Identifying models of dielectric breakdown strength from high-throughput data via genetic programming

Fenglin Yuan,Tim Mueller

doi:10.1038/s41598-017-17535-3

Abstract

The identification of models capable of rapidly predicting material properties enables rapid screening of large numbers of materials and facilitates the design of new materials. One of the leading challenges for computational researchers is determining the best ways to analyze large material data sets to identify models that can rapidly predict a given property. In this paper, we demonstrate the use of genetic programming to generate simple models of dielectric breakdown based on 82 representative dielectric materials. We identified the band gap Eg and phonon cut-off frequency ωmax as the two most relevant features, and new classes of models featuring functions of Eg and ωmax were uncovered. The genetic programming approach was found to outperform other approaches for generating models, and we discuss some of the advantages of this approach.

Highlights

With the ever-increasing power of supercomputers, materials scientists are able to perform high-throughput density functional theory (DFT) calculations[1,2] and build up online databases[3,4,5,6] of important materials properties including structure parameters, thermodynamic and transport properties, and electronic structures and properties
We demonstrate the power of a supervised learning algorithm known as “genetic programming”, in which an evolutionary algorithm is used to perform symbolic regression, for identifying simple models for dielectric breakdown strength of materials
Mueller et al.[19] have previously demonstrated that genetic programming can be used to identify important structural descriptors for hole trap depths in hydrogenated nanocrystalline and amorphous silicon, and here we evaluate the genetic programming approach for the identification of predictive models of dielectric breakdown strength

Summary

Introduction

With the ever-increasing power of supercomputers, materials scientists are able to perform high-throughput density functional theory (DFT) calculations[1,2] and build up online databases[3,4,5,6] of important materials properties including structure parameters, thermodynamic and transport properties, and electronic structures and properties Such vast amounts of materials information enables the use of machine learning methods to identify simple predictive models of more complex material properties[7,8,9,10,11,12,13,14,15,16,17,18]. Mueller et al.[19] have previously demonstrated that genetic programming can be used to identify important structural descriptors for hole trap depths in hydrogenated nanocrystalline and amorphous silicon, and here we evaluate the genetic programming approach for the identification of predictive models of dielectric breakdown strength

Methods

Results

Conclusion