Multiscale Modeling of Metal-Oxide-Metal Conductive Bridging Random-Access Memory Cells: From Ab Initio to Finite-Element Calculations

Computational Study on Filament Growth Dynamics in Microstructure-Controlled Storage Media of Resistive Switching Memories

“…The nanoscale ECM system, an active metal electrode (e.g., Ag)/dielectric (e.g., SiO 2 with volatile microstructures)/inert metal electrode (e.g., Pt), is employed to discuss its possible filament growth mechanism. In the traditional rate-limiting picture, relative rates of active metal migration/oxidation/reduction/nucleation, which can be affected by an applied voltage, a humid environment, etc., have been regarded as the most important factors to determine the filament growth mode. ,,,,,, These factors can be tailored by changing the involved materials. One more crucial factor, void concentration, corresponding to changeable microstructures of dielectric layers, is located here to induce the growth mode transition in ECM systems without involving materials and growth conditions (applied electric field or temperature) changed by using the KMC simulations.…”

“…Fortunately, our KMC method still has the potential to incorporate the environmental humidity, just as the previous FEM-based simulations did, taking humidity-related ionic species as descriptors of input parameters in modeling. 31 Furthermore, the correct description of charge in ions/atoms is also a challenge for methods like KMC, FEM, etc. For example, the chemical composition of the active electrode interface can experience changes involving chemical reactions with the electrolyte material and then affect active metal cation generation at this interface, as the filaments grow.…”

“…At present, various computational methods are available, e.g., density functional theory (DFT), first-principles molecular dynamics (FPMD), , reactive MD, , kinetic Monte Carlo (KMC), ,,,,, phase field method (PF), , finite element method (FEM), ,,, and multiscale simulation method to investigate CF growth in ECM systems or supply I – V information on ECM systems. Among them, the first-principles , or reactive , MD simulations have their advantages in providing microscopic information on ECM systems on the atomic scale, such as evolutionary active metal coordination ,,, and charge distribution of nanoclusters or electrode surfaces during cluster formation or filament growth.…”

“…As Moore’s law goes up to its limit, the nanoscale resistive switching (RS) devices are expected to be a powerful candidate for nonvolatile memories and artificial synapses in artificial intelligence. − Among them, one type of RS memories relying on the electrochemical metallization (ECM) process, called conductive-bridge random access memory (CBRAM), has attracted great experimental − and computational − research interests in characteristics of filament growth dynamics, filament morphology, and even filament composition. One of the most important processes in ECM cells is the initial electroforming, which would determine the subsequent repeated connection (on-state) and disconnection (off-state) of conducting filaments (CFs) during operation.…”

“…Their simulation results are difficult to compare with experimental results on the large scale. The PF method , or FEM ,,, ignores detailed structural information inside ECM devices but can supply comparable I – V information to experimental observation. At this stage, the coarse-grained KMC simulations can overcome the temporal and spatial limitation compared to atomistic-scale MD simulations and try their best to approach the real ECM systems.…”