betaFeedback

WorksPhysics and Astronomy

Exciton design via topological spin textures

Abstract

Excitons carry energy and information through a semiconductor without carrying charge. This makes them attractive for optoelectronics, but also hard to guide. Magnetic skyrmions, on the other hand, are moved by small electrical currents. Here we develop a theory of excitons interacting with a skyrmion in which the electron and the hole are treated as distinct constituents, each with its own spin–orbit coupling. We find that the topology of the skyrmion is transferred to the exciton only when the exchange coupling exceeds the relevant spin–orbit splitting. The exciton then experiences a full flux quantum, with opposite signs in the two valleys. The quantum metric of the spin texture creates a ring-shaped barrier that traps excitons, whereas the topological charge fixes the angular momentum of the trapped states and hence their optical selection rules. Because the trap moves with the skyrmion, excitons can be transported by current pulses, and small skyrmions accommodate a single exciton, which opens the door for potential hybrid applications in future information technology intertwining topological magnetic bits with optical excitations.

Introduction

Magnetic skyrmions are localized entities consisting of topologically non-trivial spin swirling textures in which the magnetization whirls around a reversed core [1, 2, 3, 4]. They are stabilized by the Dzyaloshinskii–Moriya interaction[5, 6] and have been observed on length scales from micrometres down to a few nanometres [7, 8, 9, 10, 11]. Individual skyrmions can be written and deleted with a spin-polarized tip[12], and they are set in motion by electrical currents of modest density [13, 14]. This mobility is at the origin of the racetrack concept [15], in which information is encoded in magnetic objects that are shifted along a track by current pulses.

Skyrmions do not only carry information themselves; they also act on particles passing through them. An electron whose spin follows the texture adiabatically acquires a Berry phase, which is equivalent to an emergent magnetic field and gives rise to the topological Hall effect [16, 17, 18, 19]. Further magneto-transport effects, associated to chirality, non-collinearity and spin-mixing can emerge such as the chiral and non-collinear Hall effects [20, 21] as well as a whole family of spin-mixing magnetoresistances[22, 23, 24]. Even when the electron spin cannot follow the texture, the scattering remains asymmetric[25]. Similar effects have been predicted for magnons [26, 27, 28, 29]. In all these cases the particle is elementary, and its coupling to the texture is characterized by a single exchange constant. Excitons, the bound electron–hole pairs that dominate the optical response of two-dimensional semiconductors [30, 31, 32], are a natural next candidate, but they raise questions that do not arise for electrons or magnons. An exciton transports energy without net charge [33], which is precisely why it is so difficult to control: neutral excitons are insensitive to electric fields, and guiding them has so far required spatially indirect excitons in electrostatic landscapes [34, 35], thermal gradients [36] or surface acoustic waves [37]. A skyrmion that could hold an exciton and carry it along a racetrack would provide a very different handle. Whether this is possible depends, however, on how an exciton perceives a spin texture in the first place, and here the composite nature of the exciton becomes essential. In transition-metal dichalcogenides, for instance, the spin of the hole is locked to the valley by an Ising spin–orbit splitting of a few hundred meV, while the spin of the electron is split by only a few meV. The two constituents of one and the same exciton therefore respond very differently to the texture, and it is not obvious what the exciton as a whole experiences.

In this work, we address this question by building the theory of an exciton in a skyrmion from its constituents up. We find that the emergent flux felt by the exciton is not guaranteed by topology alone. It appears only when the exchange coupling exceeds the relevant spin–orbit splitting, and it then takes opposite signs in the two valleys. Once the flux is present, the skyrmion acts as a trap for excitons. The two geometric ingredients of the problem play distinct roles in this trap: the quantum metric of the texture produces the confining barrier, while the Berry curvature, that is the topology, determines the angular momentum of the trapped states and therefore which light can address them. Since the trap is the skyrmion itself, it can be moved by a current, and we show that the exciton follows with negligible loss. The theory depends on only a few parameters, namely the exchange coupling, the spin–orbit splitting, the size of the skyrmion and the temperature. This allows us to summarize the different regimes in a single map (Figure 1) and to discuss which of the known two-dimensional excitonic materials are suited to realize them.

Figure 1. Excitons in a skyrmion: a map. a, Illustration of the proposed scheme in a heterostructure of a two-dimensional semiconductor (TMD), an insulating skyrmion host and a heavy-metal layer that drives the skyrmions. (1) Circularly polarized light creates a bright exciton (electron e−e^{-} and hole h+h^{+}). (2) In the skyrmion, bright and dark excitons are mixed and the exciton is held in the core. (3) A current pulse jj moves the skyrmion together with the exciton. (4) The exciton is released and recombines radiatively. The rendering is schematic and not to scale. b, Regimes as a function of the exchange coupling JJ and the skyrmion radius RR. Shaded areas: exciton trapping with binding energy above kBTk_{\mathrm{B}}T and above 10 kBT10\,k_{\mathrm{B}}T (transport window) at 4.2 K, for an exchange well of depth 2J2J and wall width R/5R/5. Vertical lines: onset of adiabatic following for wall widths of 1.5 and 5 nm. Dash-dotted lines: radii beyond which the trapped resonance lives longer than 10 and 100 ps (walls of 1.5 nm; profile dependent). Bars on top: indicative exchange scales for proximity coupling (orange, 1–5 meV) and for magnetic semiconductors with magnetic carriers (purple, above 50 meV). c, Flux-switch threshold J∗J^{*} of two-dimensional excitonic materials: ∣ΔBD∣/2|\Delta_{BD}|/2 from measured bright–dark splittings for intralayer excitons (X) [38], and the conduction-band splitting Δc\Delta_{\mathrm{c}} of the electron layer for interlayer excitons (IX) [39]; weak spin–orbit semiconductors, magnetic semiconductors and antiferromagnets are indicated qualitatively. See Supplementary Tables S2 and S3.

Results

Emergent flux of a composite exciton

We consider an exciton in a two-dimensional semiconductor placed on, or next to, a magnetic layer hosting a skyrmion with magnetization direction n^(r)\hat{\mathbf{n}}(\mathbf{r}). Near a valley τ=±1\tau=\pm1, the electron (i=ci=c) and the hole (i=vi=v) are described by

hi=p22mi+bi⋅σ,bi=Jin^(r)+ηiτΔiz^,(1)h_i=\frac{\mathbf{p}^2}{2m_i}+\mathbf{b}_i\cdot\boldsymbol{\sigma},\qquad\mathbf{b}_i=J_i\hat{\mathbf{n}}(\mathbf{r})+\eta_i\tau\Delta_i\hat{\mathbf{z}}, \tag*{(1)}

where JiJ_i is the exchange coupling to the texture, 2Δi2\Delta_i the Ising spin–orbit splitting of the corresponding band and ηi=±1\eta_i=\pm1 its ordering. In transition-metal dichalcogenides, Δv\Delta_v ranges from several tens to a few hundred meV, while Δc\Delta_c lies between about 1.5 and 20 meV (Figure 2a). The spin of the hole is therefore locked to the valley, whereas the spin of the electron can be reoriented by a modest exchange field. Bright excitons are formed by an electron and a hole of the same spin and couple to circularly polarized light, σ+\sigma^+ in KK and σ−\sigma^- in K′K'; dark excitons involve an electron of the opposite spin. In semiconductors with weak spin–orbit coupling, both splittings are small.

Figure 2. Bands, excitons and the flux switch. a, Band edges of a transition-metal dichalcogenide at the KK and K′K' valleys. The valence band has a large Ising splitting 2Δv2\Delta_{\mathrm{v}} and the conduction band a small one, 2Δc2\Delta_{\mathrm{c}}, with the spin ordering reversed between the valleys. Solid arrows: bright, spin-conserving transitions addressed by σ±\sigma^{\pm} light; dashed arrows: spin-forbidden dark transitions. The band ordering shown is that of the Mo-based compounds; the bright–dark splitting of the excitons, ΔBD\Delta_{\mathrm{BD}}, additionally contains the electron–hole exchange, which in MoS2_2 places the dark exciton below the bright one [38]. b, Path of the direction of the effective field bi\mathbf{b}_{i} from outside the skyrmion to its core. For Ji<ΔiJ_{i}<\Delta_{i} the field only tilts in the wall and returns, and the emergent flux vanishes; for Ji>ΔiJ_{i}>\Delta_{i} it is reversed in the core, sweeps the full sphere and carries one flux quantum. c, Local energies of bright and dark excitons in the two valleys across a skyrmion (radius 5 nm, wall 1 nm), obtained from Eq. (5) (Methods) with Jc=1J_{\mathrm{c}}=1 meV, Jv=0.5J_{\mathrm{v}}=0.5 meV and ΔBD=5\Delta_{\mathrm{BD}}=5 meV. The valley ordering is reversed in the core, and bright and dark states are mixed in the wall.

The important observation is that the spin of a carrier does not follow the magnetization n^\hat{\mathbf{n}} itself, but the total effective field bi\mathbf{b}_i, which also contains the Ising term. Far from the skyrmion, both n^\hat{\mathbf{n}} and bi\mathbf{b}_i point along +z^+\hat{\mathbf{z}}. In the core, n^\hat{\mathbf{n}} is reversed, but bi\mathbf{b}_i is reversed only if Ji>ΔiJ_i>\Delta_i. Otherwise the Ising field dominates and bi\mathbf{b}_i keeps pointing up, so that its direction merely tilts in the skyrmion wall and returns. The emergent flux experienced by the carrier is given by the solid angle swept by bi\mathbf{b}_i (Figure 2b), which leads to

ΦiΦ0=siν2[cos⁡Θi(0)−cos⁡Θi(∞)]={±ν,Ji>Δi,0,Ji<Δi,(2)\frac{\Phi_i}{\Phi_0}=\frac{s_i\nu}{2}\left[\cos\Theta_i(0)-\cos\Theta_i(\infty)\right]= \begin{cases} \pm\nu, & J_i>\Delta_i,\\ 0, & J_i<\Delta_i, \end{cases} \tag*{(2)}

where ν\nu is the topological charge of the skyrmion and Θi\Theta_i the polar angle of bi\mathbf{b}_i. As expected from topology, the flux is quantized. It is, however, either fully present or completely absent, and which

of the two applies is decided by the competition between exchange and spin–orbit coupling. For an isolated carrier, the relevant quantity is the band splitting. In MoS2_2, which has the smallest conduction-band splitting among the dichalcogenides, an exchange of a few meV is sufficient to switch on the flux for the electron, whereas the hole practically never winds with the texture. For an intralayer exciton, the relevant scale turns out to be the splitting between bright and dark excitons, which also contains the electron–hole exchange interaction, as we discuss in the section on materials.

We note that the sharp step in Eq. (2) is a property of the adiabatic, projected description. Close to Ji≈ΔiJ_i \approx\Delta_i, the gap between the two spin branches closes in the skyrmion core, the carrier can no longer follow adiabatically, and the true crossover is smooth. The exact solution presented below shows that the emergent-flux picture becomes quantitative once JiJ_i exceeds both Δi\Delta_i and the kinetic energy scale ℏ2/(2miw2)\hbar^2/(2m_iw^2) associated with a wall of width ww (Supplementary Note 2).

To describe the exciton as a whole, we project each constituent on its adiabatic branch, introduce centre-of-mass and relative coordinates R\mathbf{R} and r\mathbf{r}, and expand the emergent vector potentials to first order in r\mathbf{r} (Methods and Supplementary Note 1). This leads to the two-body Hamiltonian

H=[P−ℏaX−ℏΩ−z^×r]22M+[p−ℏ2Ω+z^×r]22μ+VC(r)+∑i(εi+Wi),(3)H = \frac{\left[\mathbf{P} - \hbar\mathbf{a}_{X} - \hbar\Omega_{-}\hat{\mathbf{z}} \times\mathbf{r}\right]^2}{2M} + \frac{\left[\mathbf{p} - \frac{\hbar}{2}\Omega_{+}\hat{\mathbf{z}} \times\mathbf{r}\right]^2}{2\mu} + V_{\mathrm{C}}(\mathbf{r}) + \sum_i (\varepsilon_i + W_i), \tag*{(3)}

in which aX=ae+ah\mathbf{a}_{X} = \mathbf{a}_{e} + \mathbf{a}_{h} is the emergent vector potential acting on the exciton, Ω±\Omega_{\pm} are mass-weighted combinations of the emergent fields of the electron and the hole, εi\varepsilon_i are the exchange energies and WiW_i the geometric scalar potentials. For a real, uniform magnetic field, Eq. (3) reduces to the Hamiltonian of Gor'kov and Dzyaloshinskii [40], and aX\mathbf{a}_{X} vanishes because the exciton is neutral. In a skyrmion this is no longer the case. The exciton carries no electric charge, but it does carry an emergent charge, and this charge depends on spin, valley and the ratio Ji/ΔiJ_i/\Delta_i of each constituent.

The same physics can be formulated in the basis of bright and dark excitons, which differ by the spin of the electron. The in-plane component of the texture, Jcsin⁡θe−iτνϕJ_{\mathrm{c}}\sin\theta e^{-i\tau\nu\phi}, couples the two states, and the phase winding of this coupling imposes a selection rule: a bright exciton with angular momentum ℓ\ell can only be scattered into a dark exciton with angular momentum ℓ+τν\ell+\tau\nu. In other words, each scattering event transfers one unit of angular momentum per unit of topological charge. The diagonal part of the coupling shifts the bright and dark excitons in opposite directions in the two valleys and reverses their valley ordering in the skyrmion core, while the off-diagonal part mixes them in the wall (Figure 2c). In the weak-coupling regime, a single skyrmion therefore appears as an island of reversed valley ordering surrounded by a ring of brightened dark excitons (Supplementary Fig. S4). We have solved this two-channel scattering problem exactly and compared it with the gauge-field description of Eq. (3) (Figure 3 and Supplementary Note 5). Below the switch, the scattering is weak but chiral; its asymmetry reverses for an antiskyrmion and differs between the two valleys (Supplementary Fig. S3). Close to the switch, the two descriptions disagree, and a substantial fraction of bright excitons is converted into dark ones. Well above the switch, this conversion drops below 10−310^{-3}, and the skew of the scattering obtained from the exact solution (24–27%) agrees with the gauge-field result (23–25%). In K′K', the bright exciton lies below the dark one once the flux is switched on, so that the dark channel is closed at thermal energies and no conversion occurs. The skew is smaller there and the cross-section about twice as large as in KK.

Figure 3. Spin–orbit coupling switches the emergent flux. Skyrmion radius 10 nm, wall parameter 4 nm, Jv=0.5J_{\mathrm{v}}=0.5 meV, bright–dark splitting ΔBD=5\Delta_{\mathrm{BD}}=5 meV, kinetic energy kBTk_{\mathrm{B}}T at 4.2 K; KK valley in blue, K′K' in orange. The shaded region, Jc>J∗=ΔBD/2J_{\mathrm{c}}>J^{*}=\Delta_{\mathrm{BD}}/2, marks the dark core. a, Net emergent flux of the branch connected to the bright exciton versus Jc/ΔBDJ_{\mathrm{c}}/\Delta_{\mathrm{BD}}. b, Scattering skew and c, elastic cross-section from the exact two-channel solution (symbols) and from the gauge-field theory (dashed lines); the narrow gauge-field resonances in K′K' are clipped at 150 nm. d, Bright-to-dark leakage of the exact solution relative to elastic scattering, which signals the breakdown of adiabatic following. In K′K', the dark channel is closed for Jc/ΔBD>0.54J_{\mathrm{c}}/\Delta_{\mathrm{BD}}>0.54 (dotted line), and there is no leakage.

This strong-coupling regime, in which every constituent winds with the texture, corresponds to the situation considered in an earlier study of excitons scattered by skyrmions in a CrI3_3 monolayer, where the exciton was treated as a structureless spinor [41]. The composite description contains this limit as a special case and adds the features that a structureless description cannot capture, namely the flux switch, the valley dependence and the crossover to weak coupling. We also note that for wave vectors accessible to optical excitation, kR≪1kR \ll1, a gauge-consistent solution of the structureless model gives scattering asymmetries below 10−310^{-3} (Supplementary Note 4 and Supplementary Fig. S2). Sizeable chirality requires kR≳1kR \gtrsim1, as in the calculations of Figure 3.

Trapping by quantum geometry

We now turn to the fate of an exciton once the flux is switched on. Equation (3) contains two geometric contributions of different nature. The emergent vector potential aX\mathbf{a}_{X} is the Berry connection of the texture, and its curl, the Berry curvature, integrates to the topological charge. The scalar potentials Wi=ℏ28mi(∇b^i)2W_i = \frac{\hbar^2}{8m_i}(\nabla\hat{\mathbf{b}}_i)^2, on the other hand, are proportional to the quantum metric of the texture. They are largest where the spins rotate rapidly, that is, in the skyrmion wall. Both contributions enter the Hamiltonian as non-negative terms. On their own, they cannot bind an exciton, but they can hold it for a finite time behind a barrier.

This is indeed what we find (Figure 4). The quantum metric produces a ring-shaped barrier in the wall, and an exciton located in the core is trapped behind it. For a skyrmion with a radius of 20 nm, the scattering phase shifts reveal a resonance at 0.43 meV with a lifetime of about 9 ps. To clarify the role of topology, we performed two control calculations (Figure 4b). In the first, we removed the emergent vector potential and kept the metric. The resonance remains at the same energy, but it now appears in the channel ℓ=0\ell= 0 instead of ℓ=−1\ell= -1. In the second, we replaced the skyrmion by a skyrmionium, which has walls of the same type but no net topological charge. The skyrmionium also traps an exciton, again in ℓ=0\ell= 0. We conclude that the barrier originates from the quantum metric, while the topology shifts the angular momentum under which the trapped state appears from ℓ\ell to ℓ−ν\ell- \nu. This shift is exact and fixed by the topological charge. The energy and the lifetime of the resonance, in contrast, depend on the details of the wall. They vary between 3 and 28 ps when the wall width changes from 4 to 1 nm and reach about 50 ps for a linear wall profile (Figure 4c). Within a given family of profiles, the lifetime increases with the topological charge and with the size of the skyrmion, from picoseconds to hundreds of picoseconds (Supplementary Note 7 and Supplementary Fig. S7). These are useful trends, but they are not protected by topology.

