AlN interlayers (ILs) not only affect the compressive stress introduction to overlying GaN, but also affect the quality of GaN on them. AlN interlayer was first applied to reducing the dislocation density in GaN and the control of stress to yield crack free AlGaN on sapphire [45, 46]. Some study reported that AlN IL was favorable to improving the quality of GaN [47], but based on our study it is not true for any AlN IL and dependent on the growth conditions of them. In this section, the effect of AlN ILs on the quality of overlying GaN was studied by XRD rocking curve measurement and detailed TEM observation about the dislocation generation and propagation. The investigated conditions and structures were the same as those for stress study of effects of AlN IL, as sketched in Table 3.2 and Fig. 3.17 in chapter 3, including conditions of thickness, growth temperature and V/III. Multi-condition and single-condition samples had been grown. XRD rocking curve measurement was carried out only for the single-condition-AlN -IL sample, while the TEM observation was executed only for multi-condition-AlN-IL sample. Prior to introducing XRD results, the curvature curves of the single-condition-AlN-IL samples were introduced first for the later discussion.
As shown in Fig. 5.8, the curvature increase for the 1st GaN 𝑑𝑑𝜅𝜅𝐺𝐺𝑎𝑎𝐺𝐺1 was extracted to evaluate Fig. 5.8 Structure and curvature curve of single-condition AlN IL sample.
128
its quality which may influence the behavior of the following layers. Because based on the results in the previous section, for GaN on AlN buffer layer, large 𝑑𝑑𝜅𝜅𝐺𝐺𝑎𝑎𝐺𝐺1 usually indicates high quality of it. Only the first 200 nm of the 2nd GaN layer was sampled to compare the compressive curvature increment brought by AlN interlayers grown under various conditions, as the curvature curves in the first 200 nm were linear, and the sampled thickness was the same with that of the GaN layer in between AlN interlayers in multi-condition samples.
The first investigated condition is the thickness of AlN ILs, which was varied from 4.5 nm to 45 nm. The results of XRD rocking curve as well as the curvature increase to the compressive side in the first and second GaN were extracted in Fig. 5.9a and b respectively.
From Fig. 5.9a, the quality of GaN became higher as the thickness of AlN IL increased from 4.5 nm to 22 nm and after that it degraded. But such description is not reliable without examining the measurement carefully. For confirming the effects of AlN ILs on GaN quality, the quality of only the 2nd GaN layer should be accessed. But in the measurement for the planes of (0002) and (10-12), as sketched by Fig. 5.10a, b and d, because the X-ray incident angel 𝜃𝜃 is very large, so the penetration depth of X-ray 𝑡𝑡 would be very large [48]. 𝑡𝑡 can be calculated by the following equation [49]:
𝑡𝑡 =−sin 𝜃𝜃 ln(1−𝐺𝐺𝑡𝑡)
2𝜇𝜇 , 𝜇𝜇= (𝑚𝑚𝐺𝐺𝑎𝑎∙ 𝑢𝑢𝐺𝐺𝑎𝑎 +𝑚𝑚𝐺𝐺∙ 𝑢𝑢𝐺𝐺)∙ 𝑁𝑁 ∙ 𝑚𝑚/𝑉𝑉𝐺𝐺𝑎𝑎𝐺𝐺 (5-2) 𝜃𝜃 is the incident angel, 𝐺𝐺𝑡𝑡 is the fraction of , 𝜇𝜇 is mass absorption coefficient, 𝑚𝑚𝐺𝐺𝑎𝑎 and 𝑚𝑚𝐺𝐺
are the atomic mass of gallium and nitrogen respectively, 𝑢𝑢𝐺𝐺𝑎𝑎 and 𝑢𝑢𝐺𝐺 are mass absorption of gallium and nitrogen respectively, 𝑁𝑁 is the number of gallium or nitrogen in a unit cell of GaN crystal, 𝑚𝑚 is a unit of atomic mass (1 amu), 𝑉𝑉𝐺𝐺𝑎𝑎𝐺𝐺 is the volume of a unit cell of GaN crystal. By calculation, the largest penetration depth (𝜃𝜃 = 90°) can be 42.59 um. The penetration depth of the measurement of plane (0002) (𝜃𝜃≈ 17.3°) can be 12.665 um. So for the measurements of planes of (0002) and (10-12), the diffraction information came from not only the 2nd GaN, but also from the 1st GaN. Due to larger thickness of the 1st GaN than the 2nd, the information from the 1st GaN layer is dominant in the results of planes of (0002) and (10-12). The information of the results of measuring (10-10) plane comes from only the top
Fig. 5.9 (a) FWHM of XRD rocking curve of GaN samples with single-layer AlN IL for thickness study; (b) curvature increase to the compressive side in the 1st GaN and the first 200 nm in the 2nd
GaN layer.
