Effect of thermal annealing on the optical and structural properties of (311)B and (001) GaAsBi/GaAs single quantum wells grown by MBE

The effect of Furnace Annealing (FA) and Rapid Thermal annealing (RTA) on the structural and optical properties of GaAs 1-x Bi x /GaAs single quantum wells (SQW) grown on (001) and (311)B substrates by molecular beam epitaxy (MBE) was investigated. The structural properties were investigated by high-resolution X-ray diffraction (HR-XRD) and Transmission Electron Microscopy (TEM). achieved. The enhancement of PL intensity in annealed samples was interpreted in terms of PL activation energies, with a reduction of the alloy disorder and an increase of the Bi cluster.


Introduction
Recently, dilute bismide alloys grown on GaAs have attracted much attention due to their physical properties and potential applications in photonic devices and long-wavelength optoelectronics [1,2]. The incorporation of a few atomic percent of Bi into GaAs (substitution of As atoms with Bi atoms) strongly affects the properties of GaAs. For example, a giant band gap bowing effect was observed leading to a strong reduction of the band gap energy by 88meV per % Bi [1]. In addition, the increase of the spin-orbit (SO) splitting energy with Bi concentration [3,4] makes GaAsBi-based structures suitable for spintronic applications [3,51].
Furthermore, due to the relatively large size of bismuth atoms, a strong deformation of the GaAs lattice occurs producing an increase of the carrier-phonon coupling [5]. However, incorporation of Bi into GaAs causes a strong effect on the carrier recombination processes due to an increase of the density of the localized states [6][7][8]. As a result of this effect, an extension of the band-edge toward lower energies, the so-called band-tail states is created [9].
The localized states in GaAsBi are attributed to the variations of the alloy composition together with clustering of Bi [10]. Mostly, the effects of localization on the system lead to considerable changes in carrier dynamics, therefore affecting the performance of optoelectronic devices.
In order to substitute As atoms with Bi atoms in GaAs, GaAsBi alloys are grown at low temperatures (TG = 200 °C -400 °C) because Bi atoms are much bigger than As atoms.
However, adopting this unconventional growth regime to incorporate Bi into the host GaAs lattice creates structural defects which behave as nonradiative centers such as native point defects and clusters [25] affecting their optical efficiency. It is well-known that post-growth thermal treatment improves the optical properties of semiconductor structures by dissociating complex defects and annihilating some other non-radiative recombination centers in order to reduce their concentrations. At the same time, the incorporated Bi atoms often segregate and form Bi-rich clusters with various structures sizes in the range of 5-20 nm during thermal annealing process, depending on the Bi content and on the annealing conditions [50].
The aim of this work is to investigate the effect of two thermal annealing processes on the optical and structural properties of GaAsBi/GaAs QWs. Previous studies have shown an improvement of the PL peak intensity in lattice mismatched dilute nitride GaAsN/GaAs materials by using RTA [26,27]. The influence of thermal annealing was also investigated in dilute bismide III-V materials both in bulk and QW structures grown by MBE on (001) GaAs substrates [18,[28][29][30]52] using RTA. Most of the reported work on annealing was carried out at very high temperatures above the growth temperature of the samples. However, J.F. Rodrigo et al [16] reported that (311)B GaAsBi layers grown at 350 o C and annealed at 200 o C for 3 hours resulted in a decrease of the structural defects in the epilayer with a huge increase of the photoluminescence signal. It is worth pointing out that the influence of annealing procedures on the optical and structural properties of MBE grown GaAsBi QWs have not been reported as a function of substrate orientation.
In this paper, we study the effect of thermal annealing temperatures, annealing times and thermal annealing techniques on the optical and the structural properties of a GaAs(1-x)Bix/GaAs single QW grown at 320 °C by MBE on semi-insulating (001) and (311)B GaAs substrates with a nominal Bi content x = 3%. The approach adopted for the annealing process was the use of low and high temperatures for long and short annealing times, respectively. The best material quality was obtained by annealing at 300 °C and 2 minutes using RTA. The electronic bandprofiles are calculated using the vertical Bi profile which were determined by simulating the high-resolution X-ray diffraction (HR-XRD) data using dynamical scattering theory.
Correlations between HR-XRD and TEM results will be discussed.

