Integrating Physics-based Models with Data-driven Methods for Materials Discovery

Metal-insulator   transition   (MIT)   compounds   are   materials   that   can undergo an electronic phase changes and are promising platforms to build next-generation   low-power   microelectronics.   Accelerated   discovery   is challenging    using    high-throughput    screening    because    high-fidelity quantum-mechanical    simulations    are    computationally    prohibitive    to perform. We solved this problem by building a supervised machine-learning model  that  can  classify  whether  a  material,  given  its  structure  as  input, would exhibit a thermal MIT.

  • •We   created   the   first   the   public   dataset   of   thermally-driven   MIT compounds     using     natural     language     processing     schemes     in collaboration  with  the  Olivetti  group  (at  MIT)  and  publicly  disseminated the new database.
  • •Compounds  were  featurized  using  new  physics-based  descriptors, which  were  integrated  into  the matminer  Python  library  for data  mining the properties of materials.
  • •We also deployed an easy-to-use online pipeline that can enable quick probabilistic of any crystalline material. Promising compounds identified with    classification  model  are  undergoing  experimental  validation,  in collaboration with the Wilson group at UCSB.

Additional Materials