Memory Cell Modelling Cuts Reliance on Guesswork with Advanced Simulations

Researchers are increasingly focused on understanding the complex resistance switching behaviour exhibited by metal/oxide/metal structures central to conductive bridging random access memory (CBRAM) cells. Jan Aeschlimann, Fabian Durch, and Christoph Weilenmann, all from the Integrated Systems Laboratory at ETH Z urich, alongside Alexandros Emboras, Mathieu Luisier, and Juerg Leuthold et al., present a novel multiscale simulation framework to accurately compute current-voltage characteristics and illuminate the underlying physics. This work is significant because it bridges ab initio calculations with finite element modelling, substantially reducing the need for empirical fitting parameters and yielding a more predictive modelling environment. Applying this framework to an Ag/a-SiO2/Pt CBRAM device, the team successfully reproduces experimental data and assesses the critical role of Joule heating, particularly in devices featuring nanoscale filaments and current concentrations in the microampere range, paving the way for the rational design and optimisation of future memory technologies.

This innovative approach computes current versus voltage (I-V) characteristics of metal/oxide/metal structures, providing insights into their resistance switching properties.

The research addresses a critical need for advanced modelling tools in the field of non-volatile memory technologies, particularly as devices are scaled down to the atomic level. The framework integrates techniques ranging from molecular dynamics and nudged elastic band theory to electronic and thermal transport calculations.
Input parameters for a finite element model are derived from both ab initio calculations and machine-learned empirical data, significantly reducing the reliance on fitting parameters and enhancing overall accuracy. This methodology allows for more reliable predictions than traditional modelling environments, offering a substantial improvement in simulating CBRAM cell performance.

Applying this framework to an Ag/a-SiO2/Pt CBRAM device, researchers successfully reproduced experimental data with high fidelity. Detailed analysis also assessed the relevance of Joule heating within the cell, revealing that self-heating becomes prominent in devices featuring thin conductive filaments with diameters of only a few nanometres and at current concentrations in the 10s-microampere range.

These findings are crucial for understanding and optimising the thermal behaviour of nanoscale CBRAM structures. This computational methodology now enables exploration of the potential of novel, not-yet-fabricated memory cells and facilitates the reliable optimisation of their design. By bridging the gap between atomic-scale phenomena and device-level performance, the study paves the way for advancements in energy-efficient, non-volatile storage solutions applicable to edge computing, neuromorphic systems, and in-memory computing architectures. The ability to accurately simulate CBRAM behaviour at multiple scales represents a significant step towards realising the full potential of this promising memory technology.

Computational modelling of Ag/a-SiO2/Pt CBRAM resistance switching behaviour

A multiscale simulation framework was developed to compute the current versus voltage (I-V) characteristics of metal/oxide/metal structures forming the core of conductive bridging random access memory (CBRAM) cells and to elucidate their resistance switching properties. The research began with a finite element model, with input parameters derived from both ab initio calculations and machine-learned empirical calculations to minimise the need for extensive fitting.

Molecular dynamics and nudged elastic band methods were employed to characterise atomic-level behaviour, while electronic and thermal transport calculations were integrated to model charge and heat flow within the CBRAM structure. This computational framework was then applied to an Ag/a-SiO2/Pt CBRAM device, successfully reproducing experimental I-V data with high fidelity.

The relevance of Joule heating was assessed by simulating devices with varying cell geometries, revealing that self-heating becomes significant in devices containing thin conductive filaments with diameters in the few-nanometer range and at current concentrations in the 10s-microampere range. The study utilised a scanning electron microscope to visualise a fabricated Ag/a-SiO2/Pt CBRAM cell on a silicon-on-insulator wafer, with 20nm thick amorphous SiO2 acting as the insulating layer.

A DC voltage was applied to the Ag contact while the Pt contact was grounded, simulating the formation of a nanoscale Ag filament bridging the two electrodes through the SiO2 network. The I-V characteristics were then measured, demonstrating a hysteretic behaviour with a clear SET and RESET phase between the high-resistance state (HRS) and low-resistance state (LRS), exhibiting a high-to-low resistance ratio exceeding 105. This methodology enables exploration of novel memory cell designs and reliable optimisation of their performance characteristics.

Silver ion diffusion parameters in amorphous silicon dioxide determined via multiscale modelling

Diffusion coefficients for Ag+ cations in amorphous silicon dioxide were determined using multiscale simulations and analysis of ion trajectories. Arrhenius plots, constructed from smoothed ion trajectories, revealed a pre-exponential factor, D0, of 2.962x 10−5 cm2/s and an activation energy, ∆ED, of 0.196 ±0.013 eV for Ag+ diffusion.

These values were validated through nudged elastic band calculations, yielding an average diffusion barrier of 0.234 ±0.122 eV, demonstrating strong agreement with the ab initio molecular dynamics results. The study employed a computational framework combining molecular dynamics, nudged elastic band theory, and electronic transport calculations to investigate the resistance switching properties of Ag/a-SiO2/Pt CBRAM cells.

Analysis of Ag+ ion diffusion occurred within a 1.5nm cube of a-SiO2 at 1200 K, with ion positions recorded every 0.1ps for 39ps to map the diffusion trajectory. Square mean free paths were calculated from these trajectories, and the diffusion coefficient was determined for each sample and temperature.

To model electrical current, atomistic filamentary CBRAM structures were generated with varying oxide lengths, ranging from 3000 atoms in size. These structures were then used to calculate the transmission function using the OMEN quantum transport solver with an energy resolution of 1 meV. The ballistic electrical current was obtained by applying the Landauer-Büttiker formula, considering a voltage of 1mV and utilising the Fermi-Dirac distribution function to account for electron behaviour in the contacts. Joule heating was found to be relevant in devices with thin conductive filaments exhibiting diameters in the few-nanometer range and current concentrations in the 10s-microampere range.

Joule heating and filament geometry influence resistive switching in CBRAM devices

A multiscale simulation framework has been developed to model the current-voltage characteristics of metal/oxide/metal structures central to conductive bridging random access memory (CBRAM) cells and to elucidate their resistance switching behaviour. This approach integrates a finite element model with input parameters derived from both first-principles calculations and machine learning techniques, encompassing molecular dynamics, nudged elastic band methods, and electronic and thermal transport analysis.

By combining these techniques, the number of required fitting parameters is substantially reduced, leading to a more accurate modelling environment compared to conventional methods. The framework was applied to an Ag/amorphous-silicon dioxide/platinum CBRAM cell, successfully reproducing experimental data and assessing the impact of Joule heating through variations in cell geometry.

Results indicate that self-heating is significant in devices featuring thin conductive filaments, specifically those with diameters of a few nanometres and current concentrations in the tens of microamperes range. This methodology facilitates the exploration and optimisation of the design of novel, yet-to-be-fabricated memory cells.

The authors acknowledge that the model’s accuracy relies on the precision of the input parameters obtained from the ab initio and machine-learned calculations. Furthermore, the simulations assume axial symmetry, which may not fully capture the complexities of filament formation in all scenarios. Future research could focus on extending the model to three-dimensional geometries and incorporating more detailed representations of the electrochemical processes occurring at the metal/oxide interfaces, potentially improving the predictive power and broadening the applicability of the framework.