Impact of Pocket Geometry on Quantum Dot Lasers Grown on Silicon Wafers

Epitaxially grown quantum dot (QD) lasers in narrow pockets on patterned silicon photonics wafers present a key step toward full monolithic integration of on‐chip light sources. However, InAs QD lasers grown in deep and narrow pockets demonstrate limited performance and reliability compared to planar‐grown counterparts. Herein, InAs QD lasers are grown in patterned SiO2 pockets atop planar thermal cyclic annealed GaAs on (001) Si substrate with reduced threading dislocation density, enabling detailed study of how pocket geometry impacts device performance. Fabry–Pérot lasers with cleaved facets exhibit strong variation in performance based on the dimensions of the pocket, wherein thermal and optical metrics improve with increasing pocket width. Devices lase up to a maximum stage temperature of 115 °C with an extrapolated lifetime of 2.2 years at 80 °C for material grown in 50 μm by 3900 μm pockets. This study addresses, the ongoing challenge of optimizing pocket‐grown devices to planar equivalent performance.


Introduction
[9] Mismatched lattice constants between silicon and III-V materials, e.g., GaAs, result in threading dislocations (TD) that can penetrate all device layers, including the device active region.The tail ends of TDs pinned at the active region may glide under thermal strain induced during growth temperature changes, resulting in the formation of misfit dislocations (MDs) that act as nonradiative recombination centers. [10,11][14] Applying TCA to a GaAs virtual substrate buffer atop the Si(001) wafer causes TDs throughout the GaAs to thermally glide and annihilate, thus reducing the threading dislocation density (TDD) of the material. [15,16]Adding asymmetric step-graded InGaAs DFLs grown atop the TCA GaAs further reduces TDD by creating a strained barrier that diverts climbing TDs. [17,18]Lastly, TLs inserted on either side of the QD gain material displace thermal stress-generated MDs away from the active region, thus minimizing the impact of MDs as nonradiative recombination centers. [11]Using these methods, planar-grown InAs QD on unpatterned on-axis Si(001) substrate achieve record lasing temperatures up to 150 °C [19] and operational lifetimes up to 22 years at 80 °C. [14]ecent integration efforts have successfully coupled light from on-chip lasers into waveguides by patterning large-area, millimeterscale III-V windows onto a silicon-on-insulator wafer containing Si or Si 3 N 4 waveguides. [2,3]An alternate integration strategy places the epitaxial III-V material (epi) inside narrow rectangular SiO 2 pockets of millimeter-scale length by under 50 μm width.Figure 1 illustrates planar growth on an unpatterned wafer (Figure 1a), as compared to pocket growth on pre-patterned template (Figure 1b).Previously, cleaved facet Fabry-Pérot InAs QD lasers were successfully demonstrated in narrow SiO 2 pockets on a 300 mm Si(001) patterned wafer. [4,5]However, these narrow pocket InAs QD lasers on patterned Si(001) have limited performance compared to planar InAs QD lasers grown on unpatterned Si(001).
Epitaxial growth in narrow pockets invokes subtleties not evident in planar or large-area window growth. [4,6]In situ monitoring becomes challenging when the majority of the sample surface is amorphous SiO 2 instead of crystalline III-V.QD morphology changes based on the crystallographic orientation and the width of the pocket due to faster adatom diffusion along the [110] direction. [6]Previous narrow pocket growth on Si(001) template also did not use TCA, resulting in a TDD of 1.5 Â 10 7 cm À2 , [4] an order of magnitude larger than the TDD of the longest lifetime planar lasers. [14]TCA becomes less practical in pocket-only growth, as by the Kelvin effect, higher vapor pressure at the convex corners of the pocket leads to increased evaporation and requires higher V/III overpressure to preserve the material quality during the anneal. [20]This creates ambiguity about whether the limited performance of previous narrow pocket devices arises from elevated TDD for lack of TCA, or whether it is an intrinsic shortcoming of narrow pocket growth.
In this work, we introduce the "semi-pocket" device platform, where InAs QD lasers are grown in pockets atop planar annealed GaAs/GaP/Si to reduce device TDD and determine whether pocket geometry is a critical contributor to laser performance and reliability.First, a planar TCA GaAs virtual substrate is grown atop commercial GaP/Si(001).Narrow SiO 2 pockets of 3900 μm length by various widths from 10 to 50 μm are defined above the planar annealed virtual substrate.This range provides a much broader set of pocket dimensions for characterization and device fabrication than previously studied.InAs QD laser epi is grown inside these pockets, as seen in Figure 1c with the detailed epitaxial structure in Figure 1e.The annealed planar GaAs layer provides an epitaxially smooth surface with equal reduction of TDD across all areas of the sample prior to pocket definition.Thus, the semi-pocket platform eliminates ambiguity as to whether pocket device performance trends arise from fundamental material properties or from template fabrication, e.g., etch damage in direct growth on the recessed silicon pocket-only platforms previously studied.Growth conditions used in this experiment are chosen such that 30 μm pockets have the best photoluminescence (PL) achievable for a 30 μm pocket, as optimized following the method of Shang et al. [4,6] However, we still find that despite this optimization, the wider 50 μm pockets perform better than 30 μm pockets in PL brightness and narrowness, and in laser device metrics.This indicates that narrower pockets are inherently less favorable than wider pockets.The as-grown sample is fabricated into cleaved facet Fabry-Pérot lasers with topside electrical contacts (Figure 1d).The semi-pocket platform achieves 2Â TDD reduction with use of the planar annealed virtual substrate, but, unexpectedly, device optical performance is still comparable to non-annealed counterparts of the same pocket dimensions.This similarity illustrates the key role of pocket geometry in device quality.Here, it is found that device performance metrics exhibit a remarkable dependence on pocket geometry: QD PL, laser thermal tolerance, and laser lifetime all improve with increasing pocket width.This is to date the first report of how pocket size impacts laser thermal and reliability characteristics.

