Exploration of threshold and resistive-switching behaviors in MXene/ BaFe12O19 ferroelectric memristors

Ferroelectric memristors have great potentials to be the key computational element of the brain-inspired neuromorphic system due to their excellent endurance, multi-bit storage ability, ultrafast speed, and ultra-low energy consumption. In this work, novel barium ferrite (BFO: BaFe12O19)-based ferroelectric memristors have been fabricated. Meanwhile, two-dimensional MXene Ti3C2 was inserted onto the dielectric layer for performance enhancement. More importantly, under the regulation of compliance currents, the Cu/Ti3C2/BFO/Pt ferroelectric memristors can convert from stable threshold-switching (TS) to resistive-switching (RS) behavior. The coexistence of the TS and RS demonstrated in this work facilitates the simulation of biological synapses, i.e., short/longterm plasticity, paired-pulse facilitation, and spike-time-dependent plasticity. In addition, the one-bit adder circuit by fitting the non-volatile I-V curve (RS behavior) was demonstrated in the Cadence Virtuoso environment. The deep neural network based on our ferroelectric crossbar arrays has been architected for the image classification (on-chip: 85 %) of the CIFAR-10 dataset. The internal mechanisms are attributed to the combination of barrier variation induced by the ferroelectric polarization and motion of oxygen vacancies, which have been verified by the first-principles calculation of density functional theory. These results may provide possible suggestions for ferroelectric memristors-based neuro-inspired computing systems.

Automated summary

Novel barium ferrite-based ferroelectric memristors with MXene Ti3C2 enhancement have been fabricated. These memristors can convert from stable threshold-switching to resistive-switching behavior under compliance currents. A one-bit adder circuit was demonstrated and a deep neural network was designed for image classification using these ferroelectric crossbar arrays.