Double-Wall Carbon Nanotubes for Wide-Band, Ultrafast Pulse Generation

We demonstrate wide-band ultrafast optical pulse generation at 1, 1.5, and 2 μm using a single-polymer composite saturable absorber based on double-wall carbon nanotubes (DWNTs). The freestanding optical quality polymer composite is prepared from nanotubes dispersed in water with poly(vinyl alcohol) as the host matrix. The composite is then integrated into ytterbium-, erbium-, and thulium-doped fiber laser cavities. Using this single DWNT–polymer composite, we achieve 4.85 ps, 532 fs, and 1.6 ps mode-locked pulses at 1066, 1559, and 1883 nm, respectively, highlighting the potential of DWNTs for wide-band ultrafast photonics.

M aterials with nonlinear optical properties are of critical importance for a diverse range of photonic applications, 1 such as optical regeneration, 2À4 switching, 2,5,6 modulation, 7,8 sampling, 9,10 and noise suppression. 11 In this field, one of the most sought-after applications involves generation of ultrafast laser pulses. 12,13 Indeed, laser sources producing nano-to sub-picosecond optical pulses are a major component in the product portfolio of leading laser manufacturers. 12 Many of the relevant applications, ranging from basic scientific research to materials processing, from eye surgery to printed circuit board manufacturing, from metrology to trimming of electronic components (e.g., resistors and capacitors) currently employ laser sources utilizing a mode-locking technique based on a nonlinear optical material, called saturable absorber (SA). These SAs, when placed in a laser cavity, modify the laser continuouswave output into a train of ultrashort optical pulses. 12 The key requirements for such nonlinear optical materials are fast response time, large nonlinearity, broad wavelength range, low optical loss, high power handling, low cost, and ease of integration into an optical system. 12 Currently, the dominant commercial SA technology is based on semiconductor saturable absorber mirrors (SESAMs). 14 However, these typically have limited operation bandwidths (a few tens of nanometers 12,13 ) and require complex fabrication and packaging. 12 A simpler and cost-effective alternative relies on using single-wall carbon nanotubes (SWNTs) 15À28 or graphene. 15,28À40 While wide-band operation in SWNT-based devices can be achieved using a distribution of tube diameters, 15,21,28,41,42 this is an intrinsic property of graphene, due to the gapless linear dispersion of Dirac electrons. 31,42,43 In general, a good SA should have a high (e.g., ∼10% for fiber lasers 13 ) modulation depth (the absorption change between high and low intensity optical irradiation). 12,42 However, for a pristine single-layer graphene device, the optical absorption is relatively low (∼2.3%), making it unsuitable for fiber lasers where large optical absorption and modulation depth are typically needed. 13 For a given optical absorption, high modulation depth can be typically achieved by minimizing the nonsaturable losses (the optical loss of SAs at high irradiation intensity). 44 For SWNT-based SAs, nanotube bundles and aggregations mostly contribute to these. The most widely employed approach to avoid excessive nonsaturable losses is debundling the SWNTs via solution processing techniques and embedding them into polymer matrices. 42 Indeed, this strategy, coupled with matching the SWNT absorption peak with the operation wavelength, is followed in the majority of the SWNT-based photonic devices, 15,36,42 giving a typical nonsaturable loss of ∼50% of total linear absorption. 15,42 However, when using a wide range of tube diameters to achieve a "wide-band" SWNT SA, the high loading of SWNTs required in the devices results in instability of nanotube dispersion during the composite preparation, leading to aggregation and, therefore, high nonsaturable absorption losses and small modulation depth.
Good SAs should also have a low value of saturation intensity, 45 I sat . This is defined as the optical intensity required to reduce the SA absorption coefficient to half of the initial value, considering zero nonsaturable absorption losses. 44 In graphene, I sat is estimated to be in the range of a few tens 30,46,47 of MW/cm 2 . For SWNTs, I sat is in the range of ∼30 GW/cm 2 in aqueous dispersions. 48 For multiwall nanotubes (MWNTs) with ∼40 nm outer diameter, I sat is >100 GW/cm 2 in aqueous dispersions. 49,50 Therefore, when compared with SWNTs and graphene, MWNTs require higher irradiation intensity to reach absorption saturation. 49,50 This is why, except for only a handful of reports, 51 MWNTs have not been traditionally considered as SAs for passive mode-locking.
The concentric tube arrangement makes DWNTs an interesting class of nanomaterials, with wide-ranging potential applications, including in (opto)electronics. 52À54 Of particular interest relevant to this work, DWNTs also exhibit ultrafast carrier dynamics. 48,55À57 Kamaraju et al. 48 measured a linear limit of saturable absorption (R 0 ∼ 6.1 Â 10 3 and ∼6.4 Â 10 3 cm À1 ) and I sat (∼68 and 14 GW/cm 2 ) for aqueous dispersions and thin films of DWNTs, respectively. This is similar to the values reported for SWNT aqueous dispersions by the same authors (R 0 ∼ 5.6 Â 10 4 cm À1 and I sat ∼ 33 GW/cm 2 , respectively). 48 Thus, in terms of carrier dynamics, DWNTs are comparable to SWNTs.
Further, DWNTs can have outer and inner wall combinations with different electronic types (semiconducting, s, or metallic, m) in their structures (outer-inner: s-s, s-m, m-s, and m-m), resulting in different charge transfer behaviors between the tubes. 58 With a semiconducting outer tube for s-s and s-m combinations, optical absorption from excitonic transition energies of both the inner and outer walls is expected to contribute to the overall optical absorption of the resultant DWNT structure, making them suitable as wide-band SAs. Conversely, with metallic outer wall for the m-s and m-m combinations, the outer wall has zero band gap with wide absorption range and may create a screening effect, 59 suppressing optical absorption from the inner s-or m-nanotubes. Nevertheless, such combination may also work as an advantage for ultrafast photonic applications of DWNTs. This is because the presence of inner or outer m-nanotubes (s-m, m-s) in the same structure can increase the carrier relaxation speed of the s-nanotubes as electrons and holes can tunnel from them to their metallic counterparts. 60,61 Indeed, experimental observations indicate that the relaxation times of inner nanotubes of DWNTs are comparable or shorter 60,62,63 than the SWNTs of same species. For example, Nakamura et al. 57 showed that, under the same experimental conditions, the exciton decay time for (7,6) inner tubes in DWNTs is 0.65 ps, compared to 3.2 ps for a (7,6) SWNT species due to shorter exciton decay time and energy relaxation from inner to outer tubes 57 via exciton energy transfer (EET). 64,65 The strong third-order optical nonlinearity, ultrafast carrier dynamics, 48,55À57,63 and wide optical absorption 57,66 make DWNTs with ∼1.6À1.8 nm outer diameter very attractive for ultrafast photonic applications in the 1 to 2 μm range. With an interwall distance of ∼0.36À0.38 nm, 67,68 the diameter difference between inner and outer tubes is ∼0.7À0.8 nm. Therefore, DWNTs with 0.8À1.1 nm inner diameter have outer tubes with 1.6À1.8 nm diameter with two distinct and strong eh 22 and eh 11 s-tube absorption bands at ∼1.1 and ∼2.0 μm, 69 respectively. The eh 11 from inner stubes at ∼0.8À1.1 μm are expected to be overlapped by eh 22 of the outer tubes. 66,69 This makes DWNTs efficient SAs at ∼1 and ∼2 μm and for larger diameter tubes, potentially beyond this range. This is very attractive for biomedical and biosensing applications where significant demand exists for portable, tunable, pulsed laser sources from ∼2 to up to 10 μm. 70 Similar to other nanoparticles, the aggregation phenomenon of nanotubes, especially in low viscosity dispersions, is largely governed by the diffusion process of nanotubes and nanotubeÀnanotube interactions in a certain medium. 71 This is in addition to the effect of solvent properties (e.g., pH) and stabilization by dispersant (e.g., surfactants). In low viscosity dispersions, aggregation between nanotubes can therefore increase significantly with increased nanotube concentration because of more "exposed" nanotube surfaces. 72 The processing technique we use here involves slow evaporation of solvents from a low viscosity (1.6 mPa 3 s at 25°C) mixture. This highlights the need for a stable dispersion to avoid large aggregation during solvent evaporation. DWNTs allow two ARTICLE nanotubes in a single structure, minimizing the possibility of such large (>1 μm) aggregations and bundle formation during composite fabrication and thus scattering losses while potentially offering high modulation depth at a range of wavelengths. Thus, they are potentially an attractive class of carbon nanomaterial for wide-band ultrafast pulse generation. Here, we demonstrate DWNTÀpolymer composites as wideband passive mode-lockers for ultrafast pulse generation at 1, 1.5, and 2 μm in Yb-, Er-, and Tm-doped fiber laser cavities, respectively.

