Nucleation mechanisms and the elimination of misfit dislocations at mismatched interfaces by reduction in growth area

Abstract

To investigate the effect of growth area on interface dislocation density in strained-layer epitaxy, we have fabricated 2-μm-high mesas of varying lateral dimensions and geometry in (001) GaAs substrates with dislocation densities of 1.5×105, 104, and 102 cm-2. 3500-, 7000-, and 8250-Å-thick In 0.05Ga0.95As layers, corresponding to 5, 10, and 11 times the experimental critical layer thickness as measured for large-area samples, were then deposited by molecular-beam epitaxy. For the 3500-Å layers, the linear interface dislocation density, defined as the inverse of the average dislocation spacing, was reduced from greater than 5000 to less than 800 cm -1 for mesas as large as 100 μm. A pronounced difference in the linear interface dislocation densities along the two interface 〈110〉 directions indicates that α dislocations nucleate about twice as much as β dislocations. For samples grown on the highest dislocation density substrates, the linear interface-dislocation density was found to vary linearly with mesa width and to extrapolate to a zero linear interface-dislocation density for a mesa width of zero. This behavior excludes dislocation multiplication or the nucleation of surface half-loops as operative nucleation sources for misfit dislocations in these layers. Only nucleation sources that scale with area (termed fixed sources) are active. In specimens with lower substrate dislocation densities, the density of interface dislocations still varies linearly with mesa size, but the slope becomes independent of substrate dislocation density, indicating that surface inhomogeneities now act as the dominant source for misfit dislocations. Thus, in 3500-Å-thick overlayers, substrate dislocations and substrate inhomogeneities are the active fixed nucleation sources. Since only fixed nucleation sources are active, a single strained layer will dramatically reduce the threading dislocation density in the epilayer. For the 7000-Å layers, we observe a superlinear increase in linear interface-dislocation density with mesa size for mesas greater than 200 μm, indicating that dislocation multiplication occurs in large mesas. For mesas less than 200 μm in width, linear interface-dislocation density decreases linearly with mesa size, but extrapolates to a nonzero linear interface-dislocation density for a mesa size of zero. This nonzero extrapolation suggests an additional active source which generates a dislocation density that cannot be decreased to zero by decreasing the mesa size. Cathodoluminescence (CL) images using radiative recombination indicate that the additional source is nucleation from the mesa edges. Despite a doubling in epilayer thickness from 3500 to 7000 Å, the linear interface-dislocation density for mesas 100 μm in width is still very low, approximately 1500 cm-1. The 8250-Å layers possess interface-dislocation densities too high to be accurately determined with CL. However, increases in CL intensity as mesa width is reduced indicate that the interface-dislocation density is decreasing and that growth on small areas produces higher-quality layers than growth on large areas. Our investigations show that different sources for misfit dislocations become active at different epilayer strain levels. The critical thickness depends on which type of nucleation source becomes activated first; therefore, different critical thicknesses can be observed depending on which kind of source is present in a specimen.