Partha Sarathi Dutta1,2,Aditya Koneru1,2,Sukriti Manna1,2,Adil Muhammed1,2,Karthik Balasubramanian1,2,Troy David Loeffler1,2,Henry Chan2,Subramanian Sankaranarayanan1,2
University of Illinois at Chicago1,Argonne National Laboratory2
Partha Sarathi Dutta1,2,Aditya Koneru1,2,Sukriti Manna1,2,Adil Muhammed1,2,Karthik Balasubramanian1,2,Troy David Loeffler1,2,Henry Chan2,Subramanian Sankaranarayanan1,2
University of Illinois at Chicago1,Argonne National Laboratory2
Bismuthene is a two-dimensional layer of bismuth arranged in a honeycomb lattice configuration. It possesses some unique properties – for example, it is a topological insulator which means it is conductive or allows electrons to flow on the edges but acts as an insulator in its bulk. Also, bismuthene has a direct bandgap, strong light-matter interaction, and its optical properties can be tailored by varying its layer thickness which makes it an exciting candidate for optoelectronics applications. Despite its technological significance, there is a scarcity of models that can accurately describe the structure and energetics of bismuthene. Using a Tersoff formalism and leveraging the recent advances in reinforcement learning, we introduce a potential model that captures the structures and energetics of several different bismuthene polymorphs. The training data consisted of ab-initio calculations of the crystal lattice structure, cohesive energy, equation of state, elastic constants of five polymorphs, and phonon dispersion curve of the most stable low-buckled structure of bismuthene. We deploy our in-house reinforcement learning-based continuous-Monte Carlo Tree Search Algorithm (c-MCTS) because of its suitability to model complex non-linear interatomic interactions and proven performance in solving high-dimensional parameter space optimization problems. The RL-optimized bond order potential model performs well in capturing the crystal lattice structure and energetics of all the polymorphs. It especially gives an excellent prediction of the elastic constants of the low-buckled polymorph and a reasonable agreement between the phonon dispersion predicted by the model relative to the DFT reference training data. The optimal parameters are used to study the mechanical behavior, mechanical performance in electronic or optoelectronic applications, and thermal properties of bismuthene for applications in microelectronics.