We experimentally demonstrate a silicon photonic chip-scale 16-channel wavelength division multiplexer (WDM) operating in the O-band. The silicon photonic chip consists of a common-input bus waveguide integrated with a sequence of 16 spectral add-drop filters implemented by 4-port contra-directional Bragg couplers and resonant cladding modulated perturbations. The combination of these features reduces the spectral bandwidth of the filters and improves the crosstalk. An apodization of the cladding modulated perturbations between the bus and the add/drop waveguides is used to optimize the strength of the coupling coefficient in the propagation direction to reduce the intra-channel crosstalk on adjacent channels. The fabricated chip was validated experimentally with a measured intra-channel crosstalk of ∼−18.9 dB for a channel spacing of 2.6 nm. The multiplexer/demultiplexer chip was also experimentally tested with a 10 Gbps data waveform. The resulting eye-pattern indicates that this approach is suitable for datacenter WDM-based interconnects in the O-band with large aggregate bandwidths.
© 2020 Optical Society of America under the terms of the OSA Open Access Publishing Agreement
Wavelength division multiplexing (WDM) is commonly used to increase the aggregate data rate for datacenter interconnects. [1–5]. These interconnects have traditionally operated in the O-band of the optical spectrum using coarse WDM (CWDM) so that expensive thermo-electric coolers (TECs) are not required. Further increasing the aggregate data rate while maintaining the cost-effective CWDM grid spacing requires broadband, low-loss and low crosstalk WDM multiplexing and demultiplexing components. To achieve these goals for O-band WDM components, there is a growing trend in exploiting cost-effective silicon photonic solutions. For example, it is predicted that silicon photonics is a viable candidate for data speeds of 800 Gb/s and above, which is a 2x enhancement of the current state-of-the-art [6,7].
Common realizations of silicon photonic WDM multiplexing and demultiplexing components rely on Mach-Zehnder interferometers (MZIs) [3,8–10], arrayed waveguide gratings (AWGs) [11,12], ring resonators [4,13], and Bragg based structures [1,2,5] to perform (de-)multiplexing. However, MZI and ring-based devices become limited in free spectral range (FSR) while reducing channel bandwidth due to increasing cavity lengths, which is undesirable for increasing the number of channels in WDM. Additionally, AWG-based approaches have a large footprint in comparison to these of MZI and ring-based approaches. To achieve high-aggregate data rate WDM interconnects in O-band that can approach a DWDM channel spacing, it is critical to design and demonstrate efficient, chip-scale integrated add/drop filters capable of (de-)multiplexing with low intra-channel crosstalk, low power consumption, and the high thermal stability required so that a TEC is not needed. However, at present the silicon photonics based DWDM realizations with more than 8 channels operating in O-band are still an emerging research topic .
In this manuscript we introduce and experimentally validate a scalable semi-dense WDM 16-channel (de-)multiplexer using the silicon photonic chip-scale platform that advances the technology towards the ultimate goal of a DWDM interconnect that can operate in O-band. We utilize concatenated and apodized 4-port Bragg add/drop filters to provide a large FSR and wavelength channel spacing of about 2.6 nm. While the ultimate goal for DWDM channel spacing in O-band may be comparable to the 50-100 GHz channel spacing commonly used in C-band, in this manuscript we target demonstrating an O-band (de-)multiplexer with a channel spacing of 450 GHz (2.6 nm). This spacing can be considered a semi-dense WDM grid and was chosen to tolerate thermal fluctuations and reduce fabrication constraints. We observe an average channel crosstalk of −18.9 dB and record an eye diagram on one channel of the system at a data rate of 10 Gbps. In the next section we introduce our technique and provide simulations of our design. In Section 3 experimental validation of the fabricated device is presented for a single channel as well as a 16-channel silicon photonic circuit. Summary and perspective on this approach is given in Section 4.
2. Simulation and design of the apodized filter
The 16-channel filter circuit, depicted schematically in Fig. 1(a), is composed of 16 concatenated filters [see Fig. 1(b)] with each filter centered on a 450 GHz semi-dense WDM grid in O-band. Conceptually, these filters are similar to these discussed in prior literature that are optimized for C-band operation with a lower channel count [5,15]. Here we combine prior work that can achieve a large FSR with a drastic increase of the channel count in O-band. In addition, to assure a robust system performance, we develop a novel apodization procedure for side lobe suppression that is essential for increasing the SNR. Both single channel and multiple channel designs are presented. For the 16-channel design, the apodization leads to low intra-channel cross talk for all 16-channels.
