Atomistic Models of Amorphous Semiconductors

Crystalline silicon is probably the best studied material, widely used by the semiconductor industry. The subject of this thesis is an intriguing form of this element namely amorphous silicon. It can contain a varying amount of hydrogen and is denoted as a-Si:H. It completely lacks the neat long range order of the crystal, yet its structure is not random. Almost all silicon atoms have four neighbors and the average bond angle is identical to the tetrahedral angle in the crystal. Order is thus preserved over several bond lengths. The motivations to study a-Si:H are two-fold. Firstly some of its properties are different from the crystalline form and we do not understand them completely. For example, the electronic properties degrade after exposure to intense light, but can be recovered reversibly by heat treatment. The microscopic process of this is not known. Secondly, research on a-Si:H is motivated by its applications. These are mostly large area devices such as liquid crystal displays and solar cells. The latter are in use already today, the former are waiting to be widely used in future. Amorphous semiconductors can be deposited over large areas from vapor. On the other hand, the size of c-Si devices is limited by the much smaller size of the wafers. The production of a-Si:H is also cheaper and consumes less energy. Unlike its crystalline counterpart a-Si:H has a direct band gap, leading to an increased light absorption. Consequently, a-Si:H solar cells are ~ 1000 times thinner than c-Si cells, resembling more a foil than a semiconductor device. The methods used in the thesis are computational, largely relying on algorithms and powerful computers. The structural models are atomistic, where the interaction between electrons and nuclei is treated on the level of Density Functional Theory. This is a first-principles methods, meaning that it does not use any adjustable parameters. The chemical bonding, even of complex structures is described accurately. Calculation of total energies and forces allows us to find equilibrium structures and perform molecular dynamics calculations. The models of a-Si:H are prepared by cooling a melt to room temperature. This method resembles the preparation of glasses. We find that the structure is strongly in influenced by the cooling rate. Using slower cooling rates we improved existing models that contained excessive strain and a high defect concentration. Using a cooling rate of ~ 0.02 K/fs we were even able to prepare small defect-free models. The structure was in good agreement with available neutron scattering data. Calculated density of states shows a pronounced band gap. After the generation of structural models we turn our attention to defects. Defects in an amorphous solid are defined as atoms that deviate from the normal coordination. We find 3-fold and 5-fold coordinated Si atoms and 2-fold coordinated H atoms. We focus only on the 3-fold coordinated Si, also called the dangling bond (DB), that is believed to be the major defect in a-Si:H. We have calculated formation of the DB defect in the negative, neutral and positive charge state. By averaging over 25 distinct DB models we find a considerable spread in the energies of 0.2 eV. Another related property of a defect is its correlation energy U. A positive value of U means that we have to invest energy to add an extra electron to the defect. The size and sign of U are still a subject of controversy. On average we find a positive U value of 0.1 eV. Four models, however, have a negative correlation energy, suggesting large relaxations in the defect structure. Amorphous silicon readily forms compounds with nitrogen and carbon. We have investigated silicon-rich nitride (a-SiN:H) at two different densities of 2.0 and 3.0 g/cm3. Features in the pair-distribution functions can be related to "square structures". These are planar structures consisting of two Si in opposite corners of a square and two N in the remaining corners. The dense phase shows signs of phase separation into silicon and stoichiometric nitride. Both valence and conduction band edges are dominated by Si states. This is corroborated by the fact that by increasing the nitrogen content the band gap of the nitride can be varied from 1.8 to 5.3 eV. Recently there has been a considerable interest in man-made materials. Examples are multilayers (ML) formed by two semiconductors with a different band gap. By adjusting the thickness of the small band gap material (the well) one can tune the band gap of the ML due to quantum confinement effects. This concept is well established in crystalline semiconductors. The existence of quantum confinement in amorphous structures is, however, being still debated. Using models prepared previously we have constructed a model of a silicon/nitride ML. This allowed us to study confinement effects directly without using transport or optical measurements that can obscure the observations. Comparing our model to an experimental system with the same composition gave almost identical band gaps. This confirmed the existence of quantum confinement in a amorphous multilayer. The calculation of band offsets between the materials revealed that there is almost no barrier for the electrons and the confinement originated solely from holes