Ultrafast All-Optical Switching in a Silicon-Nanocrystal-Based Silicon Slot Waveguide at Telecom Wavelengths

Ultrafast All-Optical Switching in aSilicon-Nanocrystal-Based Silicon SlotWaveguide at Telecom WavelengthsAlejandro Martınez,†,* Javier Blasco,† Pablo Sanchis,† Jose V. Galan,† Jaime Garcıa-Ruperez,†Emmanuel Jordana,‡ Pauline Gautier,‡ Youcef Lebour,§ Sergi Hernandez,§ Romain Guider,|Nicola Daldosso,| Blas Garrido,§ Jean Marc Fedeli,‡ Lorenzo Pavesi,| and Javier Martı†

†Nanophotonics Technology Center (NTC), Universidad Politecnica de Valencia, Camino de Vera s/n,46022, Valencia, Spain, ‡CEA-LETI, Minatec, 17 rue des Martyrs, Grenoble, France, §Departament d’Electronica,Universitat de Barcelona, Carrer Martı i Franques 1 08028 Barcelona, Spain, and |Nanoscience Laboratory,Department of Physics, University of Trento, via Sommarive 14, 38122 Trento, Italy

ABSTRACT We demonstrate experimentally all-optical switching on a silicon chip at telecom wavelengths. The switching devicecomprises a compact ring resonator formed by horizontal silicon slot waveguides filled with highly nonlinear silicon nanocrystals insilica. When pumping at power levels about 100 mW using 10 ps pulses, more than 50% modulation depth is observed at the switchoutput. The switch performs about 1 order of magnitude faster than previous approaches on silicon and is fully fabricated usingcomplementary metal oxide semiconductor technologies.

KEYWORDS CMOS photonics, all-optical switching, silicon nanocrystals, silicon slot waveguides

In the last years, silicon photonics has become one of themost active fields within optics. This can be explainedby considering that the development of photonic inte-

grated circuits on silicon-on-insulator (SOI) wafers usingconventional complementary metal-oxide-semiconductor(CMOS) tools and processes would result in a significantreduction in costs because of the economies of mass-manufacturing. Recently, several high-speed electro-opticalfunctionalities have been demonstrated, such as a 40 Gbit/smodulator based on the free-carrier (FC) plasma dispersioneffect.1 However, in order to reach higher speeds (100 Gbit/sand beyond) the electrical domain must be completelyavoided and the information processing must be carried outentirely in the optical domain (all-optical processing). To thisend, ultrafast third-order nonlinear effects need to be exploited.Unfortunately, silicon suffers from a weak ultrafast Kerr coef-ficient (n2 ) 4 × 10-14 cm2/W at a wavelength λ ) 1550 nm)and so two-photon absorption (TPA) becomes the dominantnonlinear optical effect at telecom wavelengths.2 As a result,the generation of FCs by single or two photon absorption hasbeen demonstrated as an effective means of achieving all-optical switching and modulation.3-7 Typically, a photoniccavity is used to increase light-matter interaction,3 whichresults in a more efficient nonlinear performance. However,the FC-based approach has two important drawbacks, lowswitching speeds due to the relatively long (∼ 1 ns) FCrecombination time and low fan-out due to relatively strong

light absorption. The introduction of nanometric defects inthe silicon waveguide shortens the recombination time to afew tens of picoseconds,5-7 but simultaneously increaseslight scattering, that is, optical losses. Final device perfor-mance is thus poor with few prospects for improvement.

Since in silicon TPA dominates over the Kerr effect attelecom wavelengths, other materials with high Kerr coef-ficients have to be considered. A main requirement is thatthe chosen material should also be fully processed in a CMOSfoundry to meet the key requirement of mass-manufactur-ability. From this point of view, silicon nanocrystals (Si-nc)embedded in silica (SiO2), Si-nc/SiO2, is a material whichaccomplishes both requirements. First, it has been shownthat Si-nc/SiO2 displays a high n2 at telecom wavelengths asa consequence of both quantum confinement and dielectricmismatch effects.8 And second, this material is produced inconventional CMOS foundries by standard methods such aslow-pressure chemical vapor deposition (LPCVD) or plasma-enhanced chemical vapor deposition (PECVD). Indeed, sili-con nanocrystals are used to manufacture CMOS flashmemories.9 Since the refractive index of Si-nc/SiO2 is closeto that of silica for low silicon content, waveguides made ofthis material would not allow for strong field confinement.To overcome this issue, the slot waveguide has been dem-onstrated to be a suitable way to strongly confine and guidelight in a low index medium.10

