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

Abstract Polymorphic HfxZr(1-x)O2 thin films have been widely used as dielectric layers in the semiconductor industry for the high-k, ferroelectric, and antiferroelectric properties in the metastable non-monoclinic phases. To maximize the non-monoclinic components, we optimize the composition depth profile of 20 nm PVD HfxZr(1-x)O2 through closed-loop experiments by using parallel Bayesian optimization (BO) with the advanced noisy expected improvement (NEI) acquisition function. Within 40 data points, the ratio of non-monoclinic phases is improved from ~30% in pure 20 nm HfO2 and ZrO2 to nearly 100%. The optimal sample has a 5 nm Hf0.06Zr0.94O2 capping layer over 15 nm Hf0.91Zr0.09O2. The composition and thickness effect of the capping layer has been spontaneously explored by BO. We prove that machine-learning-guided fine-tuning of composition depth profile has the potential to improve film performance beyond uniform or laminated pure crystals.