Figure 4. A trap made of quantum geometry (radius 20 nm, wall 2 nm, ν=1\nu= 1). a, The quantum-metric potential WW forms a ring-shaped barrier in the wall; the effective potentials UℓU_\ell include the centrifugal term shifted by the emergent flux. b, Wigner time delay: the resonance of the skyrmion in ℓ=−1\ell= -1 survives when the emergent vector potential is removed, but reappears in ℓ=0\ell= 0; a skyrmionium without net flux traps in ℓ=0\ell= 0 as well. c, Dependence of the resonance on the wall width and shape. d, Optical weight seen by a 30 nm near-field spot carrying orbital angular momentum mm; the trapped ss-mode (dotted line) appears as a feature only for m=−νm = -\nu.

If the exchange energies εi\varepsilon_i add an attractive well, the resonances become true bound states and the same rule applies. The ground state carries the angular momentum ℓ=−ν\ell= -\nu, and the excited pp-like states, which would be degenerate in an ordinary circular trap, are split by the emergent field (Supplementary Fig. S7).

The angular-momentum label has a direct optical consequence. Light couples to the bright component of the trapped exciton, and light carrying an orbital angular momentum mm couples only to states with ℓ=m\ell= m. We have therefore computed (Supplementary Note 7) the optical weight seen by a near-field spot of 30 nm diameter carrying orbital angular momentum mm (Figure 4d). The ss-like trapped mode appears as a feature at 0.43 meV only for m=−νm = -\nu. Light without orbital angular momentum addresses a different trapped mode near 1 meV, and light of the opposite handedness addresses neither. Reversing the topological charge of the skyrmion interchanges the roles of m=+1m = +1 and m=−1m = -1. The corresponding signals are weak, at most a few per cent of the optical weight at the band edge, because the trapped modes are located in the core, where the exciton is spin-dark, and because they lie at momenta outside the light cone. They nevertheless constitute a direct optical fingerprint of the topological charge.

A trap that moves

An exciton bound in a quantum dot stays where the dot is. A skyrmion, in contrast, can be displaced along a track by a current, and the natural question is whether a bound exciton follows it.

For transport, the metastable resonances of the purely geometric trap are not sufficient, and we consider the bound states that form when the exchange energy of the constituents adds an attractive well of depth V0V_0 in the skyrmion core, as discussed above. We treat V0V_0, which is of the order of the exchange coupling, as a parameter and use 2 and 5 meV. As a simple reference, we first omit the Berry and geometric terms and keep only the exchange well; we then compare with the full model, which includes both (solid and dashed lines in Figure 5).

Figure 5. The skyrmion as a mobile carrier of excitons. Interlayer excitons in MoS2_2/WSe2_2 (M=0.83m0M = 0.83m_0) bound in the exchange well of a skyrmion with wall width 1.5 nm. The well depth is V0=2V_0 = 2 meV unless stated otherwise. Solid lines: reference model with the exchange well only; dashed lines: full model including the Berry and geometric terms (Methods). a, Upper bound for the probability of losing the exciton when the skyrmion velocity changes instantaneously by Δv\Delta v, for skyrmion radii of 10, 20 and 40 nm. The grey band marks typical velocities of current-driven skyrmions. b, Distance vτescv\tau_{\mathrm{esc}} over which the exciton is carried at v=100v = 100 m/s before it escapes thermally, as a function of temperature, for a skyrmion radius of 20 nm and well depths of 2 and 5 meV, which give binding energies of 1.6 and 4.5 meV. The coloured bands show the distances vτXv\tau_X allowed by the lifetimes τX\tau_X of dark intralayer (0.1–1 ns) and interlayer (10–100 ns) excitons. The achievable range is the smaller of the two. c, Dipolar repulsion between two interlayer excitons occupying the ground state of the same skyrmion (layer separation 0.65 nm, dielectric constant 4.5), compared with the binding energy of a single exciton. The dotted line marks kBTk_BT at 4.2 K. In the shaded region, the full model gives a binding energy above kBTk_BT and a repulsion above the binding energy, so that the skyrmion holds exactly one exciton.

A simple and exact argument answers the first part of this question. Since the problem is invariant under Galilean transformations, a skyrmion moving at constant velocity carries its bound exciton without any loss. Losses can only arise while the skyrmion accelerates. The most unfavourable case is an instantaneous change of velocity Δv\Delta v, for which the probability of retaining the exciton is ∣⟨ψ0∣eiMΔv⋅x/ℏ∣ψ0⟩∣2|\langle\psi_0|e^{iM\Delta\mathbf{v}\cdot\mathbf{x}/\hbar}|\psi_0\rangle|^2. We evaluated this bound for interlayer excitons, which turn out to be the most suitable candidates, as discussed below (Supplementary Note 8). Even an instantaneous jump to 100 m/s removes at most about 10−410^{-4} of the exciton (Figure 5a), and losses become significant only above about 10 km/s, far beyond any racetrack velocity. Similarly, tunnelling out of the trap during acceleration would require accelerations of the order of 101710^{17} m/s2^2, whereas a skyrmion that reaches 100 m/s [42] within the 0.1–10 ns rise time of a current pulse experiences only a≈Δv/τ≈1010a \approx\Delta v/\tau\approx10^{10}–101210^{12} m/s2^2. The same holds in the full model. With the Berry potential, the moving texture also produces an emergent electric field, but in the frame moving with the skyrmion this only adds the static term −ℏv⋅a-\hbar\mathbf{v}\cdot\mathbf{a}, of a few μ\mueV at 100 m/s, and the losses remain of the same order (Figure 5a). For a rigid skyrmion in a clean system, the motion itself is therefore not a limitation. Disorder, deformations of the skyrmion during its motion and exciton–phonon scattering are not included in this analysis and will reduce the transport distance.

What does limit transport is time. The distance over which an exciton can be carried is the skyrmion velocity multiplied by the shorter of two times, the lifetime of the exciton and the time it needs to escape thermally from the trap. Figure 5b compares the two. The curves show the distance covered before thermal escape, and the bands the distance covered within the exciton lifetime; whichever is shorter sets the range. Bright intralayer excitons live only for picoseconds, and even dark intralayer excitons for at most about a nanosecond. Interlayer excitons in heterobilayers of transition-metal dichalcogenides, in which the electron and the hole reside in different layers, live for tens of nanoseconds or longer [43]. They are attractive for a second reason. If the layer with the smaller Ising splitting faces the magnet, only the electron couples to the texture and carries the full emergent charge, and the flux switch then requires no more than a few meV of exchange. For a binding energy of 1.6 meV, thermal escape limits the range to a few micrometres at 2 K and to well below a micrometre above about 3 K. For a binding energy of 4.5 meV, thermal escape becomes slower than the lifetime of interlayer excitons below roughly 5 K, and the exciton can be carried over several micrometres at 100 m/s, limited by its lifetime. At 4.2 K, the escape-limited distance is then about 50 μ\mum. Including the Berry and geometric terms reduces the binding energy at this radius by 10% for V0=2V_0 = 2 meV and by 2% for V0=5V_0 = 5 meV, which shortens the escape time by less than a factor of two, well within the uncertainty of the prefactor. These estimates inherit the uncertainty of the escape prefactor, which we estimate to be about one order of magnitude, but the resulting design rule is robust: the binding energy should exceed about ten times kBTk_{\mathrm{B}}T.

Interlayer excitons have one further property that matters here. They carry a permanent outof-plane electric dipole and therefore repel each other. We evaluated this repulsion for two excitons occupying the ground state of the same skyrmion. For skyrmions with radii below about 25–27 nm, it exceeds the binding energy of a single exciton (Figure 5c), and such a skyrmion accepts one exciton but not a second one. Here the full model adds an important lower bound. In small skyrmions, the geometric barrier in the wall pushes the bound state up, and for a well of 2 meV the binding energy falls below kBTk_BT at 4.2 K for radii below about 12 nm. A single exciton is therefore held stably in skyrmions with radii between about 12 and 27 nm. Deeper wells extend this window to smaller radii; for V0=5V_0 = 5 meV the binding energy at a radius of 10 nm is still 2.8 meV. The precise crossover radius depends on the dielectric environment and on the layer separation, and it should be regarded as an estimate to be refined for a specific heterostructure. Within these limits, a small skyrmion becomes a mobile carrier of single excitons, which can be loaded optically, moved by a current pulse and emptied at a chosen position. In combination with skyrmion logic operations such as merging and duplication [44], this blockade could provide the nonlinearity needed to process excitonic information.

Candidate materials

The map in Figure 1 summarizes the conditions identified so far. First, the flux has to be switched on, which requires an exchange coupling above the relevant spin–orbit splitting. Second, the exciton has to follow the texture adiabatically, which favours wide skyrmion walls. Third, trapping and transport require a binding energy above kBTk_BT, and preferably above 10kBT10 k_BT, which favours large skyrmions. Two-dimensional excitonic materials fall into a few families with respect to these criteria (Figure 1c, Supplementary Note 9 and Supplementary Tables S2 and S3).

The transition-metal dichalcogenides are the best characterized excitonic materials, and in their case a subtlety has to be taken into account. For an intralayer exciton, the texture mixes bright and dark states, and the flux switch is therefore controlled by the bright–dark splitting ΔBD\Delta_{BD}, with a threshold J∗=∣ΔBD∣/2J^{*} = |\Delta_{BD}|/2, rather than by the bare splitting of the conduction band. The two quantities differ because ΔBD\Delta_{BD} also contains the short-range electron–hole exchange and the different binding energies of bright and dark excitons. Magneto-optical measurements on hBN-encapsulated monolayers give ΔBD=+14\Delta_{BD} = +14 meV for MoS2_2, with the dark exciton below the bright one, and ΔBD=−1.4\Delta_{BD} = -1.4 meV for MoSe2_2, with the dark exciton slightly above, even though the bare conduction-band splittings are 3 and 22 meV, respectively [38, 39]. For intralayer excitons, MoSe2_2 is therefore the most favourable dichalcogenide, with a threshold below 1 meV. MoS2_2 requires about 7 meV, and the W-based compounds, in which the dark excitons lie 40–55 meV below the bright ones, 20 meV or more [38].

Our GW–Bethe–Salpeter equation (BSE) calculations for free-standing MoS2_2 support this picture. They give a bright A exciton at 1.91 eV, bound by 0.82 eV with respect to the quasiparticle gap, and place the spin-forbidden dark A exciton 23 meV below it (Supplementary Note 3 and Supplementary Fig. S1). This is the same ordering as observed in hBN-encapsulated samples, where the additional screening reduces the splitting to 14 meV [38]. The same experiments also provide a useful calibration of the coupling between bright and dark states that underlies our results. An in-plane magnetic field brightens the dark exciton of MoS2_2, which becomes visible above about 12 T [38]. With an in-plane electron gg-factor of 2, an exchange coupling of 1 meV acting on the electron is equivalent to an in-plane field of about 17 T. A skyrmion wall with this exchange therefore acts on the exciton like a local in-plane field in the range where the brightened dark exciton has already been observed.

Heterobilayers offer an additional degree of freedom, because the electron and the hole of an interlayer exciton reside in different layers and only the layer in contact with the magnet matters. If MoS2_2 faces the magnet and WSe2_2 hosts the hole, only the electron couples to the texture. Since the electron–hole exchange is weak for spatially separated carriers, the threshold is then close to the bare conduction-band value of MoS2_2, about 1.5 meV. In MoSe2_2/WSe2_2, the most thoroughly characterized heterobilayer, the electron resides in MoSe2_2 and the threshold increases to about 10 meV, so that the most studied system is not the most favourable one in this context. Interlayer excitons are larger than intralayer ones, live for nanoseconds or longer [45, 46] and diffuse over micrometres [47, 48], which makes them well suited as cargo. Two aspects require care. In aligned heterobilayers, moiré potentials of tens of meV trap interlayer excitons by themselves [49]. At large twist angles, however, the moiré period of a few nanometres is averaged out over a skyrmion of tens of nanometres. In addition, the optical selection rules of interlayer excitons depend on the stacking registry, and spin-triplet configurations can be optically active. In that case, the spin-dark core of the intralayer case would become a core emitting light of the opposite circular polarization, which would be an equally useful signature.

Other families are located in different parts of the map. Semiconductors with weak spin–orbit coupling, such as black phosphorus or InSe, have no Ising splitting to overcome, so that any exchange coupling winds the carrier spins, but they lack the valley selectivity of the dichalcogenides. In magnetic semiconductors such as CrSBr, or CrI3_3, in which skyrmions have been proposed [50, 51], the carriers are themselves magnetic, and these materials sit naturally in the strong-coupling corner considered in ref. 41. Antiferromagnetic excitonic materials such as NiPS3_3 and MnPS3_3 form a special case, since the fluxes from the two sublattices cancel and only valley- or spin-resolved effects are expected to survive.

A further question is whether skyrmion hosts and excitonic layers can be combined without destroying the skyrmions. As a test case, we considered Pd/Fe/Ir(111), in which individual skyrmions can be written and deleted on demand [9, 12], covered by two hBN layers and a MoS2_2 monolayer. Our first-principles calculations, performed using the fully relativistic Korringa–Kohn–Rostoker (KKR) Green-function method [52], show that the isotropic exchange and the Dzyaloshinskii–Moriya interaction that stabilize the skyrmions are essentially unaffected by the van der Waals overlayer (Supplementary Note 6 and Supplementary Fig. S5). Pd/Fe/Ir(111) also illustrates the small-size limit of the map, with skyrmion diameters of only 2–4 nm. Even in this limit, the valley-split island in the core and the ring of brightened dark excitons in the wall retain most of their contrast after averaging over the extent of the exciton (Supplementary Fig. S6). For the excitonic coupling itself, however, an insulating skyrmion host in direct contact with the excitonic layer is preferable to a metallic one, which requires a spacer to protect the exciton. These considerations suggest a concrete experiment. A MoSe2\mathrm{MoSe}_2 monolayer, or a MoS2/WSe2\mathrm{MoS}_2/\mathrm{WSe}_2 heterobilayer for transport, is placed in direct contact with an insulating magnet that hosts skyrmions, for instance the chiral magnet Cu2OSeO3\mathrm{Cu}_2\mathrm{OSeO}_3, in which skyrmions were first observed in an insulator [53]. Polarization-resolved near-field photoluminescence could then image the valley-split island and the ring of brightened dark excitons, the latter being the local counterpart of the brightening observed in uniform in-plane fields [38]. Writing and erasing individual skyrmions [12] would switch these signatures on and off, and the trapped states would appear as emission lines whose energy follows the size of the skyrmion as the magnetic field is varied.

Discussion

The picture that emerges from our analysis differs in several respects from what one would expect by analogy with electrons. Because the exciton is a composite particle, spin–orbit coupling decides whether it experiences the topology of a skyrmion at all, and the two valleys experience it with opposite signs. When the flux is present, the quantum geometry of the texture traps the exciton. The quantum metric provides the barrier, while the topology fixes the angular momentum of the trapped states and, through it, the light that can address them. In contrast to a conventional quantum dot, this trap can be moved by a current together with its exciton. Throughout this work, we have tried to distinguish clearly between the properties fixed by topology, namely the flux quantum and the shift of angular momentum, and those that depend on details of the system, such as energies, lifetimes and transport distances.

Several extensions appear natural. The quantitative values of the proximity exchange at interfaces between excitonic semiconductors and insulating skyrmion hosts remain to be determined, and first-principles calculations for specific heterostructures would sharpen the predictions of Figure 1. Disorder, deformations of moving skyrmions and exciton–phonon scattering, which we have not included, will set the practical limits of transport. Finally, the results are not restricted to excitons. Any neutral composite particle in a texture-induced gauge field, including cold atoms in synthetic gauge potentials [54], should display the same flux switch and the same type of geometric trap.

Excitons and skyrmions have so far been studied largely in separate communities, one optical and one magnetic. The possibility of trapping excitons in skyrmions, labelling them by topology and moving them with currents could motivate experiments that bring the two together. This defines an exciting research road for the realization of hybrid devices for future information technology, where both worlds meet: topological magnetism and optical excitations.

Methods

The essential steps of the theory and the equations used in the calculations are summarized here. The full derivations are given in Supplementary Notes 1, 2, 5, 7 and 8, and Supplementary Note 10 lists, for every curve shown, the equation that was evaluated and the parameters that were used.

Excitons from first principles. Ground-state calculations of monolayer MoS2\mathrm{MoS}_2 were performed with Quantum Espresso [55, 56] using Optimized Norm-Conserving Vanderbilt (ONCV) pseudopotentials [57], the Perdew–Burke–Ernzerhof (PBE) functional [58] and spin–orbit coupling. Quasiparticle energies were obtained within G0W0G_0W_0 [59, 60, 61, 62] with the generalized plasmon-pole model [63] and a truncated Coulomb interaction [64], and the BSE [61, 62, 65] was solved with Yambo [66] on a 27×27×127 \times27 \times1 kk-mesh [67], including spin–orbit coupling, so that bright and spin-forbidden dark excitons are obtained on the same footing. At KK we find spin–orbit splittings of 149.5 meV (valence) and 1.6 meV (conduction) in GWGW. For the model, we use the parameters of hBN-encapsulated monolayers measured by high-field magneto-optics [68]: for MoS2\mathrm{MoS}_2, a reduced mass μ=0.275m0\mu= 0.275m_0, a binding energy of 221 meV and an rms radius of the 1s exciton of 1.2 nm, consistent with the extent of the BSE density in Supplementary Figure S1. Band masses are taken from k⋅p\mathbf{k}\cdot\mathbf{p} fits to first-principles bands [39] (me≈0.44m_{\mathrm{e}}\approx0.44–0.47m00.47m_0, mh≈0.54m0m_{\mathrm{h}}\approx0.54m_0 for MoS2_2, M≈1.0m0M\approx1.0m_0).

Figure S1. Excitons of free-standing monolayer MoS₂ from GW–BSE. a, Imaginary part of the in-plane polarizability with (solid) and without (dashed) electron–hole interaction, both on top of GW quasiparticle energies; A and B mark the bright 1s excitons at 1.91 and 2.03 eV, the dotted line the quasiparticle gap at KK (2.73 eV). b, Oscillator strengths of the lowest exciton states relative to the bright A exciton: the spin-forbidden dark A exciton lies 23 meV below the bright one, and the dark B exciton 15 meV below the bright B exciton. c, Real-space electron density of the A exciton with the hole fixed near a Mo atom (top: side view; bottom: top view).

