Ab initiomodels of amorphous InN

B. Cai, D. A. Drabold
2009 Physical Review B  
In this paper, we present the first structural model of amorphous indium nitride obtained from first-principles simulation. We created a small 64-atom model by quenching from the melt and analyzed a chemically ordered 250-atom model of Mousseau and Barkema. We find that both N and In atoms tend to be fourfold. Upon relaxation, we find no homopolar bonds in the small cell and only one in the 250-atom cell. The topology of the models is analyzed with pair-correlation functions, bond angle
more » ... bond angle distributions, and ring statistics. The vibrational and electronic properties are also obtained. We found that density-functional methods in the local-density approximation predict a very small gap for amorphous InN, similar to the case for crystalline InN. Amorphous materials and glasses play ever more important roles in technology. Due to photovoltaic, infrared detection and imaging applications, optoelectronic devices, and potential utility for next generation flash memory devices, the physics of amorphous semiconductors has drawn renewed interest. Since most properties of amorphous semiconductors are determined by topology, the beginning of any such study is the creation of experimentally credible structural model. A material of considerable current interest is the narrow gap semiconductor InN. Also, since GaN is an established wide-gap material, it is appealing to consider InGaN alloys for photovoltaic and other applications. Studies in this direction 1 would benefit from basic information about amorphous InN ͑a-InN͒. These materials might possess a continuously variable range of optical gaps to optimize absorption of the solar spectrum. 2 There has been controversy over the band gap of zincblende crystalline-InN ͑c-InN͒ both in experimental and theoretical works. In experiment, a narrow band gap of 0.7 eV ͑Ref. 3 and 4͒ was reported, which contrasts with previous values near 1.89 eV. 5 Subsequently, these small gaps have been confirmed by additional experiments. 6-8 In theoretical work, calculations based on density-functional theory within the local-density approximation ͑LDA͒ always yield a tiny or even negative gap. 9 Methods using self-interaction and relaxation corrected pseudopotentials ͑SIRC͒ report a large gap around 1.3eV, 10 but semiempirical LDA methods show a gap around 0.85eV. 11 For amorphous InN ͑a-InN͒, a large optical gap around 1.7eV was measured in 2006. 12 However, no further experiments have been performed. No theoretical work has appeared on a-InN. In this paper, we present atomic models of amorphous InN obtained from ab initio molecular dynamics based on plane-wave LDA. The structural, dynamical, and electronic properties are discussed. To our knowledge, there is neither theoretical nor experimental work on structural properties or vibrational modes. After creating small but reasonable models of a-InN, we predict the vibrational spectrum and electronic properties. We particularly seek to connect the electronic structure to the topology of the network to better comprehend electronic and optical experiments. We demonstrate by direct calculation that the topology of a-InN is a chemically ordered continuous random network very much along the lines proposed by Mousseau and Barkema. 13 The simulations presented in our work are performed with the Vienna Ab initio simulation package ͑VASP͒ ͑Ref. 14͒ based on density-functional theory within the LDA and Vanderbilt's ultrasoft pseudopotentials. To construct a realistic a-InN model, we began with the quench from the melt technique for a 64-atom model. 15 The initial configuration was melted at 2000 K and equilibrated for 800 steps. Then, the system is quenched to 400K, with a mean quench rate 61 K / ps. The system initially possesses a nitrogen dimer. According to Mott's eight-N rule and the electronegativity of these two elements, in theory, it is unlikely to form homopolar bond. Thus, we artificially moved this pair apart, then re-equilibrated the system at 400 K for 1000 steps ͑2.5 ps͒. Then the 64-atom a-InN model was relaxed to an energy minimum. During the MD procedure, the volume of the cell was constant. During the final relaxation, we allowed the volume and shape of the cell to change to ensure a zero-pressure model with no artificially imposed symmetries imposed on the shape of the cell. To check our small model, a 250-atom model was formed by relaxing an a-GaAs model 13 with Ga and As atoms replaced by In and N atoms, respectively. We rescaled the cell to reproduce the density of c-InN ͑also the density for 64-atom model͒ and relaxed the system at constant volume. We present the topology of our final 64-atom model in Fig. 1 . Because the shape of the cell is allowed to change during relaxation, the final cell is not quite cubic, but nearly so. The density of our final model is 6.97 g / cm 3 which is modestly larger than 6.81Ϯ ͑0.05͒ g / cm 3 , the density of c-InN. Where coordination is concerned, we note that all N atoms are fourfold and all but two In atoms are fourfold. There are no "wrong" ͑homopolar͒ bonds such as N-N or In-In in our model. It is gratifying to see chemical order emerge so unambiguously from an unbiased melt-quench procedure, a strong indication that homopolar bonds are rare in the material. For comparison, in the 250-atom model, we PHYSICAL REVIEW B 79, 195204 ͑2009͒
doi:10.1103/physrevb.79.195204 fatcat:vpktiuac6nda3kcpm7564obs2m