1. Introduction

The growth of high quality III-V light emitting devices on Si is highly desirable, permitting the full integration of opto-electronics and Si electronics with applications including on-chip lasers for very high rate intra-chip and inter-server data distribution [1–3]. However, the optimum approach, direct III-V epitaxial growth on Si, is very challenging, due to the high dislocation densities formed at the Si/III-V interface; a result of the very different lattice constants and thermal expansion coefficients. These dislocations can propagate through the III-V epilayers and significantly degrade the optical quality of the active region [4]. Hence steps to reduce both the dislocation density and the impact of remaining dislocations on the optical quality of the III-V active region are critical if viable devices are to be obtained. Replacing quantum wells (QWs) with quantum dots (QDs) as the active region significantly reduces the sensitivity to defects because the QDs strongly localize carriers, reducing their migration to the non-radiative centres associated with the dislocations [5,6]. The density of dislocations reaching the III-V active region can be significantly reduced by placing dislocation filter layers (DFLs), between the Si/III-V interface and the active region [7–10]. The combination of QD active regions and DFLs have resulted in several reports demonstrating high output powers and high temperature operation of III-V lasers grown directly on Si substrates [11–14].

DFLs consist of strained-layer superlattices (SLSs), which inhibit dislocation propagation by bending them into the growth plane, allowing two dislocations to meet and annihilate. It has been proposed that the annihilation efficiency can be further enhanced by introducing elevated temperature steps (annealing) during growth; the increased kinetic energy of the dislocations increases their inplane movement. Practical devices will require a careful optimization of both the DFL design and annealing steps to obtain optimum device characteristics and minimize defect related lifetime degradation. Despite the importance of this optimization there have been no reports of the effects of different annealing strategies on the structural, optical and devices properties of III-V QD lasers grown on Si. Previous work has reported the effects of DFLs and annealing on the structural and optical properties of bulk GaAs layers grown on Si [15–18] and there is a report of an InGaAs/InGaAsP multiple quantum well laser grown on Si which includes a DFL but with no reference to annealing [19]. Because of their reduced sensitivity to defects and dislocations QDs are much more suited to the active region of lasers grown on Si and hence further optimisation of their optical and structural properties is very important.

In this paper we compare the effects of different annealing cycles on the properties of 1.3 μm emitting InAs/GaAs QD structures and lasers grown on Si substrates. All samples have identical DFLs but differ in the number of elevated temperature annealing steps applied during the growth. Comparison is made with a reference structure having an identical active region but grown on GaAs. The structural and optical properties and laser characteristics all exhibit a significant improvement as the number of annealing steps are increased. Our studies also allow the active region residual strain for structures grown on Si to be determined. This is shown to be controlled by the different thermal expansion coefficients of Si and GaAs. This strain produces only a small modification of the electronic states but will limit the thickness of III-V material that can be grown before relaxation occurs.

2. Experiments and results

The samples were grown on n-doped Si (100) substrates with 4° offcut to the [011] plane using a solid-source III-V molecular beam epitaxy system (MBE). Oxide desorption of the Si substrates was performed at 900 °C for 15 minutes, following which the substrates were cooled down to 370 °C for the growth of a GaAs nucleation layer comprising an optimized two-step growth scheme. For the test structures a 1000 nm GaAs buffer layer was deposited followed by 3 DFLs each separated by a 350 nm GaAs spacer layer. The DFLs consisted of a 5 period superlattice of alternating 10 nm In_0.18Ga_0.82As and 10 nm GaAs layers (5 DFLs were used for the laser structures). Following the final 350 nm of GaAs a 5 layer InAs/GaAs dot-in-a-well (DWELL) structure was grown, clad on both sides by a 100 nm Al_0.4Ga_0.6As layer grown at 610 °C and a 50 nm GaAs layer grown at 580 °C [20,21]. The samples were terminated with a 50 nm layer of GaAs and uncapped surface InAs QDs. The DWELLs consisted of 3 monolayers of InAs, which formed the QDs, with 2 nm of In_0.15Ga_0.85As layer below and 6 nm of In_0.15Ga_0.85As layer above, all grown at ~510 °C. Each DWELL layer was separated by 50 nm of undoped GaAs. This DWELL structure gives room temperature emission close to 1.3 μm. In the laser structures the AlGaAs thickness was increased to 1200 nm and 2 × 10¹⁸ cm⁻³ n dopant was added to the lower cladding and 2 × 10¹⁸ cm⁻³ p dopant was added to the upper cladding, with 1 × 10¹⁹ cm⁻³ p dopant in the capping layer. In addition, a control sample was grown on a GaAs substrate with a thinner GaAs buffer layer of 200 nm grown at 580 °C and no DFLs but with the same 5 layer QD DWELL active region and cladding layers.

