Thickness effect

AlN IL, there is a compromise between the benefits and disadvantages brought by the relaxation of it.

0 5 10 15 20 25 30 35 40 45 50 -200

-150 -100 -50 0 50 100 150 200 250 300 350 400 450 500 550 600 650 700

curvature (km-1 )

thickness (nm) dC_GaN-s

dC_AlN-s C_RT-s dC_GaN-m dC_AlN-m dC_GaN-m'

(dC_GaN-m+dC_GaN-m')/2 dC_AlN-m'

Fig. 3.22 Compressive curvature increase in GaN dCGaN and tensile curvature increase in AlN ILs dCAlN during thickness comparison of multi- (-m) and single-condition (-s) growth. “-m” represents the multi-condition sample with AlN IL thickness sequence from bottom to up of 4.5, 9, 12.5, 22.5 and 45 nm, and “-m’” means the sample with reversed AlN IL thickness sequence. CRT-s is the final

curvature after growth at room temperature for single-AlN-interlayer samples.

effects of AlN ILs grown under various conditions on the properties of overlying GaN layer, the quality of GaN under all AlN ILs should be exactly identical, which is very hard to achieve. The higher GaN quality leads to less negative influence on multi-condition AlN IL study. To check such influence, the multi-condition sample for AlN IL thickness investigation was grown with AlN IL thickness sequence from bottom to up of 4.5, 9, 12.5, 22.5 and 45 nm, and the reversed sequence, which marked as “-m” and “-m’” respectively. To eliminate the unfavorable influence of multi-condition structure, the average of dCGaN-m and dCGaN-m’ is more reliable and close to the result of single-condition series except the thickness of 9 nm. From Fig. 3.22, about the amount of compressive curvature introduction in GaN layers, in “-m” sample and single-condition series, the most effective AlN interlayer was of thickness of 9 nm. While in “-m’” sample, this was shown to be 12.5 nm. In spite of such difference, this doesn’t change the overall tendency that the optimal thickness for AlN interlayer which can introduce the most compressive stress may be in the range from 9 to 12.5 nm; growth sequence of AlN ILs doesn’t change the result significantly. About the tensile curvature increase in AlN interlayers, both multi-AlN-interlayer test and single-AlN-interlayer test showed that it increased as the thickness was doubled from 4.5 nm to 9 nm and after that it almost saturated.

When the thickness is too small, around 4.5 nm for example, AlN IL is strained mostly compared with other layers so its lattice constant gets closer to that of GaN. Smaller lattice

64

constant difference induces minimal compressive curvature increase in overlying GaN. When the thickness rises to the range of 9 to 12.5 nm, it relaxes more that it strains the overlying GaN layer more compressively. Compressive curvature increase in GaN dropped drastically from the thickness of 12.5 nm. In Fig. 3.18 and Fig. 3.21, the tensile curvature increase at AlN interlayers saturated. Especially at interlayers with thickness of 22.5 and 45 nm, there are saturation points around thickness of 11 nm.

It is supposed that cracking occurred around the saturation point. This has been verified by AFM scanning of the surface morphology of AlN ILs (Fig. 3.23c). Roughness of root mean square (RMS) was also marked in the pictures. For 4.5-nm-thick AlN IL, because it was too thin, only about 10 monolayers, the crystal grains were very small and the surface was relatively smooth. The growth of AlN ILs on GaN was not pure epitaxial layer-by-layer mode, but started with very small grain domains with size of about 10 nm. Because the amount of grains was very large and boundaries between them could act as defect sources, GaN grown on it would get relaxed quickly. When the thickness rose to 9 nm, coalescence of original fine grains started, it exhibited continuous but larger crystal domains with clear grain boundaries.

No cracks could be observed in these two samples. In Fig. 3.18, it can be figured out that in 22.5-nm- and 45-nm-thick AlN ILs, cracking occurred around the thickness of 11 nm. So, for the 12.5-nm-thick AlN IL, cracking has already started. This may be why dCGaN starts to decrease from AlN IL thickness of 12.5 nm. When the thickness was increased to 22.5 nm, coalescence of the small and fine grains has completed and the surface was quiet smooth.

