Open Access
Add to BibSonomyAbstract
We experimentally demonstrate a 16 × 16 non-blocking optical switch fabric with a footprint of 10.7 × 4.4 mm2. The switch fabric is composed of 56 2 × 2 silicon Mach-Zehnder interferometers (MZIs), with each integrated with a pair of TiN resistive micro-heaters and a p-i-n diode. The average on-chip insertion loss at 1560 nm wavelength is ~6.7 dB and ~14 dB for the “all-cross” and “all-bar” states, respectively, with a loss variation of ± 1 dB over all routing paths. The measured rise/fall time of the switch upon electrical tuning is 3.2/2.5 ns. The switching functionality is verified by transmission of 20 Gb/s on-off keying (OOK) and 50 Gb/s quadrature phase-shift keying (QPSK) optical signals.
© 2016 Optical Society of America
1. Introduction
Optical switches for non-blocking optical signal routing from multiple sources to multiple destinations are critical components for both long-haul and short-reach optical communications [1–3]. Given the increasing data traffic in telecom and datacom, high-port-count switch fabrics with a fast reconfigurable time in the order of nanoseconds are highly desired [4–6]. Optical switches on the Silicon-on-Insulator (SOI) platform have the merits of compact size, low power consumption, and compatibility with complementary metal-oxide-semiconductor (CMOS) fabrication process, providing viable solutions for future high-speed optical networks.
In the past decade, several N × N optical switch fabrics have been demonstrated by using various building blocks, such as Mach-Zehnder interferometers (MZIs) [7–17], microring resonators (MRRs) [18–20], multimode interferometers (MMIs) [21–23], double-ring-assisted MZIs [24, 25], and MEMS-actuated adiabatic couplers [26, 27]. Among them, the demonstrated highest-port-count silicon optical switches are 64 × 64 [27], which are based on MEMS actuation. However, the reconfigurable time is in the order of microsecond, intrinsically limited by the mechanical motion of the MEMS actuated waveguide. Such a response is not fast enough to support a typical packet duration of less than 20 ns in optical packet switching networks and computer communications systems [5, 6]. Besides, switch fabrics with reconfigurably non-blocking topologies require fast switching to lower rearranging latency. Silicon electro-optic (EO) switches based on free carrier dispersion (FCD) effect own the merit of short switching time of only nanoseconds. The reported highest-port-count silicon EO optical switches are 8 × 8 switches [15]. It is quite challenging to scale up the EO optical switches to a higher port-count, as the free carrier absorption (FCA)-induced loss will deteriorate the switch performances in terms of insertion loss and crosstalk.
In this paper, we report the experimental realization of a silicon 16 × 16 EO reconfigurably non-blocking switch based on 2 × 2 MZIs. Such switches have potential applications in optical packet switching networks and computer communications systems. To the best of our knowledge, this is the first reported non-blocking 16 × 16 silicon switch fabric integrated with both EO and thermo-optic (TO) tuners. Compared with our previous 4 × 4 silicon EO switch [13], the waveguide crossings and 2 × 2 MMI couplers in the MZIs are optimized. The insertion losses of the waveguide crossings and 2 × 2 MMI couplers are reduced to 0.05 dB and 0.22 dB at 1560 nm wavelength, respectively. The measured crosstalk of the 2 × 2 MZI switch elements is ~-30 dB and ~-18 dB around 30 nm wavelength range for the “cross” and the “bar” states, respectively. The footprint of the fabricated 16 × 16 switch chip is 10.7 × 4.4 mm2. The average on-chip insertion loss of the 16 × 16 switch at 1560 nm wavelength is ~6.7 dB and ~14 dB for the “all-cross” and the “all-bar” states, respectively. The measured rise/fall time for EO switching is 3.2/2.5 ns. Data transmission experiments using a 20 Gb/s on-off keying (OOK) optical signal and a 50 Gb/s quadrature phase-shift keying (QPSK) optical signal verify the signal integrity.
The rest of the paper is organized as follows. Section 2 presents the design of the switch fabric, including the switch architecture, passive components, 2 × 2 MZI switch elements. Section 3 presents an overview of the fabricated and packaged 16 × 16 switch chip. Next in Section 4, we characterize the optical performances of the constitutive components and the entire switch chip, including its spectral response, power consumption, dynamic switching and transmission of OOK and QPSK signals. Finally, we make our concluding statements in Section 5.
