Defect‐Engineered Electrically‐Injected Germanium‐on‐Insulator Waveguide Light Emitters at Telecom Wavelengths

Ge‐on‐insulators (GOIs) have been extensively explored as a potential platform for electronic‐photonic integrated circuits (EPICs), enabling various emerging applications. Although an efficient electrically‐injected light source is highly desirable, realizing such devices with optimal light emission efficiency remains challenging. Here, the first room‐temperature electrically‐injected Ge waveguide light emitters consisting of a lateral p–i–n homojunction on a GOI platform that can be monolithically integrated with EPICs are demonstrated. A high‐quality Ge active layer is transferred onto an insulator layer with the misfit dislocations in the Ge active layer eliminated to suppress unwanted nonradiative recombination. A 0.165% tensile strain is introduced to enhance the directness of the band structure and improve the light emission efficiency. The device comprises a waveguide structure with a significantly improved optical confinement as the optical resonator and a lateral p–i–n homojunction structure as the electrical injection structure. Under continuous‐wave electrical current injection at room temperature, enhanced electroluminescence is successfully observed at telecommunications wavelengths covering the C, L, and U bands, with improved efficiency. Theoretical analysis suggests that the quantum efficiency of Ge light emitters is dramatically affected by the defect density. These results pave the way for developing efficient, room‐temperature, electrically‐injected light emitters for next‐generation GOI‐based EPICs.


Introduction
[3] Considering that the materials and fabrication processes are compatible with standard complementary metal-oxidesemiconductor (CMOS) processing, group-IV semiconductors are highly preferred for the seamless manufacturing of EPICs in modern CMOS foundries.Although several fundamental building blocks of photonic devices, including photodetectors, [4][5][6] modulators, [7][8][9] and passive waveguides, [10] with excellent performance have been developed, the key device for generating light, that is, an efficient electrically-injected light emitter, remains challenging, primarily because of the indirectness of the band structure of group-IV Si and Ge semiconductors.
Fortunately, although Ge is an indirect bandgap semiconductor, its direct conduction band (CB) lies just 136.5 meV above the lowest L-valley indirect CB at room temperature. [11]In addition, the direct bandgap energy of Ge is 0.8 eV at room temperature, allowing the generation of light emission around  = 1550 nm via direct-gap band-to-band transitions; thus, it is useful for applications in the telecommunication C-band, the most important telecommunication window widely used in dense wavelength division multiplexing (DWDM) technologies to enhance data transport capacity.These unique advantages make Ge a promising candidate for efficient light emitters.Despite the large lattice mismatch (≈4.2%) between Si and Ge, epitaxy techniques have been continuously advanced to enable the direct growth of Ge on Si substrates (Ge-on-Si) or SOI substrates.Photoluminescence, [12] electroluminescence (EL), [13,14] optically pumped lasing, [15] and electrically injected lasing [16,17] at room temperature from a Geon-Si structure have been demonstrated, showing great promises in the development of efficient light emitters.However, the performance is still unsatisfactory because fundamental material challenges have not been fully addressed.First, the large lattice mismatch between Si and Ge leads to the development of significant permanent defects as high as 10 9 cm −2 in the Ge active layer, [18] which act as Shockley-Read-Hall (SRH) nonradiative recombination centers that can significantly reduce the light emission efficiency.Although various approaches, such as the postannealing [19,20] two-step growth method [21][22][23] and compositiongraded buffer, [24] have been developed, high defect densities remain near the Ge/Si interfaces.Second, the relatively small difference in the refractive indices (RIs) of Ge and Si leads to weak optical confinement of the Ge active layer, thus limiting the lightmatter interaction.
[27][28][29][30][31] The top Ge is of high quality as the defect region has been removed.In addition, the insulator layer can provide strong optical confinement to the top Ge layer, which is suitable for fabricating efficient photonic devices.Furthermore, a residual strain of ≈0.2% can be introduced into the top Ge layer to enhance the directness of the band structure.Utilizing GOI platforms, researchers have demonstrated high-performance Ge photodetectors operating at room temperature [27][28][29][30][31] as well asoptically pumped Ge nanowire lasers at cryogenic temperatures of up to 83 K, [32] highlighting the great potential in the development of efficient photonic devices needed for EPICs.However, to meet the requirements of EPICs, electrically injected planar Ge light emitters are highly desirable, which have not yet been realized on GOI platforms.
In this study, to the best of our knowledge, we demonstrated the first electrically-injected Ge edge-emitting waveguide lightemitting diodes (WGLED) on a GOI platform as a key building block for EPICs.A residual tensile strain of 0.165% was introduced into the Ge active layer to enhance the directness of the band structure, thereby enhancing direct-gap transitions.The defective region was eliminated to suppress the unfavorable nonradiative recombination.The insulator layer provides strong optical confinement for the Ge active layer, thus enhancing the lightmatter interaction.A waveguide structure was used as the optical resonator to enhance EL.A lateral p-i-n diode structure was developed for current injection, which avoids the problem of significant optical loss caused by the heavily doped region and metal pads, thus improving the light emission efficiency.Under electrical injection, EL was observed at room temperature with enhanced light emission efficiency.The electrical and optical properties of the GOI WGLEDs were systematically characterized and discussed.

