF\ m}^{-2} for a t=14t = 14 nm-thick Al2_2O3_3 dielectric layer (εr≈7\varepsilon_{\mathrm{r}} \approx7). For the reference device without a metallic island (Device C), we estimate an effective interaction length Leff≈11 μmL_{\mathrm{eff}} \approx11\ \mu\mathrm{m} to account for the spatial extent of the residual inter-edge coupling (see § 16 of the Supplementary Material). The resulting effective velocity vs≈80 km s−1v_{\mathrm{s}} \approx80\ \mathrm{km\ s}^{-1} remains below the ve≈(200–500) km s−1v_{\mathrm{e}} \approx(200\text{--}500)\ \mathrm{km\ s}^{-1} reported for isolated QAH edge states [4]. This reduction is consistent with the formation of hybridized collective plasmon modes, whose propagation is modified by Coulomb coupling between the counter-propagating edge channels [34, 35].

Besides reducing the propagation velocity, the metallic island also introduces relaxation, quantified by the time constant τr,s\tau_{\mathrm{r,s}}. This relaxation limits the coherence of the coupled plasmon modes and suppresses the oscillations above the characteristic frequency fc≃12πτp,sτr,s≈1 GHzf_{\mathrm{c}} \simeq\frac{1}{2\pi\sqrt{\tau_{\mathrm{p,s}}\tau_{\mathrm{r,s}}}} \approx1\ \mathrm{GHz}, where the transmission gradually approaches the symmetric limit S21=S41S_{21}=S_{41}, corresponding to a splitting ratio η=1\eta= 1 of the incoming plasmon.

Having established the intrinsic properties of the plasmon splitter, we next consider the complete transmission model. The previously determined values of τp,s\tau_{\mathrm{p,s}} and τr,s\tau_{\mathrm{r,s}} are kept fixed, while the remaining parameters describing bare-edge propagation and microwave coupling are obtained by fitting the complete transmission according to Eq. (3). The excellent quantitative agreement with both amplitude and phase over the full frequency range in Figure 3A-C provides a stringent test and validates the model. Table 1 summarizes the extracted fit parameters. For all three devices, the bare-edge propagation times τp,e\tau_{p,e} are of comparable magnitude, ranging from approximately 30 to 60 ps across the two transmission paths, which correspond to bare-edge velocities of ve≈(300−600)v_e \approx(300 - 600) km s−1^{-1}, consistent with previous measurements on isolated QAH edge states [4]. This agreement supports the separation of the transmission into propagation along the bare edge and propagation within the coupling region. The extracted relaxation times τr,e\tau_{r,e} are likewise comparable across the devices, although some variation between the two transmission paths remains. The fit additionally yields different gate capacitances CgC_g for the individual detection gates, reflecting differences in their coupling to the edge. These asymmetries account for the small residual phase mismatch observed in the simplified ratio analysis (see Figure 5B).

DeviceA (L=6L = 6 µm)B (L=12L = 12 µm)C (no island)
S21S_{21}S41S_{41}S21S_{21}S41S_{41}S21S_{21}S41S_{41}
τp,s\tau_{p,s} [ps]630 ± 101030 ± 10144 ± 5
τr,s\tau_{r,s} [ps]31 ± 283 ± 14−
CiC_i [fF]−−3.0 ± 0.1
τp,e\tau_{p,e} [ps]43 ± 136 ± 159 ± 133 ± 150 ± 246 ± 1
τr,e\tau_{r,e} [ps]39 ± 127 ± 250 ± 118 ± 245 ± 137 ± 1
CgC_g [fF]94 ± 315 ± 142 ± 217 ± 111.8 ± 0.713.1 ± 0.3

Table 1. Table 1: Extracted fit parameters from the complete transmission model. For Device A (6 µm long metallic island), Device B (12 µm long metallic island), and the reference device without a metallic island (Device C). For Devices A and B, τp,s\tau_{p,s} and τr,s\tau_{r,s} are extracted from fits of the splitting ratio η=S21/S41\eta= S_{21}/S_{41} with Eq. (2) in the high-frequency limit ωτc,s≫1\omega\tau_{c,s} \gg1. For Device C, τp,s\tau_{p,s} and CiC_i are extracted from fits of η\eta with Eq. (2) in the limit τr,s→0\tau_{r,s} \to0. The bare-edge parameters τp,e\tau_{p,e}, τr,e\tau_{r,e}, and the gate capacitance CgC_g are then extracted from fits of the full transmission S21S_{21} and S41S_{41} with Eq. (3), while fixing the splitter parameters τp,s\tau_{p,s}, τr,s\tau_{r,s}, and τc,s\tau_{c,s} to the values extracted from the corresponding fits of η\eta.

Discussion

We have demonstrated an interaction-mediated plasmon splitter based on two counter-propagating QAH edge states without electron tunneling between the channels. A floating metallic island provides a common electrostatic environment for two spatially separated edges, giving rise to frequency-dependent redistribution of an injected plasmon between the two output channels. The observed oscillations in the splitting ratio are quantitatively reproduced by a long-range interaction model and provide direct access to the propagation and relaxation of the hybridized plasmon modes. A central consequence of the engineered electrostatic environment is the strong reduction of the plasmon velocity. We find vs=(12±4)v_s = (12 \pm4) km s−1^{-1} below the metallic island, compared with (300−500)(300 - 500) km s−1^{-1} reported for isolated QAH edge states. The strongly reduced velocity of the propagation mode demonstrates that the metallic island provides an efficient means to engineer Coulomb interactions between chiral edge states. The coupling geometry controls the interaction length and accumulated plasmon phase, providing a direct means to engineer the response of the splitter. This approach provides a route toward interaction-engineered chiral plasmonic circuits at zero magnetic field. Future control over the coupling geometry and electrostatic environment could enable coherent mode conversion, plasmon interferometry, and interaction-induced fractionalization in QAH systems. The present device also highlights important limitations for future quantum applications. The observed approach toward a symmetric splitting (η=1\eta= 1) at high frequency is not a lossless beam-splitter operation, but results from dissipation in the plasmon splitter, which suppresses coherent oscillations and drives the system toward equal partitioning. Reducing these losses and extending the coherence of the collective plasmon modes will be essential for realizing fully coherent plasmonic quantum circuits.

