Challenge: Defects and scarcity of dopants often are the Achilles heel to realizing the theoretical potential of new semiconductors (eg transparent conductors, thermoelectrics). Further, native defects and dopants are computationally expensive to accurately calculate.  Two approaches to overcome this challenge are in progress:  

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.

Union: Looking forward, we will combine these two training sets to provide greater predictive accuracy.