We analyze the structure of dislocations in electrically aged InAs quantum dot (QD) lasers on silicon to understand gradual device degradation. We find that misfit dislocations lengthen due to carrier injection, experiencing a combination of recombination-enhanced climb and glide processes constrained by the epitaxial structure. An examination of the dislocation geometry reveals that the climb process involves the addition of atoms to the extra half plane of the dislocation. Spontaneous emission from the QDs is also dimmer after aging. Additionally, the signature of misfit dislocations in the unaged laser, discernible as sharp dark lines in spatially resolved cathodoluminescence, is replaced by finer, more inhomogeneous contrast upon aging. We speculate that this change arises from vacancy clouds expelled from the dislocation during climb. With this insight, we evaluate the driving forces for dislocation climb that could be at play and discuss the origins of slow degradation in QD lasers.

Dislocations in crystals climb by absorbing or emitting point defects in response to mechanical stress.1,2 This typically occurs when the configuration of stress cannot lead to deformation via easy glide of dislocations and when point defect diffusion is rapid.3 Dislocations can also climb in response to a purely chemical driving force (osmotic force) arising from a point defect population that is not in equilibrium, such as after rapidly quenching the sample following high temperature growth.4 Here, the dislocation acts as a source or a sink for point defects and helps drive the system toward equilibrium. Dislocation climb is not usually considered in semiconductors as they have high inter-atomic barriers5,6 that make point defect diffusion to or from the dislocation sluggish even at growth temperatures, let alone at device operating temperatures. Therefore, it is both remarkable and unexpected to see overwhelming evidence for fast climb of dislocations upon device operation at room temperature in early quantum-well (QW) and double heterostructure GaAs-based semiconductor lasers.7–9 These devices contain rapidly climbing networks of dislocations often originating from individual threading dislocations, representing a significant increase in the total dislocation line length during device operation. The outcome is an increase of non-radiative minority carrier recombination and a corresponding increase in the operating current that culminates in rapid device failure.10 

Early laser reliability research identified that these devices fail by a class of mechanisms described as recombination-enhanced defect reactions.7,11–14 Here, the electronic energy released during non-radiative minority carrier recombination helps surmount the large inter-atomic barriers to point defect diffusion and enables dislocation climb.15 These studies find that the dislocation climb network for GaAs has an interstitial character (negative climb), with the dislocations absorbing both gallium and arsenic atoms from the bulk to preserve lattice polarity.12 However, further research in this area slowed dramatically as improvements in the threading dislocation density of III–V substrates and careful processing made the problem all but obsolete. In recent years, recombination-enhanced defect reactions have once again become a critical research problem in the push for heteroepitaxially integrated III–V QW and quantum dot (QD) lasers on silicon. These devices are leading candidates for integrated light sources on a CMOS-compatible platform for silicon photonics,16–21 but mismatch in lattice constant and thermal expansion between III–V and Si materials lead to large numbers of misfit and threading dislocations in the active region.22–27 Yet, despite these high dislocation densities (and in stark contrast to rapid degradation seen in QW lasers), degradation in InAs QD lasers on silicon appears to be slow.28–30 These results suggest that a more complete understanding of how dislocations contribute to QD laser degradation will enable reliable lasers on silicon for the first time.

In this paper, we describe the structure of the as-grown dislocations in the QD laser and our observations of recombination-enhanced dislocation growth and provide a view into the local luminescence landscape in unaged and aged lasers via measurements of spatially resolved cathodoluminescence (CL) intensity. We conclude with an updated discussion on the driving force for recombination-enhanced dislocation climb and the opportunities afforded by InAs QDs toward reliable devices.

We grow the InAs QD laser diodes on silicon substrates using molecular beam epitaxy (MBE). The layer structures, growth conditions, and fabrication methods have been published previously.16,31 We focus this paper on data from four- and five-QD layer aged laser diodes (devices 1 and 2, respectively) that have threading dislocation densities of 7 × 106/cm2, and briefly discuss an older generation seven-QD layer device28 with a threading dislocation density that is on the order of 108/cm2 (device 3). Devices 1 and 2 have been aged at a temperature of 35° C at a bias of 40 mA (≈1 kA/cm2). We perform structural characterization of climbed misfit dislocations in device 1 after 500 h of aging, upon which time the peak power has dropped by 50%. We perform cathodoluminescence imaging of device 2 aged for a similar duration of time, experiencing a 10% drop in power 300 h into the aging process. Device 3 was aged for 600 h under a bias of 100 mA (≈2.5 kA/cm2), leading to a tripling of the threshold current. We do not know if these samples degrade uniformly or in localized hotspots or the facets. Therefore, we do not attempt to correlate individual microstructural features to the device-level electrical data.

Sample preparation: Plan-view (PV)-STEM foils were prepared from both unaged and aged laser samples using a FEI Helios Dualbeam Nanolab 600 and standard lift out procedures. The samples were thinned to an approximate thickness of 0.8 μm for use in tomographic reconstructions. PV-STEM imaging parameters: Images were collected on two different systems. Figure 1(b), 3(a)3(g), 4(a), and 5(a)5(d) were collected using on zone axis imaging with a Hitachi HD-2300 dedicated 200 kV STEM using an axial bright field (BF) detector. Figures 1(c)1(e) were collected using a ThermoFisher Talos G2 200X TEM/STEM and a standard BF STEM circular detector. Using a camera length of 125 mm, samples were tilted (α) from −35° to 35° and imaged at 7° steps along the g = 220 Kikuchi band for tomographic reconstructions. Tomographic reconstructions: We used Tomviz (https://tomviz.org) to obtain reconstructions of the dislocations imaged as described previously. The image stacks were manually aligned, optimizing the tilt axis for the set of 11 images. We then used the software's “Simple back projection” algorithm, using the ability to resolve individual QDs and QD layers as feedback for the quality of reconstruction. Thin slices of the reconstruction in plan-view are presented.

FIG. 1.

(a) Schematic of stacks of InAs quantum dot (QD) layers showing misfit dislocations at the top and bottom layers, respectively. We also show threading dislocations cutting across the QD layers. (b) Plan-view scanning transmission electron microscopy (PV-STEM) of a single misfit dislocation from an unaged and aged four-QD-layer laser (device 1), the latter showing protrusions on one side of the dislocation. (c) PV-STEM image taken through an aged five-QD layer laser (device 2) showing misfit dislocations with aging related protrusions; the QDs in the different stacks overlap in this projection. The yellow and red dashed boxes correspond to regions further analyzed in (d) and (e) respectively. (d) and (e) Single thin horizontal slice containing the misfit dislocations marked in (c) from 3D-tomographic reconstructions using PV-STEM images taken at multiple sample tilts. These slices now contain only the bottom (d) and top (e) single QD layers (no overlapping QDs) and confirms the vertical position of the misfit dislocations and that the protrusions lie in the (001) plane during aging.

FIG. 1.