Annealing was carried out in the MBE reactor with the growth paused and in an As rich environment, by ramping the substrate temperature to 600 °C for 6 minutes. Sample A had no annealing, sample B had a annealing step after each DFL (dashed lines) and sample C had an additional annealing step above the GaAs buffer region as well as the anneal above each DFL (dashed lines and dot/dash line), the positions of these are indicated in the TEM image of Fig. 1. This figure shows that the dislocations (dark features) propagate upwards from the Si/III-V interface and exhibit a reduction in density as they pass through the DFLs. After 3 DFLs their density is reduced below the sensitivity of the current images.

 

Fig. 1 Cross sectional TEM image of a PL sample, with a GaAs buffer layer grown on Si followed by 3 DFLs and an active region containg 5 DWELL QD layers. The dislocations can be seen originating at the Si-GaAs interface and progating up through the structure. The dashed lines indicate the positions at which the in situ annealing was performed.

Download Full Size | PDF

In order to quantify the effect of the annealing on the defect propagation TEM images, similar to Fig. 1 but for a much greater area, were recorded for each sample. Cross section TEM specimens parallel to {110} were prepared using standard protocols, each of which typically yielded 30–60 μm of electron transparent area along the Si/III-V interface. A series of images were taken using bright field g 220 diffraction conditions with the specimen tilted 5°–10° from the {110} zone axis, capturing the complete structure as shown in Fig. 1 over the entire electron transparent area (see [22] for further details). From these images it is possible to accurately determine the number of defects present at different levels in the structure thereby quantifying the efficacy of the DFLs and annealing process. In sample A, where no annealing was applied, the defect reduction in the active region relative to the initial number at the Si-GaAs interface was 95.6%. The addition of annealing in sample B increased this reduction to 98.9% and for sample C, where a further annealing step was added, the reduction was 99.9%. These results clearly demonstrate the benefits of including the in situ annealing during the growth of the structures.

Room temperature photoluminescence (PL) was measured for the test structures, to determine their optical quality. The spectra, obtained for 10 mW of 633nm excitation (32 Wcm²), are plotted Fig. 2. As expected the control sample has the brightest emission, consistent with no dislocations being present in this III-V only structure. This is followed by sample B and then sample C, which both have very similar peak intensities but with sample C having an integrated intensity ~1.2 times greater than that of sample B. Sample A, where no annealing was included, has the weakest emission, consistent with the highest dislocation density propagating through to the active region. The relative integrated intensities of the four samples are 1, 0.66, 0.54 and 0.26 for the control sample and samples C, B and A respectively. These values for the samples grown on Si are consistent with the TEM determined active region dislocation densities presented above. For reasons that remain unclear sample C has a much broader ground state (GS) linewidth (66 nm) than both sample B (39 nm), sample A (45 nm) and the control sample (44 nm). This appears to be a result of inter-sample variations in the growth of the QDs and not a result of the different annealing steps. Figure 2 also shows that the emission wavelengths of the Si grown structures are slightly shorter than that of the GaAs grown control structure (1314 nm, 1296 nm, 1277 nm and 1288 nm for the control structure and samples C, B and A respectively). This small blue shift in the emission of the Si grown structures may be a result of a slightly different strain state of the QDs as will be discussed below. The variation of 19 nm (≡ 14 meV) in the ground state emission of the three Si grown samples is attributed to small inter-run variations of the growth conditions. As the QDs are deposited after the annealing steps they will not be directly affected by the annealing.

