Lasing from Glassy Ge Quantum Dots in Crystalline Si

Semiconductor light-emitters compatible with standard Si integration technology (SIT) are of particular interest for overcoming limitations in the operating speed of microelectronic devices. Light sources based on group IV elements would be SIT-compatible, but suffer from the poor optoelectronic properties of bulk Si and Ge. Here we demonstrate that epitaxially grown Ge quantum dots (QDs) in a defect-free Si matrix show extraordinary optical properties if partially amorphized by Ge-ion bombardment (GIB). In contrast to conventional SiGe nanostructures, these QDs exhibit dramatically shortened carrier lifetimes and negligible thermal quenching of the photoluminescence (PL) up to room temperature. Microdisk resonators with embedded GIB-QDs exhibit threshold behavior as well as a superlinear increase of the integrated PL intensity with concomitant line width narrowing as the pump power increases. These findings demonstrate light amplification by stimulated emission in a fully SIT-compatible group IV nanosystem.

S ilicon micro-and nanophotonics is a field of tremendous applied and basic research interest: it connects Si photonics with Si-based microelectronics and aims at the incorporation of optical functionality into integrated circuits. Widespread applications range from power electronics, and sensors all the way toward improving on-chip data communication and processing by using guided light for data transfer instead of copper wires. 1−3 The main problem arises from the poor light emission from Si and Ge, which results from the indirect bandgap of crystalline group IV materials. Recent advances in Si photonics include demonstrations of highly doped or laserannealed Si light-emitting diodes, 4,5 an electrically pumped Ge laser, 6 group III−V QD lasers that are either grown epitaxially 7 or bonded onto Si substrates 8 and PL from strained Ge. 9−11 Most recently, a SnGe laser grown on Si was demonstrated 12 to operate up to 90 K, but shows thermal quenching by about 2 orders of magnitude between 20 and 300 K. 12 Utilizing group IV nanostructures in Si would eliminate the growth of thick, dislocation-rich GaAs or SiGe buffer layers, necessary for some of the approaches described above. The development of crystalline Ge-QDs on Si (Ge/Si-QDs) grown by strain-driven self-organization 13−20 gave hope that quantumconfinement effects in group IV nanostructures could be used to produce high-performance optical devices. However, their optical properties never matched expectations. Because of the relatively small lattice mismatch between Si and Ge (∼4%), the resulting QDs become at least in one spatial direction larger than several tens of nm. Thus, their carrier wave functions have one-or two-dimensional (1D, 2D) character rather than the zero-dimensional (0D) one aimed at. Another drawback is the spatially indirect recombination path in Ge/Si-QDs due to a type-II band alignment, 21 where only holes are confined in the Ge-QDs. Consequently, efficient room-temperature PL from crystalline Ge/Si-QDs was never observed. 14,15,[18][19][20]22 Because of strong confinement in all three spatial dimensions as well as surface termination effects, porous silicon, and Si-and Genanocrystals (Si-NCs and Ge-NCs) show significantly better optical properties at room temperature (RT). 23−28 Nevertheless, no RT continuous-wave lasing was reported so far.
In this work, we close the gap between SIT-compatible Ge/ Si-QDs 29 and porous Si/Si-NCs with their superior PL properties by bombarding self-organized Ge/Si-QDs during epitaxial growth with Ge ions. This leads to partially amorphized Ge-QDs embedded in a dislocation-free Si matrix. The nucleation of strain-driven Ge/Si-QDs 13−18 occurs on a supersaturated Ge wetting layer (WL), once a critical thickness of ∼4.2 monolayers (∼0.6 nm) is exceeded. 30 Deposition of 0.7 nm of Ge at 500°C leads to small hut-shaped QDs 13 with a very narrow height distribution of 1.95 ± 0.29 nm (see Supporting Information). The dot-density can be tuned between 2 × 10 10 cm −2 ( Figure 1A) and 2 × 10 11 cm −2 ( Figure 1B) by varying the Ge coverage. During Ge deposition, the sample is bombarded by positively charged Ge ions (dose: ∼10 4 μm −2 ) that are accelerated by voltages V GIB of down to −2.8 kV. The high-resolution transmission electron microscopy (TEM) image in Figure 1C reveals a partly crystalline, partly amorphized GIB-QD embedded in a crystalline Si matrix. Amorphization results in a blurring of the {111}-crystal-lattice fringes due to displaced atomic positions, in contrast to the regular atomic arrangement in the Si layer below and above the GIB-QD.