Methods

Fabrication

For the device fabrication, we use 8 nm thin films of V-doped (Bi1−xSbx)2Te3\mathrm{(Bi_{1-x}Sb_x)_2Te_3} grown via molecular beam epitaxy on InP (111)A substrates. The Bi:Sb ratio is optimized to place the chemical potential within the magnetic exchange gap without electrostatic gating, as confirmed by the observation of the QAH state in Hall measurements. The V-BST layer is protected by a 4 nm-thick layer of Al2O3\mathrm{Al_2O_3}, which is deposited globally via low-temperature atomic layer deposition (ALD). The device fabrication follows the cleanroom recipes described in Ref. [4]. All patterning steps were performed using electron beam lithography with the e-beam resist PMMA A4. After patterning, the resist was developed using a mixture of methyl isobutyl ketone (MIBK) and IPA in a 1:3 ratio, followed by rinsing in IPA. Each lift-off of residual resist was performed using the solvent N-methyl-2-pyrrolidone (NMP). The fabrication comprises the following steps: First, ohmic contacts are established by wet etching the capping layer with Transene aluminum etchant type D and immediately depositing 5 nm/30 nm of Pt/Au via sputtering afterwards. Second, the mesa was dry etched via Ar-sputtering, yielding two V-BST bars separated by a 1 μ\mum-wide trench. Third, a 10 nm-thick layer of Al2O3\mathrm{Al_2O_3} was globally deposited using ALD. It serves as the dielectric layer for the metallic island and the microwave finger gates. On top of the Al2O3\mathrm{Al_2O_3} layer, the rectangular-shaped metallic island and the 3 μ\mum-wide finger gates are deposited by sputtering 5 nm/30 nm of Pt/Au. Thus, the metallic island is capacitively coupled to the edge channels of both V-BST parts as indicated in Figure 4B. The geometry of the finger gates is optimized to minimize direct microwave crosstalk, ensuring that the measured transmission predominantly probes propagation through the edge channels. In the final step, most of the Al2O3\mathrm{Al_2O_3} layer was wet-etched such that it only covers the V-BST mesa and parts around it, and the coplanar waveguides and bonding pads were patterned using a 5 nm/45 nm-thick layer of Pt/Au.

Measurement

The devices are measured in a Bluefors dilution refrigerator LD250 at a base temperature below 20 mK. The fabricated devices are glued on a microwave-compatible PCB comprising 12 dc lines and 4 microwave lines connecting to the sample via coplanar waveguides. This allows us to characterize both V-BST bars during the same cooldown, as well as connecting and measuring all microwave ports simultaneously. A scheme of the setup and the components inside the dilution refrigerator is shown in Figure 6. Table 2 lists the state of quantization assessed by the dc Hall measurements for each device.

Figure 6. DC and microwave setup. A four-port VNA is connected to all four finger gates of the device, enabling measurement of the three transmission parameters S21S_{21}, S31S_{31}, and S41S_{41}. The input line (connected to port 1 of the VNA) contains attenuators of total -41 dB. Each of the detection lines is equipped with a cryogenic low noise amplifier (+30 dB). Detection lines 2 and 4 additionally comprise an attenuator (-10 dB) in order to dampen the standing wave forming between amplifier and sample. The dc wiring is exemplified for one V-BST bar on the right-hand side.

We study microwave transport using a four-port vector network analyzer (VNA), which is connected via broadband coaxial cables to the wiring of the dilution refrigerator. It allows us to inject microwave signals into port 1, creating edge plasmons that propagate along the edge channels of the two V-BST bars. We then measure the transmission to the remaining three ports 2, 3, and 4 by relating the input signal of the VNA to its detected signal and obtain the transmission coefficients S21S_{21}, S31S_{31}, and S41S_{41}. In the main text, we focus on an input power of P=−10P = -10 dBm applied from the VNA. Inside the cryostat, the input signal is attenuated at all stages, leading to an attenuation of −41-41 dB. This corresponds to about −51-51 dBm on the excitation port. Additional attenuators (−10-10 dB) are placed on the output lines to dampen the standing wave forming between the device and the broadband amplifiers (see Ref. [4]). To correct the signal for stray couplings, we apply the robust calibration scheme described in Ref. [4], based on the following principle: for counterclockwise chirality (at +2+2 T), the edge plasmon will propagate along a long path away from the splitter and dissipate. In this case, the measured signal is merely the stray signal, which we subtract from the transmission for clockwise chirality (at −2-2 T). In the main text, we only show the calibrated transmission data.

To measure the longitudinal RxxR_{xx} and transverse resistance RyxR_{yx}, we apply a current bias of 1 nA and record the voltages VxxV_{xx} and VyxV_{yx} using standard lock-in techniques. Due to the geometry of the ohmic contacts, mixing of RxxR_{xx} and RyxR_{yx} occurs, which becomes visible when the bulk becomes conducting, e.g., as small peaks in the hysteresis loop of RyxR_{yx} at the coercive field. The Hall resistance reaches Ryx=h/e2R_{yx} = h/e^2 within an accuracy of 2 %, while the longitudinal resistance remains below Rxx≲1R_{xx} \lesssim1 kΩ. Devices C and E are an exception to that and exhibit a larger longitudinal resistance of Rxx≈1.2R_{xx} \approx1.2 kΩ and not a quantized Hall resistance Ryx≈1.29h/e2R_{yx} \approx1.29h/e^2. The hysteretic reversal of RyxR_{yx} occurs at a coercive field of approximately μ0Hc≈1\mu_0H_c \approx1 T. Table 2 in the method section lists the state of quantization assessed by the dc Hall measurements for each device.

Fitting procedure

We determine the parameters of the transmission model in two successive fitting steps. First, we extract the parameters describing the plasmon splitter from the ratio of the two output signals, S21/S41S_{21}/S_{41}. Since both transmission paths share the same excitation and propagation outside the coupling region, these contributions cancel, up to minor asymmetries in the device. We therefore fit the measured complex ratio using the splitter model of Eq. (2). For devices containing a metallic island, we work in the limit ωτc,s≫1\omega\tau_{\mathrm{c,s}} \gg1 for which the dependence on the charging time τc,s\tau_{\mathrm{c,s}} drops out. We therefore only use the propagation time τp,s\tau_{\mathrm{p,s}} and relaxation time τr,s\tau_{\mathrm{r,s}} as fit parameter. For the reference devices without a metallic island, τr,s\tau_{\mathrm{r,s}} is set to zero and τc,s\tau_{\mathrm{c,s}} is retained as a fit parameter to account for residual capacitive coupling between the two edge channels.

In a second step, the extracted parameters are kept fixed, and the complete transmission model of Eq. (3) is fitted simultaneously to the measured scattering parameters. This fit determines the parameters describing propagation along the bare edge channels, τp,e\tau_{p,e} and τr,e\tau_{r,e}, as well as the gate capacitances CgC_g and the attenuation due to residual bulk conduction. The two output paths are fitted independently where required to account for differences in the excitation and detection gates.

