A Novel Flexible Artificial Synapse Based on Pseudocapacitor for High-Accuracy Neuromorphic Computing

Memristors have been suggested as competitive candidates to implement artificial synapses, which can be integrated into the neuromorphic computing architectures. However, several non-ideal factors including non-linearity and asymmetry of weight update during long-term potentiation (LTP) and long-term depression (LTD) processes restrict the practical application of the memristor based artificial synapses. In this work, a novel flexible artificial synapse based on polyaniline (PANI) pseudocapacitor is proposed. The device exhibits typical pseudocapacitive characteristics during cyclic voltammetry (CV) tests. In addition, essential synaptic plasticity such as short-term plasticity, long-term plasticity and experience-dependent learning are successfully emulated based on the electrochemical reactions of hydrogen ions. More importantly, this artificial synapse shows low non-linearity (0.36/-1.23 for LTP/LTD) and asymmetry (0.19) of weight update. By incorporating the LTP/LTD characteristics of our artificial synapse, the three-layer artificial neural network (ANN) achieves a high recognition accuracy of 97% on the MNIST data set, demonstrating its great potential in neuromorphic computing systems.