Skip to Main Content
OPTICAL interconnects fabricated using Si photonics technology are a promising means of dealing with bandwidth bottlenecks in LSI chips at inter-chip interconnections due to the intrinsic properties of optical signals, including wide bandwidth, low latency, low power consumption, and low mutual interference . We previously proposed a photonics–electronics convergence system in which photodetectors (PDs), optical modulators, and light sources are linked by optical waveguides on a Si substrate, and bare LSI chips are mounted on the substrate by flip-chip bonding and electrically connected to the PDs and optical modulators . This system requires a small, multi-channel, high-density light source capable of high optical output power operation. Since Si itself has no light emission mechanism, much research has been devoted to forming light sources on a Si substrate that are capable of such operation. For example, light sources have been fabricated monolithically on Si substrates by direct growth of emission material , , , laser structures have been formed after wafer bonding of a III/V active region onto a Si or a silicon-on-insulator (SOI) wafer , , , and a hybrid integrated light source has been fabricated by butt coupling the laser facet and the Si waveguide through the use of spot-size converters (SSCs) , , , , . As we previously reported , while the butt-coupling scheme requires using flip-chip bonding and passive alignment technology, it provides a number of advantages, including higher optical output power and lower power consumption than other schemes.
The butt-coupling scheme enables a laser diode (LD) and a Si waveguide platform to be individually optimized to obtain maximum performance without mutual restriction. High performance is of course a significant requirement; another is a simple process for fabricating the LD mounting stage to prevent cost escalation and yield degradation in wafer manufacturing. The SSC structure, the other Si waveguides for optical wiring, and portions of the active devices must be simultaneously formed in the lithography and etching processes. One SSC structure that enables this requirement to be met is the conventional inverse taper waveguide . With this structure, however, the spot size is larger and more sensitive to the width of the waveguide tip. This makes the fabrication tolerance too small for practical application.
In response to these problems, we developed an SSC structure with a simple tridentate Si waveguide with which we were able to achieve low-loss optical coupling and high-density, multi-channel operation of a hybrid integrated light source . We reported its basic characteristics for application to 1.55 μm wavelength light sources, but its application is not limited to light sources of this wavelength. For example, it is possible to use this structure to obtain low coupling loss between a 1.3 μm LD and a trident SSC designed for that wavelength .
In this paper, we discuss the trident SSC design and structure for 1.55 μm and 1.3 μm wavelengths, particularly aspects that were not heretofore described in detail. We then present the SSC fabrication process and the basic optical characteristics of the SSC. The device fabrication is described on the basis of hybrid integration technology applied to an SOI substrate. The fabricated integrated light source operating at 1.55 μm demonstrated over 50 channel output with power variation less than 2.2 dB. The one operating at 1.3 μm and mounting a quantum dot (QD) laser demonstrated the capacity to operate at temperatures up to 100 °C.
The structure of the proposed SSC (shown in Fig. 1), consisting of three tapered Si waveguides, is similar to that of a trident fork. The waveguides are surrounded with an SiO 2 cladding layer. We calculated the guiding mode in this waveguide area for a wavelength of 1.55 μm and the transverse electric (TE) mode.
Fig. 2 shows modal analysis simulation results for a 220 nm waveguide height and 1 μm spacing between the two side waveguides. The inset shows the cross-section structure of the trident SSC at the incident facet. The simulation showed that only the fundamental mode was excited between 100 and 200 nm tip widths. A higher mode existed when the tip width exceeded 200 nm. In contrast, a tip width of 100 nm or less resulted in increased light leakage from the waveguide and consequently increased radiation loss. The side waveguide widths were therefore set to 150 nm in the fabrication process.
The input light coupled to the SSC propagates through the twin waveguide, gradually changing the mode shape, as shown in Fig. 3. There are two optical discontinuities in the lightwave propagation: the facet of the center waveguide and the end facet where the two side waveguides end. The center waveguide should be narrower to reduce the first discontinuity. The center waveguide width was thus set to 100 nm since this width enables superior reproducibility of the waveguide width for fabrication using electron beam (EB) lithography.
