harnessing mode selective nonlinear optics for on chip multi channel all optical signal processing

Chen Department of Electrical and Computer Engineering, McGill University, 3480 University Street, Montr´eal, Qu´ebec H3A 0E9, Canada Received 13 September 2016; accepted 24 October 2016

Harnessing mode-selective nonlinear optics for on-chip multi-channel all-optical signal processing

Ming Ma and Lawrence R Chen

Citation: APL Photonics 1, 086104 (2016); doi: 10.1063/1.4967205

View online: http://dx.doi.org/10.1063/1.4967205

View Table of Contents: http://aip.scitation.org/toc/app/1/8

Published by the American Institute of Physics

Articles you may be interested in

Invited Article: Electrically tunable silicon-based on-chip microdisk resonator for integrated microwave photonic applications

APL Photonics 1, 080801080801 (2016); 10.1063/1.4961685

Hybrid waveguide-bulk multi-path interferometer with switchable amplitude and phase

APL Photonics 1, 081302081302 (2016); 10.1063/1.4960204

160 Gbit/s photonics wireless transmission in the 300-500 GHz band

APL Photonics 1, 081301081301 (2016); 10.1063/1.4960136

Sub-megahertz linewidth single photon source

APL Photonics 1, 096101096101 (2016); 10.1063/1.4966915

Harnessing mode-selective nonlinear optics for on-chip multi-channel all-optical signal processing

Ming Ma and Lawrence R Chen

Department of Electrical and Computer Engineering, McGill University,

3480 University Street, Montr´eal, Qu´ebec H3A 0E9, Canada

(Received 13 September 2016; accepted 24 October 2016; published online 9 November 2016)

All-optical signal processing based on nonlinear optical effects allows for the real-ization of important functions in telecommunications including wavelength conver-sion, optical multiplexing/demultiplexing, Fourier transformation, and regeneration, amongst others, on ultrafast time scales to support high data rate transmission In integrated photonic subsystems, the majority of all-optical signal processing systems demonstrated to date typically process only a single channel at a time or perform a single processing function, which imposes a serious limitation on the functionality

of integrated solutions Here, we demonstrate how nonlinear optical effects can be harnessed in a mode-selective manner to perform simultaneous multi-channel (two) and multi-functional optical signal processing (i.e., regenerative wavelength conver-sion) in an integrated silicon photonic device This approach, which can be scaled

to a higher number of channels, opens up a new degree of freedom for performing

a broad range of multi-channel nonlinear optical signal processing functions using a

single integrated photonic device © 2016 Author(s) All article content, except where

otherwise noted, is licensed under a Creative Commons Attribution (CC BY) license ( http://creativecommons.org/licenses/by/4.0/ ) [http://dx.doi.org/10.1063/1.4967205]

All-optical signal processing, which can take advantage of ultrafast nonlinear effects such

as self-phase modulation (SPM), cross-phase modulation (XPM), and four-wave mixing (FWM), has potential applications in optical communications as it enables functions such as pulse shap-ing/waveform generation, wavelength conversion, optical (de-)multiplexing, Fourier transformation, true-time delay, regeneration, and equalization, to be realized at very high data rates.1 Although electronical approaches are mature and accelerated by large-scale IC integration, all-optical signal processing offers a number of advantages, most notably avoiding the need for optical-to-electrical-to-optical (OEO) conversion The regeneration of signals, which compensates for transmission impairments including dispersion, nonlinear effects, and accumulation of amplified spontaneous emission (ASE) noise, will extend reach and wavelength conversion is of interest for solving the problem of wavelength blocking at network nodes With higher network efficiency and greater scala-bility than their OEO counterparts, all-optical signal regeneration and wavelength conversion are two functions that will be necessary to support the ever-increasing aggregate bandwidth demand in optical communications.2Furthermore, in terms of processing efficiency, it makes greater sense to process signals on different channels in a wavelength-division-multiplexed (WDM) transmission scenario simultaneously rather than handling each channel individually Consequently, many fiber-based solu-tions for multi-channel signal regeneration3 8and multi-channel wavelength conversion9 14have been proposed While these approaches indeed show the possibility to achieve multi-channel all-optical signal processing, they require careful dispersion management (which increases design complexity)

