We demonstrate an integrated triplexer on silicon with a compact size of 1mm by 3.5mm by utilizing a selective area wafer bonding technique. The wavelength demultiplexer on the triplexer chip successfully separates signals at wavelengths of 1310nm, 1490nm and 1550nm with more than 10dB extinction ratio. The measured 3dB bandwidth of the integrated laser and photodetectors are 2GHz and 16GHz, respectively. Open eye diagrams are also measured for the integrated photodetector up to 12.5GHz PRBS inputs.
© 2010 OSA
A triplexer that provides a triple play service (data, voice, and video) is considered a key component for a fiber to the home (FTTH) network. The architecture diagram of a FTTH network is shown in Fig. 1 . The triplexer is used at the customer premise and it is capable of uploading Internet data through the 1310nm channel, downloading Internet data and voice data through the 1490nm channel, and receiving video signals through the 1550nm channel. Therefore, three key components are required for a triplexer: a wavelength demultiplexer which that can separate 1310nm, 1490nm, and 1550nm channels [1–5], a 1310nm laser and two photodetectors responsive at the wavelengths of 1490nm and 1550nm.
Lasers and photodetectors with different epitaxial layer structures may be easy to be realized in pure III-V materials individually, but are harder to be implemented if we want to monolithically integrate both of them on a silicon chip. Traditionally, researchers developed these three parts separately [2–5] and combined them together using passive assembly . Here we demonstrate a novel selective area wafer bonding technique to integrate both the laser and photodetector III-V films on a patterned silicon-based triplexer chip using a single bonding process. After the III-V dies and the silicon wafer are cleaned, the bonding process can be done in less than 5 hours, which includes the substrate removal process for the bonded III-V dies. Our layout results in 18 triplexers/cm2. Much higher density is possible for a more efficient mask design. The integrated lasers and photodetectors were both designed on a hybrid silicon platform and most post-bond processes, such as dry etching, wet etching, and dielectric coating processes used here, are compatible with the standard industrial CMOS process allowing for this device to be integrated with CMOS circuits. The present approach uses gold metallization, which is not CMOS compatible. The demonstration of the selective area wafer bonding on an integrated triplexer also shows the integration possibility of multiple different III-V films on a host wafer to achieve more complex and functional silicon photonic circuits for future applications.
2. Selective area wafer bonding
A novel selective area wafer bonding technique has been developed to integrate both the 1310nm laser and the longer wavelength photodetector III-V films onto a patterned silicon-on-insulator (SOI) triplexer chip with a single bonding step. This is similar to the single III-V die bonding process described in  with the addition of a thin and compressible Al plate in the bonding fixture to compensate for the thickness differences between these different III-V dies. Since the bonding process will be performed in a 300°C oven, the graphite fixtures are used to supply the bonding pressure in this oven. The thickness differences between different III-V dies result from the thickness variation of the InP substrate for different wafers and also from different III-V epitaxial structures causing a variation of bonding pressure across these different III-V dies. The Al plate helps to compensate these thickness differences and allows the bonding pressure to be transferred uniformly to each bonded die. Figure 2 shows the diagram of the selective area wafer bonding process.
The functionality of the selective area wafer bonding process was first demonstrated by bonding three identical III-V die onto an SOI chip, as shown in Fig. 3 , after III-V substrate removal. Figure 3(a) shows the bonding result using the bonding process described in  without a thin Al plate and the picture shows significant InP delamination and peeling after substrate removal and is not suitable for device fabrication. The problem occurs because even for three identical dies from the same III-V wafer, dies from different locations of the same wafer will still have thickness difference from a few hundred nanometers to couple microns due to the non-uniformity at the backside of the wafer or the non-uniformity from the variation of epitaxial layer growth process. Figure 3(b) shows good bonding results after III-V substrate removal when a thin Al plate is used in the bonding fixture. The compressible Al plate will deform allowing better contact between the III-V dies and graphite fixtures so that the bonding pressure is more uniformly transferred to each bonded III-V die. Almost 100% of the III-V area was bonded after substrate removal demonstrating the improvement obtainable by using this bonding technique and the ability to bond multiple III-V dies with a single bonding step.
Figure 4(a) shows the selective area wafer bonding result of two 1310nm laser III-V dies and one long wavelength photodetector III-V die onto a 1cm2 SOI triplexer chip after substrate removal, showing good bonding result with more than 95% bonded area. This technique allows multiple different III-V films to be bonded onto any area on a silicon chip. Figure 4(b) and Fig. 4(c) show the schematic and top view micrograph of an SOI triplexer with an integrated 1310nm ring laser and integrated 1490nm and 1550nm photodetectors on it.