129
2nd GaN layer since the incident angel of in-plane measurement is very small around the critical angle and the sampled depth can be only tens of nanometers.
Another point should be noticed is that the factors of the quality of the 2nd GaN layer, or, the source of dislocations in the 2nd GaN layer. Dislocations in the 2nd GaN can come from the 1st GaN layer, i.e., by propagation of threading dislocations; or originate from the AlN IL.
Consequently, the quality of the 2nd GaN is determined by the quality of the 1st GaN and the AlN IL. In order to survey the influence only from the AlN IL, the quality of the 1st GaN should be as identical as possible, which usually is not the case but with some minimal variation from round to round due to the fluctuation of reactor conditions. However, basically, such variation doesn’t affect the reliability of studies principally. Based on these preliminaries, the discussion of effects of AlN ILs can be started.
Fig. 5.9a is the results of XRD rocking curve measurements. The tendency of (0002) plane and (10-10) plane showed slight difference. There was minimal variation of the FWHM of (0002) plane which included the information from the 1st GaN dominantly. By comparing Fig. 5.9a and b, it can be found that the tendency of FWHM of (0002) plane and 𝑑𝑑𝜅𝜅𝐺𝐺𝑎𝑎𝐺𝐺−1 as functions of AlN IL thickness was the similar. There was some variation of 𝑑𝑑𝜅𝜅𝐺𝐺𝑎𝑎𝐺𝐺−1 due to some reactor condition fluctuation or contamination of the wafer surface, although the conditions were designed to be identical. From the discussion in last section, 𝑑𝑑𝜅𝜅𝐺𝐺𝑎𝑎𝐺𝐺−1 is an indication of the quality of the 1st GaN layer. Except the sample of 9-nm-thick AlN
Fig. 5.10 Schematic diagram of XRD rocking curve measurement for the planes of (a) (0002), (b) (10-12) and (10-10) of GaN.
130
interlayer, 𝑑𝑑𝜅𝜅𝐺𝐺𝑎𝑎𝐺𝐺−1 and then the quality of the first GaN layer was on the same level. This proved that FWHM of (0002) included the information from the 2nd GaN layer but was dominated by the quality of the 1st GaN layer.
However, the results of the planes of (10-12) and (10-10) differed. Their tendency was more evident that the quality of the 2nd GaN became higher as the thickness of AlN IL increased from 4.5 nm to 22 nm, and after that the quality degraded. FWHM of (10-10) plane included the quality information from only the 2nd GaN. Nevertheless, it was influenced by the 1st GaN and showed a strong coupling or combination effect of the 1st GaN layer and an AlN IL on the quality of the 2nd GaN. FWHM of (10-10) plane also had the same tendency with that of 𝑑𝑑𝜅𝜅𝐺𝐺𝑎𝑎𝐺𝐺−1 partially, for example as a function of AlN IL thickness, the improvement from 4.5 nm to 9 nm, from 13 nm to 22 nm and the degradation from 22 nm to 45 nm. The quality improvement of the plane of (10-10) was determined by the dominant influence of the 1st GaN for the samples with AlN interlayer thickness increased from 4.5 nm to 9 nm. Thin AlN interlayer doesn’t contribute to the quality enhancement or even introduce new dislocation, as shown in Fig. 5.11. The difference emerged when the thickness was increased from 9 nm to 13 nm. The quality of the 1st GaN degraded while that of the 2nd GaN was almost constant. Here indicated the quality improvement of the 2nd GaN brought by the favorable effects of 13-nm-thick AlN IL which compensated the degradation caused by the lower quality of the corresponding 1st GaN. AlN IL showed dominant influence on the compressive stress introduction in the 2nd GaN, as shown by the plot of 𝑑𝑑𝜅𝜅𝐺𝐺𝑎𝑎𝐺𝐺−2 in Fig. 5.9b which held different tendency with that of XRD results.