2. Switch design
The 16 × 16 optical switch fabric is based on a Benes architecture. Figure 1(a) shows the topological structure of the switch, incorporating 56 elements of 2 × 2 MZI switches. It is a reconfigurably non-blocking switch. As each switch element has two states, the 16 × 16 switch hence has 256 states in total, among which 16! states are necessary for the complete mapping of all input to all output ports. In comparison with the structures such as Switch-and-Selected, PILOSS, Fat-tree, Clos and so on, the Benes architecture is the most compact as it requires the least switch elements to realize the full connections, and thus has the merits of lower insertion loss and less components to actuate. However, there is a drawback for reconfigurably non-blocking switches like the Benes topology, as the existing connections may be interrupted and reestablished when the switching state is changed.
Fig. 1 (a) Architecture of the 16 × 16 optical switch based on a Benes architecture. (b) MMI-based waveguide crossing. (c) 2 × 2 MZI switch element. The upper arm is integrated with both a p-i-n diode and a TiN micro-heater. The bottom arm is integrated with a TiN micro-heater.
For a 16 × 16 switch, each optical path goes through seven stages of switching elements. The input and output ports are defined as Ii and Oi (i = 1, 2,…, 16), respectively. As can be seen from Fig. 1(a), waveguide crossings are indispensable for the two-dimensional connection of all 56 switching elements. The maximum number of crossings in an optical path is 22. We carefully design the waveguide crossings to reduce the insertion loss and crosstalk. 90°-crossed 1 × 1 MMIs based on the self-imaging principle are utilized for light to cross over the waveguide junctions [28]. Linear tapers are employed to minimize the transition loss between the MMI section and the input/output waveguides. In order to reduce the footprint and insertion loss of the crossings, the taper width is designed to be the same with the MMI width [29, 30]. The MMI-based waveguide crossing is optimized for transverse electric (TE) polarization using particle swarm optimization (PSO) method together with 3-D finite-difference time-domain (FDTD) simulation [31]. In the optimization, the width and length of the 1 × 1 MMI and the length of the tapers are set as variables, and the figure of merit (FOM) is defined as normalized power transmission minus crosstalk at the wavelength of 1560 nm. Figure 1(b) shows the layout structure of the MMI-based waveguide crossing. The silicon waveguide width is 0.5 μm and the height is 0.22 μm with a slab layer thickness of 60 nm. The MMI width and length are 1.73 μm and 8.14 μm, respectively. The optimized taper length is 2.39 μm.
Figure 1(c) shows the schematic of a 2 × 2 MZI switch element, which consists of two 2 × 2 MMI couplers connected by two equal-length waveguide arms for broadband operation. The four input/output ports are denoted as “1”, “2”, “3”, and “4”. The 2 × 2 MMI couplers are used as 3-dB couplers for its high fabrication tolerance and broad optical bandwidth. We optimize the 2 × 2 MMI couplers to get low insertion loss and low power imbalance. The width and length of the 2 × 2 MMI couplers are 5 μm and 29.5 μm, respectively. The center-to-center distance between the two access waveguides is optimized to be 1.72 μm. The input and output waveguides are linearly tapered to 1.2 μm in a length of 10 μm to lower the insertion loss.
The routing path 1-4 (2-3), named as the “cross” state, is established when the phases of the two arms are equal. When the phase difference of the two arms is π, light from 1 (2) is switched to 3 (4), leading to the “bar” state. As the initial state of the MZI switch is not exactly at the “cross” state due to fabrication imperfections, TO tuners are used to correct the fabrication-induced phase errors without inducing excess loss. The TO tuners are made of TiN micro-heaters with a length of 300 μm positioned directly above MZI arms. Air trenches surround waveguides to improve the thermal tuning efficiency. Besides the TO tuners, an EO tuner based on a lateral p-i-n diode is also integrated in one of the MZI arms for high-speed switching operation. We choose the p-i-n diode instead of the reverse-biased p-n junctions for EO tuners because of its higher modulation efficiency, lower “off state” absorption loss and less fabrication complexity. The cross-sectional views of the top and bottom arms of the MZI are shown in the insets of Fig. 1(c). Free-carriers are injected into the waveguide when the p-i-n diode is turned on with current flowing through the waveguide as illustrated by the white dashed arrow. The heavily doped regions have a separation distance of 0.8 μm from the waveguide edges. The length of the p-i-n diodes is 380 μm.