to reduce or avoid unwanted inter-channel nonlinear interactions which otherwise degrade perfor-mance On the other side, significant efforts have been devoted to develop integrated solutions for all-optical signal processing to address issues related to compactness, energy efficiency, and cost efficiency Single-channel all-optical signal regeneration and wavelength conversion have been successfully demonstrated in passive chalcogenide waveguides15 – 19and silicon-on-insulator (SOI) waveguides.20 – 27 In order to realize multi-channel processing in an integrated waveguide, signals 2378-0967/2016/1(8)/086104/9 1, 086104-1 © Author(s) 2016

086104-2 M Ma and L R Chen APL Photonics 1, 086104 (2016)

can propagate bi-directionally in the nonlinear waveguide but this scheme is limited in terms of the number of channels that can be processed

All-optical switching based on a nonlinear differential phase shift between two propagating modes in an elliptical-core two-mode fiber was reported in Ref 28 Similar to this multi-mode fiber-based nonlinear switch, integrated waveguides can be engineered to support a few propagating (spatial) modes which we can exploit to realize “parallel” signal processing This creates a new degree

of freedom for scaling the number of channels that can be processed simultaneously In this paper, we harness nonlinear optical effects in a mode-selective manner in an integrated silicon photonic device

to realize on-chip multi-channel and multi-functional signal processing In particular, we demonstrate XPM-based regenerative wavelength conversion of 10 Gb/s return-to-zero on-off keying (RZ-OOK) signals and achieve large conversion bandwidth (20 nm) on two channels with up to 1.9 dB improve-ment in receiver sensitivity during simultaneous regeneration and a power penalty of <0.5 dB due to inter-channel cross talk

Figure1(a)shows a schematic of our integrated device in SOI, referred to hereafter as the mode selective nonlinear device (MSND) The MSND was fabricated using e-beam lithography with a single etch and occupies a total size of 5.24 mm × 0.46 mm on a wafer The MSND integrates the following components: vertical grating couplers (VGCs) for input and output coupling to single mode fibers, a mode multiplexer and demultiplexer (m-MUX/m-deMUX), and a length (2 cm) of multi-mode nonlinear waveguide (mm-NLWG) All waveguide components sit on top of a 3 µm buried oxide (BOX) layer with a 2 µm thick index-matched top oxide cladding (see Fig.1(c)); the width and

height of the silicon waveguides are denoted as W and H, respectively Since the waveguide height H

in the MSND is fixed (H = 220 nm) and supports one mode in this dimension, we refer to the TE fields along the waveguide width W as TE0, TE1, etc The VGCs that are connected to the input and output waveguides are optimized for TE0mode operation.29The VGC for input/output coupling has a 3-dB bandwidth of ∼30 nm and a coupling loss of ∼6 dB/fiber The total fiber-to-fiber insertion loss of the MSND is 17.5 dB on Channel #1 and 20.1 dB on Channel #2 at 1550 nm (we denote Channel #1 as the transmission from Port #1 to Port #2 and Channel #2 as that from Port #3 to Port #4) All the VGCs were aligned and placed on one side (not shown in the figure schematic); coupling light into and out

of the VGCs is done using a fiber-ribbon array with a 127 µm fiber pitch and a 25◦polishingangle

FIG 1 (a) Schematic of MSND comprising VGCs, m-MUX/m-deMUX, and mm-NLWG (b) ADC structure (c) Cross section of SOI wafer (d) Simulated mode effective indices as a function of waveguide width (e) Intra-channel transmit-tance and inter-channel transmittransmit-tance (i.e., linear inter-channel cross talk) over the wavelength span from 1500 to 1600 nm Simulated dispersion and mode profile (as insets) of the (f) TE and (g) TE modes in the mm-NLWG.

