Origin and suppressing methodology of intrinsic variations in metal-oxide RRAM based synaptic devices

Metal-oxide RRAM device is promising for synaptic application. The intrinsic variations during synaptic training process affect the accuracy of neuromorphic computation. In this work, we investigate the influence of the resistance variations on the metal-oxide RRAM-based neuromorphic computation system. Atomic KMC simulation is performed to study the random migration of oxygen vacancy, which is responsible for the origin of the variation. A novel methodology is proposed to improve the robustness of the metal-oxide RRAM-based neuromorphic computation system.