WorksPhysics and Astronomy
Stacking-Directed Polarization and Excitonic Engineering in MoS/MoSe van der Waals Heterostructures
Abstract
The stacking-dependent polarization and excitonic response of MoS/MoSe heterostructures were investigated using GW+BSE many-body perturbation theory. While homobilayer MoS exhibited a switchable interlayer dipole driven by registry-induced symmetry breaking, the MoS/MoSe hetero-interface remained pinned by the intrinsic chemical potential mismatch between sulfur and selenium. In 2L-MoS/MoSe trilayers, the stacking sequence enabled a deterministic control of photogenerated electrons between the central and bottom MoS layers, governed by internal electric fields and quasiparticle band-edge shifts of 60–70~meV. Our calculations predicted a 36~meV interlayer excitonic shift, in remarkable agreement with recent experiments. These results elucidate the microscopic link between atomic registry and many-body interactions, establishing transition metal dichalcogenide trilayers as a potential platform for sliding ferroelectricity and programmable optoelectronic functionalities.
Introduction
Two-dimensional (2D) van der Waals (vdW) heterostructures, formed by stacking atomically thin crystals via weak interlayer interactions, have emerged as highly versatile platforms for exploring quantum phenomena and engineering optoelectronic functionalities beyond what bulk materials can offer [1, 2, 3, 4, 5, 6, 7]. The vdW nature of these interactions facilitates precise control over stacking sequence, rotational alignment, and interlayer spacing, giving rise to Moiré superlattices and enabling atomic-scale tuning of electronic band structures, excitonic resonances, and interfacial polarization. These properties make vdW heterostructures ideal for manipulating light–matter interactions at the atomic scale.
Among 2D materials, transition metal dichalcogenides (TMDs) such as , , , and have attracted particular attention due to their direct band gaps in the visible–near-infrared range, strong spin–orbit coupling, and tightly bound excitons with binding energies of 0.5–0.9 eV [8, 9, 10, 11, 12]. These high binding energies, nearly an order of magnitude larger than in conventional bulk semiconductors, arise from reduced dielectric screening and enhanced Coulomb interactions in two dimensions. Consequently, monolayer TMDs exhibit exceptionally strong light absorption, up to 5–10% of incident light per layer, and robust valley-selective optical transitions, making them ideal candidates for excitonic and valleytronic applications [10, 11, 13].
When stacked into heterostructures, TMD monolayers often display type-II band alignment, which spatially separates electrons and holes and stabilizes interlayer excitons with large out-of-plane dipole moments and extended lifetimes [7, 11]. For instance, in -based heterostructures, conduction- and valence-band offsets of 0.3–0.6 eV drive ultrafast interlayer charge transfer on sub-100 fs timescales [14, 15]. The spatial separation of carriers suppresses radiative recombination and renders interlayer excitons highly sensitive to interfacial asymmetry and external fields, providing versatile opportunities for optoelectronic and quantum-functional devices.
The electronic and excitonic properties of vdW heterostructures are strongly dependent on interlayer arrangement, atomic registry, and stacking symmetry. Variations in stacking alter interlayer orbital hybridization, local electrostatics, and charge distribution, which in turn modify band edges, exciton binding energies, and optical selection rules. In twisted heterobilayers, small relative rotations produce long-period Moiré superlattices that confine electrons, holes, and interlayer excitons, flatten excitonic minibands, and enable discrete exciton arrays with correlated phases and single-photon emission [16, 17, 18, 19, 20]. For twist angles below approximately , Moiré periods can exceed 10–15 nm, generating deep periodic potentials that strongly influence exciton localization and recombination dynamics.
Interlayer excitons in TMD heterobilayers exhibit lifetimes that are orders of magnitude longer than those of intralayer excitons, reflecting their spatially indirect character. For example, vertically stacked – heterostructures show interlayer exciton lifetimes of approximately 1.8 ns at low temperature, compared to tens of picoseconds for intralayer excitons [21]. Moiré engineering at small twist angles can further extend lifetimes into the tens-of-nanoseconds range by suppressing nonradiative decay pathways [22]. These long-lived excitons, combined with strong binding energies, underpin robust light–matter interactions and highlight the potential of vdW heterostructures for excitonic devices.
While Moiré superlattices provide a static landscape for particle localization, lateral interlayer sliding offers a continuous and dynamical pathway to sample the stacking configuration space without changing the Moiré period [23, 24]. By systematically shifting the stacking configuration, sliding can reversibly modulate interfacial polarization, band topology, charge transfer, and exciton energies. This registry-dependent control complements Moiré engineering, enabling the design of heterostructures with electrically controllable dipoles and tailored excitonic landscapes [25, 26].
The interplay between interlayer polarization and excitonic dynamics opens additional avenues for quantum-functional design. In inversion-asymmetric or 3R-stacked TMD heterostructures, spontaneous out-of-plane polarizations on the order of – C m produce internal fields up to – V nm, sufficient to shift band edges, redistribute charge, and tune interlayer exciton energies [27, 28]. Controlled stacking and polarization engineering thus enable electrically programmable excitonic landscapes, ferroelectric switching, and multifunctional optoelectronic behavior [29].
Despite these advances, a systematic understanding of how atomic registry governs many-body excitonic behavior in MoS-based heterostructures remains elusive. Existing theoretical frameworks often rely on mean-field approximations that fail to account for the non-local screening and enhanced Coulomb interactions inherent to these 2D interfaces. Consequently, the fundamental relationship between local stacking order and the resulting interfacial dipoles particularly at the many-body GW–BSE level requires in-depth elucidation. This motivates a rigorous computational study of the electronic and excitonic landscapes in these systems.
Here, we employ a high-fidelity theoretical framework to investigate stacking-dependent polarization and excitonic properties in MoS-based vdW heterostructures, building upon recent experimental demonstrations of ferroelectric control in 3R-stacked systems [29]. Utilizing density functional theory (DFT) with hybrid functionals and many-body perturbation theory within the EVGW (eigenvalue-self-consistent GW)–BSE framework, we move beyond standard approximations to precisely quantify how atomic registry and interlayer asymmetry govern interfacial dipoles and quasiparticle band alignments. We demonstrate that AB and BA registries generate interfacial dipoles of opposite polarity, resulting in switchable band-edge shifts while maintaining the robust type-II alignment necessary for charge separation. Furthermore, we show that in trilayer configurations, stacking-induced asymmetry allows for the fine-tuning of polarization states and excitonic transitions, offering a predictive road map for the design of non-volatile excitonic transistors and atomically controlled ferroelectric optoelectronic logic.
Computational Framework
The structural and electronic properties of MoS/MoSe heterostructures were investigated within the framework of DFT, as implemented in the Vienna Ab initio Simulation Package (VASP) [30, 31, 32, 33]. Electron–ion interactions were described using the projector-augmented wave (PAW) formalism, with valence configurations of Mo (), S (), and Se (). Exchange-correlation effects were treated within the generalized gradient approximation using the Perdew–Burke–Ernzerhof functional [34], supplemented by Grimme’s D3 scheme to account for long-range dispersion interactions.
All calculations were performed using a plane-wave kinetic-energy cutoff of 450 eV and -centered Monkhorst–Pack -point meshes of . A vacuum region exceeding 20 Å was introduced along the out-of-plane direction to suppress spurious interactions between periodic replicas. Atomic coordinates were fully optimized until residual forces were below 0.01 eV Å and total-energy variations were smaller than eV. Systematic convergence tests with respect to the -point grid (ranging from to ) confirm that total energy differences remain within less than 0.3 meV per unit cell from onward, validating our production grid selection (see Supplemental Material (Suppl. Mat.), Fig. S1).
Commensurate heterostructure models were constructed by first optimizing isolated MoS and MoSe monolayers. Monolayers of MoS and MoSe were found to have lattice constants of 3.166 Å and 3.295 Å, respectively. For bilayers and heterobilayers, AA, AB, and BA stacking configurations were considered, while multiple trilayer stackings (AAA, ABA, ABB, BAA, BAB, AAB, and BBA) were systematically analyzed. Lattice matching in heterobilayers was achieved by adopting a common in-plane lattice parameter, which corresponds to biaxial strains of +2.1% in MoS and -1.9% in MoSe. While such strain magnitudes can quantitatively shift absolute band edges and excitonic transition energies, the relative energetic trends and stability hierarchies among competing local stacking registries remain robust, as all compared configurations experience identical uniform strain fields within the rigid commensurate model. Furthermore, while experimental samples with small lattice mismatches typically form long-period moiré superlattices accompanied by local atomic and strain relaxation, our commensurate framework provides fundamental insights into the local stacking-dependent potential energy landscape and intrinsic energetic hierarchies governing these van der Waals heterostructures.
Electronic band structures and band gaps were evaluated at multiple theoretical levels. DFT calculations were performed using PBE, including spin–orbit coupling (SOC) where indicated, as well as the screened hybrid HSE06 functional without SOC. Many-body effects beyond DFT were incorporated through EVGW@PBE [35, 36], in which quasiparticle energies in the Green’s function were iteratively updated while the orbitals and screened Coulomb interaction were kept fixed at the PBE level. Excitonic effects in the optical response were explicitly treated by solving the Bethe–Salpeter equation (BSE) [37, 38].
All EVGW and BSE calculations employed PAW datasets optimized for many-body perturbation theory, including semicore states, together with carefully converged -point meshes and numbers of unoccupied bands. Bilayer MoS and MoS/MoSe heterobilayer were computed using a -centered -point grid with 32 unoccupied bands, while the 2L-MoS/MoSe systems used a -centered grid with 64 unoccupied bands. All numerical parameters were systematically tested to ensure the reliability of quasiparticle energies and excitonic spectra.
Results and Discussion
Bilayer MoS and MoS/MoSe Heterobilayers
Stacking-Dependent Polarization and Interlayer Charge Redistribution
The impact of stacking registry on interlayer coupling in both homobilayer MoS and MoS/MoSe heterobilayers was systematically evaluated for AA, AB, and BA configurations. The calculated in-plane lattice constants () and out-of-plane lattice parameters () for monolayer and bilayer stackings are summarized in Table S1, providing a quantitative basis for subsequent analysis of interlayer interactions.
| System | Stacking | () | () | () | () |
| Bilayer MoS | AA | 3.165 | 35.45 | 6.74 | – |
| Bilayer MoS | AB | 3.161 | 35.82 | 6.43 | – |
| Bilayer MoS | BA | 3.164 | 35.76 | 6.43 | – |
| MoS/MoSe Bilayer | AA | 3.227 | 35.87 | 6.94 | – |
| MoS/MoSe Bilayer | AB | 3.228 | 35.83 | 6.62 | – |
| MoS/MoSe Bilayer | BA | 3.228 | 35.83 | 6.62 | – |
| Trilayer 2L-MoS/MoSe | AAA | 3.205 | 32.19 | 6.70 | 6.96 |
| Trilayer 2L-MoS/MoSe | ABA | 3.207 | 36.31 | 6.41 | 6.64 |
| Trilayer 2L-MoS/MoSe | BAA | 3.206 | 36.34 | 6.39 | 6.66 |
| Trilayer 2L-MoS/MoSe | ABB | 3.202 | 36.58 | 6.42 | 6.65 |
| Trilayer 2L-MoS/MoSe | BAB | 3.208 | 36.29 | 6.40 | 6.63 |
| Trilayer 2L-MoS/MoSe | AAB | 3.206 | 36.61 | 6.43 | 6.67 |
| Trilayer 2L-MoS/MoSe | BBA | 3.207 | 36.31 | 6.41 | 6.64 |
Table S1. Calculated in-plane lattice constants (), out-of-plane supercell parameters (), and interlayer distances (, ) for bilayer MoS, MoS/MoSe heterobilayers, and 2L-MoS/MoSe trilayers.
Following full structural relaxation, the AA registry preserves high structural symmetry, whereas the staggered AB and BA configurations break global inversion symmetry, thereby inducing a finite out-of-plane electric polarization, . The optimized geometries for these polar registries, including both top and side views, are illustrated for the homobilayer in Figure 1a,Figure 1b and for the heterobilayer in Figure 1c,Figure 1d, respectively. Corresponding views of the non-polar AA stacking are provided in Figure S2,Figure S2 of the Suppl. Mat.

