betaFeedback

WorksPhysics and Astronomy

Engineering nonrelativistic spin splitting in semiconductors by altermagnetic proximity

Abstract

Altermagnets combine vanishing net magnetization with momentum-dependent, nonrelativistic spin splitting, offering functionalities attractive for spintronics and quantum devices. Their practical use, however, is constrained by the limited availability of materials that simultaneously possess the required altermagnetic symmetry, semiconducting electronic structure, and weak spin-orbit coupling. Here we show that these requirements can instead be separated between constituent materials through proximity engineering. We develop an analytical theory for a conventional semiconductor coupled to a gapped dd-wave altermagnet and demonstrate that virtual tunneling transfers the altermagnetic spin splitting to the semiconductor. Crucially, the induced nonrelativistic and relativistic spin splittings exhibit qualitatively different dependences on band alignment. Virtual processes involving the valence and conduction bands of the altermagnet can reinforce the altermagnetic splitting while contributing with opposite signs to the interfacial Rashba coupling. Consequently, electrostatic tuning can access a regime in which the Rashba splitting vanishes while a sizable dd-wave spin splitting persists. Exact diagonalization of the full heterostructure model confirms the analytical theory and establishes the robustness of this mechanism beyond the perturbative regime. Our results identify altermagnetic proximity as a route for engineering nonrelativistic spin-split electronic states in established semiconductor platforms, providing a general strategy for disentangling magnetic and relativistic spin interactions in quantum heterostructures.

Introduction

Assembling quantum materials with disparate physical properties into heterostructures can enable devices with novel or highly improved functionalities. Notable examples of this paradigm include the interface between two insulating perovskite oxides LaAlO3_3 and SrTiO3_3 which forms a high-mobility electron gas [1], Weyl semimetals that can be engineered in thin film multilayers of a magnetically doped 3D topological insulator separated by ordinary-insulator spacer layers [2], and topological superconductors in one and two spatial dimensions that can be formed by combining various existing ingredients [3, 4, 5, 6, 7, 8, 9]. Here we employ this time-tested strategy to engineer an altermagnetic semiconductor by interfacing an ordinary non-magnetic semiconductor with an insulating altermagnet, as schematically depicted in Figure 1. As we discuss below semiconductors with altermagnetic properties could be technologically valuable in spintronics and qubit design but no single material is currently known to be suitable for such applications.

Proposed semiconductor–altermagnet setup and band structure

Figure 1. The proposed setup. (a) A thin semiconductor film is supported on an insulating altermagnet. Electrostatic top gate can be used to define an elliptical quantum dot that can function as a field-free spin qubit. (b) Band structure: conduction band of the semiconductor (solid black line) is assumed to lie inside the gap of the insulating altermagnet. The band alignment is described by two parameters Δv\Delta_v and Δc\Delta_c that measure the position of the semiconductor band edge with respect to the ALM valence and conduction band, respectively.

Altermagnets are a new class of collinear magnetically ordered materials whose magnetic sublattices are related by rotation rather than translation or inversion [10, 11, 12, 13]. Their characteristic signature is a net zero magnetization accompanied by momentum-dependent, single-electron band spin splitting that can have dd-, gg- or ii-wave symmetry in the Brillouin zone, depending on the material. MnTe is a canonical example of a semiconducting altermagnet [14, 15, 16] while CrSb is an altermagnetic metal [17, 18]. Importantly, both are gg-wave altermagnets which limits their technological usefulness. The most interesting applications, such as the spin-splitter effect [19, 20] and its superconducting counterpart [21], have been predicted to occur in dd-wave altermagnets and they are symmetry-forbidden in gg- and ii-wave materials.

The goal, therefore, is to engineer a dd-wave altermagnetic semiconductor with suitable properties from materials terial ingredients that are available. In support of this goal we consider here a minimal model of an insulating dd-wave altermagnet, henceforth referred to as the “donor”, brought into contact with a thin-film single-valley semiconductor, the “host”, in a heterostructure shown schematically in Figure 1(a). Our model calculations show that significant spin-splitting can be induced in the semiconducting host through the altermagnetic proximity effect.

The underlying microscopic mechanism relies on virtual tunneling of electrons from the host to the donor and back, which produces dd-wave altermagnetic splitting in the former proportional to the second power of the tunneling amplitude. For realistic material parameters, the energy scale governing this splitting, which we denote η0ind\eta_0^{\mathrm{ind}}, can be significant, typically tens of meV, reflecting values of spin splitting in the host altermagnet that can be as large as 1 eV. The actual spin splitting experienced by electrons near the Fermi surface depends on the carrier density in the semiconductor through its dependence on the Fermi momentum ∼η0indkF2\sim\eta_0^{\mathrm{ind}} k_F^2. This implies that materials with higher carrier densities are capable of supporting more significant spin splitting.

Importantly, we find that any spin-orbit coupling (SOC) produced by inversion symmetry breaking at the interface can be mitigated by controlling the band alignment between the two materials, as illustrated in Figure 1(b). Specifically, if we define a band alignment parameter

x=ΔvΔALM∈(0,1),ΔALM=Δv+Δc,(1)x = \frac{\Delta_v}{\Delta_{\mathrm{ALM}}} \in(0,1), \qquad\Delta_{\mathrm{ALM}} = \Delta_v + \Delta_c, \tag*{(1)}

to quantify the position of the host band edge we find that there is always a value of xx that results in vanishing Rashba SOC at the Fermi surface. Thus, band alignment control can be used to transform the host into a pure non-relativistic altermagnet. Since relative band alignment in semiconductor heterostructures can be manipulated by a transverse electric field, including electrically driven transitions between distinct band-alignment regimes, demonstrated experimentally in van der Waals heterobilayers [22], the parameter xx constitutes a potentially gate-tunable control knob. As we show below, this provides a direct means of electrically tuning the heterostructure toward what we call the SOC “sweet spot” characterized by predominantly altermagnetic spin splitting. We note that in generic systems the cancellation that leads to negligible Rashba SOC acquires a weak momentum dependence, which we characterize below.

Motivation

Engineering a dd-wave semiconducting altermagnet is an interesting fundamental challenge in its own right but also has potential technological applications. Below we briefly outline two such applications in spintronics and spin qubit design. We also discuss broader implications of our work as it relates to general design principles for engineering non-relativistic spin splitting in semiconductors.

Spintronic applications

The field of spintronics seeks to leverage electron spin degree of freedom in storing, transmitting and processing classical information [23, 24]. Altermagnets represent an attractive material platform for spintronic applications due to their ability to generate and carry spin-polarized currents without producing stray magnetic fields that cause unwanted cross-talk between the individual circuit elements and tend to be difficult to mitigate or control. Spin valve, spin filter and spin splitter effects that underlie various key spintronic functionalities require the lowest-symmetry, dd-wave altermagnets for optimal operation and many proposed applications also rely on electrostatic gating to control these effects [25, 26]. In sum, spintronic applications would be best served by semiconducting dd-wave altermagnets.

Although no single material known today falls in this class there are several good candidates for insulating dd-wave altermagnets. These include a correlated layered insulator La2_2O3_3Mn2_2Se2_2 [27, 28, 29], iron-based transition metal chalcogenides Fe2_2MoX4_4 (X=SX = \mathrm{S}, Se, Te) and Fe2_2WTe4_4 [30], as well as monolayer oxychalcogenide V2_2Se2_2O [31]. We envision interfacing these materials with well-characterized and commonly used nonmagnetic single valley semiconductors such as GaAs, InAs, InP, black phosphorus or wide-bandgap nitrides (GaN, InN, AlN), all of which have a conduction band centered at the Γ\Gamma point. GaAs and the nitrides additionally exhibit weak intrinsic SOC which makes them suitable hosts if one is interested in engineering primarily non-relativistic spin splitting. As we demonstrate below such a heterostructure will behave as an altermagnetic semiconductor with controllable spin splitting and electron density.

Spin qubit design

Conventional electron-spin qubits typically rely on an applied magnetic field to generate the Zeeman splitting between their logical states. Although local electrical control can be achieved through spin-orbit coupling, magnetic-field gradients, or variations of the effective gg-factor, these approaches introduce additional materials and device constraints. In addition, the applied field is often difficult to reconcile with the presence of superconducting resonators desired for high-fidelity readout, which tolerate stray magnetic fields poorly. Altermagnetism offers a qualitatively different possibility: a sizable spin splitting can exist at zero applied magnetic field and, because it depends on momentum, can be controlled through the orbital confinement of the electron. Recently, proposals have been advanced [32, 33] that employ altermagnets in spin qubit design to create field-free, electrically tunable and individually addressable qubits.

When an electron in a dd-wave altermagnet is confined to an elliptical quantum dot, the dot geometry selects which momenta it samples, converting this momentum-dependent splitting into an electrically tunable single-particle spectrum with field-free spin splitting. This in turn leads to a possibility of harnessing such quantum dots as spin qubits with individually addressable qubit frequency. With the net magnetization being zero, there are no stray fields, preserving high-fidelity superconducting-resonator readout.

An important hurdle facing development of such qubits is the choice of a suitable material. An intrinsic altermagnetic host must be, all at once, a dd-wave altermagnet that is also gapped and semiconducting, single-valley with a simple band edge, low in nuclear-spin isotopes (to suppress hyperfine dephasing), and clean enough to control effectively electron density by electrostatic gating. No presently known materials satisfy these stringent conditions.

One proposal, mentioned in [32], is to decouple the host from the source of spin splitting by proximitizing a semiconductor with an altermagnet. In this geometry, the confined electron lives in a conventional, well-optimized semiconductor, while an adjacent altermagnet imprints its dd-wave spin splitting across the interface via the altermagnetic proximity effect. In this context the altermagnetic proximity effect has already been discussed in Ref. [34] based on the first-principles calculation. Here we advance an analytical theory of this phenomenon by deriving the effective Hamiltonian of the proximitized semiconductor and studying its properties. This approach helps us establish the general conditions under which a useful altermagnetic semiconductor can be created as well as understand its sensitivity to perturbations such as the relativistic SOC.

General considerations

For both applications it is desirable that the inherited altermagnetic splitting dominate relativistic spin-orbit effects. A sizable Rashba interaction changes the characteristic spin texture of a dd-wave altermagnet, lifts the spin degeneracy along its nonrelativistic nodal directions, and can complicate protocols that rely on the momentum-space structure of the altermagnetic splitting. Interfaces, however, generically break inversion symmetry and therefore provide precisely such a source of Rashba coupling. The ability to suppress this interfacial contribution without reducing the altermagnetic proximity effect is consequently an important design requirement.