The input and output waveguides of the MSND, identified by port numbers as shown in Fig.1(a), have a width of 500 nm and support only the lowest order TE mode The waveguides at Ports #3 and

#4 are tapered down to a width of 360 nm over a length of 220 µm while the waveguides at Ports #1 and #2 are tapered up to a width of 700 nm over a length of 95 µm before and after the m-MUX and m-deMUX The mm-NLWG has a width of 800 nm and supports the TE0and TE1modes, as well as the first two TM modes which are not excited due to the polarization selective nature of the VGCs The m-MUX and m-deMUX are each based on an asymmetric directional coupler (ADC) as shown

in Fig.1(b)

The general principle of operation in the ADC is as follows.30Based on propagation constant matching, an optical mode in a single-mode waveguide can be evanescently coupled to a particular mode in an adjacent multi-mode waveguide if the effective indices of these two modes are equal Fig.1(d)shows simulated effective indices for different propagated modes over a range of waveguide width and there are a variety of phase matching conditions whereby the TE0can be transformed into the TE1mode In our case, we choose the width of the upper waveguide in the ADC as 360 nm; the simulated effective index of TE0in this waveguide is 2.09 In order to fulfil the mode transformation condition, the lower waveguide varies linearly from 700 nm to 800 nm, in which the effective index of

TE1mode in the waveguide distributes continuously over a range of 1.95–2.17 Furthermore, in order

to maximize the coupling efficiency of TE0to TE1, the coupling gap g and coupling length L cof the ADCs are optimized and determined as 100 nm and 30 µm, respectively In this case, the coupling efficiency from TE0to TE1can reach as high as 91% over a wavelength range of 1500–1600 nm

An input TE0 mode launched in the upper waveguide on the input side of the ADC (i.e., via Port #3) will be converted to the TE1mode on the output side and propagates as the TE1mode in the mm-NLWG On the other hand, a TE0mode launched in the lower waveguide on the input side of the ADC (i.e., via Port #1) will maintain its mode profile at the ADC output and propagate as the TE0

mode in the mm-NLWG If we launch a high power pump along with a signal in Port #3, nonlinear interactions (e.g., XPM and/or FWM) can occur in the mm-NLWG (both pump and signal propagate

as TE1 modes); similarly, nonlinear interactions can occur if both pump and signal are launched in Port #1 (where they propagate as TE0 modes in the mm-NLWG) On the other hand, no nonlinear interactions occur between the pump and signal if they are launched in separate ports In this way,

we can take advantage of being able to excite nonlinear effects in a mode-selective manner to realize simultaneous multi-channel and multi-functional optical signal processing (particularly, XPM-based regenerative wavelength conversion)

Figure1(e)compares the intra-channel and the inter-channel transmittance of the MSND mea-sured from 1500 nm to 1600 nm The linear inter-channel cross talk (i.e., transmission from Port #1 to

#4 or from Port #3 to #2) is at least 15 dB lower than the corresponding intra-channel transmittance Figures1(f)and1(g)show the simulated dispersion and mode profile of the TE0 and TE1modes, respectively, in the mm-NLWG The dispersion and dispersion slope at 1550 nm for the TE0mode are5.55 × 104ps/(nm cm) and 1.75 × 106ps/(nm2cm), respectively, whereas these values for the

TE1mode are 1.56 × 103ps/(nm cm) and1.51 × 105ps/(nm2cm), respectively Using 3D-FDTD, the effective mode areas of the TE0and TE1modes at 1550 nm are calculated to be 0.094 µm2 and 0.122 µm2, respectively