Pocket-Grown Material
The insertion of a planar TCA GaAs virtual substrate beneath the patterned SiO 2 in the semi-pocket template aims to reduce TDD compared to the non-annealed, fully pocket-grown scenario.In planar III-V on Si growth, a 2.55 μm GaAs buffer with 1.6 μm TCA GaAs yields TDD of 3 Â 10 7 cm À2 , and adding In x Ga 1Àx As DFLs above the TCA can reduce TDD to as low as 1 Â 10 6 cm À2 . [17,18]Laser epi grown completely in-pocket yields 1.5 Â 10 7 cm À2 TDD. [4]For semi-pocket growth, electron channeling contrast imaging of a test sample with planar TCA GaAs and in-pocket DFLs indicates TDD of 7 Â 10 6 cm À2 for this structure.The difference from planar TDD may arise from in-pocket growth temperature uncertainty or an imperfect planar-to-pocket regrowth interface after SiO 2 template deposition.Optimization of pocket-specific growth conditions and template processing is needed for further TDD reduction both semi-pocket and full pocket platforms.
At the device active region, pinned TDs under thermal stress can glide to generate MDs that act as nonradiative recombination centers, harming device performance. [21]To determine the MD density (MDD) affecting the active region, scanning-tunneling electron microscopy (STEM) foils are taken across or along pockets, shown schematically in Figure 2a.Note that these imaging conditions highlight MDs running perpendicular to the STEM foil in which they are observed; for example, the along-pocket foil in Figure 2b shows MDs running perpendicular to the pocket as a result of elevated thermal stress parallel to the pocket.The QDs in the active region are examined at 7°tilt along (220).Figure 2c shows a known high-performing 50 μm [110]-oriented pocket laser under STEM, with along-pocket and across-pocket closeups shown in Figure 2b,d.Counting MDs trapped in the TLs as well as inside the QD active region, this particular device has MDD of 1.85 μm À1 aligned with the [110] direction, perpendicular to the pocket (seen on-end as dark contrast points in the "Along" foil; Figure 2b); and MDD of 0.25 μm À1 aligned with [110], parallel to the pocket ("Across" foil; Figure 2d).Comparing this result with previous observations of planar material MDD, planar laser epi from a sample with over 3Â higher TDD than the semi-pocket growth reported here (planar TDD: 2-4 Â 10 7 cm À2 ) shows MDD of 1.8 μm À1 aligned with [110], and 2.6 μm À1 aligned with [110]. [6,22]This crystallographicorientation-based asymmetry in MD presence is inherent to laser growth independent of using planar, full pocket, or semi-pocket growth platforms.In this study, semi-pocket-grown epi has comparable MDD to that planar sample for [110]-aligned MDs, and heavily reduced MDD for [110]-aligned MDs.Reduced [110]  MDD running parallel to the pocket indicates reduced thermal stress across the narrower [110] pocket-perpendicular direction during molecular beam epitaxy (MBE) growth, which is consistent with previous geometry-based stress calculations and MD observation in rectangular pocket growth. [6]Of the observed MDs, 95% occur at the TLs, indicating strong trapping efficiency where the TLs successfully protect the active region from MD generation.Counting only MDs observed within the QD layers, MDD is 0.06 μm À1 running perpendicular to the pocket, with negligible MD presence running parallel to the pocket.to [110], with more prominent orientation-dependent divergence in smaller pockets.

