講演情報
[17a-K505-6][The 46th Young Scientist Award Speech] Phase and property control of heterogeneous HfxZr(1-x)O2 thin films by machine learning
〇Zeyuan Ni1, Hidefumi Matsui1 (1.TTS)
キーワード：
materials informatics、high-k、phase control
HfxZr(1-x)O2 is renowned for its polymorphic nature, which confer a range of electronic and dielectric properties. Its metastable non-monoclinic (NM) phases exhibit enhanced properties such as high-k, ferroelectricity, anti-ferroelectricity, but they could only be stabilized under specific conditions. To explore a more complex heterogeneous HfZrO film, machine learning-incorporated closed-loop experiments is suitable but had yet to be reported in 2020.
As our first trail to stabilize and maximize the NM phases of PVD HfZrO films, we adopted closed-loop experiments through Bayesian optimization (BO). Within 10 cycles, we observed an impressive enhancement in NM phase, measured by XRD, from approximately 30% to nearly 100%. The NM phase ratio of the optimized composition depth profile significantly outperforms that of a similar structure by stacking of pure ZrO2 and HfO2. Furthermore, we moved on to atomic layer deposition (ALD) and multi-object optimization of electronic properties, such as k and leakage. With our home-made algorithm designed for ALD, we successfully pushed the Pareto frontier, demonstrating the effectiveness of our approach.
As our first trail to stabilize and maximize the NM phases of PVD HfZrO films, we adopted closed-loop experiments through Bayesian optimization (BO). Within 10 cycles, we observed an impressive enhancement in NM phase, measured by XRD, from approximately 30% to nearly 100%. The NM phase ratio of the optimized composition depth profile significantly outperforms that of a similar structure by stacking of pure ZrO2 and HfO2. Furthermore, we moved on to atomic layer deposition (ALD) and multi-object optimization of electronic properties, such as k and leakage. With our home-made algorithm designed for ALD, we successfully pushed the Pareto frontier, demonstrating the effectiveness of our approach.