For the fits, we simultaneously consider the in-phase and quadrature components and fit over the complete experimental frequency range using a nonlinear least-squares minimization provided by the scipy package in Python. Parameter uncertainties are obtained from the covariance matrix of the fit. The same procedure is applied to all devices, allowing the extracted propagation and interaction parameters to be compared consistently between devices with and without a metallic island. For Device E, featuring a 2 μ\mum-long island, parameter bounds were imposed (zero to infinity) to prevent the optimization from converging to negative values.

Data, code and materials availability

The datasets generated and analyzed during the current study are available in the Zenodo repository with the DOI 10.5281/zenodo.22676232.

Acknowledgments

Funding: This work has been supported by the Germany Excellence Strategy (Cluster of Excellence Matter and Light for Quantum Computing ML4Q, EXC 2004/1 - 390534769) and the DFG (SFB1238 Control and Dynamics of Quantum Materials, 277146847, projects A04 and B07).

Author contributions: D.H. fabricated the devices, and A.A.T. grew the films. D.H. and T.R. performed the measurements and analyzed the data. F.H. and E.B. assisted with the data analysis and interpretation. D.H. and T.R. prepared the manuscript. E.B. supervised the project. All authors discussed the results and contributed to the writing of the manuscript.

Competing interests: All authors declare no financial or non-financial competing interests.

Supplementary information

The article contains supplementary material available at [insert URL later]

Supplementary Materials for: Zero-field plasmon splitter: Controlled interaction between quantum anomalous Hall edge states

Dominik Hoborka; Torsten Röper; Fabian Hassler; Alexey A. Taskin; Yoichi Ando; Erwann Bocquillon

This PDF file includes:

§ 13. Derivation of the long-range model.

§ 14. Comparison with the short-range interaction model.

§ 15. A fit of the splitting ratio including asymmetric finger gate capacitances.

§ 16. Estimation of the effective interaction length for reference devices.

§ 17. Further information on other devices and reproducibility.

Figure S1. Comparison of the measured splitting ratio with the short-range interaction model.

Comparison of the measured splitting ratio with the short-range interaction model

Figure S1. Comparison of the measured splitting ratio with the short-range interaction model. The green lines correspond to the measured amplitude and phase of the splitting ratio S21/S41S_{21}/S_{41} for Device A with a 6 μ\mum long metallic island. The short-range model in Eq. (15) is simulated with different time constants τr,s\tau_{\mathrm{r,s}} describing the damping while the coupling strength is set to r=0.75r = 0.75 and the propagation time τp,s=400\tau_{\mathrm{p,s}} = 400 ps. Three different dampings are compared to the amplitude and phase of the ratio of the S-parameters for Device A. (a) Small τr,s=1\tau_{\mathrm{r,s}} = 1 ps produces an oscillating splitting ratio proportional to sin⁡ωτp,s\sin\omega\tau_{\mathrm{p,s}}. (b) Using the damping time constant obtained from the long-range model (τr,s=31\tau_{\mathrm{r,s}} = 31 ps) leads to a diverging ratio. (c) The short-range model saturates for a large damping time τr,s=500\tau_{\mathrm{r,s}} = 500 ps to a value determined by the coupling strength rr and the ratio τp,s/τr,s\tau_{\mathrm{p,s}}/\tau_{\mathrm{r,s}}. However, the peak describing the oscillation of S21S_{21} and S41S_{41} at low frequencies is not captured. The experimental data corresponds to the data shown in Figure 5 of the main text.

Figure S2. Comparison of the splitting ratio fits with asymmetric gate capacitance.

Comparison of splitting ratio fits

Figure S2. Comparison of splitting ratio fits. The ratio S21/S41S_{21}/S_{41} (a) and phase difference (b) are fitted and shown as a function of frequency. The black lines correspond to the simplified model, which assumes Sge,2=Sge,4S_{\mathrm{ge},2} = S_{\mathrm{ge},4}. The dashed blue lines correspond to the fit accounting for different gate capacitances Cg,2=(91±2) fFC_{\mathrm{g},2} = (91 \pm2)\,\mathrm{fF} and Cg,4=(10±5) fFC_{\mathrm{g},4} = (10 \pm5)\,\mathrm{fF}. The data are the same as in Figure 5 of the main text.

Figure S3. Comparison of detection lines.

Comparison of detection lines showing the amplitudes of $S_{21}$ and $S_{41}$ for two measuring configurations as a function of frequency

Figure S3. Comparison of detection lines. The amplitudes of the transmission parameters S21S_{21} and S41S_{41} are plotted as a function of frequency, and the two measuring configurations are compared. Both configurations were measured for Device A with a 6 μm6\,\mu\mathrm{m}-long metallic island at μ0H=−2\mu_{0}H=-2 T and T≈20T\approx20 mK. The data of configuration 2 correspond to the data shown in Figure 3B of the main text.

Figure S4. Ratio and total transmission measured for device D

Ratio and total transmission measured for device D

Figure S4. Ratio and total transmission measured for device D. (a) The amplitude of the ratio and phase difference plotted with the fit result (black solid line) using the same equation for S21/S41S_{21}/S_{41} as discussed previously. (b) Amplitude and phase of the total transmission compared to the fit result (dark blue and dark red solid lines). Experiments were performed on device D with a 6 μ\mum metallic island at μ0H=−2\mu_{0}H=-2 T and T≈20T\approx20 mK.

Figure S5. Ratio and total transmission measured for device E

Ratio and total transmission measured for device E

Figure S5. Ratio and total transmission measured for device E. (a) The amplitude of the ratio and phase difference plotted with the fit result (black solid line) using the same equation for S21/S41S_{21}/S_{41} as discussed previously. (b) Amplitude and phase of the total transmission compared to the fit result (dark blue and dark red solid lines). Experiments were performed on device E with a 2 µm metallic island at μ0H=−2\mu_{0}H=-2 T and T≈20T \approx20 mK.

Figure S6. Ratio and total transmission measured for device F

Ratio and total transmission measured for device F

Figure S6. Ratio and total transmission measured for device F. (a) The amplitude of the ratio and phase difference plotted with the fit result (black solid line) using the same equation for S21/S41S_{21}/S_{41} as discussed previously. (b) Amplitude and phase of the total transmission compared to the fit result (dark blue and dark red solid lines). Experiments were performed on device F without a metallic island at μ0H=−2\mu_{0}H=-2 T and T≈20T \approx20 mK.

Table S1. Extracted fit parameters from the transmission model for Device D, E, F.