Device Results
Devices fabricated within pockets of 20-50 μm show distinct lasing peaks when electrically pumped.Figure 3d shows the subthreshold net modal gain and GS optical emission spectrum at 20 °C of an electrically pumped 4 μm by 2300 μm device fabricated inside a 30 μm [110] pocket.Both gain and GS emission peaks fall within the telecom O band.Although these devices are designed as standard edge-emitting Fabry-Pérot cavities, without any grating or intentional mode-selective construction, this 30 μm pocket laser and other devices show peculiar triple peak GS emission regardless of pocket geometry.As can be seen in the PL spectra of Figure 3a,b, a slight multi-peak ripple also appears superimposed over the distinct GS and excited state (ES) emission peaks expected from a uniform QD distribution.This suggests that the multi-peak nature of the lasing spectra arises from intrinsic variation in the QDs, rather than a device-specific or fabrication-induced issue.The three peaks observed in the lasing spectra may be related to temperature control during MBE growth: if QD morphology differs in each period due to a variation in growth temperature across the five QD self-assembly steps, some sets of QDs may emit at different wavelengths when electrically pumped.This phenomenon does appear in previously reported spectra of narrow pocket-grown cleaved facet Fabry-Pérot lasers, but it has not been studied in depth. [4,5]urther investigation will be needed to determine the exact mechanism behind the multi-peak appearance of both PL and lasing emissions.
Figure 4a,b shows light-current-voltage (L-I-V ) measurements of devices with mesa dimensions 4 μm by 2300 μm, fabricated in adjacent 30 and 50 μm pockets.The devices are tested under continuous wave (CW) operation on a temperaturecontrolled stage.Note the comparable I-V characteristics of both devices shown.Throughout the sample set, minimal I-V variation occurs between devices of [110] versus [110] orientations across 20-50 μm pocket widths, indicating similar resistive heating across these devices during operation.The 30 μm pocket yields a maximum GS lasing temperature of 65 °C.This is consistent with previously reported pocket-grown lasers of 20-30 μm by 2000 μm pocket dimensions with 4 μm device mesa, which support CW lasing up to 60 °C. [4,5]The 50 μm pocket vastly exceeds this performance with GS lasing up to 95 °C and ES lasing to 115 °C.In the inset in Figure 4b, the 50 μm device also shows superior thermal stability (T 0 ) of threshold current across both low-temperature (20-55 °C) and high-temperature (over 60 °C) regimes.
Overall, device thermal stability and maximum GS lasing temperature increase linearly as pockets widen (Figure 4c).The variation is largely due to the QD density variation among the different pocket sizes, although the QD density is challenging to measure within the InGaAs-capped InAs QD layers found in a full laser stack.In past characterization of uncapped InAs QD test structures grown in narrow rectangular pockets on patterned SiO 2 /GaAs substrate, QD density increased approximately linearly with pocket width. [6]Since QD density is proportional to optical gain, and the maximum lasing temperature a device can sustain increases monotonically with gain, the linear trend between QD density and pocket width explains why wider pockets have better thermal performance in Figure 4c.Growth in these wider 50 μm pockets leads one step closer to bridging the performance gap between pocket lasers and pure planar-grown devices.For comparison, the highest reported CW lasing temperature of planar MBE-grown InAs QD lasers on Si is 150 °C, achieved by using an epitaxial structure with eight QD periods instead of five. [19]Improving the thermal performance of pocket InAs QD lasers on Si to match planar performance should be possible by redesigning the pocket epi and reoptimizing pocket MBE growth conditions to permit more QD layers for enhanced optical gain.

