A detailed study of the impact of the inert electrode on the performance of ReRAM memory cells has been performed. Starting with the baseline device Cu/TaO_x/Pt/Ti, seven more devices have been manufactured with different inert electrode constructions while the Cu electrode and the TaO_x dielectric layer have been processed at the same time and are considered identical for all devices. The Pt/Ti inert electrode has been replaced by Pt/Cr, Rh/Cr, Rh/Ti, Rh/Al₂O₃, Ir/Ti, Ir/Cr, and Ru/Ti. Although the second layer Ti, Cr or Al₂O₃ is not an inherent part of the device proper (needed as a glue layer for integration reasons), it has a tangible impact on the device performance and its endurance. While the inert electrode materials (Pt, Rh, Ir, Ru) have similar inertness properties in terms of immiscibility with copper, high ionization energy, and high work function, their thermal conduction coefficients differ widely. In units of W/(m·K), their bulk thermal conductivities are: Rh – 150, Pt – 72, Ir – 147, Ru – 117, and Cu – 385, for comparison. Thermal conductivities of the glue layers (Ti – 20, Cr – 94, and Al₂O₃ – ~12) have impact too on the cell endurance and finer aspects of the device performance. Thus, those memory cells are sensitive to their integration on the wafer. We find that, when subjecting a memory cell to repeated write and erase cycles the memory cell with the highest thermal conductivity, i.e. Cu/TaO_x/Rh/Cr performs best, allowing virtually unlimited number of switching cycles, and the worst performance is obtained for the base line device, Cu/TaO_x/Pt/Ti, characterized by the lowest thermal conductivity. This is in contrast with the finding that the initial set/reset characteristics for all the devices are very similar. The different degradation over time is related to different heat dissipation modes and can be attributed to the thermal conductivity variations among the selected composite inert electrodes. Our voltage sweep experiments for the set operation are characterized by voltage ramp rate rr, varied between 0.1V/s and 10V/s, and by compliance current I_cc, varied between 5mA and 5mA. All the devices are confirmed to follow R_on=const/Iⁿ_cc relationship. It is well-known that the reset operation is mainly effectuated by Joules heating and the maximum local temperature of the conductive Cu filament (CCF) is known to be around 700 ^oC. It is also known that in most cases the geometrical shape of the CCF is that of truncated cone with the base on the surface of the inert electrode and with a narrow apex touching the active Cu electrode. Our electric characterization experiments indicate that the rupturing temperature is a function of the thermal conductivity of the inert electrode suggesting that for an inert electrode with high thermal conductivity the temperature of the CCF for the same amount of Joules heating is lower than for an inert electrode with a low thermal conductivity.

The cross-comparison of the data affords a finer insight into the formation mechanisms of the CCF formation as well. We find that the final shape of the CCF is determined in a very short time interval by two fluxes of copper - one constructive and the other one destructive in terms of the CCF’s robustness: 1) When the tip of the filament makes barely contact with Cu electrode the resistance of the tip is high and causes a large voltage drop. The large voltage drop creates a large electric field that can support a high Cu⁺ ion flux into the tip; hence the tip widens, the resistance drops, and the Cu⁺ flux comes eventually to a halt. This Cu⁺ ion flux is large when I_cc and the ramp rate are both low. 2) Once the current begins to flow through the CCF, it deposits Joules heat in the CCF. The diffusion flux may occur along the inert electrode surface and/or into the inert metal. The Joules heating triggers Cu atom diffusion flux weakening the base of CCF. The Cu diffusion flux increases with increasing I_cc and decreasing rr. In addition, all other parameters being otherwise equal, the Cu-diffusion flux depends on the thermal conductivity of the inert electrode too. The competition between the two fluxes engenders differences in the observed data which can be now consistently explained. Based on those results we propose guidelines for optimum inert electrode engineering to achieve high endurance at high frequency of set/reset switching cycles. Our analysis indicates that Rh/Cr, and possibly Rh/Cu, are the best choice for the inert electrode irrespective of the choice of the dielectric and active electrode.