However, it showed considerable cracks. In Fig. 3.23c, there are two kinds of cracks. Some crack originated during cool-down due to the huge tensile stress caused by thermal mismatch between nitrides and Si substrate, like the wide and straight crack lines marked as type-I by yellow dash circle. Such crack was caused by the cracking of underlying GaN, but not the top AlN IL itself. The other type of crack was very fine and with irregular shape, as marked as type-II by red dash line circles. Type-II crack arose during the growth of AlN IL itself but not after growth. As marked in Fig. 3.18, for AlN ILs with thickness of 22.5 nm and 45 nm, there was a sudden drop of the slope of curvature curve, namely the drop of tensile stress in AlN ILs. This can be observed also in single-condition growth in Fig. 3.21. Sudden stress drop can be a proof of the occurrence of cracking to release the strain energy in AlN ILs, because only relaxation by cracking is not continuous. This was consistent with the results from other researches that cracking of AlN epitaxial films on GaN occurs during growth but not post-growth cooling in samples grown by MOCVD [59]. During the initial post-growth stage, due to relatively low temperature and reduced Al ad-atom mobility, there are some dynamical platelets and the thin AlN IL is formed by the coalescence of them. The regions of such coalescence have been proposed as sites introducing misfit dislocation. It also has been postulated that deep grooves between grains can evolve into cracks with increasing AlN IL thickness [52]. After cracking, AlN ILs could not sustain enough shear stress to strain the overlying GaN layers on them highly compressively. Cracking in AlN ILs also indicated the difference between the growth of AlN on GaN and Si. Because lattice constant of AlN 𝑎𝑎𝐴𝐴𝑡𝑡𝐺𝐺 is much closer to 𝑎𝑎_{𝐺𝐺𝑎𝑎𝐺𝐺} than 𝑎𝑎_{𝑆𝑆𝑆𝑆}, growth of AlN on GaN at mediate and high temperature is more epitaxial by 2D layer-by-layer growth or the mixture mode of 2D and 3D, but it is completely 3D growth on Si. Then the strain energy releasing mechanism between them also

65

differs significantly. As shown in curvature curves, the stress of AlN buffer layer grown on Si held a constant stress without any sudden change up to 110 nm. Cracking in thick AlN ILs make them more relaxed, but also reduced their ability to strain the overlying GaN. Due to crack net, it became not as continuous as thin AlN ILs and made the overlying GaN get relaxed more rapidly.

On the other hand, the theoretical critical thickness for AlN on GaN is 7.56 nm [58].

Experimentally, in MBE grown samples, no line defects could be observed at AlN ILs on GaN with thickness below 6 nm [52], although the detailed growth process may differs slightly between MOCVD and MBE grown materials. Due to the lower misfit dislocation density at AlN ILs, threading dislocation propagated into following GaN might be less thus the relaxation of it might be also slower. The small tensile curvature close to 0, namely the convex bowing, is our final target.

For the single-AlN-IL series, the final tensile curvature at room temperature CRT-s was also extracted in Fig. 3.22. The one with 9-nm-thick AlN IL achieved smallest tensile curvature increase and yielded crack-free GaN. The tendency was that the samples with higher dCGaN

hold lower CRT. CRT was not determined only by AlN ILs, but also by the AlN buffer.

Although the growth conditions for AlN buffer layer and the 1^st GaN was designed to be the same, but due to some unfixable reasons from Si wafer and the MOVPE system, it differed slightly from growth to growth. Such systematic condition fluctuation led to slight variation of the stress in AlN buffers and the 1^st GaN. Such fluctuation and variation doesn’t affect the Fig. 3.23 AFM images of the surface of AlN IL with thickness of (a) 4.5 nm, (b) 9 nm and (c) 22.5

nm, grown under V/III of 1505 and at temperature of 900 ℃.

RMS = 0.238 nm RMS = 0.646 nm

RMS = 0.177 nm

(a) thickness = 4.5 nm (b) thickness = 9 nm

(c) thickness = 22.5 nm

type-I type-II type-II

66

tendency of the results. The sum of compressive curvature increment of the 1^st and the 2^nd GaN layer determines the final curvature CRT. The stress fluctuation in AlN buffer and 1^st GaN doesn’t change the effect tendency of AlN ILs. The tensile stress during cool-down 𝜎𝜎^t arose from the all two GaN layers and so did the compressive stress 𝜎𝜎^c. If the AlN IL is not proper, 𝜎𝜎^c cannot compensate tensile thermal mismatch stress 𝜎𝜎𝑡𝑡 and C_RT-s will be much larger than net compressive curvature increase in the 2^nd GaN. If AlN IL is optimal and 𝜎𝜎^c is large enough to compensate 𝜎𝜎^t, then CRT-s would be around zero.

By comparing the results of multi-condition and single-condition samples, although strictly speaking, in multi-condition sample it is unavoidable that the preceding layers influence the following ones, it still leads to the same tendency as that of single-condition samples. It means that in multi-condition sample, the influence from the bottom layers could not change the properties of top AlN ILs substantially and this method is reliable.