3. Demultiplexer design
Many solutions have been investigated to demultiplex 1310nm, 1490nm and 1550nm channels, such as thin film filter, cascaded Mach-Zehnders (Fourier transform filters), and gratings [1–5]. Here, by tapering the input and output waveguide widths of the multimode interferometers (MMIs), 1310nm, 1490nm and 1550nm channels can be separated by using an MMI multiplexer cascaded with a Mach-Zehnder interferometer (MZI) demultiplexer in a compact footprint of 0.15mm by 2.5mm. In this paper, ridge waveguides on SOI are used with a waveguide height of 0.4μm and rib etch depth of 0.2μm. The MMI multiplexer is used to combine the 1310nm channel with the 1490nm/1550nm channels, cascaded with the MZI to separate the 1490nm and 1550nm channels.
The MMI used to multiplex the 1310nm upstream channel with the 1490nm and 1550nm download channels uses tapered waveguide widths for its input and output (I/O) waveguides, which is as shown in Fig. 5(a) . This will increase the depth of focus for each channel in the MMI and allow the 1490nm and 1550nm channels to be imaged at the same output port while the 1310nm channel is imaged at the other port with a shorter MMI length than traditional MMI designs. The simulation is done with Beamprop simulation software and the final dimension of the MMI multiplexer is 6μm by 629μm with the I/O waveguides being tapered from 1μm to 3μm over a 30μm distance. The schematic and simulated and measured spectral responses of the MMI multiplexer are shown in Fig. 5. The measured response in Fig. 5(c) shows a reasonable match to the simulated result in Fig. 5(b). Over 100nm pass-band centered at 1310nm and 10dB extinction ratio between 1310nm and 1490/1550nm responses are achieved, which satisfies the ITU-T G.984.2 standard for a triplexer. The insertion loss of this MMI is measured to be ~8dB.
The cascaded MZI used to separate the 1490nm and 1550nm channels consists of a 1x2 and a 2x2 3dB MMI power splitters respectively at the input and output of two delay lines with a length difference of 15.24μm. The schematic and the measured spectral response of the MZI demultiplexer are shown in Fig. 6 . These two phase delay lines are designed to have a zero phase shift for 1550nm channel and a π phase shift for 1490nm channel so that they can be separated at different output ports as indicated in Fig. 6(a). The dimension of the 2x2 3dB power splitter is 7.92 μm by 90 μm with the I/O waveguides being tapered from 1.5μm to 2μm over a 30μm distance and the waveguide width of the delay lines is 1.5μm. The dimension of the 1x2 3dB power splitter is the same as the 2x2 3dB power splitter except for it has only one input waveguide. The detailed dimension of the 2x2 3dB power splitter is described in Fig. 6(b). The measured spectral response of the MZI in Fig. 6(c) shows ~20dB extinction ratio for both the response at 1490nm and 1550nm ports, which satisfies the ITU-T G.984.2 standard for a triplexer. The insertion loss of this MZI is measured to be ~4dB.
The designed MMI multiplexer and the cascaded MZI demultiplexer successfully separate 1310nm, 1490nm and 1550nm channels with more than 10dB extinction ratio which satisfies the ITU standards for a triplexer and the only insufficiency is the insertion loss. The insertion loss of the MMI multiplexer and the cascaded MZI demultiplexer are higher than what we expected and this could be due to the discrepancy between the real and simulated optical image spot sizes and positions in the MMIs used for both sections. This could be further improved by adding a heater on top of the MMI multiplexer or MZI demultiplexer to fine tune the performance of both sections or by iterating the design to achieve the optimum device dimension.
4. Integrated 1310nm ring laser on hybrid silicon platform
A detailed profile and SEM picture of the integrated 1310nm hybrid ring laser on an SOI triplexer chip are shown in Fig. 7 . It consists of an SOI directional coupler section, two taper sections for smooth optical mode transition from the pure silicon waveguides to the hybrid III-V section, and a hybrid III-V gain section. The waveguide width in the hybrid section is 1.5μm and the 1310nm III-V layers are designed so that the optical mode in the hybrid section has 5% overlap with the quantum wells . In the directional coupler section, the waveguide width is 0.8μm and the gap is 0.6μm. The ring radius is 150μm, the directional coupler length is 173μm, and the final cavity length is ~1600μm.
