Synergistic Enhancement of HZO Ferroelectric Properties via La2O3 Interface Engineering for Neuromorphic Computing

Abstract While Hf0.5Zr0.5O2 (HZO)-based ferroelectric memristors show great potential for neuromorphic computing, their practical application is currently hindered by low tunneling electroresistance (TER) ratios and limited endurance. In this study, a high-performance artificial synapse utilizing a Pt/HZO/La2O3/n+-Si heterostructure is demonstrated. The incorporation of a La2O3 interlayer effectively stabilizes the ferroelectric orthorhombic phase. This interface engineering yields a robust two remanent polarization (2Pr) of 39.5 μC/cm2, extends the cycle endurance to 4.3 × 109, and achieves an exceptional TER ratio of 1868 (a 120-fold enhancement). Consequently, the device successfully emulates key synaptic functions, including long-term potentiation and depression (LTP/LTD), spike-timing-dependent plasticity (STDP), and Pavlovian conditioning. Neuromorphic computing simulations using a ResNet-18 architecture achieve 94.4% image classification accuracy on the CIFAR-10 dataset after 100 training epochs, and demonstrate exceptional robustness by maintaining 75% accuracy under 40% Gaussian noise. Ultimately, this work presents a viable pathway toward the development of highly reliable and dense neuromorphic hardware.