Reliability
Beyond as-fabricated performance characterization, select devices underwent continuous operation for extended time to evaluate their lifetime and degradation profiles.Bars of 4 μm by 1500 μm devices are electrically pumped at a nominal bias of twice threshold current in an 80 °C temperature-controlled environment.The test pauses every 15 h of continuous operation to evaluate L-I-V characteristics of each device at a reference temperature of 35 °C. Figure 4d summarizes 750 h of device aging at 80 °C.The threshold current (I th ) of wider pocket devices increases more slowly than narrow pocket devices, indicating reduced degradation in wider pockets.Devices in equal width pockets of [110] or [110] orientation experience a similar percentage increase in I th overtime.Let device lifetime be defined as the time needed for I th at 35 °C to double from its initial value.The 20 μm pocket device has 600 h extrapolated lifetime, while the 50 μm pocket has 19 000 h (2.2 years) extrapolated lifetime.Lifetimes are calculated using a power fit of the form a Â t b for time t and fit parameters a, b. [13] Note the elevation in I th of the 20 μm device at the beginning of aging compared to the wider pocket devices is due to prolonged device burn-in within the first 15 h, but after the first 1 h conventional burnin period, of aging.
On planar-grown InAs QD lasers aged at twice-threshold bias at 80 °C, the extrapolated lifetime for I th doubling in devices with an expected TDD of 7 Â 10 6 cm À2 is 10 years. [14]Hence, the pocket lifetime is unexpectedly low compared to planar devices with equivalent TDD.Since dislocation presence near the active region is a strong predictor of device lifetime in planar devices, [13,14] one possibility is that in-pocket TLs or DFLs are less effective or less well calibrated due to the challenges of pocket MBE growth monitoring.However, this sample has roughly 2Â less TDD from the annealed virtual substrate compared to previous pocket lasers that demonstrate over 50 000 h (5.7 years) lifetime at 35 °C. [4,6]Note that previous pocket studies have not aged devices at 80 °C for an exact comparison.Elevated temperature causes faster device degradation; for instance, aging at 60 °C shortens lifetime by over three orders of magnitude compared to 35 °C. [13]he relatively short lifetime and prominent lifetime dependence on pocket width may arise from nonradiative recombination-enhanced degradation mechanisms.Nonradiative recombination produces local heat and lattice vibration, which enhance dislocation climb and point defect generation around existing recombination centers. [23,24]Indeed, previous studies of GaAs heterostructure and InAs QD lasers attribute the emergence of dark line defects in the device active region to dislocation growth after extended operation. [24,25]As seen in Figure 4c, as-fabricated devices in narrower pockets have a lower maximum lasing temperature than wide pocket devices, likely due to reduced gain from lower QD density when growing in narrow pockets.For instance, the 50 μm pocket device lases at roughly half its room-temperature output power at 80 °C, whereas the 30 μm pocket device effectively becomes a light-emitting diode (LED) with no coherent mode at 80 °C.If the same injection current is applied to both devices, the narrower pocket LED experiences a higher proportion of nonradiative recombination than the wider pocket laser.Narrower pocket devices are therefore more susceptible to recombination-enhanced degradation.When aged at elevated temperature and continuous injection current, this susceptibility manifests as a faster rise in narrow pocket I th compared to wider pocket devices under the same test conditions.