3. Chip fabrication and package
The 16 × 16 optical switch chip was fabricated on a SOI wafer with a top silicon layer thickness of 220 nm and a buried oxide layer thickness of 2 μm. The ridge waveguides were patterned by 248-nm deep ultra-violet (DUV) photolithography, and then plasma dry etched with a depth of ~160 nm. Ion implantations of boron and phosphorus were used to form the p+ doped regions and the n+ doped regions of p-i-n diodes, respectively. The resulting doping concentration is ~1020 cm−3. Rapid thermal annealing (RTA) at 1030°C for 5 seconds was used after ion implantation to activate the dopants. After that, a 1.5 μm thick oxide was deposited on the waveguides by using the plasma-enhanced chemical vapor deposition (PECVD). Subsequently, heaters were patterned by deposition and dry etching of a 120-nm-thickness TiN layer. Another oxide layer of 0.73 μm thickness was deposited above the TiN layer by PECVD. Finally, contact holes were etched and aluminum connection was formed by sputtering and dry etching. The fabrication was done using the CMOS compatible process in IME Singapore.
Figure 2(a) shows the optical microscope image of the fabricated 16 × 16 switch chip. Light is coupled in and out of the switch chip through grating couplers with apodized structures in order to improve the coupling efficiency. The grating couplers are positioned in an array with a period of 127 μm to match with a commercial fiber array. The TO and EO electrodes in the seven stages of switch elements are connected out to electrical pads positioned along the chip edges. The electrical pad size is 160 × 160 μm2 arranged in two rows with a separation of 100 μm. The footprint of the device is 10.7 × 4.4 mm2, which can be further reduced by using smaller electrical pads. The inset of Fig. 2(a) shows the zoom-in view of the MZI switch elements. In order to reduce the light propagation loss between switching elements, the straight sections of connection waveguides are widened to 2 μm using 200-μm-long linear tapers. In order to monitor the states of MZI switch elements, we add four waveguide taps terminated with local grating couplers before and after each MZI switch element. Thus, the transmission of each MZI can be independently measured, making it possible to optimize the switching performance of each unit. The tap is made of a curved waveguide evanescently coupled to a straight waveguide with a gap size of 0.2 μm. The splitting loss per tap is measured to be ~0.4 dB at 1560 nm wavelength, corresponding to a power splitting ratio of about 10%. Hence, the excess loss introduced by the taps for the 16 × 16 switch is about 5.6 dB. The excess loss can be eliminated by implementing contactless integrated photonic probe (CLIPP) to non-invasively monitor on-chip light power [32, 33]. The working mechanism of the CLIPP is based on the change of the waveguide conductance induced by surface state absorption at the Si-SiO2 interface. The CLIPP exploits contactless capacitive access to the waveguide, so it avoids perturbation to the optical mode inside the waveguide core.
Fig. 2 (a) Optical microscope image of the fabricated 16 × 16 optical switch. The inset shows the zoom-in view of one stage. (b) Photo of the 16 × 16 switch after electrical and optical package. The inset shows the zoom-in view of the packaged switch chip.
Figure 2(b) shows the photo of the switch chip after electrical and optical package. The electrical pads are wire-bonded to a printed circuit board (PCB). The device is connected to multi-voltage sources by high-speed micro-coax cables. A 32-channel 127-μm-pitch fiber array is attached to the switch chip by using ultra-violet (UV) light curable adhesive, whose refractive index is close to silicon dioxide. The coupling loss after optical packaging is around 5 dB/facet, similar to that before packaging.