DeviceD (L=6 μmL = 6\,\mu\mathrm{m})E (L=2 μmL = 2\,\mu\mathrm{m})F (no island)
S21S_{21}S41S_{41}S21S_{21}S41S_{41}S21S_{21}S41S_{41}
τp,s\tau_{\mathrm{p,s}} [ps]1170±301170 \pm30520±15520 \pm1560±4060 \pm40
τr,s\tau_{\mathrm{r,s}} [ps]80±1080 \pm1056.9±0.156.9 \pm0.1−-
CiC_{\mathrm{i}} [fF]−-−-20±10020 \pm100
τp,e\tau_{\mathrm{p,e}} [ps]47±247 \pm264±464 \pm440.7±0.840.7 \pm0.853±353 \pm334±134 \pm126±126 \pm1
τr,e\tau_{\mathrm{r,e}} [ps]35±235 \pm253±353 \pm331±131 \pm160±360 \pm3−-−-
CgC_{\mathrm{g}} [fF]34±234 \pm211±111 \pm125±125 \pm114±114 \pm177±477 \pm426±226 \pm2

Table S1. Extracted fit parameters from the complete transmission model. The table summarizes the fit parameters for Device D (6 μm6\,\mu\mathrm{m} long metallic island), Device E (2 μm2\,\mu\mathrm{m} long metallic island), and the reference device without a metallic island (Device F). For Devices D and E, τp,s\tau_{\mathrm{p,s}} and τr,s\tau_{\mathrm{r,s}} are extracted from fits of S21/S41S_{21}/S_{41} in the high-frequency limit 1/(ωτc,s)→01/(\omega\tau_{\mathrm{c,s}}) \to0. For Device F, τp,s\tau_{\mathrm{p,s}} and CiC_{\mathrm{i}} are extracted from fits of S21/S41S_{21}/S_{41} in the limit τr,s→0\tau_{\mathrm{r,s}} \to0. The bare-edge parameters τp,e\tau_{\mathrm{p,e}}, τr,e\tau_{\mathrm{r,e}}, and the gate capacitance CgC_{\mathrm{g}} are then extracted from fits of the full transmission S21S_{21} and S41S_{41} with the same equation presented in the main text, while fixing the splitter parameters τp,s\tau_{\mathrm{p,s}}, τr,s\tau_{\mathrm{r,s}}, and τc,s\tau_{\mathrm{c,s}} to the values extracted from the corresponding S21/S41S_{21}/S_{41} fits. For device F, the best fit result was obtained by setting the damping time constant τr,s=0\tau_{\mathrm{r,s}} = 0.

Derivation of long-range model

To derive the long-range model, we use a distributed circuit model for two chiral edge channels running beneath a floating metallic island (assuming that the capacitance to ground is negligible). Each edge state follows the chiral relation between current and potential ia=ua/RKi_a = u_a/R_{\mathrm{K}} and the charge conservation ∂xia(x)=iωρa(x)\partial_x i_a(x) = i\omega\rho_a(x), which yields a chiral wave equation driven by the electrostatic potential uau_a:

∂xia(x)=−iks[ia(x)−uaRK],(4)\partial_x i_a(x) = -i k_s \left[ i_a(x) - \frac{u_a}{R_{\mathrm{K}}} \right], \tag*{(4)}

where the dispersion relation below the gate determines the wave vector ksk_s:

ksL=ωτp,s1+iωτr,s,(5)k_s L = \frac{\omega\tau_{\mathrm{p,s}}}{1+i\omega\tau_{\mathrm{r,s}}}, \tag*{(5)}

in analogy to the dispersion relation of a bare edge channel shown in Eq. (1) of the main text. Here τp,s\tau_{\mathrm{p,s}} is the propagation time through the splitter and τr,s\tau_{\mathrm{r,s}} the relaxation time describing the frequency-dependent damping. Since the metallic island is an equipotential conductor at the frequencies of interest (its internal RCRC time is much shorter than the measured plasmon propagation times of several hundred picoseconds), it responds to the total charge qa=∫0Lρa dxq_a = \int_0^L \rho_a\,\mathrm{d}x accumulated in each channel beneath it, rather than to the local charge density. With VV the island potential and CiC_i the island–edge capacitance,

qa=Ci(ua−V),q1+q2=0.(6)q_a = C_i(u_a - V), \qquad q_1 + q_2 = 0. \tag*{(6)}

where the second relation expresses that the island is floating. Subtracting the two relations in Eq. (6) eliminates VV and yields:

u1−u2=2qCi,q≡q1=−q2.(7)u_1-u_2 = \frac{2q}{C_i}, \qquad q \equiv q_1 = -q_2. \tag*{(7)}

We can find a solution for the current by integrating Eq. (4), using that uau_a is independent of xx:

ia(L)=uaRK+[ia(0)−uaRK]e−iksL,(8)i_a(L) = \frac{u_a}{R_{\mathrm{K}}} + \left[i_a(0)-\frac{u_a}{R_{\mathrm{K}}}\right]e^{-ik_sL}, \tag*{(8)}

It shows that the edge current relaxes from its input value ia(0)i_a(0) towards the value imposed by the island potential ua/RKu_a/R_{\mathrm{K}}. Using charge conservation, the accumulated charge yields:

qa=ia(L)−ia(0)iω=e−iksL−1iω[ia(0)−uaRK].(9)q_a = \frac{i_a(L)-i_a(0)}{i\omega} = \frac{e^{-ik_sL}-1}{i\omega}\left[i_a(0)-\frac{u_a}{R_{\mathrm{K}}}\right]. \tag*{(9)}

Now we have to consider the boundary conditions of our experiment. We excite only from port 1, yielding i1(0)=iini_1(0)=i_{\mathrm{in}} and i2(0)=0i_2(0)=0. Using the latter condition, Eq. (9) yields:

q=(e−iksL−1)u2iωRK(10)q=\frac{(e^{-ik_sL}-1)u_2}{i\omega R_K} \tag*{(10)}

Using the definition of u2u_2 in Eq. (7), we find a solution for qq:

q=(e−iksL−1)τc2(iωτc+(e−iksL−1))iin.(11)q=\frac{(e^{-ik_sL}-1)\tau_c}{2\left(i\omega\tau_c+(e^{-ik_sL}-1)\right)}i_{\mathrm{in}}. \tag*{(11)}

with τc=RKCi\tau_c=R_KC_i. We can now define our transmission coefficients. Channel 2 carries no incoming signal, so that i2(L)=−iωqi_2(L)=-i\omega q. Using Eq. (11), we find

Ss,2=i2(L)iin=1−e−iksL2(1+iωτc,s(1−e−iksL))(12)S_{\mathrm{s},2}=\frac{i_2(L)}{i_{\mathrm{in}}}=\frac{1-e^{-ik_sL}}{2\left(1+\frac{i}{\omega\tau_{\mathrm{c,s}}}(1-e^{-ik_sL})\right)} \tag*{(12)}

which is Eq. (2) of the main text. Similarly, we can find the scattering parameter Ss,4S_{\mathrm{s},4}:

Ss,4=i1(L)iin=1+iωqiin=1−Ss,2.(13)S_{\mathrm{s},4}=\frac{i_1(L)}{i_{\mathrm{in}}}=1+\frac{i\omega q}{i_{\mathrm{in}}}=1-S_{\mathrm{s},2}. \tag*{(13)}

The coupling capacitance Cs=Ci/2C_{\mathrm{s}}=C_i/2 thus corresponds to the series capacitance that includes the two island–edge capacitances CiC_i, i.e. the effective capacitance between the two edge channels mediated by the island.