2.1 Single channel concept
The single channel Bragg-based add/drop filter, shown in Fig. 1(b), is realized by implementing periodic cladding perturbation to ensure spectral 3-dB bandwidths of about ∼2 nm to realize a semi-dense WDM grid. The spectral bandwidth is determined by the magnitude of the modal perturbation in the cladding, which is minimized by its distance from the input waveguide (e.g., ∼75 nm). The period Λ of the perturbations is chosen to be ∼255 nm to ensure operation within the O-band. Each perturbation has an oval shape (460 nm width, 150 nm length) to enable the implementation of adiabatic changes to the optical mode, which results in a smooth apodization that increases the side lobe suppression. The number of periods, N, is chosen to be 2500. To ensure a large FSR, the propagation constants in the bus waveguide and the drop waveguide need to be very different which is achieved by the choice of the waveguide widths (e.g., 325 nm and 450 nm, for the bus and the drop waveguides, respectively). The apodization of the drop waveguide is performed with a Gaussian profile where the gap between the two waveguides in the propagation direction, z is defined by1(c), demonstrating a sufficient side lobe suppression over a wideband spectra. Next we developed an apodization optimization procedure to maximize the side-lobe suppression.
2.2 Performance optimization
Optimization of the drop-port apodization requires careful selection of the 6 design parameters (drop waveguide width, H, perturbation width, Gmin, a, and N), which is accomplished through parametric sweeps. Each variable is set to a default value, as described in Table 1, and each parameter is swept to evaluate the filter FSR, drop-port power, channel bandwidth, and side- lobe suppression ratio (SLSR). An example sweep of one parameter as the other five parameters remained fixed is shown in Fig. 2. Due to the extensive design space, a local optimization is performed as opposed to a global optimization. First, to ensure a large FSR the widths of the waveguides are used to control the propagation constants. Within the O-band there are two Bragg matching conditions that are satisfied simultaneously: Λ(nthrough+ ndrop) = λ0 and 2Λnthrough = λ1 where nthrough and ndrop correspond to the effective refractive indices of the through and drop waveguides, λ0 and λ1 correspond to the center wavelength of the reflected response in the drop and the input ports, respectively. Increasing the width of the drop waveguide allows us to increase ndrop (i.e., propagation constant) which increases the distance between λ0 and λ1 which we define for this type of devices as an equivalent to FSR [see Fig. 2(a)]. A drop waveguide width of 450 nm is chosen to reduce the effects of conversion to higher-order modes, while maintaining high drop-port power.
The value for H is swept from 0.25 µm to 2.75 µm [see Fig. 2(b)], which decreases the strength of coupling between the waveguides. As H is increased, the estimated drop-port power varies from nearly full coupling to ∼3 dB of coupling. At the values of H where coupled power to the drop-port is maximized, it is observed that there is a tradeoff of the SLSR, which is used to estimate the channel crosstalk in a multi-channel system. A value of 2.5 is chosen for H to prioritize the SLSR. Next, the width of the oval perturbations is swept with the results summarized in Fig. 2(c). As the width increases, the coupling coefficient increases. This increases the power coupled to the output of the drop-port. An enhanced coupling coefficient also increases the 3-dB bandwidth of the channel. Accordingly, a perturbation width of 460 nm is chosen to increase the coupled drop-port power while still supporting a semi-dense WDM grid.
Next, in Fig. 2(d), the effect of the minimum gap between the waveguides is investigated, showing that there exists a maximum at ∼660 nm for both the drop-port power and SLSR. At this value the waveguide system, which behaves as a resonator due to the phase matched signals, are critically coupled. Below this value the two waveguides are over-coupled, resulting in low drop-port coupled power. Similarly, above this value the waveguides are under-coupled and experience degraded power at the drop-port. This value is thus chosen for Gmin. The investigation of the effects of the apodization parameter, depicted in Fig. 2(e), shows that increasing this value reduces both the SLSR and drop-port power for high values of a. However, for lower values of a there is a maximum of ∼3 dB. As a tradeoff we choose a value of a=2.5 as opposed to a=3, to allow a higher drop-port power. Finally, the length of the filter, as controlled by the number of periods N, is swept in Fig. 2(f). Although both the drop-port power and SLSR are increased by increasing the device length, 2500 periods is chosen to limit the device footprint.
