Ab initio Modeling of MoS2/Oxide Device Interfaces with Machine Learned Electronic Structures

Manasa Kaniselvan; Mauro Dossena; Denghui Lu; Alexander Maeder; Nicolas Vetsch; Alexandros Nikolaos Ziogas; Mathieu Luisier

cond-mat.mtrl-sciAug 27, 2026

Ab initio Modeling of MoS2/Oxide Device Interfaces with Machine Learned Electronic Structures

Authors: Manasa Kaniselvan, Mauro Dossena, Denghui Lu, Alexander Maeder, Nicolas Vetsch, Alexandros Nikolaos Ziogas, Mathieu Luisier

Organizations: Integrated Systems Laboratory, ETH Zurich, Zurich, Switzerland

Abstract

We introduce a new ab initio approach to simulate semiconductor devices that integrates scalable machine-learned (ML) electronic structure models with an advanced quantum transport (QT) solver. The developed framework enables 10,000X speedups over density functional theory to produce the Hamiltonian matrix of devices made of >20,000 atoms, while offering high prediction accuracy. We use its unique features to investigate MoS2/oxide samples and single-layer MoS2 field-effect transistors, where the surrounding oxide layers, here, HfO2 or Al2O3, are explicitly included into the QT domain. In particular, we reveal that the presence of undercoordinated metal atoms (Hf or Al) close to the semiconductor-oxide interface significantly affects the magnitude of the electronic current and its propagation through MoS2.

Figures & tables

Fig. 1 : Overview of the ‘atomic structure ( $\mathbf{xyz}$ ) $\rightarrow$ learned $\mathbf{H}$ $\rightarrow$ quantum transport’ workflow. The relevant transport regions, a monolayer MoS 2 placed between oxide layers, here Al 2 O 3 , are first assembled, atom by atom, and placed within a single-gate transistor architecture with a gate length $L_{g}$ , source and drain extensions $L_{s}$ and $L_{d}$ , oxide thickness $t_{ox}$ , and substrate height $t_{sub}$ . The amorphous oxide layers are created via a standard melt-and-quench procedure using molecular dynamics with machine-learned interatomic potentials as inputs, before being stacked with MoS 2 and undergoing a geometry relaxation. The resulting set of atomic coordinates is passed to the MALOQ package, which builds node- and edge-wise embeddings through SO(2)-equivariant graph convolutions. These embeddings are converted into diagonal/off-diagonal blocks [ $\mathbf{H_{ii}}$ ]/[ $\mathbf{H_{ij}}$ ], which are used to reconstruct the Hamiltonian matrix $\mathbf{H_{ML}}$ of the device of interest. Only $\mathbf{H_{ML}}$ is learned, as $\mathbf{S}$ can be determined analytically from the orbital basis. The $\mathbf{H_{ML}}$ and $\mathbf{S}$ matrices serve as inputs to QuaTrEx , a quantum transport solver implementing the non-equilibrium Green’s function equations. The output quantities are the $I-V$ characteristics and the bond-resolved electronic current.

Fig. 2 : DFT-level dataset generation for MoS 2 /oxide electronic structures. (a) Atomic samples selected from the generated training dataset, which consists of a total of 50 $\sim$ 2,000-atom 2D/oxide structures. The amorphous oxides were all created with an MD melt-quench process: melting for 20 ps at 4000 K, and quenching for 400 ps in a constant ramp down to 300 K. The oxide is then cleaved, sandwiched with the 2D monolayer, annealed for 20 ps at 1000 K with DFT D3 corrections, and finally relaxed. (b) Heatmap of the Hamiltonian matrix $\mathbf{H}$ for one of these MoS 2 /HfO 2 structures, as computed with CP2K using a DZVP basis for MoS 2 and a SZV basis for the oxides, resulting in a total size of 16,608 $\times$ 16,608 and 138 million nonzero electronic orbital interactions. Each of these matrices undergoes a decomposition into [ $\mathbf{H_{ii}}$ ]/[ $\mathbf{H_{ij}}$ ] sub-matrices, followed by an analytical basis transformation into a space of uncoupled spherical tensors. MALOQ is trained to minimize the $L2+\sqrt{L2}$ loss over these tensor elements up to an interatomic distance of 10 Å. (c) 1 million randomly sub-sampled nonzero matrix elements of $\mathbf{H}$ ( $|H_{ij}|$ ) from the 138 million in (b), as a function of the interatomic distance ( $r_{ij}$ ) for one MoS 2 /HfO 2 sample. Each scatter point is colored according to the atomic element interaction it originates from in the Hamiltonian. The regular vertical patterns in the scatter points indicate Mo-S interactions, while the scattered background contains the orbital interactions within the oxide and at the interfaces with MoS 2 . We note that a minimum cell dimension of 2 nm (corresponding to a distance of 10 Å ) between periodic images is necessary to fully saturate the decay of $\mathbf{H_{ij}}$ interaction terms and avoid artefacts stemming from the periodic boundary conditions of DFT. (d) Same as (c), but for MoS 2 /Al 2 O 3 .

