Executive Summary : | Thermoelectric materials play a crucial role in the development of sustainable energy-efficient technologies like power generators, sensors, and waste heat recovery systems. However, conventional methods face challenges due to higher computational costs and time required. To efficiently screen known and undiscovered 2D thermoelectrics with high figure of merit (zT) values, advanced data-driven approaches must be fused with material modeling. The poor mechanistic understanding of inferior TE efficiencies in bulk materials hinders the successful discovery of 2D TEs. This work proposes a novel fusion strategy of artificial intelligence and physics-based materials simulation, using a powerful reinforcement learning-based Monte Carlo Tree search (MCTs) algorithm, DFT, and quasi-classical Boltzmann models. The MCTs algorithm is combined with materials' uniqueness criteria and an on-the-fly surrogate model to minimize computational resources and improve the search quality. The study aims to analyze the transition from 2D to bulk form, layer thickness, and interaction processes between electron and phonon energy carriers for new TEs. The output will include a new 2D TE database and a streamlit web-application for quick access to trained ML models. |