1. Introduction

Compact silicon waveguides are the key elements to form the backbone in the integrated silicon photonics circuits [1]. One of the significant advantages, by employing silicon as a waveguide core material and silicon oxide as a cladding material, is that a wide range of effective refractive index of the waveguide can be realized from as low as 1.45 (oxide index) to as high as 3.48 (silicon index) at the wavelength of 1550 nm. This wide range of effective refractive index can be achieved by controlling the cross sections of silicon cores with advanced CMOS fabrication processes. The large span of the waveguide index allows efficient evanescent coupling of silicon waveguides to various types of other waveguides such as silicon nitride, SiON, and III-V, because the evanescent coupling between different waveguides relies on phase matching conditions where the effective indices of two waveguides are close. This unique property enables silicon photonics to be a desirable integration platform for large-scale photonic integration. Indeed, with only one decade of development, silicon photonics has demonstrated remarkable integration capability with many demonstrated large-scale photonic integration circuits. For examples, a monolithic silicon photonic wafer was demonstrated to integrate many high performance photonic function blocks such as high-speed modulators, photodetectors, variable optical attenuators, thermo-optic phase tuners, arrayed waveguide gratings, microring filters, polarization rotators, polarization beam splitters, and various passive optical elements [1,2]. It also consists of different types of waveguides, such as silicon, silicon nitride, silicon oxide, and germanium waveguides, to form an interconnected optical network on the same chip [1,2]. Silicon/III-V hybrid lasers, where silicon waveguides function as part of the laser cavities, have been also developed, thanks to the effective coupling between III-V and silicon waveguides [3–11].

For Indium Phosphide (InP)-based gain waveguides, the effective refractive index is typically larger than 3.2 if the waveguide width and height are more than 1 μm because its cladding InP layer has an index of 3.17. In order to achieve this index for silicon waveguides for effective coupling to the InP-based gain region, the corresponding thickness of silicon waveguide layer needs to be sufficiently large. The effective refractive index of silicon slab waveguides with different thicknesses is shown in Fig. 1, indicating that the required silicon thickness needs to be larger than ~400 nm to achieve an effective index of 3.2. This can be observed in the previously reported silicon/III-V hybrids lasers employing at least 400 nm starting silicon thicknesses [3–13]. However, this thick starting silicon layer does not match well with many other silicon photonic components, such as electro-optic modulators, inverse tapers, and polarization elements, where the silicon thickness is typically between 200 nm to 300 nm. This large mismatch of the silicon waveguide thickness is very challenging for implementing efficient coupling between InP waveguide and other silicon elements. Although it is still possible to couple light from 220 nm silicon to InP waveguides by using very narrow InP waveguides (~200 nm) to push down the value of effective index, as simulated in [14], the fabrication is extremely difficult to form these narrow InP waveguides. Therefore, a more flexible integration approach is desirable to realize efficient coupling without demanding thick silicon thickness and very fine lithography for InP waveguides, in order to achieve large-scale photonic integration of all these functional components.

 

Fig. 1 Effective refractive index of the fundamental mode of silicon slab waveguides versus silicon slab thicknesses. Both top and bottom claddings are oxide (n = 1.444) and silicon index is chosen as 3.48 for the wavelength of 1550 nm.

Download Full Size | PDF

In this paper, we propose and demonstrate a novel fabrication technique and associated transition structure to realize flexible coupling without the constraint of starting silicon thickness. This innovative technique employs epitaxial growth of silicon in a pre-defined trench with taper structures. Using the proposed technique, we demonstrate a long-cavity hybrid laser with a narrow linewidth of 130 kHz and an output power of 1.5 mW. The long-cavity design aims for applications in long-haul/metro coherent transmission where narrow-linewidth lasers are required especially for high-level quadrature amplitude modulation (QAM) formats such as 16-QAM in next-generation optical transport networks.

2. A new proposal for integration approach between silicon and III-V waveguides

In order to implement hybrid silicon/III-V lasers where silicon waveguides are part of the laser cavities, the starting silicon thickness needs to be larger than ~400 nm. If these lasers need to be integrated with other photonic circuits built on thinner silicon thickness, typically, a wafer with thick silicon thickness is used for coupling to InP-based gain material, then it is etched down to a thin silicon thickness for non-laser circuits, illustrated in Figs. 2(a)–2(c). The III-V gain waveguides can then efficiently couple to thick silicon waveguides first, which can then couple to thin silicon waveguides. This approach has been employed successfully in [7–9], where a hybrid laser is monolithically integrated with silicon-based Mach-Zehnder modulators or ring filters. However, this integration approach has some disadvantages: (1) the etching process introduces surface roughness which could induce significant waveguide losses, and (2) the etching process causes large thickness variations in the resultant thin silicon which are more problematic to obtain precision optical devices such as filters, couplers, etc. In order to solve these problems, we propose a new integration approach, explained in Figs. 2(d)–2(f). In this proposal, the starting silicon thickness is that required for non-laser circuits. This allows the use of the precision silicon top layers with well-controlled thickness and smoothness from wafer venders. For the region where III-V waveguides need to be coupled to silicon waveguides, a controlled layer of silicon (or other high index materials) is deposited by epitaxial growth or other deposition techniques to increase the total thickness of the silicon in the selected region to obtain desirable index of the silicon waveguide. Any transition structures reported by the previous etching approaches can readily be realized by this proposed growth technique. The growth window can be a trench etched in silicon oxide with pre-defined transition structures. By doing this, the non-laser circuits can be fabricated from well-controlled silicon thickness and also smooth silicon surfaces.