Simulation showed that setting both side waveguides widths to 300 nm at the first discontinuity resulted in an excess loss of less than 0.01 dB, as shown in the lower graph of Fig. 3. At the second discontinuity, the excess loss estimated from the simulation results was less than 0.005 dB when the tip width of the side waveguides was less than 100 nm, as shown in the upper graph. These results indicate that optimizing the width of each waveguide in the trident SSC enables low-loss mode conversion.
The tip width and spacing of the two side waveguides were set so that the spot size (3 μm) was the same as that of the LD to be mounted on the Si platform. The spot size of the trident SSC depends on both the tip width and spacing parameters while that of a conventional SSC with an inverse taper waveguide depends on only the tip width. This means the spot size variation of the trident SSC due to manufacturing variation in the tip width should be smaller than that of a conventional SSC. Finally, the coupling efficiency of the trident SSC should be more stable against the variations than that of a conventional one.
We examined the variation in coupling loss with SOI thickness variation. Calculation showed that an SOI thickness variation of about ±10 nm in one SOI wafer results in a variation in the coupling loss of less than ±0.15 dB.
We also calculated the coupling tolerance due to misalignment between the trident SSC and the LD for a 3 μm spot size. The tolerance up to a 1 dB loss increase was ±0.8 μm in both the horizontal and vertical directions. This tolerance is almost the same as that of a double-core SSC structure using SiON  and is larger than the alignment accuracy of the commonly used flip-chip mounting technique , .
We fabricated a trident SSC on an SOI substrate with a 3 μm thick buried oxide (BOX) layer. The waveguides for the trident SSC and those for the optical wiring were simultaneously formed by EB lithography and dry etching. The total length of the trident SSC was 150 μm, including the 40 μm long tapered waveguides. A 2 μm thick SiO2 upper cladding layer was deposited on the Si waveguides. The widths (W in Fig. 2; W1, W2, and W c in Fig. 3) were 150, 300, 100, and 100 nm.
To measure the excess mode conversion losses of the fabricated trident SSC, we formed SSC chains in the Si wire waveguides with various numbers of SSCs. The edge facets of the chains were formed by polishing. This measurement was carried out using lensed fibers with a spot size of about 3 μm and input light with a polarization extinction ratio of over 27 dB. Fig. 4 shows the dependence of the transmittance on the number of incorporated SSCs at wavelengths of around 1.55 μm. Linear fitting showed that the lensed fiber-SSC coupling losses for the TE and transverse magnetic (TM) modes were 0.92 dB and 0.94 dB while the calculated coupling losses between the lensed fiber and SSC were 0.34 dB for the TE mode and 0.62 dB for the TM mode. The excess mode conversion losses for the two modes were 0.55 dB and 0.24 dB at a wavelength of 1.550 μm. Almost the same loss values were obtained in the 30 nm wavelength region. These results show that the polarization dependent characteristic of the trident SSC was relatively low in a wide wavelength region at around 1.55 μm.
We measured the coupling loss between the trident SSC waveguides and the LD by direct butt coupling. The trident SSC facet was formed by polishing. The LD had a Fabry–Perot structure with a 1.55 μm lasing wavelength and a spot size of around 3 μm. Optical coupling tolerance was measured by moving the LD in the horizontal or vertical direction. As shown in Fig. 5, the minimum coupling loss was about 2.3 dB at zero deviation. The tolerance up to a 1 dB loss increase was ±0.9 μm in the horizontal direction and ±0.85 μm in the vertical direction. Because the accuracy obtained in flip-chip mounting an LD can be controlled to less than ±0.5 μm in the horizontal direction by using a precise alignment technique  and less than ± 0.1 μm in the vertical direction by fabrication technology, it is possible to suppress the maximum coupling loss to less than 1 dB.