Comparison to short-range model

The long-range interaction model is motivated by the continuous metallic island, which is expected to remain at a nearly uniform electrostatic potential. For completeness, we also compare our data to the short-range interaction model developed for counter-propagating quantum Hall edge channels in Ref. 25. In this model, the interaction is assumed to occur locally along the propagation direction and is described by a distributed mutual capacitance between the two edge channels. The transmission amplitudes into the two output channels are given by

S41=e−iksL−1+r2r2−e−2iksL,S21=r(e−2iksL−1)r2−e−2iksL.(14)\begin{aligned} S_{41} &= e^{-i k_{\mathrm{s}} L}\frac{-1+r^{2}}{r^{2}-e^{-2i k_{\mathrm{s}}L}},\\ S_{21} &= \frac{r\left(e^{-2i k_{\mathrm{s}}L}-1\right)}{r^{2}-e^{-2i k_{\mathrm{s}}L}}. \tag*{(14)} \end{aligned}

with parameter rr describing the coupling strength and ksk_{\mathrm{s}} corresponding to the dispersion given in (1) of the main text. To eliminate the influence of the excitation and detection circuitry, we consider the ratio S21/S41S_{21}/S_{41}:

S21S41=r(e−iksL−eiksL)r2−1(15)\frac{S_{21}}{S_{41}}=\frac{r\left(e^{-i k_{\mathrm{s}}L}-e^{i k_{\mathrm{s}}L}\right)}{r^{2}-1} \tag*{(15)}

Figure S1 shows the data compared to the short-range model simulated with different damping times τr,s\tau_{\mathrm{r,s}} and fixed r=0.75r=0.75 and τp,s=400 ps\tau_{\mathrm{p,s}}=400\ \mathrm{ps}, which reproduces the position and the height of the first peak in the splitting ratio as shown in fig:S1a. Three distinct regimes can be identified within the investigated frequency range: (i) For a small τr,s\tau_{\mathrm{r,s}} (τr,s=1 ps\tau_{\mathrm{r,s}}=1\ \mathrm{ps}), the dispersion reduces to the simple form ksL=iωτp,sk_{\mathrm{s}}L=i\omega\tau_{\mathrm{p,s}} yielding the ratio S21/S41=−i⋅2rr2−1sin⁡ωτp,sS_{21}/S_{41}=-\frac{i\cdot2r}{r^{2}-1}\sin\omega\tau_{\mathrm{p,s}}. The preferred transmission channel therefore depends on the frequency, and in the absence of damping, equal splitting is never achieved, as shown in fig:S1a. (ii) For τr,s=31 ps\tau_{\mathrm{r,s}}=31\ \mathrm{ps}, corresponding to the value obtained from the long-range model fit, the splitting ratio diverges within the investigated frequency range, as shown in fig:S1b. (iii) For large τr,s=500 ps\tau_{\mathrm{r,s}}=500\ \mathrm{ps}, for which ωτr,s≫1\omega\tau_{\mathrm{r,s}}\gg1 and τp,s<τr,s\tau_{\mathrm{p,s}}<\tau_{\mathrm{r,s}}, the ratio increases at low frequencies and approaches the high-frequency limit S21/S41=rr2−1(eτp,s/τr,s−e−τp,s/τr,s)S_{21}/S_{41}=\frac{r}{r^{2}-1}\left(e^{\tau_{\mathrm{p,s}}/\tau_{\mathrm{r,s}}}-e^{-\tau_{\mathrm{p,s}}/\tau_{\mathrm{r,s}}}\right), as shown in fig:S1c. None of these regimes reproduces the experimentally observed frequency dependence, and we find no set of parameters for which the short-range model provides a satisfactory description of the data. Within the physically relevant parameter range explored here, the short-range model does not reproduce the observed response, supporting the long-range description as the dominant coupling mechanism.

Ratio fit with finger gate capacitance

The fit results for the total transmission presented in the main text show that the coupling capacitance of the finger gates differs for S21S_{21} and S41S_{41} (Cg,2≠Cg,4C_{\mathrm{g},2} \ne C_{\mathrm{g},4}). This explains the overestimation of the phase difference between S21S_{21} and S41S_{41} in Figure 5b, which we aim to address in this part. Therefore, we extend the model for fitting the ratio by keeping the transmission parameter from the edge channel to the respective finger gate (Sge,2S_{\mathrm{ge},2} and Sge,4S_{\mathrm{ge},4}, with the subscript signaling the respective port). Then, the ratio S21/S41S_{21}/S_{41} is expressed as:

S21S41=Sge,2Ss,2Sge,4Ss,4\frac{S_{21}}{S_{41}} = \frac{S_{\mathrm{ge},2}S_{\mathrm{s},2}}{S_{\mathrm{ge},4}S_{\mathrm{s},4}}
Sge,i=2iωZgCg,iλi+iωλiZgCg,i(16)S_{\mathrm{ge},i} = \frac{2i\omega Z_{\mathrm{g}}C_{\mathrm{g},i}}{\lambda_i + i\omega\lambda_i Z_{\mathrm{g}}C_{\mathrm{g},i}} \tag*{(16)}

using the transmission Sge,iS_{\mathrm{ge},i} from Ref. 30 with Zg=50 ΩZ_{\mathrm{g}} = 50\,\Omega and λi=1+iωRKCg,i\lambda_i = 1+i\omega R_{\mathrm{K}}C_{\mathrm{g},i}. We fix the previously found parameters for τp,s\tau_{\mathrm{p,s}} and τr,s\tau_{\mathrm{r,s}} and perform the fit with the finger gate coupling capacitances Cg,iC_{\mathrm{g},i}. Figure S2 compares the ratio fits with and without including the finger gate transmission.

