Towards an universal artificial synapse using MXene-PZT based ferroelectric memristor

Abstract

To address the challenge of memory wall, memristor is a breakthrough for the hardware realization of computation in memory (CIM). As a promising candidate for the resistive-switching layer of memristor, ferroelectric material has recently received extensive attention. However, the performance of ferroelectric memristors is limited by rigid device structure based on metal/ferroelectric material interface. In this work, the hybrid ferroelectric Cu/MXene/PZT memristor has been firstly demonstrated. Two-dimensional (2D) material Ti3C2 MXene was synthesized and inserted into traditional PZT (PbZr0.52Ti0.48O3) ferroelectric memristors (Cu/PZT/Pt) for performance enhancement. By comparison, the ferroelectric devices based on Cu/Ti3C2/PZT/Pt exhibit enhanced performance, i. e., lower switching voltage, lower power consumption, reproducing RS behaviors, and higher switching ratio (106%). The effect of the insertion of the MXene layer has been investigated by theoretical analysis about switching mechanisms of the devices and first-principles calculations of the Ti3C2/PZT atomic structure. Additionally, functions of analogy biological synapse, i. e., long-term potentiation (LTP), long-term depression (LTD), spike-timing-dependent plasticity (STDP), and paired-pulse facilitation (PPF) have been mimicked using these MXene-PZT based devices. Based on synaptic behaviors in MXene-PZT based memristors, the learning accuracy of pattern recognition with handwritten data can reach 95.13%. Our results are expected to inspire the development of MXene for performance enhancement of ferroelectric memristors and their applications in neuromorphic computing systems.