4. Experiments
4.1 Characterization of basic components
Firstly, we characterized the performance of the basic passive components, namely, the MMI waveguide crossing and the 2 × 2 MMI coupler. Figures 3(a) and 3(b) show the microscope images of the test structures. In the measurement, light was adjusted to TE polarization before coupled in and out of the chip via grating couplers. The insertion loss of a waveguide crossing is derived from the linear fitting of 80, 160, 240 and 320 series-connected crossings. Figure 3(c) shows the wavelength-dependent insertion loss and crosstalk of a MMI crossing. The insertion loss is below ~0.05 dB in the wavelength range of 1530 to 1590 nm, much lower than our previous design [13]. The crosstalk is lower than −30 dB. The insertion loss of a 2 × 2 MMI coupler was also measured from the test structure made of a series of cascaded couplers with the results shown in Fig. 3(d). The average insertion loss is around 0.4 dB in the 60 nm wavelength range. The fluctuation of insertion loss with wavelength may be caused by several factors, such as the FP resonances induced by reflection from grating couplers and MMIs, and the shift of grating coupler central wavelength due to fabrication errors, which as a result generates uncertainties in the extracted insertion loss. The inset shows the measured normalized transmission decreasing with the number of MMI couplers at 1560 nm. The insertion loss deduced from the linear fitting is hence around 0.22 dB.
Fig. 3 (a) Microscope view of MMI waveguide crossing test structures. (b) Microscope image of a 2 × 2 MMI coupler test structure. (c) Measured insertion loss and crosstalk of the MMI waveguide crossing. (d) Measured insertion loss of the 2 × 2 MMI coupler. The inset shows the linear fitting of the normalized transmission at 1560 nm wavelength.
Next we measured the switching performance of the basic 2 × 2 MZI switch element with the microscope image shown in Fig. 4(a). Figure 4(b) presents the measured normalized transmission spectra of the MZI switch at the “cross” state after phase correction by TO tuners. Crosstalk less than −30 dB is obtained over a wavelength range of almost 30 nm, indicating the 2 × 2 MMI couplers have a relatively balanced splitting ratio. The 2 × 2 MZI is switched to the “bar” state after tuning on the EO tuner. Figure 4(c) presents the measured “bar” state transmission spectra. Due to the FCA effect, the insertion loss is increased by about ~1 dB compared with the “cross” state. The crosstalk is deteriorated to ~-18 dB as a result of the unbalanced optical power in the two arms.
Fig. 4 (a) Microscope image of the 2 × 2 MZI switch element. (b) and (c) Output transmission spectra at the (b) “cross” and (c) “bar” states.
4.2 Transmission spectra of the 16 × 16 switch
Due to fabrication imperfection-induced random phase errors in the waveguides, the 2 × 2 MZIs are not exactly at the “cross” state for the as-fabricated devices. We therefore first corrected these phase errors by applying proper voltages to TO tuners of all the 56 MZI switches. After correction, the switch is at the “all-cross” state. Figure 5 shows the measured transmission spectra of the 16 × 16 optical switch at the “all-cross” state. Each plot groups 16 transmission spectra from all 16 input ports to one output port. The spectra are normalized to a test waveguide with fourteen taps. Therefore, the average on-chip insertion loss of the 16 × 16 optical switch without taps at 1560 nm wavelength is ~6.7 dB, with a loss variation of about ± 1 dB. The insertion loss is composed of the following parts: 0.22 × 14 = 3.08 dB from the 2 × 2 MMI couplers, 0.05 × 11 = 0.55 dB from the waveguide crossings (11 waveguide crossings on average for one routing path), and 3.07 dB from connection waveguides. As the total waveguide length of a light path is about 2.6 cm, the estimated average waveguide propagation loss is hence ~1.18 dB/cm. It should be noted that 2-μm-wide waveguides are adopted for long straight sections to reduce light propagation loss. The crosstalk at 1560 nm wavelength is all below ~-30 dB. Here, the crosstalk of the input port m (m≠i) to the routing path Ii-Oj (i,j = 1,2,…,16) is defined as the ratio of the leaked output power, Pout(m→j), to the output power, Pout(i→j). The optical bandwidth for <-20 dB crosstalk is over 30 nm.
We then applied voltages to the EO tuners of all MZI switch elements while TO voltages were kept constant. The 16 × 16 switch is flipped to the “all-bar” state. Figure 6 shows the normalized transmission spectra at the “all-bar” state. As the FCA effect leads to extra loss, the average insertion loss is increased to ~14 dB compared with the “all-cross” state. The loss per MZI switch is increased by ~1 dB, consistent with the single MZI measurement. The crosstalk of the 16 × 16 switch is < −10 dB over the measured wavelength range. The crosstalk deterioration is induced by both the limited extinction ratio of the MZI switch elements and the multipath interference in the fabric. Due to some unknown reasons, the extinction ratio of S3-6 (the sixth switch element in stage S3) is lower than normal after turning on the p-i-n diode. Therefore, the leaked light will eventually interfere more severely with the light in the main routing path, leading to the more apparent ripples in the O10 and O14 spectra. It should be noted that the “all-bar” state is the worst state among all 264 states of the 16 × 16 switch. As a certain input-to-output mapping can be achieved by multiple different configurations, thus, in practice, we can choose the best routing paths to get an optimal switching performance [34]. The crosstalk of the 16 × 16 switch can be further suppressed by using a push-pull driving manner [16], or resonator-assisted MZI structures [25].