The magneto-optical reduced mass, μ=0.275m0\mu=0.275m_0, points to a heavier electron (me≈0.56m0m_{\mathrm{e}}\approx0.56m_0 for mh=0.54m0m_{\mathrm{h}}=0.54m_0) [68], i.e., M≈1.1m0M\approx1.1m_0; heavier excitons bind more strongly in a given trap, so that our estimates of binding and transport are conservative. In the model calculations we use a representative bright–dark splitting ΔBD=5\Delta_{BD}=5 meV, between the measured values of MoSe2_2 (−1.4-1.4 meV) and MoS2_2 (+14+14 meV) [38]; the dark-exciton ring weight scales as (Jc/ΔBD)2(J_{\mathrm{c}}/\Delta_{BD})^2, so it is weaker in MoS2_2 and stronger in MoSe2_2.

Magnetic interactions from first principles. We modelled Pd/Fe/Ir(111), without and with an overlayer of two hBN layers and a MoS2_2 monolayer, by a slab of five Ir(111) layers covered by Fe, Pd and the overlayer, with a 2×22\times2 in-plane supercell (a=2.715a=2.715 Å) and about 24 Å of vacuum. The structure was relaxed with Quantum Espresso [55] and PAW pseudopotentials [69] (plane-wave cutoff 90 Ry, 12×12×112\times12\times1 kk-points), optimizing the atomic positions and the in-plane lattice vectors until the forces were below 10−510^{-5} Ry/a0a_0. The magnetic interactions were then computed with the full-potential Korringa–Kohn–Rostoker Green-function method (JuKKR [52]) in the local density approximation [70], with spin–orbit coupling included self-consistently, ℓmax⁡=3\ell_{\max}=3 and an electronic temperature of 502.78 K on the complex energy contour. The isotropic exchange JijJ_{ij} and the Dzyaloshinskii–Moriya vectors Dij\mathbf{D}_{ij} [71] were obtained from infinitesimal rotations in the ferromagnetic state [72, 73] on a 200×200×1200\times200\times1 kk-mesh, for all Fe pairs up to 8a8a.

Skyrmion profiles. We describe a Néel skyrmion of charge ν\nu by n^=(sin⁡θcos⁡νφ,sin⁡θsin⁡νφ,cos⁡θ)\hat{\mathbf{n}}=(\sin\theta\cos\nu\varphi,\sin\theta\sin\nu\varphi,\cos\theta) with the 360∘360^\circ-wall profile θ(R)=π+∑±arcsin⁡tanh⁡[(R±c)/w]\theta(R)=\pi+\sum_{\pm}\arcsin\tanh[(R\pm c)/w], which describes measured skyrmions accurately [9]; cc is the radius and ww the wall parameter. For Pd/Fe/Ir(111), cc and ww were obtained by minimizing the micromagnetic energy of ref. [9] within this profile (Supplementary Note 6).

Emergent fields of a single carrier. The effective field bi\mathbf{b}_i of Eq. (1) lies in the vertical plane containing n^\hat{\mathbf{n}}. Its azimuth is νφ\nu\varphi and its polar angle Θi\Theta_i follows from cos⁡Θi=(Jicos⁡θ+ηiτΔi)/∣bi∣\cos\Theta_i=(J_i\cos\theta+\eta_i\tau\Delta_i)/|\mathbf{b}_i|. Projecting hi\mathbf{h}_i onto the spinor χi\chi_i of the branch si=±1s_i=\pm1 gives

hiad=(p−ℏai)22mi+si∣bi∣+Wi,ai=−1−cos⁡Θi2∇φi,Wi=ℏ28mi[(∇Θi)2+sin⁡2Θi(∇φi)2],(4)h_i^{\mathrm{ad}}=\frac{(\mathbf{p}-\hbar\mathbf{a}_i)^2}{2m_i}+s_i|\mathbf{b}_i|+W_i,\qquad\mathbf{a}_i=-\frac{1-\cos\Theta_i}{2}\nabla\varphi_i,\qquad W_i=\frac{\hbar^2}{8m_i}\left[(\nabla\Theta_i)^2+\sin^2\Theta_i(\nabla\varphi_i)^2\right], \tag*{(4)}

where ai=i⟨χi∣∇χi⟩\mathbf{a}_i=i\langle\chi_i|\nabla\chi_i\rangle is the Berry connection and WiW_i is ℏ2/2mi\hbar^2/2m_i times the trace of the quantum metric of χi\chi_i. For the hole, ah=−ava_h=-a_v. The flux of ∇×ai\nabla\times\mathbf{a}_i is the solid angle swept by b^i\hat{\mathbf{b}}_i, which gives Eq. (2). The zz component of bi\mathbf{b}_i equals Ji+ηiτΔiJ_i+\eta_i\tau\Delta_i outside the skyrmion and −Ji+ηiτΔi-J_i+\eta_i\tau\Delta_i in the core; the two have opposite signs, and the flux is a full quantum, only if Ji>ΔiJ_i>\Delta_i.

Two-body Hamiltonian. With R=αre+βrh\mathbf{R}=\alpha\mathbf{r}_{\mathrm{e}}+\beta\mathbf{r}_{\mathrm{h}}, r=re−rh\mathbf{r}=\mathbf{r}_{\mathrm{e}}-\mathbf{r}_{\mathrm{h}}, α=me/M\alpha=m_{\mathrm{e}}/M, β=mh/M\beta=m_{\mathrm{h}}/M and μ=memh/M\mu=m_{\mathrm{e}}m_{\mathrm{h}}/M, we expand ae(R+βr)\mathbf{a}_{\mathrm{e}}(\mathbf{R}+\beta\mathbf{r}) and ah(R−αr)\mathbf{a}_{\mathrm{h}}(\mathbf{R}-\alpha\mathbf{r}) to first order in r\mathbf{r}. The symmetric part of (r⋅∇)a(\mathbf{r}\cdot\nabla)\mathbf{a} is a gradient in r\mathbf{r} and is gauged away, leaving 12Ωz^×r\frac{1}{2}\Omega\hat{\mathbf{z}}\times\mathbf{r} with Ω=z^⋅(∇×a)\Omega=\hat{\mathbf{z}}\cdot(\nabla\times\mathbf{a}). The kinetic energy then separates exactly into Π2/2M+π2/2μ\boldsymbol{\Pi}^2/2M+\boldsymbol{\pi}^2/2\mu, with Π=P−ℏaX−ℏ2Ωz^×r\boldsymbol{\Pi}=\mathbf{P}-\hbar\mathbf{a}_{X}-\frac{\hbar}{2}\Omega\hat{\mathbf{z}}\times\mathbf{r} and π=p−ℏ(βae−αah)−ℏ2Ωz^×r\boldsymbol{\pi}=\mathbf{p}-\hbar(\beta\mathbf{a}_{\mathrm{e}}-\alpha\mathbf{a}_{\mathrm{h}})-\frac{\hbar}{2}\Omega\hat{\mathbf{z}}\times\mathbf{r}. The r\mathbf{r}-independent term of π\boldsymbol{\pi} is removed by U=exp⁡[i⋅(βae−αah)]U=\exp[i\cdot(\beta\mathbf{a}_{\mathrm{e}}-\alpha\mathbf{a}_{\mathrm{h}})]. Because UU depends on R\mathbf{R}, it also shifts P\mathbf{P} by ℏ∇R⋅(βae−αah)\hbar\nabla_{\mathbf{R}}\cdot(\beta\mathbf{a}_{\mathrm{e}}-\alpha\mathbf{a}_{\mathrm{h}}), whose antisymmetric part is −ℏ2Ωz^×r-\frac{\hbar}{2}\Omega\hat{\mathbf{z}}\times\mathbf{r}. This doubles the corresponding term in Π\boldsymbol{\Pi} and gives Eq. (3). Averaging over the 1s state, terms linear in r\mathbf{r} and the orbital Zeeman term vanish, and the emergent diamagnetic shift ℏ2⟨r2⟩[Ω2/2M+Ω2/8μ]\hbar^2\langle r^2\rangle[\Omega^2/2M+\Omega^2/8\mu] remains. In a uniform field, ℏae,h=∓eA\hbar\mathbf{a}_{\mathrm{e,h}}=\mp e\mathbf{A}, and Eq. (3) reduces to the Gor’kov–Dzyaloshinskii Hamiltonian [40].

Bright–dark Hamiltonian and threshold. With the hole spin pinned, the texture acts on the bright (BB) and dark (DD) excitons through the electron spin. For an exciton small compared with the skyrmion, and measuring energies from the bright exciton,

Hτ(R)=(τ(Jc−Jv)cos⁡θJcsin⁡θe−iτνφJcsin⁡θeiτνφ−ΔBD−τ(Jc+Jv)cos⁡θ).(5)H_{\tau}(\mathbf{R}) = \begin{pmatrix} \tau(J_c-J_v)\cos\theta& J_c\sin\theta e^{-i\tau\nu\varphi} \\ J_c\sin\theta e^{i\tau\nu\varphi} & -\Delta_{\mathrm{BD}}-\tau(J_c+J_v)\cos\theta \end{pmatrix}. \tag*{(5)}

Written as d0+b⋅σd_0+\mathbf{b}\cdot\boldsymbol{\sigma}, the pseudospin field has bx−iby=Jcsin⁡θe−iτνφb_x-i b_y=J_c\sin\theta e^{-i\tau\nu\varphi} and bz=τJccos⁡θ+ΔBD/2b_z=\tau J_c\cos\theta+\Delta_{\mathrm{BD}}/2. This is the single-carrier problem with exchange JcJ_c and Ising term ΔBD/2\Delta_{\mathrm{BD}}/2, so the branch connected to the bright exciton acquires a flux quantum when Jc>J∗=∣ΔBD∣/2J_c>J^*=|\Delta_{\mathrm{BD}}|/2, the threshold used in Figure 1c. The phase eiτνφe^{i\tau\nu\varphi} couples a bright partial wave ℓ\ell only to a dark wave ℓ+τν\ell+\tau\nu. For skyrmions comparable to the exciton, n^\hat{\mathbf{n}} is replaced by its average over the 1s density as seen by the electron at R+βr\mathbf{R}+\beta\mathbf{r} and by the hole at R−αr\mathbf{R}-\alpha\mathbf{r} (Supplementary Note 5).

Exact and adiabatic scattering. For each ℓ\ell, the radial functions uBu_B (angular momentum ℓ\ell) and uDu_D (ℓ+τν\ell+\tau\nu) obey

uc′′=[mc2R2−14R2+Ec(R)−EtM]uc+Jcsin⁡θtMucˉ,tM=ℏ22M,(6)u_c''=\left[\frac{m_c^2}{R^2}-\frac{1}{4R^2}+\frac{E_c(R)-E}{t_M}\right]u_c+\frac{J_c\sin\theta}{t_M}u_{\bar{c}}, \qquad t_M=\frac{\hbar^2}{2M}, \tag*{(6)}

with Ec(R)E_c(R) the diagonal elements of Eq. (5) and E=τ(Jc−Jv)+εE=\tau(J_c-J_v)+\varepsilon. Two regular solutions are integrated outward (DOP853) and matched to RHmc(2)(kcR)\sqrt{R}H_{m_c}^{(2)}(k_cR) (amplitude AcA_c) and RHmc(1)(kcR)\sqrt{R}H_{m_c}^{(1)}(k_cR) (amplitude BcB_c). Since the outgoing flux is proportional to ∣Bc∣2|B_c|^2 independently of kck_c, S=BA−1S=BA^{-1} is flux-normalized; it is unitary to better than 10−310^{-3}. The cross-sections are σel=kB−1∑ℓ∣SBBℓ−1∣2\sigma_{\mathrm{el}}=k_B^{-1}\sum_{\ell}|S_{BB}^{\ell}-1|^2 and σB→D=kB−1∑ℓ∣SDBℓ∣2\sigma_{B\to D}=k_B^{-1}\sum_{\ell}|S_{DB}^{\ell}|^2, and the skew is ⟨sin⁡θs⟩=∫sin⁡θs∣fBB∣2/∫∣fBB∣2\langle\sin\theta_s\rangle=\int\sin\theta_s|f_{BB}|^2/\int|f_{BB}|^2 with fBB=(2πkB)−1/2∑ℓ(SBBℓ−1)eiℓθsf_{BB}=(2\pi k_B)^{-1/2}\sum_{\ell}(S_{BB}^{\ell}-1)e^{i\ell\theta_s}. In the adiabatic limit, we follow the eigenvector v=(vB,vD)\mathbf{v}=(v_B,v_D) of Eq. (5) that is bright at infinity and solve one radial equation with the Berry potential A(R)=−τν[∣vD(R)∣2−∣vD(∞)∣2]A(R)=-\tau\nu[|v_D(R)|^2-|v_D(\infty)|^2] (gauge regular at infinity) and the geometric potential W=tM[∣∂Rv∣2+ν2∣vB∣2∣vD∣2/R2]W=t_M[|\partial_R\mathbf{v}|^2+\nu^2|v_B|^2|v_D|^2/R^2]. The net flux is τν∣vD(0)∣2\tau\nu|v_D(0)|^2.

Resonances, bound states and optical weights. For an exciton that follows the texture completely, the effective radial potential is

Uℓ(R)=tM(ℓ−A)2−14R2+W+V,A=ν2(cos⁡θ−1),W=tM4[θ′2+ν2sin⁡2θR2],(7)U_{\ell}(R)=t_M\frac{(\ell-A)^2-\frac{1}{4}}{R^2}+W+V,\qquad A=\frac{\nu}{2}(\cos\theta-1),\qquad W=\frac{t_M}{4}\left[\theta'^2+\nu^2\frac{\sin^2\theta}{R^2}\right], \tag*{(7)}

with an optional exchange well V=−V0(1−cos⁡θ)/2V=-V_0(1-\cos\theta)/2. Phase shifts δℓ\delta_\ell are obtained by matching to R[aJ∣ℓ∣(kR)+bY∣ℓ∣(kR)]\sqrt{R}[aJ_{|\ell|}(kR)+bY_{|\ell|}(kR)], tan⁡δℓ=−b/a\tan\delta_\ell=-b/a, and resonances from the Wigner time delay τW=2ℏ dδℓ/dE\tau_W=2\hbar\,d\delta_\ell/dE, with Γ=4ℏ/τWmax⁡\Gamma=4\hbar/\tau_W^{\max} and lifetime ℏ/Γ\hbar/\Gamma. The controls use A≡0A\equiv0 (no flux) and the skyrmionium profile θ=π2[2+tanh⁡R−c1w+tanh⁡R−c2w]\theta=\frac{\pi}{2}[2+\tanh\frac{R-c_1}{w}+\tanh\frac{R-c_2}{w}]. Bound states are obtained on a cell-centred finite-volume grid (Rmax⁡=300R_{\max}=300–400400 nm). The optical weight for a near-field spot with orbital angular momentum mm is Sm(E)=k−1∣∫dRR gm vBuE∣2S_m(E)=k^{-1}|\int dR\sqrt{R}\,g_m\,v_Bu_E|^2, with gm=(R/s)me−R2/2s2g_m=(R/s)^m e^{-R^2/2s^2}, s=30s=30 nm, the bright amplitude vB=(1+cos⁡θ)/2v_B=(1+\cos\theta)/2 and uEu_E normalized to unit asymptotic amplitude.

Transport. For a trap moving at constant velocity v\mathbf{v}, ψ=exp⁡[iℏ(Mv⋅R−12Mv2t−E0t)]ψ0(R−vt)\psi=\exp[\frac{i}{\hbar}(M\mathbf{v}\cdot\mathbf{R}-\frac{1}{2}Mv^2t-E_0t)]\psi_0(\mathbf{R}-\mathbf{v}t) solves the time-dependent problem exactly, so no loss occurs. After an instantaneous velocity change Δv\Delta\mathbf{v}, the probability of remaining bound is

P(Δv)=∣∫0∞2πR dR ∣ψ0(R)∣2J0(qR)∣2,q=MΔvℏ,(8)P(\Delta v)=\left|\int_0^\infty2\pi R\,dR\,|\psi_0(R)|^2J_0(qR)\right|^2,\qquad q=\frac{M\Delta v}{\hbar}, \tag*{(8)}

an upper bound for the loss at any finite acceleration. In the full model, the ground state carries ℓ=−ν\ell=-\nu, but its density is still rotationally symmetric, so the same expression applies. In the reference model of Figure 5, AA and WW are set to zero. Thermal escape is estimated as τesc=ν0−1eEb/kBT\tau_{\mathrm{esc}}=\nu_0^{-1}e^{E_b/k_BT} with ν0=(E1−E0)/h\nu_0=(E_1-E_0)/h from the two lowest ℓ=0\ell=0 levels; this prefactor is uncertain by about an order of magnitude. The interaction of two interlayer excitons in the trap is evaluated in the Hartree approximation, Udip=⟨e24πϵ0ϵ(2s−2s2+d2)⟩U_{\mathrm{dip}}=\left\langle\frac{e^2}{4\pi\epsilon_0\epsilon}\left(\frac{2}{s}-\frac{2}{\sqrt{s^2+d^2}}\right)\right\rangle, by Monte Carlo sampling of the ground-state density, with d=0.65 nmd = 0.65\ \mathrm{nm} and ε=4.5\varepsilon= 4.5. For interlayer excitons in MoS2/WSe2\mathrm{MoS_2}/\mathrm{WSe_2} we use M=0.83m0M = 0.83m_0, the sum of the first-principles electron mass of MoS2\mathrm{MoS_2} and hole mass of WSe2\mathrm{WSe_2}[39] (me≈0.44m_e \approx0.44–0.470.47, mh≈0.36m0m_h \approx0.36m_0); with the heavier electron mass inferred from magneto-optics [68], M≈1.06m0M \approx1.06m_0, which again makes our estimates conservative (Supplementary Table S3).

Data availability

The data that support the findings of this study are present in the paper and the Supplementary Information.

Code availability

The first-principles codes Quantum Espresso[56] and Yambo[66] are freely available. The Korringa–Kohn–Rostoker Green-function code JuKKR, developed in Jülich, is also available [52]; It is a rather complex ab-initio DFT-based code, which is in general impossible to use without proper training on the theory behind it and on the practical utilization of the code. The Python codes used for the model calculations and for all figures, together with the input files needed to reproduce the figures, are available at https://doi.org/10.5281/zenodo.24011147.

