Defect & Dopant Predictions for Thermoelectric Materials

Approach  1: 
Experimental  training  sets  concerning  dopability  have 
been assembled and serve as the basis for machine learning models 
to predict the maximum dopability range in materials (left Figure). We 
have   achieved   predictive   dopability,   with   carrier   concentration 
predicted within one order of magnitude. 
Approach 2: 
Rather  than  relying  on  experimental  literature  and  our 
on-going   experimental   efforts   to   build   a   training   set,   a   similar 
database  has  been  constructed  using  DFT  calculations  of  defects  in 
model  systems.    We  have  validated  the  defect  calculations  against 
experimental measurements.
Looking  forward,  we  will  combine  these  two  training  sets  to 
provide greater predictive accuracy.