Results and Discussion
The GOI samples used in this study were prepared using layer-transfer and wafer-bonding techniques.The procedure for preparing the GOI sample is illustrated in Figure 1a.First, an intrinsic Ge layer was epitaxially grown on a Si(001) substrate through reduced-pressure chemical vapor deposition (RPCVD) (Figure 1a-i).A SiO 2 interfacial layer was then deposited on the Ge-on-Si sample as well as a Si handle substrate using plasmaenhanced chemical vapor deposition (PECVD) (Figures 1a-ii,a-iii).Subsequently, a SiN layer was deposited on the Ge-on-Si sample using PECVD (Figure 1a-ii) to suppress the formation of interfacial voids during the bonding process.After cleaning the samples with O 2 plasma for 15 s and rinsing with deionized water and spin drying, the Ge-on-Si sample was flipped and bonded onto the Si handle sample (Figure 1a-iv).Thermal annealing at 300 °C was then performed for 3 h in an N 2 environment to enhance wafer binding.Subsequently, the sample was grinded, etched with tetramethylammonium hydroxide, and polished using chemical-mechanical polishing to remove the Si handle wafer and the defective region of the Ge layer (Figure 1a-vi) to complete the GOI sample.Figure 1b,c shows the comparison of the cross-sectional transmission electron microscopy (XTEM) images of the as-grown Ge-on-Si samples and fabricated GOI samples, respectively.For the Ge-on-Si sample, dense misfit dislocations, as indicated by the arrows, were observed at the Ge/Si interface (Figure 1b).By contrast, for the GOI sample, no apparent defects were observed in the Ge layer (Figure 1c-i) as the defective region was eliminated, indicating good material quality.The threading defect density of the Ge layer was obtained to be ≈2 × 10 7 cm −2 by etch pit density measurements, showing improved material quality compared to the Ge-on-Si sample.The XTEM image also revealed that the top Ge layer has a thickness of 960 nm, and the insulator layer is composed of a 312-nmthick SiO 2 layer, a 50-nm-thick SiN layer, and a 312-nm-thick SiO 2 layer with flat and abrupt interfaces between the layers.Figure 1cii shows the selected-area electron diffraction pattern of the Ge layer, providing evidence regarding the single-crystalline nature of the Ge layer.Raman microscopy was employed to measure the strain in the Ge layer. [33]Figure 1d shows the Raman spectra of the prepared GOI material and a bulk Ge wafer.Fitting the data with Lorentzian functions yielded the peak positions.For the bulk Ge, a peak was observed at 300.01 cm −1 , which was assigned to the longitudinal-optical (LO) phonon modes.By contrast, the Raman spectrum of the GOI material peaked at 299.32 cm −1 , suggesting the existence of strain.The in-plane biaxial strain ( || ) in a material can be deduced from the Raman shift (Δ) using the following expression: Using b = −415 cm −1 , [34] we obtained  || = 0.165%.This residual tensile strain is attributed to the mismatch between the thermal expansion coefficients of Si and Ge during the annealing process of the Ge-on-Si growth. [5]igure 2a,b shows the schematic three-dimensional and frontview diagrams of our designed GOI WGLEDs.The device has a lateral p-i-n diode structure for current injection, wherein the thickness of the Ge layer is t.In the intrinsic Ge region, a rib structure with a height h and width w was designed for the effective confinement of light generated by electron-hole recombination through direct-gap band-to-band transitions.The surface was passivated with a SiO 2 layer to reduce surface nonradiative recombination via surface defect states.To evaluate the optical confinement factor (OCF) of the Ge waveguide, a finite element method simulation was performed, wherein the wavelength-dependent refractive indices (RIs) of the materials were adopted from those reported in previous studies. [35,36]The inset in Figure 2c shows the simulated energy distribution for the   quasi-transverse-electric (TE) fundamental mode of the waveguide at  = 1550 nm for w = 5 μm and h = 500 nm.The optical field distribution clearly shows that light can be properly confirmed in the waveguide, as the low-RI-deposited SiO 2 and the insulator layer provide strong optical confinement for the Ge active layer.Figure 2c shows the calculated optical confinement factor for the quasi-TE fundamental mode as a function of the thickness of the active layer.As t increased, the optical OCF of the Ge active layer significantly increased and tended to saturate beyond t > 700 nm.An excellent optical confinement factor of >95% can be achieved with t > 360 nm, which is beneficial for enhancing light-matter interactions.The GOI samples were fabricated into WGLEDs through CMOScompatible processing (Experimental Section).A scanning electron microscopy images of the fabricated device is shown in Figure 2d.