Method and Experimental details
In order to substitute As atoms with Bi atoms in GaAs, GaAsBi alloys are usually grown at low temperatures between 200 °C -400 °C because Bi atoms are much bigger than As atoms.
This means that the incorporation of Bi into the GaAs lattice is enhanced at growth temperatures lower than 400 °C. In this work the growth of GaAsBi single quantum well (SQW) samples started with the deposition of a 370 nm GaAs buffer layer. The first 300 nm was grown at high As overpressure and at a temperature of 580 °C which the optimum growth temperature of GaAs. The growth temperature of GaAs was then lowered from 580 °C to 320 °C for the last 70 nm before the growth of GaAsBi QW in order to reach the growth temperature for optimum incorporation of Bi into GaAs. This was followed by a 10 nm thick GaAsBi SQW layer and a 50 nm GaAs cap grown at the same temperature (see Figure 1). The growth rate was ~ 0.4μm/h for all layers. For efficient Bi incorporation, the atomic Ga/As flux ratio was adjusted close to the stoichiometric value before the growth of the QW. All our samples were grown using As4 species. The effect of Bi localization due to alloy fluctuations and clustering on the optical properties is investigated for dilute bismuth GaAsBi QWs using PL measurements as a function of temperature and excitation power. A 532 nm green laser is used as an excitation power (PEXC) source ranging from 0.5 mW to 33 mW. All samples were checked for uniformity before annealing them using PL measurements. The PL energy and PL intensity were similar at different positions in the sample. The size of the original samples were 8×8 mm 2 squares. These were cleaved into four pieces for different annealing treatments.
For this purpose, we have applied RTA at different temperatures and durations (500 °C for 1min, 300 °C for 2mins and 700 o C for 30 sec) and FA in a tubular furnace at 200 °C for 3h.
All heat treatments were carried out under N2 gas conditions. It is worth pointing out that the samples were placed between two protective GaAs substrates in order to avoid As evaporation from the top GaAs cap layer and formation of Ga droplets.

GaAs (001) or (311)B Semi-Insulating substrate
The PL measurements were performed at temperatures between 10 and 300 K using a highresolution optical spectrometer, an InGaAs dual detector, and a computer-controlled lock-in amplifier.
To gain an insight into the evolution of the structural properties of the samples upon annealing, a Bruker D8 Discover HR-XRD diffractometer system was used. An asymmetric 2-bounce Ge 220 channel-cut monochromator was used to select the Cu Kα1 line at λ= 1.540 56 Å and to collimate the incident X-ray beam. Symmetrical 2θ-ω scans were performed from which the layer thicknesses, vertical strains and vertical Bi concentration profiles were determined.
Cross-sectional TEM specimens were prepared using a Focused Ion Beam (FIB), model Quanta

3.1.Structural Properties
HR-XRD diffraction curves (2θ-ω scans) are shown in Figure 2. The sharp peaks at about 2θ = 66.05° and 53.73° are the GaAs (004) and (311) substrate Bragg reflections, respectively. In both cases, we observe a weak shoulder at low angle side of the strong substrate Bragg reflections which can be attributed to the compressively strained GaAsBi QWs. However, the well-defined intensity fringes around the sharp substrate Bragg peaks clearly prove a smooth interface between the epitaxial layer and the substrate.
In order to evaluate the vertical Bi concentration profiles, the experimental diffraction curves were simulated using dynamical scattering theory. In these simulations we have determined the Bi concentration through a linear interpolation between the lattice parameters of GaAs (a = 5.653 Å) and of a hypothetical zincblende GaBi (a = 6.33 Å) [53]. Linear elasticity theory has been applied. Owing to the low Bi content we have used, as a good approximation, throughout the elastic coefficients of GaAs [54].
The best fits, which are in excellent agreement with the experimental data, are shown as solid lines in Figure 2. It is worth noting that the obtained Bi concentration profiles shown in Figure   2 contain a number of fitting parameters making the fitting routine presumably not reliable.
For that reason we have performed many extended simulations in which we tried to further refine the simulations. In all of these simulations we obtained very similar Bi concentration profiles. The simulations show that the maximum Bi concentrations in the QWs are 2.15% and 2.10% for the samples grown on (001) and (311)B substrates, respectively. However, as seen in the inset of Figure 2, the Bi content spreads out from the GaAsBi QW to the GaAs cap.
Therefore, for both samples, the total thickness of the Bi-containing layer is much larger than the nominal GaAsBi QW thickness. Since the GaAs cap was grown at the same temperature as the GaAsBi QW, it can be assumed that the Bi atoms on the surface continue to incorporate after the nominal QW growth. The thickness of the Bi containing layers is almost the same for both samples; however, the amount of Bi in the extended/top layer region in the sample grown on (311)B GaAs is higher than that in the sample grown on (001) GaAs. The HR-XRD studies described above show no measurable impact of annealing on the vertical Bi-concentration profile. In the X-ray measurements, however, a lateral averaging over large sample areas of typically a few mm 2 is carried out. Therefore, possible Bi clustering may not be directly visible in the X-ray data. In order to obtain local information on atomic scale highresolution. For that reason high-resolution TEM measurements were performed in order to obtain local structural information on the atomic scale. GaAs substrates, respectively. The area represented by the dashed rectangular shown in Figure   4(b) provides a strong indication of Bi clustering of about 10 nm in size, probably due to the Bi-segregation during annealing. This characteristic has been observed also in the work of Baladés et al [31]. In this paper, the authors reported the Bi distribution in epitaxial GaAsBi-

