A follow-up on the sulphur atom popping model for MoS$_2$ memristor

TL;DR

This work defends the sulphur atom popping mechanism as the intrinsic driver of resistive switching in MoS2 memristors, using DFT-based analyses to complement prior ReaxFF MD and to critique claims about universal ML interatomic potentials. It emphasizes the necessity of dynamic charge treatment (e.g., QEq) in simulations and shows that without extra electrons, ground-state DFT cannot reproduce the field-induced popped state; field-induced charge localization and electron trapping emerge as key stabilizers. Through targeted ab-initio calculations and careful examination of NEB results, the authors provide a coherent, field-driven mechanism that reconciles ReaxFF observations with quantum-mechanical insights, while acknowledging the limitations of static DFT and MLIPs for such reactive, defect-rich systems. The work underscores the practical significance of incorporating charge dynamics and field effects to realistically model non-volatile switching in 2D memristors and cautions against over-relying on bulk-trained, charge-agnostic ML interatomic potentials for complex 2D/defective systems.

Abstract

The mechanism of resistive switching in two-dimensional (2D) semiconductor-based memristors is intriguing, and our conventional knowledge of bulk-oxide based memristors does not apply to these devices. Experimental data indicate that the genesis of resistive switching may be intrinsic to the 2D semiconducting active layer, as well as resulting from the movement of electrode atoms. Employing reactive-force field (ReaxFF) molecular dynamics simulations, we introduced the "sulphur atom popping model" [npj 2D Mater. Appl. 5, 33 (2021)] to elucidate the intrinsic nature of non-volatile resistive switching in 2D molybdenum disulfide-based memristors. In this paper we provide additional perspective to this model using density functional theory. We also discuss the limitations of universal machine learning interatomic potentials in reproducing ReaxFF simulation results.

Figures (2)

• Figure 1: a Relaxation of an 8$\times$8 supercell of monolayer MoS$_2$, where a popped state is created by manually moving the S atom opposite to the vacancy to the Mo plane. After relaxation, the S atom remains within the Mo layer. b An 8$\times$8 supercell of monolayer MoS$_2$ containing an S vacancy (1) the vacancy structure is first relaxed (2), then a popped state is created by manually positioning an S atom in the Mo plane (3), the popped state is then relaxed without adding extra electrons, (4) and relaxed again with two extra electrons added, (5) without extra electrons, the popped state returns to the original S plane, whereas with extra electrons, the popped atom remains in the Mo plane. c Isosurface plot of the charge density difference between popped state stabilized MoS$_2$ layer with two added electrons and popped state stabilized MoS$_2$ layer without added electrons. The isosurface level is set as 0.0007 eVÅ$^{-3}$.
• Figure 2: a Plot of the total energy values for the vacancy and popped structures under a static electric field (E) applied along the negative z-direction. The left panel (blue) represents the energy values of the parent vacancy states, while the right panel (red) shows the energy values of the popped states as the electric field increases. b CINEB calculation for S atom popping without an electric field (blue) and with an electric field (red). c Isosurface plot of the charge density difference between E=0.5 eV/Å and E=0 eV/Å. The isosurface level is set as 0.0004 eVÅ$^{-3}$.