Berlin 2024 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
MM: Fachverband Metall- und Materialphysik
MM 17: Poster Ib
MM 17.6: Poster
Montag, 18. März 2024, 18:30–20:30, Poster F
Machine-learning based optimization of hafnium-zirconium oxide compositions for stable ferroelectric applications in non-volatile memories — •Arunima Singh and Patrick Rinke — Department of Applied Physics, Aalto University
CMOS-compatible hafnium dioxide (HfO2) based ferroelectrics (FE) are widely studied for use in non-volatile memories. Zr-doping of HfO2 i.e. hafnium-zirconium oxide (HZO) has further improved the FE properties. However, the optimum HZO structure and stoichiometry (Hf1-xZrxO2) still needs to be found. In this work, we used machine-learning assisted density-functional theory (DFT) calculations [1] to map out and characterize Hf1-xZrxO2 and to find optimal configurations in the materials space. We built an initial dataset of Hf1-xZrxO2 structures and their corresponding single point DFT energies. A kernel ridge regression (KRR) machine learning model is trained on this dataset and then further refined with active learning that picks appropriate structures from DFT relaxation trajectories. The atomic structures are represented in the vector form with many-body tensor representation (MBTR). With the refined KRR model, we explored the full HZO configuration space and optimized for structural stability and FE (also calculated with DFT).
[1] J. Laakso, M. Todorović, J. Li, G. X. Zhang & P. Rinke, *Compositional engineering of perovskites with machine learning*, Physical Review Materials, 6(11), 113801, (2022).
Keywords: Machine learning; Bayesian optimization; HZO; Ferroelectric memory