Figure 3 shows the current-voltage (I-V) characteristics of the fabricated devices measured at room-temperature using a sourcemeter (Keithley, 2400).The I-V characteristics exhibited significantly different trends in the forward and reverse bias regions with a high on-off ratio of 390 at ±1 V, demonstrating clear rectifying behavior and excellent Ohmic contact.To further understand the electrical characteristics, the I-V curve was modeled using the following standard diode equation [37] : where I 0 is the saturation current, R s is the series resistance, and  is the ideal diode factor.Equation (2) can then be transformed to The inset of Figure 3 shows a plot of dV/dI versus I −1 , which was fitted using Equation (3).From the slope and intercept, the ideality factor and series resistance were calculated to be  = 1.73 and R s = 36 Ω, respectively.For typical p-i-n junctions,  = 1 and  = 2 indicate that the dominant currents are the SRH nonradiative current and band-to-band radiative current, respectively. [37]Thus, the obtained ideality factor of our device indicates improved light-emitting efficiency.In addition, the low R s value suggests an excellent metal-semiconductor interface, which could help suppress Joule heating during the current injection.With the series resistance, the turn voltage was then determined to be 0.52 V by fitting the fixed slope of 1/R s .
Figure 4a,b shows the EL spectra of the devices collected from the top surface and one facet of the GOI WGLED, respectively.From the top surface EL spectrum shown in Figure 4a, a strong peak was observed at  = 1619 nm with a relatively weak peak at shoulder  ≈ 1800 nm, which were assigned to the directand indirect-gap interband transitions, respectively.The directgap emission covered the telecommunication C, L, and U bands, which matched the most useful DWDM bands.Thus, the developed GOI WGLED is compatible with the currently available fiber-optic networks and useful for telecommunication applications.On the contrary, the edge EL spectrum shown in Figure 4b, exhibited behavior similar to that of the top-surface EL spectrum; however, several ripple structures were observed.The ripple structures are evidence of the Fabry-Pérot modes in the waveguide.Figure 5c shows the edge EL spectra of the GOI WGLED at different injected currents.As the injected current increased, the EL intensity increased owing to the increased injected current density.In addition, the direct-gap emission peak redshifted slightly as the injected current increased.This observation was primarily attributed to the Joule heating effect.Figure 5d shows the integrated edge EL intensities of the devices.The light-current (L-I) curve can be characterized by L = C 0 I m , where C 0 is a constant and m is a fitting parameter that represents the emission efficiency of the device.Fitting of the experimental data yielded an m value of 1.57, which is higher compared with the m value of ≈1 for Ge-on-Si waveguide edge light emitters reported in a previous study. [14]This indicates a higher light emission efficiency of the proposed device, which is a result of the removal of the defect region and better optical confinement.
Herein, we discussed the effect of defects on the light emission efficiency of the device.[40][41] The parameters used in the calculations can be found in a previous study. [11]Figure 5a shows the schematic of the band structure of the Ge active layer.The incorporation of 0.165% tensile strain decreased both the direct and indirect conduction bands, and their energy difference reduced from 136.5 to 121 meV at T = 300 K.As a result, more electrons can populate the direct conduction band for electron-hole recombination via efficient direct-gap transitions, leading to enhanced light emission.On the contrary, tensile strain puhed the lighthole (LH) band above the heavy-hole (HH) band.As a result, the lowest direct transition was from the Γ-CB to the LH band (cΓ→LH), with a calculated transition energy of 770 meV, which agrees well with the experimental direct bandgap of 765 meV obtained from the EL results, confirming that the observed emission originated from the direct-gap cΓ→LH interband transition.Next, we analyzed the effect of defect density (N D ) on light emis- sion efficiency.The injected current density (J inj ) can be associated with the injected carrier density (n) using the following equation: where J SRH is the SRH nonradiative current density, J rad is the radiative current density, J Auger is the nonradiative Auger recombination current density, J ind is the indirect-gap recombination current density, A is the SRH recombination coefficient, R Γ is the direct-gap radiative recombination coefficient, C is the Auger nonradiative recombination coefficient, and R L is the indirectgap radiative recombination coefficient.Here, n Γ and n L are the carrier densities in