The best quality of 2nd GaN turned up when the thickness of AlN IL was raised to 22 nm, in spite of lower 𝑑𝑑𝜅𝜅𝐺𝐺𝑎𝑎𝐺𝐺−1 than the sample with 9-nm-thick AlN IL. As AlN IL thickness became larger to 45 nm, quality of the 2nd GaN started degrading. On the contrary, 𝑑𝑑𝜅𝜅𝐺𝐺𝑎𝑎𝐺𝐺−2 dropped a lot when the thickness of AlN IL was thickened from 13 nm to 22 nm and 45 nm.
Fig. 5.11 TEM image of the cross section of multi-condition sample of AlN-IL thickness study.
1 1 1
1 3 1
3 3
2 2
2
3
2 3 2
2 v 2
AlN AlN AlN AlN AlN GaN
GaN GaN GaN GaN
GaN
131
The overall conclusion is that thicker AlN IL leads to higher quality of the top GaN, while there a proper thickness around 22 nm exists. Nonetheless, thicker AlN IL brings better quality but low compressive stress (curvature). In terms of compressive stress in GaN, the optimal AlN IL thickness exists in the range from 9 nm to 13 nm. Thicker AlN IL is of higher quality and then higher quality of overlying GaN. Based on the strain relaxation theory in chapter 3, higher quality is favorable to keep strain in epilayer and sustain larger compressive strain in GaN. There is some other mechanism has counteracted the merit brought by higher crystal quality which can cause more compressive stress in GaN, for example cracking in thick AlN ILs as discussed in chapter 3. Therefore, from the XRD rocking curve measurement of single-AlN-IL samples, in terms of the quality of 2nd GaN, due to the influence from the 1st GaN, it is very hard to conclude that what thickness is the best for AlN interlayer. We can only say that 13 nm was better than 9 nm. The thickness which is favorable to yield higher GaN quality might be not proper to induce larger compressive stress in overlying GaN.
More direct information can be accessed by TEM observations. For convenience, only the multi-condition samples were characterized by TEM to observe the dislocation behavior in GaN on Si with multiple AlN ILs. Fig. 5.11 is the cross section image of the sample for AlN-IL thickness investigation. The dark lines are threading dislocations. From the TEM image, we can see the interaction between dislocations and AlN ILs, which can be classified into 3 types as marked in Fig. 5.11. Interaction type 1 is the threading of dislocations through GaN and AlN ILs from bottom to the up. Type 2 is the origination of new dislocations from AlN ILs at the interfaces of GaN/AlN, which is unfavorable. The first AlN IL with thickness of 4.5 nm generated the most new dislocations and after that as the thickness increased, the newly generated dislocations became less and less. This is consistent with the AFM results in Fig. 3.23 in chapter 3. As the thickness increased, the sizes of AlN grains became larger with higher quality and then the dislocation source density at GaN/AlN interface decreased and
Fig. 5.12 Interfaces of AlN ILs with varied thickness.
132
lead to less new dislocations. Interaction type 3 is blocking of the propagation of threading dislocations, which is advantageous to the reduction of dislocation density. The dislocation blocking effect became stronger and stronger as the thickness of AlN ILs increases.
Combining interaction type 2 and 3, thicker AlN IL is beneficial to reducing the dislocation density while this is not the case for introducing higher compressive stress in GaN layers. On the other hand, in AlN ILs with thickness of 22 nm and 45 nm, some V-pits and cracks have been observed and circled in Fig. 5.11, which have been observed also in AFM scanning of AlN-IL surfaces in Fig. 3.23. This can prove that the fine crackles in Fig. 3.23 were generated during growth but not cooling after growth. In the discussion of chapter 3, these crackles are responsible for the reduction of compressive stress in their overlying GaN.