on-GaAs by using High Angle Annular Dark-Field Scanning Transmission Electron
Microscopy (HAADF-STEM) technique. They reported a non-homogeneous Bi distribution at the interfaces as well as within the GaAsBi layer. In that case the clusters may lead to a partially relaxed GaAsBi/GaAs system around them. However, in our samples we identify in Figure 4 (b) a structural disorder due to annealing that can be attributed to the Bi-segregation followed by a structural relaxation in the QW region. This evidence is compelling if we compare Figures4 (c) with 4 (d): the distributions of dark regions are more disperse in (c), but after annealing a major separation of these dark regions is observed that correspond to Bisegregation. As can be seen from these HRTEM images, it is not possible to identify structural defects such as extended defects. Finally, the results indicated that the Bi clusters segregation is more pronounced in (001) samples after annealing [ Figure 4(b)]. This event can be associated with Bi clusters reported by Beladés et al [31]. Nevertheless, corroborating with HR-XRD, the GaAs buffer layer is not visible in the TEM images indicating, therefore, the existence of the same vertical lattice parameter than that from GaAs substrate. sample and confirmed by HR-XRD results, this feature does not explain the energy difference observed in the PL peak. However, it is difficult to compare straight the PL of (001) sample with (311)B because the latter has more spatial variations in Bi incorporation. This may be due to more clustering of Bi and carrier localization at Bi pairs and clusters [32,33]. This feature seems to affect the PL peak intensity which is 9 times higher in (001) sample than in (311)B sample, probably due to a smaller amount of non-radiative defects in (001) sample [11]. It is important to point out that the quantum confinement could affect the PL peak position but less than Bi content, since the nominal QW width is around 10 nm. Furthermore, the Bi segregation from the GaAsBi QW to the GaAs cap could lead to a larger QW width and therefore there is a reduction of quantum confinement. In addition, the reduction of the band gap energy seems be higher in (001) than in (311)B sample, resulting in a less energetic PL peak. Therefore, it is very difficult to separate all effects that contribute to the observed difference in the PL peak in this case. The excitation power dependence of the PL spectra has been investigated in detail. Figure 6 shows the normalized PL spectra for the QW samples at 10K as a function of the excitation To analyse the emission from the (001) and (311)B samples, the integrated PL intensity was plotted as a function of excitation power as shown in Figure 6 for (d) (001) and (e) (311)B