Acknowledgements

Funding has been provided by the Priority Programme SPP 2137 “Skyrmionics” (Project LO 1659/8-1) of the Deutsche Forschungsgemeinschaft (DFG). We also acknowledge funding through Project No. 328545488 – CRC/TRR 227, Project No. B12, and by the German Excellence Strategy – EXC 3112/1 – 533767171 (Center for Chiral Electronics). K.R. acknowledges funding from the Open Research Fund of the General Direction of Research and Technological Development (DGRSDT) of the Ministry of Higher Education and Scientific Research, Algeria. The authors disclose the use of AI tools to assist with drafting and refining portions of the manuscript based on author-provided scientific ideas and context that were critically reviewed, verified, and revised.

Author contributions

S.L. initiated, designed, and supervised the project. K.R. developed the theoretical framework with the support of S.Z. and supervision of S.L. Simulations were performed by S.Z. and K.R. The three authors discussed the results and wrote the manuscript.

Competing interests

The authors declare no competing interests.

Emergent fields of a composite exciton

Lengths are given in nm and energies in meV. We use the kinetic constant

tM≡ℏ22M=38.0998M/m0 meV nm2,(9)t_M \equiv\frac{\hbar^2}{2M} = \frac{38.0998}{M/m_0}\ \mathrm{meV\ nm^2}, \tag*{(9)}

ℏ=0.65821 meV ps\hbar= 0.65821\ \mathrm{meV\ ps} and kBT=0.362 meVk_B T = 0.362\ \mathrm{meV} at 4.2 K. Unless stated otherwise, the exciton mass is M=1.01m0M = 1.01m_0 (intralayer exciton of MoS2_2, me=0.47m_e = 0.47, mh=0.54m_h = 0.54) and M=0.83m0M = 0.83m_0 for interlayer excitons in MoS2_2/WSe2_2. The magnetization of a rotationally symmetric Néel skyrmion of charge ν\nu is

n^(R)=(sin⁡θ(R)cos⁡νφ, sin⁡θ(R)sin⁡νφ, cos⁡θ(R)),θ(R)=π+∑±arcsin⁡tanh⁡R±cw,(10)\hat{\mathbf{n}}(\mathbf{R}) = \left(\sin\theta(R)\cos\nu\varphi,\ \sin\theta(R)\sin\nu\varphi,\ \cos\theta(R)\right),\qquad\theta(R) = \pi+ \sum_{\pm}\arcsin\tanh\frac{R\pm c}{w}, \tag*{(10)}

with θ(0)=π\theta(0) = \pi (core pointing down) and θ(∞)=2π\theta(\infty) = 2\pi (up). The parameter cc is the skyrmion radius and ww the wall parameter. This 360°-wall profile describes measured skyrmions accurately.

Model. Near a valley τ=±1\tau= \pm1, a conduction electron (i=ci = c) or valence electron (i=vi = v) of a transition-metal dichalcogenide placed next to the texture is described by

hi=p22mi+bi(r)⋅σ,bi=Jin^(r)+ηiτΔiz^.(11)h_i = \frac{\mathbf{p}^2}{2m_i} + \mathbf{b}_i(\mathbf{r})\cdot\boldsymbol{\sigma},\qquad\mathbf{b}_i = J_i\hat{\mathbf{n}}(\mathbf{r}) + \eta_i\tau\Delta_i\hat{\mathbf{z}}. \tag*{(11)}

The first term of bi\mathbf{b}_i is the exchange field of the texture, the second the Ising spin–orbit field, which points in opposite directions in the two valleys; ηi=±1\eta_i = \pm1 fixes the band ordering.

Direction of the effective field. Because bi\mathbf{b}_i is the sum of Jin^J_i\hat{\mathbf{n}} and a field along z^\hat{\mathbf{z}}, it lies in the same vertical plane as n^\hat{\mathbf{n}}. Its azimuth is therefore that of the texture, ϕi=νφ\phi_i = \nu\varphi, and its polar angle Θi\Theta_i follows from

cos⁡Θi=Jicos⁡θ+ηiτΔi∣bi∣,sin⁡Θi=Jisin⁡θ∣bi∣,∣bi∣=Ji2+Δi2+2ηiτJiΔicos⁡θ.(12)\cos\Theta_i = \frac{J_i\cos\theta+\eta_i\tau\Delta_i}{\lvert\mathbf{b}_i\rvert},\qquad\sin\Theta_i = \frac{J_i\sin\theta}{\lvert\mathbf{b}_i\rvert},\qquad\lvert\mathbf{b}_i\rvert= \sqrt{J_i^2+\Delta_i^2+2\eta_i\tau J_i\Delta_i\cos\theta}. \tag*{(12)}

Adiabatic projection. We write the spinor of the branch parallel (si=+1s_i = +1) or antiparallel (si=−1s_i = -1) to bi\mathbf{b}_i. For si=+1s_i = +1,

χi=(cos⁡(Θi/2)eiϕisin⁡(Θi/2)).(13)\chi_i = \begin{pmatrix} \cos(\Theta_i/2)\\ e^{i\phi_i}\sin(\Theta_i/2) \end{pmatrix}. \tag*{(13)}

Projecting hih_i onto χi\chi_i, i.e., writing ψ=χif\psi= \chi_i f and multiplying from the left by χi†\chi_i^\dagger, gives

hiad=(p−ℏai)22mi+ϵi+Wi,(14)h_i^{\mathrm{ad}} = \frac{(\mathbf{p}-\hbar\mathbf{a}_i)^2}{2m_i}+\epsilon_i+W_i, \tag*{(14)}

with the Berry connection, the branch energy and the geometric potential

ai=i⟨χi∣∇χi⟩=−sin⁡2Θi2∇ϕi=−1−cos⁡Θi2∇ϕi,(15)\mathbf{a}_i = i\langle\chi_i\lvert\nabla\chi_i\rangle= -\sin^2\frac{\Theta_i}{2}\nabla\phi_i = -\frac{1-\cos\Theta_i}{2}\nabla\phi_i, \tag*{(15)}
ϵi=si∣bi∣,(16)\epsilon_i = s_i\lvert\mathbf{b}_i\rvert, \tag*{(16)}
Wi=ℏ22mi(⟨∇χi∣∇χi⟩−∣ai∣2)=ℏ28mi[(∇Θi)2+sin⁡2Θi(∇ϕi)2].(17)W_i = \frac{\hbar^2}{2m_i}\left(\langle\nabla\chi_i\lvert\nabla\chi_i\rangle-\lvert\mathbf{a}_i\rvert^2\right)=\frac{\hbar^2}{8m_i}\left[(\nabla\Theta_i)^2+\sin^2\Theta_i(\nabla\phi_i)^2\right]. \tag*{(17)}

The bracket in Eq. (17) is the trace of the quantum metric of χi\chi_i, gμμ(i)=Re⁡⟨∂μχi∣(1−∣χi⟩⟨χi∣)∣∂μχi⟩g_{\mu\mu}^{(i)} = \operatorname{Re}\langle\partial_\mu\chi_i\lvert(1-\lvert\chi_i\rangle\langle\chi_i\rvert)\rvert\partial_\mu\chi_i\rangle. Hence Wi=(ℏ2/2mi)Tr⁡g(i)W_i = (\hbar^2/2m_i)\operatorname{Tr}g^{(i)} is non-negative and largest where the effective field rotates quickly. The hole is the absence of a valence electron; its connection and energy are ah=−av\mathbf{a}_h = -\mathbf{a}_v and ϵh=−ϵv\epsilon_h = -\epsilon_v.

Emergent field and flux. With ∇ϕi=(ν/R)ϕ^\nabla\phi_i=(\nu/R)\hat{\boldsymbol{\phi}}, Eq. (15) gives ai,ϕ=−ν(1−cos⁡Θi)/(2R)a_{i,\phi}=-\nu(1-\cos\Theta_i)/(2R), and the emergent field is

Ωi=z^⋅(∇×ai)=1RddR(Rai,ϕ)=−ν2Rsin⁡ΘidΘidR.(18)\Omega_i=\hat{\mathbf{z}}\cdot(\nabla\times\mathbf{a}_i)=\frac{1}{R}\frac{d}{dR}(Ra_{i,\phi})=-\frac{\nu}{2R}\sin\Theta_i\frac{d\Theta_i}{dR}. \tag*{(18)}

Integrating over the plane, the flux in units of Φ0=2π\Phi_0=2\pi (in units where the emergent charge is one) is

ΦiΦ0=12π∫d2R Ωi=siν2[cos⁡Θi(0)−cos⁡Θi(∞)],(19)\frac{\Phi_i}{\Phi_0}=\frac{1}{2\pi}\int d^2R\,\Omega_i=\frac{s_i\nu}{2}\left[\cos\Theta_i(0)-\cos\Theta_i(\infty)\right], \tag*{(19)}

which is Eq. (2) of the main text; the factor sis_i accounts for the branch.

Coordinates. The two-body Hamiltonian in the adiabatic limit is

H=∑i=e,h[(pi−ℏai(ri))22mi+εi(ri)+Wi(ri)]+VC(re−rh).(20)H=\sum_{i=e,h}\left[\frac{(\mathbf{p}_i-\hbar\mathbf{a}_i(\mathbf{r}_i))^2}{2m_i}+\varepsilon_i(\mathbf{r}_i)+W_i(\mathbf{r}_i)\right]+V_C(\mathbf{r}_e-\mathbf{r}_h). \tag*{(20)}

We introduce R=αre+βrh\mathbf{R}=\alpha\mathbf{r}_e+\beta\mathbf{r}_h and r=re−rh\mathbf{r}=\mathbf{r}_e-\mathbf{r}_h, with α=me/M\alpha=m_e/M, β=mh/M\beta=m_h/M, M=me+mhM=m_e+m_h and μ=memh/M\mu=m_em_h/M. Then re=R+βr\mathbf{r}_e=\mathbf{R}+\beta\mathbf{r}, rh=R−αr\mathbf{r}_h=\mathbf{R}-\alpha\mathbf{r}, and the momenta are pe=αP+p\mathbf{p}_e=\alpha\mathbf{P}+\mathbf{p}, ph=βP−p\mathbf{p}_h=\beta\mathbf{P}-\mathbf{p}.

Expansion to first order in r\mathbf{r}. We expand ae(R+βr)≃ae+β(r⋅∇)ae\mathbf{a}_e(\mathbf{R}+\beta\mathbf{r})\simeq\mathbf{a}_e+\beta(\mathbf{r}\cdot\nabla)\mathbf{a}_e and ah(R−αr)≃ah−α(r⋅∇)ah\mathbf{a}_h(\mathbf{R}-\alpha\mathbf{r})\simeq\mathbf{a}_h-\alpha(\mathbf{r}\cdot\nabla)\mathbf{a}_h, all quantities on the right being evaluated at R\mathbf{R}. The tensor ∂kaj\partial_k a_j is split into symmetric and antisymmetric parts,

rk∂kaj=12rk(∂kaj+∂jak)+12rk(∂kaj−∂jak)=∂∂rj[12rkrl∂kal]+12Ω(z^×r)j.(21)r_k\partial_k a_j=\frac{1}{2}r_k(\partial_k a_j+\partial_j a_k)+\frac{1}{2}r_k(\partial_k a_j-\partial_j a_k)=\frac{\partial}{\partial r_j}\left[\frac{1}{2}r_k r_l\partial_k a_l\right]+\frac{1}{2}\Omega(\hat{\mathbf{z}}\times\mathbf{r})_j. \tag*{(21)}

The symmetric part is a gradient with respect to r\mathbf{r} and is removed by a gauge transformation in the relative coordinate; it contributes only at second order in r\mathbf{r} to the centre-of-mass motion, which we neglect. The kinetic momenta become

πe=αP+p−ℏae−ℏβ2Ωez^×r,(22)\boldsymbol{\pi}_e=\alpha\mathbf{P}+\mathbf{p}-\hbar\mathbf{a}_e-\frac{\hbar\beta}{2}\Omega_e\hat{\mathbf{z}}\times\mathbf{r}, \tag*{(22)}
πh=βP−p−ℏah+ℏα2Ωhz^×r.(23)\boldsymbol{\pi}_h=\beta\mathbf{P}-\mathbf{p}-\hbar\mathbf{a}_h+\frac{\hbar\alpha}{2}\Omega_h\hat{\mathbf{z}}\times\mathbf{r}. \tag*{(23)}

Exact separation of the kinetic energy. For any pair of momenta, πe2/2me+πh2/2mh=Π2/2M+π2/2μ\boldsymbol{\pi}_e^2/2m_e+\boldsymbol{\pi}_h^2/2m_h=\boldsymbol{\Pi}^2/2M+\boldsymbol{\pi}^2/2\mu with Π=πe+πh\boldsymbol{\Pi}=\boldsymbol{\pi}_e+\boldsymbol{\pi}_h and π=βπe−απh\boldsymbol{\pi}=\beta\boldsymbol{\pi}_e-\alpha\boldsymbol{\pi}_h. Here

Π=P−ℏaX−ℏ2Ω−z^×r,aX=ae+ah,Ω−=βΩe−αΩh,(24)\boldsymbol{\Pi}=\mathbf{P}-\hbar\mathbf{a}_X-\frac{\hbar}{2}\Omega_-\hat{\mathbf{z}}\times\mathbf{r},\qquad\mathbf{a}_X=\mathbf{a}_e+\mathbf{a}_h,\qquad\Omega_-=\beta\Omega_e-\alpha\Omega_h, \tag*{(24)}
π=p−ℏc−ℏ2Ω+z^×r,c=βae−αah,Ω+=β2Ωe+α2Ωh.(25)\boldsymbol{\pi}=\mathbf{p}-\hbar\mathbf{c}-\frac{\hbar}{2}\Omega_+\hat{\mathbf{z}}\times\mathbf{r},\qquad\mathbf{c}=\beta\mathbf{a}_e-\alpha\mathbf{a}_h,\qquad\Omega_+=\beta^2\Omega_e+\alpha^2\Omega_h. \tag*{(25)}

Gauge transformation. The r\mathbf{r}-independent potential c(R)\mathbf{c}(\mathbf{R}) in the relative momentum is removed by ψ=Uψ~\psi=U\tilde{\psi} with U=exp⁡[ir⋅c(R)]U=\exp[i\mathbf{r}\cdot\mathbf{c}(\mathbf{R})]. Since UU depends on R\mathbf{R}, it also shifts the centre-of-mass momentum, U†PU=P+ℏ∇R(r⋅c)U^\dagger\mathbf{P}U=\mathbf{P}+\hbar\nabla_{\mathbf{R}}(\mathbf{r}\cdot\mathbf{c}). Using ∇(r⋅c)=(r⋅∇)c+r×(∇×c)\nabla(\mathbf{r}\cdot\mathbf{c})=(\mathbf{r}\cdot\nabla)\mathbf{c}+\mathbf{r}\times(\nabla\times\mathbf{c}) at fixed r\mathbf{r}, and ∇×c=Ω−z^\nabla\times\mathbf{c}=\Omega_-\hat{\mathbf{z}}, we find ∇R(r⋅c)=[gradient in r]−12Ω−z^×r\nabla_{\mathbf{R}}(\mathbf{r}\cdot\mathbf{c})=[\text{gradient in }\mathbf{r}]-\frac{1}{2}\Omega_-\hat{\mathbf{z}}\times\mathbf{r}. The antisymmetric part therefore doubles the coefficient of z^×r\hat{\mathbf{z}}\times\mathbf{r} in Π\boldsymbol{\Pi}, and we obtain Eq. (3) of the main text,

H=[P−ℏaX−ℏΩ−z^×r]22M+[p−ℏc+ℏ2Ω+z^×r]22μ+VC(r)+∑i=e,h(εi+Wi).(26)H=\frac{\left[\mathbf{P}-\hbar\mathbf{a}_X-\hbar\Omega_-\hat{\mathbf{z}}\times\mathbf{r}\right]^2}{2M}+\frac{\left[\mathbf{p}-\hbar\mathbf{c}+\frac{\hbar}{2}\Omega_+\hat{\mathbf{z}}\times\mathbf{r}\right]^2}{2\mu}+V_C(\mathbf{r})+\sum_{i=e,h}(\varepsilon_i+W_i). \tag*{(26)}

Gor’kov–Dzyaloshinskii limit. For a uniform magnetic field, ℏae=−eA\hbar\mathbf{a}_e=-e\mathbf{A} and ℏah=+eA\hbar\mathbf{a}_h=+e\mathbf{A}, so that aX=0\mathbf{a}_X=0, ℏΩ−=−eB\hbar\Omega_-=-eB and ℏΩ+=−eB(mh−me)/M\hbar\Omega_+=-eB(m_h-m_e)/M. Equation (26) then reduces to the Hamiltonian of an exciton in a magnetic field. In a skyrmion, ae\mathbf{a}_e and ah\mathbf{a}_h do not cancel, because each depends on its own Ji/ΔiJ_i/\Delta_i, spin and valley.

Averaging over the 1s state. For an exciton small compared with the wall, we average Eq. (26) over the 1s state ϕ1s(r)\phi_{1s}(\mathbf{r}). Terms linear in r\mathbf{r} vanish by symmetry, and so does the orbital Zeeman term ∝Ω+(z^×r)⋅p\propto\Omega_{+}(\hat{\mathbf{z}} \times\mathbf{r}) \cdot\mathbf{p} for an ss state. With ⟨∣z^×r∣2⟩=⟨r2⟩\langle|\hat{\mathbf{z}} \times\mathbf{r}|^{2}\rangle= \langle r^{2}\rangle,

HCM=(P−ℏaX)22M+∑i(εi+Wi)+ℏ2⟨r2⟩[Ω−22M+Ω+28μ].(27)H_{\mathrm{CM}} = \frac{(\mathbf{P} - \hbar\mathbf{a}_{X})^{2}}{2M} + \sum_{i}(\varepsilon_{i} + W_{i}) + \hbar^{2}\langle r^{2}\rangle\left[\frac{\Omega_{-}^{2}}{2M} + \frac{\Omega_{+}^{2}}{8\mu}\right]. \tag*{(27)}

The last term is the emergent diamagnetic shift. The first, second (geometric) and last terms are non-negative; only the exchange energies εi\varepsilon_i can produce an attractive well.

