1. Introduction

There is intense ongoing research on developing efficient light sources for silicon photonics. This would enable optical inter-chip and intra-chip communications on Si-based CMOS circuits with large bandwidth and low power consumption, both necessary for today’s seemingly insatiable needs of data communication. Since silicon itself has an indirect bandgap, a lot of research interest has been directed to use III-V semiconductors on silicon as the light emitting material despite the problems arising from the large lattice mismatch. Two main approaches have been considered: bonding and heteroepitaxy [1]. Although the bonding approach has been proven to be successful, heteroepitaxial techniques are preferred in the long run to enhance the flexibility and reduce the cost [2–6]. A recent review article by Liu and Bowers emphasizes the epitaxial growth on silicon as a promising method to minimize the cost, the size, the weight and the power dissipation of photonic integrated circuits [7]. The review also points out that the epitaxial growth of III-V on silicon eliminates the need for expensive III-V substrates.

Electrically pumped light sources from heteroepitaxial III-V materials on silicon so far are dominated by the use of quantum dot lasers emitting at 1.3 µm, e.g., [7–8]. Similar works with 1.55 µm emission are scanty. Shi et al. demonstrated InAs/InAlGaAs quantum dot microdisk lasers epitaxially grown on silicon but with optical pumping [9]. Electrically pumped 1.55 µm ridge lasers have been achieved with GaSb-based multi quantum wells (MQW) on Si [10], InGaAs/InGaAsP MQW on Si with GaAs buffer layer [11], InGaAs/InGaAsP MQW [12] and InGaAs/InAlGaAs MQW [13] on III-V-on-V-grooved-silicon and microring lasers with InAs quantum dash on Si [14]. The 1.55 µm vertical cavity surface emitting lasers (VCSEL) on Si are attractive for high speed and energy efficient optical interconnect in datacenter and high-performance computing [15,16] because of their low threshold current and radially symmetric light beam suitable for low loss fiber coupling. We employed epitaxial lateral overgrowth (ELOG) to achieve a high quality InP on Si where a threading dislocation density of < 4×10⁷ cm^-2 can be achieved [17]. MQWs grown on InP arising out of ELOG can be directly integrated with the waveguide or distributed Bragg reflector (DBR) acting as the mask in the ELOG process to enable monolithic evanescently coupled silicon laser (MECSL) [18] or InP based 1.55 µm VCSEL on Si [19]. The optical properties of MQW structures grown on ELOG InP were studied in [20], however, electroluminescence/electrical injection and the gain properties of MQW on ELOG InP have not been investigated. Here, we report on InGaAsP/InP based MQW laser structures grown on ELOG InP on 3” Si (001) with 4° off cut towards <111> direction. We employed the ELOG method to deposit InP on Si by hydride vapor phase epitaxy (HVPE) [21] on which molecular beam epitaxy (MBE) growth of InGaAsP/InP MQWs was conducted. Due to the uneven surface morphology of the ELOG InP, it was subjected to chemical mechanical polishing (CMP) prior to the MBE growth. We also investigate the material properties prior to and after CMP. Photoluminescence, Raman and X-ray diffraction were employed for this purpose. The wafer was processed to light emitting diodes (LED) and we present the room temperature electroluminescence results of the ensuing devices. We demonstrate that the MQW structure grown on ELOG InP/Si indeed exhibits gain with longitudinal cavity modes, which is promising for realizing VCSEL on Si with further optimization.

2. Experimental

An n-type (001) Si substrate with 4° off cut toward <111> direction was provided with a 2 µm thick InP seed layer by metal organic vapor phase epitaxy (MOVPE). Multiple circular discs of 400 nm thick SiO₂ with a diameter of 30 µm were formed on the seed InP layer on Si by using plasma enhanced chemical vapor deposition (PECVD), photolithography, and reactive ion etching (RIE). Circular mask was used to achieve fully coalesced ELOG layer and reduce the coalescence defects over the mask [22]. The circular discs were arranged in a triangular lattice with a center-to-center distance of 33 µm between two adjacent openings. A schematic view of the patterns and the ELOG are presented in Fig. 1(a).

