Figure shows the 15-K PL spectra of both samples. Adding N induced a redshift of 80 meV, but it also significantly increased the full width at half maximum (FWHM) and reduced the integrated intensity by a factor of 12. Although the large redshift can be partially explained by the reduced QD-barrier conduction band offset in the presence of N [7
], nitrogen should also decrease the SRL redshift effect, the strong differences in the PL lineshape and intensity point to the presence of N-induced structural changes. In order to elucidate the possible reasons of this behavior, the samples were analyzed by transmission electron microscopy techniques to determine the strain fields and compositional distribution.
15-K PL spectra of the two InAs/GaAs QD layers capped with GaAsSb and GaAsSbN. The nominal Sb content is the same in both samples. The nominal N content in S-SbN is 2%.
From CTEM analysis, no extended defects such as dislocations or planar defects were found in either of the samples (<108
), denoting that the degradation of the PL efficiency by the addition of N should not be attributed to these kinds of defects. In Figure a,b, low-magnification HAADF-STEM images show representative views of samples S-Sb and S-SbN, respectively. In these Z-contrast images, regions with higher content of heavy elements appear brighter, thus making possible a clear distinction between the position and shape of the QDs and the CL surrounded by GaAs. After analyzing 75 QDs, the average QD base and height were measured as 13
2 nm and 3.4
0.6 nm in both samples, and no appreciable differences in the QD size are observed within the statistical error. The addition of N to the GaAsSb CL does not seem to influence the structure of the QD itself. However, this was not exactly the case for the CL. It should be noted that the CL in the region between QDs was thicker in the S-Sb sample (8.7 nm) than in the S-SbN (7.2 nm). Furthermore, sample S-Sb shows a planar top interface of the CL even when it wraps up the QD, that is to say, the CL over the QDs is thinner than in between QDs. A strong Sb segregation in the growth direction was observed in areas between QDs. By contrast, the CL of the S-SbN sample is adapted to the wetting layer QD surface, trying to keep the thickness constant, and therefore causing a growth front undulation when it covers the QDs. This is the first clue that the combination of N and Sb in the capping layer does not only lead to a band offset and a strain modification effect but also has a strong effect on the morphology of the structure.
HAADF images of samples (a) S-Sb and (b) S-SbN acquired at low magnification. Dark solid arrows mark the QDs position, and dashed arrows in (b) show the GaAsSbN lobules formed over the QDs.
In order to evaluate the strain in the CL, HRTEM images were acquired on the  zone axis far from any QD (Figure a,b) and processed to obtain strain maps using GPA [13
]. From the measurement of εzz
and assuming the compliance with Vegard's law at these contents, the Sb composition in the CL of sample S-Sb was estimated. Thus, in the CL regions between QDs for sample S-Sb (Figure c), the average strain measured, εzz
0.2%, corresponds with a Sb content of 10.3
0.4%, close to the designed value of 12%. In contrast, a reduced εzz
0.1% was observed in comparable areas of the S-SbN sample (Figure d). This large reduction needs to be attributed to N incorporation, which rises to 2.6
0.4% (in reasonable agreement with the nominal composition of 2%). In addition, as is observed in Figure d, the strain distribution in the CL of the S-SbN sample becomes more inhomogeneous, which points to a composition modulation with the presence of Sb-rich and Sb-poor regions in the range of a few nanometers. Certainly, the addition of N enhances even more the large miscibility gap of GaAsSb [15
] and gives rise to strong composition fluctuations. The observed clusters can act as traps for carrier, reducing the injection efficiency in the InAs QDs. The presence of the N-induced non-radiative centers (characteristic in N-diluted alloys) [16
] together with the higher compositional inhomogeneities is a possible explanation for the degradation of the PL.
HRTEM images and strain maps of CLs. The images were acquired along the  zone axis of the CLs in the samples (a) S-Sb and (b) S-SbN, and (c,d) their strain maps, respectively. Color scales of the strain are shown on the right.
However, the situation changes in and around the QDs. Figure shows the HRTEM and strain maps of QDs from both samples. Firstly, the average strain εzz
evaluated within the core of up to five QDs of the S-Sb sample was 3.8
0.2%, which is very similar to the average strain of 3.5
0.2% measured in several QDs in S-SbN. Secondly, the average strain in the CL region above the QDs was also similar in both cases, with the εzz
0.2% and 1.5
0.2% for the S-Sb and S-SbN samples, respectively. Important conclusions can be extracted from the strain data. The marked strain release in the CL between QDs when adding N (εzz
lowers from 1.5% to 0.6% as mentioned in the previous paragraph) should entail a decrease in εzz
inside the QD due to the reduced SRL effect [17
]. This did not happen, and the explanation of the observed result implies that the CL must undergo clear changes in the surroundings of the QDs. Assuming a homogenous distribution of N over the CL in S-SbN, the measured strain implies that the Sb content in that region must be approximately 17%, which indicates a strong Sb migration towards the top of the QDs in the S-SbN sample. This assumption of a homogeneous N distribution is in agreement with the results of Ciatto et al. [18
], who found that the Sb local environment is essentially random and that no significant preferential Sb-N pairing occurs over random statistics. The higher Sb content on top of the QDs compensates the counteracting effect of N on the strain reduction effect of the CL, explaining the similar strain values also observed inside the QDs. It seems that the system rearranges itself through composition modulation in the CL in order to keep the total strain in the QD region constant and equal to what would likely be the most stable value.
