Dislocation Densities within RLG Buffers

Figure 4.14: Examples of etched sample surfaces which have different TDD levels due to their different grading rates. a) Sample 4054 with a grading rate of 4.69 % µm-1 and TDD to the order of 106 cm-2. b) Sample 4130 with a grading rate of 324 % µm-1 and TDD to the order of

108 cm-2.

A 5 second diluted iodine etchant (3:1 ratio, Section 3.5) was used to reveal threading dislocations in the SiGe cap layer. An example of an etched sample is shown in Figure 4.14a) for sample 4054 with a low GRRLG of 4.69 % µm-1 which has

a low TDD of order 106 cm-2. Figure 4.14b) shows a DIC photo of etched sample

4130 with a high grading rate of 324 % µm-1with a high TDD of ~108 cm-2

Figure 4.15 shows the measured TDD as a function of grading rate. The horizontal grey line represents the TDD of 2 x 107 cm-2

4.2.2

from the Ge underlayer (sample 4045, Section ). Increasing the grading rate within RLG buffers shows that the TDD shifts from dislocation gliding to the nucleation regime [44]. This conclusion is drawn because the TDD levels within the SiGe cap are greater than those grown into

the buffer from the Ge underlayer. The lower energy barrier to dislocation nucleation of high Ge composition SiGe alloys suggests that reverse graded buffers are more susceptible to nucleation of dislocations.

Figure 4.15: TDD variation with grading rate. The horizontal line shows the initial TDD within the Ge layer. A grading rate below 124 % µm-1 allows the system to enter the glide regime and hence achieve lower-than-initial layer dislocation densities. The lines shown are guides for the

eye.

A constant composition Si0.21Ge0.79 cap which is deposited directly on the Ge

underlayer (sample 3695) is calculated to have a GReff of about 2100 % µm-1 and is

measured with a TDD of 7.7 x 107 cm-2. This shows that a direct cap of constant

composition material deposited at 850°C will have a higher dislocation nucleation rate than the threading arm annihilation rate, as it contains a TDD higher than that of the Ge underlayer (2 x 107 cm-2). The misfit between the cap layer and the Ge

As the grading rate is decreased, the TDD of the samples reaches a maximum value of 1.5 x 108 cm-2 at a GRRLG of 174 % µm-1. Section 4.4.2 reports that as the GRRLG

is decreased to 174 % µm-1 the roughness of the buffer surface increases. It is

postulated that this high density of dislocations is likely to be due to the increase of the heterogeneous nucleation rate, promoted by the extra roughness incurred on the buffer surface.

Figure 4.16: Cross sectional TEM micrographs of samples a) 4129 (tRLG = 136 nm) and b)

sample 4128 (tRLG = 193 nm). It can be seen that the MFR mechanism decreases from a) to b)

and that the number of resulting threading dislocations decrease as well. The second observation to be noted is that the stacking fault density increases from a) to b). Lines shown are

only guides for the eye.

Further decreasing the GRRLG to 124 % µm-1 (sample 4128) reduces the TDD to

2 x 107 cm-2. This sample has been measured to have roughest surface (Section 4.3.2)

and is speculated to have a high heterogeneous nucleation rate. As the TDD in the final SiGe cap layer is the same as the TDD from the underlying Ge layer the total nucleation and annihilation rates are assumed to be equal to each other. Therefore the reduction in TDD could only be due to a reduction in the multiplication nucleation

rate (Section 2.4.8). This mechanism is shown in Figure 4.16, where two TEM cross sectional micrographs of sample 4129 (GRRLG = 172 % µm-1) and sample 4128

(GRRLG = 124 % µm-1

2.6.3

) are shown. It is seen that as the thickness of the graded layer is increased the MFR nucleation mechanism is lowered. As with forward graded buffers (Section ) the thickness of the graded region will determine the [001] spacing between the misfit dislocations in the RLG region. This determines the interaction between misfit dislocations which directly affects the MFR multiplication nucleation.

Stacking faults are also seen within Figure 4.16a) and b) and are seen to increase in density as the graded region thickness is increased. Stacking fault density (SFD) is reported in Section 4.3.4. It is speculated that the SFD is too low to severely hinder the annihilation rate of dislocations by blocking mobile dislocations.

When the GRRLG within the system reaches 61.3 % µm-1 and is decreased to

11.4 % µm-1

Figure 4.7

the buffers are grown in a 2D mode, therefore heterogeneous nucleation is assumed to be low. It is also seen in , Figure 4.8 and Figure 4.9 that the nucleation of dislocations through multiplication is limited, as lower grading rates are used. Threading dislocations are not seen in TEM micrographs of thicker layers (Figure 4.7, Figure 4.8 and Figure 4.9), indicating a low TDD. As the GRRLG is

reduced from 61.3 % µm-1 to 11.4 % µm-1 the TDD is calculated from EPD

measurements to reduce from 4.5 x 106 cm-2 to 3.3 x 106 cm-2. This indicates that

only if there is a residual macroscopic strain in the layer and a high temperature is applied (Equation 2.14). It is reasonable to assume that the reduction of the TDD with reduced grading rates is due to the extra thermal budget applied to the buffer during the growth of the thicker structures. Within this GRRLG

Figure 4.15

range the system is thought to be within a temperature-limited sub-regime ( ). Further reducing the GRRLG to 4.65 % µm-1 increases the TDD to 4.6 x 106 cm-2. It is

surmised that the annihilation rate limited by the strain gradient of the RLG layer for this GRRLG and is deemed to be in a strain-limited sub-regime.

An anneal of 750°C for 10 hours on a RLG buffer with a GRRLG of 29.0 % µm-1

(sample 4050) lowers the TDD to a minimum of 2.4 x 106 cm-2

2.6.7

. These TDD values are comparable to the low temperature technique and traditional linear grading without a CMP mid-step (Sections and 2.6.3) whilst superior to the Ge condensation technique and the two temperature method (Sections 2.6.5 and 2.6.8). However, when CMP is applied to conventional forward graded buffers (Section 2.6.3), dislocations in pile ups are no longer impeded, which allows mobile threading dislocations to achieve a TDD of 105 cm-2.

Pile-up of defects has been shown in forward graded structures to result from severe crosshatch on the buffer [49]. Characterisation of pile up is accomplished by etching of the top layer as described in Section 3.5 and then observation though a DIC microscope. Pile up has not been observed on RLG buffers.