Indeed, the phase difference is captured more accurately by including Sge,iS_{\mathrm{ge},i} in the fit, and different finger gate coupling capacitances are found (Cg,2=(91±2)C_{\mathrm{g},2} = (91 \pm2) fF, Cg,4=(10±5)C_{\mathrm{g},4} = (10 \pm5) fF) as listed in the table of the main text. We note that bounds had to be imposed to obtain a physically meaningful value for Cg,2C_{\mathrm{g},2}. Also, in return, the fit overestimates the amplitude of the ratio, which we ascribe to the non-trivial form of the model due to high correlation of the fit parameters Cg,iC_{\mathrm{g},i}. Nevertheless, the position of the peak, and hence the time constant τp,s\tau_{\mathrm{p,s}}, are approximately the same for both fitting methods, justifying the validity of the simplified version for fitting the ratio.

Effective interaction length

For the reference device C, which does not contain a metallic island, the spatial extent of the residual inter-edge coupling is not directly defined by the device geometry. In the main text, we therefore introduce an effective interaction length Leff≈11 μmL_{\mathrm{eff}} \approx11\ \mu\mathrm{m} to estimate the corresponding plasmon velocity. Here, we motivate this estimate from the electrostatic capacitance between the two edge channels CsC_{\mathrm{s}}. We approximate the two edge channels as parallel cylindrical conductors with an effective radius rr and a position-dependent center-to-center separation s(x)s(x). The capacitance per unit length of two parallel wires is given by

c(x)=πε0εrcosh⁡−1[s(x)/(2r)],(17)c(x) = \frac{\pi\varepsilon_0 \varepsilon_r}{\cosh^{-1}\left[s(x)/(2r)\right]}, \tag*{(17)}

where εr\varepsilon_r is the effective dielectric constant of the surrounding medium (εr≈7\varepsilon_r \approx7 for Al2O3\mathrm{Al_2O_3}). The device geometry is divided into three regions: a central section in which the two edges run approximately parallel at a constant separation of s0=2 μms_0 = 2\ \mu\mathrm{m}, and two adjoining sections in which their separation increases away from the coupling region. The spatially varying separation can therefore be written as

s(x)={s0+2d2−2x1,0<x1<d2,s0=2 μm,d2≤x2≤d2+L,s0−2L−2d2+2x3,d2+L<x3<2d2+L,(18)s(x) = \begin{cases} s_0 + \dfrac{2d}{\sqrt{2}} - 2x_1, & 0 < x_1 < \dfrac{d}{\sqrt{2}},\\ s_0 = 2\ \mu\mathrm{m}, & \dfrac{d}{\sqrt{2}} \le x_2 \le\dfrac{d}{\sqrt{2}} + L,\\ s_0 - 2L - \dfrac{2d}{\sqrt{2}} + 2x_3, & \dfrac{d}{\sqrt{2}} + L < x_3 < \dfrac{2d}{\sqrt{2}} + L, \end{cases} \tag*{(18)}

where d=20 μmd = 20\ \mu\mathrm{m} is the bare edge channel length from the coupling region to the finger gate. The total inter-edge capacitance is then obtained by integrating the local capacitance per unit length along the two edge channels for each section,

Ctot=∫c(x) dx≈1.57 fF(19)C_{\mathrm{tot}} = \int c(x)\,\mathrm{d}x \approx1.57\ \mathrm{fF} \tag*{(19)}

The solution of the integral was obtained by evaluating the hyperbolic sine integral using the scipy package in Python. To express this distributed coupling in terms of a single characteristic length, we define an effective interaction length LeffL_{\mathrm{eff}} as the length of two parallel edges with a constant separation s=2 μms = 2\ \mu\mathrm{m} that gives the same total capacitance CtotC_{\mathrm{tot}}. Using an approximate radius of r=500 nmr = 500\ \mathrm{nm} (half of the overlap to the island) and a dielectric constant of εr≈7\varepsilon_r \approx7, we find Leff≈11 μmL_{\mathrm{eff}} \approx11\ \mu\mathrm{m}. We use this effective length in the main text to estimate the plasmon velocity associated with the residual coupling in the reference device.

Other devices and reproducibility

We first address and show that both detection lines are approximately equal. Therefore, we measured the same device in two wiring configurations. The first configuration corresponds to the wiring illustrated in Figure 6. In the second configuration, we swapped the cryostat line and amplifier of Ports 2 and 4 by exchanging the respective cables on the PCB. This enables us to estimate the systematic error due to the use of two different detection lines. As shown in Figure S3, we indeed observe the same phenomenology in both configurations, with only minor discrepancies at low frequencies, and confirm the equivalence of both detection lines.

Next, we present the results of the remaining devices (D, E, F) and discuss the reproducibility of the splitter device. The same analysis procedure was applied as described in the main text. We fit the ratio S21/S41S_{21}/S_{41} to extract the time constants τp,s\tau_{\mathrm{p,s}} and τr,s\tau_{\mathrm{r,s}} for the devices D and E containing a metallic island, respectively τp,s\tau_{\mathrm{p,s}} and CiC_{\mathrm{i}} for the reference device without a metallic island. These parameters are then used to fit the total transmission S21S_{21} and S41S_{41}. The results are shown in Figs. S4, S5, and S6 and the obtained fit parameters are summarized in Table S1. As pointed out in Ref. 4 the quality of the device, for instance, the plasmon dissipation, is strongly dependent on the sample, having a clear impact on our measurements and the respective fit results. Despite this, we observe the same main features that are discussed in the main text.

(i) Oscillations of S21S_{21} and S41S_{41} are observed for the devices containing a metallic island, indicating strong coupling between both edge states. Note that the position of the peak of Device D is at lower frequencies compared to Device A despite having the same metallic island length. The reference device F without a metallic island does not show the oscillatory feature but instead shows high transmission via S41S_{41}, and only at high frequencies S21S_{21} becomes significant. This observation is in accordance with the data of the other reference device C.

(ii) The propagation time inside the coupling region, τp,s\tau_{\mathrm{p,s}}, is large compared to the bare edge propagation time τp,e\tau_{\mathrm{p,e}}. This results in the slowed-down plasmon mode in the coupling region of the splitter. Also, the damping time constants are comparable for each device with a metallic island, leading to equal splitting of the amplitude at high frequencies. Device E is one exception, which shows a broadened peak and a decreasing ratio towards large frequencies. We ascribe this to lower sample quality and larger uncertainties in the correction of the raw data at low frequencies. Performing a manual fit allows us to improve on matching the data and yields a plasmon velocity of approximately 10 km s−110\ \mathrm{km\ s^{-1}}, which fits the value obtained from the linear regression presented in the main text.