More broadly, the heterostructure approach separates material requirements that are difficult to satisfy simultaneously in a single compound. The semiconductor host can be selected for a simple band structure, low disorder, favorable nuclear-spin composition, and mature electrostatic control, while the magnetic donor can be optimized independently for a large dd-wave altermagnetic splitting and an insulating gap. This separation is especially useful because the proximity-induced nonrelativistic and relativistic spin interactions need not track one another. As we show below, their distinct dependence on the host-donor band alignment makes it possible to strongly suppress the interfacial Rashba coupling without simultaneously suppressing the desired altermagnetic splitting. The heterostructure therefore provides not only a route around the scarcity of suitable intrinsic materials, but also a means of engineering the relativistic and nonrelativistic components of the spin splitting independently.

The induced altermagnetic state should moreover be directly accessible to experiment. Momentum-resolved spectroscopy of the semiconductor layer can probe its characteristic anisotropic spin splitting, while transport measurements can detect the corresponding spin-splitter response. Electrostatic control of the relative band alignment provides an additional diagnostic: tuning through the SOC sweet spot should suppress relativistic spin splitting on the altermagnetic nodal directions while leaving the dd-wave splitting away from those directions finite.

Model

We describe the heterostructure depicted in Figure 1(a) by a minimal continuum model Hamiltonian consisting of a semiconductor, altermagnet and the mutual tunnel coupling term,

H=Hs+HALM+Hg.(2)H = H_s + H_{\mathrm{ALM}} + H_g. \tag*{(2)}

The semiconductor host is modeled as a low-density degenerate electron gas with a parabolic band and no intrinsic SOC,

Hs(k)=ξksσ0,ξks=k22ms.(3)H_s(\mathbf{k}) = \xi_{\mathbf{k}}^s \sigma_0,\qquad\xi_{\mathbf{k}}^s = \frac{k^2}{2m_s}. \tag*{(3)}

Here σa\sigma_a denotes Pauli matrices in the spin space, with σ0\sigma_0 being a unit matrix. We adopt units such that ℏ=1\hbar= 1 and the lattice constant a0=1a_0 = 1, making kk dimensionless, and msm_s in the units of inverse energy.

We make a key assumption that the donor is a gapped altermagnetic insulator and describe it using a two-band Hamiltonian

HALM(k)=∑n=v,c[En(k)σ0+ηn(k)σz]Pn,(4)H_{\mathrm{ALM}}(\mathbf{k}) = \sum_{n=v,c} \left[E_n(\mathbf{k})\sigma_0 + \eta_n(\mathbf{k})\sigma_z\right]P_n, \tag*{(4)}

where

ηn(k)=ηn0(ky2−kx2)(5)\eta_n(\mathbf{k}) = \eta_n^0\left(k_y^2-k_x^2\right) \tag*{(5)}

accounts for the dd-wave spin splitting and PnP_n projects onto donor valence and conduction band n∈{v,c}n \in\{v,c\}. The band energies En(k)E_n(\mathbf{k}) are taken as

En(k)=Enedge+k22mn,(6)E_n(\mathbf{k})=E_n^{\mathrm{edge}}+\frac{k^2}{2m_n}, \tag*{(6)}

where En=v,cedgeE_{n=v,c}^{\mathrm{edge}} denotes the position of the band edge and mn=v,cm_{n=v,c} are the corresponding effective masses, such that mv<0m_v<0 and mc>0m_c>0. The altermagnetic strength of each band can be characterized by a dimensionless anisotropy factor

αn≡2mnηn0.(7)\alpha_n\equiv2m_n\eta_n^0. \tag*{(7)}

Throughout we assume ∣αn∣<1|\alpha_n|<1, so that effective inverse masses keep the sign of mnm_n along all directions and the constant-energy contours are ellipses, 90°-rotated between the two spins. In addition, we will assume that

ΔALM=Ecedge−Evedge>0(8)\Delta_{\mathrm{ALM}}=E_c^{\mathrm{edge}}-E_v^{\mathrm{edge}}>0 \tag*{(8)}

so that the spectrum is fully gapped. Typical spin-split spectra and constant energy contours of the Hamiltonian (4) are shown in Figure 2. They exhibit the characteristic C4TC_4\mathcal{T} symmetry of a dd-wave altermagnet where C4C_4 and T\mathcal{T} denote the in-plane 4-fold rotation and time reversal, respectively.

Figure 2. Spectral properties of the altermagnet Hamiltonian defined in Eq. (4). (a,b) Energy levels along kxk_x and kyk_y: both spins are exactly degenerate at each band edge and differ only in curvature; comparing (a) with (b) shows the C4TC_4\mathcal{T} altermagnetic symmetry. Shading marks the spin splitting. (c) Constant-energy contours of the conduction band: ellipses 90°-rotated between the spins, centered on the Γ\Gamma point represented by the dot. Valence band: blue; conduction band: red; spin ↑: solid; spin ↓: dotted. Parameters are: ΔALM=1\Delta_{\mathrm{ALM}}=1 eV with Ecedge=−Evedge=ΔALM/2E_c^{\mathrm{edge}}=-E_v^{\mathrm{edge}}=\Delta_{\mathrm{ALM}}/2, so that the gap is centered at zero energy, (mv,mc)=(−1,1)(m_v,m_c)=(-1,1) eV−1^{-1}, and ηv0=ηc0=150\eta_v^0=\eta_c^0=150 meV, giving ∣αn∣=0.3|\alpha_n|=0.3, with the contour in (c) taken at Ecedge+0.25ΔALME_c^{\mathrm{edge}}+0.25\Delta_{\mathrm{ALM}}.

Finally, we assume that the coupling between the semiconducting and altermagnetic layers consists of spin conserving electron tunneling parametrized by tunneling amplitudes gng_n and a Rashba-type spin flip tunneling that is allowed at the interface due to the locally broken inversion symmetry, parametrized by λR\lambda_R. This specific choice is physically motivated: if one is interested in generating a predominantly non-relativistic spin splitting in the semiconductor then, presumably, SOC strength in the host and the donor can be minimized by a judicious choice of materials. However, breaking of the inversion symmetry at the interface will inevitably give rise to a non-vanishing interfacial Rashba SOC which we include in our model.

With these assumptions the total heterostructure Hamiltonian can be written compactly in the second-quantized notation as H=∑kΨk†H(k)Ψk\mathcal{H}=\sum_{\mathbf{k}}\Psi_{\mathbf{k}}^\dagger\mathcal{H}(\mathbf{k})\Psi_{\mathbf{k}} where Ψk=(ak↑,ak↓,cvk↑,cvk↓,cck↑,cck↓)T\Psi_{\mathbf{k}}=(a_{\mathbf{k}\uparrow},a_{\mathbf{k}\downarrow},c_{v\mathbf{k}\uparrow},c_{v\mathbf{k}\downarrow},c_{c\mathbf{k}\uparrow},c_{c\mathbf{k}\downarrow})^{\mathsf{T}} and

H(k)=(Hs(k)V(k)V†(k)HALM(k)),V=(Vv,Vc).(9)\mathcal{H}(\mathbf{k})= \begin{pmatrix} H_s(\mathbf{k}) & V(\mathbf{k})\\ V^\dagger(\mathbf{k}) & H_{\mathrm{ALM}}(\mathbf{k}) \end{pmatrix}, \qquad V=(V_v,V_c). \tag*{(9)}

Here akσ†a_{\mathbf{k}\sigma}^\dagger and cnkσ†c_{n\mathbf{k}\sigma}^\dagger create electrons with crystal momentum k\mathbf{k} and spin σ\sigma in the semiconductor and the altermagnet, respectively. VnV_n are 2×22\times2 matrices in spin space describing the tunneling processes,

Vn(k)=gnσ0+λR(σxky−σykx).(10)V_n(\mathbf{k})=g_n\sigma_0+\lambda_R\left(\sigma^x k_y-\sigma^y k_x\right). \tag*{(10)}

Various model parameters are chosen such that the host conduction band lies within the donor gap as illustrated in Figure 1(b) and the band alignment is quantified by parameter x∈(0,1)x\in(0,1) defined in Eq. (1). We remark that within our conventions ηn0\eta_n^0 and λR\lambda_R both have units of energy.

Effective Semiconductor Hamiltonian

We now wish to derive the effective Hamiltonian for the proximitized semiconductor by formally integrating out the gapped altermagnetic degrees of freedom. This is most easily achieved by performing a unitary Schrieffer-Wolff type transformation on the 6×66\times6 matrix Hamiltonian H(k)\mathcal{H}(\mathbf{k}) that eliminates the off-diagonal term to leading order in the interlayer coupling. In doing this we will follow steps outlined in Refs. [35, 36].

Derivation of the effective Hamiltonian

To reduce clutter we temporarily suppress the crystal momentum k\mathbf{k} and write H=H0+V\mathcal{H}=\mathcal{H}_0+\mathcal{V} with

H0=(Hs00HALM),V=(0VV†0).(11)\mathcal{H}_0= \begin{pmatrix} H_s & 0\\ 0 & H_{\mathrm{ALM}} \end{pmatrix}, \qquad \mathcal{V}= \begin{pmatrix} 0 & V\\ V^\dagger& 0 \end{pmatrix}. \tag*{(11)}

We next seek an anti-Hermitian generator SS such that H~=eSHe−S\widetilde{H}=e^{S}He^{-S} is block diagonal. Expanding,

H~=H0+V+[S,H0]+[S,V]+12[S,[S,H0]]+…(12)\widetilde{H}=H_0+V+[S,H_0]+[S,V]+\frac{1}{2}[S,[S,H_0]]+\ldots \tag*{(12)}

we see that the first-order off-diagonal part vanishes if we choose SS such that

[S,H0]=−V.(13)[S,H_0]=-V. \tag*{(13)}

With the block-off-diagonal ansatz S=(0S−S†0)S=\begin{pmatrix}0&S\\-S^\dagger&0\end{pmatrix}, where SS is a 2×42\times4 block like VV, Eq. (13) becomes a linear equation

SHALM−HsS=−V.(14)SH_{\mathrm{ALM}}-H_sS=-V. \tag*{(14)}

To solve Eq. (14), we write out the donor structure explicitly. In the band basis, HALMH_{\mathrm{ALM}} is block diagonal with one 2×22\times2 spin block per band,

HALM=(hv00hc),hn=Enσ0+ηnσz.(15)H_{\mathrm{ALM}}=\begin{pmatrix}h_v&0\\0&h_c\end{pmatrix},\qquad h_n=E_n\sigma_0+\eta_n\sigma_z. \tag*{(15)}

Consequently, SS and VV split into the corresponding column blocks with S=(Sv,Sc)S=(S_v,S_c) alongside V=(Vv,Vc)V=(V_v,V_c). Because HsH_s is proportional to the identity at fixed k\mathbf{k}, Eq. (14) decouples into one 2×22\times2 equation per band,

Snhn−ξksSn=−Vn.(16)S_nh_n-\xi_{\mathbf{k}}^sS_n=-V_n. \tag*{(16)}

Each can be solved by right-multiplying with the inverse of the 2×22\times2 matrix

Gn=(ξksσ0−hn)−1(17)G_n=(\xi_{\mathbf{k}}^s\sigma_0-h_n)^{-1} \tag*{(17)}

which exists as long as the level lies inside the donor gap. Reassembling the blocks, we finally obtain

S=V(ξks(σ0⊗σ0)−HALM)−1=(VvGv,VcGc).(18)S=V\left(\xi_{\mathbf{k}}^s(\sigma_0\otimes\sigma_0)-H_{\mathrm{ALM}}\right)^{-1}=(V_vG_v,V_cG_c). \tag*{(18)}

We observe that each band carries its own generator Sn=VnGnS_n=V_nG_n and there are no vv–cc cross terms at this order. Physically, this can be understood by considering that at second order in interlayer tunneling the relevant processes involve virtual transitions to and from another band and hence cannot involve both conduction and valence band simultaneously.