The underlying mechanism for partial amorphization is based on a collision-cascade of elliptical shape consisting of matrix atoms that become displaced by impact of either the original Ge ion or a recoil atom with an energy exceeding the displacement energy of ∼14 eV 31,32 ( Figure 1D). The cascade of a single Ge ion (≤3 keV) produces one small, a few nm wide amorphous zone. During Ge deposition and ramp-down to the growth-temperature of the Si capping layer, vertical and lateral solid-phase epitaxial regrowth (SPER) takes place that leads to a recrystallization of the WL and parts of the Ge islands 33−35 ( Figure 1D). This allows for partial overgrowth with a fully crystalline Si cap layer ( Figure 1C). In this way, the original hut-cluster becomes divided into small crystalline Ge regions with diameters <4 nm separated by glassy Ge regions with higher energy gap. 36 In Figure 2, we discuss the main PL features of such GIB-QDs. Figure 2A shows excitation-power (P exc ) dependent PL spectra, normalized to the intensity maximum of the GIB-QD-PL, as obtained at a sample temperature T PL of 6 K. An increase of P exc from 1.5 to 1600 μW shifts the onset of the PL from 1450 to 1250 nm, whereas the intensity-maximum of the GIB-QD-PL moves from 1560 to 1340 nm. The PL spectra were fitted using four Gaussians with maxima at (i) 1552, (ii) 1488,

ACS Photonics
Article (iii) 1420, and (iv) 1352 nm and a common full-width-at-halfmaximum (fwhm) of 70 nm, as shown in the inset of Figure 2B for P exc = 1070 μW. In Figure 2B, the integrated PL intensity, I PL , of the GIB-QDs as well as the sum of the two Gaussians with longer and shorter wavelength (i) + (ii) and (iii) + (iv), respectively, are plotted versus P exc . Power laws of the form I PL (P exc ) ∼ P exc m are found for the two sums, the former (i) + (ii) with m ≈ 0.6, the latter (iii) + (iv) with m ≈ 1. In Figure  2C, we present PL spectra (P exc = 1600 μW) obtained at 20, 80, 197, and 300 K. The spectra are to scale, that is, the PL intensity hardly decreases up to 300 K. At higher T PL , the onset of the GIB-QD-PL shifts to about 1200 nm. The temperaturedependent behavior of the GIB-PL for P exc of 1600, 430, and 136 μW is depicted in the Arrhenius plots in Figure 2D. In the inset the data are plotted on a double logarithmic scale to emphasize the PL quenching at high T PL . The solid red lines are fits to the data corresponding to activation energies E A of ∼350 meV for both P exc = 136 and 430 μW. Details about the calculations of the activation energies are presented in the Supporting Information. The open blue circles in the inset of Figure 2D depict the quenching behavior (P exc = 430 μW) of Ge-QDs that were grown in the same way as the GIB-QDs, but under V GIB = 0 kV. Note the increase of 2 orders of magnitude of the GIB-QD PL intensity as compared to the one of the Ge-QDs at 200 K (see blue-dashed arrow in the inset of Figure  2D). For the latter, no PL is observed at RT, in agreement with previous reports in literature. 14,15,22,37−39 We attribute the pronounced shift of the GIB-QD-PL to shorter wavelengths for increasing P exc to a combination of (i) progressive filling of smaller QDs, created by the Ge ion bombardment, that, due to higher confinement energies, exhibit higher PL transition energies, (ii) state-filling within the GIB-QDs, and (iii) photoinduced band-bending effects. 22,37 In crystalline epitaxial SiGe-QDs electrons and holes are spatially separated and band filling 37 of about 100 meV is not uncommon, even for larger QDs than the ones investigated in this work, see ref 38 and the Supporting Information. In order to estimate which effect dominates, more investigations would be required, which is beyond the scope of this paper.