(iii) The total transmission S21S_{21} and S41S_{41} are captured by the model in Eq. (3) of the main text. The extracted propagation times τp,e\tau_{\mathrm{p,e}} correspond to plasmon velocities (ve=(310±20)−(430±20) km s−1v_{\mathrm{e}} = (310 \pm20) - (430 \pm20)\ \mathrm{km\ s^{-1}} for device D and ve=(380±20)−(490±10) km s−1v_{\mathrm{e}} = (380 \pm20) - (490 \pm10)\ \mathrm{km\ s^{-1}} for device E) that are comparable to Ref. 4. Device F represents an exception to the observations. We first note that the fit of the total transmission was performed by neglecting the damping time constant τr,e\tau_{\mathrm{r,e}} since the obtained values are too small and carry a large error. Second, the observed plasmon velocity is larger compared to other devices, ranging from ve=(610±20)−(770±30) km s−1v_{\mathrm{e}} = (610 \pm20) - (770 \pm30)\ \mathrm{km\ s^{-1}} indicating a stronger bulk contribution for this device. Furthermore, the coupling capacitance of Device F is poorly constrained by the fit and should be regarded only as evidence for weak residual coupling.

References

  1. [1]C.Z. Chang, J. Zhang, X. Feng, J. Shen, Z. Zhang, M. Guo, K. Li, Y. Ou, P. Wei, L.L. Wang, Z.Q. Ji, Y. Feng, S. Ji, X. Chen, J. Jia, X. Dai, Z. Fang, S.C. Zhang, K. He, Y. Wang, L. Lu, X.C. Ma, Q.K. Xue, Experimental Observation of the Quantum Anomalous Hall Effect in a Magnetic Topological Insulator. Science 340(6129), 167–170 (2013). https://doi.org/10.1126/science.1234414.
  2. [2]C.Z. Chang, W. Zhao, D.Y. Kim, H. Zhang, B.A. Assaf, D. Heiman, S.C. Zhang, C. Liu, M.H.W. Chan, J.S. Moodera, High-precision realization of robust quantum anomalous Hall state in a hard ferromagnetic topological insulator. Nature Materials 14(5), 473–477 (2015). https://doi.org/10.1038/nmat4204.
  3. [3]R. Yu, W. Zhang, H.J. Zhang, S.C. Zhang, X. Dai, Z. Fang, Quantized Anomalous Hall Effect in Magnetic Topological Insulators. Science 329(5987), 61–64 (2010). https://doi.org/10.1126/science.1187485.
  4. [4]T. Röper, H. Thomas, D. Rosenbach, A. Uday, G. Lippertz, A. Denis, P. Morfin, A.A. Taskin, Y. Ando, E. Bocquillon, Propagation, dissipation, and breakdown in quantum anomalous Hall edge states probed by microwave edge plasmons. Physical Review B 110(16), L161403 (2024). https://doi.org/10.1103/PhysRevB.110.L161403.
  5. [5]A.C. Mahoney, J.I. Colless, L. Peeters, S.J. Pauka, E.J. Fox, X. Kou, L. Pan, K.L. Wang, D. Goldhaber-Gordon, D.J. Reilly, Zero-field edge plasmons in a magnetic topological insulator. Nature Communications 8(1), 1836 (2017). https://doi.org/10.1038/s41467-017-01984-5. Number: 1
  6. [6]L.A. Martinez, G. Qiu, P. Deng, P. Zhang, K.G. Ray, L. Tai, M.T. Wei, H. He, K.L. Wang, J.L. DuBois, D.X. Qu. Edge Magnetoplasmon Dispersion and Time-Resolved Plasmon Transport in a Quantum Anomalous Hall Insulator (2023). ArXiv:2308.15665 [cond-mat]arxiv.org/abs/2308.15665
  7. [7]A. Rupp, D. Rosenbach, T. Röper, D. Hoborka, A.A. Taskin, Y. Ando, E. Bocquillon, Dynamics of Current-Induced Switching in the Quantum Anomalous Hall Effect. Physical Review Letters 136(8), 086602 (2026). https://doi.org/10.1103/nlqq-w287.
  8. [8]W. Yuan, L.J. Zhou, K. Yang, Y.F. Zhao, R. Zhang, Z. Yan, D. Zhuo, R. Mei, Y. Wang, H. Yi, M.H.W. Chan, M. Kayyalha, C.X. Liu, C.Z. Chang, Electrical switching of the edge current chirality in quantum anomalous Hall insulators. Nature Materials 23(1), 58–64 (2024). https://doi.org/10.1038/s41563-023-01694-y.
  9. [9]E. Zimmermann, J. Teller, M. Schleenvoigt, G. Behner, P. Schüffelgen, H. Lüth, D. Grützmacher, T. Schäpers, Current-induced magnetization switching in a magnetic topological insulator heterostructure. Physical Review Materials 8(2), 026201 (2024). https://doi.org/10.1103/PhysRevMaterials.8.026201.
  10. [10]G. Viola, D.P. DiVincenzo, Hall Effect Gyrators and Circulators. Physical Review X 4(2), 021019 (2014). https://doi.org/10.1103/PhysRevX.4.021019.
  11. [11]A.C. Mahoney, J.I. Coless, S.J. Pauka, J.M. Hornibrook, J.D. Watson, G.C. Gardner, M.J. Manfra, A.C. Doherty, D.J. Reilly, On-chip microwave quantum hall circulator. Physical Review X 7(1), 011007 (2017)DOI
  12. [12]L.K. Rodenbach, A.R. Panna, S.U. Payagala, I.T. Rosen, M.P. Andersen, P. Zhang, L. Tai, K.L. Wang, D.G. Jarrett, R.E. Elmquist, D.B. Newell, D. Goldhaber-Gordon, A.F. Rigosi, Metrological Assessment of Quantum Anomalous Hall Properties. Physical Review Applied 18(3), 034008 (2022). https://doi.org/10.1103/PhysRevApplied.18.034008.
  13. [13]N.J. Huáng, J.L. Boland, K.M. Fijalkowski, C. Gould, T. Hjesedal, O. Kazakova, S. Kumar, H. Scherer, Quantum anomalous Hall effect for metrology. Applied Physics Letters 126(4), 040501 (2025). https://doi.org/10.1063/5.0233689.
  14. [14]C. Beenakker, P. Baireuther, Y. Herasymenko, I. Adagideli, L. Wang, A. Akhmerov, Deterministic creation and braiding of chiral edge vortices. Physical Review Letters 122(14), 146803 (2019). https://doi.org/10.1103/PhysRevLett.122.146803.