Substituting Eq. (13) back into Eq. (12), the two second-order terms combine, [S,V]+12[S,[S,H0]]=12[S,V][S,V]+\frac{1}{2}[S,[S,H_0]]=\frac{1}{2}[S,V], and the effective host Hamiltonian becomes

Heff=Hs+12([S,V])ss=Hs+Σ,(19)H_{\mathrm{eff}}=H_s+\frac{1}{2}([S,V])_{ss}=H_s+\Sigma, \tag*{(19)}

where (⋅)ss(\cdot)_{ss} denotes the semiconductor block. The correction is contained in the surviving diagonal (semiconductor) block and reads

Σ=V(ξks−HALM)−1V†,(20)\Sigma=V(\xi_{\mathbf{k}}^s-H_{\mathrm{ALM}})^{-1}V^\dagger, \tag*{(20)}

Because HALMH_{\mathrm{ALM}} is diagonal in the band index, the self-energy can be expressed as a sum over bands,

Σ(k)=∑n=v,cVn(k)Gn(k)Vn(k)†,(21)\Sigma(\mathbf{k})=\sum_{n=v,c}V_n(\mathbf{k})G_n(\mathbf{k})V_n(\mathbf{k})^\dagger, \tag*{(21)}

with no vv–cc cross terms, and we restored the momentum dependence. We note that the expansion is controlled by

gn≪min⁡(Δv,Δc)=ΔALMmin⁡(x,1−x)(22)g_n\ll\min(\Delta_v,\Delta_c)=\Delta_{\mathrm{ALM}}\min(x,1-x) \tag*{(22)}

as each factor of SS in the commutators is of order gn/Δng_n/\Delta_n.

Performing explicitly the inversion indicated in Eq. (17) the Green’s function becomes

Gn(k)=(ξks−En(k))σ0+ηn(k)σz(ξks−En(k))2−ηn(k)2.(23)G_n(\mathbf{k})=\frac{(\xi_{\mathbf{k}}^s-E_n(\mathbf{k}))\sigma_0+\eta_n(\mathbf{k})\sigma_z}{(\xi_{\mathbf{k}}^s-E_n(\mathbf{k}))^2-\eta_n(\mathbf{k})^2}. \tag*{(23)}

In order to make further analytical progress it is useful to adopt an additional assumption, namely that the altermagnetic spin splitting ηn(k)\eta_n(\mathbf{k}) is small compared to both Δv\Delta_v and Δc\Delta_c. Then, it is permissible to expand the above expression to leading order as

Gv(k)≃σ0Δv(k)+ηv(k)Δv(k)2σz,G_v(\mathbf{k})\simeq\frac{\sigma_0}{\Delta_v(\mathbf{k})}+\frac{\eta_v(\mathbf{k})}{\Delta_v(\mathbf{k})^2}\sigma_z,
Gc(k)≃−σ0Δc(k)+ηc(k)Δc(k)2σz.(24)G_c(\mathbf{k})\simeq-\frac{\sigma_0}{\Delta_c(\mathbf{k})}+\frac{\eta_c(\mathbf{k})}{\Delta_c(\mathbf{k})^2}\sigma_z. \tag*{(24)}

where Δv(k)\Delta_v(\mathbf{k}) and Δc(k)\Delta_c(\mathbf{k}) are momentum-dependent gaps defined in Figure 3(a) that satisfy Δn(0)=Δn\Delta_n(0)=\Delta_n. Evaluating the channel sum indicated in Eq. (21) we finally obtain the self energy correction to the semiconductor Hamiltonian of the form

Figure 3. Altermagnetic proximity effect. (a) In the absence of tunneling, the donor and and the host are independent: the spin-degenerate semiconductor band ϵk\epsilon_{\mathbf{k}} (black) lies inside the gap of the altermagnet, between its spin-split valence (blue) and conduction (red) bands. It is useful to define momentum-dependent gaps Δv(k)\Delta_v(\mathbf{k}) and Δc(k)\Delta_c(\mathbf{k}) separating it from the two donor bands, measured at the same momentum. At the band edge they reduce to Δv(0)=Δv\Delta_v(0)=\Delta_v and Δc(0)=Δc\Delta_c(0)=\Delta_c. (b) When the tunneling is turned on, the host band undergoes a level shift δϵk\delta\epsilon_{\mathbf{k}} and acquires a spin splitting inherited from the altermagnet (spin ↑\uparrow solid, spin ↓\downarrow dashed). Because the induced couplings depend on momentum through Δn(k)\Delta_n(\mathbf{k}), the band alignment cancels the induced Rashba coupling at one momentum only, k∗\mathbf{k}^{*} (orange), where the splitting is purely altermagnetic, Δϵk∗=ηk∗\Delta\epsilon_{\mathbf{k}^{*}}=\eta_{\mathbf{k}^{*}}. Eqs. (31),(32) neglect this momentum dependence by evaluating all quantities at the band edge, k=0\mathbf{k}=0.

Σ(k)=δϵk+ηkind(ky2−kx2)σz+λkind(σxky−σykx).(25)\Sigma(\mathbf{k})=\delta\epsilon_{\mathbf{k}}+\eta_{\mathbf{k}}^{\mathrm{ind}}(k_y^2-k_x^2)\sigma_z+\lambda_{\mathbf{k}}^{\mathrm{ind}}(\sigma_xk_y-\sigma_yk_x). \tag*{(25)}

Here

δϵk=gv2+λR2k2Δv(k)−gc2+λR2k2Δc(k),(26)\delta\epsilon_k = \frac{g_v^2+\lambda_R^2 k^2}{\Delta_v(k)}-\frac{g_c^2+\lambda_R^2 k^2}{\Delta_c(k)}, \tag*{(26)}

represents a spin-independent correction to the semiconductor band energy, while

ηkind=(gv2−λR2k2)ηv0Δv(k)2+(gc2−λR2k2)ηc0Δc(k)2,(27)\eta_k^{\mathrm{ind}} = \frac{(g_v^2-\lambda_R^2 k^2)\eta_v^0}{\Delta_v(k)^2}+\frac{(g_c^2-\lambda_R^2 k^2)\eta_c^0}{\Delta_c(k)^2}, \tag*{(27)}

and

λkind=2λR[gvΔv(k)−gcΔc(k)].(28)\lambda_k^{\mathrm{ind}} = 2\lambda_R\left[\frac{g_v}{\Delta_v(k)}-\frac{g_c}{\Delta_c(k)}\right]. \tag*{(28)}

give the induced altermagnetic splitting and Rashba SOC, respectively.

Effective Hamiltonian properties

The spin-independent correction δϵk\delta\epsilon_k to the band energy is unimportant as long as it remains small compared to ΔALM\Delta_{\mathrm{ALM}} such that the host band remains well inside the altermagnet gap as illustrated in Figure 3(b). The altermagnetic proximity effect and the interface-induced SOC coupling in the semiconductor host are quantified by Eqs. (27),(28) respectively. We observe that these receive contributions from both donor bands. Importantly, the band contributions to ηkind\eta_k^{\mathrm{ind}} add up whereas they subtract in λkind\lambda_k^{\mathrm{ind}}. Right away we see that this structure allows for a vanishing Rashba SOC accompanied by a large non-relativistic spin splitting.

As the above sign structure is of key importance, it is worthwhile to pause here to understand its physical origin. Each term in the self-energy describes a virtual transition in which the host electron tunnels into the donor band nn and back. The level shift δϵk\delta\epsilon_k and the induced Rashba coupling λkind\lambda_k^{\mathrm{ind}} are energies produced by tunneling itself, so their sign is set by the energy denominator ξks−En\xi_k^s-E_n, which is positive for the valence band and negative for the conduction band. This is the origin of the negative sign between contributions in Eqs. (26),(28). For the level shift, the transition either conserves spin in both tunneling events, giving the terms proportional to gn2g_n^2, or flips it twice, giving terms proportional to λR2k2\lambda_R^2k^2; in either case there is no net spin flip. The Rashba coupling instead requires exactly one spin flip: the spin is flipped on entering or on leaving the donor, complemented by a spin-conserving process. Hence λkind\lambda_k^{\mathrm{ind}} is linear in gng_n and the two bands again contribute with opposite signs. The altermagnetic splitting is different in kind. Neither tunneling process treats the two host spin states differently: the coupling gng_n has the same amplitude for a host electron of either spin, and the interface Rashba term flips host spin up to down and down to up with equal strength. Without the donor splitting, ηn=0\eta_n=0, both host spin states would be shifted by the same amount, (gn2+λR2k2)/Δn(g_n^2+\lambda_R^2k^2)/\Delta_n. The spin dependence of the host must therefore come from the donor itself, where ηn(k)σz\eta_n(k)\sigma_z acts as a momentum-dependent exchange field. A host electron of given spin reaches the donor band either with its spin conserved, with probability gn2/Δn2g_n^2/\Delta_n^2, or with its spin flipped, with probability λR2k2/Δn2\lambda_R^2k^2/\Delta_n^2; in the second case it occupies the opposite spin branch and feels the exchange field with the opposite sign. Averaging gives (gn2−λR2k2)ηn/Δn2(g_n^2-\lambda_R^2k^2)\eta_n/\Delta_n^2, the two terms of Eq. (27). Each band therefore contributes with the sign of its own ηn\eta_n, and the two contributions to ηkind\eta_k^{\mathrm{ind}} reinforce each other when ηv0\eta_v^0 and ηc0\eta_c^0 have the same sign.

Importantly, symmetry alone does not fix this relative sign; whether they tend to cancel or reinforce each other depends on the microscopic structure of the altermagnet,

as we show with a lattice model in Appendix B. We also note that La2_2O3_3Mn2_2Se2_2 [27] exhibits the band structure indicative of the same sign of ηv0\eta_v^0 and ηc0\eta_c^0. We thus expect the two contributions to η0ind\eta_0^{\mathrm{ind}} to act in concert in this material.

