Open Access
Add to BibSonomyAbstract
Optical switches based on ring resonator cavities were fabricated by a silicon photonics foundry process and analyzed for optical crosstalk at various data rates and channel spacings. These devices were compared to commercial bandpass filters and at 20Gb/s, 0.5dB power penalty is observed due to spectral filtering for bit error ratio threshold of 1 × 10−9. Concurrent modulation at 20Gb/s with a channel spacing as narrow as 40GHz shows error-free transmission with 1dB power penalty as compared to wider channel spacing for the ring-based switch.
© 2016 Optical Society of America
1. Introduction
As internet traffic continues to increase and approaches the zettabyte volume [1] state-of-the-art data centers are under strain and their network infrastructure will need to rely more heavily on optical interconnects. Current data center architectures use electrical switches with internal optical to electrical (O/E) conversion mechanisms which are a major source of power consumption and capital expense [2]. This conversion has inherent latency and increased power consumption but enables buffering in the electrical domain, greatly simplifying switching algorithms [3]. To circumvent the inefficiencies associated with O/E conversion, all-optical switching platforms based on micro-electrical mechanical systems (MEMS) [4], Mach-Zehnder interferometer [5] and hybrid III-V/silicon optical gate [6] devices have been proposed. Each of these platforms has unique advantages such as scalability and cost but comes at the expense of slow switching times, high insertion loss or complex fabrication procedures, among others.
This work presents initial experimental results for an all-optical switch platform based on ring resonators. These structures have low insertion loss and are inherently wavelength-sensitive, eliminating the need of optical multiplexers. Previous work on ring modulators has also shown bi-directional resonance wavelength tuning with both thermal and carrier injection mechanisms [7]. These characteristics make ring-based switches an attractive candidate for development of optical switches and their system-level characteristics have been modeled extensively using a theoretical analysis previously [8]. Previous experimental work [9] has been done to understand the crosstalk of ring-based switches. This work extends on the works done in [8] and [9] by initially experimentally investigating the power penalty in bit error ratio (BER) due to spectral filtering by high cavity Q ring resonator structures. Subsequently, the crosstalk of these filters for various data rates and channel spacing is investigated and compared to a commercial bench-top bandpass filter (BPF). It is found that at 20Gb/s, the ring switch has 0.5dB increased power penalty as compared to a near-square shaped BPF due to spectral filtering. At the same data rate, concurrent modulation with a channel spacing as narrow as 40GHz is possible with an additional 1dB power penalty as compared to wider channel spacing for the studied ring-based switch.
2. Experimental
2.1. Setup
The two filters used in these experiments are based on the ring resonator cavity (Ring) and a commercially available bench-top thin-film dielectric BPF. The spectral responses of these two filters is shown in Figs. 1(a) and 1(b). The BPF has a 3dB bandwidth of 40GHz, outside of which the near-square transmission shape has a slope of 2dB/GHz. The ring cavity has a diameter of 14μm with a drop port cavity Q on the order of 8, 000 and through port cavity Q on the order of 10, 000. It is formed with a 450nm wide waveguide on a 300nm SOI wafer, fabricated at CEA-Leti. The add/drop waveguides of the ring cavity are 150nm away, resulting in ∼2% coupling between the bus waveguide and the ring cavity. This particular device is a part of a larger design of experiments to understand cavity Q dependence on add/drop gaps in order to identify critical coupling for the Leti SOI process technology and was chosen for its reasonably high Q factor. All modulated signals are injected into the ring resonator via the input port and collected via the drop port, as shown in Fig. 1(c) by using output grating couplers and single-mode fibers.
Fig. 1 Output optical spectra showing normalized through port spectrum of the a) commercial bandpass thin-film filter and b) both drop and through port transmission of the ring filter and c) a microscope image of the ring filter test structure outlining the location of the ports and d) experimental setup used for crosstalk analysis of ring resonator switch.
The experimental setup is shown in Fig. 1(d) where two lasers (Santec TCL-510) and modulators (EOSpace) form channels that are independently modulated and combined by a fiber combiner. One output of the combiner is used to accurately monitor each laser wavelength via an optical spectrum analyzer (OSA). The other is injected into either the BPF or Ring filter where the variable optical attenuator (VOA) and photodetector (PD) are used to generate an eye diagram whose BER versus input power is measured by the BER scope. Minimal amount of single-mode fiber is used between the photonic chip and PD to avoid significant effects on BER degradation from group velocity dispersion and other similar mechanisms.