The flux switch and the validity of the adiabatic picture

From Eq. (12), the zz component of bi\mathbf{b}_i is Ji+ηiτΔiJ_i+\eta_i\tau\Delta_i far from the skyrmion (θ=2π\theta=2\pi) and −Ji+ηiτΔi-J_i+\eta_i\tau\Delta_i in the core (θ=π\theta=\pi), while sin⁡Θi=0\sin\Theta_i=0 in both places. Therefore cos⁡Θi(0)=sign⁡(−Ji+ηiτΔi)\cos\Theta_i(0)=\operatorname{sign}(-J_i+\eta_i\tau\Delta_i) and cos⁡Θi(∞)=sign⁡(Ji+ηiτΔi)\cos\Theta_i(\infty)=\operatorname{sign}(J_i+\eta_i\tau\Delta_i). The two signs differ only if Ji>ΔiJ_i>\Delta_i, and Eq. (19) gives

ΦiΦ0={±ν,Ji>Δi,0,Ji<Δi.(28)\frac{\Phi_i}{\Phi_0} = \begin{cases} \pm\nu, & J_i > \Delta_i,\\ 0, & J_i < \Delta_i. \end{cases} \tag*{(28)}

For Ji<ΔiJ_i<\Delta_i, bi\mathbf{b}_i tilts in the wall and returns to its original direction, so the swept solid angle vanishes. For Ji>ΔiJ_i>\Delta_i, it is reversed in the core and sweeps the full sphere (main text Figure 2b). Because of spin–valley locking, the band-edge carrier occupies opposite branches in KK and K′K', so that sis_i, and with it the flux, has opposite signs in the two valleys.

Validity. The projection requires the local splitting 2∣bi∣2|\mathbf{b}_i| to exceed the kinetic energy associated with the rotation of b^i\hat{\mathbf{b}}_i, ℏ2/(2miw2)\hbar^2/(2m_iw^2). Near Ji≈ΔiJ_i\approx\Delta_i, ∣bi∣|\mathbf{b}_i| becomes small in the core and the condition fails. The adiabatic description is quantitative when Ji>max⁡{Δi,ℏ2/(2miw2)}J_i>\max\{\Delta_i,\hbar^2/(2m_iw^2)\}. For a mass m0m_0, ℏ2/(2m0w2)=38.1/w2\hbar^2/(2m_0w^2)=38.1/w^2 meV, i.e., 16.9 meV for w=1.5w=1.5 nm and 1.52 meV for w=5w=5 nm.

Excitons of monolayer MoS2 from first principles

To anchor the model parameters, we computed the excitons of free-standing monolayer MoS2_2 within many-body perturbation theory. Ground-state calculations were performed with Quantum Espresso[55, 56] using ONCV pseudopotentials[57], the PBE functional[58] and spin–orbit coupling. Quasiparticle energies were obtained within G0W0G_0W_0[60], with the generalized plasmon-pole model[63] and a truncated Coulomb interaction[64], and the Bethe–Salpeter equation was solved with Yambo[66] on a 27×27×127\times27\times1 kk-mesh. Since spin–orbit coupling is included throughout, bright and spin-forbidden dark excitons are obtained on the same footing.

The results are summarized in Supplementary Figure S1. The quasiparticle gap at KK is 2.73 eV, and the spin–orbit splittings at KK are 149.5 meV in the valence band and 1.6 meV in the conduction band. The bright A and B excitons appear at 1.91 and 2.03 eV, so that the A exciton is bound by 0.82 eV in the free-standing layer. The most relevant result for the present work concerns the fine structure. The lowest exciton is not the bright A exciton but a spin-forbidden dark state 23 meV below it, with an oscillator strength smaller by about six orders of magnitude. A similar dark state lies 15 meV below the bright B exciton. The ordering agrees with magneto-optical experiments on hBN-encapsulated MoS2_2, which find the dark exciton 14 meV below the bright one[38]. The smaller experimental value is consistent with the additional screening by hBN, which reduces the short-range electron–hole exchange that contributes to the bright–dark splitting. The real-space density of the A exciton (Supplementary Figure S1) extends over about a nanometre, in line with the radius of 1.2 nm inferred from high-field magneto-optics[68].

The structureless-spinor limit

When every constituent winds with the texture, the exciton can be described as a structureless spinor with a single exchange coupling, as in ref.[41]. In this limit, the emergent vector potential has a long-range tail, a∝ν/ρ\mathbf{a} \propto\nu/\rho, which makes the scattering problem delicate: the result depends on whether the tail is treated exactly or perturbatively, and on the gauge in which the calculation is performed. To clarify this point, we solved the structureless model exactly in partial waves, including the geometric scalar potential, in the gauge regular at the origin and in the gauge regular at infinity. Both gauges give identical cross-sections, and the results are converged in the number of partial waves and in the matching radius.

For the parameters of ref.[41] (exciton mass 5.6m05.6m_0, skyrmion radius of about 5 nm5\ \mathrm{nm}, wave vectors between 3.33.3 and 16.5 μm−116.5\ \mu\mathrm{m}^{-1}), the exact differential cross-section is smooth and nearly isotropic (Supplementary Fig. S2a). It does not show the forward divergence 4/(kθ2)4/(k\theta^2) that results when the 1/ρ1/\rho tail is treated to first order. The left–right asymmetry remains below 10−310^{-3} for all angles (Supplementary Fig. S2b). This is what one expects physically. At these wave vectors kR≪1kR \ll1, the exciton cannot resolve the texture, and chirality in the scattering requires kR≳1kR \gtrsim1, as in the calculations of Figure 3 of the main text.

Bright and dark excitons: from weak to strong coupling

Local Hamiltonian. When the hole spin is pinned by the large valence splitting, the texture acts on the exciton through the electron spin, which distinguishes bright (BB, electron spin parallel to the hole) from dark (DD) states. Measuring energies from the bright exciton in zero field, the local Hamiltonian in valley τ\tau is

Hτ(R)=(τJcnze−τJvnzhJc(nxe−iτnye)Jc(nxe+iτnye)−τJcnze−τJvnzh−ΔBD).(29)H_\tau(\mathbf{R}) = \begin{pmatrix} \tau J_c n_z^e-\tau J_v n_z^h & J_c(n_x^e-i\tau n_y^e) \\ J_c(n_x^e+i\tau n_y^e) & -\tau J_c n_z^e-\tau J_v n_z^h-\Delta_{\mathrm{BD}} \end{pmatrix}. \tag*{(29)}

Here ne\mathbf{n}^e and nzhn_z^h are the texture seen by the electron and the hole. For an exciton small compared with the skyrmion, both are the local texture at R\mathbf{R}, nz=cos⁡θn_z=\cos\theta and nx−iτny=sin⁡θ e−iτνφn_x-i\tau n_y=\sin\theta\,e^{-i\tau\nu\varphi}. This gives

Hτ(R)=(τ(Jc−Jv)cos⁡θJcsin⁡θ e−iτνφJcsin⁡θ eiτνφ−ΔBD−τ(Jc+Jv)cos⁡θ).(30)H_\tau(\mathbf{R}) = \begin{pmatrix} \tau(J_c-J_v)\cos\theta& J_c\sin\theta\,e^{-i\tau\nu\varphi} \\ J_c\sin\theta\,e^{i\tau\nu\varphi} & -\Delta_{\mathrm{BD}}-\tau(J_c+J_v)\cos\theta \end{pmatrix}. \tag*{(30)}

In a uniform magnetization (cos⁡θ=1\cos\theta=1), the bright exciton is split between the valleys by 2(Jc−Jv)2(J_c-J_v).

Pseudospin form and threshold. Writing Hτ=d0+b⋅σH_\tau=d_0+\mathbf{b}\cdot\boldsymbol{\sigma} in the bright–dark pseudospin space gives

d0=−τJvcos⁡θ−ΔBD2,bz=τJccos⁡θ+ΔBD2,bx−iby=Jcsin⁡θ e−iτνφ.(31)d_0=-\tau J_v\cos\theta-\frac{\Delta_{\mathrm{BD}}}{2}, \qquad b_z=\tau J_c\cos\theta+\frac{\Delta_{\mathrm{BD}}}{2}, \qquad b_x-i b_y=J_c\sin\theta\,e^{-i\tau\nu\varphi}. \tag*{(31)}

This is Eq. (11) for a pseudospin, with exchange JcJ_c and Ising term ΔBD/2\Delta_{\mathrm{BD}}/2. Outside the skyrmion bz=τJc+ΔBD/2b_z=\tau J_c+\Delta_{\mathrm{BD}}/2, in the core bz=−τJc+ΔBD/2b_z=-\tau J_c+\Delta_{\mathrm{BD}}/2. The pseudospin field is reversed, and the branch connected to the bright exciton carries a full flux quantum, if and only if

Jc>J∗=∣ΔBD∣2,(32)J_c > J^*=\frac{|\Delta_{\mathrm{BD}}|}{2}, \tag*{(32)}

which is the threshold used in Figure 1c for intralayer excitons. For interlayer excitons the electron–hole exchange is weak and the threshold reduces to the single-particle condition Jc>ΔcJ_c>\Delta_c of Supplementary Note 2.

Angular-momentum selection rule. Writing ψB=R−1/2uB(R)eiℓφ\psi_B=R^{-1/2}u_B(R)e^{i\ell\varphi}, the off-diagonal term eiτνφe^{i\tau\nu\varphi} of Eq. (30) couples it only to ψD=R−1/2uD(R)ei(ℓ+τν)φ\psi_D=R^{-1/2}u_D(R)e^{i(\ell+\tau\nu)\varphi}. Each scattering event therefore transfers τν\tau\nu units of angular momentum.

Figure S2. Gauge-consistent exact solution of the structureless-spinor model. Model of ref. [41] including the geometric scalar potential, with exciton mass 5.6m05.6m_0, skyrmion radius ≈5\approx5 nm and the wave vectors of that work. a, Exact differential cross-section (solid) compared with the small-angle form 4/(kθ2)4/(k\theta^2) obtained when the 1/ρ1/\rho tail of the vector potential is treated perturbatively (dotted). b, Exact left–right asymmetry. Results are identical in the gauges regular at the origin and at infinity, and are converged in partial waves and matching radius.

Weak-coupling ring. For Jc≪∣ΔBD∣J_c \ll|\Delta_{\mathrm{BD}}|, first-order perturbation theory gives the bright admixture of the dark branch, ∣vB∣2≃(Jcsin⁡θ/ΔBD)2|v_B|^2 \simeq(J_c \sin\theta/\Delta_{\mathrm{BD}})^2, which is maximal in the wall. The exact maximum over the wall (θ=π/2\theta= \pi/2, diagonal terms set to the bare splitting) is

∣vB∣max⁡2=12[1−ΔBDΔBD2+4Jc2].(33)|v_B|_{\max}^2 = \frac{1}{2}\left[1-\frac{\Delta_{\mathrm{BD}}}{\sqrt{\Delta_{\mathrm{BD}}^2+4J_c^2}}\right]. \tag*{(33)}

Three consequences follow directly from Eq. (30). First, the diagonal terms form a scalar landscape that is odd in the valley index: a skyrmion creates a region in its core where the valley ordering of the bright exciton is reversed (Supplementary Fig. S4a). Second, the off-diagonal term mixes the dark exciton into the bright one, and vice versa, where the magnetization lies in plane. For Jc≪∣ΔBD∣J_c \ll|\Delta_{\mathrm{BD}}|, the dark exciton acquires a bright weight of about (Jcsin⁡θ/ΔBD)2(J_c\sin\theta/\Delta_{\mathrm{BD}})^2, which is largest in the wall and gives a ring of brightened dark excitons (Supplementary Fig. S4b,d). Third, the phase e−iτνφe^{-i\tau\nu\varphi} imposes the selection rule ℓD=ℓB+τν\ell_D = \ell_B + \tau\nu on the angular momenta of the coupled channels.

Averaging over the internal motion of the exciton. For skyrmions comparable in size to the exciton, the electron at R+βr\mathbf{R}+\beta\mathbf{r} and the hole at R−αr\mathbf{R}-\alpha\mathbf{r} see different parts of the texture. The fields in Eq. (29) are then

nˉe(R)=∫d2r ∣ϕ1s(r)∣2n^(R+βr),nˉzh(R)=∫d2r ∣ϕ1s(r)∣2nz(R−αr),(34)\bar{\mathbf{n}}^{e}(\mathbf{R})=\int d^2r\,|\phi_{1s}(\mathbf{r})|^2\hat{\mathbf{n}}(\mathbf{R}+\beta\mathbf{r}),\qquad\bar{n}^{h}_{z}(\mathbf{R})=\int d^2r\,|\phi_{1s}(\mathbf{r})|^2 n_z(\mathbf{R}-\alpha\mathbf{r}), \tag*{(34)}

with the two-dimensional 1s density ∣ϕ1s∣2∝e−4r/aB|\phi_{1s}|^2\propto e^{-4r/a_B}. Equivalently, nˉe\bar{\mathbf{n}}^e is the texture convolved with e−4∣s∣/(βaB)e^{-4|\mathbf{s}|/(\beta a_B)} and nˉzh\bar{n}^h_z with e−4∣s∣/(αaB)e^{-4|\mathbf{s}|/(\alpha a_B)} (normalized kernels). The local eigenvalues and eigenvectors of Eq. (29) with these fields give the maps of Supplementary Figs. S4 and S6.

The adiabatic exciton. Diagonalizing Eq. (30) at each RR (at φ=0\varphi=0) and following the eigenvector v(R)=(vB,vD)v(R)=(v_B,v_D) that is bright at infinity, the projection of Supplementary Note 1 applied to the pseudospin gives the single-channel radial problem

−u′′+[(ℓ−A(R))2−14R2+Ebr(R)−Ebr(∞)+W(R)−εtM]u=0,(35)-u^{\prime\prime}+\left[\frac{(\ell-A(R))^2-\frac{1}{4}}{R^2}+\frac{E_{\mathrm{br}}(R)-E_{\mathrm{br}}(\infty)+W(R)-\varepsilon}{t_M}\right]u=0, \tag*{(35)}

with the Berry potential (flux quanta enclosed within RR), in the gauge regular at the origin,

Areg(R)=−τν[∣vD(R)∣2−∣vD(0)∣2],(36)A_{\mathrm{reg}}(R)=-\tau\nu\left[|v_D(R)|^2-|v_D(0)|^2\right], \tag*{(36)}

and in the gauge regular at infinity, used in the calculations, A(R)=Areg(R)−Areg(∞)A(R)=A_{\mathrm{reg}}(R)-A_{\mathrm{reg}}(\infty). The geometric potential is

W(R)=tM[∣dvdR∣2+ν2∣vB∣2∣vD∣2R2].(37)W(R)=t_M\left[\left|\frac{dv}{dR}\right|^2+\nu^2\frac{|v_B|^2|v_D|^2}{R^2}\right]. \tag*{(37)}

The net flux of the branch is Φ/Φ0=Areg(∞)=τν∣vD(0)∣2\Phi/\Phi_0=A_{\mathrm{reg}}(\infty)=\tau\nu|v_D(0)|^2, i.e., 00 below the threshold of Eq. (32) and τν\tau\nu above it.

For a texture that the exciton follows completely (∣vD∣2=sin⁡2(θ/2)|v_D|^2=\sin^2(\theta/2) along the profile), these expressions reduce to

A(R)=ν2[cos⁡θ(R)−cos⁡θ(∞)],W(R)=tM4[θ′2+ν2sin⁡2θR2],(38)A(R)=\frac{\nu}{2}\left[\cos\theta(R)-\cos\theta(\infty)\right],\qquad W(R)=\frac{t_M}{4}\left[\theta^{\prime2}+\nu^2\frac{\sin^2\theta}{R^2}\right], \tag*{(38)}

and the effective radial potential becomes

Uℓ(R)=tM(ℓ−A(R))2−14R2+W(R)+V(R),V(R)=−V01−cos⁡θ(R)2,(39)U_\ell(R)=t_M\frac{(\ell-A(R))^2-\frac{1}{4}}{R^2}+W(R)+V(R),\qquad V(R)=-V_0\frac{1-\cos\theta(R)}{2}, \tag*{(39)}

where VV is an optional exchange well of depth V0V_0 in the core. In the core A=−νA=-\nu, so the channel ℓ=−ν\ell=-\nu is s-like inside and meets the centrifugal barrier and WW outside.

Exact two-channel scattering

Radial equations. With ψB\psi_B and ψD\psi_D as above and mB=ℓm_B=\ell, mD=ℓ+τνm_D=\ell+\tau\nu, the Schrödinger equation for −tM∇2+Hτ-t_M\nabla^2+H_\tau at energy EE becomes

uB′′=[mB2−14R2+EB(R)−EtM]uB+Jcsin⁡θtMuD,(40)u_B^{\prime\prime}=\left[\frac{m_B^2-\frac{1}{4}}{R^2}+\frac{E_B(R)-E}{t_M}\right]u_B+\frac{J_c\sin\theta}{t_M}u_D, \tag*{(40)}
uD′′=[mD2−14R2+ED(R)−EtM]uD+Jcsin⁡θtMuB,(41)u_D^{\prime\prime}=\left[\frac{m_D^2-\frac{1}{4}}{R^2}+\frac{E_D(R)-E}{t_M}\right]u_D+\frac{J_c\sin\theta}{t_M}u_B, \tag*{(41)}

with EB=τ(Jc−Jv)cos⁡θE_B=\tau(J_c-J_v)\cos\theta and ED=−ΔBD−τ(Jc+Jv)cos⁡θE_D=-\Delta_{\mathrm{BD}}-\tau(J_c+J_v)\cos\theta. The channel momenta are kc=[(E−Ec(∞))/tM]1/2k_c=[(E-E_c(\infty))/t_M]^{1/2}, with EB(∞)=τ(Jc−Jv)E_B(\infty)=\tau(J_c-J_v) and ED(∞)=−ΔBD−τ(Jc+Jv)E_D(\infty)=-\Delta_{\mathrm{BD}}-\tau(J_c+J_v). The incident bright exciton has kinetic energy ε\varepsilon, i.e., E=τ(Jc−Jv)+εE=\tau(J_c-J_v)+\varepsilon, and we use ε=kBT=0.36\varepsilon=k_BT=0.36 meV unless stated otherwise.

S matrix. Two regular solutions, uc∝R∣mc∣+1/2u_c\propto R^{|m_c|+1/2} near the origin in one channel and zero in the other, are integrated outward. At large RR, each component is written as uc=AcRHmc(2)(kcR)+BcRHmc(1)(kcR)u_c=A_c\sqrt{R}H_{m_c}^{(2)}(k_cR)+B_c\sqrt{R}H_{m_c}^{(1)}(k_cR). Since RHm(1,2)(kR)\sqrt{R}H_{m}^{(1,2)}(kR) has an amplitude proportional to k−1/2k^{-1/2}, the outgoing flux in channel cc is proportional to ∣Bc∣2|B_c|^2 independently of kck_c. The matrix S=BA−1S=BA^{-1} is therefore flux-normalized and unitary for open channels; we checked max⁡∣S†S−1∣<10−3\max|S^\dagger S-1|<10^{-3}.