Fig. 3 : MALOQ training validation. (a) Parity plots between the DFT and machine-learned matrix elements of $\mathbf{H}$ , showing perfect alignment for both MoS 2 /HfO 2 (top) and MoS 2 /Al 2 O 3 structures, with an average matrix element error of 39 $\mu$ eV. (b) Comparison between all eigenvalues (16,608) of $\mathbf{H_{DFT}}$ (blue circles) and $\mathbf{H_{ML}}$ (red crosses) for the MoS 2 /HfO 2 system made of 2,088 atoms. The inset zooms into the edge of the MoS 2 conduction band. (c) Energy difference (error) between all eigenvalues of $\mathbf{H_{DFT}}$ and $\mathbf{H_{ML}}$ in (b) . On average, the error is 38 meV, but it is largely concentrated at high energies, outside the range of interest. (d) Conduction bandstructure of the MoS 2 /HfO 2 sample from (a)-(c), as obtained from CP2K’s (solid blue lines) and MALOQ ’s (red crosses) Hamiltonian matrix. (e) Transmission function through the same MoS 2 /HfO 2 structure as in (d) computed with QuaTrEx and using $\mathbf{H_{DFT}}$ (blue line) and $\mathbf{H_{ML}}$ (red line) as inputs. Note that pure MoS 2 extensions are added on each side of the MoS 2 /HfO 2 central part and act as electron injection contacts.

Fig. 4 : Computational complexity and generalization of MALOQ . (a) Computational cost in node hours of CP2K (blue circles) and MALOQ (red crosses) to produce the Hamiltonian matrix of systems made of a different number of atoms, from 2,088 to 16,704, on GH200 GPUs. The DFT calculations converge on average within 100 self-consistent field (SCF) iterations, while the ML approach is single-shot. The dashed lines serve as visual guides and show a measured time complexity of $\mathcal{O}(N^{2.82})$ for CP2K and $\mathcal{O}(N^{1.01})$ for MALOQ . The speedups brought by MALOQ with respect to DFT are indicated as double-arrows, with a reference to the figures using the annotated sizes. (b) 12.8 $\times$ 3.2 $\times$ 2.8 nm 3 MoS 2 /HfO 2 and MoS 2 /Al 2 O 3 structures consisting of the repetition of four identical 3.2 $\times$ 3.2 $\times$ 2.8 nm 3 cells, containing $\sim$ 8,000 atoms. (c) Transmission function through the structures in (b), near their conduction band edges. QuaTrEx was used to compute all curves, with inputs from CP2K (blue line) and MALOQ (red line). Importantly, CP2K only returned the Hamiltonian matrix of the small cell, which was then tiled, while MALOQ received the whole system at once, without knowledge of its internal periodicity.

Fig. 5 : Current flow analysis in MoS 2 /oxide devices. (a) Top view ( $x$ - $z$ plane) of an MoS 2 /HfO 2 atomic structure inserted into the $n$ -type FET architecture of Fig. 1 with $L_{g}$ =15 nm, $L_{s}$ = $L_{d}$ =11.5 nm (total length $L_{ch}$ =38 nm), and $t_{sub}$ =20 nm. The source and drain are doped with a donor concentration $N_{D}$ =5e13 cm -2 . The equivalent oxide thickness (EOT) is set to 0.8 nm ( $t_{ox}$ =4 nm for HfO 2 with $\epsilon_{R}$ =20 or $t_{ox}$ =2 nm for Al 2 O 3 with $\epsilon_{R}$ =10). (b) Transfer characteristics of MoS 2 /HfO 2 and MoS 2 /Al 2 O 3 FETs with the specifications in (a) at $V_{ds}$ =0.6 V. The solid lines refer to QT simulations where only the MoS 2 channel is part of the transport domain and the dielectric layers are treated as ideal insulators. The total number of atoms is 2,160, with the size of $\mathbf{H_{DFT}}$ equal to 37,440. The symbols (error bars) at $V_{gs}$ =0.6 V are the averages (standard deviations) of three QT calculations per device type, explicitly including the oxides in the NEGF equations ( $>$ 20,000 atoms, size of $\mathbf{H_{ML}}$ $>$ 180,000). They were computed at $V_{gs}$ =0.6 V using the electrostatic potential profiles shown as inset. (c) Top view ( $x$ - $y$ plane) of the electrostatic potential landscape experienced by the MoS 2 atomic plane for the structure in (a) under flat band conditions. The formation of hills (red) and valleys (blue) can be observed. (d) Top view of current isosurfaces at a high (top) and low (bottom) value. In the upper plot, current hot spots are clearly visible that correlate with the presence of potential valleys in (c). In the lower one, regions with no current align well with hill’s locations.

