Open Access
Add to BibSonomy
Abstract
We propose a novel approach to demonstrate simultaneous multi-wavelength locking during temperature changes in a silicon photonic polarization insensitive microring-based wavelength division multiplexing (WDM) receiver. The DC component of a single monitoring photodetector at the through port of the microring filter array is exploited as a feedback signal with no additional power consumption. This feedback signal is used in control circuitry to properly tune the microring filters using ohmic heating, thus creating a feedback loop for thermal adaptation. We describe the necessary information, specifically each microring filter’s room temperature resonant wavelength and tunability, which can be used to calibrate and achieve proper wavelength configurability and locking. In addition, we describe a simple control algorithm based on an adaptive gradient method often used in machine learning, allowing the receiver to endlessly demultiplex at different temperatures. We successfully achieve thermal adaptation over a temperature range >37°C and demultiplex a 4 × 25 Gb/s on-off-keying signal of 150 GHz channel spacing, all while the polarization is scrambling.
© 2017 Optical Society of America
1. Introduction
Silicon photonics is a disruptive technology platform for optical communications and interconnects, from solving the capacity crunches of fiber telecommunication networks to enabling increased data traffic in datacenters/supercomputers and fulfilling stringent power/latency/capacity requirements for multi-core processors. Wavelength division multiplexing (WDM) is a key technique to scale up the capacity; however, silicon photonics often encounter significant difficulties to offer robust WDM circuits [1]. Firstly, the effective index of silicon waveguides is highly sensitive to fabrication deviations. In WDM filters, this makes the design wavelength difficult to achieve. This also leads to high insertion loss and high channel crosstalk for conventional WDM filters such as arrayed waveguide gratings (AWGs) [2,3]. Secondly, the thermo-optic coefficient of silicon is large, attributing to high sensitivity to ambient and on-chip temperatures. Thirdly, silicon waveguides are highly birefringent, requiring polarization diversity especially for receiver applications. In summary, robust polarization and temperature insensitive silicon photonic WDM receivers have yet to be developed.
To mitigate these effects, one approach that can be used is through the use of ohmic heaters to thermally tune microring filters [1,4]. Thermal tuning in polarization insensitive WDM receivers has been done in previous works [5–7], and automated microring filter tuning has been explored [8–19]. Designs of future microring-based WDM receivers will require compact, simple, and low power closed-loop control systems for thermal adaptation. Also, they will require systems that manage multi-wavelength locking and photonic circuit branch synchronization for TE and TM polarizations. Compact and simple photonic integrated circuits can increase the bandwidth per footprint and minimize the energy per bit; and low power systems are much needed for short-reach optical communications. In this paper, we take a polarization insensitive WDM receiver similar to that of [6] and [7], but develop an automated system with some key differences. Inspired by our approach for an automatically tuned microring-based WDM transmitter with a single monitor [20], we utilize a single monitoring photodetector (MPD) for the microring-based WDM receiver in this paper. The MPD is integrated at the combined through port of the WDM receiver allows locking of multiple wavelengths simultaneously. Instead of using the optical WDM signal’s RF power such as in [20], we exploit the DC component of the signal to provide closed-loop control. This demonstrated technique, together with that in [20], highlights the feasibility of wavelength locking of a photonics system with multiple rings rather than a single ring. The use of a single monitoring signal could reduce the power consumption and also simplify the control circuitry.
2. Microring closed-loop locking system
A schematic of the polarization insensitive WDM receiver and closed-loop locking system used in this paper is depicted in Fig. 1. The input WDM signal is split into two polarization branches by a polarization beam splitter (PBS), and each branch is wavelength demultiplexed by cascaded second-order microring filters and photocurrent combined with germanium photodetectors (PDs) at the drop ports in between the branches [6]. The through ports of each branch are combined by a polarization beam combiner (PBC) and then are detected by an MPD. The MPD outputs an electrical signal that is proportional to the amount of optical power. This electrical signal can be used as feedback to the circuitry that simultaneously controls all the microring heaters. When no microrings are locked to the input wavelengths, the MPD returns the maximum signal; and when all microrings are properly locked and all the wavelength channels are dropped to the PDs, the MPD returns the minimum signal.
Fig. 1 Proposed closed-loop control of a polarization insensitive microring WDM receiver with the detection of the DC signal in the combined through port. PBS/C: polarization beam splitter/combiner; PD: photodetector; MPD: monitoring photodetector. Picture of an unpackaged fabricated WDM receiver in the inset.