3. Experimental results
3.1 Device fabrication and testing configuration
The designed devices were fabricated at Applied Nanotools under multi-project-wafer (MPW) process. Waveguides are patterned using electron beam lithography on a silicon-on-insulator wafer with a 220 nm thick device layer, as shown in Fig. 3(b), and are able to clearly define the perturbations. TiW alloy heaters are deposited above the Bragg add/drop filters for tuning purposes. To verify filter performance, the through-port and drop-port of four concatenated filters are measured. Each filter is identical in design and only differs in Λ, which is increased by 1 nm in each successive channel to shift the center of the operating spectral band. In this configuration, the through-port is connected to the input of the next filter allowing the drop-ports to be freely measured. The measured response of each port was obtained using a broadband amplified spontaneous emission source (Thorlabs S5FC1021S) edge coupled to the chip and collected by a spectrometer (ANDO AQQ6317B). These responses are normalized to the maximum value of the through-port and are shown in Fig. 3.
3.2 Single channel spectra
The measured device response for a single channel, centered at 1310 nm, is shown in Fig. 3. To control the spectral width of each WDM channel, which is assumed to be within 450 GHz, the cladding corrugations are chosen to weakly perturb the waveguide mode, resulting in a 1.5 nm channel width. The channel width is well coordinated with the simulated width of 1.6 nm. The optimization of the apodization of the Bragg filter suppresses the sidelobe response of the drop-port to less than −25 dB, whereas a non-apodized structure can exhibit sidelobe suppression as high as −8 dB .
The reduction in crosstalk reduces the interference level. Under the common assumption that this interference can be treated as noise, the crosstalk lowers the signal-to-noise ratio and increases the required power to maintain a given bit error rate (BER). The crosstalk at adjacent channels correspond to −24 dB and −31 dB, which achieves our system-design goal of −18 dB, which was chosen to reduce the system power as determined by the LEED initiative this work is linked to . Insertion loss from the through-port is −2.8 dB, which is attributed to device length and the low coupling coefficient between the through waveguide and drop waveguide, which is discussed in depth in Section 2.2. The coupling coefficient can be increased by decreasing the value of the parameter H such that the area of the device is maintained at the expense of increase in crosstalk.
3.3 Sixteen-channel add-drop filters circuit
To demonstrate a 16-channel semi-dense WDM (de-)multiplexer, we fabricated a concatenated 16 add/drop filters in series. Each filter is identical in design to the single channel filter discussed in the previous section with the exception of the grating period Λ, which ranges from 250 nm to 265 nm. Increasing Λ by 1 nm has the effect of red shifting the channel by ∼2.6 nm, which falls within the 450 GHz goal of the semi-dense WDM grid. This allows all channels to be spaced between 1290 nm and 1330 nm. Although a 1 nm resolution is below the capabilities of most lithography processes, the filters avoid suffering from this limitation by noticeably increasing L by 2.5 µm over 2500 periods. However, if the difference in length between two filters, N(Λ2-Λ1), is smaller than the lithography resolution the channel spacing will not be resolved. As such, all 16 channels are defined clearly in wavelength without need for active (i. e., temperature) tuning to control the center wavelength. To mitigate lithographic errors, heaters are fabricated above each filter with the capacity to independently redshift the central wavelength of each add/drop filter under its corresponding heater in the range of 15 nm. Average crosstalk falls for all channels is −18.9 dB whereas all crosstalk values can be found in Table 2. This increase over the single channel fiber was caused by the spectral bunching of channels 3-6 due to fabrication tolerance imperfections. Although the channels could potentially be separated through use of the fabricated heaters, simultaneous control of each heater is outside the range of the available testing equipment for this experiment.
Figure 4 shows that the measured through-port response of the device degrades below 1290 nm due to the reflection of the incident power backwards into the input waveguide because of the multiple Bragg phase matching conditions. For our waveguide geometry designs, the FSR for a single channel is about 40 nm as observed in Fig. 4 in the through-port response. A reduction in power on the through-port is recorded at wavelengths below 1292 nm. This corresponds to interference due to the FSR of channels centered around 1330 nm, as visualized in Fig. 2(a). The effects of this loss are mitigated by routing the channels in successive order in wavelength such that channels 1-15 have interacted with the input before channel 16, avoiding the loss caused by the FSR of channel 16.
