Extended crystal defects often play a critical role in determining semiconductor device performance. This dissertation describes the application of transmission electron microscopy (TEM) and aberration-corrected scanning TEM (AC-STEM) to study defect clusters and the atomic-scale structure of defects in compound semiconductors.
An extensive effort was made to identify specific locations of crystal defects in epitaxial CdTe that might contribute to degraded light-conversion efficiency. Electroluminescence (EL) mapping and the creation of surface etch pits through chemical treatment were combined in attempts to identify specific structural defects for subsequent TEM examination. Observations of these specimens revealed only surface etch pits, without any visible indication of extended defects near their base. While chemical etch pits could be helpful for precisely locating extended defects that intersect with the treated surface, this study concluded that surface roughness surrounding etch pits would likely mitigate against their usefulness.
Defect locations in GaAs solar-cell devices were identified using combinations of EL, photoluminescence, and Raman scattering, and then studied more closely using TEM. Observations showed that device degradation was invariably associated with a cluster of extended defects, rather than a single defect, as previously assumed. AC-STEM observations revealed that individual defects within each cluster consisted primarily of intrinsic stacking faults terminated by 30° and 90° partial dislocations, although other defect structures were also identified. Lomer dislocations were identified near locations where two lines of strain contrast intersected in a large cluster, and a comparatively shallow cluster, largely constrained to the GaAs emitter layer, contained 60° perfect dislocations associated with localized strain contrast.
In another study, misfit dislocations at II-VI/III-V heterovalent interfaces were investigated and characterized using AC-STEM. Misfit strain at ZnTe/GaAs interfaces, which have relatively high lattice mismatch (7.38%), was relieved primarily through Lomer dislocations, while ZnTe/InP interfaces, with only 3.85% lattice mismatch, were relaxed by a mixture of 60° perfect dislocations, 30° partial dislocations, and Lomer dislocations. These results were consistent with the previous findings that misfit strain was relaxed primarily through 60° perfect dislocations that had either dissociated into partial dislocations or interacted to form Lomer dislocations as the amount of misfit strain increased.