Conclusion
In this article, the semi-pocket growth platform has been used to analyze the defect dependence and pocket geometry dependence of MBE-grown InAs QD lasers on patterned SiO 2 on GaAs on silicon.Wider pockets result in enhanced optical and thermal performance, whereas narrower pockets perform more poorly despite having consistent electrical characteristics.Pocket width displays a much more pronounced effect on device reliability than crystallographic orientation.The planar annealed GaAs layer between the silicon substrate and patterned SiO 2 pockets indeed reduces the presence of harmful defects compared to fully pocket-grown counterparts, as observed through the lower TDD below the laser stack and lower MDD near the device active region.The defect reduction achieved in the semi-pocket platform brings patterned template-grown laser performance one step closer to that of state-of-the-art planar-grown InAs QD lasers.With further refinement of pocket template layout and in-pocket MBE growth conditions, additional enhancement of narrow pocket device performance is within reach.

Experimental Section
Planar Growth: The initial substrate is a commercially available on-axis Si(001) 300 mm wafer with 200 nm GaP grown through metal-organic chemical vapor deposition at NAsP III-V GmbH. [26]The GaP crystal orientation with respect to the Si wafer notch was determined by manufacturer surface treatment and step edge distribution on the original Si(001) wafer.The exact GaP crystal orientation was confirmed by inspecting InGaAs/GaAs test samples for crack and faceted trench formation, which preferentially occurred parallel to the ½110 direction on highly strained films. [4,18]A planar 1.5 μm GaAs virtual substrate was grown atop the GaP/Si(001) sample with MBE.The GaAs film underwent 16 cycles of TCA between 400 and 700 °C, which reduced TDD to 3 Â 10 7 cm À2 . [18]ocket Patterning: Atop the TCA GaAs buffer, 4 μm of SiO 2 was deposited and patterned into narrow rectangles of 3.9 mm length and 10-50 μm width.The pockets were oriented along the substrate [110] or [110] crystallographic orientations and comprised less than 10% of total sample surface area.The SiO 2 was dry etched with CHF 3 =CF 4 chemistry to define recessed pockets.O 2 plasma and a buffered hydrofluoric acid (HF) wet etch were used to remove any remaining in-pocket SiO 2 to expose epiready GaAs surfaces suitable for MBE growth.
Pocket Growth: The patterned surface complicated temperature control and in situ MBE growth monitoring.Emissivity shifts between SiO 2 , crystalline III-V, and polycrystalline III-V formed atop the SiO 2 during growth could inhibit pyrometer accuracy.Due to the majority SiO 2 surface, refraction high-energy electron diffraction could not be used to confirm oxide desorption, surface reconstruction, or QD formation.Here, growth temperatures were estimated by applying an experimentally determined offset to pyrometer readouts of the SiO 2 surface, using the method and growth conditions previously optimized for full pocket growth in 30 μm pockets in ref. [4].After a standard pre-growth 420 °C hydrogen clean and native oxide desorb, MBE growth began with asymmetric graded In x Ga 1Àx As DFLs: 150 nm at x ¼ 0.05, 200 nm at x ¼ 0.10, and 300 nm at x ¼ 0.05, followed by the laser stack.The laser active region contained five periods of InAs QDs embedded in 7 nm In 0.15 Ga 0.85 As quantum well (QW), spaced apart with p-modulation-doped barriers at a level of five holes per dot.In 0.15 Ga 0.85 As QW TLs were inserted on either side of the active region to discourage MD formation near the QDs. [11]The active region was surrounded on either side by 1400 nm of n-or p-Al 0.4 Ga 0.6 As, followed by respective n-or p-GaAs contact layers.Figure 1e shows the complete epitaxial structure.It was noted that the semi-pocket platform used here only performed in-pocket growth after the formation of the initial GaP film and 1500 nm GaAs buffer, resulting in an overall thinner in-pocket epi and out-of-pocket polycrystalline III-V layer than if the buffer had been grown fully in pocket.
Device Fabrication and Testing: Polycrystalline III-V deposited atop the SiO 2 template was removed with a photoresist mask and H 3 PO 4 -based wet etch, leaving behind only the epitaxial pocket-grown III-V material.Polycrystal removal decreased height variation in the pocket sample surface to enable more consistent photolithography and reduce the risk of contact metal gaps on abrupt height transitions near the pocket edges.Next, Pd/Ti/Pd/Au p-contact metal was deposited directly on the p-GaAs layer.Rectangular 4 μm by 3900 μm mesas were dry etched with Cl 2 =N 2 chemistry straight down to the n-GaAs contact layer (Figure 1d  and 2c).The device mesa was centered along the width of the pocket because previous studies have observed considerable nonuniformity in across-pocket QD density within a few micrometers of pocket edges. [6]idewalls are passivated with Al 2 O 3 and SiO 2 .Vias were opened to deposit Pd/Ge/Au n-contact metal on the n-GaAs layer, followed by a 350 °C rapid thermal anneal.Lastly, vias were opened and Ti/Au probe metal was deposited for electrical probe contact.The completed sample was thinned and cleaved into bars, resulting in cleaved facet Fabry-Pérot lasers of cavity length 1500 or 2300 μm.Devices were tested under electrical injection on a temperature-controlled stage.Light was collected with an integration sphere or lens fiber.For long-term aging tests, devices were operated under constant injection current in an 80 °C aging rack, and L-I-V characteristics were periodically collected to evaluate device degradation overtime.