The advantage of this locking mechanism is simplification of the optical circuit by exploiting only one monitor signal. Also, we only require the DC components of the WDM signal, therefore making the mechanism independent of the modulation format. More importantly, by only using the DC components of the WDM signal, we can convert the photocurrent outputted by the MPD to a voltage with a single resistor, requiring no additional power to amplify the MPD signal. Interestingly, this design has an inherent pitfall due to the polarization diversity. Figure 2 illustrates the range of MPD responsivities of a WDM receiver configured for 2 input channels. When a multi-wavelength signal is present, the MPD signal will have multiple minima.
Fig. 2 Illustrated calculations of MPD responsivity versus the heating powers of the first pair of microring heaters for (a) TE, (b) TM, and (c) mixed polarization inputs. An intersection of a solid and dashed line represents a minimum with no demultiplexing.
When the input signal is purely TE-polarized or TM-polarized, a minimized MPD signal will result in proper demultiplexing – see Figs. 2(a) and 2(b). However, when the input signal is of a mixed polarization, potentially no demultiplexing could occur despite having a minimized MPD signal – see Fig. 2(c). The first filter in the TE branch could properly lock on to the first wavelength channel, but the first filter in the TM branch could lock on to the second wavelength channel. We prevent this pitfall by tuning each TM microring filter proportionally to the tuning of its corresponding TE microring filter and vice versa. This is possible as long as we can calibrate the device based on its microring filters’ room temperature resonances and tunabilities. By using proper calibration of the TE and TM microring heaters, we can demultiplex at any polarization input.
3. Wavelength locking and tracking algorithm
The key method that is used in the wavelength locking and tracking algorithm is the adaptive gradient (ADAGRAD) method – an optimization algorithm which is often applied to machine learning problems [21]. ADAGRAD maintains a good balance between computation simplicity and convergence rate. Using other vanilla gradient methods such as stochastic gradient descent (SGD) and mini-batch gradient descent may result in increased fluctuations of the feedback signal and the demultiplexed signals, and therefore, an increase in the BER. Quasi-Newton methods such as the Broyden-Fletcher-Goldfarb-Shanno (BFGS) algorithm and its limited memory version (L-BFGS) may prove to be overly complex computationally and therefore be counterintuitive approaches for simple microring-based systems. For WDM receiver control, we implement the ADAGRAD method as
where P is the power applied to a microring heater and VMPD is the feedback voltage signal obtained from the MPD. η is the learning rate of the method (by default, η = 0.01), Gn is the sum of the squares of the past gradients up to the nth iteration, and ε is a smoothing term (typically ε = 1 × 10−8) to avoid zero in the denominator. The nth gradient ∇VMPD(Pn) is approximated using the forward difference method.The algorithm is programmed into a microcontroller that reads the feedback voltage signal from one of its analog-to-digital converters (ADCs) and sets a voltage on the microring heaters via its digital-to-analog converters (DACs). We illustrate the algorithm as a flow chart in Fig. 3. At initialization, the heaters are set such that the resonance of each microring filter is set at a point away from the WDM channels to establish that none are filtering at the same wavelength, sending all the optical power to the MPD. This allows the largest dynamic range to lock on to each wavelength channel. The ADAGRAD method is then used to minimize the signal, consecutively tuning and locking each microring filter in the TE branch on to a wavelength channel. The order of the locking can be configured into the algorithm. Each TM microring filter is paired (or mapped) to its corresponding TE microring filter, based on their room temperature resonances and tunabilities. Therefore, when the resonance of a TE microring filter is changed by a certain heating power, the TM microring filter is tuned proportionally to that change, locking on to the same wavelength channel and avoiding the possibilities of partially demultiplexed signals. Subsequent fine tuning of the TM microring filters is done with the ADAGRAD method. After the final microring filter is initially tuned, the ADAGRAD method is endlessly run over each microring filter to maintain the best microring tuning with temperature drift.
Fig. 3 Algorithm flow for wavelength locking and tracking. The algorithm minimizes the MPD signal by tuning each microring filter, dropping all the optical power to the high-speed germanium PDs.
