Analytic Model for Statistical State Instability and Retention Behaviors of Filamentary Analog RRAM Array and Its Applications in Design of Neural Network

For the first time, an analytic model is presented for the statistical state instability and retention behaviors of filamentary analog resistive random access memory (RRAM) array. In the model, the diffusion of oxygen vacancy (1), the Brownian-like hopping of V
_O
during the diffusion process and the recombination of V
_O
are considered. The statistical state instability and retention behaviors of different states under various temperatures are accurately described by the model, which is verified by the measured data of 1Kb filamentary analog RRAM (FA-RRAM) array. Furthermore, the analytic model is successfully implemented to evaluate and optimize the reliability of FA-RRAM based multi-layer neural network. Guided by the model, optimized synapse structures and refresh operation are proposed to significantly enhance the reliability of FA-RRAM based neural network.