the Γ-and L-valley conduction bands, respectively, and n = n Γ + n .p (= n) is the hole density.Carrier occupation was calculated using the model described in a previous study. [11]The SRH coefficient was strongly related to the defect density via the expression A = GN D , where G is a constant, calculated to be G = 0.19 cm 2 s −1 for Ge. [42,43]For Ge, the parameters used in the calculations were R Γ = 1.3 × 10 −10 cm 3 s −1 , R L = 5.1 × 10 −15 cm 3 s −1 , and C = 1.6 × 10 −30 cm 6 s −1 . [40,44]Next, we calculated the spontaneous emission rate per unit volume per unit energy interval with a Lorentzian broadening function as follows: [45] where n r is the refractive index, c is the velocity of light in free space, q is the electronic charge, m 0 is the mass of the free electron,  0 is free-space permittivity, ℏ is the reduced Planck's constant,  is the angular frequency of incident light, |ê ⋅ p cv | 2 = m 0 E p ∕6 is the momentum matrix with E p being the optical energy, f n (K) and f m (K) are the electron occupation probabilities of Γ-conduction band and valence band, respectively,  is the fullwidth-at-half-maximum of the Lorentzian line shape, and E CΓ (K) and E m (K) are the electron and hole energies in the Γ-valley CB and VB, respectively.Figure 5b shows the calculated spontaneous emission spectra at different defect densities at T = 300 K under a fixed injected current density of J inj = 10 kA cm −2 .The spontaneous emission intensity as a function of the defect density is shown in Figure 5c.We observed that, at a high defect density of N D = 10 9 cm −2 , the spontaneous emission rate is extremely small, as most of the injected electrons and holes recombine nonradiatively through SRH recombination.As the defect density was improved from N D = 10 9 cm −2 to N D = 10 8 cm −2 and to N D = 10 7 cm −2 , the spontaneous emission intensity significantly increased by a factor of 88 and 545, respectively.When the defect density was further improved to N D = 10 6 cm −2 , the spontaneous emission intensity also increased but was not as impressive.
Figures 6a-d shows the calculated current densities as a function of the total injected current density for different defect densities in the range of N D = 10 6 − 10 9 cm −2 at T = 300 K. We observed that, at a low defect density of N D = 10 6 cm −2 , J SRH is dominant at a low injected current density.Furthermore, J ind is negligibly small because the indirect-gap transition is a slow process owing to the need for additional phonons to conserve the momentum.As the injected current density increased, J rad and J Auger increased rapidly owing to the increased injected carriers and subsequently became dominant current densities.As the defect density increased, J SRH increased significantly and eventually outperformed other current densities at N D = 10 9 cm −2 .Using the calculated current densities, the internal quantum efficiency (IQE) can be evaluated as follows: Figure 6e shows the calculated IQE as a function of the injected current density for the GOI WGLED at varying defect densities.At N D = 10 6 cm −2 , the IQE increased with an increase in the injected current density as J rad surpassed J SRH .As the injected current further increased, J Auger became increasingly significant, thereby lowering the IQE.As the defect density increased, more injected electrons and holes recombined through nonradiative SRH recombination, thus considerably lowering the IQE.Although increasing the injected current density increased the emission intensity, J Auger became dominant at high injected current densities, thus limiting the IQE.These results suggest that it is essential to improve the defect density in the Ge active layer to the level of ≈N D = 10 7 cm −2 to enhance the light emission efficiency of Ge light emitters, highlighting the importance of defect engineering in our GOI WGLEDs.
The realization of our planar GOI WGLEDs represents an important advancement in GOI-based EPICs.Recently, GOIbased photonic devices including low-loss strip waveguides, [46] modulators, [47] and photodetectors [48] have been successfully demonstrated.Additionally, Ge metal-oxide-semiconductor fieldeffect transistors (MOSFET), [49] planar field-effect transistors (FET), [50] and fin field-effect transistors (FinFET) [51] have also been demonstrated on GOI platforms with excellent electrical performance.This noteworthy progress paves the way for monolithic integration of electronics and photonics on the GOI platform.Another advantage is the capability of manufacturing GOI substrates up to 8″ in diameter. [52]Since GOI is fully-compatible with standard CMOS processing, [53] it is anticipated that the widespread adoption of the GOI platform in modern CMOS foundries will facilitate the realization of largescale and functional GOI-based EPICs for a wide range of applications.