4. Simultaneous four-channel demultiplexing
The proposed locking method is demonstrated using a silicon photonic polarization insensitive WDM receiver chip based on an inverse taper for edge coupling, an on-chip PBS and PBC, wavelength-tunable second-order microring filters, and monolithic germanium PDs [6]. The free spectral range (FSR) for each microring is about 6.4 nm with room temperature resonances and tunability experimental data listed in Table 1. TE branch microring tunability is inherently larger than that of the TM branch due to larger optical confinement in the waveguides. Tunability variation can be attributed to different metal volume in the microrings’ proximities. The average resistances of the metal heaters are 485 Ω and 588 Ω for TE and TM microrings respectively. The heater metal width is 2 μm and each second-order ring has a single heater which is symmetrically laid on its two rings to relax the misalignment impact from fabrication errors. Similar devices have shown that the worst-case channel crosstalk for 200 GHz and 100 GHz channel spacing is less than −21 dB and −12 dB respectively [6]. The fiber-to-PD responsivities of the MPD as the device temperature changes from 23 °C to 60 °C are shown in Fig. 4. The wavelength red shift is approximately 0.092 nm/°C.
Table 1. Microring Filter Resonance and Tunability
Fig. 4 Fiber-to-PD responsivity of the MPD when the WDM receiver is at 23 °C (blue) and 60 °C (red) – intermediate responsivities in light gray. The microrings in the TE branch are more sensitive to the device temperature than those in the TM branch, resulting in a change to the general responsivity profile at higher temperatures.
In this experiment, we launch four WDM channels of 1550.52 nm, 1551.72 nm, 1552.93 nm, and 1554.13 nm (150 GHz channel spacing) through a wideband optical modulator driven by an on-off-keying (OOK) pseudorandom binary sequence (PRBS) with a length of 215-1, generating a 4 × 25 Gb/s WDM signal. The modulated WDM signal passes through a 5 km long optical fiber with a dispersion of 17 ps/km/nm to decorrelate the signals. The WDM signal also passes through a polarization controller/scrambler, is amplified by an erbium-doped fiber amplifier (EDFA), and is finally launched into a packaged receiver chip.
Each drop port PD is reverse biased at 2 V and can be used to observe the electrical eye diagram when their corresponding microrings are locked onto a wavelength channel. Since the signals are decorrelated by the 5-km fiber, we can determine if there is any significant channel crosstalk in the eye diagrams. The through ports of each polarization branch are combined to be detected by the MPD, and a 1 kΩ resistor converts the MPD’s photocurrent into a voltage signal. This simple conversion provides sufficient dynamic range, so a transimpedance amplifier (TIA) is unnecessary, resulting in a zero-power consumption feedback signal. A microcontroller with a clock speed of 41.78 MHz is used as the control circuit. One of its 12-bit ADCs is used to read the signal from the MPD, and eight of its 12-bit DACs are used to control the heaters over the microring filters. The number of bits in the ADC and the chosen feedback resistor can limit the detection sensitivity of the feedback signal. Also, the number of bits in the DACs limits the resolution of the voltage driving the microring heaters. Heating power is the square of the DAC output voltage, so this is even more of a concern for heaters tuning at a high voltage and therefore potential instability in the received bit stream. However, these can be easily improved with higher-bit ADCs and DAC, and in our experiments, the current bit number is sufficient enough. Also, the electrical bandwidth of the single control circuit loop – including the MPD, the microcontroller, and all microring heaters – is approximately 120 kHz.
We demonstrate active wavelength locking of a 4 × 25 Gb/s WDM signal as shown in Fig. 5. At initialization, we de-tune the microring filters and send most of the optical power to the MPD. At this point, the microring filters are set to have some resonant wavelength between 1555.33 nm and 1555.72 nm. Then we consecutively lock on to each wavelength first on the TE branch and finally on the TM branch. While locking, the heaters have ambient effects – thermal crosstalk – towards microrings for which they are not designed to tune. The algorithm systematically compensates for this and fine tunes the microrings. We configure the system such that the locked wavelength for PD1 is 1550.52 nm, PD2 is 1551.72 nm, PD3 is 1552.93 nm, and PD4 is 1554.13 nm. In this configuration, their corresponding microring filters have the maximum tuning range for when the device temperature increases from room temperature. The heaters are configured to tune each microring filter approximately over a 7.5 nm range – more than one FSR.
Fig. 5 Locking and demultiplexing of a mixed polarization 4 × 25 Gb/s WDM signal. Initialization runs only a few times, hence the eye diagram for PD3 not being completely closed before locking. After each microring filter locks on a wavelength, the corresponding eye diagram widens. In some cases, due to thermal crosstalk, the currently controlled heater slightly tunes/detunes neighboring microrings, but the algorithm fine tunes the microrings after initial locking.
