Boosting the Memory Window of Memristive Stacks via Engineered Interfaces with High Ionic Mobility

TL;DR

This paper demonstrates that engineering interfaces with high ionic mobility in memristive stacks significantly expands the memory window, improves switching characteristics, and enhances endurance, with practical neural network applications and transferability to other materials.

Contribution

Introducing high ionic mobility materials at interfaces to substantially increase the number of stable resistance states and improve device performance in memristive stacks.

Findings

Memory window increased from 8 to 22 states

SET/RESET voltage reduced by 50%

Achieved below 7% classification error on MNIST

Abstract

The great potential of memristive devices for real-world applications still relies on overcoming key technical challenges, including the need for a larger number of stable resistance states, faster switching speeds, lower SET/RESET voltages, improved endurance, and reduced variability. One material optimization strategy that has still been quite overlooked is interface engineering, specifically, tailoring the electrode/dielectric interface to modulate oxygen exchange. Here, we demonstrate that introducing materials with high ionic mobility can significantly expand the accessible oxygen concentration range within the dielectric layer, significantly broadening the memory window. Using SrTiO3-based memristive stacks, we integrated an ion-conducting SrCoO3 interfacial layer to facilitate oxygen transfer, increasing the number of distinguishable resistance states from 8 to 22. This modification also reduced the SET/RESET voltage by 50% and markedly improved device endurance, albeit with a trade-off of reduced state retention. To assess the practical implications of this trade-off, we trained a two-layer fully connected neural network using the experimental SrTiO3/SrCoO3 memristor characteristics on the MNIST handwritten digit dataset. Networks with hidden-layer sizes between 64 and 256 memristive elements achieved classification errors below 7%. The observed temporal drift means the functional state must be updated at intervals of less than 1 h to maintain reliable operation. Finally, we confirmed the transferability of this interface-engineering approach by applying it to HfOx-based devices, achieving a similarly enhanced memory window.