Herein, we present simulations of conductive filament formation in resistive random-access memory using a finite element solver. We consider the switching material, which is typically an oxide, as a two-phase material comprising low- and high-resistance phases. The low-resistance phase corresponds to a defective and conducting region with a high anion vacancy concentration, whereas the high-resistance phase corresponds to a non-defective and insulating region with a low anion-vacancy concentration. We adopt a phase variable corresponding to 0 and 1 in the insulating and conducting phases, respectively, and we change the phase variable suitably when new defects are introduced during voltage ramp-up for forming. Initially, some defects are embedded in the switching material. When the applied voltage is ramped up, the phase variable changes from 0 to 1 at locations wherein the electric field exceeds a critical value, which corresponds to the introduction of new defects via vacancy generation. The applied voltage at which the defects percolate to form a filament is considered as the forming voltage. Here, we study the forming-voltage uniformity using simulations, and we find that for typical planar-electrode devices, the forming voltage varies significantly owing to the stochastic location of the initial defects at which the electric field is “crowded.” On the other hand, a protruding electrode can improve the switching uniformity drastically via facilitating the deterministic location of electric-field crowding, which also supported by the reported experimental results.
We investigated the intrinsic variation of retention in phase-change memory using a phase-field method that is capable of depicting stochastic nucleation followed by the growth of the nucleated crystallites. We found that the median of the retention time increases upon scaling unlike the Monte Carlo results previously reported. More importantly, the industrial requirement, which is ten years at 70 °C, can be guaranteed for 1-ppb tail bits down to the 10-nm technology, although statistical spread increases.