Furthermore, we demonstrate tracking while a temperature controller cycles the device temperature between 23 °C and 60 °C and a polarization controller scrambles the input polarization. The 150 GHz WDM spectrum observed at the input of the device is shown in Fig. 6(a), modulating at 25 Gb/s, and the temperature applied to the entire device over time is shown in Fig. 6(b). Figure 6(c) illustrates the electrical eye diagrams from the four high-speed PDs with actively tracked and simultaneously demultiplexed 25 Gb/s OOK channels. All eye diagrams are shown to have a clean opening during polarization scrambling and over a temperature change >37 °C. PD3 has the worst eye diagram due to higher polarization dependent loss from its microrings. The main limitation to the temperature control range is the used epoxy between the input/out fibers and the facet of the device, which begins to liquefy at temperatures near 70 °C. To avoid increased insertion loss due to misalignment, we set the max temperature to a safe 60 °C. We have also characterized the bit error ratio (BER) for each wavelength channel as depicted in Fig. 6(d). The 4 × 25 Gb/s WDM signal is launched with various launching powers, demonstrating reasonable demultiplexing. No TIAs are packaged with chip, so instead we use an external power amplifier with a gain of 30 dB, limiting the sensitivity. In these experiments, we have maintained tracking with temperature rates ≤0.64 °C/second with BER rippling roughly maintained within an order of the average BER. The tuning speed of silicon photonic devices with on-chip ohmic heaters is typically on the order of microseconds [4], so the speed bottleneck would come mainly from the feedback loop containing the DACs and ADCs as well as the efficiency of the tracking algorithm. Channel power fluctuation should not drastically affect the closed-loop control unless the fluctuation is on the order of the control circuitry’s electrical bandwidth, which may result in an incorrect estimation of the gradient.
Fig. 6 (a) Modulated optical spectrum. (b) Temperature cycle during wavelength locking. (c) 25 Gb/s received electrical eye diagrams and (d) BERs from the 4 high-speed PDs during both temperature cycling and polarization scrambling.
5. Conclusion
In summary, we have proposed and demonstrated simultaneous wavelength locking and tracking with a polarization insensitive WDM receiver. By using a single MPD at the combined through port of the microring filter array, we can implement a feedback control loop for automatic control of all the microrings, which distinguishes this work from previous reports of single microring filter locking techniques. This novel approach enables low power consumption by the use of a single monitoring signal for multiple channels and has no additional power requirement to amplify this signal. Moreover, we synchronized the tuning between two polarization branches. Opposite microring filters from two branches may mix two wavelength channels, but synchronizing the tuning avoids this pitfall. With this current setup, the full dynamic range of a microring filter can be detected under 7 bits of resolution. If we utilize most of the ADC’s 12-bit dynamic range, we estimate that this setup can lock as many as 36 microring filters (or 18 wavelength channels). With a higher-bit ADC, we can potentially push the amount of microring filters to higher numbers. Also, if there were no limitations in the temperature control setup, this method should be able to track over a temperature range as high as 70 °C.
Funding
Part of this project is funded by Intelligence Advanced Research Projects Activity (IARPA) under the SPAWAR contract number N66001-12-C-2011.
References and links
1. P. Dong, “Silicon Photonic Integrated Circuits for Wavelength-Division Multiplexing Applications,” IEEE J. Sel. Top. Quantum Electron. 22(6), 370–378 (2016).
2. W. Bogaerts, S. K. Selvaraja, P. Dumon, J. Brouckaert, K. De Vos, D. Van Thourhout, and R. Baets, “Silicon-on-insulator spectral filters fabricated with CMOS technology,” IEEE J. Sel. Top. Quantum Electron. 16(1), 33–44 (2010).
3. M. Gehl, D. Trotter, A. Starbuck, A. Pomerene, A. L. Lentine, and C. DeRose, “Active phase correction of high resolution silicon photonic arrayed waveguide gratings,” Opt. Express 25(6), 6320–6334 (2017). [PubMed]
4. F. Gan, T. Barwicz, M. A. Popovic, M. S. Dahlem, C. W. Holzwarth, P. T. Rakich, H. I. Smith, E. P. Ippen, and F. X. Kartner, “Maximizing the Thermo-Optic Tuning Range of Silicon Photonic Structures,” in Photonics in Switching (IEEE, 2007), pp. 67–68.
5. P. De Heyn, J. De Coster, P. Verheyen, G. Lepage, M. Pantouvaki, P. Absil, W. Bogaerts, D. Van Thourhout, and J. Van Campenhout, “Polarization-Insensitive 5x20Gb/s WDM Ge Receiver using Compact Si Ring Filters with Collective Thermal Tuning,” in Optical Fiber Communication Conference (OSA, 2014), paper Th4C.5.