Figure 1 .
Figure 1.a) Schematic of planar III-V laser grown on Si. b) Fully pocket-grown laser on Si. c) Semi-pocket growth on Si combines planar annealed virtual substrate with pocket growth of III-V device epi.d) Scanning electron microscopy (SEM) cross section of a 50 μm semi-pocket epi fabricated into an edge emitting cleaved facet Fabry-Pérot laser.e) Detailed structure of the planar and pocket epitaxial stacks.

Figure
Figure 3a,b shows the room-temperature PL of as-grown laser epi for 3900 μm long pockets of varied width from 10 to 50 μm across the same chip, as summarized in Figure 3c.The center wavelength of ground state (GS) emission redshifts as the pocket widens, following an approximately exponential relation (a Â expðb Â wÞ þ c Â expðd Â wÞ given numerical fit parameters a, b, c, d, with jdj < 0:02 and width w) indicated by dashed fit lines drawn as guides to the eye.For each pocket width of 20 μm and up, the GS wavelength falls within the telecommunications O band.GS full width at half maximum (FWHM) also varies exponentially by pocket width, with wider pockets showing smaller FWHM and narrower PL spectra.Pockets oriented with the long axis parallel to the [110] crystallographic direction have smaller FWHM and redshifted GS center wavelengths compared

Figure 2 .
Figure 2. a) Pocket schematic with scanning-tunneling electron microscope (STEM) foil orientations for misfit dislocation density (MDD) characterization.b) STEM along the ridge of a known working laser in a 50 μm [110] pocket.Imaging when tilted at 7°along (220) shows MDs at the top and bottom trapping layers (TL), indicated by arrows.c) STEM foil taken across the pocket and device ridge.d) STEM across the pocket tilted along (220) shows lower MDD than (b).

Figure 3 .
Figure 3. a) Photoluminescence (PL) spectra taken at 293 K at the center of [110]-oriented pockets of 10-50 μm width.b) PL at 293 K for [110]-oriented pockets.c) PL peak wavelength and full width at half maximum (FWHM) as a function of pocket width.Dashed lines are exponential fits drawn as guides to the eye.d) Ground-state (GS) lasing spectrum for a 30 μm [110]-oriented pocket laser under electrical injection, and corresponding net modal gain calculated from subthreshold emission.

Figure 4 .
Figure 4. a) T-dependent light-current-voltage (L-I-V ) measurements starting from 20 °C in increments of 5 °C, with inset threshold current I th and thermal tolerance parameter T 0 fits for device of 2300 μm by 4 μm mesa grown inside a 30 μm pocket.b) T-dependent L-I-V with inset T 0 fits for an adjacent 50 μm device of the same dimensions and laser bar as (a).c) Device thermal performance by pocket width: maximum lasing temperature of the GS (T max ,GS ) and T 0 .d) Shift in threshold current measured at 35 °C for various pocket sizes aged at 750 h and 80 °C under twice-threshold bias current.Dashed lines are power fits of the form a Â t b for time t.