The measured L-I curve and frequency response are shown in Fig. 8 . The lasing wavelength is centered at 1348nm and the threshold current is ~330mA with a maximum on-chip measured power of 6.3mW at 540mA current injection. The red-shift of the lasing wavelength is due to the heating effect of the ring laser when it is biased at a high current. The laser 3dB bandwidth is measured to be ~2GHz by direct modulation, which satisfies the 1.25GHz requirement of ITU-T G.984.2 standard for a triplexer.
5. Integrated PIN photodetector on hybrid silicon platform
The device profile and epitaxial layer structure of a PIN photodetector on the hybrid silicon platform are shown in Fig. 9 where w is the mesa width and d is the intrinsic layer thickness. The III-V epitaxial layers of the PIN photodetector were first bonded on to the silicon ridge waveguide with a waveguide width of 1.5μm and the post-bond process, such as epitaxial layer etching and metal deposition, were performed afterwards to complete the photodetector. Since the absorption band of InGaAs covers 1490nm and 1550nm, the same photodetector design was used for 1490nm and 1550nm channels and signals from these two channels were demultiplexed by the MZI demultiplexer to their corresponding photodetectors. Two characteristics of a photodetector are important when designing a high-speed photodetector: one is the responsivity and the other is the 3dB bandwidth. Higher responsivity can be achieved by increasing the mode overlap with the absorption layer of a PIN photodetector by increasing the thickness of this layer. The high speed performance of a PIN photodetector is mainly limited by two factors: one is the transit-time needed for carriers to transit between p and n layers and the other is the RC limitation. Generally, the transit-time response and the RC response of a PIN photodetector are given in [7,8].First, we choose the mesa width w to be 12μm for the device mechanical stability since the trench width of the ridge waveguide is 10μm and the device length to be 120μm to guarantee that most of the light could be absorbed over this length. The only parameter remains to be designed is the intrinsic layer thickness. The optimum choice of the intrinsic layer thickness of a PIN photodetector should be designed to achieve both high responsivity and high 3dB bandwidth. The 3dB bandwidth simulation of a PIN photodetector with a mesa width of 12μm and a length of 120μm for different intrinsic layer thicknesses is shown in Fig. 10 . The intrinsic layer thickness is chosen as 0.5μm in this work which results in a simulated responsivity of 0.9A/W and a simulated 3dB bandwidth of ~12GHz.
Figure 11 shows the measured frequency response, eye diagrams, and bit error rate results of the integrated PIN photodetector of the triplexer measured at 1550nm. The 3dB bandwidth of the PIN photodetector is ~16GHz, which is close to the simulated 3dB bandwidth of 12GHz and satisfies the 2.5GHz requirement of ITU-T G.984.2 standard. The photodetector eye diagram measurement setup is shown in Fig. 12 and open eye diagrams are measured up to 12.5Gb/s PRBS inputs. The measured bit error rates at 5Gb/s and 10Gb/s PRBS are shown in Fig. 11(c).
6. Link tests
The link tests of the integrated triplexer are done by measuring the upstream channel and downstream channel responses of the triplexer chip. For the upstream channel, the light generated from the electrically biased and modulated 1310nm ring laser will go through the MMI multiplexer for the 1310nm and 1490/1550nm signals and be collected at the single I/O port of the triplexer by a lensed fiber, which is shown in Fig. 4(b). During this measurement, the amplified spontaneous emission (ASE) light source will be forward biased at 30mA and seed the ring laser in the clockwise direction so that the ring laser will be lasing only in the clockwise direction . For the semiconductor optical amplifier (SOA) in Fig. 4(b), in order to simplify the measurement complexity, it will be slightly forward biased at 2mA to prevent optical absorption and sending too much feedback to the ring laser cavity. The measured 3dB frequency response for the upstream channel is shown in Fig. 8(b) and it shows a 3dB bandwidth of 2GHz which satisfies the 1.25GHz requirement of ITU-T G.984.2 standard for a triplexer.
The downstream channels were characterized by launching the input light at the input port of the triplexer and monitoring the received photocurrents of the integrated photodetectors at the 1490nm and 1550nm output ports of the MZI demultiplexer. However, since the maximum available input power over the wide wavelength range from 1490nm to 1570nm is only 2mW and the additional 10dB fiber coupling loss exists in this measurement, the maximum measured photocurrent is ~16μA and further high speed measurements for the downstream link tests are thus prohibited. The maximum received photocurrent can be improved by applying the double stage taper structure at the I/O waveguide of the triplexer to reduce the fiber coupling loss for the future design . Since the integrated PIN photodetector shows a 3dB bandwidth of 16GHz, the next version of the triplexer with the double stage taper structure design will be more applicable for real commercial use.