To measure the manufacturing tolerance, we prepared trident SSCs and conventional inverse taper SSCs with various waveguide tip widths. The inverse taper SSCs were fabricated by EB lithography, dry etching, and deposition of a SiO 2 cladding layer. We measured the coupling loss between them and the LD. As shown in Fig. 6, the coupling loss of the trident SSC remained almost constant even when the tip width was varied from 130 to 170 nm. In contrast, the coupling loss of the conventional SSCs significantly changed when the tip width was varied by only 10 nm. These results indicate that the proposed trident SSC is structurally superior to a conventional SSC in terms of manufacturing tolerance.
We fabricated a hybrid integrated light source on a Si platform. It included our trident SSC and a mounting stage with pedestals to mount an arrayed LD on an SOI substrate. The LD was mounted by passive alignment flip-chip bonding. The trident SSC facet was formed by dry etching. The LD mounting stage had Si alignment marks and Si pedestals that were formed by dry etching after fabricating the trident SSC and removing the BOX layer from the mounting area. An electrode was then formed on the Si substrate . The arrayed LD was mounted on the Si optical waveguide platform with solder bumps by flip-chip bonding and faced the trident SSC facet. The LD consisted of a conventional InP-based quantum well active region. The cavity length and width of the LD were 400 and 600 μm, respectively. The LD had 13-channel arrayed stripes with a 30 μm pitch. Each stripe was a Fabry–Perot laser with a SSC (spot size 3 μm). Current was injected into each active region through a single electrode, and the 13 stripe lasers emitted light simultaneously under uncooled conditions (room temperature). The alignment marks placed on the arrayed LD and the mounting stage made it possible to horizontally position the LD with high accuracy. The Si pedestals enabled highly accurate vertical positioning. The gap between the LD and trident SSC was about 1 μm and was not filled with resin. The measured coupling loss between the mounted LD and SSC was about 2.5 dB. The reflectivity of the SSC facet due to Fresnel reflection was estimated to about 3.5%. It is possible to reduce the reflectivity by filling the gap between the LD facet and SSC face with a gel matched to the refractivity of the SSC facet.
Fig. 7 shows the characteristics of the light power coupled into the Si waveguide and the corresponding values for an arrayed LD mounted on a chemical vapor deposition diamond heat sink with 2.5 dB coupling loss. The threshold currents of the lasers were 11 mA, and the slope efficiencies were 0.3 mW/mA for the LD mounted on the heat sink and 0.26 mW/mA for the LD mounted on the Si platform. The results shown indicate that the light source fabricated using our integration scheme has advantages for application to inter-chip optical interconnects. A comparison of the output power characteristics shows that no degradation occurred as a result of the flip-chip bonding, so high optical output power operation can be obtained by using a low coupling loss SSC. We can use an LD with low threshold and low resistance characteristics without degrading the characteristics of the LD. The light source has superior temperature characteristics due to the relatively high thermal conductivity of the Si stage. In addition, low power consumption operation is feasible because a low power consumption LD can be used without degradation occurring. These features make the hybrid integrated light source well suited for large-scale integrated devices, especially from the viewpoint of low power operation.
To achieve high-density integration of the light-emitting output ports, the Si waveguide platform was designed to have 52 output channels formed by using a doubly cascaded 1 × 2 multi-mode interferometer (MMI) optical splitter for each 13-channel waveguide. The photograph in Fig. 8 shows the fabricated light source. The foot print is about 1.5 mm × 1.5 mm. The observed near field pattern (NFP) and measured uniformity of the output power of all 52 channels are also shown in Fig. 8. The typical spot size of the trident SSC was 3.6 μm in the horizontal direction and 3.0 μm in the vertical direction. The output power variation of the 52 channels was less than 2.2 dB although this uniformity included the variation in coupling loss, waveguide loss, excess branching loss at the MMI, and LD output power. In other experiments, we successfully demonstrated an optical link by using a Si optical interposer monolithically integrated with a 1 × 4 optical splitter, optical modulators with side-wall gratings , optical waveguides , PDs  and trident SSCs, and a 13-channel arrayed LD was hybridly integrated , . The experiments demonstrated that the trident SSC is suitable for integrated device fabrication.