 

Fig. 1. Schematic drawing of (a) ELOG template on silicon substrate and (b) after ELOG.

Download Full Size | PDF

ELOGs of unintentionally doped InP were realized on the InP seed through openings in SiO₂ mask by HVPE, which were fully coalesced over SiO₂ mask (Fig. 1(b)). The root mean square surface roughness of the as grown ELOG layer was 40 nm in a scanning area of 30 µm × 30 µm, which was reduced to 1.5 nm by applying CMP. The targeted thickness of the ELOG InP layer in HVPE growth was 20 µm. Due to the tensile strain of the layer, the wafer experienced warping and caused breakage into two pieces during CMP. This also resulted in uneven polishing throughout the wafer. The average thickness of the ELOG layer after CMP was ∼12 µm as revealed by SEM cross-section (not shown here). The growth of the MQW structure was done by gas-source molecular beam epitaxy (GSMBE) on these two pieces held side by side. A commercial MQW laser structure is grown on ELOG InP/Si as following: p-InGaAs (100 nm)/p-InGaAsP (λ=1.3 µm, 20 nm)/p-InP (1.6 µm)/6×InGa(0.2)As(0.4 in barriers, 0.7 in wells)P QWs/n⁺-InP (2.2 µm)/n-InGaAsP (λ=1.1 µm, 10 nm)/n-InP (2 µm)/ELOG InP(12 µm)/Si. The growth was conducted at 500^°C using a standard recipe used for growing qualified telecom lasers on InP substrates.

Since only certain detached regions of the grown whole wafer exhibited good planarity after MQW laser structure growth, this led to challenges in defining features with precision of less than 5 µm and did not permit us to process it further for ridge laser. Hence, one half of the wafer was used for material characterization and the other half was processed to light emitting diodes.

2.1 InP/Si and MQW/InP/Si characterization

After CMP, reciprocal lattice mapping (RLM) from high resolution X-ray diffraction (HRXRD) studies (ω-scan rocking curves and ω-2θ scans), photoluminescence (PL) mapping and Raman spectroscopy were performed on the ELOG InP to assess the InP crystal quality. Raman studies were conducted on the seed InP, ELOG InP prior to and after CMP and on p-InP cladding layer of the MQW structure after removing the ternary contact layer by selective etching. PL measurements on the MQW structure were also conducted.

2.2 LED processing

We processed the material to make LEDs with mesa diameters from 30 µm to 200 µm. The process started with a lift-off of Ti-Au (20-200 nm) to form disk and annular shaped contacts to the p-type InGaAs. To form the mesas a 760 nm thick oxide mask was deposited and patterned using a fluorine based etch. Then the InP material was etched to 3.44 µm (through the quantum well active region) using a CH₄/Cl₂/H₂ inductively coupled plasma (ICP) etch chemistry. A short etch of InP was performed (HCl:H₃PO₄ 1:4) in an attempt to reach the stop layer. The oxide mask was removed with buffered HF solution. A shadow mask was used to evaporate Au-Ge-Ni-Au n-type contact. The contact metals were annealed at 380°C for 5 minutes in a tube furnace. The schematic view and the scanning electron microscope images of the processed devices are shown in Figs. 2(a) and 2(b). The rough surface morphology after MQW growth by surface kinetically controlled GSMBE technology is evident in Fig. 2(b).

 

Fig. 2. (a) Schematic view of the MQW structure processed as LED. (b) SEM view of the wafer processed into electrically contacted LEDs.

