First-Principles Calculations of the Thermodynamic, Structural, Electronic, and Optical Properties of Compositionally Disordered Semiconductor Materials

Abstract

Alloyed semiconductor systems can provide improved properties beyond their unalloyed counterparts. Atomistic modeling of such systems was previously infeasible, but increases in computational power and improvements in modeling methods for random alloys have opened up this field of study, as well as unalloyed systems with natural disorder. Using these techniques, we analyze the structure-composition-property relationships in many technologically relevant alloyed and disordered semiconductor systems. First, we show the band gap insensitivity of the solid-state Li electrolyte LLZO to tantalum and aluminum doping. Next we show that antisite defects play a crucial role in ferromagnetic ordering within the high Tc ferromagnetic semiconductor FeSb2Se4. We then predict the large absorption coefficient and optimal band gap of Cu4TiSe4 for function as a thin film solar cell absorber material, but also its tendency to form copper-related defects that would hinder functionality as a photovoltaic. We analyze new inorganic compounds, SrHfSe3 and (Ge, Sn, Pb)(S, Se, Te). Finally, we explore boron incorporation in group-III nitrides. We show that the incorporation of boron into InGaN, the material that is the active layer in blue LEDs and won the 2014 Nobel prize in physics, can maintain the electronic properties of InGaN while better matching its structural properties to the surrounding GaN. This will allow for thicker active layers to be grown in devices, improving overall device efficiency and reducing device size. Additionally, BAlGaN alloys are shown to be easy to lattice match to AlN with band gaps similar to AlGaN alloys of the same gallium content, but with a smaller structural u parameter that correlates to a TE polarized light emission that is easier to extract from LEDs. Additionally, we analyze the polarization of BAlGaN alloys. Lastly, we also use machine learning on combined experimental and computational data for exploring vast chemical spaces for potential scintillator host materials. By adding local structural information into a machine learning model, we show improved ability to predict doping Ce d energy levels inside various host materials.