4.3 OOK signal switching
We then examined the dynamic routing performance of the 16 × 16 optical switch. Figure 7 shows the system setup for the experiment. A continuous wave (CW) light at 1560 nm wavelength is generated by a tunable laser source (TLS). The light is modulated by an amplitude modulator to generate a continuous optical OOK data stream. The modulator is driven by a 20 Gb/s 231-1 pseudo-random bit sequence (PRBS) radio frequency (RF) signal from a pulse-pattern generator (PPG). The optical signal is first amplified by an erbium-doped fiber amplifier (EDFA) and set to transverse electric (TE) polarization before coupling into the input port I1 of the switch. A 10 MHz electrical square-wave gate signal is applied to the p-i-n diode of S7-5 (the fifth MZI switch element in stage S7), while all the other MZIs are at the “cross” state. The peak-to-peak voltage of the gate signal is 0.3 V biased at 1.01 V. Thus, the optical data stream is switched between output ports O9 and O10 by the control gate. The out-transmitted signal from O9 or O10 is amplified by another EDFA followed by a variable optical attenuator (VOA) to adjust the optical power. Finally, the optical signal is converted back to an electrical one in a photodetector with a bandwidth of 100 GHz, and received by a sampling oscilloscope (Keysight 86100D).
Fig. 7 Experimental setup for demonstration of optical signal switching. Solid and dashed lines represent the optical and electrical signals, respectively.
Figures 8 show the measured OOK signal dynamic switching in response to the gate signal. The optical packets are routed either to O9 or O10 as expected. From the measured waveforms, the 10%-90% rise and fall times are 3.2 and 2.5 ns, respectively. The switch speed is almost three orders faster than that based on the TO effect [7, 14] or the MEMS actuation [26]. Clear and open eye diagrams are observed from both routing paths. The measured peak-to-peak jitter is ~12 ps.
Fig. 8 Measured high speed OOK signal switched by the 16 × 16 switch. The top row shows the signal through routing path I1-O9 and the bottom row I1-O10.
4.4 QPSK signal switching
We also performed optical data transmission experiments using high-speed QPSK optical signals. The high-order modulation formats such as QPSK are widely used in coherent fiber-optic communications due to their high spectral efficiency. The QPSK signal is generated by modulating a CW light at 1560 nm wavelength using a LiNbO3 IQ modulator. The modulator is driven by two 25 Gb/s 231-1 PRBS RF signals from two PPGs. The bit rate of the optical signal is 50 Gb/s. The other experimental setup is similar to that shown in Fig. 7. The received optical signal is analyzed by an optical modulation analyzer (Keysight, N4392A). The error-vector-magnitude (EVM) is obtained from the measured constellation diagrams.
Figure 9(a) depicts the constellation diagram of the system back-to-back (BtB) transmission, with an EVM of 8.57%. Figures 9(b) and 9(c) show the constellation diagrams for all the 16 optical paths at the “all-cross” and “all-bar” states, respectively. There is no observable signal degradation from the constellation diagrams. The measured EVMs are all better than 11.3%, indicating the signal is only degraded slightly after passing through the chip. The EVMs at the “all-bar” state are a little higher than those at the “all-cross” state, which is caused by the increased loss and crosstalk at the “all-bar” state. It hence demonstrates that our 16 × 16 switch is capable of switching a 50 Gb/s QPSK signal with high fidelity.
Fig. 9 Measured constellation diagrams of a 50Gb/s QPSK signal. (a) BtB transmission; (b) the “all-cross” state; (c) the “all-bar state.”