Fig. 6 : Scattering at oxide interfaces (a) Current flux passing through each atom of the three MoS 2 /HfO 2 (red dots) and three MoS 2 /Al 2 O 3 (blue dots) structures used in Fig. 5 (b), with $z$ =0 corresponding to the average location of the Mo plane along $x$ . The dashed box indicates the flux through the Mo and S atoms. Lateral variation in the current flux through MoS 2 along $z$ is from macroscopic bending of each monolayer. The dashed ovals highlight large current fluxes through interface oxide atoms. (b) Same as in (a), but for the flux through the Hf and Al atoms in close contact with MoS 2 , sorted as a function of the metal atom’s oxygen coordination number. (c) Isosurface of the current flowing through the MoS 2 /HfO 2 structure in Fig. 5 (c), at the region marked with a star, with focus on the trajectory deformation induced by a surface undercoordinated Hf atom. (d) Same as in (c), but for an isosurface taken at a lower current value.

Computational Materials Science

Computational materials science uses computation and data-driven models to predict, explain, and design materials and their properties.

See all 144 papers

Physics-Informed Neural Network Surrogate for Oxygen Vacancy Dynamics in epitaxial $\mathrm{SrTiO_3}$ on Si memristors via Dynamic Spectral Optimization
Sep 2, 2026Rodion Podorozhny, Nikoleta Theodoropoulou, Jelena TešićPDE Surrogate Modeling Materials Science
Dept. of Computer Science, Texas State University, San Marcos, TX, USA · Dept. of Physics, Texas State University, San Marcos, TX, USA
MALOQ: Massively Accelerated Learning of Operators for Quantum Transport
Jun 27, 2026Manasa Kaniselvan, Alexander Maeder, Denghui Lu +2Quantum Chemistry High-Performance Computing
D-ITET, ETH Zurich, Zurich, Switzerland
Agentic multi-fidelity learning of quasiparticle and excitonic properties
Jun 5, 2026Arnab Neogi, Aaron Forde, Christopher A. Lane +2AI Agents for Scientific Discovery Materials Property Prediction
Theoretical Division, Los Alamos National Laboratory, Los Alamos, New Mexico 87545, USA · Center for Integrated Nanotechnologies, Materials Physics and Applications Division, Los Alamos National Laboratory, Los Alamos, New Mexico 87545, USA

Sep 2, 2026J/K move · Enter open · S save

Physics-Informed Neural Network Surrogate for Oxygen Vacancy Dynamics in epitaxial $\mathrm{SrTiO_3}$ on Si memristors via Dynamic Spectral Optimization

Rodion Podorozhny, Nikoleta Theodoropoulou, Jelena Tešić

Dept. of Computer Science, Texas State University, San Marcos, TX, USA · Dept. of Physics, Texas State University, San Marcos, TX, USA

PDE Surrogate Modeling Materials Science Physics-Informed ML

Open paper AskPDF arXiv

Physics-informed neural networks (PINNs) offer a promising framework for modeling semiconductor devices, yet standard architectures struggle with severe numerical stiffness and multiscale spatial discrepancies inherent to oxide heterostructures. Here, we demonstrate a cascaded PINN architecture coupled with a custom second-order Chebyshev second kind polynomial spectral optimizer (DSO V2 Hybrid) to model ion-electronic drift-diffusion transport in Pt/SrTiO

_3

/Si memristive heterostructures across a 20 nm STO film on a 380

μ

m Si substrate. By isolating potential, carrier density, and vacancy transport into four sequentially trained sub-neural-networks, our model circumvents condition numbers exceeding

10^{16}

without operator splitting. The trained surrogate reproduces experimental conductive-AFM current-voltage hysteresis (

R^2 > 0.96

) while ensuring strict Poisson consistency across continuous space. Compared to conventional finite-element solvers (e.g., COMSOL), the PINN surrogate enables differentiable inverse parameter estimation and linear time inference.

Abstract

Computational Materials Science

Explore similar work

Physics-Informed Neural Network Surrogate for Oxygen Vacancy Dynamics in epitaxial SrTiO3\mathrm{SrTiO_3}SrTiO3​ on Si memristors via Dynamic Spectral Optimization

MALOQ: Massively Accelerated Learning of Operators for Quantum Transport

Agentic multi-fidelity learning of quasiparticle and excitonic properties

Physics-Informed Neural Network Surrogate for Oxygen Vacancy Dynamics in epitaxial $\mathrm{SrTiO_3}$ on Si memristors via Dynamic Spectral Optimization