42nd 2000 Electronic Materials Conference
Symposium: Point and Extended Defects in Mismatched Materials (STUDENT)


P. Feichtinger, B. Poust, M.S. Goorsky, Dept of Materials Science and Engineering, University of California, Los Angeles; D. Oster, T. D'Silva, and J. Moreland, Wacker Siltronic Corporation, Portland, OR.

Substrates with a high degree of miscut are often used with mismatched layers. In this study, the interactions among misfit dislocations at strained layer / miscut substrate interfaces were examined to understand how interaction of extended defects in mismatched layers grown on miscut substrates differs from the interactions that are observed for layers grown using on-axis substrates. Misfit dislocations with opposite tilt components (that are parallel for a structure grown on an on-axis substrate) are inclined by opposite angles with respect to the <110> when the structure is grown on a miscut substrate. We determined that these non-orthogonal misfit segments could act to block each other and to cause cross-slip of those segments and change their relaxation direction. This can be shown to lead to localized preferential tilt of the lattice due to certain sets of interactions in one area of the wafer, while other areas show an opposite tilt due to other sets of dislocation interactions. Nominally (001) silicon epitaxial layers were deposited on highly p-type (2.61019 cm-3) silicon substrates. This system has a mismatch of about 1.510-4; this low level of mismatch mimics the early stages of relaxation in both graded buffer layers and in strained single layers. Substrate miscuts of zero, 2.3, and 4.6 were used with the miscut direction along either a <110> or <100> direction. Layers were grown at different thicknesses and the as-grown wafers were also subjected to rapid thermal annealing to separate the influences of temperature and the stress acting on the dislocations. Double crystal x-ray topography is sensitive to the tilt and screw components of the different misfit segments and clearly delineates their interactions across the entire 150 mm wafer. The statistical advantage of examining dislocation interactions across an entire wafer confirmed that an excess of a certain type of dislocations (i.e., with the same Burger's vector) in one region does not necessarily represent the status of the dislocation distribution in other regions. This demonstrates why recent models proposed for the formation of tilt in partially relaxed layers that are based on measurements that use small sampling areas (such as TEM or single spot XRD measurements) are inconsistent with this and other experiments [F. Romanato, et al., J. Appl. Phys. 86 4748 (1999)] that show variation in the tilt distribution at different locations.