For very low P exc , only ground states of the largest GIB-QDs are filled. For epitaxial Ge/Si-QDs, I PL usually increases with P exc , following a sublinear power law with m ∼ 0.6, caused by Auger-recombination. 39 A power law with m = 1, as found here for the shorter wavelength part of the GIB-QD-PL spectrum, is usually observed in direct-gap semiconductor QDs, for example, in the III−V material system. 40 Thermal quenching with E A ∼ 60−80 meV was reported for epitaxial Ge/Si-QDs (see inset of Figure 2D), 22,39 in sharp contrast to E A ≈ 350 meV found here. We tentatively ascribe this high E A to electron-occupancy of deep levels in the glassy region of GIB-QDs (see discussion of Figure 3), in analogy to deep levels caused by dangling bonds (∼250−350 meV) in Si. 41 From the discussion of Figure 2, we conclude that small GIB-QDs exhibit a different band structure scheme than the indirect and type-II band alignment of crystalline QDs. This assignment is supported by (i) the observed power law of m = 1, (ii) the high E A in combination with the temperature stability of the GIB-QD-PL, and (iii) the fact that the k-selection rule of the indirect band gap will be softened due to Heisenberg's uncertainty principle because of strong localization of both the electron at the deep level and the holes in the GIB-QDs with diameters <∼4 nm. Further support stems (iv) from the strong decrease of the average PL-decay lifetime, τ av , with decreasing GIB-PL-emission wavelength λ PL . Figure 3A depicts temperature-and wavelength-dependent time-decay PL spectra. For low T PL < 180 K, τ av is strongly wavelength-dependent: it amounts to several hundreds of ns for λ PL = 1565 nm and decreases to about 600 ps for λ PL = 1330 nm. For high T PL >

ACS Photonics
Article 300 K, τ av converges to a level of about 1−2 ns, independent of the wavelength. This is consistent with the onset of thermal quenching of I PL under pulsed excitation ( Figure 3A, second ordinate), attributed to the increased role of nonradiative recombination processes. This is further supported by the findings of Figure 3B. There, the logarithm of the half-width at half-maximum (hwhm) of the fitted log-normal distribution functions is plotted, which gives a qualitative measure of the amount of different transition processes participating in the PL time decay. 42 A value of 0 implies a single exponential decay, higher values for hwhm imply that higher number of transition processes have to be taken into account. For T PL > 250 K, the hwhm starts to converge to a level below 1, owed to nonradiative recombination processes. Figure 3C,D depicts selected time-decay spectra and fits using the log-normal distribution function for T PL = 6 and 315 K and λ PL of 1565, 1445, and 1330 nm. The short PL decay time of about 600 ps at 1330 nm for T PL < 180 K ( Figure 3A) is significantly faster than the ∼20 ns observed for Ge/Si-QD-multilayers. 19 The short τ av is also consistent with the fact that the PL signal does not saturate, even in the available P exc range ( Figure 2B), that is, radiative recombination can compete with Auger-recombination.
To assess the potential of the GIB-QDs for lasers, we embedded them into microdisk resonators 43 of 1.8 μm diameter. The excited resonant modes are whispering-gallery modes (WGM) running around the disk's circumference and radial Fabry-Peŕot modes (FPM) across the disks (see inset in Figure 4A). The former appear as sharp emission lines, and the latter as broader peaks. Figure 4A displays PL spectra of a microdisk for T PL = 10 K and RT, and PL spectra for increasing P exc at T PL = 10 K. In the following, we will focus on the emergence of the Lorentzian-shaped transversal electrical mode TE (12,1) emitting at about 1323 nm. In Figure 4B, its integrated PL intensity, I WGM , is plotted versus P exc on a doublelogarithmic scale, and, in the inset, on a double linear scale. The P exc dependence displays threshold behavior at about 100 μW. For higher P exc , I WGM tends to saturate toward m = 1 ( Figure  4B). Both the threshold behavior and the s-shaped I PL curves are indicative of stimulated emission and can be observed up to RT. However, due to the strong filling effects of smaller GIB-QDs with higher P exc , a natural threshold behavior of the cavitymode emitting at 1323 nm is also expected (see Figure 2A). To unfold the influence of GIB-QD-filling on the threshold behavior of the TE(12,1) mode, we plotted in the inset of Figure 4B the ratio I WGM /I FPM versus P exc . Here, I FPM stands for the I PL of the FPM and both I WGM and I FPM emit at 1323 nm, but couple to different GIB-QDs. Thus, if the threshold behavior would be caused by P exc -driven filling only, we would expect I WGM /I FPM to remain constant versus P exc . However, at the threshold, a distinct increase of I WGM /I FPM from 3 to about 11 is observed. Finally, we observe also line-width-narrowing (see Supporting Information) of the emission mode, which is expected for stimulated emission.
Based on these results, we expect that GIB-QDs will bridge the gap between epitaxial group IV QDs and Si-NC systems to open new paths for Si photonics based on group IV nanostructures. GIB-QDs with their separated glassy and crystalline Ge regions within an original Ge/Si-QD, are small enough to exhibit 0D-quantum confinement, similar to that of porous Si and Si-NCs. As the GIB-QDs are fully compatible with standard Si technologies, they can be monolithically integrated as electrically pumped light sources into an environment of highly complex devices together with other passive optoelectronic components based on Si.
■ METHODS Sample Growth. Although, in this work, the samples were grown by molecular beam epitaxy (MBE) in combination with in situ low energy Ge-ion bombardment, all fabrication steps can, in principle, be performed with SIT-compatible fabrication methods. Ge quantum dots can be grown by CVD, 16 and ion implantation and annealing are standard procedures of Si device technology.