Figure 1. Top and side views of the atomic configurations for (a) AB-stacked and (b) BA-stacked MoS homobilayers, and (c) AB-stacked and (d) BA-stacked MoS/MoSe heterobilayers. Panels (e)–(h) display the corresponding electron localization function (ELF) contour plots with an interval of 0.1, illustrating the covalent nature of intralayer bonding and the vdW gap. Panels (i)–(l) show the charge density difference () isosurfaces for each configuration. Red and green regions represent charge depletion and accumulation, respectively, visualized at an isovalue of 0.001 e/Å. The variations in stacking order and chemical composition significantly modulate the interlayer coupling and the resulting interfacial charge redistribution.

Figure S2. (a, b) Top and side views of the AA-stacked bilayer MoS (a) and the MoS/MoSe heterobilayer (b), along with their corresponding charge density difference (CDD) isosurfaces. (c, d) Planar-averaged CDD profiles for the bilayer MoS (c) and MoS/MoSe heterobilayer (d). Red and green indicate charge depletion and accumulation, respectively.
The evolution of the lattice parameters from monolayer to bilayer provides insight into the interplay between intralayer strain, interlayer steric repulsion, and vdW interactions. As listed in Table S1, monolayer MoS and MoSe exhibit equilibrium in-plane lattice constants of 3.166 Å and 3.295 Å, respectively. Upon forming a heterobilayer, the in-plane lattice constant converges to approximately 3.228 Å, effectively representing an arithmetic mean of the constituent monolayers. This indicates a commensurate strain regime in which the energy cost of stretching MoS is balanced by compressing MoSe, enabling commensurable stacking with minimal total elastic energy. For the homobilayer MoS, AB and BA stackings show only minor deviations from the monolayer due to symmetric strain distribution between identical layers.
The out-of-plane lattice constant is strongly influenced by stacking registry. In the AA configuration, Mo atoms and chalcogen atoms from opposing layers are vertically aligned, maximizing electron cloud overlap and inducing strong Pauli repulsion. This steric effect pushes the layers apart, resulting in the largest values and Mo–Mo interlayer separations (), reaching 6.74 Å for homobilayer MoS and 6.94 Å for MoS/MoSe. In contrast, the staggered AB and BA configurations allow lateral atomic offsets, reducing localized electronic repulsion and enabling attractive vdW interactions to bring the layers closer to each other. In these registries, decreases to 6.43 Å for MoS and 6.62 Å for MoS/MoSe, with corresponding contractions in . The systematic expansion of approximately 0.20 Å in the heterobilayer relative to the homobilayer reflects the larger vdW radius of Se compared to S.
To characterize the bonding nature and spatial distribution of electrons, the electron localization function (ELF) was analyzed, as illustrated in Figure 1e–Figure 1h. ELF values approaching unity indicate regions of strong covalent bonding or lone-pair character, whereas values near zero denote regions of low electron density. Across all configurations, a pronounced low-ELF region (ELF ) is observed within the interlayer space, identifying the vdW gap and confirming the absence of significant orbital overlap.
The intralayer bonding characteristics are further elucidated by these ELF distributions. In the AB-stacked systems (Figure 1e,Figure 1g), the first layer exhibits moderate localization around chalcogen sites, while the second layer shows stronger, more spatially extended localization, indicated by orange-to-red regions. In the BA stacking (Figure 1f,Figure 1h), this behavior is inverted, demonstrating a stacking-dependent redistribution of the local electronic environment and highlighting the sensitivity of intralayer charge to the vertical atomic registry.
Interlayer charge redistribution was quantified using charge density difference (CDD) analysis. The three-dimensional CDD isosurfaces, illustrated in Figure 1i–Figure 1l, use red and green contours to indicate charge depletion and accumulation, respectively (isovalue = 0.001 e/Å). The CDD, , is defined as
where , , and are the charge densities of the bilayer and isolated monolayers at the same atomic coordinates. The planar-averaged redistribution perpendicular to the interface (along the -axis) is
where denotes the in-plane unit-cell area, yielding units of e/Å for . The associated out-of-plane (in stacking direction) electric polarization (dipole moment per unit area) is given by
yielding surface polarization units of e/Å (equivalently converted to pC/m). The stacking registry dictates the electronic landscape of the bilayers through a competition between structural symmetry and local orbital hybridization. In AA-stacked bilayers, the preservation of an interlayer mirror symmetry plane () enforces a symmetric charge redistribution, , fundamentally precluding an out-of-plane dipole (). This is corroborated by the symmetry-protected CDD isosurfaces (Figure S2,Figure S2) and planar-averaged profiles (Figure S2,Figure S2), which exhibit distributions centered symmetrically across the vdW gap. The vertical alignment of Mo atoms and opposing chalcogen planes in layers L1 and L2 maintains a balanced interlayer hybridization, preserving a non-polar, purely vdW-mediated interaction.
In contrast, the staggered AB and BA registries facilitate a pronounced, asymmetric interlayer charge transfer. For the MoS homobilayer in the AB configuration (Figure 1i and Figure 2a), a spontaneous electronic redistribution occurs, with charge depletion (holes) localized near the lower layer (L1, red region) and charge accumulation (electrons) concentrated near the top layer (L2, green region). This spatial separation of charge establishes a net upward-oriented electric polarization, (↑). The redistribution originates from the broken global inversion symmetry inherent to staggered registries, which permits asymmetric hybridization between the metal Mo-4 and chalcogen S-3 orbitals. This effect is quantified by the planar-averaged , which exhibits extrema of +0.005 and −0.007 e/Å. A structural transition to the BA configuration (Figure 1j and Figure 2b) effectively inverts this electronic profile, reversing the dipole orientation to (↓). The existence of these two energetically degenerate states with opposing dipole moments confirms a switchable, bistable interlayer configuration, representing the essential microscopic prerequisite for sliding ferroelectricity in MoS bilayers.