More enlarged images at the interfaces can be seen in Fig. 5.12. For thin AlN ILs with thickness less than 13 nm, their interfaces are hazy and not abrupt. In most locations it is hard to tell where the interface is. When the thickness was 4.5 nm, it was too thin to form a continuous AlN film due to 3D growth. It could not separate the GaN layers sufficiently and then GaN could not be strained intensively. As shown by the AFM images in Fig. 3.23, AlN IL with the smallest thickness of 4.5 nm has the smallest grains and highest grain density as well as grain boundary density. Then at the grain boundaries, the most dislocations were generated. As the thickness increased to mediate thickness of about 9 nm, grains grew bigger and started to coalesce and the roughness also increased. After the thickness increased to 22 nm and higher, the coalescence has completed and the interfaces were more smooth and abrupt. There was no grain boundary to act as the source of dislocations. This is why thinnest AlN IL led to highest dislocation density in GaN and it decreased with the thickness of AlN IL. So, larger thickness is good for the improvement of GaN quality. However, it might not favorable for inducing higher compressive stress in GaN. Thin AlN ILs with thickness less than 13 nm are continuous 3D grains and are more relaxed. When the thickness increased to more than 22 nm, cracking occurred, as shown in AFM image of Fig. 3.23 as well as TEM image of Fig. 5. 12. After cracking, the AlN IL is not a whole piece of continuous film anymore and it could not sustain the tensile stress in itself, as indicated in Fig. 3. 18. As the AFM result of Fig. 3.23 showed, the crack density was high. With large amount of crack, although the tensile stress in AlN IL was relaxed, it could not induce much shear stress in
Fig. 5.13 (a) FWHM of XRD rocking curve of GaN samples with a single-layer AlN IL for temperature study; (b) curvature increase to the compressive side in the 1st GaN and the first 200
nm in the 2nd GaN layer.
133
lateral direction in GaN as AlN IL consisted of separated domains.
The same study has also been carried out as a series of growth temperature of AlN ILs. The results of FWHM of XRD rocking curve measurement for single-AlN-IL samples in Fig.
5.13a showed the same tendency as that of 𝑑𝑑𝜅𝜅𝐺𝐺𝑎𝑎𝐺𝐺−1for all the measured planes. This indicated that the influence of the 1st GaN was more dominant than that of AlN ILs or the growth temperature of AlN IL was not very influential on the quality of GaN above it. It is hard to interpret properly because the quality and morphology of AlN grown at low and high temperature was substantially different, as shown by the surface morphology in Fig. 3.27.
Compared with growth temperature, the influence on overlying GaN of AlN-IL thickness is more effective. As the thickness of AlN ILs was fixed to be around 8 nm which was sufficiently thin, they didn’t coalesce completely. The density of grain boundary which can be the source sites of dislocations were on the same level regardless of growth temperature.
This accounts for the weak dependence of GaN quality on the growth temperature of AlN ILs.
On the other hand, 𝑑𝑑𝜅𝜅𝐺𝐺𝑎𝑎𝐺𝐺−2 was very sensitive to the growth temperature of AlN IL, which the relative optimal value was around 900 ℃. In the TEM image of the cross section of the multi-condition sample (Fig. 5.14), the density of newly generated dislocation at the upper interface of AlN ILs increased slightly as the growth temperature increased from bottom to up while with no substantial difference. 1200-℃ grown IL caused the least dislocation. Besides the highest growth temperature and better quality, the thickness of it was also larger than others which may be due to the incorporation of gallium at high temperature, as shown in Fig. 5.15, and larger thickness causes less dislocation. Most of the new dislocations were blocked by the following AlN IL. After 5 layers of AlN IL, the dislocation
Fig. 5.14 TEM image of the cross section of multi-condition sample of AlN-IL temperature study. AlN ILs are marked by arrows and their growth temperature (bottom to top) are 1200 ℃,
1050 ℃, 900 ℃, 750 ℃ and 600 ℃ respectively.
134
density in GaN increased. The weak dependence of the density of newly generated dislocation may be due to the small thickness of about 8 nm of them. They did not coalesce well and the grain density may on the same level. The interfaces were more abrupt and clear at higher growth temperature, which is an indication of quality improvement of AlN ILs.
However, this also verified that high abruptness or smoothness is not the most impacting factor to introducing high compressive stress in GaN, compared with lattice constant difference.
To summarize this section, because the tendency of FWHM of XRD rocking curve measurement of the 2nd GaN is similar to that of 𝑑𝑑𝜅𝜅𝐺𝐺𝑎𝑎𝐺𝐺−1 but different from that of 𝑑𝑑𝜅𝜅𝐺𝐺𝑎𝑎𝐺𝐺−2, the effects of AlN ILs on the quality of overlying GaN is not as strong as that on the compressive stress introduction. The quality of upper GaN layers are determined by the underlying GaN layers dominantly if the thickness of AlN IL is small (< 13 nm). Certainly, AlN could introduce huge amount of new dislocations at the upper interface of AlN ILs if the growth conditions of it are not proper. AlN ILs can also block the propagation of threading dislocations.