(a) Schematic of stacks of InAs quantum dot (QD) layers showing misfit dislocations at the top and bottom layers, respectively. We also show threading dislocations cutting across the QD layers. (b) Plan-view scanning transmission electron microscopy (PV-STEM) of a single misfit dislocation from an unaged and aged four-QD-layer laser (device 1), the latter showing protrusions on one side of the dislocation. (c) PV-STEM image taken through an aged five-QD layer laser (device 2) showing misfit dislocations with aging related protrusions; the QDs in the different stacks overlap in this projection. The yellow and red dashed boxes correspond to regions further analyzed in (d) and (e) respectively. (d) and (e) Single thin horizontal slice containing the misfit dislocations marked in (c) from 3D-tomographic reconstructions using PV-STEM images taken at multiple sample tilts. These slices now contain only the bottom (d) and top (e) single QD layers (no overlapping QDs) and confirms the vertical position of the misfit dislocations and that the protrusions lie in the (001) plane during aging.

Close modal
FIG. 2.

(a) The black arrow represents a 60° misfit dislocation in GaAs relieving tensile strain along 110 with a Burgers vector b1. This vector is resolved into tilt (edge), misfit (edge), and twist (screw) components to aid the discussion of climb in the text. b2b4 are the other Burgers vectors that relieve an equivalent amount of strain in this direction (same misfit component) and are equally probable but have different tilt and screw components. (b) Similar to (a), but this dislocation relieves compressive strain using four equally probable Burgers vectors b5b8 that can likewise be resolved into various tilt and twist components. (c) A 60°dislocation with a positive (upward) tilt component, the extra half plane is shown. The dislocation undergoes positive climb, contracting this extra half plane by absorbing vacancies into the bulk. (d) Similar to (c) but corresponding to a negative (downward) tilt component. This dislocation is undergoing negative climb, expanding the extra half plane by emitting vacancies into the bulk. Both (c) and (d) will appear similar using conventional PV-STEM employed in this study but corresponds to different climb mechanisms.

FIG. 2.

(a) The black arrow represents a 60° misfit dislocation in GaAs relieving tensile strain along 110 with a Burgers vector b1. This vector is resolved into tilt (edge), misfit (edge), and twist (screw) components to aid the discussion of climb in the text. b2b4 are the other Burgers vectors that relieve an equivalent amount of strain in this direction (same misfit component) and are equally probable but have different tilt and screw components. (b) Similar to (a), but this dislocation relieves compressive strain using four equally probable Burgers vectors b5b8 that can likewise be resolved into various tilt and twist components. (c) A 60°dislocation with a positive (upward) tilt component, the extra half plane is shown. The dislocation undergoes positive climb, contracting this extra half plane by absorbing vacancies into the bulk. (d) Similar to (c) but corresponding to a negative (downward) tilt component. This dislocation is undergoing negative climb, expanding the extra half plane by emitting vacancies into the bulk. Both (c) and (d) will appear similar using conventional PV-STEM employed in this study but corresponds to different climb mechanisms.

Close modal
FIG. 3.

(a) A single dislocation A with a misfit segment along 110 imaged in an aged four-QD-layer laser (device 1) using plan-view scanning transmission electron micrograph (PV-STEM). The threading segment of this dislocation lying on a {111}-family plane is seen at the bottom. The lighter region at the top of this image corresponds to a thinner un-injected portion of the laser and hence the misfit dislocation remains straight; only the injected part of the misfit dislocation experiences dislocation climb. Additionally, this dislocation cross-slips at the very top of the image, highlighted by the orange arrow. (b) Magnified view of dislocation A. (c) and (d) The dislocation A is imaged after rotating the sample along the directions indicated (black arrows correspond to the axis of rotation following the right hand rule). (e) PV-STEM of a single dislocation B with a misfit segment along 110 also with a threading segment. Protrusions can be seen. (f) and (g) The misfit dislocation B imaged after tilting the sample along the directions indicated. A sketch showing the dislocations (h) A and (i) B deduced from these tilted images; they thread up toward the surface and down into the substrate, respectively. The misfit segments of A and B that lie on the orange planes correspond to (001) planes and represent the top and bottom QD/cladding interfaces, respectively. The protrusions are also constrained to lie on the (001) planes. The blue planes are {111}-family glide planes that must contain both the misfit and threading segments of the dislocation and, by definition, the Burgers vectors (shown as red arrows).

FIG. 3.

(a) A single dislocation A with a misfit segment along 110 imaged in an aged four-QD-layer laser (device 1) using plan-view scanning transmission electron micrograph (PV-STEM). The threading segment of this dislocation lying on a {111}-family plane is seen at the bottom. The lighter region at the top of this image corresponds to a thinner un-injected portion of the laser and hence the misfit dislocation remains straight; only the injected part of the misfit dislocation experiences dislocation climb. Additionally, this dislocation cross-slips at the very top of the image, highlighted by the orange arrow. (b) Magnified view of dislocation A. (c) and (d) The dislocation A is imaged after rotating the sample along the directions indicated (black arrows correspond to the axis of rotation following the right hand rule). (e) PV-STEM of a single dislocation B with a misfit segment along 110 also with a threading segment. Protrusions can be seen. (f) and (g) The misfit dislocation B imaged after tilting the sample along the directions indicated. A sketch showing the dislocations (h) A and (i) B deduced from these tilted images; they thread up toward the surface and down into the substrate, respectively. The misfit segments of A and B that lie on the orange planes correspond to (001) planes and represent the top and bottom QD/cladding interfaces, respectively. The protrusions are also constrained to lie on the (001) planes. The blue planes are {111}-family glide planes that must contain both the misfit and threading segments of the dislocation and, by definition, the Burgers vectors (shown as red arrows).

Close modal
FIG. 4.

(a) Plan-view scanning transmission electron micrograph (PV-STEM) of two coplanar misfit dislocations in an aged InAs quantum dot laser grown on a 6° miscut silicon substrate. These dislocations have evidence of recombination-enhanced growth on opposite sides due to aging (white arrows). Note that protrusions grow at select points of the dislocation while large portions of the dislocation remain unaffected/straight. (b) Schematic diagram of these misfit dislocations assuming they relieve compressive strain and have the line directions as shown. The blue planes represent two {111} glide planes and the single orange plane represents the growth interface tilted from the (001) due to miscut. The two misfit dislocations have different {111} glide planes because they have opposite tilt components of the Burgers vector (red arrows).

FIG. 4.

(a) Plan-view scanning transmission electron micrograph (PV-STEM) of two coplanar misfit dislocations in an aged InAs quantum dot laser grown on a 6° miscut silicon substrate. These dislocations have evidence of recombination-enhanced growth on opposite sides due to aging (white arrows). Note that protrusions grow at select points of the dislocation while large portions of the dislocation remain unaffected/straight. (b) Schematic diagram of these misfit dislocations assuming they relieve compressive strain and have the line directions as shown. The blue planes represent two {111} glide planes and the single orange plane represents the growth interface tilted from the (001) due to miscut. The two misfit dislocations have different {111} glide planes because they have opposite tilt components of the Burgers vector (red arrows).

Close modal
FIG. 5.

(a) and (b) Dislocation loops visible as darker circles amidst lighter quantum dots form adjacent to climbed segments of misfit dislocations in aged device 3, imaged in plan-view using scanning transmission electron microscopy. (c) Magnified view of the dislocation loops in (b). (d) Another magnified view of dislocation loops formed near climbed dislocations. The dislocation loops form right above a quantum dot and are identified by the Ashby–Brown strain contrast. (e) Schematic of dislocations loops (black circles) forming over quantum dots (yellow squares) in the vicinity of a climbed misfit dislocation segment. Invisible point defects (black dots) dissolved in the bulk that have not coalesced into dislocation loops are also shown.

FIG. 5.

