A universal re-annealing method for enhancing endurance in hafnia ferroelectric memories: Insights from stochastic noise analysis

Abstract

Ferroelectric memories based on hafnium oxide (HfO₂) are promising for next-generation non-volatile memory due to their compatibility with complementary metal-oxide semiconductor (CMOS) processes and scalability to nanometer-thin films. However, cycling endurance remains a critical challenge, largely limited by defect generation and ferroelectric fatigue. In this work, we demonstrate a universal re-annealing process that significantly enhances the endurance of ferroelectric HfO2 memory devices. By employing stochastic noise analysis, specifically low-frequency noise (LFN) spectroscopy, as a diagnostic tool, we uncover the microscopic mechanisms by which thermal re-annealing mitigates degradation. This stochastic diagnostic approach serves as a crucial technique for process optimization, turning the inherent randomness of defect generation into actionable insights. The re-annealing treatment, optimized at 600 °C, effectively repairs ferroelectric thin films, reducing trap densities and improving ferroelectric phase stability without inducing the adverse effects encountered at higher annealing temperatures. This optimization was guided by noise measurements that sensitively detect trap-related fluctuations, revealing how an overly aggressive anneal at 800 °C introduces new defects, eventually degrading device performance. The effectiveness of this approach is validated across standalone ferroelectric films and integrated devices (ferroelectric tunnel junctions), highlighting its broad applicability.