We demonstrate the first integrated triplexer on silicon using the hybrid silicon platform with a compact size of 1mm by 3.5mm. One 1310nm Laser and two photodetectors responsive at 1490nm and 1550nm wavelengths are successfully integrated on a silicon triplexer chip by using selective area wafer bonding technique. The demultiplexer of the triplexer successfully separates signals at 1310nm, 1490nm and 1550nm channels with over 10dB extinction ratio. The measured 3dB bandwidth of the integrated 1310nm ring laser and integrated 1490nm/1550nm photodetector are 2GHz and 16GHz respectively. Open eye diagram of the PIN photodetector was measured up to 12.5Gb/s. The developed selective area wafer bonding technique makes it possible to integrate multiple different III-V films on a single silicon chip with a compact size and versatile functionalities for future applications.
The authors thank Jag Shah, Michael Haney, Alexander Fang, and Hui-Wen Chen for useful discussions. The UCSB research was supported by Intel and the Defense Advanced Research Projects Agency (DARPA) contract W911NF-05-1-0175.
References and links
1. Y. Han, Y. Park, S. O. Park, J. Shin, D. Kim, C. Park, S. Park, Y. Kwon, D. Lee, W. Hwang, and H. Sung, “A PLC-based optical sub-assembly of triplexer using TFF-attached WDM and PD Carriers,” ETRI J. 28(1), 103–106 (2006). [CrossRef]
2. W. Chen, B. Little, W. Chen, S. Chu, J. Hryniewicz, D. Gill, O. King, F. Johnson, R. Davidson, K. Donovan, and J. Gibson, “Compact, Low Cost Chip Scale Triplexer WDM Filters,” OFC PDP12 (2006).
3. S. Bidnyk, D. Feng, A. Balakrishnan, M. Pearson, M. Gao, H. Liang, W. Qian, C.-C. Kung, J. Fong, J. Yin, and M. Asghari, “SOI waveguide based planar reflective grating demultiplexer for FTTH,” Proc. SPIE 6477, 64770F 1–6 (2007)
4. N.-N. Feng, D. Feng, H. Liang, W. Qian, C.-C. Kung, J. Fong, and M. Asghari, “Low-loss polarization –insensitive silicon –on-insulator-base WDM filter for triplexer applications,” IEEE Photon. Technol. Lett. 20(23), 1968–1970 (2008). [CrossRef]
5. T. H. Lee, D. H. Lee, and Y. C. Chung, “Design and simulation of fabrication-error-tolerant triplexer based on cascaded Mach-Zehnder interferometers,” IEEE Photon. Technol. Lett. 20(1), 33–35 (2008). [CrossRef]
6. H.-H. Chang, A. W. Fang, M. N. Sysak, H. Park, R. Jones, O. Cohen, O. Raday, M. J. Paniccia, and J. E. Bowers, “1310nm silicon evanescent laser,” Opt. Express 15(18), 11466–11471 (2007). [CrossRef] [PubMed]
7. J. E. Bowers and C. A. Burrus, “Ultrawide-Band Long-Wavelength p-i-n Photodetectors,” J. Lightwave Technol. 5(10), 1339–1350 (1987). [CrossRef]
8. K. S. Giboney, M. J. W. Rodwell, and J. E. Bowers, “Traveling-Wave Photodetector Design and Measurements,” IEEE J. Sel. Top. Quantum Electron. 2(3), 622–629 (1996). [CrossRef]
9. A. W. Fang, R. Jones, H. Park, O. Cohen, O. Raday, M. J. Paniccia, and J. E. Bowers, “Integrated AlGaInAs-silicon evanescent race track laser and photodetector,” Opt. Express 15(5), 2315–2322 (2007). [CrossRef] [PubMed]
10. A. Barkai, A. Liu, D. Kim, R. Cohen, N. Elek, H.-H. Chang, B. H. Malik, R. Gabay, R. Jones, M. Paniccia, and N. Izhaky, “Efficient mode converter for coupling between fiber and micrometer size silicon waveguides,” IEEE International Conference on Group IV Photonics, WC4 (2007).