Inthiswork,weproposeahorizontalsiliconslotwaveguidefilled with Si-nc/SiO2 as a highly Kerr nonlinear medium toachieve ultrafast all-optical switching on silicon. The termhorizontal means that the electric field discontinuity that

* To whom correspondence should be addressed. E-mail: [email protected] for review: 12/11/2009Published on Web: 03/31/2010

pubs.acs.org/NanoLett

© 2010 American Chemical Society 1506 DOI: 10.1021/nl9041017 | Nano Lett. 2010, 10, 1506–1511

provides a strong field confinement inside the slot occurs athorizontal interfaces and so, quasi transverse-magnetic (TM)polarization must be used. A cross-sectional view of thiswaveguide is schematically depicted in Figure 1a. Comparedto the common vertical slot waveguide as originally pro-posed in ref 10, in the horizontal configuration there is nolimit for the minimum achievable slot thickness ts so it canbe reduced well below 100 nm to minimize the effective areaAeff of the guided mode and, as a result, maximize thenonlinear interaction.11 By means of numerical simulationsusing the beam propagation method, we determined thatthe nonlinear interaction (in terms of modification of theeffective index of the waveguide) was maximized when thets was in the range between 25 and 50 nm. In our experi-ments, we chose ts ) 50 nm to ensure uniformity in thedeposition of Si-nc as well as to facilitate the coupling toexternal fiber. As shown in the electric field pattern depictedin Figure 1a, when ts ) 50 nm then the optical field is

strongly confined inside the slot for quasi-TM polarization,resulting in Aeff ≈ 0.06 µm2 (see the electric field pattern inFigure 1a).11 In addition, the horizontal slot can be perfectlyfilled with the nonlinear material unlike its vertical counter-part, as seen in Figure 1b, which shows a scanning electronmicroscope image of the cross-section of one of the fabri-cated waveguides. The fabrication process is fully CMOS(details about the fabrication procedure can be found in ref12) and so all-optical circuits based on slot waveguides filledwith Si-nc/SiO2 can be entirely manufactured in a micro-electronics foundry by use of mainstream processes andtools.

Before fabricating the horizontal slot waveguide circuits,the Si-nc/SiO2 material was deposited onto a transparentsilica substrate forming 500 nm layers in order to test theirlinear and nonlinear properties. The Kerr coefficient n2 aswell as the TPA parameter were measured at 1550 nm usingthe Z-scan technique.13 A femtosecond laser source wasused to differentiate between the Kerr (fast), the free-carriers(slow), and the thermal (even slower) contributions to then2 coefficient. More details about the experimental measure-ments can be found in ref 8. It was found that, at low opticalintensities, the Kerr effect dominates the nonlinear re-sponse. The experimentally determined values (maximum)of the Kerr coefficient n2 and the TPA coefficient � are shownin Table 1 for both deposition methods, PECVD and LPCVD.It should be noticed that the nonlinear performance stronglydepends on silicon excess and the annealing temperature.8

However, values of n2 between one and 2 orders of magni-tude higher than in bulk silicon2 are attainable by properlytailoring the silicon excess, the annealing temperature, andthe deposition process.

Previous results were obtained for relatively thick (incomparison with the slot) Si-nc/SiO2 layers, where the nano-crystals are expected to be homogeneously distributed overthe whole volume. However, it was not clear if similarconditions (mainly in terms of mean size and homogeneity)would be achieved when depositing the Si-nc/SiO2 in thin(50 nm) slots. An energy-filtered transmission electronmicroscopy (EFTEM) analysis of the 50 nm thick LPCVD slotlayer was performed with a 200 kV field-emission Jeol 2010Felectron microscope in order to monitor the location and sizedistribution of Si-nc. The Si-nc contrast in the EFTEM images

FIGURE 1. Description of the nonlinear horizontal slot waveguides.(a) Schematic of the cross-section of the waveguide. The waveguidewidth is w. The lower (upper) strip is made of crystalline (amor-phous) silicon and has a thickness t1 (t2). The slot is filled withSi-nc/SiO2 material and its thickness is ts. The whole waveguide isplaced over a SiO2 layer and covered by tetra-ethyl-ortho-silicate(TEOS). The computed transverse profile of the electric field forquasi-TM polarization for t1 ) t2 ) 220 nm, ts ) 50 nm, w ) 500nm and a 0.08 silicon excess is superimposed. (b) Scanning electronmicroscope (SEM) image of the cross-section of a fabricated slot-waveguide. Measured dimensions are found on the image (ts ) 45nm).