To gain additional insight into the results obtained in the previous subsection it is helpful to focus on the case of a dilute electron density in the semiconductor, which is also the relevant limit for potential applications. Referring to Figure 1(b) we see that the Fermi energy then lies very close to the bottom of the host band and one can thus safely neglect all the momentum dependence entering δϵk\delta\epsilon_k, ηkind\eta_k^{\mathrm{ind}} and λkind\lambda_k^{\mathrm{ind}}, i.e. evaluate them at the band edge k=0k=0 in Figure 3. Focusing on the effect of the band alignment parametrized by xx the effective host Hamiltonian thus becomes

Heff≃Hs′+η0ind(ky2−kx2)σz+λRind(σxky−σykx),(29)H_{\mathrm{eff}} \simeq H_s' + \eta_0^{\mathrm{ind}}(k_y^2-k_x^2)\sigma_z + \lambda_R^{\mathrm{ind}}(\sigma_x k_y-\sigma_y k_x), \tag*{(29)}

where Hs′=Hs+δϵ0H_s'=H_s+\delta\epsilon_0 with

δϵ0=1ΔALM(gv2x−gc21−x),(30)\delta\epsilon_0=\frac{1}{\Delta_{\mathrm{ALM}}}\left(\frac{g_v^2}{x}-\frac{g_c^2}{1-x}\right), \tag*{(30)}

and

η0ind=1ΔALM2(gv2ηv0x2+gc2ηc0(1−x)2),(31)\eta_0^{\mathrm{ind}}=\frac{1}{\Delta_{\mathrm{ALM}}^2}\left(\frac{g_v^2\eta_v^0}{x^2}+\frac{g_c^2\eta_c^0}{(1-x)^2}\right), \tag*{(31)}
λRind=2λRΔALM(gvx−gc1−x).(32)\lambda_R^{\mathrm{ind}}=\frac{2\lambda_R}{\Delta_{\mathrm{ALM}}}\left(\frac{g_v}{x}-\frac{g_c}{1-x}\right). \tag*{(32)}

We observe that δϵ0\delta\epsilon_0 simply moves the host band relative to the altermagnet band edge and, assuming its value is small compared to ΔALM\Delta_{\mathrm{ALM}}, we can ignore it henceforth. η0ind\eta_0^{\mathrm{ind}} on the other hand is important as it transforms the host to an altermagnetic semiconductor. λRind\lambda_R^{\mathrm{ind}} quantifies the amount relativistic Rashba SOC induced by the inversion symmetry breaking at the interface. It is important to note that the above expressions are valid only away from x=0,1x=0,1 as per the expansion criterion given in Eq. (22). Typical dependences of η0ind\eta_0^{\mathrm{ind}} and λRind\lambda_R^{\mathrm{ind}} on the band alignment are illustrated in Figure 4(a).

Figure 4. Properties of the effective Hamiltonian compared to the exact solution. Solid lines represent the effective theory Eqs. (31,32) while dashed lines are extracted from the exact spectrum of Eq. (9) via Eq. (43) at ∣k∣=0.1|k|=0.1. (a) The altermagnetic coupling η0ind\eta_0^{\mathrm{ind}} (red) and the Rashba coupling λRind\lambda_R^{\mathrm{ind}} (blue). (b) The ratio ∣η0ind/λRind∣|\eta_0^{\mathrm{ind}}/\lambda_R^{\mathrm{ind}}|. Two vertical lines mark the sweet spot: x∗=0.667x^*=0.667 from Eq. (33) (solid) and the exact value x∗=0.637x^*=0.637 from Eq. (A7) (dashed), at which the exact λRind\lambda_R^{\mathrm{ind}} vanishes; the inset zooms into the two peaks. The shaded band highlights the region where Eq. (36) is satisfied for kF=0.1k_F=0.1 and the system may be regarded as predominantly altermagnetic. Parameters used are: gv=0.20g_v=0.20 eV, gc=0.10g_c=0.10 eV, ηv0=ηc0=150\eta_v^0=\eta_c^0=150 meV, λR=20\lambda_R=20 meV, ΔALM=1\Delta_{\mathrm{ALM}}=1 eV, ms=3m_s=3 eV−1^{-1}, and (mˉv,mˉc)=(−1,1)(\bar m_v,\bar m_c)=(-1,1) eV−1^{-1}.

Because η0ind\eta_0^{\mathrm{ind}} and λRind\lambda_R^{\mathrm{ind}} exhibit different dependences on xx it is possible to use band alignment to effectively control the host properties. For example assuming that gcg_c and gvg_v are both positive one can completely eliminate Rashba SOC by taking x=x∗x=x^* in Eq. (32) with

x∗=gvgv+gc.(33)x^*=\frac{g_v}{g_v+g_c}. \tag*{(33)}

Inspecting Eq. (31) it is clear that there is no reason to expect that η0ind\eta_0^{\mathrm{ind}} would vanish for this value of xx – hence a band alignment characterized by x=x∗x=x^* gives a pure non-relativistic spin splitting in the host semiconductor. We henceforth refer to this situation as the SOC “sweet spot”. In addition, because λRind\lambda_R^{\mathrm{ind}} can have either sign, it should be possible to achieve the same SOC-free situation even if the host Hamiltonian itself contains a small Rashba component.

If ηv0\eta_v^0 and ηc0\eta_c^0 have the same sign then the two terms in Eq. (31) act in concert and one may expect to obtain a large value of η0ind\eta_0^{\mathrm{ind}}. As mentioned above, the relative sign of ηv0\eta_v^0 and ηc0\eta_c^0 is a material specific property and can be understood based on a simple physical model.

More generally, one may be interested in engineering a semiconductor such that the electrons near the Fermi surface experience predominantly non-relativistic spin splitting. A reasonable criterion for this is that the altermagnetic splitting at the Fermi surface

ΔEALM≃2η0indkF2(34)\Delta E_{\mathrm{ALM}} \simeq2\eta_0^{\mathrm{ind}} k_F^2 \tag*{(34)}

is larger than the corresponding Rashba splitting

ΔER≃2λRindkF.(35)\Delta E_R \simeq2\lambda_R^{\mathrm{ind}} k_F. \tag*{(35)}

Requiring ΔEALM>ΔER\Delta E_{\mathrm{ALM}}>\Delta E_R leads to the condition

η0ind/λRind>kF−1.(36)\eta_0^{\mathrm{ind}}/\lambda_R^{\mathrm{ind}}>k_F^{-1}. \tag*{(36)}

Linear vs. quadratic growth with kFk_F of ΔER\Delta E_R and ΔEALM\Delta E_{\mathrm{ALM}} implies that the condition is easier to satisfy at higher electron densities. Figure 4(b) shows the ratio η0ind/λRind\eta_0^{\mathrm{ind}}/\lambda_R^{\mathrm{ind}} evaluated from Eqs. (31),(32). Assuming, as an example, that the electron density is such that kF=0.1k_F=0.1 the system will be predominantly altermagnetic for band alignments inside the shaded region centered at x∗x^*. We note that for the chosen material parameters, Eq. (36) will be satisfied for all xx when kF≳0.3k_F\gtrsim0.3.

To get some intuition for these values we recall that in a 2D electron gas kF2=2πna02k_F^2=2\pi n a_0^2 where nn denotes the electron density and a0a_0 is the lattice constant. Taking GaN with a0≃0.32a_0\simeq0.32 nm as an example, we find that kF=0.1k_F=0.1 requires the electron density n≃1012 cm−2n\simeq10^{12}\,\mathrm{cm}^{-2} which is achievable in AlGaN/GaN heterostructures [37]. Assuming xx is tuned close to its sweet spot value, Figure 4(a) gives η0ind≃25\eta_0^{\mathrm{ind}}\simeq25 meV which according to Eq. (34) translates to ΔEALM≃0.5\Delta E_{\mathrm{ALM}}\simeq0.5 meV. This is a small splitting which may nevertheless be observable in transport experiments at low temperature. Larger electron densities will produce larger altermagnetic spin splitting.

As a consistency check on the effective Hamiltonian approach we extracted the η0ind/λRind\eta_0^{\mathrm{ind}}/\lambda_R^{\mathrm{ind}} ratio from the exact numerical solution of the full 6×66\times6 Bloch Hamiltonian Eq. (9), as described at the end of Sec. V. These results are plotted as dashed lines in Figure 4. We observe that the exact result largely agrees with the approximation near the x=x∗x=x^* sweet spot but the two disagree in the vicinity of x=0,1x=0,1. The disagreement is due to the fact that, in view of Eq. (22), the procedure leading to the effective Hamiltonian breaks down in the limit x→0,1x\to0,1. The results shown in Figure 4 nevertheless indicate that, generically, there will be a range of favorable band alignments near x=x∗x=x^* where the ratio η0ind/λRind\eta_0^{\mathrm{ind}}/\lambda_R^{\mathrm{ind}} can be made arbitrarily large and the heterostructure behaves as a pure non-relativistic altermagnet.

We note that Figure 4 also shows small deviations between the approximate and the exact solution near x=x∗x=x^*. These can be attributed to two effects. First, the sweet spot position x∗x^* given in Eq. (33) is based on approximate values of η0ind\eta_0^{\mathrm{ind}} and λRind\lambda_R^{\mathrm{ind}} given in Eqs. (31),(32) which neglect the momentum dependence of these quantities indicated in more accurate Eqs. (27),(28). The upshot is that the precise position of the SOC sweet spot is in general kk-dependent: for a given band alignment the induced Rashba coupling vanishes at a single momentum only, k∗k^* in Figure 3(b), where the splitting is purely altermagnetic. Second, these expressions measure the gaps from the bare semiconductor band edge, whereas the proximitized level is shifted by δϵ0\delta\epsilon_0, Eq. (30). For the parameters of Figure 4 this moves the sweet spot from x∗=0.667x^*=0.667 to x=0.637x=0.637. Importantly, the approximate x∗x^* still lies inside the shaded region of Figure 4, where the exact ratio η0ind/λRind\eta_0^{\mathrm{ind}}/\lambda_R^{\mathrm{ind}} exceeds 10. For most practical purposes, therefore, the simple approximate expression in Eq. (33) should be more than adequate. For the sake of completeness we provide a more detailed analysis that takes into account the additional kk-dependence and the level shift in Appendix A.

Importantly, the momentum dependence of the sweet spot is not unavoidable. As shown in Appendix A, the leading drift vanishes when in the “symmetric limit” defined therein; the sweet spot location x∗=1/2x^*=1/2 becomes exact and momentum-independent. Thus, suitable engineering of the band curvatures and tunneling amplitudes can realize an extended SOC-free altermagnetic regime rather than cancellation at a single momentum.