To further increase the number of interconnect channels, multiple 13-channel arrayed LD chips were mounted on a single Si substrate by using flip-chip bonding . We fabricated an integrated light source with two LD array chips, one with a 20 μm pitch (Array LD1) and one with a 30 μm pitch (Array LD2). A photograph of the fabricated 3.2 mm × 1.5 mm light source is shown in Fig. 9. The channels numbered 104 in total since doubly cascaded 1 × 2 MMIs were embedded in the Si waveguides. The NFPs were measured under uncooled conditions (room temperature); optical outputs were observed for all 104 channels. A light source of over 100 channels corresponds to a 1 Tb/s transmitter, assuming 10 Gb/s for each port.
The power consumption of a light source on a Si optical interposer containing mounted LSI chips increases as a result of a temperature increase due to the heat generated in the LSIs (up to about 85 °C or higher) . Because an LD serving as a light source is particularly sensitive to the effects of a temperature increase, we examined a temperature-insensitive hybrid integrated light source on a Si platform to apply a p-type QD LD with high-temperature characteristics . In this section, we describe the design of a trident SSC for a wavelength of 1.3 μm and describe a hybrid integrated light source on a Si platform using a QD LD that introduces output power into a Si waveguide.
Since a high-temperature characteristic QD LD operates at a wavelength of 1.3 μm, an SSC designed for this wavelength is needed to couple a QD LD and a Si waveguide with low coupling loss. To address this need, we redesigned an optimum SSC structure to enable a trident SSC to be applied to a QD LD with 1.3 μm wavelength lasing. We calculated the guiding mode in the dual-core waveguide area in the same way as shown in Fig. 2. Modal analysis simulation showed that only the fundamental mode is excited between 100 and 150 nm tip widths when the waveguide thickness is 200 nm. We set the side waveguide tip width to 120 nm in this fabrication process. The width of the wiring Si waveguide was 350 nm, and the other structural parameters were the same as shown in Fig. 3.
Fig. 10 shows the simulated coupling loss tolerance between a trident SSC and a QD LD when the thickness and tip width of the Si waveguide for the trident SSC were set to 200 nm and 120 nm, respectively. The spot size of the QD LD was set to 4 μm in the horizontal direction and 1 μm in the vertical direction. The minimum coupling loss of the SSC and QD LD without misalignment was estimated to be less than 4 dB. Because the QD LD is characterized as having strong optical confinement in the vertical direction, the spot size is a quite flattened elliptical shape. Therefore, the coupling tolerance in the vertical direction is smaller than that in the horizontal direction. It is possible to suppress the maximum coupling loss to less than 4–5 dB in consideration of the alignment accuracy.
We fabricated a trident SSC for a 1.3 μm wavelength on an SOI substrate with a 200 nm thick Si layer and a 3 μm thick BOX layer by EB lithography, dry etching, and deposition of a cladding layer. The SSC facet was formed by polishing. We measured the coupling loss between the SSC waveguide and a QD LD by direct butt coupling. The QD LD (fabricated at QD Laser Inc.) had a Fabry–Perot laser structure with a 1.3 μm lasing wavelength. The optical coupling tolerance was measured by moving the LD in the horizontal or vertical direction. The minimum coupling loss was about 3.9 dB at zero deviation, as shown in Fig. 11. The tolerance up to a 1 dB loss increase was ±0.7 μm in the horizontal direction and ±0.5 μm in the vertical direction. We also measured the manufacturing tolerance and found that the coupling loss remained almost constant even when the tip width varied from 110 to 130 nm, as shown in Fig. 12 (the same as shown in Fig. 6). These results indicate that the trident SSC is structurally superior in terms of manufacturing tolerance even for a wavelength of 1.3 μm.