Download Full Size | PDF

3. Results and discussions

3.1 ELOG InP/Si and MQW/ELOG InP/Si characterization

The crystalline quality of ELOG InP/Si was characterized by HRXRD and µ-PL mapping. The full width at half maximum (FWHM) of the ω-rocking curve obtained from reciprocal lattice mapping (RLM) of (004) reflection of ELOG InP from HRXRD (Fig. 3) is found to be 239 arcsec and that of the ω-2θ curve, 75 arcsec. The values for InP layer on Si using quantum dots as dislocation filters obtained by Shi et al. are 507 arcsec and 78 arcsec, respectively [23]; similar values obtained for heteroepitaxial InP on nanostructured (001) silicon are 360 arcsec and 78 arcsec, respectively [13,24]. To the authors’ knowledge, our values are among the best reported so far. The upper bound of threading dislocation density (TDD) can be estimated by the FWHM of ω-rocking curve according to Ayers’ model [25], which is 1.8 × 10⁸ cm^-2 for 239 arcsec in ELOG InP/Si. ELOG InP/Si has evidently improved crystal quality than the seed InP grown by MOVPE with TDD in the range of 10⁹ cm^-2. In the HRXRD RLM at (004) reflection of ELOG InP/Si shown in Fig. 3, (a) single InP layer peak is observed indicating no tilt angle between the lateral overgrowth and the vertical growth from window openings [26] due to the single coalesced ELOG layer throughout the wafer.

 

Fig. 3. HRXRD reciprocal lattice mapping at (004) reflection of the InP ELOG layer on Si.

Download Full Size | PDF

Figure 4 presents the µ-PL mapped area (45 µm × 45 µm) of ELOG InP after CMP. The step size was 3 µm. The contour was taken for the InP band-edge emission of 921 nm. The side bar shows the corresponding intensity in arbitrary units. It is clear from this figure that ELOG region is of much better quality than that of the seed region. The area with high PL intensity is smaller than the size of the mask with 30 µm diameter. This could be caused by the threading dislocations in InP seed propagating or escaping to the ELOG over mask area, which was observed in ELOG InP/Si grown from annular openings [27]. Considering the rough InP seed surface, the conformal SiO₂ deposition could also be rough, which may introduce a high density of dangling bonds and nucleation sites and results in extra defects. By applying CMP on the seed prior to SiO₂ mask deposition and subsequent ELOG growth, lower surface roughness can be obtained [28]. We point out that the quality and the surface roughness of the ELOG layer can be further improved by introducing aspect ratio trapping (ART) with thicker SiO₂ masks [6,29].

 

Fig. 4. PL mapping of 921 nm emission from the ELOG InP on Si. The side bar shows the intensity scale.

Download Full Size | PDF

In order to analyze if the CMP resulted in a perfectly planar surface, we also undertook Raman studies since transverse optical (TO) phonon modes will be absent in the case of a perfect planar (001) surface. Figure 5 presents the Raman spectra of the sample at different stages of processing: a) plain seed-InP (prior to ELOG), b) InP after ELOG, c) ELOG InP after CMP, d) InP cladding after MQW growth.

 

Fig. 5. Raman shift of InP-seed/Si, ELOG InP/Si after growth, ELOG InP/Si after CMP and InP on MQW measured from the top (001) surface. A 514.5 nm Ar⁺ laser beam was used as the excitation source.

Download Full Size | PDF

The exposure of the combination of (110) and (111) planes on (001) oriented wafer surface could cause non-uniform morphology of MQW grown by surface kinetically controlled GSMBE technology, which has different growth rates on different crystal planes.

The PL spectrum with its peak at 1529 nm and PL intensity mapping of MQW at this emission wavelength are shown in Fig. 6(a). All these data were collected after selectively etching away the p-InGaAs contact layer and the p-cladding layer. The PL mapping reveals that the MQW on the selective area of the coalesced ELOG region (large circle in the inset of Fig. 6) is of better optical quality. The PL spectra of MQW on the better quality ELOG region in the large encircled region and that on the seed region (small encircled region) are presented. The peak intensity of MQW on the good quality region above the ELOG regions is much stronger than that on the seed region.

 

Fig. 6. (a) PL spectra of the MQW. The peak is observed at 1529 nm. Inset map shows the PL intensity distribution of the MQW emission (1529 nm); the large encircled region is the ELOG region and the small one is the seed region. (b) PL spectra of MQW under various optical pump power.