BER is a unitless number defined as the ratio between incorrect and overall bits transmitted [10]. Due to the statistical nature of BER, an integration time is associated with the reported number. More accurately, a confidence limit of a reported BER value is defined by the integration time, T (in seconds) as [10]:
where f is the data rate, BER is the desired error ratio minimum value to be resolved, and CL is the confidence limit (between 0–1). For this work, an integration resulting in a 95% confidence interval is used for a BER minimum of 1 × 10−12 for all the reported data rates.2.2. Spectral Filtering
The presented experiments quantify the optical power penalty at 10, 15 and 20Gb/s for the Ring filter as compared to the BPF due to spectral filtering. The PD (Discovery Semiconductor) used incorporates a trans-impedance amplifier and a linear amplifier, enabling a low noise floor for error-free operation. Specifically, it has a 3dB bandwidth of 43GHz and a specification of BER < 1 × 10−12 at 25Gb/s with an optical input power noise floor of −10dBm and. However, at lower data rates, the noise floor will be lower. In order to quantify this dependency, several baseline measurements were performed using a commercial Lithium Niobate modulator (MOD) with a 20Gb/s modulation bandwidth. Using the setup shown in Fig. 1(d), only one wavelength/MOD channel is used to inject directly into the PD a modulated signal at the three data rates as a means to quantify the inherent power penalty for each data rate from the PD itself. Subsequently, this experiment is repeated but the modulated signal is filtered by either the BPF or the Ring. This is done to then further understand the increased power penalty due to spectral filtering of the modulated signal by the transmission spectrum of each filter. Figure 2(a) shows the eye diagram for all three conditions of PD alone (no filtering), BPF, and Ring filter at all three data rates for maximum input power into the PD.
Fig. 2 a) eye diagrams for each condition of PD alone (no filtering), BPF and Ring filter and b) modulated spectra as input to the PD for data rates of 10, 15, and 20Gb/s, c) BER of the modulated signal for each condition of PD alone (no filtering), BPF and Ring filter for data rates of 10, 15, and 20Gb/s and the d) power penalty for each data rate and condition for BER of 1 × 10−9..
It is expected that the optical sidebands that are created as a result of modulation will be attenuated due to the BPF or Ring filter due to the aforementioned spectral filtering mechanism. In order to compare the spectral signal received by the PD for each of the three test conditions and data rates, the modulated signal that is input into the PD is captured by the OSA, normalized, and shown in Fig. 2(b). From this figure, we can see clear effects of spectral filtering between the condition of PD alone versus either of the filters. From the spectra of the BPF compared to the Ring, we see comparable behavior for the first order harmonics. However, the BPF shows stronger attenuation for higher order modulation sidebands created outside of its bandwidth.
The impact of spectral filtering due to the BPF versus the Ring filter is especially pronounced for higher data rates due to increased spectral content outside of the filter bandwidth. This filter bandwidth can be numerically simulated and recreated in a Bode plot to understand the maximum data rate for a given resonator cavity, as done in [11]. To quantify this effect, the BER of the modulated signal for all three conditions and data rates is measured and shown in Fig. 2. The increase in noise floor for the PD is shown in Fig. 2(c), as well as the increase associated with each filter for increasing data rates. The power penalty in going from 10 to 15Gb/s can be attributed primarily to the PD bandwidth itself for all three cases. However, in going from 15 to 20Gb/s, the Ring filter shows a higher power penalty as compared to the BPF. Figure 2(d) summarizes the power penalty for a BER of 1 × 10−9 at higher data rates relative to the baseline data rate of 10Gb/s. From this figure, we can see 1dB power penalty for the Ring at 20Gb/s, compared to 0.5dB for the BPF at the same data rate with respect to the case of an unfiltered signal directly into the PD.
2.3. Crosstalk Measurement
In order to understand the crosstalk behavior of the two filters, concurrent modulation is done with both channels using the setup shown in Fig. 1d. The BPF bandwidth and channel 1 (Laser 1 and MOD 1) are tuned to the center frequency (228.92THz) with channel 2 set to specific spectral separation with both channels modulated at the three aforementioned data rates. The spectra captured by the OSA after the fiber combiner, as shown in Fig. 1(d), is shown in Fig. 3 for various values of spectral detuning and modulation data rates. The detuning value is defined as the frequency of channel 1 less the frequency of channel 2.
Fig. 3 Spectra of concurrently modulated optical signals for various detuning values for each data rate of 10, 15, and 20Gb/s.
Figure 4 shows the measured BER for each filter at the denoted data rates versus input power to the PD. Each plot shows the BER for the respective data rate for positive and negative de-tuning values. from these plots, an increase in error for a fixed input power can be seen. On the same token, in order to achieve the same BER value, the input power to the PD must be increased as the channel spacing is reduced. This is thus defined as the optical power penalty. Figure 5 shows the power penalty for the bandpass filter at various data rates and channel de-tuning values for a BER of 1 × 10−9. No significant power penalty can be observed for any data rate for channel spacing greater than 40GHz. The sharp increase observed near the 20GHz spacing is due to the overlap with the transmission bandwidth of the filter itself. As a result, an open eye diagram was not observed for the −18GHz detuning value at 20Gb/s. It should be noted that the presented power penalty for this data point was calculated by extrapolation of the measured data.
Fig. 4 BER vs. input power for channel 1 as filtered by bandpass filter for each data rate of a) 20Gb/s, b) 15Gb/s, and c) 10Gb/s and Ring filter for d) 20Gb/s, e) 15Gb/s, and f) 10Gb/s for various detuning conditions with respect to channel 2.