3.4 High-speed performance of the system
To evaluate the high-speed performance of the (de-)multiplexer, an eye diagram was measured on Channel 9 while a signal was absent from all other channels. A commercial transceiver (Eoptolink EOLP-1696-30-31) with a 10 Gbps non-return-to-zero (NRZ) pattern is used to generate the input signal for the channel. Insertion losses due to fiber coupling (∼11 dB), propagation (∼5 dB), and uncoupled light remaining in the through-port (2.8 dB) lead to a total measured loss of −18.8 dB. The drop-port response is boosted using an optical amplifier (Thorlabs S9FC1132P) and measured using a photodetector (Agilent 1198A). The eye-diagram of the photodetected electrical waveform was generated using a real-time oscilloscope (Agilent DSA91304A). The eye diagram, as shown in Fig. 5, is open at 10 Gbps. Asymmetries in the eye are due to the original signal from the transceiver, which is a directly-modulated laser diode. Eye quality can be improved by reducing insertion loss from chip-to-fiber coupling and reducing the propagation loss via reducing the system footprint by placing filtering two wavelengths at the same space, which is seen previously in literature [17,18]. Additionally, triggering the oscilloscope using a known pattern clock can reduce the signal jitter opening the eye pattern in time. Each channel has a bandwidth determined by the time spent in the filter, which is anticipated to allow operation at higher data rates and frequencies due to this time being short.
In conclusion, a 16-channel O-band WDM (de-)multiplexer with attainable aggregate data speeds of 160 Gbps has been presented. This design has significant potential to be scaled to 400 Gbps for a 16-channel system and 800 Gb/s for a 32-channel system. Individual channels have an insertion loss of ∼−2.8 dB from the input signal, average channel width of 1.5 nm (266 GHz @ 1.3 µm), average channel spacing of 2.6 nm, and an average crosstalk of −18.9 dB. Although the channels are designed to be static, a heater is included for individual channel tuning. Due to the lack of silicon-based O-band (de-)multiplexer that can approach the grid spacing of a DWDM system, this Bragg-based system is strong candidate for future high-capacity O-band WDM interconnects.
Advanced Research Projects Agency - Energy; Defense Advanced Research Projects Agency; Office of Naval Research; National Science Foundation (CBET-1704085, CCF-1640227, DMR-1707641, ECCS-1542148, ECCS-180789, ECCS-190184); Semiconductor Research Corporation; Army Research Office; Cymer.
The authors thank George Papen, Xiaoxi Wang, and Yossef Ehrlichman for fruitful discussions.
The authors declare no conflicts of interest.
1. D. Charron and W. Shi, “O-band add-drop filter in Bragg-grating-assisted Mach-Zehnder interferometers for CWDM,” in Conference on Lasers and Electro-Optics (OSA, 2019), p. JTh2A.44.
2. M. G. Saber, Z. Xing, D. Patel, E. El-Fiky, N. Abadia, Y. Wang, M. Jacques, M. Morsy-Osman, and D. V. Plant, “A CMOS Compatible Ultracompact Silicon Photonic Optical Add-Drop Multiplexer with Misaligned Sidewall Bragg Gratings,” IEEE Photonics J. 9(3), 1–10 (2017). [CrossRef]
3. F. Horst, W. M. J. Green, S. Assefa, S. M. Shank, Y. A. Vlasov, and B. J. Offrein, “Cascaded Mach-Zehnder wavelength filters in silicon photonics for low loss and flat pass-band WDM (de-)multiplexing,” Opt. Express 21(10), 11652 (2013). [CrossRef]
4. P. P. Absil, P. De Heyn, P. Dumon, D. Van Thourhout, P. Verheyen, S. Selvaraja, G. Lepage, M. Pantouvaki, M. Rakowski, and J. Van Campenhout, “Advances in silicon photonics WDM devices,” Proc. SPIE 9010, 90100J (2014).
5. D. T. H. Tan, K. Ikeda, S. Zamek, A. Mizrahi, M. P. Nezhad, A. V. Krishnamoorthy, K. Raj, J. E. Cunningham, X. Zheng, I. Shubin, Y. Luo, and Y. Fainman, “Wide bandwidth, low loss 1 by 4 wavelength division multiplexer on silicon for optical interconnects,” Opt. Express 19(3), 2401 (2011). [CrossRef]
6. N. Eiselt, J. Wei, H. Griesser, A. Dochhan, M. Eiselt, J.-P. Elbers, J. J. V. Olmos, and I. T. Monroy, “First Real-Time 400G PAM-4 Demonstration for Inter-Data Center Transmission over 100 km of SSMF at 1550 nm,” in Optical Fiber Communication Conference (OSA, 2016), p. W1 K.5.