Conclusion
In this study, we demonstrated the first electrically-injected Ge waveguide light emitters with a lateral p-i-n junction on a Geon-insulator platform.The Ge layer was transferred onto an insulator layer with a low refractive index, achieving excellent optical confinement for the Ge active layer.A significant biaxial tensile of 0.165% was introduced to the Ge active layer, which reduced the energy difference between the direct and indirect conduction bands, thereby enhancing the direct-gap transition and light emission efficiency.The defective region in the Ge active layer was also eliminated to suppress nonradiative recombination, leading to enhanced light emission efficiency.Consequently, strong electroluminescence under continuous-wave current injection was obtained at room temperature with enhanced efficiency.Theoretical analysis of the effect of defect density on light-emitting efficiency indicates that defect density plays a crucial role in light-emitting efficiency and internal quantum efficiency.Our demonstration of Ge lateral p-i-n diodes provides a promising solution for efficient electrically injected light emitters operating at room temperature to enable the fabrication of GOIbased EPICs.

Experimental Section
Device Fabrication: The sample was fabricated into lateral p-i-n diodes using a CMOS-compatible fabrication process.The process began with the definition of a ridge structure with a width of w = 5 μm and a height of h = 450 nm using standard optical lithography and reactive ion etching techniques.Ion implantation techniques were employed to introduce n-and p-doped regions by implanting boron and phosphorous ions, respectively, at a dose of 10 15 cm −2 and an ion implantation energy of 20 keV to create a lateral p-i-n junction.Microwave annealing with a power of 1650 W was subsequently performed to effectively activate the dopant.SiO 2 passivation was then performed using PECVD, followed by wet etching with a buffered oxide etch (BOE) solution to expose the surface of the n-and p-doped Ge regions.Au/Cr metal pads with thicknesses of 200/20 nm were then deposited using an e-beam evaporator and patterned using the lift-off method to complete the device.The sample was then cleaved into devices of different lengths, which were packaged on chip carriers and wire-bonded for measurements.The device used in this study has a length of L = 3 mm.
Raman Microscopy Experiments: Raman experiments were performed using a μ-Raman microscope (Jobin Yvon, Labram/HR800) under a backscattering scheme.A 532-nm laser was used as the excitation light source, which was focused to a spot size of ≈1 μm and an objective of 100 × .The laser power was maintained at a lower level to avoid unwanted thermal heating.
Electroluminescence Experiments: The EL from the GOI devices was measured at room temperature by injecting an AC current with a frequency of 1 KHz into the GOI device using a Keithley 2400 source meter.The emitted light from the device was then sent to a Fourier-transform infrared spectrometer (Thermo-Fisher, IS50R) equipped with an LN2-cooled extended-InGaAs photodetector in a step-scan mode, which can eliminate unwanted thermal emission.