6. P. Dong, Y.-K. Chen, and L. L. Buhl, “Reconfigurable Four-Channel Polarization Diversity Silicon Photonic WDM Receiver,” in Optical Fiber Communication Conference (OSA, 2015), paper W3A.2.
7. D. Y. Lee, X. Zheng, J. Yao, Y. Luo, J.-H. Lee, S. Lin, H. Thacker, J. Bovington, I. Shubin, S. S. Djordjevic, J. E. Cunningham, K. Raj, and A. V. Krishnamoorthy, “Error-free operation of a polarization-insensitive 4λ x 25 Gbps silicon photonic WDM receiver with closed-loop thermal stabilization of Si microrings,” Opt. Express 24(12), 13204–13209 (2016). [PubMed]
8. J. A. Cox, A. L. Lentine, D. C. Trotter, and A. L. Starbuck, “Control of integrated micro-resonator wavelength via balanced homodyne locking,” Opt. Express 22(9), 11279–11289 (2014). [PubMed]
9. K. Padmaraju, D. F. Logan, T. Shiraishi, J. J. Ackert, A. P. Knights, and K. Bergman, “Wavelength Locking and Thermally Stabilizing Microring Resonators Using Dithering Signals,” J. Lightwave Technol. 32(3), 505–512 (2014).
10. S. Grillanda, M. Carminati, F. Morichetti, P. Ciccarella, A. Annoni, G. Ferrari, M. Strain, M. Sorel, M. Sampietro, and A. Melloni, “Non-invasive monitoring and control in silicon photonics using CMOS integrated electronics,” Optica 1(3), 129–136 (2014).
11. C. V. Poulton, P. Dong, and Y.-K. Chen, “Photoresistive Microring Heater with Resonance Control Loop,” in CLEO:2015 (OSA, 2015), paper SM2I.3.
12. J. C. C. Mak, W. D. Sacher, T. Xue, J. C. Mikkelsen, Z. Yong, and J. K. S. Poon, “Automatic Resonance Alignment of High-Order Microring Filters,” IEEE J. Quantum Electron. 51(11), 1–11 (2015).
13. H. Jayatilleka, K. Murray, M. Á. Guillén-Torres, M. Caverley, R. Hu, N. A. F. Jaeger, L. Chrostowski, and S. Shekhar, “Wavelength tuning and stabilization of microring-based filters using silicon in-resonator photoconductive heaters,” Opt. Express 23(19), 25084–25097 (2015). [PubMed]
14. Y. Li and A. W. Poon, “Active resonance wavelength stabilization for silicon microring resonators with an in-resonator defect-state-absorption-based photodetector,” Opt. Express 23(1), 360–372 (2015). [PubMed]
15. R. G. Beausoleil, J. Ahn, N. Binkert, A. Davis, D. Fattal, M. Fiorentino, N. P. Jouppi, M. McLaren, C. M. Santori, R. S. Schreiber, S. M. Spillane, D. Vantrease, and Q. Xu, “A Nanophotonic Interconnect for High-Performance Many-Core Computation,” in 16th IEEE Symposium on High Performance Interconnects (IEEE, 2008), pp. 182–189.
16. D. Miller, “Device requirements for optical interconnects to silicon chips,” Proc. IEEE 97, 1166–1185 (2009).
17. G. T. Reed, G. Mashanovich, F. Y. Gardes, and D. J. Thomson, “Silicon optical modulators,” Nat. Photonics 4(8), 518–526 (2010).
18. F. Morichetti, S. Grillanda, and A. Melloni, “Breakthroughs in photonics 2013: toward feedback-controlled integrated photonics,” IEEE Photonics J. 6(2), 1–6 (2014).
19. K. Padmaraju and K. Bergman, “Resolving the thermal challenges for silicon microring resonator devices,” Nanophotonics 3(4–5), 269–281 (2014).
20. P. Dong, R. Gatdula, K. Kim, J. H. Sinsky, A. Melikyan, Y.-K. Chen, G. de Valicourt, and J. Lee, “Simultaneous wavelength locking of microring modulator array with a single monitoring signal,” Opt. Express 25(14), 16040–16046 (2017). [PubMed]
21. J. Duchi, E. Hazan, and Y. Singer, “Adaptive subgradient methods for online learning and stochastic optimization,” J. Mach. Learn. Res. 12, 2121–2159 (2011).