TABLE 1. Experimentally Determined Material NonlinearProperties for Thick (500 nm) Si-nc/SiO2 Layers by Z-ScanMeasurements at λ ) 1550 nma

depositionmethod n2 (cm2/W) � (cm/GW) medium characteristics

PECVD 4 × 10-13 5 0.08 silicon excess, annealingtemperature ) 800 °C

LPCVD 2 × 10-12 50 0.17 silicon excess, annealingtemperature ) 800 °C

a For each deposition method, the observed maximum vales of n2

are given. Silicon excess refers to the amount of silicon in thedeposited silicon-rich oxide in excess with respect to stoichiometricSiO2.

© 2010 American Chemical Society 1507 DOI: 10.1021/nl9041017 | Nano Lett. 2010, 10, 1506-–1511

was enhanced by energetically filtering the electron energy-loss spectra by using a Gatan Image Filter (GIF 2000) aroundthe Si plasmon (placed at 17 eV), which is well separatedfrom the SiO2 plasmon (placed at 22 eV). In this way, it waspossible to detect the appearance of amorphous clusters insidethe oxide matrix and determine their actual size. The resultsare shown in Figure 2. It can be observed that the precipitationis uniform and well achieved (Figure 2a). The observed Si-nchave a medium diameter of 2.4 nm (Figure 2b) and arehomogeneously distributed. This is mainly because LPCVDproceeds at low deposition rate so very thin layers with nm-scale control may be deposited with uniform compositionover the whole thickness. We should mention that this ismore difficult to achieve when using PECVD deposition,which is a faster process and does not provide uniformconcentration of Si-nc in thin layers.

We fabricated a very simple and compact photonicstructure to demonstrate all-optical switching: a ring-resona-tor (RR) coupled to a waveguide (see the diagram depicted

in Figure 3a). It has been proposed that this switchingstructure can perform very efficiently with moderate powerlevels when using slot waveguides filled with Si-nc/SiO2.14

The Si-nc/SiO2 material was deposited by LPCVD, whichallowed for a higher Kerr coefficient as shown previously.After deposition of SiOx with a 0.17 silicon excess, thematerial was annealed at 1000 °C during 3 min 30s underpure N2 ambiance to allow for the formation of these Si-nc.The dimensions of the fabricated horizontal slot waveguideswere w ) 565 nm, t1 ) 220 nm, t2 ) 252 nm, and tS ) 50nm. The measured propagation losses were of the order of20 dB/cm. However, it deserves to be mentioned thatsignificantly lower losses values were also obtained for otherfabrication runs using different Si-nc deposition techniquesand procedures (e.g., about 4 dB/cm for slot waveguides withSi-nc deposited by PECVD15), showing that 500 nm widesilicon horizontal slot waveguides filled by Si-nc (with lowdensity) fabricated with CMOS processes can display muchlower propagation losses. The RR was spaced 250 nm fromthe waveguide and its radius was 20 µm. The transmissionspectrum of the fabricated structure in the linear regime forTM-polarized light is shown in Figure 3b (black curve). Theobserved RR resonance is characterized by a loaded qualityfactor of 1875 (corresponding to a photon lifetime about1.54 ps inside the cavity) and a 17 dB transmission dip. Thelow observed value of the Q factor is mainly due to therelatively large propagation losses of the waveguide, whichin fact become even larger in the RR owing to the curvature.To test the resonance behavior in the dynamic regime wealso measured the optical transmission of a probe signalwhen a high-power pump signal (1 ps pulses at 10 GHzrepetition rate) tuned to an adjacent RR resonance (λpump )1557.5 nm) was injected into the input waveguide. Thepulsed pump signal was generated from a fiber laser with a10 GHz repetition rate and tunable in the range 1530-1565nm. At the output of the chip an optical filter was used toremove the remaining pump signal. It is observed in Figure3b that the resonance dip was clearly red shifted when thepump power was increased (values of the on-chip peakpump power are indicated). This means that the effectiveindex of the RR is increased, which could be attributed tothermal effects or the Kerr nonlinearity, but not to FCexcitation. Additionally, the transmission dip was decreaseddown to values around 6 dB. Since the spectra were recordedwith an optical spectrum analyzer that operates by averagingover a temporal interval much larger than the pulse duration,it can be concluded that the red shift is mainly originatedby slow thermal effects. This leads us to conclude that theRR must be critically coupled when operating under anonlinear regime in order to achieve the optimal perfor-mance of an all-optical switch. Such a switch would resultfrom a more sophisticated design that should take intoaccount thermal effects, as well as dynamic generation ofFCs, inside the RR.

