High Performance of Ferroelectric Synaptic Device in Hf_0.7Zr_0.3O₂ (HZO)/La_0.7Sr_0.3MnO₃ (LSMO)/SrTiO₃ (STO) Heterostructure by Optimizing Defect-Mediated Ferroelectric Switch

High-performance artificial synaptic devices with linear synaptic weight and high precision are in high demand for hardware neural network (HNN) implementation. Hf-based ferroelectric memristor has robust stable non-volatile analog resistive switching driven by its gradual polarization reversal. However, the high linearity and precision of the synaptic weight update are required to improve the accuracy of the neuromorphic chip application. Here, we demonstrate the high performance of ferroelectric synaptic device in Hf_0.7Zr_0.3O₂(HZO)/La_0.7Sr_0.3MnO₃(LSMO)/SrTiO₃ (STO) heterostructure, showing the high linearity (α=0.5), effective multiple conductance states (&gt; 40), and low cycle-to-cycle variation (2.06%). The synaptic characteristics of the device are dominantly attributed to the inhomogeneous domain nucleation, ultimately affecting the precision and linear weight update achievable in continuous polarization switching. As an artificial synapse, experimental results with electrical pulse modulation closely mimic the neuro-inspired signal process dependent on spike timing and spike rate. These results supposed that oxygen vacancy engineering in ferroelectric synaptic devices is a powerful approach to improving effective precision and overcoming the limitations nonlinear synaptic behavior for the implementation of HNN. In addition, when applied to a hardware neural network with a crossbar array (20X20), the recognition accuracy of MNIST handwriting data was recorded as 94.8%.<br/><br/><b>Acknowledgments</b> This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government(MSIT) (NRF; Grant Mo. NRF-2021R1A2C2010781).