Download Full Size | PDF

The interference fringes observed in the PL spectra are due to the Fabry-Perot (FP) cavity formed between the air/MQW interface and the SiO₂ mask layer on the seed InP used for ELOG. The FP modes are distributed evenly at the interval of ∼ 26 nm. The FP mode spacing can be calculated using the equation, Δν = $\frac{c}{{2{n_g}L}}$, where Δν is the frequency difference between two adjacent modes, c is the speed of light, n_g is the group index of the medium (InP) and L is the cavity length (InP thickness). Taking λ_peak as 1529 nm, n_g as 3.338 and L as 14 µm (sum of the thicknesses of the ELOG InP layer after CMP and the n-InP cladding layer), Δλ is calculated as 24.6 nm which is close to the observed fringe interval. The peak wavelength shifts around ± 5 nm due to the variation of the medium thickness. As shown in Fig. 6(b), PL spectra of MQW were measured under various optical pump power between 8-40 mW. By increasing the pump laser power, the intensity of the FP cavity modes increases simultaneously.

3.2 LED characterization

The MQW on ELOG InP/Si wafer was processed to LED in the manner described above for electroluminescence studies. The electrical characteristics are shown for a 200 µm diameter mesa with an annular contact in Fig. 7. The I-V curve is linear, and the resistance of the device is 23 Ω. The device shows a low leakage of < 1 µA at -1 V (< 3 × 10⁻³ A/cm²) while in forward bias the diode ideality is 1.7 indicating bulk recombination at low currents.

 

Fig. 7. Electrical characteristics of 200 µm diameter mesa MQW on ELOG InP/Si. The insets show (i) the low reverse leakage and (ii) the subthreshold diode characteristic on a log scale with I ∼ exp(qV/1.7kT).

Download Full Size | PDF

The electroluminescence spectra are shown in Fig. 8(a) for different applied currents plotted in logarithmic scale. The emitted light was collected from the top of the mesa using a multimode fiber and measured using an Ocean Optics spectrometer. No temperature control was employed.

 

Fig. 8. (a) Evolution of electroluminescence spectra from 200 µm diameter LED on Si for currents from 10 mA to 90 mA plotted in logarithmic scale. (b) Light-Current characteristic of the device.

Download Full Size | PDF

It is clear from Fig. 8(a) that the MQW emission is around 1.53 µm with a full width at half maximum (FWHM) of 200 nm. Hence the MQW growth is according to the targeted value. A small spectral shift of the maximum and mainly a broadening of the spectrum on the high energy side are observed with increasing current. The slope of the high energy side is proportional to exp(-E/kT). The slope remains similar at the high currents suggesting that the junction temperature is fairly constant. The light intensity shown in Fig. 8(b) increases with current up to 150 mA (500 A/cm²), which is well above normal LED operation range. Auger recombination will become significant at higher currents and result in saturation. It is noted that the power initially increases superlinearly with current due to the saturation of non-radiative recombination associated with defects. It should be noted that the InGaAs contact layer is still present and thus absorbing part of the emitted light. By growing thinner less indium-containing InGaAs alloy or using highly doped 1.3 µm wavelength InGaAsP lattice-matched alloy as p-contact, emitted light absorption can be avoided. According to our knowledge, this is the first electrically injected InGaAsP/InP MQW structure on ELOG InP useful for MECSL and 1550 nm VCSEL on Si. A recent publication by Megalini et al. [31] shows photoluminescence from InGaAsP/InP MQW on Si (from an SOI substrate) and the full width half maximum (FWHM) of their spectrum is ∼120 nm. The FWHM of our electroluminescence spectrum is ∼200 nm which is slightly higher than their value. Light emission was achieved from all diameter mesas. The spectra in Fig. 8(a) are modulated with a fringe spacing of 25 nm as for the PL spectra (Fig. 6(b)) due to the longitudinal FP cavity formed between air/semiconductor interface and SiO₂/seed InP interface buried under ELOG InP. In this work, the total thickness of semicondutor layer above SiO₂ mask is more than 20 µm, which gives rise to a close spacing between the adjacent FP cavity modes. By using a short cavity with 1550 nm MQW on top of a dielectric distributed Bragg reflector as ELOG mask, the frequency spacing of longitudinal modes can be wider than the gain bandwidth. Thus, 1550 nm VCSEL with single mode operation can be achieved on ELOG InP/Si.