(a) and (b) Dislocation loops visible as darker circles amidst lighter quantum dots form adjacent to climbed segments of misfit dislocations in aged device 3, imaged in plan-view using scanning transmission electron microscopy. (c) Magnified view of the dislocation loops in (b). (d) Another magnified view of dislocation loops formed near climbed dislocations. The dislocation loops form right above a quantum dot and are identified by the Ashby–Brown strain contrast. (e) Schematic of dislocations loops (black circles) forming over quantum dots (yellow squares) in the vicinity of a climbed misfit dislocation segment. Invisible point defects (black dots) dissolved in the bulk that have not coalesced into dislocation loops are also shown.

Close modal

Focused ion beam laser preparation: Samples were prepared using a FEI Helios Dualbeam Nanolab 600 (FIB) in several steps. Laser bars were removed from carrier wafers, Pt coated, and mounted flat in the FIB. To view the center of the laser bar, far from possible irregularities at the facets, trenches were cut from the side of the chip, perpendicular to the bar, to the bar's edge. The edge of the laser bar was polished, and 30° cuts were made into the bar at either end of the trench to judge the depth of the height of the surface across the ridge. Samples were unloaded and remounted at 90°. Pt was locally deposited to protect the edge of the bar and then the contact metal, isolator, and some of the p-cladding were removed such that optical emission generated in the cathodoluminescence (CL) could be collected. Cathodoluminescence spectroscopy: All CL maps were collected at room temperature using an Attolight Rosa system. An incident beam accelerating voltage of 8 kV was used with an approximate probe current of ∼5 nA. The emission at each spatial location (pixel) was collected over a 262 nm range, centered at 1250 nm, in 0.5 nm increments using an Andor iDus InGaAs detector. The exposure time was 0.25 s per pixel. Due to some drift in the incident electron beam over the long scans, a slight shifting can be observed in all images.

We introduce the structure of the dislocation network in devices 1 and 2 schematically in Fig. 1(a). These samples have a threading dislocation density of 7 × 106/cm2. They also have misfit dislocations that we have previously identified as lying close to either the top or bottom QD layers29,32 and have proposed a mechanism for their formation.27 We suggest that these misfit dislocations form during sample cooldown in response to the buildup of 0.15%–0.2% tensile strain in the AlGaAs and GaAs layers caused by a mismatch in thermal expansion between the III–V layers and Si. The strain in the upper p-doped AlGaAs cladding is relieved partially by the glide of some of the many threading dislocations to create a new tensile strain relieving misfit dislocation segments at the top QD layer/cladding interface. Pre-existing compressive strain relieving misfit dislocation segments rise to near the bottom QD layer/cladding interface to relieve the tensile strain in the lower n-doped AlGaAs cladding. Both types of misfit dislocations are stuck at the QD layers due to a combination of compressive stress and threading dislocation pinning in the QD layers.

STEM analysis of an unaged and aged sample of device 1 in Fig. 1(b) reveals that electrically aging such a laser leads to the previously straight misfit dislocations developing small curved segments or protrusions that are suggestive of recombination-enhanced dislocation climb. These protrude only on one side of the straight dislocation and are not quite semi-circular in shape but rather slightly tilted or stretched. Figure 1(c) shows a plan-view STEM image of the orthogonal network of misfit dislocations with protrusions from an aged device 2. We perform a 3D-tomographic reconstruction of the dislocation and the QD layers using images taken at multiple tilt angles while staying close to a 220 two-beam diffraction condition. Using this, we examine two climbed misfit dislocations highlighted by boxes in Figs. 1(c)1(e) showing thin (001)-oriented slices of this reconstruction, revealing that these protrusions, although curved, lie entirely in horizontal (001) planes at the bottom and top QD layers, respectively, and are not helical in nature. The dislocation appears to become increasingly serpentine in this plane during aging. We also see that the rate of climb is not uniform along a dislocation. As an example, overall, the dislocation in Fig. 1(e) is ∼16% longer than a straight line (manually traced) and that in Fig. 1(d) is 12% longer, but large segments appear straight experiencing no climb.

Identifying the nature of dislocation climb is important to uncover the driving force behind it. We use PV-STEM to determine if climb causes the extra half plane corresponding to the misfit dislocation to expand (atoms attach, negative climb) or contract (atoms leave, positive climb) in the following way. Briefly, the low thermal expansion strain is relieved by misfit dislocations that have their Burgers vectors inclined 60° to the line direction.33Figure 2(a) shows a sketch of a tensile strain relieving 60° misfit dislocation along the [110] direction with the four possible Burgers vector directions (b1b4) it can have (right-hand/finish-start convention).34 For instance, we can resolve b1 into three orthogonal components, a positive b1_tilt (edge), a positive b1_twist (screw), and a tensile-relieving b1_misfit (edge). From similar decomposition of all four Burgers vectors, we see that b1andb4 result in positive (upward pointing) tilt components, whereas b2andb3 result in negative tilt components. We focus on this tilt component as this directly leads to climb in the (001)plane as we have observed. Figures 2(c) and 2(d) show a sketch of the tilt component of these dislocations corresponding to the positive and negative tilts, respectively; they differ only in the arrangement of the extra half plane. An outwardly similar protrusion in the misfit dislocation is obtained after climb if the positive tilt dislocation absorbs vacancies (or emits interstitials) or if the negative tilt dislocation emits vacancies (or absorbs interstitials). While both cases will appear identical to the PV-STEM technique used here, we can discriminate between whether the extra half plane associated with this tilt component is contracting [Fig. 2(c)] or expanding [Fig. 2(d)] if we can first obtain the sign of this tilt component of the Burgers vector. Section IV describes our method to identify the sign of the tilt component and reveal the nature of dislocation climb in QD lasers.

Figures 3(a)3(e) show PV-STEM images of two orthogonal misfit dislocations labeled A and B, respectively, that have experienced recombination-enhanced dislocation climb in device 1. We calculate dislocation A to be 15% longer than a straight line segment and B to be 11% longer. These images also show threading dislocation segments attached to the misfit dislocations. Figures 3(c)3(g) show images of these same dislocations taken by tilting the TEM foil in the microscope along two axes. From these tilted images, we deduce that the threading segment for dislocation A threads up toward the surface while that for dislocation B threads down into the substrate. The threading segments together with the misfit segments must be contained within the same glide plane, which we can now ascertain. Together from cross-sectional STEM images (not shown), Figs. 3(h)3(i) shows our interpretation of the structure of the dislocations. The blue planes indicate the particular {111}-family glide planes of dislocations A and B. The orange planes correspond to the horizontal (001) planes representing the top and bottom QD layers. We have identified that dislocation A is located at the top QD interface and hence relieves tensile strain. That the misfit segment is able to cross-slip means the threading dislocation must be screw type and implies its Burgers vector has a negative tilt component [b2 in Fig. 2(a)]. We conclude that dislocation A climbs via attaching atoms to the half plane as it has protrusions on its left side as shown [compare with Fig. 2(d)]. Likewise, misfit dislocation B is located at the bottom QD layer and hence relieves compressive strain. Based on its glide plane, it too has a negative tilt component of its Burgers vector (b6orb7). The sense of the protrusions once again indicates an expanding half plane of atoms, confirming the nature of climb in this sample.