For nonlinear inter-channel cross talk measurements, a mode-locked fiber laser generating pulses

at 1550 nm with a repetition rate of 9.95328 GHz is amplified before being merged with a CW wave-length tunable laser at 1540 nm or 1560 nm Then, both the CW probe (average power 10 mW) and the pulsed pump (average power 51 mW) are launched onto the input port of either Chan-nel #1 or ChanChan-nel #2 and the output spectra are measured using an optical spectrum analyzer with a 0.1 nm resolution bandwidth at the corresponding output port of each channel Figure2shows when XPM-induced broadening occurs: for example, in (a) and (b) both pump and probe are launched onto the same channel (i.e., both at Port #1 for Channel #1 or Port #3 for Channel #2) whereas in (c) and (d), they are on different channels, i.e., in (c) the probe is on Channel #1 while the pump is on Channel #2 and in (d), the probe is on Channel #2 while the pump is on Channel #1 In all cases, the pump is fixed at a wavelength of 1550 nm while the probe is centered either at 1540 nm (red dashed curves) or 1560 nm (blue solid curves) The spectral measurements are observed at the output port which shares the same channel with the probe input port Clearly, significant XPM-induced spectral

086104-4 M Ma and L R Chen APL Photonics 1, 086104 (2016)

FIG 2 Comparison of the spectral output from the MSND when the probe (at 1540 nm in red dashed curve and at 1560 nm

in blue solid curve) and pump (at 1550 nm) are on the same channel, i.e., (a) on Channel #1 and (b) on Channel #2, and

on different channels, i.e., (c) probe on Channel #1 and pump on Channel #2 and (d) probe on Channel #2 and pump on Channel #1.

broadening occurs only when both the probe and the pump are on the same channel (the estimated nonlinear phase shifts on Channel #1 and Channel #2 are ∼0.32π and ∼0.25π, respectively) On the other hand, XPM is not observed when the probe and pump are on different channels Note that by comparing Figs.2(a)to2(c)or2(b)to2(d), we determine the nonlinear inter-channel cross talk to

be <30 dB so that it can be neglected

The experimental setup for single-channel regenerative wavelength conversion is shown in Fig.3

and is essentially a probe-pump heterodyne configuration: a return-to-zero (RZ) 10 Gb/s data signal

at 1550 nm (pump) is combined with a continuous wave (CW) at 1540 nm (probe) by a WDM

FIG 3 Experimental setup for single-channel regenerative wavelength conversion Spectrum of single-channel operation on (a) Channel #1 and (b) Channel #2 (the peaks of input spectrum and output spectrum are aligned) In regenerative wavelength conversion, a portion of the optical power in the XPM-induced sideband (depicted as solid blue or red lines) is detected by the optical receiver PC: polarization controller; PRBS: pseudorandom bit sequence; MZM: Mach-Zehnder modulator; EDFA: erbium-doped fiber amplifier; VOA: variable optical attenuator; OBPF: optical bandpass filter; cw: continuous-wave; WSS: Finisar WaveShaper.

coupler and they interact nonlinearly on either Channel #1 or Channel #2 of the MSND In detail, the RZ-OOK signals with a PRBS length of 231 1 at 9.953 28 Gb/s were generated from a mode-locked fiber laser operating at 1550 nm and a LiNO3-based Mach-Zehnder modulator (MZM) Noise loading consists of an EDFA and a VOA and is used to control the optical signal to noise ratio (OSNR) of signals from 30 dB down to 11 dB The data signal was amplified and acted as a pump with an average power of 63 mW and a duty cycle of 15% into the MSND A WDM coupler combined the data signal with another CW laser at 1540 nm (an average power of 20 mW) and launched them both onto either Channel #1 or Channel #2 Intra-channel XPM occurred to the probe and a portion of XPM-induced sideband was filtered out by a 1 nm bandwidth programmer filter (Finisar Waveshaper 1000S) A VOA controlled the extracted optical power into the optical receiver that comprised of an EDFA, an OBPF (1 nm bandwidth), and a high-speed photodiode (3 dB bandwidth 45 GHz)

As shown in the insets in Fig.3, XPM occurs on the probe (FWM is also generated) and as described above, a portion of the XPM-induced sideband is retained using carrier-offset filtering In the OOK data stream, only the “1” data bits, and not the “0” data bits, give rise to XPM As a result, only “1” bits are converted to the probe wavelength Meanwhile, this nonlinear process establishes

an improvement in the Q-factor of the signal (as shown in Fig.4(c))

