The Moore’s law has slowed down significantly during the last few years, as transistor scaling becomes increasingly difficult when we hit the scale of few nanometers (i.e., few billionths of a meter). In this range of distances, surface atoms are a considerable fraction of the overall material within the device, thus surface effects might start to play a leading role.
This is what we found in our recent study ‘Surface diffusion-limited lifetime of silver and copper nanofilaments in resistive switching devices’, addressing the spontaneous retraction of metallic filaments within a certain type of memory devices, called resistive switching memory (RRAM), or simply memristors. RRAM with Ag or Cu electrodes typically show a filament disconnection after few ms, or even μs, after electrical formation of the filament. While this instability might be harmful for nonvolatile memory devices that must retain their data bit for years, it might come at hand in neuromorphic applications, where it can form the basis of short-term plasticity responsible for many of our cognitive functions, such as the recognition of speech and visual patterns.
In this work, we first conduct an atomistic study of atomic diffusion within nanosized filaments in RRAM showing that the surface plays a key role in controlling the memory lifetime. An atom can move along the filament surface, or the interface between the metallic filament and the host material, as it was a surfer on the sea waves, or a skier on a snowy slope. Compared to bulk atoms, surface atoms have a larger ability to move, as they are more loosely bonded to the surrounding atoms. Their diffusion aims at minimizing the surface-to-volume ratio, as the surface is energetically expensive compared to the bulk. The surface energy minimization thus serves as the driving force for the atomic diffusion until the filament breaks in two or more pieces, which causes the final memory loss.
In addition to the atomistic study, we present a user-friendly numerical model that can be applied to predict the device lifetime depending on the constituting material and initial filament size. The simulations can account for a wide range of experimental data both from our lab and from the literature. The interpretation and new model can be used to better engineer memory devices and artificial synapses for future circuits that can mimic the cognitive behavior in the human brain.