The microstructure in the aged device 3 grown on a 6° miscut silicon substrate further illustrates the role of the tilt component. Figure 4(a) shows a plan-view TEM image of two coplanar misfit dislocations in an aged laser that are inclined due to the substrate miscut geometry.35 The dislocations in Fig. 4(a) are considerably more serpentine, consistent with longer aging. Here, the miscut discriminates between dislocations with different glide planes (and hence opposite tilt components of the Burgers vector) by changing their line directions such that they are no longer parallel. It is not possible to say from this image alone if the tilt components of the Burgers vector are positive or negative in these dislocations as no threading dislocation segments are visible. However, we can say with certainty that the tilt components should be antiparallel with respect to each other based on the skew line directions [Fig. 4(b)]. These antiparallel tilt components do indeed result in the protrusions to grow out on opposite sides of these misfit dislocations.

We have not focused on the climb of threading dislocations in this work, as their total length near the QD layers is much lower than the misfit dislocations and they are much harder to characterize. We comment that their climbed structure appears to be visually like that of the misfit dislocations with small protrusions growing out and Fig. 3(a) shows an example. More importantly, these dislocations have not expanded into long dipoles like that seen in AlGaAs lasers12 

We estimate the density of point defects involved in this misfit dislocation climb process by assuming that recombination-enhanced defect diffusion36 facilitates point defect migration either toward or away from the dislocation line. This constrains the participating point defect population to where minority carrier recombination is active—the thin active region that contains the spacer layers, quantum wells, wetting layers, and the QDs. For example, if we assume a uniform 10 nm lateral excursion in the misfit dislocation, an average concentration of about 1018 point defects/cm3 must be involved in this climb process assuming that these point defects diffuse just 5 μm laterally in the thin active region. As we have seen in Figs. 1 and 2, the protrusions are not uniform and many are significantly larger than 10 nm. Therefore, misfit dislocation climb in the confined volume of the active area has the potential to significantly alter the point defect population.

Negative climb or growth of the extra half plane of the dislocation can arise from the dislocation either expelling vacancies into the bulk or absorbing interstitials from the bulk. We present preliminary structural evidence that suggests that vacancies are expelled. Figures 5(a) and 5(b) shows PV-STEM images taken from device 3. Tiny dislocation loops lie adjacent to the protrusions in climbed misfit segments. The characteristic coffee bean shaped Ashby–Brown contrast of these dislocation loops can be seen clearly over several of the nearby QDs in the higher magnification images in Figs. 5(c) and 5(d). Figure 5(e) shows a schematic of this observation. We do not see these loops anywhere in the unaged lasers nor are they seen in regions far away from the climbed segments in aged lasers. The tendency for these loops to form directly over the compressively strained InAs QDs suggests that the dislocation loops themselves are centers of tensile strain and thus vacancy type. Additionally, the total area of all the loops is significantly smaller than the area of the climbed protrusion of the misfit dislocation. As these loops can be might arise from vacancies precipitating near QDs, their low density and small size implies that most of the vacancies expelled during misfit dislocation climb remain dissolved in the bulk.

In summary, we find that dislocations in InAs QD lasers on Si climb by expanding the extra plane of atoms (negative climb), analogous to quantum-well and double heterostructure lasers.8 We suspect that the dislocation emits vacancies during this climb process, some of which form dislocation loops in nearby QDs while most remain dissolved in the bulk.

We now describe the impact of dislocation climb on luminescence inferred from CL imaging on unaged and aged samples of device 2. CL of fully processed laser diodes requires significant sample preparation in order to remove metallization, contact layers, cladding, etc., to expose the active region. This enables high spatial resolution carrier injection and light collection, but the depth variability inherent in this processing modifies carrier injection and the absolute CL emission intensity. We account for differences in carrier injection into the quantum dots by comparing the ratio of the excited state and ground state emission. Greater injection/pumping leads to an increase in this ratio. Figure 6(a) shows room temperature area-integrated CL spectra collected from processed unaged and aged laser bars. We see that the emission intensity from the ground state of the aged laser is dimmer on average. Furthermore, looking at the histogram of the ratios of excited state to ground state emission measured from all points on the sample in Fig. 6(b), we see that the electron beam also injects the aged laser to a greater degree. This suggests that the aged laser is even dimmer than that suggested by the CL spectra. Thus, we conclude that the aging process leads to an overall reduction in emission, consistent with the observed lowering of device peak power during aging.

FIG. 6.

(a) Cathodoluminescence (CL) spectra collected from device 2 unaged and aged InAs quantum dot (QD) lasers on silicon. The peaks at higher and lower wavelength correspond to the QD ground state and excited state emission, respectively. (b) A probability distribution plot of the ratio of ground state to excited state emission at each measurement point. (c)–(d) Spatially resolved CL intensity maps from which (a) and (b) were generated. Regions (1–4) are imaged from device 2 unaged [(c)] and aged laser [(d)], respectively, and we show both the ground state and excited state emission intensity landscape. Straight dark lines corresponding to misfit dislocations can be seen from the unaged laser, especially in the excited state maps (dotted white box). We mark a few such misfit dislocations in the dashed white box. The emission landscape is more mottled after aging; clear signatures of misfit dislocations are no longer visible.

FIG. 6.

(a) Cathodoluminescence (CL) spectra collected from device 2 unaged and aged InAs quantum dot (QD) lasers on silicon. The peaks at higher and lower wavelength correspond to the QD ground state and excited state emission, respectively. (b) A probability distribution plot of the ratio of ground state to excited state emission at each measurement point. (c)–(d) Spatially resolved CL intensity maps from which (a) and (b) were generated. Regions (1–4) are imaged from device 2 unaged [(c)] and aged laser [(d)], respectively, and we show both the ground state and excited state emission intensity landscape. Straight dark lines corresponding to misfit dislocations can be seen from the unaged laser, especially in the excited state maps (dotted white box). We mark a few such misfit dislocations in the dashed white box. The emission landscape is more mottled after aging; clear signatures of misfit dislocations are no longer visible.

Close modal

Figures 6(c) and 6(d) show spatially resolved cathodoluminescence emission maps prepared from the ground state and excited state emission from four sections of these unaged and aged devices. These maps provide a localized view of changes to the emission profiles because of aging. These maps are normalized and do not show the dimming in the aged samples. Although the data have significant variance, we can resolve orthogonal straight dark lines in the unaged lasers (dashed box) due to the misfit dislocations seen previously in STEM. The spatial resolution of this technique renders it insufficient to resolve closely spaced misfit dislocations. Nevertheless, these dark lines suggest that misfit dislocations are non-radiative in the unaged lasers.

The emission maps change quite dramatically for the aged devices. They have a distinctly mottled look and it is no longer possible to resolve dark lines. We expect that the lengthening of the dislocation would greatly increase the number of non-radiative recombination sites associated with this dislocation. Such a change should lead to much darker straight lines in CL maps because the protuberances are smaller than the spatial resolution of CL. Our results instead indicate that dispersed pockets of non-radiative recombination that we suspect are the clouds of point defects and dislocation loops generated by the misfit dislocation during climb. We believe that the misfit dislocation density in these samples is high enough that overlapping point defect impact obscures the features of any single misfit dislocation. Such mottled features were also seen in electroluminescence by Beanland et al. arising due to structurally defective-InAs QDs on GaAs substrate.37 In the only study we found that probed the recombination behavior of a single dislocation after climb, Lang et al. found a reduction in the DX-center trap density in an AlGaAs heterostructure following recombination-enhanced dislocation climb using scanning deep level transient spectroscopy.38 They hypothesize that the DX-center (now known to be a silicon–gallium vacancy complex) in AlGaAs is the source of the gallium interstitial needed for negative climb. Our results suggest the opposite: there is an increase in dispersed non-radiative recombination centers due to vacancy emission.

