Featureless Optimization of Material Properties for Small Data Sets in Complex Ceramics

Electronic   materials   that   exhibit   phase   transitions   between metastable   states   (e.g.,   metal-insulator   transition   materials   with abrupt   electrical   resistivity   transformations)   are   challenging   to decode. For these materials, conventional machine learning methods display  limited  predictive  capability  due  to  data  scarcity  and  the absence  of  features  that  impede  model  training.  In  this  work,  we developed  an  adaptive  optimization  engine  that  overcomes  these limitations. It:

  • •Directly learns the composition-property relationship
  • •Enables efficient co-design of functional materials
  • •Suitable for limited data and descriptor availability
  • •Applicable to inverse design problems

We used it to identify  12 functional and synthesizable metal-insulator transition  (MIT)  materials  without  a  large  starting  data  set  or  using hand-tuned   features.   These   materials   exhibit   the   lacunar   spinel structure  and  cation  ordered  transition  metals  cations. The  approach also led to identification of a semiconductor-to-insulator transition and improved understanding of the transport properties.

Additional Materials