Figure 2. Planar-averaged charge density difference () profiles for MoS and MoS/MoSe bilayers. Panels (a)–(d) display the out-of-plane () distribution of charge for AB- and BA-stacked configurations. The top insets map the 3D isosurfaces to the 1D profiles; red and green vertical lines correlate the quantitative maxima (Max) and minima (Min) with physical regions of electron accumulation and depletion, respectively. The sequence of these extrema (MinMax vs. MaxMin) delineates the interfacial dipole orientation, illustrating how lateral sliding between AB and BA stackings reverses or modulates the internal electric field and interlayer coupling.
For the chemically asymmetric MoS/MoSe heterobilayer Figure 1k,Figure 1l, localized electronic redistribution occurs due to the electronegativity mismatch between sulfur and selenium, exhibiting charge depletion on the MoS side and accumulation on the MoSe side. However, quantitative evaluation of the macroscopic polarization yields a negligible net value of C/m, substantially smaller than the switchable homobilayer polarization of C/m. This highlights a crucial physical distinction: while local CDD isosurfaces capture microscopic orbital asymmetry and bonding variations at the interface, they do not inherently translate into a finite macroscopic dipole moment after full unit-cell integration. Consequently, the planar-averaged electrostatic potential across the heterobilayer reveals an approximately vanishing potential offset (, ). This demonstrates that local asymmetric charge transfer is counterbalanced by the intrinsic chemical potential profile, chemically pinning the electronic structure and suppressing switchable out-of-plane ferroelectricity.
The electrostatic consequences of charge redistribution are quantified by the planaraveraged electrostatic potential , which is coupled to the charge density difference via Poisson’s equation:
The resulting out-of-plane electric field, , is determined by the first moment of the redistributed charge. A macroscopic potential shift across the interlayer distance generates an effective internal field,
For the AB-stacked homobilayer (Figure 3a), the planar-averaged potential reveals a distinct layer-to-layer offset of meV. To ensure methodological consistency, we applied a macroscopic averaging technique to filter out periodic variations; this confirms the offset is robust and yields an internal electric field of V/nm directed from L2 to L1. Inverting the registry to the BA configuration (Suppl. Mat., Fig. S3a) reverses the sign of the potential difference ( meV) and the vector of the internal field. Since the constituent layers are chemically identical, this polarization is an emergent phenomenon driven by registry-induced symmetry breaking and interlayer orbital hybridization.