 

Fig. 2 (a)-(c) Conventional etch-down process to achieve thin silicon to III-V waveguide coupling. (d)-(e) Proposed grow-up process.

Download Full Size | PDF

3. Fabrication of silicon/III-V hybrid lasers

One of the possible fabrication sequences to realize the proposed integration approach is shown in Fig. 3. Starting from an SOI wafer with thin silicon, waveguides are first fabricated [Fig. 3(a)]. A top oxide cladding is then deposited, followed by chemical-mechanical polishing (CMP) for planarization. For the next step, trenches are defined on top of silicon by lithography and etching of oxide, shown in Fig. 3(c). The trenches can have taper structures to enable adiabatic or efficient coupling, typically used in silicon/III-V hybrid lasers [7–9]. After the trench formation, epitaxial growth of silicon is employed. Afterwards, CMP is required to flatten the surface for the following wafer bonding and remove the poly-silicon on top of oxide. Figure 3(f) shows a scanning electron microscope (SEM) image after the trench formation, and Fig. 3(g) presents an SEM picture after epitaxial growth. Poly-silicon can be seen on oxide from non-selective growth, which is subsequently removed after CMP [Fig. 3(h)]. The CMP is also required to produce a bonding surface with low surface roughness on the order of a few nanometers.

 

Fig. 3 (a)-(e) Fabrication sequence for proposed integration approach between III-V and silicon waveguides. (f) SEM pictures after trench formation; (g) SEM picture after epitaxial growth of silicon; (h) SEM picture after final CMP.

Download Full Size | PDF

After the completion of the silicon processing of SOI wafer, the remaining fabrication steps to form the hybrid lasers are similar to the standard processes that we reported previously using direct wafer bonding [10]. The III-V wafer is transferred to the processed silicon wafer through low temperature oxygen plasma assisted wafer bonding. Accurate alignment is not required for this bonding technique. The InP substrate is then removed using wet etching, leaving only the active epitaxial layers on the silicon wafer. Figure 4(a) presents an optical picture after the substrate removal, showing a 2” InP wafer bonded to the processed silicon wafer and then re-sized from 8” to 3”. Very high bonding yield is achieved as more than 95% III-V layers are remaining after substrate removal. For the next step, the InP waveguides are delineated by lithography using alignment marks on the silicon wafer and dry etch. An SEM picture after InP waveguide etch is shown in Fig. 4(b). A dielectric planarization is applied afterwards and followed by etching the contact vias and metallization. Figure 4(c) exhibits a picture of the InP gain section with metal contacts after full fabrication.

 

Fig. 4 (a) Optical picture of the III-V wafer bonded on SOI processed wafer after InP substrate removal. (b) SEM picture of III-V waveguide after III-V dry etch. (c) SEM picture of the III-V gain section after full fabrication.

Download Full Size | PDF

4. Long-cavity silicon/III-V laser

The hybrid laser is designed and fabricated based on a starting silicon thickness of 300 nm with a buried oxide thickness (BOX) of 3 μm. The laser cavity consists of InP semiconductor amplifier (SOA), passive silicon waveguide, a ring mode filter, and a thermo-optic phase element, shown in Fig. 5. The ring filter provides single mode selection. High reflection (close to 100%) and partial reflection (~30%) at the two chip facets provide the required feedbacks for laser operation. The total physical length of the cavity is about 1 cm. The long cavity is designed to achieve very narrow laser linewidth. In the cavity, the ring radius is 15 μm, and the ring waveguide has a cross section of 0.5 μm x 0.3 μm. The FSR of the ring is ~6.5 μm.

 

Fig. 5 Long-cavity hybrid silicon/III-V laser cavity design. HR: High refection; SOA: semiconductor amplifier with schematic drawing shown in the bottom-left diagram. SEM picture shows the in-cavity ring filter with micro-heater.