We have shown that misfit dislocations in InAs QD lasers on silicon experience negative climb upon electrical aging, attaching atoms to the extra half plane of the dislocation, although constrained to lie in the plane of the QD layer. Dislocation loops are generated on nearby QDs in the wake of this process, but a much larger point defect concentration change is expected to remain dissolved in the bulk and this leads to significant non-radiative recombination and overall dimming of the active region.

Negative climb is also seen in traditional GaAs-based quantum-well and double heterostructure lasers;12 yet, the driving force or the source of the point defects is not clarified even in these systems. Early GaAs-based devices were grown using Ga-rich liquid phase epitaxy, a technique in which it was conceivable that an excess of gallium interstitials might be quenched in and provided a driving force for climb. This is the so-called extrinsic defect model where the point defects involved in climb pre-exist in the bulk.39 Researchers at the time hypothesized that the source of arsenic atoms necessary to complete the climb sequence came from expelling arsenic vacancies into the bulk, still energetically favorable due to the high formation energy of gallium interstitials.40 Materials grown under arsenic overpressure in MBE and metalorganic chemical vapor deposition (MOCVD) also experience recombination-enhanced dislocation climb; yet, here arsenic anti-sites and gallium vacancies are likely the dominant point defects as compared to gallium interstitials.41,42 There continues to remain support for this extrinsic defect model where the grown-in point defect population is important to reliability. Waters and Hill observe that controlling the V/III ratio, which will change the grown-in point defect populations, during laser synthesis has an impact on device reliability.43 

Citing that the extrinsic defect model requires an unreasonable 1018–1019/cm3 excess interstitials, O’Hara et al. proposed an alternate intrinsic source model where these point defects are created at the dislocation itself during non-radiative recombination.8 Non-radiative recombination dislodges both gallium and arsenic atoms from the lattice near the dislocation creating both interstitials and vacancies. Strain effects lead to interstitials preferentially migrating toward dislocations, the so-called interstitial bias widely reported in metals.44 O’Hara et al. suggest that this could be at play in semiconductors, leading to only interstitials attaching to the extra half plane and leaving behind vacancies to diffuse away. The intrinsic source model has also received criticism. Neutral point defects often have formation energies in excess of the bandgap energy, so it is not obvious how sufficiently high numbers of these point defects could be formed in this manner.45 Furthermore, it is unclear why this chemical reaction would occur preferentially in one direction (i.e., why do vacancies and interstitials not annihilate right after creation or why are vacancies not reabsorbed at the dislocation core after their creation?).46 This necessitates alternative thermal equilibrium and steady-state models (factoring in the forward bias injection) to yield a driving force for negative climb in MBE and MOCVD grown devices that we now discuss in detail.

Tan et al. first highlighted a phenomenon where moderately doped n-type GaAs crystals grown under arsenic-rich conditions could have higher acceptor-type triply negatively charged Ga vacancy (VGa) concentrations in equilibrium at room temperature than at growth temperatures.46 This is contrary to conventional wisdom that suggests that there are fewer point defects at lower temperatures. The position of the Fermi-level in the bandgap rises upon cooling a moderately n-type sample. This greatly reduces the formation energy of triply charged acceptor-type defects41 via what is known as the Fermi-level effect.47 Nearly all the vacancies have a 3− charge due this very drastic lowering of the formation energy. Gebauer et al. have experimentally validated this idea of an increasing gallium vacancy concentration upon cooling in the range of 700–1100 °C.48 Building on results by Tan et al., they show that the fractional concentration of gallium vacancy cGa is given by

cGa=0.293×T5/8(pAs4)1/4exp[(Hf0{zEFΣEa,i}TSf)/kBT],

with Sf being the formation entropy and Hf0 is the formation enthalpy of the neutral gallium vacancy. kBandT are the Boltzmann constant and temperature, respectively. The defect becoming acceptor like and acquiring a negative charge reduces this free energy of formation by a value zEFΣEa,i. Here, the defect acquires a charge z corresponding to its zth ionization level, Ea,i is the ionization energy of the ith level and is summed from 0 to z, and EF is the position of the Fermi level. The latter two energy terms are referenced to the valence band. pAs4 is the As4 vapor partial pressure in equilibrium with the system. We show this increasing trend in total gallium vacancy concentration with decreasing temperature for different extrinsic donor concentrations in Fig. 7 using two different values of pAs4, one fixed at 1 atm. consistent with the work done by Ng et al.,62 and the other set by the equilibrium over arsenic-rich GaAs as determined by Arthur.49 We have followed the calculation methodology in Tan et al.46 but use experimentally determined parameters by Gebauer et al.,48 which we find are also in agreement with recent first principles calculations.42 The equilibrium concentration of gallium vacancies increases on average when the sample cools down after growth at 600 °C.

FIG. 7.

Negative temperature dependence of the gallium vacancy calculated for different extrinsic donor concentrations in GaAs. On average, there is an increase in equilibrium the gallium vacancy concentration due to cooldown from 600 °C, leading to under saturation of gallium vacancies in typically quenched growths. We show this for two cases of the partial pressure of As4: one fixed to 1 atm and the other that corresponds to the equilibrium vapor pressure above arsenic-rich GaAs or solid arsenic.

FIG. 7.

Negative temperature dependence of the gallium vacancy calculated for different extrinsic donor concentrations in GaAs. On average, there is an increase in equilibrium the gallium vacancy concentration due to cooldown from 600 °C, leading to under saturation of gallium vacancies in typically quenched growths. We show this for two cases of the partial pressure of As4: one fixed to 1 atm and the other that corresponds to the equilibrium vapor pressure above arsenic-rich GaAs or solid arsenic.

Close modal

Tan et al. use this idea to propose an alternate (intrinsic source) model where quenched samples (such as during MBE cooldown) likely have an under saturation of gallium vacancies, and this provides the driving force for negative dislocation climb. Attaching atoms from the lattice to the extra half plane of the dislocation expels vacancies in the bulk and serves to remove this under saturation of gallium vacancies.46 This suggestion agrees with our preliminary findings. If this is the case, we think that arsenic antisite defects (AsGa0) that are also expected to be prevalent in arsenic-rich GaAs might be an important source of the arsenic atoms necessary to complete the dislocation climb process after the gallium atom has attached. The absorption of an arsenic atom from the antisite defect also liberates a gallium vacancy. Hence, this process should have a driving force at least that of the gallium climb sub-step. This driving force is further increased because the arsenic antisite defects are expected to be neutral defects42 and should be over-saturated at room temperature. The magnitude of point defects necessary to restore the crystal to equilibrium according to Fig. 7 is within range of our earlier estimate from STEM images of the point defects generated by climb (Sec. III B).