First, we degrade the data signals by deliberately tuning the bias voltage applied to the MZM away from its optimized value (e.g., to reduce the extinction ratio and create optical power for

a “0” bit without intentional ASE noise loading) Figure4(c)shows a typical degraded eye-diagram

at a received power of 31 dBm with noticeable “noise” on both “1” and “0” bits; the Q-factor

is 5.8 dB Following regenerative wavelength conversion, the noise on the “0” bits is suppressed and the Q-factor is improved by 1.2 dB and 0.9 dB on Channel #1 and Channel #2, respectively Figure4(a)shows that the corresponding improvement in receiver sensitivity is 1.7 dB and 1.0 dB

FIG 4 (a) BER measurement for the regeneration with degraded signals; receiver sensitivity after regeneration has been improved by 1.7 dB and 1.0 dB on Channel #1 and Channel #2, respectively (b) V-curve comparison for before and after regeneration (c) Degraded signals with a Q-factor of 5.8 dB have been enhanced to 7.0 dB and 6.7 dB on Channel #1 and Channel #2, respectively (d) Influence of OSNR on receiver sensitivity before and after regenerative wavelength conversion: power penalty exists above 21 dB OSNR whereas sensitivity improvement occurs as the OSNR drops below 21 dB (e) Representative eye-diagrams for input OSNR values of 11, 17, 25, and 30 dB (time scale: 10 ps/div) In (b), (c), and (e), the received power is 31 dBm.

086104-6 M Ma and L R Chen APL Photonics 1, 086104 (2016)

on Channel #1 and Channel #2, respectively We also measure the bit error rate (BER) as a function

of decision threshold voltage at a received power of31 dBm, see Fig.4(b), which shows that the V-shape moves closer to 0 V after regeneration, i.e., that the threshold voltage has been reduced This shift indicates that the noise on the “0” bits has been reduced Next, we optimize the DC bias voltage applied to the MZM for maximum extinction ratio and degrade the 10 Gb/s signal by ASE-noise load-ing only Figure4(d)illustrates the receiver sensitivity after regenerative wavelength conversion is plotted as a function of the optical signal to noise ratio (OSNR) of the input signal; representative eye-diagrams before and after regenerative wavelength conversion on both channels are shown in Fig.4(e)

[all eye-diagram measurements were made at a received power of31 dBm] As can be observed

in Fig.4(d), the sensitivity of the degraded signals worsens by 3.4 dB as the OSNR drops from

30 dB to 11 dB However, the sensitivity of regenerated signals on Channel #1 remains at31.3 dBm over the same OSNR range and it is improved by 2.2 dB when the OSNR is 11 dB; similarly, the receiver sensitivity on Channel #2 is also improved by 1.5 dB at an input OSNR of 11 dB Note that when the OSNR is higher than 21 dB, a power penalty, rather than an improvement, occurs to the nominal regenerated signals For such high values of input OSNR, the signal process is more akin to wavelength conversion, as opposed to regenerative wavelength conversion, which introduces a power penalty.31

In our experiments, the duty cycle of the RZ-OOK pulses is smaller than what might nominally

be used (15% compared to 33%-40%) For longer RZ-OOK pulses, a higher average power is nec-essary in order to obtain the same peak power (and correspondingly, nonlinear phase shift) System simulations based on the experimental setup and that account for the device characteristics show that regenerative wavelength conversion can be obtained for RZ-OOK pulses having a duty cycle of 34% with comparable improvements to the case of a duty cycle of 15% duty using ∼3.4 dB more average power

We now demonstrate simultaneous multi-channel regenerative wavelength conversion in our MSND with the setup depicted in Fig 5 The same 10 Gb/s data generation unit as that in the

FIG 5 Experimental setup for simultaneous multi-channel regenerative wavelength conversion Insets: (a) spectrum com-parison on Channel #2 when Channel #1 was active or idle; (b) spectrum comcom-parison on Channel #1 when Channel #2 was active or idle.