4.5 TO and EO Power consumptions
The TO phase correction power consumptions of all the 56 MZI switch elements are listed in Table 1. The power varies from 0 to 26 mW due to the random phase errors. The total power consumption is ~881 mW, which is dependent on fabrication accuracy and can be reduced by using advanced high-resolution processing tools. The power efficiency of TO tuners is about 35 mW/π. It can be further improved by etching off the silicon substrate or using silicon resistive microheaters instead of TiN heaters [35].
Table 1. Thermo-optic power consumption of the 16 × 16 switch (unit: mW)
The EO switching currents and powers of all the 56 MZI switch elements are listed in Table 2. The EO power per switch element varies from 3.28 to 5.88 mW, which is maybe due to doping variation and contact resistance variation induced by fabrication imperfections. The total EO power to set the 16 × 16 switch to the “all-bar” state is ~289 mW. Thus, the up limit of the operation power for our 16 × 16 optical switch chip is 1.17 W.
Table 2. Electro-optic tuning current and power consumption of the switch
As our 16 × 16 optical switch has an operation bandwidth of 30 nm, it can simultaneously switch multiple dense wavelength-division-multiplexing (DWDM) channels. Based on the ITU-T G.694.1 standard, the DWDM channel spacing is 100 GHz and thus each port can occupy 37 wavelength channels from 1545 nm to 1575 nm. Assuming 50 Gb/s data rate for each channel, the switching capacity for the 16 × 16 optical switch is thus 16 × 37 × 50Gb/s = 29.6 Tb/s. With the TO and EO power consumptions shown in Table 1 and Table 2, the maximum bit switching power is 39.5 fJ/bit. A higher data rate could result in even lower bit power consumption.
5. Conclusions
We have designed, fabricated, and experimentally demonstrated a high-speed 16 × 16 silicon optical switch fabric based on 56 2 × 2 symmetric MZIs. All the 2 × 2 MZIs are integrated with TO and EO phase shifters for phase error correction and fast switching operation, respectively. The on-chip insertion loss of the switch at 1560 nm wavelength is ~6.7 dB and the crosstalk is below −20 dB over a broad optical bandwidth of ~30 nm at the “all-cross” state after TO correction. The on-chip insertion loss increases to ~14 dB and the crosstalk increased to −10 dB at the “all-bar” state due to the FCA effect upon EO switching. It should be noted that although the TO tuners are designed for phase correction, they basically can also be used for switching operation where lower loss and crosstalk for the “all-bar” state are expectable. The total TO power consumption to compensate fabrication errors is ~881 mW, which could be reduced by using improved fabrication processes. The upper limit of EO power consumption to switch all the MZIs to the “bar” state is ~289 mW. The measured rise/fall time for EO switching is 3.2/2.5 ns. We carried out optical signal transmission experiments using both high-speed OOK and QPSK modulated signals, revealing good signal qualities after switching.
The demonstrated high-speed broadband 16 × 16 silicon optical switch fabric shows promising potential for future high-speed optical networks, such as optical packet switching and computer communications networks. For practical use, one should configure the 16 routing paths so that the optimal overall performance could be obtained for a certain switching state. In our future work, we will try to mitigate the deteriorating effect of FCA on crosstalk and insertion loss so as to implement switch fabrics with a higher port-count.
Acknowledgments
This work was supported in part by the 863 program (2013AA014402), the National Natural Science Foundation of China (NSFC) (61422508), the Shanghai Rising-Star Program (14QA1402600). We also acknowledge IME Singapore for device fabrication.