Download Full Size | PDF

The schematic SOA and transitions to the silicon waveguides with a height of 0.3 μm is shown in bottom-left part of Fig. 5. Here, the light-blue, dark-blue, and red regions represent 300 nm silicon, 600 nm silicon, and III-V. The 600-nm silicon is achieved by additional epitaxial growth of 300 nm silicon on top of 300-nm silicon. The structure and epitaxial layers of III-V waveguides were reported in [10], with the exception that the multiple quantum well (MQW) has a peak gain of ~1.6 μm. For the transitions from thin silicon waveguide to III-V waveguide, two-step tapers are used. Taper I (Fig. 5) help the light transfer from thin silicon to thick silicon, with the width tapering from 0.2 μm to 1.5 μm in a length of 100 μm. Taper II enables adiabatic light conversion from thick silicon waveguide to silicon/III-V compound waveguide. For taper II, the silicon waveguide width is tapering from 1.5 μm to 0.7 μm with a length of 150 μm, while the III-V waveguide width is tapering from 0.7 μm to 3.5 um. These taper structures are very similar to those reported in [7, 12, 15], except that the 600 nm silicon is achieved by epitaxial growth. For the III-V gain section, the total length is 950 μm and the width is 3.5 μm.

5. Laser characterization

5.1 Transition loss

To verify the proposed integration approach, a test device with 16 transition tapers (taper I) is fabricated with the hybrid laser. For taper II transitions, the highly absorptive III-V waveguides prevent us from characterizing the transition loss. Figure 6(a) presents the TE transmission spectra of the test device and also from a reference waveguide without any tapers. From these spectra, we calculate the transition loss and show the results in Fig. 6(b). The excess loss of each transition is only 0.15 dB at short wavelength of 1535 nm, and increases to 0.3 dB at 1605 nm. The slight increase at longer wavelength may come from the imperfect growth of silicon at the taper tip, which affects longer wavelengths to a greater degree because of larger mode size. The low loss of taper I transition verifies the feasibility of the proposed integration approach, although it would be more convincing if the transition loss of taper II can be characterized as well.

 

Fig. 6 Transition loss. (a) Transmission spectra for a reference waveguide (green) and a test device with 16 transition tapers. (b) Loss per transition as a function of wavelength.

Download Full Size | PDF

5.2 Laser power and spectra

The fabricated chip is diced and polished with one facet coated with high reflectivity (HR) film to increase the reflectivity. The laser is driven by a constant current source. The laser output power from the waveguide facet without HR coating is collected by a large area detector. Figure 7 shows the measured laser output power as a function of injected current at a substrate temperature of 20 °C. As shown in Fig. 7, the laser threshold current is ~80 mA (~2 kA/cm²) and the output power reaches 1.5 mW at 200 mA. The external differential quantum efficiency is calculated as 2.6%. We believe that the limited efficiency may result from (1) poor thermal dissipation because of the use of 3 μm BOX, (2) high cavity loss including transition losses from III-V gain segments to other passive segments and propagation loss for the long cavity, and (3) III-V absorption loss at the transition tapers which cannot be fully pumped by injection current. Nevertheless, we demonstrate single-mode operations up to 180 mA, shown by the spectra in Fig. 7(b), by tuning the in-cavity phase element. The side-mode suppression ratios exceed 40 dB.

 

Fig. 7 Laser power and spectrum. (a) Output power versus injection current. (b) Laser spectra under different injection currents.

Download Full Size | PDF

5.3 Laser linewidth

We measured the laser linewidth by employing a method of intra-dyne coherent detection [16,17], with the set-up presented in Fig. 8(a). The laser output is split into two paths, with a delay provided by 25 km of single mode fiber on one path to de-correlate the two optical signals, and then re-combined into a commercial optical coherent receiver. The coherent detection can produce the phase difference of the beating signals. The differential phase variance between two time instants separated by an interval Τ varies linearly with T, with a slope determined by the laser linewidth ∆ν, namely, ${\sigma }_{\varphi }^2\left(T\right)=〈{\left\{\varphi \left(t\right)-\varphi \left(t-T\right)\right\}}^2〉=\Delta vT$. With this method, we measured a linewidth of 130 kHz for the fabricated long-cavity hybrid laser, as shown in Figs. 8(b) and 8(c). Figure 8(b) presents the electrical fields over a period of 100 μs, from which the phase variations can be extracted.

 

Fig. 8 Laser linewidth measurement. (a) Measurement set-up using intra-dyne coherent detection. (b) Electrical field over a period of 100 μs. (c) Differential phase variation as a function of time interval. The slope is the lase linewidth which is 130 kHz for the fabricated hybrid laser.

Download Full Size | PDF

6. Conclusion

In this paper, we have demonstrated a long-cavity silicon/III-V hybrid laser employing novel fabrication processes which enable flexible light coupling between silicon and III-V waveguides. This new fabrication technique is based on epitaxial or other growth of high-index materials in pre-defined trenches to form low loss transition tapers with flexible silicon thickness in the designated regions without requiring a thick starting silicon thickness. This selective growth technique may be very useful to integrate hybrid lasers with other photonic integrated circuits for future large-scale photonic integration on silicon platforms.