We have carried out an in-depth structural study of Ge thick layers grown using the so-called “LT/HT” approach on Si(0 0 1), Si(0 1 1) and Si(1 1 1) wafers. The low temperature (400 °C) adopted for the first Ge layer plastically relaxes the strain in the Ge film without 3D islanding. The high temperature (750 °C) used for the growth of the second, topmost, Ge layer lowers the dislocation density and reduces the overall deposition time. High temperature thermal cycling (in-between 750 and 890 °C) was called upon to further reduce the amount of defects in the layers. Ge growth rates on (0 0 1) were consistently higher than the ones on (1 1 1), which were themselves higher than on (0 1 1) (be it at 400 °C, 100 Torr or at 750 °C, 20 Torr). The surfaces of Ge (0 0 1) layers were rather flat, as attested by their small root mean square (rms) roughness and Z ranges (<1 and 10 nm, respectively). By contrast, Ge (0 1 1) (Ge (1 1 1)) thick layers were really rough, with rms roughness and Z ranges typically 30 (60) times higher than on (0 0 1). Almost all Ge layers were in a tensile-strain configuration, with macroscopic degrees of strain relaxation in-between 101% and 106% (slightly higher on (1 1 1) than on (0 0 1) and on (0 1 1), however). Misfit dislocations were found (by cross-sectional transmission electron microscopy) to be confined in the first few hundreds of nanometres of 2.5 μm thick Ge layers on Si(0 0 1). As far as 2.5 μm thick Ge layers on (0 1 1) and (1 1 1) are concerned, numerous {1 1 1} stacking faults were present at the Ge/Si interface and also threading all the way to the surface. The threading “defect” density, which is close to 10 7 cm −2 on (0 0 1), is around 8×10 8 cm −2 on (0 1 1) and roughly 2×10 9 cm −2 on (1 1 1). The higher defect density on (0 1 1) and (1 1 1) is likely due to 60° misfit dislocations dissociation (into 30° and 90° partial dislocations). For such surface orientations, the 90° partial dislocation is expected to lead with in its trail the 30° partial dislocation (hence the stacking faults multiplication).
Read full abstract