References and links
1. S. J. B. Yoo, “Optical packet and burst switching technologies for the future photonic internet,” J. Lightwave Technol. 24(12), 4468–4492 (2006). [CrossRef]
2. A. Biberman and K. Bergman, “Optical interconnection networks for high-performance computing systems,” Rep. Prog. Phys. 75(4), 046402 (2012). [CrossRef] [PubMed]
3. R. G. Beausoleil, M. McLaren, and N. P. Jouppi, “Photonic architectures for high-performance data centers,” IEEE J. Sel. Top. Quantum Electron. 19(2), 3700109 (2013). [CrossRef]
4. M. J. O’Mahony, C. Politi, D. Klonidis, R. Nejabati, and D. Simeonidou, “Future optical networks,” J. Lightwave Technol. 24(12), 4684–4696 (2006). [CrossRef]
5. Y. Li, Y. Zhang, L. Zhang, and A. W. Poon, “Silicon and hybrid silicon photonic devices for intra-datacenter applications: state of the art and perspectives [Invited],” Photon. Res. 3(5), B10–B27 (2015). [CrossRef]
6. B. G. Lee, N. Dupuis, P. Pepeljugoski, L. Schares, R. Budd, J. R. Bickford, and C. L. Schow, “Silicon photonic switch fabrics in computer communications systems,” J. Lightwave Technol. 33(4), 768–777 (2015). [CrossRef]
7. L. Chen and Y. K. Chen, “Compact, low-loss and low-power 8×8 broadband silicon optical switch,” Opt. Express 20(17), 18977–18985 (2012). [CrossRef] [PubMed]
8. B. G. Lee, A. V. Rylyakov, W. M. J. Green, S. Assefa, C. W. Baks, R. Rimolo-Donadio, D. M. Kuchta, M. H. Khater, T. Barwicz, C. Reinholm, E. Kiewra, S. M. Shank, C. L. Schow, and Y. A. Vlasov, “Monolithic silicon integration of scaled photonic switch fabrics, CMOS logic, and device driver circuits,” J. Lightwave Technol. 32(4), 743–751 (2014). [CrossRef]
9. K. Suzuki, K. Tanizawa, T. Matsukawa, G. Cong, S. H. Kim, S. Suda, M. Ohno, T. Chiba, H. Tadokoro, M. Yanagihara, Y. Igarashi, M. Masahara, S. Namiki, and H. Kawashima, “Ultra-compact 8 × 8 strictly-non-blocking Si-wire PILOSS switch,” Opt. Express 22(4), 3887–3894 (2014). [CrossRef] [PubMed]
10. J. Xing, Z. Li, P. Zhou, X. Xiao, J. Yu, and Y. Yu, “Nonblocking 4×4 silicon electro-optic switch matrix with push-pull drive,” Opt. Lett. 38(19), 3926–3929 (2013). [CrossRef] [PubMed]
11. M. Yang, W. M. J. Green, S. Assefa, J. Van Campenhout, B. G. Lee, C. V. Jahnes, F. E. Doany, C. L. Schow, J. A. Kash, and Y. A. Vlasov, “Non-blocking 4x4 electro-optic silicon switch for on-chip photonic networks,” Opt. Express 19(1), 47–54 (2011). [CrossRef] [PubMed]
12. J. Xing, Z. Li, Y. Yu, and J. Yu, “Low cross-talk 2 × 2 silicon electro-optic switch matrix with a double-gate configuration,” Opt. Lett. 38(22), 4774–4776 (2013). [CrossRef] [PubMed]
13. L. Lu, L. Zhou, Z. Li, X. Li, and J. Chen, “Broadband 4 x 4 nonblocking silicon electrooptic switches based on Mach-Zehnder interferometers,” IEEE Photonics J. 7, 7800108 (2015). [CrossRef]
14. K. Tanizawa, K. Suzuki, M. Toyama, M. Ohtsuka, N. Yokoyama, K. Matsumaro, M. Seki, K. Koshino, T. Sugaya, S. Suda, G. Cong, T. Kimura, K. Ikeda, S. Namiki, and H. Kawashima, “Ultra-compact 32 × 32 strictly-non-blocking Si-wire optical switch with fan-out LGA interposer,” Opt. Express 23(13), 17599–17606 (2015). [CrossRef] [PubMed]
15. T. Chu, L. Qiao, and W. Tang, “High-speed 8x8 electro-optic switch matrix based on silicon PIN structure waveguides,” 12th International Conference on Group IV Photonics (GFP), 123–124, 2015.