Numerical results

Following the strategy of Ref. [36], we test the effective theory against the exact spectrum: we numerically diagonalize the full 6×66\times6 Bloch Hamiltonian, Eq. (9), and compute the electron spectral function projected onto the semiconductor block,

As(k,ω)=−1πTr⁡′(ω+iδ−H(k))−1,(37)A_s(k,\omega)=-\frac{1}{\pi}\operatorname{Tr}'\left(\omega+i\delta-\mathcal{H}(k)\right)^{-1}, \tag*{(37)}

where Tr⁡′\operatorname{Tr}' extends over the two semiconductor components. The layer-resolved As(k,ω)A_s(k,\omega) is what a spectroscopic probe, e.g. angle-resolved photoemission (ARPES), of the semiconductor would measure, and its exact in-gap dispersion accounts for the proximity effect non-perturbatively.

Here ω\omega refers to the probe frequency corresponding the energy at which one asks whether the system can absorb or emit an electron with momentum kk. Scanned over a window at fixed kk, it makes As(k,ω)A_s(k,\omega) light up whenever it coincides with an energy eigenvalue that has weight on the semiconductor. To see this, we insert the eigenbasis of the Hermitian Bloch matrix, H(k)∣m,k⟩=Em(k)∣m,k⟩\mathcal{H}(k)|m,k\rangle=E_m(k)|m,k\rangle with m=1,…,6m=1,\ldots,6, into the resolvent:

(ω+iδ−H(k))−1=∑m=16∣m,k⟩⟨m,k∣ω−Em(k)+iδ.(38)\left(\omega+i\delta-\mathcal{H}(k)\right)^{-1}=\sum_{m=1}^{6}\frac{|m,k\rangle\langle m,k|}{\omega-E_m(k)+i\delta}. \tag*{(38)}

Taking the partial trace with PsP_s the projector onto the two semiconductor components — in the basis defined above Eq. (9), explicitly Ps=diag⁡(1,1,0,0,0,0)P_s=\operatorname{diag}(1,1,0,0,0,0) — turns each numerator into the weight ⟨m,k∣Ps∣m,k⟩=∣Ps∣m,k⟩∣2\langle m,k|P_s|m,k\rangle=|P_s|m,k\rangle|^2, and J[1/(ω−Em+iδ)]=−δ/[(ω−Em)2+δ2]\mathcal{J}[1/(\omega-E_m+i\delta)]=-\delta/[(\omega-E_m)^2+\delta^2] then gives

As(k,ω)=∑m=16∣Ps∣m,k⟩∣2δ/π[ω−Em(k)]2+δ2.(39)A_s(k,\omega)=\sum_{m=1}^{6}|P_s|m,k\rangle|^2\frac{\delta/\pi}{[\omega-E_m(k)]^2+\delta^2}. \tag*{(39)}

Hence As(k,ω)A_s(k,\omega) can be interpreted as a set of Lorentzian peaks of width δ\delta centered on the six eigenvalues Em(k)E_m(k), each weighted by the semiconductor content ∣Ps∣m,k⟩∣2|P_s|m,k\rangle|^2 of its eigenstate.

We evaluate the spectral function for two sets of parameters, a symmetric and a generic, at a coupling strength ε=(gv+gc)/ΔALM=0.3\varepsilon=(g_v+g_c)/\Delta_{\mathrm{ALM}}=0.3. Both sets share

ΔALM=1 eV,mˉs=3 eV−1,∣mˉv∣=1 eV−1,\Delta_{\mathrm{ALM}}=1\ \mathrm{eV},\qquad\bar{m}_s=3\ \mathrm{eV}^{-1},\qquad|\bar{m}_v|=1\ \mathrm{eV}^{-1},
ηv0=ηc0=150 meV,λR=20 meV,(40)\eta_v^0=\eta_c^0=150\ \mathrm{meV},\qquad\lambda_R=20\ \mathrm{meV}, \tag*{(40)}

while the tunneling amplitudes and conduction band mass follow one of two rules:

symmetric: gv=gc=12εΔALM,mˉc=0.6 eV−1g_v=g_c=\frac{1}{2}\varepsilon\Delta_{\mathrm{ALM}},\qquad\bar{m}_c=0.6\ \mathrm{eV}^{-1};

generic: gv=2gc=23εΔALM,mˉc=1 eV−1g_v=2g_c=\frac{2}{3}\varepsilon\Delta_{\mathrm{ALM}},\qquad\bar{m}_c=1\ \mathrm{eV}^{-1}.

ΔE(k)=2[η0ind(ky2−kx2)]2+(λRindk)2.(41)\Delta E(k)=2\sqrt{\left[\eta_0^{\mathrm{ind}}\left(k_y^2-k_x^2\right)\right]^2+\left(\lambda_R^{\mathrm{ind}}k\right)^2}. \tag*{(41)}

The altermagnetic term vanishes on the nodal line kx=kyk_x=k_y, while the Rashba term depends only on k=∣k∣k=|{\bf k}|. The splitting on the nodal line, ΔEn=2∣λRind∣k\Delta E_n=2|\lambda_R^{\mathrm{ind}}|k, therefore measures the Rashba coupling alone. The splitting along the antinodal line ky=0k_y=0 is ΔEa=2(η0indkx2)2+(λRindkx)2\Delta E_a=2\sqrt{(\eta_0^{\mathrm{ind}}k_x^2)^2+(\lambda_R^{\mathrm{ind}}k_x)^2} and can be used to extract η0ind\eta_0^{\mathrm{ind}}. At each band alignment ‍സ xx we diagonalize Eq. (9) at the same momentum ∣k∣|k| on these two cuts, and read off the two splittings ΔEn,a\Delta E_{n,a}. From these we then calculate

Figure 5. Exact spectral function As(k,ω)A_s(k,\omega) for the semiconductor, Eq. (39), for the symmetric (top) and generic (bottom) parameter set, each at its exact Γ\Gamma-point sweet spot, Eq. (51): x=1/2x=1/2 and x=0.637x=0.637. (a,d) Color plot of As(k,ω)A_s(k,\omega) along the nodal line kx=kyk_x=k_y, with ω\omega measured from the spin-averaged in-gap dispersion Eˉ(k)\bar{E}(k). The symmetric set remains degenerate at every ∣k∣|k| while for the generic set the residual λkind\lambda_k^{\mathrm{ind}} splits the band by 2∣k∣∣λkind∣2|k||\lambda_k^{\mathrm{ind}}|. (b,c,e,f) Constant-energy cuts at ω1=E0+30\omega_1=E_0+30 meV and ω2=E0+60\omega_2=E_0+60 meV, with E0E_0 the bottom of the in-gap band (δϵ0\delta\epsilon_0 above the bare host edge), zoomed in to one of the four diagonal touching points; the insets show the full spin-split contours. We used δ=0.04\delta=0.04 meV in (a,d), 0.030.03 meV in the main panel and 11 meV in the insets. Parameters are those given in Eqs. (40),41) at ε=0.3\varepsilon=0.3.

∣λRind∣=ΔEn2k,∣η0ind∣=1k2(ΔEa2)2−(λRindk)2,(42)|\lambda_R^{\mathrm{ind}}|=\frac{\Delta E_n}{2k},\qquad|\eta_0^{\mathrm{ind}}|=\frac{1}{k^2}\sqrt{\left(\frac{\Delta E_a}{2}\right)^2-\left(\lambda_R^{\mathrm{ind}}k\right)^2}, \tag*{(42)}

and plot the extracted values in Figure 4.

Conclusions

In this work, we have proposed a route to engineer a semiconducting dd-wave altermagnet by interfacing a conventional semiconductor with an insulating altermagnetic donor. Starting from a minimal continuum model, we show that virtual interlayer tunneling transfers the momentum-dependent spin splitting of the altermagnet to the semiconductor. The resulting proximity-induced splitting is of the second order in the tunneling amplitude and the relevant altermagnetic strength η0ind\eta_0^{\mathrm{ind}} can reach tens of meV for realistic parameters. The resulting spin splitting experienced by the electrons near the Fermi surface depends on the carrier density through its dependence on kFk_F indicated in Eq. (35), suggesting that a substantial altermagnetic proximity effect can be realized at higher electron densities without requiring the semiconductor itself to possess magnetic order.

An important feature of the proposed heterostructure is the possibility of controlling the relativistic and nonrelativistic components of the induced spin splitting independently through band alignment. Inversion-symmetry breaking at the interface generically produces a Rashba spin-orbit coupling, but the virtual processes involving the valence and conduction bands of the altermagnet contribute to this coupling with opposite signs. Their contributions to the altermagnetic splitting, by contrast, can reinforce one another. This leads to an SOC sweet spot at which the induced Rashba coupling vanishes while a finite dd-wave altermagnetic splitting remains. Exact diagonalization of the full heterostructure Hamiltonian confirms this mechanism beyond the perturbative effective theory and shows how the location of the sweet spot acquires weak momentum dependence in a generic system.

It is important to note that, the relative band alignment in semiconductor heterostructures is not fixed solely by the constituent materials, but can be controlled by an applied transverse electric field or electrostatic gating, as demonstrated experimentally in van der Waals semiconductor heterobilayers [22] and explored theoretically in a variety of heterostructures [38, 39]. Thus, the all-important alignment parameter xx introduced in Eq. (1) may be regarded, at least in part, as an experimentally tunable control parameter.

These results suggest that proximity engineering can circumvent an important materials constraint confronting proposed altermagnetic devices. Rather than requiring a single material to simultaneously provide a suitable semiconductor band structure, dd-wave altermagnetism, and weak spin-orbit coupling, these functions can be divided between two materials and optimized separately. The semiconductor can provide a clean, well-characterized and gate-tunable electronic platform, while the adjacent insulating altermagnet supplies the desired nonrelativistic spin splitting. Such heterostructures could therefore provide a versatile platform for electrically controlled spintronic devices and field-free spin qubits, while also opening a broader route toward engineering altermagnetic electronic states by proximity.

Looking ahead, several research directions are suggested by the present study. It would be clearly desirable to impart a larger altermagnetic splitting to the host semiconductor. A key limiting factor in the present setup with the valley located at the Γ\Gamma point is the k2k^{2} suppression implied by Eq. (34). For realistic semiconductor carrier densities this amounts to a ∼10\sim10–100100 fold suppression compared to the bare value of 2η0ind2\eta_{0}^{\mathrm{ind}}. This suppression reflects the simple fact that spin splitting in a dd-wave altermagnet vanishes as k2k^{2} near Γ\Gamma. Using a semiconductor with valleys located near X and Y points of the Brillouin zone would avoid this suppression as the dd-wave splitting behaves as ∼(cos⁡kx−cos⁡ky)\sim(\cos k_x-\cos k_y) and is therefore maximal at these points. Several families of materials with valleys near X and Y are known to exist; these include (001) films of rock-salt group IV-VI compounds (SnTe, PbTe, PbSe, PbS) and group-IV monochalcogenides (SnS, SnSe, GeS, GeSe). A detailed study of heterostructures with these materials would be interesting.