Optical Properties
samples. The solid and dashed lines represent the fitting curves using the power law = P [34]. IPL is the integrated PL intensity and β and k are fitting parameters. Two k values (k1 and k2) for each sample were obtained, indicating two different processes for carrier radiative recombination as a function of PEXC.
As shown in Figure 6 (c), the PL peak has the lowest energy at low PEXC. Under this low power excitation, the PL emission is controlled by radiative recombination of carriers that occupy localized states close to the valence band maximum. For both samples, the value of k1 is > 1 indicating that both free excitons and free carriers contribute the recombination process.
However, as PEXC increases to a high-power regime, the obtained value of k2 is ~1, lower than k1. This means that as the PEXC is increased, the recombination process, which is limited by nonradiative recombination centres, is dominated by free excitons. This effect could be due to the fact that the localized states are being fully occupied by carriers and a strong evidence of this effect is demonstrated by a blueshift of the PL peak as PEXC is increased as shown in figure   6 (c) [17,35,36].
The temperature dependence of PL spectra of (001) and (311)B samples are shown in Figure   7. Both samples exhibit a redshift of the PL peak as the temperature is increased. This wellknown effect is due to the dependence of the bandgap energy with temperature. However, alloy disorder and potential fluctuations are the main cause of exciton localization. In this case, the temperature dependence of the PL peak position cannot be accounted for by using the standard Varshni's model [37]. The well-known 'S-shaped' behaviour observed in disordered semiconductors such as GaAsN and GaAsBi could possibly explain this effect [36, 38, -40].
It has been reported that the PL spectra are dominated by free excitons at high PEXC and temperature conditions [17,35,36]. As shown in Figure 7 (c-d), the PL peak red-shits and the PL FWHM gets larger as the temperature is increased. There is no clear evidence of S-shape behaviour in these samples. This could be due to the use of high PEXC that could make this effect less evident since the localized states are completely filled in this condition. Mazur et al also have shown that by increasing excitation intensities the S-shape becomes less noticeable, and for high power the S-shape vanishes completely [41]. Similar reported results on PL of GaAsBi QWs grown on (001) GaAs substrates [38,42] are in good agreement with ours. However, as shown in Figure 8 the PL intensity increases after annealing meaning that the optical quality is improved by both annealing procedures, except the case of (311)B sample which was RTA annealed at 500 o C for 1 min. The improvement in PL intensity can be explained in terms of decrease in the QW localized states density. Degradation of the QW interfaces with annealing and a decrease of the density of localized defects, due to bismide aggregates or alloy disorder, have been reported to occur during annealing by Mohmad et al. [43] and Mazzucato et al. [44].
As can be seen in Figure 8  PL emission was detected. This degradation of PL intensity is in accordance with what was observed in other studies [30,43], where the annealing treatment at temperatures higher than 700°C typically degrades the PL intensity with respect to the as-grown samples, since a very high annealing temperature leads to the activation of additional non-radiative centers. This could be explained by the fact that high annealing temperatures, well above the growth temperature, could lead to a re-distribution of the Bi atoms in the lattice or Bi out-diffusion from the well. The surface damage during the high temperature annealing process could also explain the disappearance of PL emission. As highlighted by Hajer et al., no room temperature PL emission is visible for the annealed samples at 750 o C and 800 o C [19] but they observed PL at 20K. On the other hand, appropriate thermal annealing is more efficient to improve the optical quality of alloys without affecting the PL peak wavelength [45].
To assess the samples quality and explain the nonradiative process for the samples grown on the different substrate planes, Arrhenius plots (integrated PL intensities (IPL) versus reciprocal temperature) are obtained for a PXEC=16mW as shown in Figure 9. Both as-grown and annealed (at 300 o C for 2 mins) (001) and (311)B GaAsBi QW samples are analysed. The behaviour is classified into thermally activated nonradiative recombination processes. The fits are derived by applying the formalism described in [46,47].
where I(T) and I(0) are the PL integrated intensity at temperature T and 0K, respectively. C1 and C2 are the non-radiative and radiative recombination probabilities, kB is the Boltzmann constant, E1 and E2 are the thermal activation energies. A plot of IPL versus 1/ kB T is shown in Figure 9 and the activation energies obtained by these fittings before and after RTA at 300 The drop of the activation energy observed for both annealed samples when compared with asgrown samples could be interpreted as (i) a reduction of the alloy disorder and (ii) an increment of the Bi cluster as discussed before. This, therefore, results in a reduction of non-radiative centers and an enhancement of the PL intensity. These findings seem to agree with the structural changes observed in our HR-TEM data, since the Bi clustering, probably due to the Bi-segregation, become more effective during the annealing process.

Conclusions
In conclusion, the effects of RTA and FA upon the HR-XRD, TEM and PL measurements from GaAsBi single quantum well with 3% Bi grown on (001) and (311)B GaAs substrates were investigated. The HR-XRD results reveal a very small difference between the as-grown and annealed samples and the Bi composition is not uniform in the epilayer and in the QW layer.
The HR-XRD spectra demonstrated that the Bi concentration inhomogeneity is more pronounced for (311)B than (001) samples. The HRTEM images show evidence of Bisegregation, which is more pronounced in (001) samples after annealing. The single QW shows a structural stability in terms of thickness for the optimum annealing temperature of 300 o C for 2 mins. RTA is found to be a more effective method to improve the PL intensity than FA, which could be due to a reduction of structural defects in the samples. The optimum annealing condition is determined to be 300°C for 2 mins for all studied samples regardless of the substrate orientations. The increase of PL intensity in annealed samples was interpreted in terms of PL activation energies, with a reduction of the alloy disorder and an increase of the Bi