Here, the growth of GIB-QDs was carried out in a Riber SIVA45 solid-source MBE system with electron-beam evaporators for Si and Ge. We use buried-oxide SOI substrates with a SiO 2 thickness of 2 μm and a high-quality Si(001) top layer with 160 nm thickness. After ex situ sample cleaning, the natural oxide was desorbed in situ at 950°C for 15 min. Thereafter, a 40 nm thick Si buffer layer was grown at a temperature that was ramped-down from 550 to 500°C. A single Ge layer was grown at 500°C with coverages of either 7 or 8.4 Å. QDs grown at such low temperatures are referred to as hut cluster or huts. 13 They are confined by 11.3°steep {105}-facets, which, due to kinetic reasons, 13 are elongated with

ACS Photonics
Article a rectangular base and, thus, they resemble huts (see Figure  1A,B).
During growth, a small fraction of the evaporated Ge atoms is ionized as they pass through the electron beam of the evaporator. Those Ge ions (dose ∼10 4 μm −2 , that is, about one ion per area of 10 × 10 nm 2 ) are then accelerated toward the substrate that is biased by an adjustable substrate bias V GIB between 0 and −2.8 keV. The crystalline reference QDs were grown in the same manner but under V GIB = 0 kV.
Finally, the GIB-QDs were embedded into a Si matrix by capping with Si. For this purpose, the substrate temperature is ramped down from the 500°C used for Ge deposition to 350°C for Si-cap deposition to preserve the shape and composition of the small QDs and the WL. 44 During ramp-down, the topmost part of the GIB-QDs recrystallizes laterally via solidphase epitaxial regrowth, which then allows for overgrowth with a fully crystalline Si capping layer (see Figure 1C and, for the WL, the Supporting Information).
Optical and Structural Investigations. The surface topography was analyzed by atomic force microscopy (AFM) using a Digital Instruments Dimension 3100 AFM. To get insight into the structural properties of the GIB dots, we performed cross-sectional transmission electron microscopy using a JEOL JEM-2011 FasTEM instrument operated at 200 kV. The cross-sectional lamellae were cut by a focused ion beam using a ZEISS 1540XB CrossBeam facility.
For PL experiments, we used an excitation diode laser operated at 442 nm and a microscope objective with a numerical aperture of 0.7, which is used both for laser focusing and for collecting the PL signal from the sample. A continuous flow cryostat allows for sample cooling down to liquid He temperature. The laser spot diameter on the sample was ∼2 μm. The signal is dispersed by a grating spectrometer and recorded by a nitrogen-cooled InGaAs line detector.
For time-resolved measurements the samples were excited by a pulsed laser (wavelength of 442 nm), with a pulse width of less than 200 ps and an average optical power ranging from 7 μW to 440 μW. The time-delayed PL signal was detected by a superconducting single photon detector (SSPD) from Scontel, operated at 1.8 K. It allows single photon detection with a quantum efficiency of approximately 12% (at a wavelength λ = 1310 nm) and a counting rate larger than 70 MHz. Different emission wavelengths λ PL of the PL were selected by band-pass filters with half-width at half-maximum (HWHM) of about 10 nm. Most of the decay curves of the GIB-QDs-related PLemission show neither single-nor double-exponential behavior. They are best described by a log-normal distribution, as discussed in detail by van Driel et al. 42 The distribution of decay rates can be extracted following eq 1.
Here, C is a normalization constant, γ is related to the hwhm of the distribution, ΔΓ, and Γ mf is the most frequent rate constant: The average lifetime of the decay is thus calculated as follows: Disk Fabrication. As a demonstrator for the incorporation of GIB-QDs into a resonant cavity, we fabricated microdisk resonators with diameters of 1.8 μm. The disk shape was written by electron-beam lithography on a LEO Supra 35 scanning electron microscope with an attached Raith Elphy plus pattern generator into a negative resist. After development, the resist acts as a mask for a reactive ion etching process in an Oxford 100 cryo-ICP etcher that is used to thin the structures down to the Si/SiO 2 interface with perpendicular sidewalls. For this process, we used the gases SF 6 , He, and O 2 . Hereafter, the microcavities are partially underetched by hydrofluoric acid (HF) in order to increase the mode confinement due to the larger refractive index contrast between the Si/Ge layer and the surrounding air. The carriers in the sample were optically excited, and the PL signal was detected perpendicular to the disk. The surface roughness of the disk sidewalls and the HF under-etching procedure are not fully optimized, which explains relatively low quality factors of about 1700 (see Supporting Information).

* S Supporting Information
The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acsphotonics.5b00671.
Evaluation of the activation energies, TEM investigation of the wetting layer between the GIB-QDs, quantitative size analysis of QDs, theoretical evaluation of transition energies in crystalline QDs, estimation of the number of excitons generated per QD, and additional data concerning PL-properties of microdisk containing GIB-QDs (PDF).