Cross-sections. The elastic and conversion amplitudes are

fBB(θs)=12πkB∑ℓ(SBBℓ−1)eiℓθs,fDB(θs)=12πkB∑ℓSDBℓei(ℓ+τν)θs,(42)f_{BB}(\theta_s)=\frac{1}{\sqrt{2\pi k_B}}\sum_\ell(S_{BB}^\ell-1)e^{i\ell\theta_s},\qquad f_{DB}(\theta_s)=\frac{1}{\sqrt{2\pi k_B}}\sum_\ell S_{DB}^\ell e^{i(\ell+\tau\nu)\theta_s}, \tag*{(42)}

and

σel=∫dθs ∣fBB∣2=1kB∑ℓ∣SBBℓ−1∣2,σB→D=1kB∑ℓ∣SDBℓ∣2.(43)\sigma_{\mathrm{el}}=\int d\theta_s\,|f_{BB}|^2=\frac{1}{k_B}\sum_\ell|S_{BB}^\ell-1|^2,\qquad\sigma_{B\to D}=\frac{1}{k_B}\sum_\ell|S_{DB}^\ell|^2. \tag*{(43)}

The skew and the left–right asymmetry are

⟨sin⁡θs⟩=∫dθs sin⁡θs ∣fBB∣2∫dθs ∣fBB∣2,γ(θs)=∣fBB(θs)∣2−∣fBB(−θs)∣2∣fBB(θs)∣2+∣fBB(−θs)∣2.(44)\langle\sin\theta_s\rangle=\frac{\int d\theta_s\,\sin\theta_s\,|f_{BB}|^2}{\int d\theta_s\,|f_{BB}|^2},\qquad\gamma(\theta_s)=\frac{|f_{BB}(\theta_s)|^2-|f_{BB}(-\theta_s)|^2}{|f_{BB}(\theta_s)|^2+|f_{BB}(-\theta_s)|^2}. \tag*{(44)}

In the adiabatic (gauge-field) description, the same formulas are used with a single channel, Sℓ=bℓ/aℓS^\ell=b_\ell/a_\ell from Eq. (35). For K′K' above the flux switch, the dark exciton lies above the bright one and the dark channel is closed at the energies considered (kD2<0k_D^2<0, i.e., Jc/ΔBD>0.54J_c/\Delta_{\mathrm{BD}}>0.54 for ε=kBT\varepsilon=k_BT). In this case, the two regular solutions are integrated outward to R=cR=c, while the decaying dark solution RK∣mD∣(κDR)\sqrt{R}K_{|m_D|}(\kappa_DR) and the two real bright solutions RJ∣mB∣(kBR)\sqrt{R}J_{|m_B|}(k_BR) and RY∣mB∣(kBR)\sqrt{R}Y_{|m_B|}(k_BR) are integrated inward from Rmax⁡R_{\max}, the stable direction for the closed channel, with Gram–Schmidt re-orthogonalization. Matching at R=cR=c gives SBBS_{BB} with ∣SBB∣=1|S_{BB}|=1 and σB→D=0\sigma_{B\to D}=0.

In the weak-coupling regime, the scattering is weak but chiral (Supplementary Fig. S3). The asymmetry reverses for an antiskyrmion and has opposite sign in the two valleys, it changes sign as a function of the skyrmion radius and of the kinetic energy, and it is accompanied by a substantial conversion of bright into dark excitons, which becomes comparable to elastic scattering for the smallest skyrmions.

Bound states in the valley-odd landscape. The diagonal of Eq. (30) also produces an attractive well in one valley when Jc≠JvJ_c\ne J_v. Its depth is 2∣Jc−Jv∣2|J_c-J_v| for a fully reversed core. We computed the binding energy of the lowest state as a function of the skyrmion radius and of ∣Jc−Jv∣|J_c-J_v| on a cell-centred finite-volume grid (Supplementary Fig. S4c). Binding above kBTk_BT at 4.2 K requires ∣Jc−Jv∣|J_c-J_v| of about 1 meV for a skyrmion radius of 5 nm, and about 0.4 meV for a radius of 15 nm. Larger skyrmions are therefore clearly favourable.

Figure S3. Chiral scattering in the weak-coupling limit (wall width 1 nm, Jc=1 meVJ_c = 1\,\mathrm{meV}, Jv=0.5 meVJ_v = 0.5\,\mathrm{meV}, ΔBD=5 meV\Delta_{\mathrm{BD}} = 5\,\mathrm{meV}, kinetic energy kBTk_{\mathrm{B}}T at 4.2 K unless stated). a, Left–right asymmetry of elastic scattering by a 10 nm skyrmion. b, Skew versus skyrmion radius; the sign reverses for antiskyrmions. c, Skew versus kinetic energy for radii of 3 and 10 nm. d, Ratio of bright-to-dark conversion to elastic scattering. Grey bands mark Pd/Fe/Ir(111).

Figure S4. Weak-coupling fingerprints of a single skyrmion (radius 5 nm, wall 1 nm, aB=1.5 nma_{\mathrm{B}} = 1.5\,\mathrm{nm}, Jc=1 meVJ_c = 1\,\mathrm{meV}, Jv=0.5 meVJ_v = 0.5\,\mathrm{meV}, ΔBD=5 meV\Delta_{\mathrm{BD}} = 5\,\mathrm{meV}). a, Local valley splitting of the bright exciton. b, Bright weight acquired by the dark exciton. c, Binding energy in the valley-odd well versus skyrmion radius and ∣Jc−Jv∣\lvert J_c - J_v\rvert; the white line marks kBTk_{\mathrm{B}}T at 4.2 K and the shaded strip Pd/Fe/Ir(111). d, Maximum bright weight of the dark exciton versus JcJ_c and ∣ΔBD∣\lvert\Delta_{\mathrm{BD}}\rvert, with 1% and 5% contours; the dashed line marks MoS2_2[38].

Pd/Fe/Ir(111) as an illustrative skyrmion host

Pd/Fe/Ir(111) is the system in which individual skyrmions were first written and deleted with a spin-polarized tip[12], and its micromagnetic parameters are well established[9]. We use it here for two purposes. The first is to test whether a skyrmion host keeps its skyrmions when it is covered by a van der Waals overlayer. To this end, we computed from first principles the magnetic interactions of Pd/Fe/Ir(111) without and with an overlayer of two hBN layers and a MoS2_2 monolayer. The isotropic exchange interactions JijJ_{ij} and the Dzyaloshinskii–Moriya interactions between the Fe atoms are essentially unchanged by the overlayer (Supplementary Fig. S5). Since these interactions determine the size and the stability of the skyrmions, we conclude that the skyrmions survive the van der Waals overlayer. We stress that we use these calculations only for this purpose. The relaxed geometry does not include van der Waals corrections, and it would not be reliable enough for quantitative values of the proximity exchange acting on MoS2_2.

The second purpose is to illustrate the small-size limit of our theory. Minimizing the micromagnetic functional with the parameters of ref. [9] (A=2.0A = 2.0 pJ/m, D=3.9D = 3.9 mJ/m2^2, K=2.5K = 2.5 MJ/m3^3, Ms=1.1M_s = 1.1 MA/m) reproduces the measured skyrmion sizes, with radii between about 1.9 and 0.8 nm for fields between 1 and 3 T (Supplementary Fig. S6a). These skyrmions are comparable in size to the exciton, so that the excitonic fingerprints of Supplementary Note 5 are averaged over the extent of the exciton. We find that the valley reversal in the core and the ring of brightened dark excitons keep between about 55 and 98% of their large-skyrmion values, depending on field and exciton radius (Supplementary Fig. S6b,c). Bright-to-dark conversion is comparable to elastic scattering in this regime (Supplementary Fig. S6d). Pd/Fe/Ir(111) is metallic, however, and a spacer is needed to protect the exciton. For the excitonic coupling itself, insulating skyrmion hosts in direct contact with the excitonic layer are therefore preferable, as discussed in the main text.

Micromagnetic profiles. The energy of a rotationally symmetric skyrmion in a film with exchange stiffness AA, interfacial DMI DD, uniaxial anisotropy KK, saturation magnetization MsM_s and field BB is, per unit thickness,

E[θ]=2π∫0∞ρ dρ[A(θ′2+sin⁡2θρ2)−D(θ′+sin⁡θcos⁡θρ)+Ksin⁡2θ−MsB(cos⁡θ−1)](45)E[\theta] = 2\pi\int_{0}^{\infty} \rho\,\mathrm{d}\rho\left[ A\left(\theta'^{2} + \frac{\sin^{2}\theta}{\rho^{2}}\right) - D\left(\theta' + \frac{\sin\theta\cos\theta}{\rho}\right) + K\sin^{2}\theta- M_sB(\cos\theta- 1) \right] \tag*{(45)}

measured from the ferromagnetic state. We minimize it within the profile θ(ρ)=π+∑±arcsin⁡tanh⁡[(ρ±c)/(wR/2)]\theta(\rho) = \pi+ \sum_{\pm} \arcsin\tanh[(\rho\pm c)/(w_R/2)] with respect to cc and the wall width wRw_R, using A=2.0A = 2.0 pJ/m, D=3.9D = 3.9 mJ/m2^2, K=2.5K = 2.5 MJ/m3^3 and Ms=1.1M_s = 1.1 MA/m. The skyrmion diameter is twice the radius at which nz=0n_z = 0. In the model of Eq. (10), this profile corresponds to ω=wR/2\omega= w_R/2.

The quantum-geometric trap

In the main text we showed, using control calculations, that the barrier of the trap originates from the quantum metric, while the topology shifts the angular momentum of the trapped state. Here we summarize how the trap depends on the topological charge and on the size of the skyrmion, and what happens when an attractive exchange well is added.

Phase shifts. For V0=0V_0 = 0, Eq. (39) is integrated outward from u∝R∣ℓ−A(0)∣+1/2u \propto R^{|\ell-A(0)|+1/2} and matched at Rmax⁡R_{\max} to R [aJℓ(kR)+bYℓ(kR)]\sqrt{R}\,[a J_{\ell}(kR) + b Y_{\ell}(kR)], k=(E/tM)1/2k = (E/t_M)^{1/2}. Because A(R→∞)=0A(R \to\infty) = 0 in this gauge, the asymptotic channel is free with angular momentum ℓ\ell. The phase shift is δℓ=arg⁡(a−ib)\delta_{\ell} = \arg(a - ib), i.e., tan⁡δℓ=−b/a\tan\delta_{\ell} = -b/a.

Time delay and lifetime. Resonances are identified from the Wigner time delay

τW(E)=2ℏdδℓdE(46)\tau_W(E) = 2\hbar\frac{\mathrm{d}\delta_{\ell}}{\mathrm{d}E} \tag*{(46)}
Two plots showing isotropic exchange $J_{ij}$ and Dzyaloshinskii–Moriya interaction magnitude versus distance for bare Pd/Fe/Ir(111) and Pd/Fe/Ir(111) covered by two hBN layers and a MoS$_2$ monolayer.

Figure S5. Skyrmion-stabilizing interactions survive a van der Waals overlayer. a, Isotropic exchange JijJ_{ij} and b, magnitude of the Dzyaloshinskii–Moriya interaction between Fe atoms versus distance, for bare Pd/Fe/Ir(111) (blue) and for Pd/Fe/Ir(111) covered by two hBN layers and a MoS2_2 monolayer (red), from first-principles KKR calculations.

Figure S6. Excitonic fingerprints of nanometric skyrmions: the example of Pd/Fe/Ir(111). a, Skyrmion diameter, radius and wall width versus field from the micromagnetic functional with the parameters of ref. 9, reproducing the measured sizes. b, Core valley reversal and ring weight relative to their large-skyrmion values, for exciton radii of 1.0 and 1.5 nm. c, Valley-splitting map at 1.15 T with contours of dark-exciton bright weight (1, 2 and 3%). d, Ratio of bright-to-dark conversion to elastic scattering versus field. Parameters as in Supplementary Fig. S4.

with width Γ=4ℏ/τWmax⁡\Gamma= 4\hbar/\tau_W^{\max}, lifetime ℏ/Γ=τWmax⁡/4\hbar/\Gamma= \tau_W^{\max}/4 and quality factor Q=Er/ΓQ = E_r/\Gamma.

Control calculations. (i) “No flux”: A≡0\mathbf{A} \equiv0, keeping WW. (ii) Skyrmionium: θ(R)=π2[2+tanh⁡R−c1w+tanh⁡R−c2w]\theta(R) = \frac{\pi}{2}\left[2 + \tanh\frac{R-c_1}{w} + \tanh\frac{R-c_2}{w}\right], which winds from 0 through π\pi back to 2π2\pi and has zero net flux. (iii) Linear wall: θ(R)=π+πclip⁡(R−(c−w)2w,0,1)\theta(R) = \pi+ \pi\operatorname{clip}\left(\frac{R-(c-w)}{2w},0,1\right).

Bound states. With V0>0V_0 > 0, the radial equation for ψ(R)\psi(R) is discretized on cell centres Rj=(j−12)hR_j = (j-\frac{1}{2})h, j=1,…,Nj=1,\ldots,N, with Rj±1/2=Rj±h/2R_{j\pm1/2}=R_j\pm h/2. The finite-volume Laplacian gives a tridiagonal symmetric matrix with

Hjj=tMRj+1/2+Rj−1/2h2Rj+tM(ℓ−A(Rj))2Rj2+W(Rj)+V(Rj),Hj,j+1=−tMRj+1/2h2RjRj+1(47)H_{jj}=t_M\frac{R_{j+1/2}+R_{j-1/2}}{h^2R_j}+t_M\frac{(\ell-A(R_j))^2}{R_j^2}+W(R_j)+V(R_j),\qquad H_{j,j+1}=-t_M\frac{R_{j+1/2}}{h^2\sqrt{R_jR_{j+1}}} \tag*{(47)}

whose lowest eigenvalues are the bound-state energies (Rmax⁡=300R_{\max}=300–400 nm400\ \mathrm{nm}, N=3N=3–4×1044\times10^4).

Optical weight. A near-field spot carrying orbital angular momentum mm couples only to ℓ=m\ell=m. The light acts on the bright component of the local exciton, vB2=(1+cos⁡θ)/2v_B^2=(1+\cos\theta)/2 for a fully winding electron and pinned hole, which vanishes in the core. With a Gaussian spot gm(R)=(R/s)∣m∣e−R2/2s2g_m(R)=(R/s)^{|m|}e^{-R^2/2s^2}, s=30 nms=30\ \mathrm{nm}, and the scattering solution uEu_E normalized to unit asymptotic amplitude, the optical weight is

Sm(E)=1k[∫0∞dRRgm(R)vB(R)uE(R)]2.(48)S_m(E)=\frac{1}{k}\left[\int_0^\infty dR\sqrt{R}g_m(R)v_B(R)u_E(R)\right]^2. \tag*{(48)}

The factor 1/k1/k converts the energy-independent asymptotic normalization into the density of states of the free channel.

Within this family of 360°-wall profiles, the lifetime of the ss-like resonance increases with the topological charge, from 9 ps for ν=1\nu=1 to 38 ps for ν=3\nu=3 at a radius of 20 nm, with quality factors between about 6 and 25 (Supplementary Fig. S7a). It also increases strongly with the size of the skyrmion, reaching hundreds of picoseconds for radii of 40 nm when the wall width is kept fixed (Supplementary Fig. S7b). These trends are useful for design, but we emphasize that they depend on the profile and are not protected by topology. When an exchange well of 2 meV is added, the resonances become bound states. The ground state then carries ℓ=−ν\ell=-\nu for all charges considered, and the two excited states with ℓ=−ν±1\ell=-\nu\pm1, which would be degenerate in an ordinary circular trap, are split by the emergent field (Supplementary Fig. S7c).

Transport and single-exciton occupation

Galilean invariance. For a trap moving at constant velocity, H=P2/2M+U(R−vt)H=\mathbf{P}^2/2M+U(\mathbf{R}-\mathbf{v}t), the function

ψ(R,t)=exp⁡[iℏ(Mv⋅R−12Mv2t)−iℏE0t]ψ0(R−vt)(49)\psi(\mathbf{R},t)=\exp\left[\frac{i}{\hbar}\left(M\mathbf{v}\cdot\mathbf{R}-\frac{1}{2}Mv^2t\right)-\frac{i}{\hbar}E_0t\right]\psi_0(\mathbf{R}-\mathbf{v}t) \tag*{(49)}

solves the time-dependent Schrödinger equation exactly if ψ0\psi_0 is a bound state of the trap at rest. A trap moving at constant velocity therefore carries its bound state without loss.

Sudden boost. If the velocity changes instantaneously by Δv\Delta\mathbf{v}, the wavefunction is unchanged in the frame at rest but acquires the phase eiMΔv⋅R/ℏe^{iM\Delta\mathbf{v}\cdot\mathbf{R}/\hbar} in the new frame. The probability of remaining bound is P=∣⟨ψ0∣eiq⋅r∣ψ0⟩∣2P=\left|\langle\psi_0|e^{i\mathbf{q}\cdot\mathbf{r}}|\psi_0\rangle\right|^2 with q=MΔv/ℏ\mathbf{q}=M\Delta\mathbf{v}/\hbar. For an ss state, the angular integral gives a Bessel function,

P(Δv)=∣∫0∞2πR dR ρ0(R)J0(qR)∣2,ρ0=∣ψ0∣2.(50)P(\Delta v)=\left|\int_0^\infty2\pi R\,dR\,\rho_0(R)J_0(qR)\right|^2,\qquad\rho_0=|\psi_0|^2. \tag*{(50)}

Any finite acceleration loses less than this bound.

Tunnelling under acceleration. An acceleration aa adds a force F=MaF = Ma in the comoving frame. For a triangular barrier, the WKB exponent is 43κEb/F\frac{4}{3}\kappa E_b/F with κ=(Eb/tM)1/2\kappa= (E_b/t_M)^{1/2}, so tunnelling becomes appreciable only for a≳a∗=43κEb/Ma \gtrsim a^* = \frac{4}{3}\kappa E_b/M, of order 1017 m s−210^{17}\ \mathrm{m\,s^{-2}} for the parameters of Figure 5.