We acknowledge that Tan's thermal point defect equilibria based on the Fermi level effect discussed here need to be modified under steady-state carrier injection experienced by the laser for an accurate description of the driving force. Indeed, there has been a surge of interest in using photoexcitation to tailor the concentration of point defects away from thermal equilibrium during growth.50–52 These effects now depend on both the electron and hole quasi-Fermi levels and are likely important during laser operation where the excess electron and hole concentrations exceed that in light stimulated growth. The gallium vacancy (2−/3−) charge transition level is close to midgap and hence the formation energy is expected to be strongly linked to the electron quasi-Fermi level as the defect is more likely to attain a 3− charge state via capture of an electron from the conduction band as opposed to emitting a hole to the valence band. The electron quasi-Fermi level lies in the conduction band in lasers and should provide an even greater driving force for compensation than in the thermal equilibrium scenario described earlier. We should consider this in future work using simpler model systems without the complex environments of quantum dots.

It is clear from the nature of the protrusions on the misfit dislocations that atoms attach to the extra half plane. We discuss further how the unique stress state in these systems keeps the misfit dislocation in the (001) plane even after climb and generates a slant to the protrusion. Figure 8(a) shows a sketch of a typical 60° misfit dislocation at the interface between the lower cladding layer and the bottom QD layer. We remind the reader that such misfit dislocations lie at the interface between n-doped cladding and the QD layer because this is where the sign of stress switches from tensile (cladding) to compressive (QD layer). The dislocation is in mechanical equilibrium as it experiences an equal and opposite resolved shear stress from each of the layers, upward from cladding and downward from the QD layer. The sign of the Burgers vector relieves compressive misfit stress, and by way of example, we assume it has a positive tilt and screw component [b5 in Fig. 2(b)]. The extra plane of atoms corresponding to the dislocation lies in the lower cladding layer. A convenient scheme to consider climb of this 60° dislocation would be the addition of a unit of point defects in the form of a prismatic dislocation loop with the same Burgers vector as the misfit dislocation.53Figure 8(a) depicts such an approaching interstitial loop (note the sense of the dislocation loop) with the same Burgers vector as the misfit dislocation and represents one method to expand the extra half plane that we identified must be happening. We break down this next step into two sequential elementary steps that in reality must happen together. The attachment of the loop leads to a segment of the dislocation protruding out of the (11¯1) glide plane of the dislocation, shown in Fig. 8(b). This corresponds to expanding the extra plane of atoms into the QD layer. Now this curved segment of dislocation is no longer at the neutral strain interface but rather inside the compressively strained QD layer and hence experiences a restoring glide force. Glide for this circular segment no longer occurs on a single plane but rather a surface of a glide cylinder also shown in Fig. 8(b) (the cylinder is defined by the curved dislocation line and the cylinder's axis is parallel to the Burgers vector and this corresponds to prismatic slip).3 Thus, after a unit of dislocation climb, the curved segment is pushed out of the QD layer by glide along the cylindrical surface [yellow arrow in Fig. 8(b)] until it once again reaches the cladding/QD neutral stress plane. The curved segment from recombination-enhanced climb is now stretched in one direction because of recombination-enhanced glide [Fig. 8(c)]. This direction is opposite to the twist component of the Burgers vector of the misfit dislocation (btwist) [Fig. 8(d)]. Revisiting Fig. 3(e) and inspecting this angle of slant of the protrusion, we confirm b2 as the correct choice of Burgers vector for dislocation A and establish b7 for dislocation B.

FIG. 8.

(a) A sketch of a misfit dislocation with a line direction (ξ) along the [110] at the interface between the bottom QD layer and the lower cladding. This dislocation relieves compressive misfit strain. One out of four possible Burgers vector that allows for this,b=a2[011], is shown with a positive tilt component and a positive twist component. We represent point defects required for negative climb, expanding the extra half plane, using a prismatic interstitial dislocation loop with the same Burgers vector as the misfit dislocation. (b) Attachment of the point defects leads to the growth of a protrusion. The glide plane for this curved segment is the surface of a cylinder shown in blue, the axis of the cylinder as the same as the Burgers vector. The protrusion experiences a glide force (yellow arrow) along this surface due to compressive stress in the QD layer. (c) The stretched protrusion formed as a result of this combined climb and glide process. (d) Plan-view of such a dislocation showing both how the stretched protrusion forms on one side of the straight segment of dislocation.

FIG. 8.

(a) A sketch of a misfit dislocation with a line direction (ξ) along the [110] at the interface between the bottom QD layer and the lower cladding. This dislocation relieves compressive misfit strain. One out of four possible Burgers vector that allows for this,b=a2[011], is shown with a positive tilt component and a positive twist component. We represent point defects required for negative climb, expanding the extra half plane, using a prismatic interstitial dislocation loop with the same Burgers vector as the misfit dislocation. (b) Attachment of the point defects leads to the growth of a protrusion. The glide plane for this curved segment is the surface of a cylinder shown in blue, the axis of the cylinder as the same as the Burgers vector. The protrusion experiences a glide force (yellow arrow) along this surface due to compressive stress in the QD layer. (c) The stretched protrusion formed as a result of this combined climb and glide process. (d) Plan-view of such a dislocation showing both how the stretched protrusion forms on one side of the straight segment of dislocation.

Close modal

It is a remarkable fact that InAs QD lasers show reliable operation in the tens of thousands of hours range as high as 60 °C, despite the presence of misfit dislocations close to the active region.54 Hundreds of hours of aging in this study led to only modest climb in the misfit dislocations, a significant improvement from the rapid dislocation growth seen after only a few hours of operating QW lasers on silicon.22 In fact, the overall morphology of the climbed misfit dislocations in our work closely resembles the degradation resistant InGaAsP system (see Fig. 2 in this reference).55 We do not fully understand why dislocation climb is slower in InAs QD systems although two possibilities are seen immediately. The first is that QDs reduce the overall flux of carriers to the dislocation,32,56–60 a point critical to the choice of QDs for integration on to silicon. The second lies in the mechanical hardening property that researchers say arise from the QD strain field.37,61 Compressive strain counteracts negative climb by offering resistance to the expansion of the extra half plane,62 and there are certainly many sources of compressive strain in the QD layers (both the InAs QDs and the InGaAs wells are compressive with respect to GaAs). This latter notion is most relevant to threading dislocations that cut the QD layers and can be pinned by the compressive strain and could be key to why threading dislocations in QD lasers on silicon do not form rapid climb networks. As for the misfit dislocations themselves, we believe they exist in the neutral strain plane between the compressive QDs and the tensile cladding. In such a position, it should experience no resistance to climb due to compressive strain. We do not see unambiguous evidence of pinning of the misfit dislocation by QDs in the TEM images. Therefore, we suggest that the former possibility of low flux of carriers that migrate to the misfit dislocations might be responsible for the slow rate of climb. There is a third possibility that we state to motivate future work. It is possible that the driving force for climb in the barriers is low in MBE grown films that have substantial p-type doping. P-type doping lowers the Fermi level and significantly raises the formation energy of gallium vacancies following the logic of Tan's model described in Sec. IV A. This could be part of the story behind the experimental observation of marked improvement in the reliability of p-doped QD lasers over undoped ones.63 Complementary to efforts in understanding dislocation climb, we recently demonstrate incorporating trapping layers as a potential route to displacing misfit dislocations;27 more work is needed to understand how this impacts device reliability.

