MULTISCALE MODELING AND MACHINE LEARNING STUDIES OF THE DIFFUSION OF SILICON AND INTRINSIC DEFECTS IN III-V SEMICONDUCTORS
Integrating III-V semiconductors into next-generation silicon-based processing is a promising alternative being considered as a route to faster and more energyefficient electronic devices. These III-V materials will be doped, typically with Si as a dopant. However, dopant activation remains an issue, compounded by the fact that there is still insufficient knowledge of the ease and preferred mechanistic pathways by which dopants, like Si, become activated within the III-V matrix. Using Density Functional Theory (DFT) and Nudge Elastic Band (NEB) calculations, we have determined many of these critically important properties, namely, the energy barriers associated with the diffusion of both intrinsic point defects and silicon impurities in prototypical III-V materials, here Zinc Blende GaAs, InAs and the CuAuI-ordered ternary In0.5Ga0.5As. Refuting assumptions in the current literature that the enhanced diffusion can be attributed primarily to an increase in vacancies, vacancy-assisted diffusion of isolated Si atoms was found to be an unfavorable mechanism for this group of semiconductor alloys. Our results show that new and highly mobile species that are created at high dopant concentration are instead responsible for the enhanced diffusion observed. Those new species include Si complexes such as Si-Si pairs and Si split interstitials which can move more easily within the crystal lattice. We use these DFT results to inform the development of a continuum model that addresses limitations in current models and shows great agreement with experimental results. We also develop a new method whereby machine learning in lieu of DFT, is used to predict forces during NEB simulations. This new method allows to compute transition pathways at a fraction of the cost while maintaining reasonable accuracy compared to the traditional DFT approach.
silicon; III-V Semiconductors; Diffusion; Chemical engineering; Materials Science; Computational Chemistry; Doping; machine learning; Ab Initio
Escobedo, Fernando; Thompson, Michael Olgar; Pepiot, Perrine
Ph. D., Chemical Engineering
Doctor of Philosophy
dissertation or thesis