single-channel operation (Fig.3) was split into 2 branches with the signals in one branch being de-correlated from the other one by a 2 m length of SMF-28 fiber We duplicated the probe generation and noise-loading units to create two similar probe-pump sets These 2 sets of probe and pump were launched into the two channels of the MSND simultaneously to simulate concurrent multi-channel operation At the output ports of the MSND, the wavelength-converted and regenerated signals were detected and measured one channel at a time while the other channel was active and maintained at

an OSNR of 14 dB

In order to exploit the potential of wavelength conversion, the signals were wavelength-converted

to 1560 nm on Channel #1 and to 1540 nm on Channel #2 The spectrum in Fig.5(a)shows the XPM-broadening for Channel #2 when Channel #1 is active or inactive; a similar spectral comparison for Channel #1 is given in Fig.5(b) As can be seen, very little inter-channel interference occurs to the XPM-broadening for the channel being processed Out of the MSND, the regenerated, wavelength-converted signals are detected and characterized one channel at a time We compared the receiver sensitivity for before and after regenerator in 4 scenarios: (1) Channel #1 operation alone, (2) Channel

#2 operation alone, (3) Channel #1 operation with Channel #2 active, and (4) Channel #2 operation with Channel #1 active

Figure6demonstrates the multi-channel regenerative wavelength conversion capability of our device and compares the performance to single channel operation For simultaneous multi-channel operation, the receiver sensitivity of Channel #1 after regenerative wavelength conversion is main-tained at31.5 dBm over an input OSNR range of 30 dB–11 dB As a result of inter-channel cross talk, however, the receiver sensitivity for multi-channel operation is 0.5 dB worse compared to single-channel operation on Channel #1 Similarly, the receiver sensitivity after regenerative wave-length conversion on Channel #2 for multi-channel operation is degraded by 0.4 dB compared to that for single-channel operation Based on these results, we determine that the regeneration performance

on each channel during multi-channel operation is not compromised as a result of inter-channel cross talk

The phase-matching condition in transforming the TE0 mode to higher-order mode does not impose an upper limit on how many higher-order modes can be excited through the ADC For exam-ple, Fig.7shows that the TE0mode can be transformed into all higher-order modes (TE1to TE9) for a

mm-NLWG with W = 4.5 µm and H = 220 nm through phase-matching condition with an effective

index of 2.3 at 1550 nm (this enables potentially simultaneous processing of 10 channels) However, as Fig.7also shows, the effective indices of the TE modes tend to approach constant values as the waveg-uide width increases and their difference decreases A smaller index difference may result in greater

FIG 6 (a) Dependence of receiver sensitivity on the OSNR of input signals in the multi-channel regenerative wavelength conversion (b) Eye-diagrams for the multi-channel operation are grouped as degraded signals, regenerated signals on Channel

#1 and regenerated signals on Channel #2 (time scale: 10 ps/div) All eye-diagrams were measured at a received power of

31 dBm.

086104-8 M Ma and L R Chen APL Photonics 1, 086104 (2016)

FIG 7 Simulated effective indices at 1550 nm for the first 10 order TE modes in the mm-NLWG with W = 4.5 µm and H

= 220 nm Based on the phase matching condition with an effective index of 2.3, the TE 0 may be transformed into 9 higher-order TE modes (TE 1 to TE 9 ) The effective modal area at 1550 nm for each mode is calculated with 3D-FDTD and given in the parentheses (in units of µm 2 ).

inter-channel cross talk during mode-conversion and hence the ADCs must be designed carefully to

support a large number of channels Experimentally, Wang et al demonstrated an ADC capable of

exciting up to 8 modes in a multi-mode waveguide (TE0-TE3and TM0-TM3).32By exploiting polar-ization diversity, bidirectional propagation, and an appropriately designed mm-NLWG, our approach

to achieve simultaneous multi-channel all-optical signal processing by harnessing nonlinear optical effects in a mode-selective manner can be scaled up to a higher channel count The corresponding effective modal areas for the aforementioned TE modes are given in the parentheses in Fig.7 and

as expected, they are larger for higher-order modes Thus, the corresponding peak power required will be greater for higher-order modes; for example, to achieve a π phase change, the peak signal (pump) power required for the TE9mode is ∼1.26 times greater than that for the TE0mode Lastly, reducing the coupling losses with optimized VGC designs (e.g., a coupling efficiency of0.58 dB was recently reported33) will help ensure that nonlinear effects can be stimulated at lower optical input power levels