Figure 3. Planar-averaged electrostatic potentials, quasiparticle band structures, and band alignments. (a)–(b) Out-of-plane electrostatic potential profiles for AB-stacked and , respectively; and indicate the vacuum level and work function. (c)–(f) Quasiparticle band structures calculated at the EVGW0@PBE level for AB and BA stacking orders. The arrows highlight stacking-dependent shifts in the and orbitals at the point. The right-hand schematics illustrate the resulting Type-II band alignments between layers L1 and L2.
In contrast, the heterobilayer (Figure 3b) exhibits a qualitatively distinct electrostatic profile. The planar-averaged potential is dominated by large variations of the peaks ( to eV) corresponding to the ionic cores of the Se and Mo planes. While these features reflect the intrinsic electronegativity mismatch between S and Se, they represent local atomic-scale variations that do not contribute to a macroscopic dipole. Using the same macroscopic averaging technique applied to the homobilayer, the potential offset for the heterobilayer is found to be negligible (Suppl. Mat., Fig. S3b), such that:
Consequently, despite the pronounced local chemical asymmetry, the interface does not support a stacking-dependent, switchable internal electric field. The intrinsic chemical potential gradient effectively pins the electronic state, suppressing the collective, registry-dependent charge transfer required for the bistable dipolar states observed in the homobilayer.
The calculated vacuum levels and work functions provide a complementary perspective on the electrostatic environment of these layered systems (Table 1). Bilayer exhibits identical vacuum levels of 4.49 eV for both AB and BA stackings, with corresponding work functions of approximately 5.10 eV, consistent with the presence of a small, reversible interlayer dipole. In contrast, the heterobilayer shows elevated vacuum levels of 4.96 eV (AB) and 4.99 eV (BA), and work functions of 5.3 eV, reflecting the chemical asymmetry of the interface. These values corroborate the electrostatic analysis since the homobilayer supports a bistable, stacking-dependent dipole manifested as the 55 meV potential step, whereas the heterobilayer lacks a switchable internal field and the macroscopic potential is dominated by intrinsic chemical offsets rather than stacking-induced charge redistribution.
| System | Stacking | (eV) | (eV) | (eV) | (eV) | (eV) | (eV) | PL Energy (eV) | (eV) |
| Bilayer MoS | AB | 4.49 | 5.10 | 1.23 | 1.22 | 1.74 | 2.26 | 2.08 | 0.18 |
| BA | 4.49 | 5.10 | 1.25 | 1.24 | 1.77 | 2.29 | 2.07 | 0.22 | |
| MoS/MoSe | AB | 4.96 | 5.30 | 0.85 | 0.76 | 1.24 | 1.89 | 1.45 | 0.44 |
| BA | 4.99 | 5.30 | 0.75 | 0.66 | 1.13 | 1.78 | 1.34 | 0.44 |
Table 1. Calculated vacuum levels (), work functions (), electronic band gaps () obtained with PBE, PBE+SOC, HSE06, and EVGW0@PBE, photoluminescence (PL) energies, and exciton binding energies () of the lowest excitonic transition for bilayer MoS and MoS/MoSe heterobilayers in AB and BA stackings.
Stacking-Dependent Electronic Structure and Band Alignment
To establish the baseline electronic landscape of the isolated components, the fundamental band gaps, absolute vacuum-referenced valence-band maxima (), and conduction-band minima () for both unstrained and strained monolayers of and were evaluated (Table S2). As summarized in Table S2, mechanical strain induces a transition from direct to indirect band gaps and shifts the band edges relative to the vacuum level. Furthermore, the corresponding directional variations in electron and hole effective masses ( and ) along the high-symmetry and paths (Table S3) reveal that lattice deformation significantly modifies band curvature, directly impacting carrier mobility and effective mass anisotropy.
| System | Character | (eV) | (eV) | (eV) |
| MoS (Unstrained) | Direct () | 1.734 | −5.923 | −4.189 |
| MoS (+2.1% Strain) | Indirect () | 1.294 | −5.8211 | −4.528 |
| MoSe (Unstrained) | Direct () | 1.514 | −5.296 | −3.782 |
| MoSe (−1.9% Strain) | Indirect | 1.482 | −5.208 | −3.726 |
Table S2. Calculated fundamental band gaps () and vacuum-referenced valence-band maximum () and conduction-band minimum () for unstrained and strained monolayers of MoS and MoSe (PBE functional).
| System | Direction | ||
| (Unstrained) | 0.824 | 1.885 | |
| (Unstrained) | 0.560 | ||
| (+2.1% Strain) | 1.676 | 1.626 | |
| (+2.1% Strain) | 1.920 | 0.568 | |
| (Unstrained) | 0.645 | 2.615 | |
| (Unstrained) | 1.352 | 0.259 | |
| ( Strain) | 0.533 | 3.848 | |
| ( Strain) | 1.051 |
Table S3. Calculated electron () and hole () effective masses (in units of ) along and for unstrained and strained monolayers of and (PBE functional).
The electronic properties of homobilayer and heterobilayers were evaluated using PBE, PBE+SOC, and HSE06 functionals to investigate the impact of stacking configuration on the band gap and band-edge localization. To accurately account for many-body effects and the reduced dielectric screening characteristic of two-dimensional systems, eigenvalue-self-consistent quasiparticle calculations were performed within the framework. Summations over unoccupied states were converged using 64 empty conduction bands, ensuring stable numerical convergence of the polarizability matrix and the BSE Hamiltonian within less than 5 meV uncertainty. Furthermore, while spin-orbit coupling (SOC) is explicitly evaluated at the PBE level (PBE+SOC), the advanced HSE06 and +BSE calculations were performed without SOC due to computational scaling limits. Because intra-atomic SOC shifts are nearly identical across rigid sliding registries (AB vs. BA), differential metrics such as the stacking-dependent interlayer exciton shifts are robust against this approximation. The resulting band gap renormalization and the complex interplay between stacking registry and electronic structure are illustrated in Figure 3c–Figure 3f, while a comprehensive comparison across the PBE, PBE+SOC, and HSE06 levels including fully layer- and orbital-resolved band structures is provided in the Suppl. Mat. (Figs. S4–S7).
For the MoS homobilayer, both AB and BA stackings exhibit an indirect – band gap. At the PBE level, the band gaps are approximately eV for AB and 1.25 eV for BA. Including SOC reduces the gaps slightly to about 1.22 eV for AB and 1.24 eV for BA, which is consistent with the moderate spin splitting of the Mo states. The HSE06 hybrid functional increases the gaps by roughly 40%, yielding eV for AB and 1.74 eV for BA, while EVGW@PBE further enhances them to 2.26 eV for AB and 2.29 eV for BA (Table 1).
This indirect character originates from orbital-selective interlayer hybridization. The VBM and CBM at the -point are dominated by in-plane Mo orbitals, which suppress interlayer coupling and effectively pin the localized states at . Stacking-dependent shifts of these and states, and , are highlighted by the arrows in Figure 3c,Figure 3d, demonstrating how registry modulates local orbital energies. In the AB configuration, the planar-averaged electrostatic potential profile (Figure 3a) reveals a distinct intrinsic potential offset of meV between the layers, indicating a stacking-dependent dipole. This macroscopic electrostatic environment acts as a perturbation that modifies the electronic states, resulting in band offsets of meV and meV (as derived from the electronic band structure). This relationship is precisely inverted in the BA configuration due to the spatial reversal of the layer symmetry.
The -point valence states are primarily composed of out-of-plane Mo and S orbitals, where the interlayer coupling produces a pronounced bonding–antibonding splitting of eV (cf. Fig. S5). The electronic coupling at can be estimated using a two-level approximation at the -point. For a homobilayer the uncoupled VBM orbitals are degenerate. However, due to the static electric field there is an offset of eV as given in the previous subsection. Thus the effective splitting is given by . From this expression the coupling (hopping matrix) follows as eV.
The considerable magnitude of this extracted hopping parameter highlights the strong interlayer electronic hybridization occurring despite the predominantly vdW nature of the interlayer assembly. This coupling arises from the non-negligible spatial tails of out-of-plane atomic orbitals into the van der Waals gap—which is distinct from covalent bond formation—and this electronic overlap drives the large bonding–antibonding splitting (). This pushes the upper hybridized state above the localized uncoupled states at , thereby establishing the definitive indirect band gap character of the homobilayer. Comparing the AB and BA configurations, the negligible variation observed in demonstrates that while the local atomic registry directly modulates the layer-dependent potential offset, the interlayer hopping matrix element remains fundamentally constrained by the global equilibrium interlayer separation rather than the relative lateral atomic displacement.
As detailed in the quasiparticle band alignment scheme (Suppl. Mat., Fig. S8), these registry-dependent potential offsets ultimately yield a type-II alignment that facilitates the formation of spatially separated interlayer excitons. These are denoted based on their specific interlayer character: (electron localized in the bottom layer), (electron localized in the top layer), and (an exciton state associated with the global lattice registry).
In contrast, the MoS/MoSe heterobilayer exhibits a robust direct type-II band alignment across all staggered registries. The potential profiles in Figure 3b show large atomic-scale oscillations; while such fluctuations are intrinsic to all van der Waals heterojunctions due to their periodic atomic structure, here they are clearly reflecting the chemical asymmetry between the layers. As depicted in the quasiparticle band structures (Figure 3e,Figure 3f), where L1 (MoS, blue) and L2 (MoSe, orange) are color-coded, the CBM is strictly localized on L1, while the VBM resides on L2. This chemical potential mismatch between S and Se effectively overrides the registry-dependent potential shifts, preserving the direct – gap and ensuring that the intrinsic band offsets indicated by the work functions () and vacuum levels () in Figure 3b remain the primary drivers of electronic localization.
Optical Properties and Excitonic Response
Excitonic effects were evaluated by solving the Bethe-Salpeter equation (BSE) atop the quasiparticle (QP) landscape (Figure 4). In these reduced-dimensional systems, the suppressed dielectric screening significantly enhances the electron–hole (e–h) interaction, as evidenced by high exciton binding energies, , and a substantial renormalization of the absorption onset. The dimensionless oscillator strength quantifies the spectral weight and transition probabilities, , defined as:

Figure 4. Many-body excitonic absorption spectra and stacking-dependent photoluminescence (PL). (a, c) Absorptivity of AB-stacked bilayer MoS and MoS/MoSe calculated at the EVGW (red dashed) and EVGW+BSE (black solid) levels, illustrating the significant excitonic enhancement and band shifts. (b, d) BSE absorptivity (black, broadened with gaussian of width 0.1 eV) with individual transitions and oscillator strengths (blue bars).
where is the electron mass, is the reduced Planck constant, and is the excitonic transition frequency. The inclusion of the energy-dependent prefactor ensures that the length units of the position matrix element are properly canceled, yielding the dimensionless values provided by ab initio codes like VASP.
To ensure that the calculated absorptivity spectra and intensities are free from artificial dependence on the out-of-plane periodic box size, the optical response of these 2D slabs is rigorously normalized. For a 2D system modeled in a periodic unit cell containing a vacuum region of height along the -direction, the standard 3D macroscopic dielectric function scales inversely with . To eliminate this volume dependence, the optical absorptivity (expressed in percent) is evaluated via the 2D sheet response, scaling as . Because all bilayer and trilayer configurations in this study were modeled using an identical out-of-plane lattice parameter and vacuum padding ( Å), the vacuum scaling factor remains uniform across all structures, ensuring that comparisons of peak intensities and oscillator strengths among different configurations are physically meaningful. In addition, the BSE matrix construction inherits the quasiparticle self-energy inputs evaluated with 64 empty conduction bands, ensuring numerical stability of the excitonic eigenvalues.
In the MoS homobilayer, the optical response is dominated by robust intralayer excitons with at 2.085 eV (AB) and 2.073 eV (BA), corresponding to exciton binding energies of eV and 0.22 eV, respectively (Table 1), as shown in Figure 4a,Figure 4b and Figure S9,Figure S9). Despite the stacking-induced band offsets, and , discussed in the previous section, the negligible energy difference of meV indicates that these intralayer states are largely decoupled from macroscopic interlayer polarization. This stability is consistent with the in-plane and orbital character of the K-point band edges (Figure 3c,Figure 3d), which enforces strong spatial e–h overlap within a single layer and prevents significant interlayer leakage (Figure 4e,Figure 4f).

Figure S9. Many-body excitonic absorption spectra and stacking-dependent photoluminescence (PL). (a, c) Absorptivity of BA-stacked bilayer MoS and MoS/MoSe calculated at the EVGW (red dashed) and EVGW+BSE (black solid) levels, illustrating the significant excitonic enhancement and band shifts. (b, d) BSE absorptivity (black, broadened) with individual transitions and oscillator strengths (blue bars).
Conversely, the MoS/MoSe heterobilayer exhibits a significantly suppressed oscillator strength, , for the lowest-energy transitions (Figure 4c,Figure 4d). This optical darkness is a direct spectroscopic signature of the intrinsic type-II alignment, where the chemical potential mismatch between S and Se strictly partitions the carriers—holes in MoSe and electrons in MoS—yielding a near-zero dipole matrix element alongside substantial exciton binding energies reaching eV (Table 1). The resulting interlayer excitons (X, X, and X; Suppl. Mat., Figure S8) show pronounced sensitivity to atomic registry, evidenced by a 107 meV stacking-dependent shift, nearly an order of magnitude larger than the variation observed in the homobilayer.

Figure S8. Quasiparticle band alignment and interlayer – excitons in AB- and BA-stacked bilayer MoS and MoS/MoSe. Layer-dependent VBM and CBM offsets ( meV, meV) result in type-II alignment and spatially separated interlayer excitons , , and .
At higher energies ( eV), hybrid excitonic states emerge with finite oscillator strengths. The observed reduction in hybrid intensity for the BA configuration (cf. Suppl. Mat., Figure S9,Figure S9)) relative to the AB configuration (cf. Figure 4d) correlates with suppressed interlayer orbital hybridization, , and the modified charge redistribution evident in the CDD profiles (Figure 2). These results confirm that lateral sliding in TMD heterostructures provides a powerful degree of freedom to modulate the extent of wavefunction delocalization across the van der Waals gap.
Trilayer 2L-MoS/MoSe Heterostructures
Stacking-Dependent Polarization in Trilayer 2L-MoS/MoSe Heterostructures
The evolution of spontaneous out-of-plane polarization was investigated in trilayer heterostructures consisting of a MoSe monolayer atop a MoS homobilayer (L1-MoS/L2-MoS/L3-MoSe). This mixed-chalcogenide architecture introduces additional interlayer degrees of freedom, enabling precise modulation of the electric dipole through the interplay of multiple van der Waals interfaces. High-symmetry stacking sequences (ABA, BAA, ABB, and BAB) are analyzed in Figure 5a–Figure 5d, while the asymmetric AAB and BBA configurations are discussed in the Supplementary Materials, Figure S11.

Figure 5. Atomic configurations of 2L-MoS/MoSe heterostructures for different stacking sequences. Panels (a–d) show the ABA, BAA, ABB, and BAB configurations, respectively, in both top and side views. The interlayer distances and denote the separations between L1–L2 and L2–L3, highlighting the structural variation induced by stacking order.

Figure S11. (a, b) Top and side views of the trilayer 2L-MoS/MoSe heterostructure with AAB stacking, and (c, d) those with BBA stacking, along with their corresponding charge density difference (CDD) isosurfaces. (e, f) Planar-averaged CDD profiles for the AAB (e) and BBA (f) configurations. Red and green indicate charge depletion and accumulation, respectively.
Stacking-Dependent Polarization in Trilayer 2L-MoS/MoSe Heterostructures
The in-plane lattice constant for all trilayer configurations converges within a narrow range of 3.202–3.208 Å, as summarized in Table S1. This indicates that observed variations in electronic and polarization properties are primarily driven by registry-dependent interlayer interactions rather than residual epitaxial strain. While the total out-of-plane supercell dimension varies between approximately 32.19 Å for the AAA arrangement and 36.29–36.61 Å for the staggered configurations (utilizing a consistent vacuum region as noted in the Computational Details), the total lattice constant of a slab model includes the vacuum padding and is therefore not an appropriate physical measure of interlayer contraction. Instead, structural compaction and interlayer proximity are rigorously evaluated via the actual interlayer distances, (between the two layers, L1 and L2) and (across the – heterointerface between L2 and L3).
Side views of the atomic arrangements (Figure 5a–Figure 5d) further illustrate that and show strong sensitivity to stacking sequence. For instance, contracts significantly to 6.41 Å in ABA and 6.39 Å in BAA relative to the AAA reference (6.70 Å), highlighting the alleviation of Pauli steric repulsion in staggered registries where lateral shifts allow adjacent layers to nest closer together. Conversely, which governs the heterofacial S–Se interface exhibits distinct values (e.g., 6.64 Å in ABA versus 6.96 Å in AAA), reflecting the larger covalent radius of Selenium. This local structural relaxation enhances interlayer orbital hybridization, facilitates charge redistribution across the heterojunction, and ultimately modulates the net out-of-plane polarization. Table S1 quantitatively captures these trends in , , and across all systems, providing a robust structural framework to rationalize the stacking-dependent electronic and polar properties.
The interlayer charge redistribution was quantified using the planar-averaged CDD
The 3D isosurfaces and corresponding 1D profiles (Figure 6a–Figure 6h) provide a clear spatial map of this redistribution. Analysis of the CDD profiles shows that the L2/L3 () interface maintains a robust, stacking-insensitive maximum-to-minimum sequence. This indicates that the interfacial dipole at the heterojunction is primarily dictated by the intrinsic chemical potential mismatch between S and Se, effectively pinning the polarization regardless of the underlying bilayer registry.