  15. [15]T. Giamarchi, Quantum Physics in One Dimension (Oxford University Press, 2003). https://doi.org/10.1093/acprof:oso/9780198525004.001.0001.
  16. [16]N. Kumada, H. Kamata, T. Fujisawa, Edge magnetoplasmon transport in gated and ungated quantum Hall systems. Physical Review B 84(4), 045314 (2011). https://doi.org/10.1103/PhysRevB.84.045314.
  17. [17]H. Kamata, T. Ota, K. Muraki, T. Fujisawa, Voltage-controlled group velocity of edge magnetoplasmon in the quantum Hall regime. Physical Review B 81(8), 085329 (2010). https://doi.org/10.1103/PhysRevB.81.085329.
  18. [18]E. Bocquillon, V. Freulon, J.M. Berroir, P. Degiovanni, B. Plaçais, A. Cavanna, Y. Jin, G. Fève, Separation of neutral and charge modes in one-dimensional chiral edge channels. Nature Communications 4(1), 1839 (2013). https://doi.org/10.1038/ncomms2788.
  19. [19]I.P. Levkivskyi, E.V. Sukhorukov, Energy relaxation at quantum Hall edge. Physical Review B 85(7), 075309 (2012). https://doi.org/10.1103/PhysRevB.85.075309.
  20. [20]E. Berg, Y. Oreg, E.A. Kim, F. Von Oppen, Fractional Charges on an Integer Quantum Hall Edge. Physical Review Letters 102(23), 236402 (2009). https://doi.org/10.1103/PhysRevLett.102.236402.
  21. [21]E. Bocquillon, V. Freulon, J.M. Berroir, P. Degiovanni, B. Plaçais, A. Cavanna, Y. Jin, G. Fève, Coherence and Indistinguishability of Single Electrons Emitted by Independent Sources. Science 339(6123), 1054–1057 (2013). https://doi.org/10.1126/science.1232572.
  22. [22]P. Glidic, O. Maillet, C. Piquard, A. Aassime, A. Cavanna, Y. Jin, U. Gennser, A. Anthore, F. Pierre, Quasiparticle Andreev scattering in the v = 1/3 fractional quantum Hall regime. Nature Communications 14(1), 514 (2023). https://doi.org/10.1038/s41467-023-36080-4
  23. [23]E. Frigerio, G. Rebora, M. Ruelle, H. Souquet-Basiège, Y. Jin, U. Gennser, A. Cavanna, B. Plaçais, E. Baudin, J.M. Berroir, I. Safi, P. Degiovanni, G. Fève, G.C. Ménard, Gate tunable edge magnetoplasmon resonators. Communications Physics 7(1), 314 (2024). https://doi.org/10.1038/s42005-024-01803-6.
  24. [24]S. Kouamé, E. Frigerio, G. Rebora, S. Purkait, Y. Jin, U. Gennser, A. Cavanna, J.M. Berroir, E. Baudin, P. Degiovanni, F.D. Parmentier, G. Fève, G.C. Ménard. Electrostatic splitting of an Edge Magnetoplasmon Resonator (2026). https://doi.org/10.48550/arXiv.2609.05188. ArXiv:2609.05188 [cond-mat.mes-hall]
  25. [25]M. Hashisaka, H. Kamata, N. Kumada, K. Washio, R. Murata, K. Muraki, T. Fujisawa, Distributed-element circuit model of edge magnetoplasmon transport. Physical Review B 88(23), 235409 (2013). https://doi.org/10.1103/PhysRevB.88.235409.
  26. [26]H. Kamata, N. Kumada, M. Hashisaka, K. Muraki, T. Fujisawa, Fraction-alized wave packets from an artificial Tomonaga–Luttinger liquid. Nature Nanotechnology 9(3), 177–181 (2014). https://doi.org/10.1038/nnano.2013.312.
  27. [27]E. Perfetto, G. Stefan ucci, H. Kamata, T. Fujisawa, Time-resolved charge fractionalization in inhomogeneous Luttinger liquids. Physical Review B 89(20), 201413 (2014). https://doi.org/10.1103/PhysRevB.89.201413.
  28. [28]L.A. Martinez, G. Qiu, P. Deng, P. Zhang, K.G. Ray, L. Tai, M.T. Wei, H. He, K.L. Wang, J.L. DuBois, D.X. Qu, Edge magnetoplasmon dispersion and time-resolved plasmon transport in a quantum anomalous Hall insulator. Physical Review Research 6(1), 013081 (2024). https://doi.org/10.1103/PhysRevResearch.6.013081.
  29. [29]T. Röper, D. Rosenbach, A. Rosch, A.A. Taskin, Y. Ando, E. Bocquillon, Breakdown of the quantum anomalous Hall effect under microwave drives. Physical Review B 112(20), L201301 (2025). https://doi.org/10.1103/3148-3vp4.
  30. [30]A. Gourmelon, H. Kamata, J.M. Berroir, G. Fève, B. Placais, E. Bocquillon, Characterization of helical Luttinger liquids in microwave stepped-impedance edge resonators. Physical Review Research 2(4), 043383 (2020). https://doi.org/10.1103/PhysRevResearch.2.043383.
  31. [31]H.C. Lee, S.R. Eric Yang, Spin-charge separation in quantum Hall edge liquids. Physical Review B 56(24), R15529–R15532 (1997). https://doi.org/10.1103/PhysRevB.56.R15529.
  32. [32]E.V. Sukhorukov, V.V. Cheianov, Resonant Dephasing in the Electronic Mach-Zehnder Interferometer. Physical Review Letters 99(15), 156801 (2007). https://doi.org/10.1103/PhysRevLett.99.156801.
  33. [33]N. Kumada, N.H. Tu, K.i. Sasaki, T. Ota, M. Hashisaka, S. Sasaki, K. Onomitsu, K. Muraki, Suppression of gate screening on edge magnetoplasmons by highly resistive ZnO gate. Physical Review B 101(20), 205205 (2020). https://doi.org/10.1103/PhysRevB.101.205205.
  34. [34]K. Flensberg, Coulomb Drag of Luttinger Liquids and Quantum Hall Edges. Physical Review Letters 81(1), 184–187 (1998). https://doi.org/10.1103/PhysRevLett.81.184.
  35. [35]Y. Fu, Y. Huang, Q.L. He, Non-reciprocal Coulomb drag between Chern insulators. Nature Communications 16(1), 3058 (2025). https://doi.org/10.1038/s41467-025-58401-5.

Paper details

vlnc:W_6mjemnvq
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

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