Figure 1 .
Figure 1.Material preparation and characteristics.a) Schematics of the fabrication process of the Ge-on-insulator sample: a-i) growth of the Ge layer on silicon via reduced-pressure chemical vapor deposition, a-ii) deposition of SiO 2 and SiN layers on Ge-on-Si; a-iii) deposition of SiO 2 on the Si handle wafer; a-iv) wafer bonding; and a-v) grinding, tetramethylammonium hydroxide (TMAH) etching, and chemical-mechanical polishing (CMP) of the bonded sample.b) Cross-section transmission electron microscopy (XTEM) image of the as-grown Ge-on-Si sample, c) XTEM image of the GOI material: c-i) Cross-sectional transmission electron microscopy (XTEM) image; c-ii) high-resolution XTEM image; and c-ii) Fast Fourier Transform (FFT) pattern of the selected region in the Ge layer.d) Raman spectra of the GOI and bulk Ge.

Figure 2 .
Figure 2. Schematic of our designed lateral Ge p-i-n waveguide diodes on a GOI platform: a) 3D view and b) side-view (not to scale).c) Calculated optical confinement factor for the quasi-TE fundamental (TE) mode of the Ge active layer as a function of thickness.The inset shows the simulated field distribution.e) Scanning electron microscopy image of the fabricated device.

Figure 3 .
Figure 3. I-V curve of the Ge waveguide diodes.The inset shows a plot of dV/dI versus I −1 , where the red dashed line represents a linear fit to the data.

Figure 4 .
Figure 4. a) Top surface and b) edge electroluminescence (EL) spectra of the GOI device with an injected current of 260 mA at room temperature.c) Room-temperature edge electroluminescence spectra of the GOI device at different injected currents.d) Integrated electroluminescence intensity and peak emission wavelength as a function of injected current.

Figure 5 .
Figure 5. Theoretical analysis.a) Modeled band structure of the Ge active layer with a tensile strain of 0.165%.b) Calculated spontaneous emission rate spectra of the Ge active layer with a tensile strain of 0.165% at different defect densities.c) Calculated spontaneous emission intensity as a function of defect density.

Figure 6 .
Figure 6.a-d) Calculated different current densities as a function of injected current density at different defect densities.e) Calculated internal quantum efficiency as a function of injected current density of Ge at different defect densities.