From volatile to nonvolatile resistive switching in a ferroelectric second-order memristor

Abstract

Ferroelectric memristors are promising candidates for neuromorphic computing, yet their practical application is hindered by the trade-off between the volatile synaptic plasticity required for learning and the nonvolatile retention needed for long-term data storage. Here, we demonstrate a ferroelectric memristor based on an a-Si:H/Hf0.5Zr0.5O2/TiN MOS structure that resolves this challenge by tuning the ferroelectric layer thickness. By increasing the HZO thickness to 8 nm, the built-in electric field arising from charged interface traps is effectively suppressed, both extending state retention and improving endurance compared to 5 nm devices. The developed memristor exhibits dual-mode operation controlled by write pulse duration: with short pulses, it emulates biological synaptic plasticity including paired-pulse facilitation and depression, while longer pulses enable stable nonvolatile data storage. Furthermore, the device supports approximately 10 distinct and time-stable multibit conductance states via partial polarization switching. This functionality arises from the memristor's true second-order nature, where ferroelectric polarization provides nonvolatile multibit memory, and the temporal dynamics of electron injection and emission into interface traps govern synaptic behavior. The combination of long-term retention, multibit storage, and biologically inspired plasticity within a single cell positions this device as a promising building block for compact and efficient neuromorphic computing systems.