FIGURE 2. EFTEM characterization of the Si-nc/SiO2 layer depositedby LPCVD inside the horizontal slot waveguide. (a) Image showinga homogeneous Si-nc density; (b) size distribution of the Si-nc insidethe slot layer.

© 2010 American Chemical Society 1508 DOI: 10.1021/nl9041017 | Nano Lett. 2010, 10, 1506-–1511

The time response of the switching structure was alsomeasured by using a pump-probe scheme. The pumpsignal was generated by modulating a 10 GHz pulse trainwith a certain digital sequence using an electro-opticalmodulator that was driven by an electrical signal generatedby a 40 Gbit/s bit pattern generator. At the modulator output,the pulse width was estimated to be about 10 ps. Themodulated signal was amplified using an erbium-doped fiberamplifier. The probe signal was generated by an externalcavity laser and was amplified by an erbium-doped fiberamplifier. The polarization of both signals was controlledindependently in order to inject only TM-polarized light intothe silicon chip. Both signals were combined before enteringthe silicon chip. In the experiment, the probe signal wastuned to λCW ) 1544.5 nm, which is close to the RRresonance shown in Figure 3b. However, the RR resonanceis shifted depending on the pump power. This shift depends,as previously argued, on the thermal effects induced by thepump power. Therefore, we could modify the resonanceposition by controlling the pump power. The time evolutionof the output probe signal when the peak pump power atthe waveguide input was 150 mW is shown in Figure 3c. Inthis case, the transmission dip due to the RR resonance wasclose to λCW ) 1544.5 nm so transmission is low in theabsence of the pump pulse. In the presence of the ultrashortpump pulse, the RR resonance is rapidly red shifted and a

positive pulse is registered at the output. These positivepeaks, corresponding to positive pulses (digital 1) of thepump signal, have a duration of 11 ps at half-maximum andhave an extinction ratio ER ) 2.2 dB (40% modulationdepth). This demonstrates the ultrafast nonlinear responseof the all-optical switching structure. Actually, this fastresponse is due to the Kerr nonlinearity being superimposedon a slow (a few nanoseconds) and weak response causedby the excitation of FCs inside the RR. Interestingly, this slowresponse shows both cumulative effects and long recoverytimes. The slow response variation is below 1 dB for theconsidered digital sequence. In a real device, this slow effectcould be completely avoided by appropriately encoding thedigital sequences (e.g., line codes such as 8B/10B)16 that arewidely used to maintain the DC-balance and remove low-frequency detrimental effects in digital communicationsystems.

We obtained the results shown in Figure 3d by narrowingthe recorded time window to that corresponding to a singleoutput pulse at the probe wavelength λCW ) 1544.5 nm. Themeasured output pulse when the on-chip pump peak powerwas 160mW is shown in blue. In this case, the probe signalis in the RR resonance and, therefore, the transmissionincreases as the RR resonance is red shifted due to the pumppulses. An ER ) 3.6 dB (more than 50% of modulationdepth) with an 11 ps duration at half-maximum is observed.

FIGURE 3. All-optical switching structure based on a RR coupled to a waveguide. (a) Scheme of the switching structure showing a SEM imageof the RR (radius 20 µm). (b) Transmission spectra of a RR resonance at different peak levels (indicated in the figure) of a pump signal injectedat an adjacent resonance (λpump )1557.5 nm). The black curve spectrum was measured by swapping the wavelength of a CW optical signaland recording the power at the output. The other spectra were measured in an optical spectrum analyzer from the amplified spontaneousemission noise coming from the erbium-doped fiber amplifier used to boost the pump signal. The probe wavelength λCW used to obtain theresults in (c) and (d) is shown as a vertical dashed arrow. (c,d) Probe-signal time waveform at the output of the switching structure. In (c), theon-chip pump peak power (Ppump) is 150 mW and λpump ) 1557.5 nm (corresponding to a RR resonance). In (d), each curve corresponds todifferent values of Ppump and λCW (indicated in the figure).