Misfit dislocations present in the top and bottom layers of InAs QD structures on silicon undergo recombination-enhanced dislocation climb during device operation and contribute to device degradation. From an analysis of the protrusions on previously straight dislocations, we surmise that the extra half plane corresponding to the dislocation grows in extent by attaching atoms during climb, like what happens in GaAs quantum-well and double heterostructure lasers. However, the climbed segments also stretch and further lengthen by recombination-enhanced glide owing to the compressive strain in the QD layers. Comparing cathodoluminescence in unaged and aged lasers reveals that aging indeed leads to an overall reduced intensity. We see structural signatures of vacancy emission from dislocation climb that may be linked to an increase in the inhomogeneity of the spatially resolved emission intensity. We speculate that the generation of gallium vacancies and the reduction of arsenic antisites to move toward point defect equilibria drives recombination-enhanced dislocation climb. On the other hand, point defect equilibrium under carrier injection is significantly different from thermal equilibrium, and more work is needed to understand the true microscopic driving force that appears to be common to all GaAs-based lasers. Notwithstanding the point defects involved, the origins of slower dislocation climb in QD lasers are likely due to greatly reduced non-radiative recombination, with threading dislocations experiencing additional slowing from the compressive indium-rich microstructure of the QDs. As QD lasers on silicon approach commercialization, the need to close the gap between the rapid device development and our lagging understanding of the materials science of defects in these complex systems becomes pressing. Likewise, these devices can re-invigorate the study of recombination-enhanced defect reactions. We hope that methods in experiment and theory developed in recent years can shed light on these complex materials phenomena and clarify decades-old problems in device reliability.

Sample growth was supported by ARPA-E, U.S. Department of Energy, under Award No. DE-AR00000843. This study is based on the work supported by the National Science Foundation Graduate Research Fellowship under Grant No. 1650114. Further support has been provided by the University of California, Santa Barbara Graduate Division through the Doctoral Scholars Program. K.M. acknowledges support from the California Nanosystems Institute SEED-TECH program and discussions with E. Spiecker. The research reported here made use of shared facilities of the UCSB MRSEC (No. NSF DMR 1720256), a member of the Materials Research Facilities Network (www.mrfn.org). A.Y.L. is with Quintessent Inc.

The data that support the findings of this study are available from the corresponding author upon reasonable request.