3.3 Gain measurement

The strong and narrow electroluminescence peak indicates that the MQW material may have good enough quality to achieve net optical gain. To investigate this, a setup was built to measure the transparency current, using the method described in Refs. [32] and [33]. A VNA (Vector Network Analyzer HP 4195A) was used to generate a sinusoidal signal (f=10 kHz) which was fed to the modulation input of a tunable laser (Agilent 81689A). The intensity of modulated light, ΔP(t), was injected from above into the LED structure via a tapered fiber. The induced voltage modulation, ΔV(t), over the LED contact was probed and coupled into the receiver port of the VNA to measure its amplitude and phase. Since the contact voltage is determined by the splitting of the quasi-Fermi levels, it is a measure of the carrier density. If the device is biased below transparency, the light will act as an optical pump that via stimulated excitation increases the carrier density and the contact voltage. Above transparency the light will instead deplete the carrier density via stimulated recombination. Hence, by noting when ΔV/ΔP changes from positive (stimulated excitation) to negative (stimulated recombination), the transparency current as a function of wavelength can be determined.

For each wavelength of the injected modulated light (1525 nm-1550 nm), the transparency current was determined in this way by adjusting the LED bias current for minimum induced voltage modulation (Fig. 9(a)). The measured transparency currents ranged from 7 mA to 25 mA depending on wavelength and sample size (Fig. 9(b)). Hence, above this current the devices exhibited net material gain. However, a quantitative measurement of the material gain would require single mode waveguide structures with known radiation losses.

 

Fig. 9. Transparency current measurement. Measurement principle (a) and measured transparency current for an LED with 50 µm diameter shown as colored dots in (b). At higher currents (red area), the device exhibits material gain.

Download Full Size | PDF

4. Conclusions

We have for the first time demonstrated electroluminescence and gain from InGaAsP MQW grown on 3-inch ELOG InP/Si wafer suitable for realizing monolithic evanescently coupled silicon lasers at 1.5 µm wavelength. The crystalline quality of InP grown on Si was enhanced by ELOG method, which is revealed by narrow FWHMs of ω-rocking curve and ω-2θ scans in HRXRD-RLM and high luminesce intensity in ELOG region in PL mapping. The uneven morphology of ELOG InP/Si wafer was planarized by CMP to a reduced surface roughness of 1.5 nm before MQW growth by GSMBE. Stronger PL intensity associated with interference fringes arising from longitudinal FP cavity modes is observed in MQW grown on ELOG region than that on seed InP. The CMP process exposed a combination of (110) and (111) planes in addition to (001) surface plane because of the wafer warping caused by thermal strain in the thick InP layer on Si, which results in a non-uniform MQW morphology after surface kinetic limited GSMBE growth. By mitigating the thermal strain in thick InP/Si to avoid wafer warping, uniform MQW growth on CMP processed ELOG InP/Si can be expected. Based on the current design, stripe laser fabrication is in principle possible in conjunction with etched facets. The more significant issue is the variation in the quality of the material over distances of >30 µm which leads to inhomogeneous emission and therefore reduction in the gain. A solution might be to make ring lasers within the 30 µm diameter zones in future. The grown MQW on ELOG InP/Si wafer was processed to electrically injected LEDs, which show very strong luminescence with a FWHM of 200 nm. Such 1.55 µm LEDs would be very useful for spectroscopy for characterization of materials (e.g., Si wafers which are nominally transparent at this wavelength) as well as free space signalling (due to low background from environment - the sun - along with low voltage operation). The electroluminescence spectra of LED are modulated by fringes with close spacing due to the FP cavity formed by the LED device structure and ELOG InP above SiO₂ mask. The transparency current measurements reveal an optical gain in MQW material grown on ELOG InP/Si, which is promising for further development of wafer level integration of lasers on silicon. Further optimizations in the process flow such as achieving a thinner ELOG layer after CMP, ART to improve crystalline quality and employing corrugated epitaxial lateral overgrowth (CELOG) [34] to improve thermal performance can be employed in the future to realise a MECSL at telecom wavelength. By reducing the thickness of the optical cavity with 1550 nm MQW above the dielectric DBR as the mask for ELOG InP, single mode telecom wavelength VCSEL can be realized on silicon.