Investigation of Multi-Level Properties of TaOx-based Memristive Devices and Optimized Programming Scheme for On-Line Training

In this paper, we present the multi-level cell characteristics of TaO _x based memristive devices as synaptic elements for neuromorphic computing. We utilized various programming conditions to modulate the conductance states with different set/reset voltage amplitude or width. Our results reveal that for LTP and LTD process, the dominant factors for write error are nonlinearity and fluctuation respectively. Furthermore, we proposed an error-aware programming scheme, thereby enabling a high accuracy for weight updating when used in ANN applications.