Nano Devices

Nanoscale resistive switching elements

Digital applications require devices that combine long retention times with fast write speeds. Emerging ReRAM elements based on functional oxides can in practice achieve a long state lifetime (>1010) and ultra-fast switching, since relatively small biases can increase the switching speed up to six orders of magnitude due to their intrinsic highly non-linear rate of switching. And although such elements can effortlessly be implemented as bi-stable switches, their operation is not limited by discrete states (ROFF and RON). A continuous spectrum of conductive states is in principle viable by collectively controlling the number of current percolation channels existing per unit cell. One of the most celebrated memristor attributes is the capability to maintain a state without requiring external biasing. This feature in combination with the attainable miniscule physical dimensions of the device (minimum reported: 10x10nm) is predominant for implementing high-density low-power memory technologies that could expedite a much-needed extension to Moore’s law.

When it comes however to employing such devices, engineers struggle with their excessively large variability that aggravates with their scaling. Most scientific reports claiming nanoscale reproducible cells allude on thin-film active cores. Yet, their lateral dimensions are several orders of magnitudes larger and thus their results are not representative of nanoscale implementations. On the contrary, cells established via e-beam lithography and/or nano-imprint lithography, demonstrate a remarkably large state-variability when compared against CMOS standards. This becomes even more evident when the devices are employed as multi-state analogue switches. On top of that, situations exist were competing switching mechanisms are triggered simultaneously, facilitating both bipolar and unipolar switching. This has so far been reported for SrTiOx-based devices and we have recently witnessed a rather similar response for TiO2-based elements that further aggravates the programming fidelity.

Extra care is required in modelling the kinetics of ions that often precipitate devices with such unconventional characteristics. A plurality of models has already been proposed for each distinct switching type scenario. While there is no single encompassing model that can explain all observations, we have recently demonstrated a versatile model that accounts for non-linear dopant kinetics and can be easily optimised towards matching experimental results. This model will be initially employed to empirically describe measured responses and will be accordingly optimised to match the dynamics of distinct prototypes. And as it is now evident that their functioning is subject to statistical laws, appropriate probabilistic components will be considered in modelling.