 

Fig. 2 Room temperature PL spectra for the three structures grown on Si substrates and the control structure grown on a GaAs substrate.

Download Full Size | PDF

Photoluminescence excitation (PLE), excited with quasi-monochromatic light from a 150 W tungsten halogen lamp and dispersed by a monochromator, was used to study the band structure of the samples. Typical spectra for detection at the peak of the QD emission and for a sample temperature of 77K are shown in Fig. 3. As all three samples grown on Si substrates exhibited identical spectra within the experimental accuracy only one is shown for clarity. The resultant absorption-like spectra have been normalised to the intensity in the region below 800 nm. The position of the GaAs exciton is shifted between the GaAs and Si grown samples, occurring at 830 nm in the Si samples and 819 nm in the GaAs control sample; this is equivalent to a ~20 meV red shift in the Si grown samples. In the Si samples the exciton also appears to be split with a shoulder appearing to longer wavelengths. This shift and splitting is attributed to strain and can be explained by assuming that the GaAs is unstrained at the growth temperature but acquires an in-plane tensile strain due to the very different thermal expansion coefficients of GaAs and Si. Using the strain model described by Chandrasekhar and Pollak [23] with the material parameters from Vurgaftman et al [24] it is possible to calculate the tensile strain by fitting both the energy shift with respect to the control sample and the splitting of the exciton. A GaAs tensile strain of 0.24% at 77K is obtained. The same value is found for all three Si grown samples indicating that the annealing has no influence on this residue strain in the GaAs layers, which are grown after the application of the annealing steps. Interpolating to room temperature gives a strain of 0.17%, in agreement with the calculated value obtained from the known thermal expansion coefficients of Si and GaAs and the growth temperature of 580 °C. Although relatively small, this residual strain will limit the thickness of III-V material that can be grown before relaxation occurs. For GaAs and/or AlGaAs this limit is estimated to be ~5 μm, which is comparable to the total thickness of a laser structure. A larger critical thickness of ~20 μm is possible by using GaAs latticed matched GaInP, which has a thermal expansion coefficient closer to that of Si. The residual strain in the GaAs may also explain the slightly shorter emission wavelength of the samples grown on Si (as seen in Fig. 2) by slightly modifying the strain state of the QDs. No signal is seen due to the DFLs in the Si substrate samples, as PLE is recorded for detection of the QD emission, a DFL related feature would require initial absorption by the DFLs followed by subsequent carrier escape, transfer and capture by the QDs. Thermal carrier escape from the DFLs at 77K is unlikely and there will also be significant non-radiative recombination in the region of the DFLs due to the remaining dislocations.

 

Fig. 3 77K PLE spectra for 5 layer InGaAs DWELL QD samples grown on Si and GaAs substrates. The spectra have been normalised and shifted vertically for clarity

Download Full Size | PDF

Features between 850 and 950nm in the PLE spectra are attributed to transitions involving a combination of the WL and InGaAs layers in the DWELL with the light and heavy hole states split by a mixture of strain and confinement. The energies and relative intensities of these features are consistent across the three samples grown on Si but are blue shifted and significantly more intense in the GaAs-substrate sample. The reduced intensity of the WL/DWELL features in the PLE of the samples grown on Si is attributed to the presence of a residue density of dislocations in the active region. Carriers created in the WL/DWELL are able to move inplane, resulting in some migration to defects where they recombine non-radiatively before capture by the QDs can occur. The lowest energy feature attributed to the WL/DWELL (920~950nm) is red shifted by ~35meV in the samples grown on Si, again this is attributed to a residue strain in these samples. As thermal excitation of carriers from the dots to WL states has been proposed as a major loss mechanism contributing to the temperature performance of QD lasers this reduced separation between the QD ground state and lowest energy WL state (272 meV for the Si samples compared to 315 meV for the GaAs sample) may impact negatively on the temperature stability of QD lasers grown on Si.