Our study also made several simplifying assumptions that could be relaxed in future works if one aims at obtaining more realistic and material-specific predictions. These include perfect momentum conservation for the interlayer tunneling processes, which clearly will not hold unless the materials are perfectly lattice matched, a neglect of any net exchange field generated by an uncompensated surface, the effects of the finite semiconductor layer thickness, and the material-specific symmetry constraints that may allow additional terms in the Hamiltonian beyond those considered here. Our work nevertheless points to rich new physics relevant to applications that is enabled by altermagnet-semiconductor heterostructures which will hopefully motivate future theoretical and experimental studies of these systems.

Acknowledgments

We are grateful to Nitin Kaushal for illuminating discussions and correspondence. This research was supported in part by NSERC and CIFAR. Claude (Anthropic) was used to assist with debugging parts of the numerical code, checking references, and improving the manuscript.

References

SOC sweet spot

The SOC sweet spot is the band alignment at which the induced Rashba coupling vanishes, so that the two in-gap bands of Eq. (41) are degenerate on the nodal lines ∣kx∣=∣ky∣|k_x| = |k_y|. Setting λRind=0\lambda_{R}^{\mathrm{ind}} = 0 in Eq. (32) gives the sweet spot of Eq. (33). As mentioned in Sec. IV, this result has two limitations. First, it holds only at Γ\Gamma: the gaps Δn(k)\Delta_n(k) change with momentum, so the sweet spot moves with kk. Second, it measures the gaps from the bare host band edge, while the tunneling also shifts the in-gap level away from that edge [Eq. (26)], so the electron sees one gap widened and the other narrowed.

In this appendix we include both effects. On the nodal lines the dd-wave term vanishes, and the 6×66 \times6 Hamiltonian of Eq. (9) decouples exactly into two 3×33 \times3 blocks, one for each spin eigenstate of the interface Rashba term. This gives the sweet spot at every momentum with the level shift included; dropping the shift recovers the second-order result. We then evaluate the induced splitting at the sweet spot and compare both with the exact spectrum in Figure 6.

Figure 6. Closed-form predictions against the exact spectrum, for the symmetric (blue) and generic (red) sets at ε=0.3\varepsilon= 0.3 (left) and ε=0.4\varepsilon= 0.4 (right). (a,b) The sweet spot: curves are Eq. (A7), circles are located from the 6×66 \times6 spectrum by closing the nodal-line splitting at each ∣k∣\lvert k\rvert; dotted, the second-order value x∗=2/3x^* = 2/3 of the generic set [Eq. (A8)]. (c,d) The splitting at the sweet spot: curves are Eq. (A12), circles are the measured ΔE/2k2\Delta E/2k^2 along kxk_x—by Eq. (42), with λkind=0\lambda_k^{\mathrm{ind}} = 0 at the sweet spot, the splitting there is 2ηvindk22\eta_v^{\mathrm{ind}} k^2. Parameters are: ΔALM=1 eV\Delta_{\mathrm{ALM}} = 1\ \mathrm{eV}, ms=3 eV−1m_s = 3\ \mathrm{eV}^{-1}, mv=−1 eV−1m_v = -1\ \mathrm{eV}^{-1}, ηv=ηc=150 meV\eta_v = \eta_c = 150\ \mathrm{meV}, λR=20 meV\lambda_R = 20\ \mathrm{meV}; symmetric set gv=gcg_v = g_c, mc=0.6 eV−1m_c = 0.6\ \mathrm{eV}^{-1}; generic set gv=2gcg_v = 2g_c, mc=1 eV−1m_c = 1\ \mathrm{eV}^{-1}; with gv+gc=εΔALMg_v + g_c = \varepsilon\Delta_{\mathrm{ALM}}, i.e. (gv,gc)=(0.15,0.15)(g_v, g_c) = (0.15, 0.15) and (0.2,0.1) eV(0.2, 0.1)\ \mathrm{eV} at ε=0.3\varepsilon= 0.3, and (0.2,0.2)(0.2, 0.2) and (0.267,0.133) eV(0.267, 0.133)\ \mathrm{eV} at ε=0.4\varepsilon= 0.4.

Exact sweet spot on the nodal line

Momentum-dependent gaps.—With Eqs. (3,6) the momentum-dependent gaps of Eq. (24) are

Δv(k)=Δv+avk2,Δc(k)=Δc+ack2,\Delta_v(k) = \Delta_v + a_v k^2,\qquad\Delta_c(k) = \Delta_c + a_c k^2,
av=12ms+12∣mv∣,ac=12mc−12ms,(43)a_v = \frac{1}{2m_s} + \frac{1}{2|m_v|},\qquad a_c = \frac{1}{2m_c} - \frac{1}{2m_s}, \tag*{(43)}

and they add up to the donor gap at the same momentum, Δv(k)+Δc(k)=ΔALM(k)≡ΔALM+(av+ac)k2\Delta_v(k) + \Delta_c(k) = \Delta_{\mathrm{ALM}}(k) \equiv\Delta_{\mathrm{ALM}} + (a_v + a_c)k^2.

Decoupling on the nodal line.—On kx=kyk_x=k_y the form factor vanishes, ηn(k)=0\eta_n(\boldsymbol{k})=0, so both donor blocks are spin-degenerate and the only spin structure left in Eq. (9) is the interface Rashba matrix R=λR(σxky−σykx)R=\lambda_R(\sigma_xk_y-\sigma_yk_x), whose eigenvalues are ±r\pm r with r=λRkr=\lambda_R k, since R2=λR2k2σ0R^2=\lambda_R^2 k^2\sigma_0. Let UU diagonalize it. Applying the same UU in all three spin spaces leaves the host block ξksσ0\xi_{\boldsymbol{k}}^s\sigma_0 and both donor blocks En(k)σ0E_n(\boldsymbol{k})\sigma_0 untouched—each is proportional to the identity—while turning every interface block Vn=gnσ0+RV_n=g_n\sigma_0+R into the diagonal 2×22\times2 matrix diag⁡(gn+r,gn−r)\operatorname{diag}(g_n+r,g_n-r). All six blocks are then diagonal in the ±\pm label, so the 6×66\times6 splits into two independent 3×33\times3 matrices, one per Rashba branch:

H±=(ξksgv±rgc±rgv±rEv(k)0gc±r0Ec(k)),(44)H_\pm= \begin{pmatrix} \xi_{\boldsymbol{k}}^s & g_v\pm r & g_c\pm r\\ g_v\pm r & E_v(\boldsymbol{k}) & 0\\ g_c\pm r & 0 & E_c(\boldsymbol{k}) \end{pmatrix}, \tag*{(44)}

in the basis (host, valence, conduction).

Sweet-spot condition.—Only the host couples to the other two, so the donor amplitudes can be eliminated exactly: for an eigenvector (cs,cv,cc)(c_s,c_v,c_c) at energy EE the last two rows give cn=(gn±r)cs/[E−En(k)]c_n=(g_n\pm r)c_s/[E-E_n(\boldsymbol{k})]. Substituting these into the first row and writing W=E−ξksW=E-\xi_{\boldsymbol{k}}^s for the in-gap eigenvalue measured from the host band, so that E−Ev(k)=W+Δv(k)E-E_v(\boldsymbol{k})=W+\Delta_v(\boldsymbol{k}) and E−Ec(k)=W−Δc(k)E-E_c(\boldsymbol{k})=W-\Delta_c(\boldsymbol{k}),

W=(gv±r)2W+Δv(k)−(gc±r)2Δc(k)−W.(45)W=\frac{(g_v\pm r)^2}{W+\Delta_v(\boldsymbol{k})}-\frac{(g_c\pm r)^2}{\Delta_c(\boldsymbol{k})-W}. \tag*{(45)}

The sweet spot is where the two branches return the same WW, so that the doublet stays degenerate. Subtracting the two versions of Eq. (45), the terms even in rr cancel and an overall 4r4r divides out, leaving

gvΔv(k)+W=gcΔc(k)−W.(46)\frac{g_v}{\Delta_v(\boldsymbol{k})+W}=\frac{g_c}{\Delta_c(\boldsymbol{k})-W}. \tag*{(46)}

The electron does not sit at the host band edge but a distance WW above it, so the gaps it sees are Δv(k)+W\Delta_v(\boldsymbol{k})+W and Δc(k)−W\Delta_c(\boldsymbol{k})-W: one widens, the other narrows by the same amount.

Sweet spot solution.—We call the common value of the two ratios in Eq. (46) ε(k)\varepsilon(\boldsymbol{k}), so that Δv(k)+W=gv/ε(k)\Delta_v(\boldsymbol{k})+W=g_v/\varepsilon(\boldsymbol{k}) and Δc(k)−W=gc/ε(k)\Delta_c(\boldsymbol{k})-W=g_c/\varepsilon(\boldsymbol{k}). Adding the two removes WW and, with the sum rule below Eq. (43), gives

ε(k)=gv+gcΔALM(k).(47)\varepsilon(\boldsymbol{k})=\frac{g_v+g_c}{\Delta_{\mathrm{ALM}}(\boldsymbol{k})}. \tag*{(47)}

The level shift therefore does not change ε\varepsilon; it only changes how ΔALM(k)\Delta_{\mathrm{ALM}}(\boldsymbol{k}) is divided between the two channels, which requires WW. With the sweet-spot denominators, each term of Eq. (45) collapses, (gn±r)2ε/gn=ε(gn±2r+r2/gn)(g_n\pm r)^2\varepsilon/g_n=\varepsilon(g_n\pm2r+r^2/g_n), and in the difference of the two channels the ±2r\pm2r pieces drop out:

W=ε(k)[(gv−gc)+λR2k2(1gv−1gc)]=δεk,(48)W=\varepsilon(\boldsymbol{k})\left[(g_v-g_c)+\lambda_R^2k^2\left(\frac{1}{g_v}-\frac{1}{g_c}\right)\right]=\delta\varepsilon_k, \tag*{(48)}

i.e. WW is the level shift of Eq. (26) evaluated at the sweet spot. Then Δv(k)=gv/ε(k)−δεk\Delta_v(\boldsymbol{k})=g_v/\varepsilon(\boldsymbol{k})-\delta\varepsilon_k, and converting back to the k=0k=0 alignment fraction x=Δv/ΔALMx=\Delta_v/\Delta_{\mathrm{ALM}} gives

x∗(k)=gvgv+gc−gcav−gvacΔALM(gv+gc)k2−δεkΔALM.(49)x^*(\boldsymbol{k})=\frac{g_v}{g_v+g_c}-\frac{g_ca_v-g_va_c}{\Delta_{\mathrm{ALM}}(g_v+g_c)}k^2-\frac{\delta\varepsilon_k}{\Delta_{\mathrm{ALM}}}. \tag*{(49)}

exact for the model of Eq. (9) at every momentum.