HRTEM images and strain maps of QDs. The images were acquired along the  zone axis of the QDs in the samples (a) S-Sb and (b) S-SbN, and (c,d) their strain maps, respectively. Color scales of the strain are shown on the right.
To confirm our hypothesis about the chemical composition in and around the QDs, HRSTEM Z-contrast imaging was carried out for both samples (Figure a,d). Anion and cation sub-nets are resolved in these images, showing the characteristic ‘dumbbells’ found at the  zone axis orientation. Figure g shows a magnified dumbbell where the upper and lower columns are the anionic and cationic components, respectively. In order to estimate column-by-column the Sb and In contents, a method similar to the one used in [9
] was performed. The R
values corresponding to each sub-net column, Ri
, are represented in Figure with colored dots, where higher values (red dots) are associated with atomic columns with higher proportion of heavier elements with respect to the corresponding atomic columns in GaAs. It is assumed in this work that the minimum Ri
measured in each image corresponds to a pure Ga or As column, respectively. Taking all this into account, Figure b,e depicts the estimated In/Ga distribution of both samples. Red dots, related to In-rich atomic columns, perfectly draw the QD and wetting layer (WL) regions. No appreciable out-diffused In from the QD was evinced. Still, in the case of the anionic positions, the panorama is very different when comparing the two samples. The redder dots in Figure c,f are associated to anionic columns with a higher content of Sb. As it is shown in Figure c for the S-Sb sample, the Sb atoms are mainly segregated to the upper region of the CL, leading to a Sb-poor region just above the WL. This is in agreement with previous results from scanning tunneling microscopy [5
], which reported that a region depleted of both Sb and In is distinguishable at the WL-CL and QD-CL interfaces. This kind of behavior - the accumulation of Sb atoms on the growth-front surface - has been reported in many III-V systems [21
], and it is a consequence of the bond energy differences. Since the Ga-Sb bond is considerably weaker than the Ga-As one, there is a preference for Sb atoms to be expelled to the surface [23
]. Moreover, the Sb concentration in the CL of the S-Sb sample was not excessively altered by the presence of QDs. This could explain the lack of undulation on the growth front observed in the capping layer above the QD. Nevertheless, the Sb distribution intensely changes in the case of sample S-SbN (Figure f). In this case, the reddest dots were located on and around the QDs in a higher density than in sample S-Sb. The higher Sb content around the QDs agrees with the higher εzz
values determined in the strain maps on the QD with respect to the CL regions in between the QDs that pointed to a lateral migration of Sb towards the QDs.
Figure 5 HR Z-contrast images of QDs and their integrated intensity (compostition related) maps. The images were acquired along the  zone axis of samples (a) S-Sb and (d) S-SbN, and their respective maps of the R values (Ri) of (b,e) cations and (c,f) anions. (more ...)
Certainly, the presence of N (with Z
7) should give rise to a small decrease of the anionic R1
quotient. However, the number of anionic columns with R1
1 in S-SbN amply exceeds the observed ones in sample S-Sb. In this sense, statistical analysis in dilute nitrides of GaAsN had shown that the incorporation of N at low contents causes negligible changes in the brightness of the atomic columns but a strong increase in the valleys between them [24
]. This fact does not affect our measurements since the pixels selected to calculate the R
quotients just avoid the area between atomic columns [9
]. However, the impossibility to detect the position and content of N around the QD, together with the non-negligible influence of static atomic displacements in the Z-signal, disables any attempt to quantify the Sb concentration in sample S-SbN by this technique [25
All these results indicate that the addition of N to the GaAsSb capping layer increases the amount of Sb on top and around the QDs, inducing an undulation in the capping layer which tends to adapt to the morphology of the QD below it. Though this migration leads to an enrichment of Sb around the QDs, the strain-compensating effect of N gives rise to similar strain fields around and inside of the S-SbN QDs to those for the S-Sb ones. When the CL covers a QD, nitrogen fosters the Sb to accumulate on top of the partially relaxed InAs QDs since GaSb has a very similar lattice constant with InAs. On the other hand, as the composition threshold for the total restraint of GaAsSb/InAs QD dissolution in the capping process is around 11% to 14% of Sb [26
], we could assume that the dissolution process of InAs QDs during the capping growth is completely suppressed for both samples and that the In atoms are not being relocated from the top of the QDs to the QD base [1
]. This explains why no significant differences in the size and morphology of the QDs are seen in both samples. The addition of N greatly enhances the lateral Sb segregation, and this fact could even induce to a transition to a type II band alignment in the VB, which is expected for Sb contents of 14% to 17% [1
]. Further work is in progress to clarify this issue.