Figure 6. Charge redistribution in 2L-MoS/MoSe heterostructures for different stacking sequences. Panels (a–d) show the charge density difference () isosurfaces for ABA, BAA, ABB, and BAB configurations, respectively, where yellow and cyan regions represent electron accumulation and depletion (isovalue: 0.001 e/Å). Panels (e–h) present the corresponding planar-averaged profiles across the trilayer, revealing the interfacial dipole characteristics associated with each stacking sequence.
In contrast, the L1/L2 () homobilayer interface acts as a tunable degree of freedom. In the AAA stacking, the L1/L2 junction exhibits a mirror-symmetric charge distribution, yielding a vanishing net dipole (Suppl. Mat., Fig. S10). In ABA and ABB sequences, the L1/L2 interface adopts an AB-type registry, producing a maximum-to-minimum CDD sequence and a corresponding upward dipole vector (Figure 6e,Figure 6j). Conversely, BAA and BAB stackings invert this sequence to minimum-to-maximum (Figure 6f,Figure 6h), generating a downward dipole vector analogous to BA-type homobilayers.
The AAB and BBA configurations further illustrate the role of structural relaxation in dipole formation. In AAB, the initial AA-like registry of the L1/L2 interface is energetically unstable due to S–S chalcogen repulsion. Subsequent relaxation induces a lateral sliding of L1 toward a BA-like configuration, breaking out-of-plane symmetry and producing a downward dipole. In BBA, the L1/L2 interface occupies a near-equilibrium staggered position, resulting in a symmetric extrema sequence and negligible net polarization (Suppl. Mat., Fig. S11).
Collectively, these results demonstrate that the 2L-MoS/MoSe trilayer behaves as a multistable dipolar system. By delineating the local dipole orientations at each interface whether pointing from the MoS layers toward MoSe or in reverse Figure 6 illustrates how the internal electric field can be significantly modulated.
To address the distinction between geometrically constructed trial configurations and genuinely metastable polarization states, full structural relaxations were systematically carried out starting from various initial registries. While unfavorable trial configurations (such as the high-symmetry AAA arrangement at meV or poorly coordinated stacking sequences) spontaneously relaxed barrierlessly into lower-energy alternative registries, the identified ABA, BAA, ABB, and BAB sequences successfully converged into distinct, stable local minima on the potential energy surface. A quantitative evaluation of their total energies (summarized in Table S4 of the Suppl. Mat.) reveals a clear energetic hierarchy: BAB and ABB form nearly degenerate ground states ( and 0.01 meV, respectively), followed by the accessible metastable ABA ( meV) and BAA ( meV) configurations. The successful convergence into distinct energy minima, combined with their unique interlayer distances (, ) and distinct charge redistribution profiles, confirms that these stacking registries represent robust, physically realizable metastable states rather than arbitrary geometric constructs.
| Stacking Registry | (eV) | (meV) |
| BAB | 0.00 | |
| ABB | 0.01 | |
| ABA | 4.70 | |
| BAA | 75.44 | |
| AAA | 143.10 |
Table S4. Calculated total energies () and relative energies () for different stacking configurations of 2L-/ trilayers evaluated relative to the BAB configuration.
Quasiparticle Band Structure Modulation by Stacking in Trilayer 2L-MoS/MoSe
The electronic properties of 2L-MoS/MoSe trilayer heterostructures exhibit a pronounced and systematic sensitivity to atomic registry, as summarized in Table 2, Figure 7, and Figure S12–Figure S14. At the PBE+SOC level, the fundamental band gaps span 0.68–0.83 eV, corresponding to a ∼22% modulation driven exclusively by stacking-dependent interlayer coupling. This persistence across electronic structure hierarchies identifies atomic registry as an effective internal boundary condition governing the electronic landscape of these quasi-2D heterostructures.

Figure 7. Quasiparticle electronic structures and band alignments of trilayer 2L-MoS/MoSe heterostructures. (a)–(d) EVGW quasiparticle band structures and corresponding type-II band alignment schematics for ABA, ABB, BAA, and BAB stacking sequences. The arrows at the point highlight the stacking-dependent localization of the conduction-band minimum (CBM) and valence-band maximum (VBM) across the L1, L2, and L3 layers.
| Stacking | PBE | PBE+SOC | HSE06 | EVGW () | PL / (eV) | (eV) | Gap Type |
| AAA | 0.82 | 0.74 | 1.25 | 1.55 | 1.362 | 0.188 | Direct |
| ABA | 0.82 | 0.73 | 1.21 | 1.51 | 1.287 | 0.223 | Direct |
| ABB | 0.86 | 0.83 | 1.31 | 1.60 | 1.391 | 0.209 | Indirect (PBE)/Direct (SOC, HSE06) |
| BAA | 0.77 | 0.68 | 1.19 | 1.51 | 1.356 | 0.154 | Direct |
| BAB | 0.82 | 0.78 | 1.28 | 1.58 | 1.395 | 0.185 | Indirect (PBE)/Direct (SOC, HSE06) |
| AAB | 0.83 | 0.79 | 1.28 | 1.59 | 1.398 | 0.192 | Indirect (PBE)/Direct (SOC, HSE06) |
| BBA | 0.82 | 0.73 | 1.22 | 1.52 | 1.278 | 0.242 | Direct |
Table 2. Calculated band gaps (, in eV) using PBE, PBE+SOC, HSE06, and EVGW (), optical/PL transition energies (), exciton binding energies (), along with gap type and photoluminescence (PL) energies for different stacking configurations of 2L-MoS/MoSe trilayer heterostructures. Values of demonstrate systematic stacking-dependent modulation across registries, reflecting subtle registry-induced variations in interlayer distances and local dielectric screening.

Figure S12. Electronic band structures of trilayer heterostructures (AAB, ABB, BAB) at the PBE level.

Figure S14. Electronic band structures calculated using HSE06, showing enhanced band gaps and preserved stacking trends.
A systematic method comparison reveals consistent changes in the band gap. Moving from PBE to HSE06 increases the band gap by ∼60–70%, while inclusion of many-body effects at the EVGW@PBE level introduces a further ∼20–25% enhancement relative to HSE06 due to quasiparticle renormalization. The resulting quasiparticle gaps (1.51–1.60 eV) exceed PBE+SOC values by ∼110–120%, reflecting the combined effects of exact exchange and reduced dielectric screening in the layered limit.
The stacking-dependent electronic structure is further illustrated in Figure 7(a–Figure 7d), which presents EVGW band dispersions and type-II alignment schematics for ABA, ABB, BAA, and BAB configurations. Complementary planar-averaged electrostatic potentials (Figure 8(a–Figure 8d)) reveal that atomic registry imposes a configuration-dependent internal electrostatic field, manifested through systematic shifts in the vacuum level relative to the Fermi level . These shifts directly quantify work-function variations across stackings (5.019 eV for ABA, 5.014 eV for ABB, 4.862 eV for BAA, and 4.975 eV for BAB), establishing electrostatics as a registry-controlled degree of freedom.