Second-order limit.—Setting W=0W=0 in Eq. (46) gives back gv/Δv(k)=gc/Δc(k)g_v/\Delta_v(\boldsymbol{k})=g_c/\Delta_c(\boldsymbol{k}), the condition λk0\lambda_k^0 of Eq. (28). Solving for the alignment,

x∗(k)=gvgv+gc−gcav−gvacΔALM(gv+gc)k2,(50)x^*(\boldsymbol{k})=\frac{g_v}{g_v+g_c}-\frac{g_ca_v-g_va_c}{\Delta_{\mathrm{ALM}}(g_v+g_c)}k^2, \tag*{(50)}

which reduces to Eq. (33) at Γ\Gamma. These are the first two terms of Eq. (49), so the exact sweet spot is the second-order curve lowered by the level shift in units of the donor gap. At Γ\Gamma the drift term drops and

xΓ∗=gvgv+gc−δε0ΔALM,δε0=gv2−gc2ΔALM,(51)x_\Gamma^*=\frac{g_v}{g_v+g_c}-\frac{\delta\varepsilon_0}{\Delta_{\mathrm{ALM}}},\qquad \delta\varepsilon_0=\frac{g_v^2-g_c^2}{\Delta_{\mathrm{ALM}}}, \tag*{(51)}

and when gv=gcg_v=g_c the shift vanishes at every momentum.

Condition for kk-independent sweet spot.—Equation (49) has two sources of momentum dependence: the k2k^2 drift and the running of the shift δεk\delta\varepsilon_k. The drift is a pure k2k^2 term, while δεk\delta\varepsilon_k carries all powers of k2k^2, so each must vanish on its own. The drift vanishes when

gcav=gvac.(52)g_ca_v=g_va_c. \tag*{(52)}

The shift is constant in kk only if gv=gcg_v=g_c, since the factor (1−λR2k2/gvgc)/ΔALM(k)(1-\lambda_R^2k^2/g_vg_c)/\Delta_{\mathrm{ALM}}(\boldsymbol{k}) in Eq. (48) decreases with k2k^2. Equation (A10) then reads av=aca_v=a_c, and xΓ∗(k)=1/2x_\Gamma^*(\boldsymbol{k})=1/2 for every kk, exactly: the two channels are indistinguishable, Δv(k)=Δc(k)\Delta_v(\boldsymbol{k})=\Delta_c(\boldsymbol{k}) at all momenta.

In a real material gv≠gcg_v\ne g_c in general. Equation (A10) alone still removes the leading drift, so we impose it in the following. Since av>0a_v>0 always, it requires ac>0a_c>0, i.e. mc<msm_c<m_s, and written out in masses with Eq. (43) it fixes the host mass,

1ms=1gv+gc(gvmc−gc∣mv∣).(53)\frac{1}{m_s}=\frac{1}{g_v+g_c}\left(\frac{g_v}{m_c}-\frac{g_c}{|m_v|}\right). \tag*{(53)}

Substituting gn/Δn(k)=ε(k)g_n/\Delta_n(\boldsymbol{k})=\varepsilon(\boldsymbol{k}) into Eq. (27) gives

ηkind∣sweet spot=(gv+gcΔALM(k))2[(1−λR2k2gv2)ηv0+(1−λR2k2gc2)ηc0].(54)\left.\eta_{\boldsymbol{k}}^{\mathrm{ind}}\right|_{\mathrm{sweet\ spot}} =\left(\frac{g_v+g_c}{\Delta_{\mathrm{ALM}}(\boldsymbol{k})}\right)^2 \left[\left(1-\frac{\lambda_R^2k^2}{g_v^2}\right)\eta_v^0+ \left(1-\frac{\lambda_R^2k^2}{g_c^2}\right)\eta_c^0\right]. \tag*{(54)}

The shifted gaps Δv(k)+W\Delta_v(\boldsymbol{k})+W and Δc(k)−W\Delta_c(\boldsymbol{k})-W of Eq. (46) have the same ratio ε(k)\varepsilon(\boldsymbol{k}) and give the same result, so we evaluate it at the exact sweet spot. At Γ\Gamma this reduces to ε2(ηv0+ηc0)\varepsilon^2(\eta_v^0+\eta_c^0).

Comparison with the exact spectrum

Figure 6 uses the symmetric and generic parameter sets in Eqs. (40,41), with coupling ε=0.3\varepsilon= 0.3 and as an additional example, with ε=0.4\varepsilon= 0.4. All four sets satisfy Eq. (A10) with the same host mass, ms=3 eV−1m_s = 3\ \mathrm{eV}^{-1}, because Eq. (A11) fixes msm_s through the ratio gv/gcg_v/g_c alone: for the symmetric sets it reduces to av=aca_v = a_c, and for the generic sets the 2:12:1 ratio gives the same value. Raising ε\varepsilon from 0.30.3 to $0.4 therefore changes only the tunneling amplitudes, with the host and donor bands unchanged.

The curves of Fig. 6 show what Eqs. (A7) and (A12) predict for the four sets; the circles are the exact-spectrum check, described in Sec. V. Three observations from the curves:

  1. The symmetric sweet spot does not move: xex∗=1/2x_{\mathrm{ex}}^* = 1/2 at every momentum and at either ε\varepsilon.

  2. The generic sweet spot sits below the second-order value 2/32/3 of (50). Tunneling pushes the in-gap level up by δεk\delta\varepsilon_k, so the electrons there sees a wider balance gap and a narrower conduction gap than the bare alignment suggests. To make the two channels hybridize equal again, the alignment has to move the level back toward the valence band, i.e. to smaller xx.

  3. In (c,d) the circles lie below the curves, and further below in (d) than in (c). The curves, (54), give the splitting ε(k)2(ηv2+ηc2)\varepsilon(k)^2(\eta_v^2+\eta_c^2) assuming the in-gap state is entirely a host state. But that is not the case as part of it sits in the donor. The effective Hamiltonian acts on the host component alone, and the energy of the full eigenstate instead has a weight factor caused by the donors, so the exact splitting is (54) reduced by the host fraction of the state. A larger coupling moves more of the state into the donor, which is why (d) deviates more than (c).

Microscopic model for η0ind\eta_0^{\mathrm{ind}}

The two terms in (31) carry the signs of ηv0\eta_v^0 and ηc0\eta_c^0, because the weights gv2/x2g_v^2/x^2 and gc2/(1−x)2g_c^2/(1-x)^2 are positive. When ηv0\eta_v^0 and ηc0\eta_c^0 have the same sign, the two terms add and η0ind\eta_0^{\mathrm{ind}} is nonzero at every band alignment xx. When they have opposite signs, η0ind\eta_0^{\mathrm{ind}} vanishes at the alignment given by

xη=gv∣ηv0∣gv∣ηv0∣+gc∣ηc0∣.(55)x_\eta= \frac{g_v\sqrt{\lvert\eta_v^0\rvert}}{g_v\sqrt{\lvert\eta_v^0\rvert}+g_c\sqrt{\lvert\eta_c^0\rvert}}. \tag*{(55)}

For ∣ηv0∣=∣ηc0∣\lvert\eta_v^0\rvert= \lvert\eta_c^0\rvert this point coincides with the SOC sweet spot x∗x^* of (33), so the alignment that removes the Rashba coupling would also remove the altermagnetic splitting. Importantly, the relative sign of ηv0\eta_v^0 and ηc0\eta_c^0 is not fixed by symmetry. The C4TC_4T symmetry only requires the spin splitting of each band to have the dd-wave form of (5), but it acts on each band separately and allows either sign of ηv0\eta_v^0 and ηc0\eta_c^0. Thus, the relative sign is determined by the details of the underlying microscopic model.

In the rest of this appendix, we illustrate this using a specific three-band lattice model of an altermagnet relevant to oxychalcogenides. Specifically, we show that both cases can occur and the relative sign of ηv0\eta_v^0 and ηc0\eta_c^0 can be controlled by adjusting the spatial range of the hopping between lattice sites.

We start from the modified Lieb-lattice Hubbard model of Ref. [40], relevant to the transition-metal oxide layers of oxychalcogenides such as La2_2O3_3Mn2_2Se2_2 [27]. Each unit cell has an oxygen site A and two transition-metal sites B and C [Fig. 7(a)], and can be described by Hamiltonian

H=t1∑⟨i,j⟩,σciσ†cjσ+t2∑⟨⟨i,j⟩⟩,σciσ†cjσ+U∑i∈{B,C}ni↑ni↓+ϵA∑i∈Ani−μ∑ini.(56)H = t_1 \sum_{\langle i,j\rangle,\sigma} c_{i\sigma}^{\dagger}c_{j\sigma} + t_2 \sum_{\langle\langle i,j\rangle\rangle,\sigma} c_{i\sigma}^{\dagger}c_{j\sigma} + U \sum_{i\in\{B,C\}} n_{i\uparrow}n_{i\downarrow} + \epsilon_A \sum_{i\in A} n_i - \mu\sum_i n_i. \tag*{(56)}

Here t1t_1 and t2t_2 follow from Figure 7(a), Hubbard UU acts only on transition metal sites B and C, ϵA\epsilon_A is the on-site energy of site A, and μ\mu is the chemical potential. As we elaborate below, this minimal model generically gives opposite signs for ηind\eta^{\mathrm{ind}} in adjacent bands. We thus add a hopping between second neighbor blue sites as illustrated in Figure 7(b),

Lattice models of the altermagnetic donor and their Hartree–Fock bands

Figure 7. Lattice models of the altermagnetic donor and their Hartree–Fock bands. (a) Modified Lieb lattice of Ref. [40] described by Eq. (56): oxygen sites A (red) and transition-metal sites B and C (blue), with hopping t1t_1 between A and B or C, and t2t_2 between B and C. (b) Lattice with the additional hopping t3t_3 of Eq. (57), B to B along $y$ and C to C along $x$. (c) Magnetic order of the Hartree–Fock ground state: each A site holds two electrons with opposite spins, the B sites carry spin up (blue arrows) and the C sites carry spin down (red arrows). (d) Hartree–Fock bands of Eq. (B2) for spin up (blue, solid) and spin down (red, dashed); the dotted line represents the chemical potential $\mu$. (e) Band structure of {xref}Eq. (B3). Parameters used: n=4n=4 electrons per unit cell, t1=−1t_1=-1, t2=0.3t_2=0.3, U=10U=10 and ϵA=0\epsilon_A=0 in all panels, and t3=0.3t_3=0.3 in (b,e).