Fig. 5 Power penalty for 1 × 10−9 BER for the bandpass filter at various data rates and channel detuning values is shown. Under this plot, the associated eye diagrams for maximum input power to the PD are shown for reference.
In comparison, Fig. 6 shows the same power penalty and associated eye diagrams observed by the Ring filter. For 10Gb/s, channel spacing greater than 40GHz shows no observable power penalty which enables these devices for integration with previously proposed ring modulators operating at 10Gb/s at 80GHz channel spacing [12]. At higher data rates, the power penalty increases which is expected due to the generation of modulation sidebands. Another mechanism contributing to the observed power penalty is due to the Lorentzian lineshape of the ring filter and optical leakage from channel 2. This effect is more carefully quantified in [12] and can be expected to contribute less than 0.5dB of the observed power penalty.
Fig. 6 Power penalty for 1 × 10−9 BER for the ring filter at various data rates and channel detuning values is shown. Under this plot, the associated eye diagrams for maximum input power to the PD are shown for reference.
3. Conclusion
The effects of spectral filtering of a high-Q ring resonator switch and the associated crosstalk for various data rates and channel spacings is presented. It is found that for a 20Gb/s data rate signal, 0.5dB excess power penalty is observed due to the spectral filtering of the ring filter as compared to a near-square shaped bandpass filter. The crosstalk versus data rate for the ring switch is also studied as a function of different channel spacing. For a 10Gb/s data rate, no observable power penalty is measured for spacing greater than 40GHz. However, at 20Gb/s a power penalty of 1dB is observed for 40GHz spacing as compared to wider channel spacing values. As a reference, the bandpass filter showed a power penalty of less than 0.5dB for the same data rate and channel spacing. These results show great promise of integration of ring-based Top-Of-Rack switches into modern data center architecture.
References and links
1. V. N. I. Cisco, “The Zettabyte Era: Trends and Analysis,” http://www.cisco.com/c/en/us/solutions/collateral/service-provider/visual-networking-index-vni/VNI_Hyperconnectivity_WP.html.
2. N. Farrington, A. Forencich, G. Porter, P. C. Sun, J. E. Ford, Y. Fainman, G. C. Papen, and A. Vahdat, “A multiport microsecond optical circuit switch for data center networking,” IEEE Photonics Technol. Lett. 25(16), 1589–1592 (2013). [CrossRef]
3. Y. Kunishige, K. i. Baba, and S. Shimojo, “Optical network configuration methods considering end-to-end latency in data centers,” IEEE Pacific Rim Conference on Communications, Computers and Signal Processing (PACRIM) (IEEE2015), pp. 210–215.
4. T. J. Seok, N. Quack, S. Han, and M. C. Wu, “50×50 digital silicon photonic switches with mems-actuated adiabatic couplers,” Optical Fiber Communication Conference (OFC) (Optical Society of America, 2015), paper M2B4.
5. N. Dupuis, “Technologies for fast, scalable silicon photonic switches,” International Conference on Photonics in Switching (PS) (IEEE2015), pp. 100–102.
6. G. de Valicourt, N. D. Moroz, P. Jennev, F. Vacondio, G. H. Duan, C. Jany, A. Leliepvre, A. Accard, and J. C. Antona, “A next-generation optical packet-switching node based on hybrid III-V/silicon optical gates,” IEEE Photonics Technol. Lett. 26(7), 678–681 (2014). [CrossRef]
7. T. C. Huang, C. Li, R. Wu, C. H. Chen, M. Fiorentino, K. T. Cheng, and R. Beausoleil, “DWDM nanophotonic interconnects: toward terabit/s chip-scale serial link,” IEEE 58th International Midwest Symposium on Circuits and Systems (MWSCAS) (IEEE2015) pp. 1–4.
8. M. Bahadori, S. Rumley, D. Nikolova, and K. Bergman, “Comprehensive design space exploration of silicon photonic interconnects,” J. Lightwave Technol. PP(99), 1–14 (2016).
9. A. Descos, M. A. Seyedi, C.H. Chen, F. Vincent, D. Penkler, M. Fiorentino, B. Szelag, and R. Beausoleil, “Crosstalk analysis of ring-resonator based optical switches,” IEEE Optical Interconnects Conference (OIC) (IEEE2016), paper TuP5.
10. S. Haykin, An Introduction to Analog and Digital Communication (Wiley & Sons, 1989)
11. B. Small, B. Lee, K. Bergman, Q. Xu, J. Shakya, and M. Lipson, “High data rate signal integrity in micron-scale silicon ring resonators,” Conference on Lasers and Electro-Optics (CLEO) (Optical Society of America, 2006), paper CTuCC4.
12. M. A. Seyedi, C.-H. Chen, M. Fiorentino, and R. Beausoleil, “Error-free DWDM transmission and crosstalk analysis for a silicon photonics transmitter,” Opt. Express 23(26), 32968–32976 (2015). [CrossRef]
