Conductive bridge random access memory devices such as Cu/SiO2/W are promising candidates for applications in neuromorphic computing due to their fast, low-voltage switching, multiple-conductance states, scalability, low off-current, and full compatibility with advanced Si CMOS technologies. The conductance states, which can be quantized, originate from the formation of a Cu filament in the SiO2electrolyte due to cation-migration-based electrochemical processes. A major challenge related to the filamentary nature is the strong variability of the voltage required to switch the device to its conducting state. Here, based on a statistical analysis of more than hundred fifty Cu/SiO2/W devices, we point to the key role of the activation energy distribution for copper ion diffusion in the amorphous SiO2. The cycle-to-cycle variability is modeled well when considering the theoretical energy landscape for Cu diffusion paths to grow the filament. Perspectives of this work point to developing strategies to narrow the distribution of activation energies in amorphous SiO2.
Keywords: CBRAM; SiO2; analytical model; electrochemical metallization cell (ECM); neuromorphic computing; quantum conductance; stochasticity.
Creative Commons Attribution license.