We fabricated a hybrid integrated light source with Si waveguides—including the trident SSC and a mounting stage with pedestals to mount a QD LD on an SOI substrate—and mounted the QD LD by passive alignment flip-chip bonding. The cavity length of the QD LD was 600 μm, and the reflectivities of the front and rear sides were 30% and 90%, respectively. Fig. 12 shows the characteristics of the light power coupled into the Si waveguide as a function of injection current, with the measurement temperature ranging from 25 to 100 °C. The threshold current was 5.1 mA at 25 °C and 14.9 mA at 100 °C. Compared to the I–L characteristic of a QD LD mounted on an AlN heat sink [shown in the inset of Fig. 13(a)], the output temperature dependence was similar to that before mounting due to the superior characteristics of QD lasing. Fig. 13(b) summarizes the temperature dependence of the threshold currents and slope efficiencies based on Fig. 13(a). The characteristic temperature from 25 to 100 °C was about 70 K for both mountings, and the degradation in slope efficiency was low. These are well known characteristics of QD LDs. Fig. 13(c) shows the light power degradation from the light power at 25 °C and 60 mA current injections. The degradation exhibited almost the same dependence. This means that there was no coupling loss temperature dependence due to the flip-chip mounting and that the light source produced using our integration scheme works well even at high temperatures. Power degradation was minimal from 25 °C on up, staying within about 1.5 dB up to 85 °C and within about 2.5 dB up to 100 °C. The light source thus has temperature-insensitive characteristics.
From these measurements, we found that coupled light output into a Si waveguide of more than 2 mW was obtained up to 85 °C. We are investigating a Si optical interposer that integrates optical modulators, PDs, and QD LDs, and, given the loss budget that has been obtained at present , have confirmed that the light power introduced into a Si waveguide is high enough to achieve an optical link at temperatures up to 85 °C.
We fabricated a hybrid integrated light source on a silicon (Si) platform using flip-chip bonding with our proposed trident SSC. This SSC can be applied to various wavelengths; we showed an optimum design for 1.55 and 1.3 μm wavelengths. It exhibited low polarization dependence and low coupling loss even though it has only a simple planar waveguide pattern. It was shown to have superior manufacturing tolerance characteristics for both wavelengths. By incorporating it we were able to fabricate a 52-channel light source using an InP-based quantum well arrayed LD operating at 1.55 μm with high output uniformity and to fabricate a 104-channel light source by bonding multiple arrayed LDs. The trident SSC was incorporated in a monolithically integrated Si optical interposer, and its simple fabrication process was demonstrated to effectively prevent manufacturing cost escalation. We also fabricated a hybrid integrated light source on a Si platform using a QD LD operating at 1.3 μm. This light source is suitable for applications with an operating temperature exceeding 100 °C without coupling loss temperature dependence due to the flip-chip mounting. This integrated light source is promising for application to optical interconnects when compact size and high power operation are required.
The authors would like to thank M. Kurihara, Y. Noguchi, and E. Saito for their sample preparations. The authors would like to thank the TIA-SCR Management Office, AIST, for their technical support and device fabrication.
This work was supported by a grant from the Japan Society for the Promotion of Science through the “Funding Program for World-Leading Innovative R&D on Science and Technology (FIRST Program),” initiated by the Council for Science and Technology Policy.
N. Hatori, T. Shimizu, M. Ishizaka, T. Yamamoto, Y. Urino, and T. Nakamura are with the Photonics Electronics Technology Research Association (PETRA), Tsukuba, Ibaraki 305-8569, Japan, and also with the Institute for Photonics-Electronics Convergence System Technology (PECST), Tsukuba, Ibaraki 305-8568, Japan (e-mail: firstname.lastname@example.org; email@example.com; firstname.lastname@example.org; email@example.com; firstname.lastname@example.org; email@example.com).
M. Okano and M. Mori are with the National Institute of Advanced Industrial Science and Technology, Tsukuba, Ibaraki 305-8568, Japan, and also with the Institute for Photonics-Electronics Convergence System Technology (PECST), Tsukuba, Ibaraki 305-8568, Japan (e-mail: firstname.lastname@example.org; email@example.com).
Y. Arakawa is with the University of Tokyo, Komaba 153-8505 Japan, and also with the Institute for Photonics-Electronics Convergence System Technology (PECST), Tsukuba, Ibaraki 305-8568, Japan (e-mail: firstname.lastname@example.org).
Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org.
Back to Top