Figure 8. Planar-averaged electrostatic potential profiles and work function analysis for trilayer 2L-MoS/MoSe heterostructures. (a)–(d) Potential energy landscapes along the -axis for ABA, ABB, BAA, and BAB stackings. Red dashed lines denote the vacuum level (), blue dash-dotted lines indicate the Fermi level (), and green bidirectional arrows represent the stacking-dependent work function (). The inset information boxes provide the corresponding energetic references obtained after applying a macroscopic averaging filter.
This registry induced electrostatic landscape originates from stacking dependent structural relaxation at the MoS/MoS interface, which continuously rebalances Pauli repulsion and van der Waals attraction. The resulting charge redistribution generates asymmetric interfacial dipoles that reshape , producing weak but finite internal fields relative to . Importantly, these fields do not primarily shift absolute band edges; instead, they act as a symmetry-breaking perturbation that competes with interlayer orbital hybridization within the MoS bilayer, while the MoS/MoSe interface remains electronically more rigid and serves as a dipolar reference layer.
This competition manifests most strongly in the conduction sector. Layer- and momentum resolved projections at the point show that the VBM remains strongly localized on the MoSe layer across all registries, effectively pinning the valence reference and stabilizing robust type-II alignment. In contrast, the CBM resides in a near-degenerate manifold distributed across the MoS bilayer, making it highly susceptible to dipole-induced symmetry breaking and interlayer hybridization.
Quantitatively, this is reflected in a systematic reduction of interlayer conduction-band splitting from 62.7–65.8 meV (ABA/ABB) to 53.4–48.5 meV (BAA/BAB), indicating progressive weakening of internal asymmetry within the MoS bilayer. Crucially, the absolute CBM energy remains nearly invariant across stackings, confirming that registry does not act as a rigid band-edge shift but instead redistributes orbital character within a fixed energetic envelope.
This orbital redistribution is governed by the same electrostatic landscape extracted from , where variations in encode the effective work-function modulation. The resulting internal potential bias subtly tunes the balance between delocalizing interlayer hybridization and dipole-induced layer polarization, selectively reshaping the conduction manifold without perturbing the pinned valence reference.
Spin–orbit coupling further refines the band-edge structure at the PBE level, lifting valence-band degeneracies at the point and, in selected registries, enhancing direct-gap character. Across all computational levels, the quasiparticle gaps are consistent with experimental photoluminescence trends, with the ABB stacking exhibiting the largest gap and correspondingly strongest optical response.
Overall, stacking registry in 2L-MoS/MoSe trilayers emerges as an internal electrostatic boundary condition that encodes atomic configuration into a self-consistent dipolar landscape. This landscape governs a constrained electronic degree of freedom: a pinned valence manifold coexisting with a dipole-sensitive conduction manifold. The resulting registry electrostatics orbital hierarchy establishes atomic-scale stacking as a deterministic control knob for internal fields, band-edge reconstruction, and optical response in layered heterostructures.
Optical Properties and Excitonic Response
The optical response of the 2L-MoS/MoSe trilayer was systematically evaluated to elucidate the impact of stacking-dependent polarization on the many-body excitonic landscape. The specific electronic structure is reflected in the many-body excitonic absorption spectra (Figure 9a–Figure 9d), calculated at the EVGW+BSE level using a Gaussian broadening of 0.1 eV. The black solid lines represent the absorption spectra, while the blue bars denote the oscillator strengths (); notably, the absorptivity is the result of the convolution of these oscillator strengths with the broadening function. For all configurations, the lowest-energy peaks exhibit strongly suppressed oscillator strengths, a hallmark of spatially separated interlayer excitons. The variation in intensity across different stackings reflects differences in interlayer coupling; for example, ABB and BAB configurations exhibit higher oscillator strengths than BAA, indicating enhanced wavefunction overlap across the vdW gap.

Figure 9. Excitonic absorption of trilayer -/ heterostructures. (e)–(h) Many-body excitonic absorption spectra (black solid lines, Gaussian broadening with width 0.1 eV) and oscillator strengths (blue bars) calculated at the EVGW+BSE level. The variations in the low-energy excitonic peaks reflect the modulation of interlayer coupling and spatial charge separation induced by the different stacking registries.
To explicitly address the influence of stacking modes on excitonic properties, exciton binding energies () were quantitatively determined across all configurations as the difference between the quasiparticle band gap and the optical transition energy (). As summarized in Table 2, exhibits a clear stacking-dependent modulation, ranging from 0.154 eV (BAA) to 0.242 eV (BBA). This variation is governed by registry-induced changes in interlayer distance and local orbital overlap, which subtly modulate the electron–hole Coulomb interaction while maintaining consistent macroscopic dielectric screening across the van der Waals stack.
Table 2 summarizes the PL energies for the lowest-energy transitions, which span a wide spectral range from 1.278 to 1.398 eV. These variations originate from the interplay between the intrinsic type-II band alignment and stacking-induced electrostatic potential modulations at the MoS bilayer interfaces (Suppl. Mat., Fig. S8). In these heterostructures, the lowest-energy excitations are dominated by interlayer excitons (ILXs), characterized by spatial separation of charge carriers, with holes localized in the MoSe valence band and electrons residing in the MoS bilayer. Consequently, the transition energies are highly sensitive to internal electric fields arising from stacking-dependent dipole moments. This can already be seen from the EVGW+BSE absorption spectra (Figure 9a–Figure 9d) where AB-type stackings (ABA and ABB) exhibit stronger excitonic peaks compared to BA-type configurations (BAA and BAB), reflecting enhanced electron–hole wavefunction overlap. To quantitatively assess the effect of polarization reversal, the average ILX energies for the two stacking families were evaluated. This approximately accounts for variable stacking configurations between and due to lattice mismatch and domain disorder in actual experiments. For AB-type configurations, the characteristic energy is defined as eV, whereas for BA-type configurations it is given by eV. This results in an interlayer excitonic energy shift meV.
The microscopic origin of this shift is schematically illustrated in Figure 10. In the AB-type stacking, the internal polarization establishes a reference electrostatic potential landscape for ILX formation. Upon switching to the BA-type configuration, the reversal of the out-of-plane dipole () modifies the electrostatic potential profile across the layers. A closer examination reveals that this shift arises from a cooperative interplay between modest quasiparticle gap variations and the systematic modulations established across registries, rather than a purely rigid band-edge shift.

Figure 10. Schematic representation of the energy level shifts for ILXs in AB- and BA-type stackings. Energy levels are based on calculations. The diagram explicitly displays the stacking-dependent modulation of the exciton binding energy (), showing a larger for AB-type ( eV) than for BA-type ( eV) configurations. The meV shift represents the calculated registry-dependent offset; experimental PL energies, which represent ensemble averages, are found to be consistent with this framework [29].
The predicted shift of 36 meV is in excellent agreement with the experimentally reported value of approximately 40 meV [29], confirming that stacking-governed internal dipoles and their subsequent impact on the excitonic landscape are the dominant mechanisms for excitonic energy tuning. The remaining discrepancy () is likely attributable to additional dielectric screening effects, such as hBN encapsulation in experimental samples, which are not explicitly included in the present calculations.
Finally, the full range of spectral tunability reaches 108 meV between the ABA and BAB configurations, demonstrating that the trilayer architecture significantly expands the accessible energy landscape compared to conventional bilayers. By switching the stacking registry between AB- and BA-type configurations, discrete and reversible shifts in ILX emission energy can be achieved, effectively stabilizing excitons at distinct energy levels.
To bridge the gap between our first-principles findings and practical device architectures, the demonstrated multi-state energetic hierarchy and stacking-dependent polarization provide a direct physical foundation for sliding ferroelectric and optoelectronic memory concepts. Specifically, the existence of multiple stable local minima with distinct relative energies ( ranging from nearly degenerate ground states to accessible metastable states up to meV, Table S4) enables multi-state non-volatile memory, where external electrical fields or mechanical shear can deterministically switch between discrete polarization states (, ). Furthermore, the robust type-II band alignment coupled with the stacking-dependent interlayer exciton energy shift ( meV) and distinct optical oscillator strengths allows for non-destructive optical read-out of the ferroelectric polarization state. This synergistic interplay between switchable dipole fields and tunable excitonic emission underpins the practical realization of multi-state optoelectronic memory cells and reconfigurable van der Waals heterostructure devices.
Conclusion
In summary, this work provides a comprehensive investigation of the stacking-dependent electronic and excitonic landscapes in -based vdW heterostructures. By employing eigenvalue-self-consistent calculations combined with the Bethe-Salpeter equation (), we elucidate the fundamental mechanisms by which atomic registry governs interfacial polarization and quasiparticle band alignments.
Our analysis of the MoS homobilayer demonstrates that lateral sliding between AB and BA registries enables a reversible inversion of the out-of-plane dipole moment. This switchability, driven by registry-induced symmetry breaking and asymmetric orbital hybridization, establishes the microscopic foundation for sliding ferroelectricity. In contrast, the MoS/MoSe heterobilayer exhibits a robust, non-switchable polarization pinned by the intrinsic chemical potential mismatch between sulfur and selenium, which effectively overrides the geometric registry and stabilizes a type-II band alignment.
In trilayer 2L-MoS/MoSe configurations, an expanded parameter space for deterministic control of charge carriers emerges. The stacking sequence dictates whether photogenerated electrons are localized in the central or bottom MoS layer, while holes remain confined to the MoSe monolayer. This layer-selective behavior can be described within an electrostatic framework, where internal electric fields induce band-edge shifts of 60–70 meV, directly linking dipole orientation to electronic localization.
The many-body excitonic spectra further corroborate these findings. Interlayer excitons exhibit strong sensitivity to atomic registry, with stacking-dependent energy shifts reaching up to 108 meV. Notably, the calculated average interlayer excitonic shift of 36 meV shows excellent agreement with the experimentally reported meV ferroelectric shift in 3R-MoS/MoSe heterostructures. This quantitative agreement confirms that stacking-governed internal dipoles are the primary drivers of excitonic tunability.
Overall, these results demonstrate that the interplay between structural sliding and many-body interactions in TMD trilayers provides a powerful degree of freedom for engineering programmable light–matter interactions. The ability to modulate excitonic energies and carrier transfer pathways through precise control of stacking sequence offers a promising route toward multi-state optoelectronic memory and next-generation sliding ferroelectric devices.
Acknowledgments
This work was funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) - SFB 1477 ”Light-Matter Interactions at Interfaces”, project number 441234705”.
Supporting Material: Stacking-Directed Polarization and Excitonic Engineering in MoS/MoSe van der Waals Heterostructures
Mohammed El Amine Miloudi; Oliver Kühn

