Phase control of heterogeneous Hf_xZr_(1-x)O₂ thin films by machine learning

Abstrsct Polymorphic Hf x Zr (1− ) O 2 thin films have been widely used as dielectric layers in the semiconductor industry for their high- k , ferroelectric, and antiferroelectric properties metastable non-monoclinic phases. To maximize components, we optimize composition depth profile of 20 nm PVD through closed-loop experiments by using parallel Bayesian optimization (BO) with advanced noisy expected improvement acquisition function. Within 40 data points, ratio phases is improved from ∼30% pure HfO ZrO to nearly 100%. The optimal sample has a 5 0.06 0.94 capping layer over 15 0.91 0.09 . thickness effect spontaneously explored BO. We prove that machine-learning-guided fine-tuning potential improve film performance beyond uniform or laminated crystals lead discovery novel phenomena.