16. N. Dupuis, B. G. Lee, A. V. Rylyakov, D. M. Kuchta, C. W. Baks, J. S. Orcutt, D. M. Gill, W. M. J. Green, and C. L. Schow, “Modeling and characterization of a nonblocking 4x4 Mach-Zehnder silicon photonic switch fabric,” J. Lightwave Technol. 33, 4329–4337 (2015). [CrossRef]
17. X. Jiejiang, Z. Peiji, G. Yuanhao, L. Zhiyong, T. Manqing, Y. Yude, and Y. Jinzhong, “Nonblocking 4 x 4 silicon electro-optic switch matrix with low power consumption,” IEEE Photon. Technol. Lett. 27, 1434–1436 (2015). [CrossRef]
18. P. Dasmahapatra, R. Stabile, A. Rohit, and K. A. Williams, “Crossbar switch matrix using fifth-order resonators,” 10th International Conference on Group IV Photonics (GFP), 11–12 (2013). [CrossRef]
19. N. Sherwood-Droz, H. Wang, L. Chen, B. G. Lee, A. Biberman, K. Bergman, and M. Lipson, “Optical 4x4 hitless slicon router for optical networks-on-chip (NoC),” Opt. Express 16(20), 15915–15922 (2008). [CrossRef] [PubMed]
20. R. Q. Ji, J. Xu, and L. Yang, “Five-port optical router based on microring switches for photonic networks-on-chip,” IEEE Photonics Technol. Lett. 25(5), 492–495 (2013). [CrossRef]
21. W. Wang, H. Zhou, J. Yang, M. Wang, and X. Jiang, “Highly integrated 3×3 silicon thermo-optical switch using a single combined phase shifter for optical interconnects,” Opt. Lett. 37(12), 2307–2309 (2012). [CrossRef] [PubMed]
22. L. Lu, L. Zhou, S. Li, Z. Li, X. Li, and J. Chen, “4 x 4 nonblocking silicon thermo-optic switches based on multimode interferometers,” J. Lightwave Technol. 33(4), 857–864 (2015). [CrossRef]
23. Z. Li, L. Zhou, L. Lu, and J. Chen, “4×4 strictly non-blocking optical switch fabric based on cascaded multimode interferometers,” 14th International Conference on Optical Communications and Networks (ICOCN), pp. 1–3, 2015.
24. L. Lu, L. Zhou, X. Li, and J. Chen, “Low-power 2×2 silicon electro-optic switches based on double-ring assisted Mach-Zehnder interferometers,” Opt. Lett. 39(6), 1633–1636 (2014). [CrossRef] [PubMed]
25. L. Lu, L. Zhou, Z. Li, D. Li, S. Zhao, X. Li, and J. Chen, “4×4 silicon optical switches based on double-ring-assisted Mach-Zehnder interferometers,” IEEE Photon. Technol. Lett. 27, 2457–2460 (2015). [CrossRef]
26. S. Han, T. J. Seok, N. Quack, B.-W. Yoo, and M. C. Wu, “Large-scale silicon photonic switches with movable directional couplers,” Optica 2(4), 370–375 (2015). [CrossRef]
27. T. J. Seok, N. Quack, S. Han, R. S. Muller, and M. C. Wu, “Large-scale broadband digital silicon photonic switches with vertical adiabatic couplers,” Optica 3(1), 64–70 (2016). [CrossRef]
28. L. B. Soldano and E. C. M. Pennings, “Optical multi-mode interference devices based on self-imaging: principles and applications,” J. Lightwave Technol. 13(4), 615–627 (1995). [CrossRef]
29. C.-H. Chiu and C.-H. Chiu, “Taper-integrated multimode-interference based waveguide crossing design,” IEEE J. Quantum Electron. 46(11), 1656–1661 (2010). [CrossRef]
30. Y. Liu, J. M. Shainline, X. Zeng, and M. A. Popović, “Ultra-low-loss CMOS-compatible waveguide crossing arrays based on multimode Bloch waves and imaginary coupling,” Opt. Lett. 39(2), 335–338 (2014). [CrossRef] [PubMed]
31. https://www.lumerical.com/.
32. F. Morichetti, S. Grillanda, M. Carminati, G. Ferrari, M. Sampietro, M. J. Strain, M. Sorel, and A. Melloni, “Non-invasive on-chip light observation by contactless waveguide conductivity monitoring,” IEEE J. Sel. Top. Quantum Electron. 20(4), 292–301 (2014). [CrossRef]
33. 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). [CrossRef]
34. P.-H. Yuen and L.-K. Chen, “Optimization of microring-based interconnection by leveraging the asymmetric behaviors of switching elements,” J. Lightwave Technol. 31(10), 1585–1592 (2013). [CrossRef]
35. L. Zhou, X. Zhang, L. Lu, and J. Chen, “Tunable vernier microring optical filters with p-i-p type microheaters,” IEEE Photonics J. 5, 6601211 (2013). [CrossRef]