Thermal escape and range. The escape time is estimated from an Arrhenius law with an attempt frequency given by the level spacing of the trap,

τesc=ν0−1eEb/kBT,ν0=E1−E0h,Eb=−E0,(51)\tau_{\mathrm{esc}} = \nu_0^{-1}e^{E_b/k_BT}, \qquad\nu_0 = \frac{E_1-E_0}{h}, \qquad E_b=-E_0, \tag*{(51)}

where E0E_0 and E1E_1 are the two lowest ℓ=0\ell=0 levels. The transport distance is L=vmin⁡(τesc,τX)L=v\min(\tau_{\mathrm{esc}},\tau_X), with τX\tau_X the exciton lifetime.

Two excitons in one trap. Interlayer excitons are dipoles of length dd perpendicular to the plane. In the Hartree approximation, two excitons in the ground state of the trap interact with energy

Udip=⟨e24πε0ε(2s−2s2+d2)⟩,s=∣R1−R2∣,(52)U_{\mathrm{dip}} = \left\langle\frac{e^2}{4\pi\varepsilon_0\varepsilon}\left(\frac{2}{s}-\frac{2}{\sqrt{s^2+d^2}}\right)\right\rangle, \qquad s=\lvert\mathbf{R}_1-\mathbf{R}_2\rvert, \tag*{(52)}

averaged over ∣ψ0(R1)∣2∣ψ0(R2)∣2\lvert\psi_0(\mathbf{R}_1)\rvert^2\lvert\psi_0(\mathbf{R}_2)\rvert^2 by Monte Carlo sampling, with d=0.65 nmd=0.65\ \mathrm{nm} and ε=4.5\varepsilon=4.5. The trap accepts a single exciton when Udip>EbU_{\mathrm{dip}}>E_b.

Results for the parameters of main text Figure 5. At 100 m/s the loss bound of Eq. (50) is at most about 10−410^{-4}, reached for a skyrmion radius of 40 nm, and it becomes of order unity only for velocity changes of about 10 km/s.

For a binding energy of 1.6 meV (V0=2V_0=2 meV, radius 20 nm), the escape-limited distance at 100 m/s is about 50 μ\mum at 1.5 K, 2.4 μ\mum at 2 K and only 0.02 μ\mum at 4.2 K. For 4.5 meV (V0=5V_0=5 meV), it is about 47 μ\mum at 4.2 K, 6 μ\mum at 5 K and 1 μ\mum at 6 K. Since interlayer excitons live for 10–100 ns, corresponding to 1–10 μ\mum at this velocity, thermal escape stops being the limiting factor below about 5 K for the deeper well. These numbers motivate the design rule Eb≳10 kBTE_b \gtrsim10\,k_BT. For radii below about 25–27 nm the dipolar repulsion of Eq. (52) exceeds the binding energy; the crossover radius depends on dd and on the dielectric environment. In the full model, the binding energy for V0=2V_0=2 meV falls below kBTk_BT at 4.2 K for radii below about 12 nm, so that single occupancy is stable between about 12 and 27 nm.

Reference model and full model. Figure 5 of the main text compares a reference model, in which only the exchange well VV is kept (A=W=0A=W=0, ℓ=0\ell=0), with the full model of Eq. (39), which includes the Berry potential and the geometric barrier (ν=1\nu=1, ground state ℓ=−1\ell=-1). In the full model, the attempt frequency is taken from the two lowest levels of the ℓ=−1\ell=-1 channel. Supplementary Table S1 summarizes the binding energies. For radii of 20 nm and above, the two models differ by at most 10%, and the escape times by less than a factor of two. For small skyrmions, the geometric barrier strongly reduces the binding: for V0=2V_0=2 meV, the exciton is no longer bound at a radius of 8 nm, and only by 0.06 meV at 10 nm. In the full model, the moving texture also produces an emergent electric field. In the frame moving with the skyrmion, it adds the static term −ℏv⋅a-\hbar\mathbf{v}\cdot\mathbf{a}, of order hv/R≈3 μeVh v/R \approx3\ \mu\mathrm{eV} for v=100v=100 m/s and R=20R=20 nm, which is negligible compared with the binding energy.

Masses. For interlayer excitons we use the first-principles band masses of ref. [39]. Magneto-optical experiments indicate heavier electrons[68], which would increase the exciton mass (Supplementary Table S3). A heavier exciton is bound more strongly in a given trap, so that our estimates of binding and transport are conservative in this respect.

Radius (nm)V0V_0 (meV)ReferenceFull model
1020.860.06
1053.392.81
2021.571.41
2054.484.39
4021.871.84
4054.854.84

Table S1. Binding energy EbE_b (meV) of an interlayer exciton (M=0.83 m0M = 0.83\,m_0, wall 1.5 nm) in the reference model and in the full model.

2Δc2\Delta_c (meV)2Δv2\Delta_v (meV)ΔBD\Delta_{BD} (meV), exp.mem_e (m0m_0)mhm_h (m0m_0)μ\mu (m0m_0), exp.r1sr_{1s} (nm), exp.
MoS2_23148+140.43–0.460.540.2751.2
MoSe2_222186−1.40.49–0.560.590.3501.1
MoTe2_236219< 00.53–0.620.660.361.3
WS2_2−32429+40–550.26–0.350.350.1751.8
WSe2_2−37466+400.28–0.390.360.201.7

Table S2. Spin–orbit splittings and band masses at the KK point from first-principles k⋅p\mathbf{k}\cdot\mathbf{p} fits (HSE/LDA, ref. 39), and exciton parameters of hBN-encapsulated monolayers from magneto-optics (refs. 38;68; WSe2_2 from ref. 27 of ref. 68). ΔBD>0\Delta_{BD} > 0 means the dark exciton lies below the bright one; for MoTe2_2 the dark exciton was not observed and is presumably above. The conduction-band splitting has opposite sign in W-based compounds, where the lowest-energy exciton is dark. Δ\Delta in the main text is half the splitting.

Material parameters

Supplementary Table S2 collects the parameters of the transition-metal dichalcogenides used in this work, and Supplementary Table S3 those of the interlayer excitons used in the transport estimates. Two points deserve attention. First, the threshold for the flux switch of intralayer excitons is set by the measured bright–dark splitting ΔBD\Delta_{BD}, which contains the electron–hole exchange, and not by the bare conduction-band splitting. The two can differ strongly, and even in sign, as the comparison of MoS2_2 and MoSe2_2 shows. Second, for interlayer excitons the electron–hole exchange is weak, and the threshold is close to the bare conduction-band splitting of the layer facing the magnet. In the model calculations of the main text we use a representative ΔBD=5\Delta_{BD} = 5 meV, between the measured values for MoSe2_2 and MoS2_2. Since the weight of the dark-exciton ring scales as (Jc/ΔBD)2(J_c/\Delta_{BD})^2, it is weaker in MoS2_2 and stronger in MoSe2_2 than in these calculations.

Equations and parameters behind each figure

For each panel we give the equation that was evaluated and the parameters. Unless stated otherwise: Jc=1J_c = 1 meV, Jv=0.5J_v = 0.5 meV, ΔBD=5\Delta_{BD} = 5 meV, ν=1\nu= 1, ε=0.36\varepsilon= 0.36 meV and the profile of Eq. (10).

Figure 1.

  • b, shaded regions: binding energy Eb(J,R)E_b(J,R) of the lowest ℓ=0\ell= 0 state of U0U_0 in Eq. (39) with A=W=0A = W = 0, V0=2JV_0 = 2J, c=Rc = R, w=R/5w = R/5, M=m0M = m_0, computed with the finite-volume method of Supplementary Note 7 on a 34×3434 \times34 grid in J∈[0.05,100]J \in[0.05,100] meV and R∈[2,200]R \in[2,200] nm. Light: Eb>kBTE_b > k_BT; dark: Eb>10kBTE_b > 10 k_BT (4.2 K). The solid lines are the two contours.

  • b, vertical lines: adiabatic onset J=ℏ2/(2m0ω2)J = \hbar^2/(2m_0\omega^2) for w=1.5w = 1.5 and 5 nm (Supplementary Note 2).

  • b, dash-dotted lines: radii at which the lifetime ℏ/Γ\hbar/\Gamma of the ℓ=−1\ell= -1 resonance reaches 10 and 100 ps for a fixed wall w=1.5w = 1.5 nm, by log-interpolation of the values of Supplementary Figure S7

Figure S7. Supplementary Figure S7: Trends within the family of 360∘360^\circ-wall profiles. a, Lifetime and quality factor of the s-like trapped mode versus topological charge (radius 20 nm, wall 2 nm). b, Lifetime versus radius for self-similar skyrmions (wall width ∝R\propto R) and for a fixed wall width of 1.5 nm. c, Bound states in an additional exchange well of 2 meV (radius 20 nm, wall 1.5 nm), labelled by ℓ\ell; the ground state carries ℓ=−ν\ell=-\nu. Energies and lifetimes depend on the wall profile (main text Figure 4c).

(Supplementary Table S3: Interlayer excitons used in the transport estimates. Total masses from the band masses of Supplementary Table S2 (first-principles) and, in brackets, with the heavier electron masses inferred from magneto-optics [68]. Threshold Δc\Delta_c is that of the electron layer facing the magnet.

Heterobilayerelectron / hole layerMM (m0m_0)Δc\Delta_c (meV)
MoS2_2/WSe2_2MoS2_2 / WSe2_20.83 (1.06)1.5
MoSe2_2/WSe2_2MoSe2_2 / WSe2_20.85 (1.2)11
WS2_2/WSe2_2WS2_2 / WSe2_20.6216

Table S3.

  • b, bars: indicative exchange ranges (proximity in direct contact 1–5 meV; magnetic semiconductors 50–100 meV).

  • c: J∗=∣ΔBD∣/2J^* = \lvert\Delta_{\mathrm{BD}}\rvert/2 for intralayer excitons (Eq. (32)) with the measured ΔBD\Delta_{\mathrm{BD}} of Supplementary Table S2; J∗=ΔcJ^* = \Delta_c of the electron layer for interlayer excitons (Supplementary Table S3).

Figure 2.

  • b: schematic of Eq. (28): path of b^i\hat{\mathbf{b}}_i for J<ΔJ < \Delta and J>ΔJ > \Delta.

  • c: eigenvalues of Eq. (30) versus RR for τ=±1\tau= \pm1; c=5c = 5 nm, w=1w = 1 nm. “Bright” and “dark” label the branches by their character outside the skyrmion.

Figure 3. Parameters: c=10c = 10 nm, w=4w = 4 nm, Jv=0.5J_v = 0.5 meV, ΔBD=5\Delta_{\mathrm{BD}} = 5 meV, M=1.01m0M = 1.01m_0, ϵ=0.36\epsilon= 0.36 meV, partial waves ∣ℓ∣≤8\lvert\ell\rvert\le8.

  • a: net flux Φ/Φ0=Areg(∞)\Phi/\Phi_0 = A_{\mathrm{reg}}(\infty) of the bright-connected branch (Supplementary Note 5) for τ=±1\tau= \pm1, at 400 values of Jc/ΔBD∈[0.04,12]J_c/\Delta_{\mathrm{BD}} \in[0.04, 12]. Shaded: dark core, ∣vD(0)∣2=1\lvert v_D(0)\rvert^2 = 1.

  • b: skew ⟨sin⁡θs⟩\langle\sin\theta_s\rangle of ∣fBB∣2\lvert f_{\mathrm{BB}}\rvert^2 (Supplementary Note 5), τ=±1\tau= \pm1, from the exact two-channel solution (circles; closed-channel matching for K′K' above Jc/ΔBD=0.54J_c/\Delta_{\mathrm{BD}} = 0.54) and from the single-channel adiabatic problem, Eq. (35) (dashed), at 64 (KK) and 161 (K′K') values of Jc/ΔBD∈[0.05,10]J_c/\Delta_{\mathrm{BD}} \in[0.05, 10]. Increasing the number of partial waves to ∣ℓ∣≤12\lvert\ell\rvert\le12 or the matching radius from c+22wc + 22w to c+30wc + 30w leaves all curves unchanged (Source Data).

  • c: σel\sigma_{\mathrm{el}} from both solutions.

  • d: σB→D/σel\sigma_{\mathrm{B}\to\mathrm{D}}/\sigma_{\mathrm{el}} of the exact solution.

Figure 4. Parameters: ν=1\nu= 1, c=20c = 20 nm, w=2w = 2 nm, M=1.01m0M = 1.01m_0, V0=0V_0 = 0, fully winding exciton, Eq. (38).

  • a: W(R)W(R) (grey) and tM(ℓ−A)2/R2+Wt_M(\ell-A)^2/R^2 + W for ℓ=−1,0,+1\ell= -1, 0, +1 (the two-dimensional form of Eq. (39), without the −14-\frac{1}{4} of the radial equation). Dotted line: resonance energy 0.43 meV.

  • b: τW(E)\tau_W(E), Eq. (46), for 1571 energies in [0.03,1.6][0.03, 1.6] meV, from the radial equation integrated to Rmax⁡=120R_{\max} = 120 nm: skyrmion, ℓ=−1\ell= -1; no-flux control (A=0A = 0), ℓ=0\ell= 0; skyrmionium with c1=10c_1 = 10, c2=20c_2 = 20 nm, w=2w = 2 nm, ℓ=0\ell= 0 and ℓ=−1\ell= -1.

  • c: τW(E)\tau_W(E) for ℓ=−1\ell= -1 with w=1w = 1, 2 and 4 nm, and for the linear wall with w=2w = 2 nm. Resonances: 0.48, 0.43 and 0.34 meV with lifetimes τWmax⁡/4=28\tau_W^{\max}/4 = 28, 9.0 and 3.4 ps; linear wall 0.51 meV, 49 ps.

  • d: Sm(E)/Sm(Emin⁡)S_m(E)/S_m(E_{\min}) from Eq. (48) for m=−1,0,+1m = -1, 0, +1, s=30s = 30 nm, Emin⁡=0.03E_{\min} = 0.03 meV, 600 energies, with uEu_E normalized to unit asymptotic amplitude of uu. An independent calculation from the eigenstates of a large box (Rmax⁡=700R_{\max} = 700 nm), Sm(En)∝∣⟨gmvB∣n⟩∣2/ΔEnS_m(E_n) \propto\lvert\langle g_m v_B \vert n\rangle\rvert^2/\Delta E_n, reproduces the curves to within 10% and the positions of all zeros.

Figure 5. Parameters: interlayer exciton, M=0.83m0M = 0.83m_0, exchange well V0=2V_0 = 2 meV, wall w=1.5w = 1.5 nm, c=Rc = R. Solid lines: reference model, A=W=0A = W = 0, ℓ=0\ell= 0. Dashed lines: full model, Eq. (39) with AA and WW from Eq. (38), ν=1\nu= 1, ground state ℓ=−1\ell= -1, attempt frequency from the two lowest ℓ=−1\ell= -1 levels.

  • a: 1−P(Δv)1-P(\Delta v) from Eq. (50) for R=10R=10, 20 and 40 nm, with the ground state from the finite-volume solver (Rmax⁡=300R_{\max}=300 nm, N=3×104N=3\times10^{4}).

  • b: L=vτesc(T)L=v\tau_{\mathrm{esc}}(T) with v=100v=100 m/s and Eq. (51), for R=20R=20 nm and V0=2V_{0}=2 and 5 meV, which give Eb=1.6E_{b}=1.6 and 4.5 meV. Bands: vτXv\tau_{X} for dark intralayer excitons (τX=0.1\tau_{X}=0.1–1 ns) and interlayer excitons (τX=10\tau_{X}=10–100 ns).

  • c: UdipU_{\mathrm{dip}} from Eq. (52) (1.5×1051.5\times10^{5} samples) and EbE_{b}, for R=8R=8, 10, 14, 20, 28, 40 and 55 nm. Dotted line: kBTk_{B}T at 4.2 K. Shaded: full-model radii with Eb>kBTE_{b}>k_{B}T and Udip>EbU_{\mathrm{dip}}>E_{b} (single occupancy).

Supplementary Figures

  • S1: GWGW–BSE data plotted directly. (a) Im⁡α(ω)\operatorname{Im}\alpha(\omega) with (BSE) and without (independent quasiparticles) electron–hole interaction, normalized to the maximum in the plotted range; quasiparticle gap =min⁡k[E27QP(k)−E26QP(k)]=\min_{k}[E_{27}^{\mathrm{QP}}(k)-E_{26}^{\mathrm{QP}}(k)]. (b) Oscillator strengths of the lowest states, energies relative to the bright A exciton. (c) Exciton density; no model.

  • S2: structureless spinor model, H=(P−ℏa)2/2μX+WH=(\mathbf{P}-\hbar\mathbf{a})^{2}/2\mu_{X}+W, with μX=5.6 m0\mu_{X}=5.6\,m_{0}, Θ(r)=2arctan⁡[cosh⁡(c/(w/2))/sinh⁡(r/(w/2))]\Theta(r)=2\arctan[\cosh(c/(w/2))/\sinh(r/(w/2))], c=4.9c=4.9 nm, w=1.0w=1.0 nm, A=−(1−cos⁡Θ)/2A=-(1-\cos\Theta)/2, W=tX4[Θ′2+sin⁡2Θ/r2]W=\frac{t_{X}}{4}[\Theta^{\prime2}+\sin^{2}\Theta/r^{2}], k=3.3k=3.3, 9.9 and 16.5 μm−1\mu\mathrm{m}^{-1}, ∣ℓ∣≤6|\ell|\leq6. (a) ∣f(θs)∣2=∣∑ℓ(Sℓ−1)eiℓθs∣2/(2πk)|f(\theta_{s})|^{2}=|\sum_{\ell}(S_{\ell}-1)e^{i\ell\theta_{s}}|^{2}/(2\pi k); dotted: 4/(kθs2)4/(k\theta_{s}^{2}). (b) γ(θs)\gamma(\theta_{s}).

  • S3: exact two-channel solution of Supplementary Note 5 with w=1w=1 nm. (a) γ(θs)\gamma(\theta_{s}) for c=10c=10 nm, τ=±1\tau=\pm1. (b) Skew versus c=1,…,14c=1,\ldots,14 nm. (c) Skew versus ε=0.1\varepsilon=0.1–3 meV for c=3c=3 and 10 nm. (d) σB→D/σel\sigma_{B\to D}/\sigma_{\mathrm{el}} versus cc.