In order to further assess the relative advantages of the annealing steps, full laser structures with thicker and doped cladding layers were grown. The wafers were processed into 10 μm ridge lasers, with the waveguide dry-etched down to the lower n-doped layer to allow the formation of an AuGe/Ni/Au contact. Facets were also dry-etched (in the same step) to circumvent the difficulty of forming cleaved-facets on the Si substrates. Benzocyclobutene (BCB) was used to planarise the samples and Cr-Au was deposited for the p-contact. Lasers were driven pulsed with a repetition rate of 1 kHz and a pulse width of 1 µs to prevent self-heating effects. The current versus emitted power characteristics for 3 mm long lasers processed from each wafer are shown Fig. 4, with the inset showing the corresponding emission wavelengths. The sample with no annealing (Sample A) had the highest threshold current of 382 mA and the shortest emission wavelength, at 1211 nm. For Sample B, the threshold current was 287 mA and emission occurred at 1228 nm. With an additional annealing step (Sample C), the threshold was reduced to 103 mA and the emission wavelength increased further, to 1286 nm. This wavelength shift is not observed in the below threshold emission for low injection currents, where, for example, sample A emits at 1280 nm. In samples A and B the lower lasing wavelength is likely to be a consequence of increased internal optical loss (α_i) (possibly caused by a larger number of dislocations in these structures), whereby the maximum ground state (GS) gain is no longer sufficient to balance the loss and hence the device lases from the excited state (ES), where the gain is larger due to the double degeneracy of this state. Furthermore, the increased gain requirement of these devices can only be provided by the ES at higher levels of carrier injection, as the GS must first be saturated before the ES begins to fill and as a result the threshold current is increased. Dislocations acting as non-radiative pathways contribute to an increase in the threshold current and, in addition, may cause dislocated dots to form, reducing the number of optically active dots and hence the available maximum gain. The external slope efficiency increases as the number of dislocations reaching the active region decreases (sample A→B→C), indicating either a decreasing α_i or a combination of decreasing α_i and increasing internal differential efficiency η_i. Dislocations may act to scatter photons, which will increase α_i, and/or non-radiative recombination centers which will decrease η_i. The current device results clearly demonstrate the improvement achieved by incorporating annealing steps to reduce the number of dislocations that reach the active region, with two steps giving a noticeable improvement in comparison to a single annealing step.

 

Fig. 4 Pulsed light vs current characteristics for 3 mm long ridge lasers fabricated from wafers grown with different in situ annealing steps. Sample A – no annealing, Sample B – one annealing step and Sample C – two annealing steps. The inset shows the corresponding lasing wavelengths.

Download Full Size | PDF

3. Conclusion

The addition of in situ annealing to DFLs has been shown to improve the structural and optical properties of InAs/GaAs QD structures grown on Si substrates. TEM studies show that increasing the number of annealing steps results in a reduced dislocation density reaching the III-V active region. The addition of two annealing steps improves the threshold current density of a 1.3 μm emitting InAs/GaAs QD laser by more than a factor of three. A comparison with a structure grown on GaAs shows a residual cool-down strain of 0.17% at room temperature for GaAs-based structures grown on Si. This strain results in a small modification of the active region electronic structure and will limit the thickness of III-V material that can be grown before relaxation occurs. Further work could study the efficiency of different annealing temperatures and increased number of annealing steps. However it is expected that at some point additional annealing steps will cease to be effective, as the number of remaining dislocations will become too small for it to be likely that two meet and annihilate.