Figure S1. Convergence of total free energy difference () per unit cell as a function of the Monkhorst-Pack -mesh (), evaluated relative to the reference mesh.

Figure S3. (a, b) Planar-averaged electrostatic potential profiles of the BA-stacked bilayer MoS (a) and the MoS/MoSe heterobilayer (b). The work function is obtained from the difference between vacuum and Fermi levels.

Figure S4. Projected electronic band structures of AA-, AB-, and BA-stacked bilayer MoS calculated with PBE, PBE+SOC, and HSE06.

Figure S5. Spin- and orbital-resolved PBE band structures of bilayer MoS₂ for AB and BA stacking, showing layer-resolved contributions and type-II band alignment.

Figure S6. Projected electronic band structures of AA-, AB-, and BA-stacked MoS/MoSe heterobilayers.

Figure S7. Spin- and orbital-resolved PBE band structures of the MoS/MoSe heterobilayer for AB and BA stacking. Panels (a,b) correspond to the AB-stacked heterobilayer for layer 1 (L1 = MoS) and layer 2 (L2 = MoSe), showing contributions from Mo , , , and orbitals and the layer-resolved localization of the valence-band maximum (VBM) and conduction-band minimum (CBM). Panels (c,d) show the BA-stacked heterobilayer, illustrating the stacking-induced redistribution of band-edge states while preserving the direct-gap character at . These results confirm type-II band alignment, with the VBM predominantly localized on MoSe and the CBM on MoS. Consistent layer-dependent band alignment is obtained using HSE06 and EVGW calculations.

Figure S10. Top and side views of the trilayer 2L-MoS/MoSe heterostructure with AAA stacking (a,b), together with the corresponding charge density difference (CDD) isosurfaces.

Figure S13. Electronic band structures including spin–orbit coupling for all stacking configurations.
References
- [1]Zhao K, He D, Fu S, Bai Z, Miao Q, Huang M, Wang Y and Zhang X 2022 Nanomaterials 12 3418
- [2]Xue F, He J H and Zhang X 2021 Applied Physics Reviews 8 021316
- [3]Vizner Stern M, Waschitz Y, Cao W, Nevo I, Watanabe K, Taniguchi T, Sela E, Urbach M, Hod O and Ben Shalom M 2021 Science 372 1462–1466
- [4]Chen X, Xuan X, Guo W and Zhang Z 2025 npj 2D Materials and Applications 9 10
- [5]Niu Y, Li L, Qi Z, Aung H H, Han X, Tenne R, Yao Y, Zak A A and Guo Y 2023 Nature Communications 14 5578
- [6]Li S, Wei K, Liu Q, Tang Y and Jiang T 2024 Frontiers of Physics 19 42501
- [7]Tran K, Choi J and Singh A 2021 2D Materials 8 022002
- [8]Amin B, Singh N and Schwingenschlögl U 2015 Physical Review B 92 075439
- [9]Berkelbach T C, Hybertsen M S and Reichman D R 2013 Physical Review B 88 045318
- [10]Mak K F, Lee C, Hone J, Shan J and Heinz T F 2010 Physical Review Letters 105 136805
- [11]Wang F, Wang J, Guo S, Zhang J, Hu Z and Chu J 2017 Scientific Reports 7 44712
- [12]Xiao D, Liu G B, Feng W, Xu X and Yao W 2012 Physical Review Letters 108 196802
- [13]Wang G, Chernikov A, Glazov M M, Heinz T F, Marie X, Amand T and Urbaszek B 2018 Reviews of Modern Physics 90 021001
- [14]Hong X, Kim J, Shi S F, Zhang Y, Jin C, Sun Y, Tongay S, Wu J, Zhang Y and Wang F 2014 Nature Nanotechnology 9 682–686
- [15]Ceballos F, Bellus M Z, Chiu H Y and Zhao H 2014 ACS Nano 8 12717–12724
- [16]Tran K, Moody G, Wu F, Lu X, Choi J and Li X 2019 Nature 567 71–75
- [17]Zhang N, Surrente A, Baranowski M, Maude D K, Gant P C, Castellanos-Gomez A and Plochocka P 2018 Nano Letters 18 7651–7657
- [18]Choi J, Florian M, Steinhoff A, Erben D, Tran K, Kim D S and Li X 2021 Physical Review Letters 126 047401
- [19]Jiang Y, Chen S, Zheng W and Pan A 2021 Light: Science & Applications 10 72
- [20]Li S, Wei K, Liu Q, Tang Y and Jiang T 2024 Frontiers of Physics 19 42501
- [21]Jung E, Park J C, Seo Y S, Kim J H, Hwang J and Lee Y H 2022 Scientific Reports 12 4543
- [22]Lin Q, Fang H, Kalaboukhov A, Liu Y, Zhang Y, Fischer M, Li J, Hagel J, Brem S, Malic E, Stenger N, Sun Z, Wubs M and Xiao S 2024 Nature Communications 15 8762
- [23]Zhang Q, Fan A, Wang Y, Wu F, Li L, Meng H and Geng D 2025 npj 2D Materials and Applications 9 76
- [24]Cao W, Hod O and Urbach M 2022 ACS Applied Materials & Interfaces 14 57492–57499
- [25]Pan J, Wang H, Zou L, Wang X, Zhang L, Dong X, Xie H B, Ding Y, Zhang Y, Watanabe K, Taniguchi T, Wang S and Wang Z 2025 Physical Review Letters 135 126603
- [26]Yang D, Liang J, Wu J and Ye Z 2024 Nature Communications 15 1389
- [27]Jiang Y, Chen S, Zheng W, Zheng B and Pan A 2021 Light: Science & Applications 10 72
- [28]Hou S, Xie X, Li S, Chen J, Zhang X and Liu Y 2025 ACS Nano 19 34954–34965
- [29]Schwandt-Krause J, Miloudi M E A, Blundo E, Deb S, Heidkamp J N, Watanabe K, Taniguchi T and Korn T 2026 Nano Letters 26 214–221
- [30]Blöchl P E 1994 Physical Review B 50 17953–17979
- [31]Kresse G and Joubert D 1999 Physical Review B 59 1758–1775
- [32]Kresse G and Furthmüller J 1996 Physical Review B 54 11169–11186
- [33]Kresse G and Furthmüller J 1996 Computational Materials Science 6 15–50
- [34]Perdew J P, Burke K and Ernzerhof M 1996 Physical Review Letters 77 3865–3868
- [35]Hybertsen M S and Louie S G 1986 Physical Review B 34 5390–5413
- [36]Wilhelm J, Seewald P and Golze D 2021 Journal of Chemical Theory and Computation 17 1662–1677
- [37]Hanke W and Sham L J 1980 Physical Review B 21 4656–4673
- [38]Strinati G 1988 Rivista del Nuovo Cimento 11 1–86