7. E. Maniloff, S. Gareau, and M. Moyer, “400G and Beyond: Coherent Evolution to High-Capacity Inter Data Center Links,” in Optical Fiber Communication Conference (OFC) 2019 (OSA, 2019), p. M3H.4.
8. J. C. Mikkelsen, A. Bois, T. Lordello, D. Mahgerefteh, S. Menezo, and J. K. S. Poon, “Polarization-insensitive silicon nitride Mach-Zehnder lattice wavelength demultiplexers for CWDM in the O-band,” Opt. Express 26(23), 30076 (2018). [CrossRef]
9. D. Munk, M. Katzman, Y. Kaganovskii, N. Inbar, A. Misra, M. Hen, M. Priel, M. Feldberg, M. Tkachev, A. Bergman, M. Vofsi, M. Rosenbluh, T. Schneider, and A. Zadok, “Eight-Channel Silicon-Photonic Wavelength Division Multiplexer With 17 GHz Spacing,” IEEE J. Sel. Top. Quantum Electron. 25(5), 1–10 (2019). [CrossRef]
10. Q. Deng, L. Liu, R. Zhang, X. Li, J. Michel, and Z. Zhou, “Athermal and flat-topped silicon Mach-Zehnder filters,” Opt. Express 24(26), 29577 (2016). [CrossRef]
11. S. Pitris, G. Dabos, C. Mitsolidou, T. Alexoudi, P. De Heyn, J. Van Campenhout, R. Broeke, G. T. Kanellos, and N. Pleros, “Silicon photonic 8 × 8 cyclic Arrayed Waveguide Grating Router for O-band on-chip communication,” Opt. Express 26(5), 6276 (2018). [CrossRef]
12. Z. Liu, J. Zhang, X. Li, L. Wang, J. Li, C. Xue, J. An, and B. Cheng, “25 × 50 Gbps wavelength division multiplexing silicon photonics receiver chip based on a silicon nanowire-arrayed waveguide grating,” Photonics Res. 7(6), 659 (2019). [CrossRef]
13. M. Rakowski, M. Pantouvaki, P. De Heyn, P. Verheyen, M. Ingels, H. Chen, J. De Coster, G. Lepage, B. Snyder, K. De Meyer, M. Steyaert, N. Pavarelli, J. S. Lee, P. O’Brien, P. Absil, and J. Van Campenhout, “A 4 × 20Gb/s WDM ring-based hybrid CMOS silicon photonics transceiver,” in 2015 IEEE International Solid-State Circuits Conference - (ISSCC) Digest of Technical Papers (IEEE, 2015), Vol. 58, pp. 1–3.
14. K. Fotiadis, S. Pitris, M. Moralis-Pegios, C. Mitsolidou, P. De Heyn, J. Van Campenhout, T. Alexoudi, and N. Pleros, “16×16 silicon photonic AWGR for dense wavelength division multiplexing (DWDM) O-band interconnects,” in Silicon Photonics XV, G. T. Reed and A. P. Knights, eds. (SPIE, 2020), p. 13.
15. W. Shi, H. Yun, C. Lin, J. Flueckiger, N. A. F. Jaeger, and L. Chrostowski, “Coupler-apodized Bragg-grating add–drop filter,” Opt. Lett. 38(16), 3068 (2013). [CrossRef]
16. Y. S. Fainman, J. Ford, W. M. Mellette, S. Mookherjea, G. Porter, A. C. Snoeren, G. Papen, S. Saeedi, J. Cunningham, A. Krishnamoorthy, M. Gehl, C. T. DeRose, P. S. Davids, D. C. Trotter, A. L. Starbuck, C. M. Dallo, D. Hood, A. Pomerene, and A. Lentine, “LEED: A Lightwave Energy-Efficient Datacenter,” in Optical Fiber Communication Conference (OFC) 2019 (OSA, 2019), p. M4D.4.
17. K. Ikeda, M. Nezhad, and Y. Fainman, “Wavelength selective coupler with vertical gratings on silicon chip,” Appl. Phys. Lett. 92(20), 201111 (2008). [CrossRef]
18. M. T. Boroojerdi, M. Ménard, and A. G. Kirk, “Two-period contra-directional grating assisted coupler,” Opt. Express 24(20), 22865 (2016). [CrossRef]