Investigation of weight updating modes on oxide-based resistive switching memory synapse towards neuromorphic computing applications

Memristors, resistive random access memory (RRAM) devices when used in memory applications, have attracted significant interest recently as a promising candidate for neuromorphic computing systems due to their excellent size scalability, fast switching speed and low energy consumption [1]. In order to obtain high learning accuracy in neural networks based on back propagation learning rule, the updating behavior of synaptic linear symmetric weights is important [2]. During weight update, the resistance of the memristor device can be adjusted incrementally by controlling the distribution of oxygen vacancies, which modulate the overall conductance of the device.