© 2010 American Chemical Society 1509 DOI: 10.1021/nl9041017 | Nano Lett. 2010, 10, 1506-–1511

This considerably overlaps the pump pulse time duration,which is further proof of the instantaneous Kerr effect. Whenthe on-chip peak pump power was reduced to 80mW, theoutput pulse shown in red was recorded at λCW. In this case,the RR resonance is slightly blue shifted due to the thermaleffects since is higher than the resonance. In both cases, atransient effect that is opposite to the Kerr nonlinearitydynamics is observed at the beginning of the output pulse.This may be attributed to the onset of the excitation of FCsthat produces a negative change in the refractive index.However, and in contrast with the behavior observed insilicon wires, the Kerr effect dominates the nonlinear dy-namic response of the switching structure so FCs onlyproduce a slow response with an amplitude variation muchsmaller than the fast variation originated by Kerr. For thesake of comparison, when the probe signal is tuned far fromthe RR resonance, no switching is observed in the presenceof the pump pulse (yellow curve in Figure 3d). Similarexperiments performed with quasi-TE polarization (i.e., nofield enhancement in the slot region) did not produce anyappreciable fast switching effect, and this rules out the roleof the silicon strips in the observed effect.

These results lead us to conclude that this all-opticalswitching structure performs about 10 times faster thanother previous silicon switches in which both pump andprobe signals are both within the third telecom window.3-5

In addition, the required on-chip pump power (<100mW forthe red curve in Figure 3d) is also lower than in refs 4 and 5.An ER of the order of those reported in ref 3 (∼8 dB) and ref5 (∼9 dB) could be also achieved by using a RR with a higherquality factor and designed to perform optimally underdynamic working conditions. Moreover, in our approachboth probe and pump signals are placed in the telecomwindow. In this sense, the presented device also improvesother approaches that make use of pump signals within thesilicon absorption band (410 nm in ref 7 or 800 nm in ref 6)and, therefore, would be useless in a realistic network. Thepresented result can therefore be considered as a proof-of-concept of ultrafast all-optical switching at telecom wave-lengths on a CMOS silicon chip.

It should be mentioned that the slot waveguide could alsobe filled with other low-index highly nonlinear materials inorder to obtain very good nonlinear performance on a siliconchip. For instance, a vertical slot waveguide filled with anonlinear polymer (n2 ) 1.7 × 10-13 cm2/W) has been usedto demultiplex a 160 Gbit/s signal by means of a four-wavemixing process.17 This vertical configuration is not straight-forward to achieve with CMOS processes because fabricationrequires the formation of the slot using deep ultravioletlithography (limited to a width of around 120 nm, which islarger than those required to minimize Aeff)11 and a subse-quent filling of the air slot with nonlinear material (whichresults in a nonuniform filling and the formation of an airchannel when filling with Si-nc).12 More importantly, thedeposition of polymers is not a typical CMOS process. In

addition, the thermal stability of organic materials is poorerthan that for inorganic materials, which means that theintegration of polymers in CMOS is really a back-end issueas the process has to take place strictly at the end.

In all the measurements, light was injected to and col-lected from the slot waveguides by making use of lensedfibers. At the chip ends, the slot waveguides have a 3 µmwidth that was adiabatically reduced to the 500 nm width.This resulted in very high coupling losses (about 10 dB percoupling point). By use of proper coupling techniques, lightcan be efficiently injected to and extracted from circuits builtwith horizontal slot waveguides. For instance, an invertedtaper can achieve theoretically a 93% coupling efficiency perfacet.18 Grating couplers are not so efficient (a 20% couplingefficiency has been demonstrated experimentally).19 How-ever, this coupling technique would allow for on-wafertesting, which is a prerequisite in the paths toward mass-scale production.

In summary,wehaveshownthat siliconslotwaveguidesfilled with Si-nc/SiO2 material displays a very fast andefficient nonlinear performance at telecom wavelengths.We have presented the proof-of-concept for ultrafast all-optical switching using Si-nc/SiO2 material as a nonlinearmedium. A modulation depth over 50% has been achievedfor on-chip optical powers of the order of 100 mW and atspeeds about 10 times faster than previous approachesusing silicon. The use of a micrometer-size RR results ina very compact structure with a footprint about 500 µm2.From the observed output pulse duration of ∼10 ps, all-optical switching on silicon at speeds well beyond 40Gbit/s appears feasible. In contrast to silicon waveguides,the ultrafast Kerr effect dominates over the slower FC-related effects. The photonic structures have been entirelymade using CMOS processes and equipment, in contrastto other approaches (e.g., those based on nonlinearpolymers that require extra back-end processes). Ourresults could pave the way to the development of all-optical processing circuits on CMOS.

Acknowledgment. The work was financially supported bythe EU through project PHOLOGIC (FP6-IST-NMP-017158).