H3=t3∑j,σ(cj+y^,Bσ†cjBσ+cj+x^,Cσ†cjCσ+h.c.).(57)H_3=t_3\sum_{j,\sigma}\left(c_{j+\hat{y},B\sigma}^{\dagger}c_{jB\sigma}+c_{j+\hat{x},C\sigma}^{\dagger}c_{jC\sigma}+\mathrm{h.c.}\right). \tag*{(57)}

This addition maintains the original C4C_4 symmetry of the lattice and we may thus expect that t3t_3 will be non-zero in realistic situations. We next solve the model with and without H3H_3 in collinear Hartree–Fock theory. At electron filling n=4n=4 both cases give the same altermagnetic spin order illustrated in Figure 7(c). The corresponding band structures are given in Figure 7(d,Figure 7e). From these we can easily deduce signs of ηind\eta^{\mathrm{ind}}. In fact, since E↑−E↓=−2ηx0kx2E_{\uparrow}-E_{\downarrow}=-2\eta_x^0 k_x^2 along Γ\Gamma–X, the bands give the signs of ηv0\eta_v^0 and ηc0\eta_c^0 directly. Without H3H_3 they are opposite, and the two contributions to ηind\eta^{\mathrm{ind}} cancel at xηx_\eta. On the other hand, with H3H_3 included and t3≳0.1t_3\gtrsim0.1 the conduction-band splitting reverses: ηv0\eta_v^0 and ηc0\eta_c^0 are both positive, as in Eq. (40), and the two contributions add.

We conclude that at the level of models, a single hopping term that leaves the magnetic order unchanged can control the sign structure, which therefore has to be determined for each material separately. Inspecting the band structures for Lieb-lattice altermagnets computed in the existing literature we find that La2_2O3_3Mn2_2Se2_2 [27] exhibit the same sign of ηv0\eta_v^0 and ηc0\eta_c^0, while iron-based transition metal chalcogenides [30] and the monolayer V2_2Se2_2O [31] exhibit opposite signs.

References

  1. [1]A. Ohtomo and H. Y. Hwang, Nature 427, 423 (2004).
  2. [2]A. A. Burkov and L. Balents, Phys. Rev. Lett. 107, 127205 (2011).
  3. [3]L. Fu and C. L. Kane, Phys. Rev. Lett. 100, 096407 (2008).
  4. [4]J. D. Sau, R. M. Lutchyn, S. Tewari, and S. Das Sarma, Physical review letters 104, 040502 (2010).
  5. [5]J. Alicea, Physical Review B—Condensed Matter and Materials Physics 81, 125318 (2010).
  6. [6]Y. Oreg, G. Refael, and F. Von Oppen, Physical review letters 105, 177002 (2010).
  7. [7]R. M. Lutchyn, J. D. Sau, and S. Das Sarma, Phys. Rev. Lett. 105, 077001 (2010).
  8. [8]Y. Cao, V. Fatemi, S. Fang, K. Watanabe, T. Taniguchi, E. Kaxiras, and P. Jarillo-Herrero, Nature 556, 43 (2018).
  9. [9]O. Can, T. Tummuru, R. P. Day, I. Elfimov, A. Damascelli, and M. Franz, Nature Physics 17, 519 (2021).
  10. [10]L. Šmejkal, J. Sinova, and T. Jungwirth, Phys. Rev. X 12, 031042 (2022).
  11. [11]L. Šmejkal, J. Sinova, and T. Jungwirth, Phys. Rev. X 12, 040501 (2022).
  12. [12]I. Mazin (The PRX Editors), Phys. Rev. X 12, 040002 (2022).
  13. [13]T. Jungwirth, R. M. Fernandes, E. Fradkin, A. H. MacDonald, J. Sinova, and L. Šmejkal, Newton, Newton 1 (2025), 10.1016/j.newton.2025.100162.
  14. [14]I. I. Mazin, Phys. Rev. B 107, L100418 (2023).
  15. [15]J. Krempaský, L. Šmejkal, S. D’Souza, M. Hajlaoui, G. Springholz, K. Uhlířová, F. Alarab, P. Constantinou, V. Strocov, D. Usanov, W. Pudelko, R. Gonzalez-Hernandez, A. Hellenes, Z. Jansa, H. Reichlová, Z. Šobáň, R. Betancourt, P. Wadley, J. Sinova, and T. Jungwirth, Nature 626, 517 (2024).
  16. [16]O. J. Amin, A. Dal Din, E. Golias, Y. Niu, A. Zakharov, S. C. Fromage, C. J. B. Fields, S. L. Heywood, R. B. Cousins, F. Macchererzo, J. Krempaský, J. H. Dil, D. Kriegner, B. Kiraly, B. P. Campion, A. W. Brushforth, K. W. Edmonds, S. S. Dhesi, L. Šmejkal, T. Jungwirth, and P. Wadley, Nature 636, 348 (2024).
  17. [17]J. Ding, X. Jiang, X. Chen, Z. Tao, Z. Liu, T. Li, J. Liu, J. Sun, J. Cheng, J. Liu, Y. Yang, R. Zhang, L. Deng, W. Jing, Y. Huang, Y. Shi, M. Ye, S. Qiao, Y. Wang, Y. Guo, D. Feng, and D. Shen, Phys. Rev. Lett. 133, 206401 (2024).
  18. [18]S. Reimers, L. Odenbreit, L. Šmejkal, V. N. Strocov, P. Constantinou, A. B. Hellenes, R. Jaeschke-Uberigo, W. H. Campos, V. K. Bharadwaj, A. Chakraborty, T. Denneulin, W. Shi, R. E. Dunin-Borkowski, S. Das, M. Kläui, J. Sinova, and M. Jourdan, Nature Communications 15, 2116 (2024).
  19. [19]R. González-Hernández, L. Šmejkal, K. Výborný, Y. Yahagi, J. Sinova, T. c. v. Jungwirth, and J. Železný, Phys. Rev. Lett. 126, 127701 (2021).
  20. [20]R. Zarzuela, R. Jaeschke-Uberigo, O. Gomonay, L. Šmejkal, and J. Sinova, Phys. Rev. B 111, 064422 (2025).
  21. [21]K. Monkman, J. Weng, N. Heinsdorf, A. Nocera, Y. Barlas, and M. Franz, Phys. Rev. X 16, 011057 (2026).
  22. [22]J. Kistner-Morris, A. Shi, E. Liu, T. Arp, F. Farahmand, T. Taniguchi, K. Watanabe, V. Aji, C. H. Lui, and N. Gabor, Nature Communications 15, 4075 (2024).
  23. [23]I. Žutić, J. Fabian, and S. Das Sarma, Rev. Mod. Phys. 76, 323 (2004).
  24. [24]J. Sinova, S. O. Valenzuela, J. Wunderlich, C. H. Back, and T. Jungwirth, Rev. Mod. Phys. 87, 1213 (2015).
  25. [25]P.-H. Fu, Q. Lv, Y. Xu, J. Cavao, J.-F. Liu, and X.-L. Yu, npj Quantum Materials 10, 111 (2025).
  26. [26]T. Jungwirth, J. Sinova, P. Wadley, D. Kriegner, H. Re- ichlová, F. Krizek, H. Ohno, and L. Šmejkal, Nature Physics 22, 1012 (2026).
  27. [27]C.-C. Wei, X. Li, S. Hatt, X. Huai, J. Liu, B. Singh, K.-M. Kim, r. M. Fernandes, P. Cardon, L. Zhao, T. T. Tran, B. A. Frandsen, K. S. Burch, F. Liu, and H. Ji, Phys. Rev. Mater. 9, 024402 (2025).
  28. [28]S. Asai, J. Hu, Z. Liu, S. Itoh, D. Ueta, J. Matsuda, T. Hatanaka, H. Watanabe, R. Arita, and T. Masuda, D. Kriegner, B. Kiraly, B. P. Campion, A. W. Brushforth, K. W. Edmonds, S. S. Dhesi, L. Šmejkal, T. Jungwirth, and P. Wadley, Nature 636, 348 (2024).
  29. [29]G. Zhang, Y. Li, X. Ye, V. C. Morano, S. T. Li, J. Choi, R. Scatena, S. Tang, M. Pi, M. Ye, M. Garcia-Fernandez, A. Bombardi, X. Qin, Z. Pan, D. G. Mazzone, Q. Wang, Y. Lu, Y. Shen, and Y. Long, “d-wave altermagnetism revealed by resonant inelastic x-ray scattering,” (2026), arXiv:2606.18731 [cond-mat.str-el].arxiv.org/abs/2606.18731
  30. [30]D.-F. Fan, Y.-K. Wang, J.-Y. Li, and S. Li, Phys. Rev. B 112, 253135 (2025).
  31. [31]H.-Y. Ma, M. Hu, N. Li, J. Liu, W. Yao, J.-F. Jia, and J. Liu, Nature Communications 12, 2846 (2021).
  32. [32]J. C. Abadillo-Uriel, A. Maiani, A. Cortijo, R. Aguado, and R. S. Souto, “All-electrical dephasing-protected spin qubits in altermagnets,” (2026), arXiv:2606.26066 [cond-mat.mes-hall].
  33. [33]H. Vakili, “Gate-controlled spin qubits in confined altermagnets,” (2026), arXiv:2606.24150 [cond-mat.mes-hall].arxiv.org/abs/2606.24150
  34. [34]Z. Zhu, R. Huang, X. Chen, Z. Cui, X. Duan, J. Zhang, I. Žutić, and T. Zhou, Phys. Rev. Lett. 136, 186702 (2026).
  35. [35]Y. Yao, F. Ye, X.-L. Qi, S.-C. Zhang, and Z. Fang, Phys. Rev. B 75, 041401(R) (2007).
  36. [36]N. Heinsdorf and M. Franz, Phys. Rev. B 113, L020501 (2026).
  37. [37]M. J. Manfra, N. G. Weimann, J. W. P. Hsu, L. N. Pfeiffer, K. W. West, S. Syed, H. L. Stormer, W. Pan, D. V. Lang, S. N. G. Chu, G. Kowach, A. M. Sergent, J. Caissie, K. M. Molvar, L. J. Mahoney, and R. J. Molnar, J. Appl. Phys. 92, 338 (2002).
  38. [38]W. Xiong, C. Xia, J. Du, T. Wang, X. Zhao, Y. Peng, Z. Wei, and J. Li, Phys. Rev. B 95, 245408 (2017).
  39. [39]C. Q. Nguyen, Y. S. Ang, S.-T. Nguyen, N. V. Hoang, N. M. Hung, and C. V. Nguyen, Phys. Rev. B 105, 045303 (2022).
  40. [40]N. Kaushal and M. Franz, Phys. Rev. Lett. 135, 156502 (2025).

Paper details

vlnc:W_s9katr9y
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

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

  • Physics and Astronomy

Claude

View statement
Looking for related papers…
Contents