We have demonstrated simultaneous multi-channel signal regenerative wavelength conversion for two 10 Gb/s RZ-OOK signals by harnessing nonlinear optical effects in a mode-selective manner First, broadband (20 nm) wavelength conversion was achieved Second, thanks to simultaneous regeneration, the receiver sensitivity of input degraded signals on Channel #1 and Channel #2 was improved by up to 1.9 dB and 1.3 dB, respectively Furthermore, the power penalty due to inter-channel cross talk is less than 0.5 dB Finally, owing to the use of ultrafast nonlinear effects, higher data rates are possible

The design of the MSND can be implemented using other nonlinear materials/platforms, e.g., silicon nitride34,35or chalcogenide.15For example, silicon nitride offers small propagation loss (on the order of 0.03 dB/cm35), relatively large nonlinearity, and negligible nonlinear loss Moreover, apart from nonlinear optical signal processing, the MSND can potentially be used in microwave photonics applications, and in particular, where nonlinear optics is used for microwave signal processing, such as XPM-based radio-frequency spectrum analysis36and FWM-based instantaneous frequency measurements.37In conclusion, the method of harnessing nonlinear optics in a mode-selective manner has the potential to scale to a higher number of channels and opens up a new degree of freedom in realizing various multi-channel all-optical signal processing and microwave photonics functionalities

in an integrated photonic device

The devices were fabricated by Richard Bokjko at the University of Washington Nanofabrication Facility, a member of the National Science Foundation’s National Nanotechnology Infrastructure Network We thank the NSERC CREATE NGON and NSERC CREATE Si-EPIC programs for financial support

1 A E Willner, S Khaleghi, M R Chitgarha, and O F Yilmaz, J Lightwave Technol. 32, 660 (2014).

2 A A M Saleh and J M Simmons, Proc IEEE 100, 1105 (2012).

3 F Parmigiani, L Provost, P Petropoulos, D J Richardson, W Freude, J Leuthold, A D Ellis, and I Tomkos, IEEE J Sel Top Quantum Electron. 18, 689 (2012).

4 L Provost, F Parmigiani, C Finot, K Mukasa, P Petropoulos, and D J Richardson, Opt Express 16, 2264 (2008).

5 K Cvecek, K Sponsel, C Stephan, G Onishchukov, R Ludwig, C Schubert, B Schmauss, and G Leuchs, Opt Express

16, 1923 (2008).

6 C Nan, X Lin, K S Berg, T Tokle, and P Jeppesen, IEEE Photonics Technol Lett. 14, 1581 (2002).

7 M Vasilyev and T I Lakoba, Opt Lett. 30, 1458 (2005).

8 L Provost, F Parmigiani, P Petropoulos, and D J Richardson, IEEE Photonics Technol Lett. 20, 270 (2008).

9 P Devgan, R Tang, V S Grigoryan, and P Kumar, J Lightwave Technol. 24, 3677 (2006).

10 H Nguyen Tan, M Matsuura, T Katafuchi, and N Kishi, Opt Express 17, 22960 (2009).

11 X Li, J Yu, Z Dong, and N Chi, Opt Express 20, 21324 (2012).

12 M.-F Huang, J Yu, and G.-K Chang, Opt Express16, 21161 (2008).

13J Yu, M F Huang, and N Cvijetic, in Proceedings of Optical Fiber Communication Conference Paper OThS7 (OFC/NFOEC

2009, San Diego) (Optical Society of America, 2009).