1.
W.
Shockley
,
Imperfections in Nearly Perfect Crystals
(
John Wiley and Sons
,
La Jolla
,
CA
,
1952
).
2.
J.
Lothe
and
J. P.
Hirth
,
J. Appl. Phys.
38
,
845
(
1967
).
3.
J. P.
Hirth
and
J.
Lothe
,
Theory of Dislocations
, Reprint ed. (
Krieger Pub Co
,
Malabar
,
FL
,
1992
).
4.
J.
Weertman
,
Philos. Mag. A
11
,
1217
(
1965
).
5.
D. V.
Lang
,
L. C.
Kimerling
, and
S. Y.
Leung
,
J. Appl. Phys.
47
,
3587
(
1976
).
6.
S. F.
Wee
,
M. K.
Chai
,
K. P.
Homewood
, and
W. P.
Gillin
,
J. Appl. Phys.
82
,
4842
(
1997
).
7.
P.
Petroff
and
R. L.
Hartman
,
Appl. Phys. Lett.
23
,
469
(
1973
).
8.
S.
O’Hara
,
P. W.
Hutchinson
, and
P. S.
Dobson
,
Appl. Phys. Lett.
30
,
368
(
1977
).
9.
O.
Ueda
,
Jpn. J. Appl. Phys.
49
,
090001
(
2010
).
10.
S. L.
Yellen
,
A. H.
Shepard
,
R. J.
Dalby
,
J. A.
Baumann
,
H. B.
Serreze
,
T. S.
Guido
,
R.
Soltz
,
K. J.
Bystrom
,
C. M.
Harding
, and
R. G.
Waters
,
IEEE J. Quantum Electron.
29
,
2058
(
1993
).
11.
H.
Yonezu
,
I.
Sakuma
,
T.
Kamejima
,
M.
Ueno
,
K.
Nishida
,
Y.
Nannichi
, and
I.
Hayashi
,
Appl. Phys. Lett.
24
,
18
(
1974
).
12.
P. W.
Hutchinson
and
P. S.
Dobson
,
Philos. Mag. A
32
,
745
(
1975
).
13.
L. C.
Kimerling
,
P.
Petroff
, and
H. J.
Leamy
,
Appl. Phys. Lett.
28
,
297
(
1976
).
14.
K.
Maeda
and
S.
Takeuchi
,
J. Phys. Colloques
44
,
C4
(
1983
).
15.
D. V.
Lang
,
Annu. Rev. Mater. Sci.
12
,
377
(
1982
).
16.
A. Y.
Liu
,
J.
Peters
,
X.
Huang
,
D.
Jung
,
J.
Norman
,
M. L.
Lee
,
A. C.
Gossard
, and
J. E.
Bowers
,
Opt. Lett.
42
,
338
(
2017
).
17.
S.
Chen
,
M.
Liao
,
M.
Tang
,
J.
Wu
,
M.
Martin
,
T.
Baron
,
A.
Seeds
, and
H.
Liu
,
Opt. Express
25
,
4632
(
2017
).
18.
A. Y.
Liu
and
J.
Bowers
,
IEEE J. Sel. Top. Quantum Electron.
24
,
1
(
2018
).
19.
J. C.
Norman
,
D.
Jung
,
Y.
Wan
, and
J. E.
Bowers
,
APL Photonics
3
,
030901
(
2018
).
20.
J.
Kwoen
,
B.
Jang
,
J.
Lee
,
T.
Kageyama
,
K.
Watanabe
, and
Y.
Arakawa
,
Opt. Express
26
,
11568
(
2018
).
21.
K.
Li
,
Z.
Liu
,
M.
Tang
,
M.
Liao
,
D.
Kim
,
H.
Deng
,
A. M.
Sanchez
,
R.
Beanland
,
M.
Martin
,
T.
Baron
,
S.
Chen
,
J.
Wu
,
A.
Seeds
, and
H.
Liu
,
J. Cryst. Growth
511
,
56
(
2019
).
22.
M. E.
Groenert
,
C. W.
Leitz
,
A. J.
Pitera
,
V.
Yang
,
H.
Lee
,
R. J.
Ram
, and
E. A.
Fitzgerald
,
J. Appl. Phys.
93
,
362
(
2003
).
23.
T.
Egawa
,
Y.
Hasegawa
,
T.
Jimbo
, and
M.
Umeno
,
Appl. Phys. Lett.
67
,
2995
(
1995
).
24.
I.
George
,
F.
Becagli
,
H. Y.
Liu
,
J.
Wu
,
M.
Tang
, and
R.
Beanland
,
Semicond. Sci. Technol.
30
,
114004
(
2015
).
25.
P. G.
Callahan
,
B. B.
Haidet
,
D.
Jung
,
G. G. E.
Seward
, and
K.
Mukherjee
,
Phys. Rev. Mater.
2
,
081601
(
2018
).
26.
E. T.
Hughes
,
R. D.
Shah
, and
K.
Mukherjee
,
J. Appl. Phys.
125
,
165702
(
2019
).
27.
J.
Selvidge
,
J.
Norman
,
E. T.
Hughes
,
C.
Shang
,
D.
Jung
,
A. A.
Taylor
,
M. J.
Kennedy
,
R.
Herrick
,
J. E.
Bowers
, and
K.
Mukherjee
, arXiv:2005.06066.
28.
A. Y.
Liu
,
R. W.
Herrick
,
O.
Ueda
,
P. M.
Petroff
,
A. C.
Gossard
, and
J. E.
Bowers
,
IEEE J. Sel. Top. Quantum Electron.
21
,
690
(
2015
).
29.
D.
Jung
,
J.
Norman
,
Y.
Wan
,
S.
Liu
,
R.
Herrick
,
J.
Selvidge
,
K.
Mukherjee
,
A. C.
Gossard
, and
J. E.
Bowers
,
Phys. Status Solidi A
216
,
1800602
(
2019
).
30.
Z.
Liu
,
C.
Hantschmann
,
M.
Tang
,
Y.
Lu
,
J.
Park
,
M.
Liao
,
S.
Pan
,
A. M.
Sanchez
,
R.
Beanland
,
M.
Martin
,
T.
Baron
,
S.
Chen
,
A. J.
Seeds
,
I.
White
,
R.
Penty
, and
H.
Liu
,
J. Lightwave Technol.
38
(2), 240–248 (
2019
).
31.
D.
Jung
,
J.
Norman
,
M. J.
Kennedy
,
C.
Shang
,
B.
Shin
,
Y.
Wan
,
A. C.
Gossard
, and
J. E.
Bowers
,
Appl. Phys. Lett.
111
,
122107
(
2017
).
32.
J.
Selvidge
,
J.
Norman
,
M. E.
Salmon
,
E. T.
Hughes
,
J. E.
Bowers
,
R.
Herrick
, and
K.
Mukherjee
,
Appl. Phys. Lett.
115
,
131102
(
2019
).
33.
R.
Hull
and
J. C.
Bean
,
J. Vac. Sci. Technol. A
7
,
2580
(
1989
).
34.
J. A.
Olsen
,
E. L.
Hu
,
S. R.
Lee
,
I. J.
Fritz
,
A. J.
Howard
,
B. E.
Hammons
, and
J. Y.
Tsao
,
J. Appl. Phys.
79
,
3578
(
1996
).
35.
P.
Kightley
,
P. J.
Goodhew
,
R. R.
Bradley
, and
P. D.
Augustus
,
J. Cryst. Growth
112
,
359
(
1991
).
36.
M.
Uematsu
and
K.
Wada
,
Appl. Phys. Lett.
58
,
2015
(
1991
).
37.
R.
Beanland
,
A. M.
Sánchez
,
D.
Childs
,
K. M.
Groom
,
H. Y.
Liu
,
D. J.
Mowbray
, and
M.
Hopkinson
,
J. Appl. Phys.
103
,
014913
(
2008
).
38.
D. V.
Lang
,
P. M.
Petroff
,
R. A.
Logan
, and
W. D.
Johnston
,
Phys. Rev. Lett.
42
,
1353
(
1979
).
39.
O.
Ueda
,
Mater. Sci. Eng. B
20
,
9
(
1993
).
40.
P. M.
Petroff
and
L. C.
Kimerling
,
Appl. Phys. Lett.
29
,
461
(
1976
).
41.
G. A.
Baraff
and
M.
Schlüter
,
Phys. Rev. Lett.
55
,
1327
(
1985
).
42.
H.-P.
Komsa
and
A.
Pasquarello
,
J. Phys. Condens. Matter
24
,
045801
(
2012
).
43.
R. G.
Waters
and
D. S.
Hill
,
JEM
17
,
239
(
1988
).
44.
P. T.
Heald
,
Philos. Mag.
31
,
551
(
1975
).
45.
W.
Frank
and
U.
Gösele
,
Appl. Phys.
23
,
303
(
1980
).
46.
T. Y.
Tan
,
H.-M.
You
, and
U. M.
Gösele
,
Appl. Phys. A
56
,
249
(
1993
).
47.
S.
Yu
,
U. M.
Gösele
, and
T. Y.
Tan
,
J. Appl. Phys.
66
,
2952
(
1989
).
48.
J.
Gebauer
,
M.
Lausmann
,
F.
Redmann
,
R.
Krause-Rehberg
,
H. S.
Leipner
,
E. R.
Weber
, and
P.
Ebert
,
Phys. Rev. B
67
,
235207
(
2003
).
49.
J. R.
Arthur
,
J. Phys. Chem. Solids
28
,
2257
(
1967
).
50.
M.
Ichimura
,
T.
Wada
,
S.
Fujita
, and
S.
Fujita
,
Jpn. J. Appl. Phys.
30
,
3475
(
1991
).
51.
Z.
Bryan
,
I.
Bryan
,
B. E.
Gaddy
,
P.
Reddy
,
L.
Hussey
,
M.
Bobea
,
W.
Guo
,
M.
Hoffmann
,
R.
Kirste
,
J.
Tweedie
,
M.
Gerhold
,
D. L.
Irving
,
Z.
Sitar
, and
R.
Collazo
,
Appl. Phys. Lett.
105
,
222101
(
2014
).
52.
K.
Alberi
and
M. A.
Scarpulla
,
J. Appl. Phys.
123
,
185702
(
2018
).
53.
R. M.
Thomson
and
R. W.
Balluffi
,
J. Appl. Phys.
33
,
803
(
1962
).
54.
D.
Jung
,
R.
Herrick
,
J.
Norman
,
K.
Turnlund
,
C.
Jan
,
K.
Feng
,
A. C.
Gossard
, and
J. E.
Bowers
,
Appl. Phys. Lett.
112
,
153507
(
2018
).
55.
K.
Ishida
,
T.
Kamejima
,
Y.
Matsumoto
, and
K.
Endo
,
Appl. Phys. Lett.
40
,
16
(
1982
).
56.
J.-M.
Gerard
, and
C.
Weisbuch
, U.S. patent 5075742A (24 December 1991).
57.
S. Y.
Hu
,
S. W.
Corzine
,
K. K.
Law
,
D. B.
Young
,
A. C.
Gossard
,
L. A.
Coldren
, and
J. L.
Merz
,
J. Appl. Phys.
76
,
4479
(
1994
).
58.
A.
Lee
,
Q.
Jiang
,
M.
Tang
,
A.
Seeds
, and
H.
Liu
,
Opt. Express
20
,
22181
(
2012
).
59.
S.
Chen
,
W.
Li
,
J.
Wu
,
Q.
Jiang
,
M.
Tang
,
S.
Shutts
,
S. N.
Elliott
,
A.
Sobiesierski
,
A. J.
Seeds
,
I.
Ross
,
P. M.
Smowton
, and
H.
Liu
,
Nat. Photonics
10
,
307
(
2016
).
60.
C.
Hantschmann
,
Z.
Liu
,
M.
Tang
,
S.
Chen
,
A. J.
Seeds
,
H.
Liu
,
I.
White
, and
R.
Penty
,
J. Lightwave Technol.
(
2020
).
61.
R.
Beanland
,
J. P. R.
David
, and
A. M.
Sanchez
,
J. Appl. Phys.
104
,
123502
(
2008
).
62.
H.
Wang
,
A. A.
Hopgood
, and
G. I.
Ng
,
J. Appl. Phys.
81
,
3117
(
1997
).
63.
J. C.
Norman
,
Z.
Zhang
,
D.
Jung
,
C.
Shang
,
M.
Kennedy
,
M.
Dumont
,
R. W.
Herrick
,
A. C.
Gossard
, and
J. E.
Bowers
,
IEEE J. Quantum Electron.
55
,
1
(
2019
).