  • S4: (a) EK−EK′E_{K}-E_{K^{\prime}} of the bright branch and (b) bright weight of the dark branch in KK, from Eq. (29) with the averaged fields of Supplementary Note 5 (c=5c=5 nm, w=1w=1 nm, aB=1.5a_{B}=1.5 nm, me=0.47m_{e}=0.47, mh=0.54m_{h}=0.54). (c) EbE_{b} of the lowest ℓ=0\ell=0 state in V=−2∣Jc−Jv∣(1−cos⁡θ)/2V=-2|J_{c}-J_{v}|(1-\cos\theta)/2, w=1w=1 nm, M=1.01 m0M=1.01\,m_{0}. (d) Eq. (33).

  • S5: first-principles JijJ_{ij} and ∣Dij∣|\mathbf{D}_{ij}| of the Heisenberg model H=−∑ijJijSi⋅Sj−∑ijDij⋅(Si×Sj)H=-\sum_{ij}J_{ij}\mathbf{S}_{i}\cdot\mathbf{S}_{j}-\sum_{ij}\mathbf{D}_{ij}\cdot(\mathbf{S}_{i}\times\mathbf{S}_{j}) from KKR; no model equation.

  • S6: (a) Supplementary Note 6. (b) Core valley reversal −[EK−EK′](0)/[2(Jc−Jv)]-[E_{K}-E_{K^{\prime}}](0)/[2(J_{c}-J_{v})] and maximal ring weight, both relative to a large skyrmion (c=20c=20 nm, w=0.75w=0.75 nm), for aB=1.0a_{B}=1.0 and 1.5 nm, using the profiles of (a) with w=wR/2w=w_{R}/2. (c) Map at 1.15 T, aB=1.5a_{B}=1.5 nm, with contours of the bright weight of the dark branch. (d) σB→D/σel\sigma_{B\to D}/\sigma_{\mathrm{el}} from Supplementary Note 5 at each field.

  • S7: (a) ℏ/Γ\hbar/\Gamma and QQ of the ℓ=−ν\ell=-\nu resonance for ν=1\nu=1, 2, 3, c=20c=20 nm, w=2w=2 nm. (b) ℏ/Γ\hbar/\Gamma versus cc for self-similar skyrmions (w=c/10w=c/10) and for w=1.5w=1.5 nm. (c) Bound states of Eq. (39) with V0=2V_{0}=2 meV, c=20c=20 nm, w=1.5w=1.5 nm, labelled by ℓ\ell.

References

  1. [1]Bogdanov, A. N. & Yablonsky, D. A. Thermodynamically stable ”vortices” in magnetically ordered crystals. the mixed state of magnets. Zh. Eksp. Teor. Fiz. 95, 178–182 (1989).
  2. [2]Bogdanov, A. & Hubert, A. Thermodynamically stable magnetic vortex states in magnetic crystals. Journal of Magnetism and Magnetic Materials 138, 255–269 (1994).DOI
  3. [3]Bogdanov, A. & Hubert, A. The properties of isolated magnetic vortices. physica status solidi (b) 186, 527–543 (1994).
  4. [4]Rössler, U. K., Leonov, A. A. & Bogdanov, A. N. Chiral skyrmionic matter in non-centrosymmetric magnets. Journal of Physics: Conference Series 303, 012105 (2011). URL https://dx.doi.org/10.1088/1742-6596/303/1/012105.
  5. [5]Dzyaloshinsky, I. A thermodynamic theory of “weak” ferromagnetism of antiferromagnetics. Journal of Physics and Chemistry of Solids 4, 241–255 (1958). URL https://www.sciencedirect.com/science/article/pii/0022369758900763.DOI
  6. [6]Moriya, T. New mechanism of anisotropic superexchange interaction. Phys. Rev. Lett. 4, 228–230 (1960). URL https://link.aps.org/doi/10.1103/PhysRevLett.4.228.DOI
  7. [7]Mühlbauer, S. et al. Skyrmion lattice in a chiral magnet. Science 323, 915–919 (2009). URL https://www.science.org/doi/abs/10.1126/science.1166767.
  8. [8]Heinze, S. et al. Spontaneous atomic-scale magnetic skyrmion lattice in two dimensions. Nature Physics 7, 713–718 (2011). URL https://doi.org/10.1038/nphys2045.
  9. [9]Romming, N., Kubetzka, A., Hanneken, C., von Bergmann, K. & Wiesendanger, R. Field-dependent size and shape of single magnetic skyrmions. Phys. Rev. Lett. 114, 177203 (2015). URL https://link.aps.org/doi/10.1103/PhysRevLett.114.177203.
  10. [10]Wiesendanger, R. Nanoscale magnetic skyrmions in metallic films and multilayers: a new twist for spintronics. Nature Reviews Materials 1, 16044 (2016).
  11. [11]Fert, A., Reyren, N. & Cros, V. Magnetic skyrmions: advances in physics and potential applications. Nature Reviews Materials 2, 17031 (2017).
  12. [12]Romming, N. et al. Writing and deleting single magnetic skyrmions. Science 341, 636–639 (2013).DOI
  13. [13]Nagaosa, N. & Tokura, Y. Topological properties and dynamics of magnetic skyrmions. Nature Nanotechnology 8, 899–911 (2013). URL https://doi.org/10.1038/nnano.2013.243.
  14. [14]Fert, A., Cros, V. & Sampaio, J. Skyrmions on the track. Nature Nanotechnology 8, 152–156 (2013). URL https://doi.org/10.1038/nnano.2013.29.
  15. [15]Parkin, S. S. P., Hayashi, M. & Thomas, L. Magnetic domain-wall racetrack memory. Science 320, 190 (2008).DOI
  16. [16]Schulz, T. et al. Emergent electrodynamics of skyrmions in a chiral magnet. Nature Physics 8, 301–304 (2012). URL https://doi.org/10.1038/nphys2231.
  17. [17]Bruno, P., Dugaev, V. K. & Taillefumier, M. Topological hall effect and berry phase in magnetic nanostructures. Phys. Rev. Lett. 93, 096806 (2004). URL https://link.aps.org/doi/10.1103/PhysRevLett.93.096806.
  18. [18]Everschör-Sitte, K. & Sitte, M. Real-space berry phases: Skyrmion soccer (invited). Journal of Applied Physics 115, 172602 (2014). URL https://doi.org/10.1063/1.4870695.
  19. [19]Verma, N., Addison, Z. & Randeria, M. Unified theory of the anomalous and topological hall effects with phase-space berry curvatures. Science Advances 8, eabq2765 (2022). URL https://www.science.org/doi/abs/10.1126/sciadv.abq2765.
  20. [20]Lux, F. R., Freimuth, F., Blügel, S. & Mokrousov, Y. The chiral hall effect of magnetic skyrmions from a cyclic cohomology approach. Physical Review Letters 124, 096602 (2020).
  21. [21]Bouaziz, J., Ishida, H., Lounis, S. & Blügel, S. Transverse transport in two-dimensional relativistic systems with non-trivial spin textures. Physical Review Letters 126, 147203 (2021).
  22. [22]Crum, D. M. et al. Perpendicular reading of single confined magnetic skyrmions. Nature Communications 6, 8541 (2015).DOI
  23. [23]Hanneken, C. et al. Electrical detection of magnetic skyrmions by tunnelling non-collinear magnetoresistance. Nature Nanotechnology 10, 1039–1042 (2015).DOI
  24. [24]Fernandes, I. L., Blügel, S. & Lounis, S. Spin-orbit enabled all-electrical readout of chiral spin-textures. Nature Communications 13, 1576 (2022).DOI
  25. [25]Denisov, K. S., Rozhansky, I. V., Averkiev, N. S. & Lähderanta, E. Electron scattering on a magnetic skyrmion in the nonadiabatic approximation. Phys. Rev. Lett. 117, 027202 (2016).DOI
  26. [26]Schütte, C. & Garst, M. Magnon-skyrmion scattering in chiral magnets. Phys. Rev. B 90, 094423 (2014). URL https://link.aps.org/doi/10.1103/PhysRevB.90.094423.DOI
  27. [27]Zhang, X. et al. Motion of skyrmions in nanowires driven by magnonic momentum-transfer forces. New Journal of Physics 19, 065001 (2017). URL https://doi.org/10.1088/1367-2630/aa6b70.
  28. [28]Pan, X.-F. et al. Magnon-skyrmion hybrid quantum systems: Tailoring interactions via magnons. Phys. Rev. Lett. 132, 193601 (2024). URL https://link.aps.org/doi/10.1103/PhysRevLett.132.193601.DOI
  29. [29]Lee, S., Nakata, K., Tchernyshyov, O. & Kim, S. K. Magnon dynamics in a skyrmion-textured domain wall of antiferromagnets. Phys. Rev. B 107, 184432 (2023). URL https://link.aps.org/doi/10.1103/PhysRevB.107.184432.DOI
  30. [30]Yu, P. Y. & Cardona, M. Fundamentals of semiconductors. Physics and materials properties. 4. ed. (Springer Berlin, Heidelberg, 2010).
  31. [31]Combescoт, M. & Shiau, S.-Y. Excitons and Cooper Pairs: Two Composite Bosons in Many-Body Physics (Oxford University Press, New York, 2016).
  32. [32]Knox, R. S. Introduction to Exciton Physics, 183–245 (Springer US, Boston, MA, 1983). URL https://doi.org/10.1007/978-1-4684-8878-4_5.
  33. [33]Mueller, T. & Malic, E. Exciton physics and device application of two-dimensional transition metal dichalcogenide semiconductors. npj 2D Materials and Applications 2, 29 (2018). URL https://doi.org/10.1038/s41699-018-0074-2.
  34. [34]High, A. A., Novitskaya, E. E., Butov, L. V., Hanson, M. & Gossard, A. C. Control of exciton fluxes in an excitonic integrated circuit. Science 321, 229–231 (2008). URL https://www.science.org/doi/abs/10.1126/science.1157845. https://www.science.org/doi/pdf/10.1126/science.1157845.DOI
  35. [35]Winbow, A. G. et al. Electrostatic conveyer for excitons. Phys. Rev. Lett. 106, 196806 (2011).DOI
  36. [36]Cadiz, F. et al. Exciton diffusion in WSe₂ monolayers embedded in a van der Waals heterostructure. Applied Physics Letters 112, 152106 (2018). URL https://doi.org/10.1063/1.5026478. https://pubs.aip.org/aip/apl/article-pdf/doi/10.1063/1.5026478/14510538/152106_1_online.pdf.
  37. [37]Peng, R. et al. Long-range transport of 2d excitons with acoustic waves. Nature Communications 13, 1334 (2022). URL https://doi.org/10.1038/s41467-022-29042-9.
  38. [38]Robert, C. et al. Measurement of the spin-forbidden dark excitons in mos₂ and mose₂ monolayers. Nat. Commun. 11, 4037 (2020).DOI
  39. [39]Kormányos, A. et al. k·p theory for two-dimensional transition metal dichalcogenide semiconductors. 2D Mater. 2, 022001 (2015).DOI
  40. [40]Gor’kov, L. P. & Dzyaloshinskii, I. E. Contribution to the theory of the mott exciton in a strong magnetic field. Sov. Phys. JETP. 26, 449 (1968).
  41. [41]Kazemi, M., Shahnazaryan, V. A., Zhumagulov, Y. V., Bessarab, P. F. & Shelykh, I. A. Interaction of excitons with magnetic topological defects in 2d magnetic monolayers: localization and anomalous hall effect. 2D Materials 10, 015003 (2022). URL https://dx.doi.org/10.1088/2053-1583/ac973d.
  42. [42]Woo, S. et al. Observation of room-temperature magnetic skyrmions and their current-driven dynamics in ultrathin metallic ferromagnets. Nat. Mater. 15, 501–506 (2016).
  43. [43]Miller, B. et al. Long-lived direct and indirect interlayer excitons in van der waals heterostructures. Nano Letters 17, 5229–5237 (2017). URL http://dx.doi.org/10.1021/acs.nanolett.7b01304.
  44. [44]Zhang, X., Ezawa, M. & Zhou, Y. Magnetic skyrmion logic gates: conversion, duplication and merging of skyrmions. Sci. Rep. 5, 9400 (2015).DOI
  45. [45]Rivera, P. et al. Observation of long-lived interlayer excitons in monolayer mose2–wse2 heterostructures. Nat. Commun. 6, 6242 (2015).DOI
  46. [46]Rivera, P. et al. Valley-polarized exciton dynamics in a 2d semiconductor heterostructure. Science 351, 688 (2016).DOI
  47. [47]Unuchek, D. et al. Room-temperature electrical control of exciton flux in a van der waals heterostructure. Nature 560, 340 (2018).DOI
  48. [48]Jauregui, L. A. et al. Electrical control of interlayer exciton dynamics in atomically thin heterostructures. Science 366, 870 (2019).DOI
  49. [49]Seyler, K. L. et al. Signatures of moiré-trapped valley excitons in mose2/wse2 heterobilayers. Nature 567, 66 (2019).DOI
  50. [50]Liu, J., Shi, M., Mo, P. & Lu, J. Electrical-field-induced magnetic skyrmion ground state in a two-dimensional chromium tri-iodide ferromagnetic monolayer. AIP Adv. 8, 055316 (2018).DOI
  51. [51]Behera, A. K., Chowdhury, S. & Das, S. R. Magnetic skyrmions in atomic thin cri3 monolayer. Appl. Phys. Lett. 114, 232402 (2019).DOI
  52. [52]Rüßmann, P. & JuDFTTeam. Judftteam/jukkr: v3.6. Zenodo (2022). URL https://doi.org/10.5281/zenodo.7284739.
  53. [53]Seki, S., Yu, X. Z., Ishiwata, S. & Tokura, Y. Observation of skyrmions in a multiferroic material. Science 336, 198 (2012).
  54. [54]Dalibard, J., Gerbier, F., Juzeliūnas, G. & Öhberg, P. Colloquium: Artificial gauge potentials for neutral atoms. Rev. Mod. Phys. 83, 1523 (2011).
  55. [55]Giannozzi, P., Baroni, S., Bonini, N. et al. Journal of Physics: Condensed Matter 21, 395502 (2009). URL https://doi.org/10.1088/0953–8984/21/39/395502.
  56. [56]Giannozzi, P. et al. Quantum espresso: a modular and open-source software project for quantum simulations of materials. Journal of physics: Condensed matter 21, 395502 (2009). URL http://dx.doi.org/10.1088/0953-8984/21/39/395502.
  57. [57]Hamann, D. R. Optimized norm-conserving vanderbilt pseudopotentials. Phys. Rev. B 88, 085117 (2013). URL https://link.aps.org/doi/10.1103/PhysRevB.88.085117.DOI
  58. [58]Perdew, J. P., Burke, K. & Ernzerhof, M. Generalized gradient approximation made simple. Phys. Rev. Lett. 77, 3865–3868 (1996). URL https://link.aps.org/doi/10.1103/PhysRevLett.77.3865.DOI
  59. [59]Hedin, L. New method for calculating the one-particle green’s function with application to the electron-gas problem. Phys. Rev. 139, A796–A823 (1965). URL https://link.aps.org/doi/10.1103/PhysRev.139.A796.DOI
  60. [60]Hybertsen, M. S. & Louie, S. G. First-principles theory of quasiparticles: Calculation of band gaps in semiconductors and insulators. Phys. Rev. Lett. 55, 1418–1421 (1985). URL https://link.aps.org/doi/10.1103/PhysRevLett.55.1418.DOI
  61. [61]Rohlfing, M. & Louie, S. G. Electron-hole excitations and optical spectra from first principles. Phys. Rev. B 62, 4927–4944 (2000). URL https://link.aps.org/doi/10.1103/PhysRevB.62.4927.
  62. [62]Onida, G., Reining, L. & Rubio, A. Electronic excitations: density-functional versus many-body green’s-function approaches. Rev. Mod. Phys. 74, 601–659 (2002). URL https://link.aps.org/doi/10.1103/RevModPhys.74.601.DOI
  63. [63]Godby, R. W. & Needs, R. Metal-insulator transition in kohn-sham theory and quasiparticle theory. Physical review letters 62, 1169 (1989).pubmed.ncbi.nlm.nih.gov/10039594
  64. [64]Rozzi, C. A., Varsano, D., Marini, A., Gross, E. K. U. & Rubio, A. Exact coulomb cutoff technique for supercell calculations. Phys. Rev. B 73, 205119 (2006). URL https://link.aps.org/doi/10.1103/PhysRevB.73.205119.DOI
  65. [65]Strinati, G. Application of the green’s functions method to the study of the optical properties of semiconductors. La Rivista del Nuovo Cimento (1978-1999) 11, 1826–9850 (1988). URL https://doi.org/10.1007/BF02725962.
  66. [66]Marini, A., Hogan, C., Grüning, M. & Varsano, D. yambo: An ab initio tool for excited state calculations. Computer Physics Communications 180, 1392–1403 (2009). URL https://www.sciencedirect.com/science/article/pii/S0010465509000472.DOI
  67. [67]Monkhorst, H. J. & Pack, J. D. Special points for brillouin-zone integrations. Physical review B 13, 5188 (1976).DOI
  68. [68]Goryca, M. et al. Revealing exciton masses and dielectric properties of monolayer semiconductors with high magnetic fields. Nat. Commun. 10, 4172 (2019).DOI
  69. [69]Dal Corso, A. Computational Materials Science 95, 337–350 (2014). URL https://www.sciencedirect.com/science/article/pii/S0927025614005187.
  70. [70]Vosko, S. H., Wilk, L. & Nusair, M. Accurate spin-dependent electron liquid correlation energies for local spin density calculations: a critical analysis. Can. J. Phys. 58, 1200 (1980).DOI
  71. [71]Dzyaloshinsky, I. A thermodynamic theory of “weak” ferromagnetism of antiferromagnetics. Journal of Physics and Chemistry of Solids 4, 241–255 (1958). URL https://www.sciencedirect.com/science/article/pii/0022369758900763.DOI
  72. [72]Liechtenstein, A., Katsnelson, M., Antropov, V. & Gubanov, V. Local spin density functional approach to the theory of exchange interactions in ferromagnetic metals and alloys. Journal of Magnetism and Magnetic Materials 67, 65–74 (1987). URL https://www.sciencedirect.com/science/article/pii/0304885387907219.DOI
  73. [73]Ebert, H. & Mankovsky, S. Anisotropic exchange coupling in diluted magnetic semiconductors: Ab initio spin-density functional theory. Phys. Rev. B 79, 045209 (2009). URL https://link.aps.org/doi/10.1103/PhysRevB.79.045209.DOI

Paper details

vlnc:W_enamv92r
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 2

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

AI tools

View statement
Looking for related papers…
Contents