14 V J F Ranca, F Parmigiani, P Petropoulos, and D J Richardson, J Lightwave Technol. 32, 3027 (2014).

15 B J Eggleton, B Luther-Davies, and K Richardson, Nat Photonics 5, 141 (2011).

16 V G Ta’eed, M Shokooh-Saremi, L Fu, D J Moss, M Rochette, I C M Littler, B J Eggleton, Y Ruan, and B Luther-Davies, Opt Lett. 30, 2900 (2005).

17 V G Ta’eed, M Shokooh-Saremi, L Fu, I C M Littler, D J Moss, M Rochette, B J Eggleton, Y Ruan, and B Luther-Davies, IEEE J Sel Top Quantum Electron. 12, 360 (2006).

18 M D Pelusi, V G Ta’eed, L Fu, E Magi, M R E Lamont, S Madden, D Y Choi, D A P Bulla, B Luther-Davies, and

B J Eggleton, IEEE J Sel Top Quantum Electron. 14, 529 (2008).

19 M D Pelusi, F Luan, S Madden, D Y Choi, D A Bulla, B Luther-Davies, and B J Eggleton, IEEE Photonics Technol Lett. 22, 3 (2010).

20 J Leuthold, C Koos, and W Freude, Nat Photonics 4, 535 (2010).

21 R Salem, M A Foster, A C Turner, D F Geraghty, M Lipson, and A L Gaeta, Opt Express 15, 7802 (2007).

22 R Salem, M A Foster, A C Turner, D F Geraghty, M Lipson, and A L Gaeta, Nat Photonics 2, 35 (2008).

23 R L Espinola, J I Dadap, R M Osgood, S J McNab, and Y A Vlasov, Opt Express 13, 4341 (2005).

24 Y.-H Kuo, H Rong, V Sih, S Xu, M Paniccia, and O Cohen, Opt Express14, 11721 (2006).

25 M A Foster, A C Turner, R Salem, M Lipson, and A L Gaeta, Opt Express 15, 12949 (2007).

26 H Hu, H Ji, M Galili, M Pu, C Peucheret, H C H Mulvad, K Yvind, J M Hvam, P Jeppesen, and L K Oxenløwe,

Opt Express 19, 19886 (2011).

27 R Adams, M Spasojevic, M Chagnon, M Malekiha, J Li, D V Plant, and L R Chen, Opt Express 22, 4083 (2014).

28 H G Park, C C Pohalski, and B Y Kim, Opt Lett. 13, 776 (1988).

29 Y Wang, X Wang, J Flueckiger, H Yun, W Shi, R Bojko, N A F Jaeger, and L Chrostowski, Opt Express 22, 20652 (2014).

30 Y Ding, J Xu, F Da Ros, B Huang, H Ou, and C Peucheret, Opt Express21, 10376 (2013).

31M Ma and L R Chen, in OSA Technical Digest of Conference on Lasers and Electro-Optics Paper STh3E.3 (CLEO 2016,

San Jose) (Optical Society of America, 2016).

32 J Wang, S He, and D Dai, Laser Photonics Rev. 8, L18 (2014).

33 Y Ding, C Peucheret, H Ou, and K Yvind, Opt Lett. 39, 5348 (2014).

34 D J Moss, R Morandotti, A L Gaeta, and M Lipson, Nat Photonics 7, 597 (2013).

35 J F Bauters, M J R Heck, D John, D Dai, M.-C Tien, J S Barton, A Leinse, R G Heideman, D J Blumenthal, and

J E Bowers, Opt Express 19, 3163 (2011).

36 C Dorrer and D N Maywar, J Lightwave Technol. 22, 266 (2004).

37 M Pagani, B Morrison, Y Zhang, A Casas-Bedoya, T Aalto, M Harjanne, M Kapulainen, B J Eggleton, and D Marpaung,

Optica 2, 751 (2015).

#1 and regenerated signals on Channel #2 (time scale: 10 ps/div) All. .. wave-length conversion on Channel #2 for multi- channel operation is degraded by 0.4 dB compared to that for single -channel operation Based on these results, we determine that the regeneration performance