REFERENCES AND NOTES(1) Liao, L.; Liu, A.; Basak, J.; Nguyen, H.; Paniccia, M.; Rubin, D.;

Chetrit, Y.; Cohen, R.; Izhaky, N. Electron. Lett. 2007, 43, 1196.(2) Dinu, M.; Quochi., F.; Garcia, H. Appl. Phys. Lett. 2003, 82, 2954.(3) Almeida, V. R.; Barrios, C. A.; Panepucci, R. R.; Lipson, M. Nature

2004, 431, 1081.(4) Preble, S. F.; Xu, Q.; Schmidt, B. S.; Lipson, M. Opt. Lett. 2005,

30, 2891.(5) Tanabe, T.; Nishiguchi, K.; Shinya, A.; Kuramochi, E.; Inokawa,

H.; Notomi, M.; Yamada, K.; Tsuchizawa, T.; Watanabe, T.;Fukuda, H.; Shinojima, H.; Itabashi, S. Appl. Phys. Lett. 2007, 90,No. 031115.

(6) Waldow, M.; Plotzing, T.; Gottheil, M.; Forst, M.; Bolten, J.;Wahlbrink, T.; Kurz, H. Opt. Express 2008, 16, 7693.

(7) Preston, K.; Dong, P.; Schmidt, B.; Lipson, M. Appl. Phys. Lett.2008, 92, 151104.

© 2010 American Chemical Society 1510 DOI: 10.1021/nl9041017 | Nano Lett. 2010, 10, 1506-–1511

(8) Spano, R.; Daldosso, N.; Cazzanelli, M.; Ferraioli, L.; Tartara, L.;Yu, J.; Degiorgio, V.; Jordana, E.; Fedeli, J. M.; Pavesi, L. Opt.Express 2009, 17, 3941.

(9) Ammendola, G.; Ancarani, V.; Triolo, V.; Bileci, M.; Corso, D.;Crupi, I.; Perniola, L.; Gerardi, C.; Lombardo, S.; DeSalvo, B. Solid-State Electron. 2004, 48, 1483.

(10) Almeida, V. R.; Xu, Q.; Barrios, C. A.; Lipson, M. Opt. Lett. 2004,29, 1209.

(11) Sanchis, P.; Blasco, J.; Martınez, A.; Martı, J. J. Lightwave Technol.2007, 25, 1298.

(12) Jordana, E.; Fedeli, J. M.; Lyan, P.; Colonna, J. P.; Gautier, P.;Daldosso, N.; Pavesi, L.; Lebour, Y.; Pellegrino, P.; Garrido, B.;Blasco, J.; Cuesta-Soto, F.; Sanchis, P. Deep-UV LithographyFabrication of Slot Waveguides and Sandwiched Waveguides forNonlinear Applications, Proceedings of IEEE International Confer-ence on Group IV Photonics, Tokio, Japan, 19–21 September2007; p 222.

(13) Sheik-Bahae, M.; Said, A. A.; Wei, T. H.; Hagan, D. J.; Van Stryland,E. W. IEEE J. Quantum Electron. 1990, 26, 760.

(14) Barrios, C. A. Electron. Lett. 2004, 40, 862.(15) Guider, R.; Daldosso, N.; Pitanti, A.; Jordana, E.; Fedeli, J.-M.;

Pavesi, L. Opt. Express 2009, 17, 20762.(16) Widmer, A. X.; Franaszek, P. A. IBM J. Res. Dev. 1983, 27, 440.(17) Koos, C.; Vorreau, P.; Vallaitis, T.; Dumon, P.; Bogaerts, W.; Baets,

R.; Esembeson, B.; Biaggio, I.; Michinobu, T.; Diederich, F.;Freude, W.; Leuthold, J. Nature Photon. 2009, 3, 216.

(18) Galan, J. V.; Sanchis, P.; Blasco, J.; Martinez, A.; Marti, J. Opt.Commun. 2008, 281, 5173.

(19) Galan, J. V.; Blasco, J.; Sanchis, P.; Martinez, A.; Marti, J.; Fedeli,J. M.; Jordana, E.; Gautier, P.; Perrin, M. Vertical grating couplersfor silicon sandwiched slot waveguides, Proceedings of IEEE Inter-national Conference on Group IV Photonics, Sorrento, Italy, 17–19 September 2008; p 105.

© 2010 American Chemical Society 1511 DOI: 10.1021/nl9041017 | Nano Lett. 2010, 10, 1506-–1511