Memristor-based LSTM network with in situ training and its applications

Xiaoyang Liu, Zhigang Zeng, Donald C. Wunsch II

Abstract: Highlights•The presented memristor-based long short-term memory (MbLSTM) has an effective inference process derived from the circuit structure.•A weight update scheme is put forward to achieve in situ training and the updating of the conductance of memristors is row-parallel.•Activation functions are approximately implemented through intentionally designing circuit parameters.