RESULTS AND DISCUSSION
We use DWNTs produced by catalytic chemical vapor deposition (CCVD) of CH 4 over Mg 1Àx Co x O solid solution containing Mo oxide. 73 After CCVD, the nanotubes are oxidized in air at 570°C for 30 min. 74 The residual material is next washed with HCl to dissolve the metal oxides. 74 Figure 1aÀc shows representative transmission electron microscopy (TEM) images of the DWNT samples at different magnifications. Statistics on ∼130 DWNTs reveal that they have ∼1.1 nm inner and ∼1.8 nm outer mean diameters (Figure 1d,e). TEM images of ∼145 tubes also indicate that the purified samples contain ∼90% DWNTs, ∼8% SWNTs, and ∼2% triple-wall carbon nanotubes (TWNTs).
Optical absorption and photoluminescence excitation (PLE) spectroscopy of this purified nanotube sample dispersed with sodium dodecylbenzenesulfonate (SDBS) surfactant in deuterium oxide (D 2 O) is then used to characterize the DWNT samples. Using D 2 O instead of water allows extension of the spectral study region of the dispersions in the NIR. Pure water begins to absorb at ∼700 nm, followed by a series of strong absorption peaks above ∼900 nm, with complete absorption just above 1100 nm. Using D 2 O instead of H 2 O means OÀH is substituted by OÀD, pushing all these water-related absorption peaks by ∼ √ 2 times toward the higher wavelength, that is, to beyond ∼1300 and ∼1800 nm for the case of the strong peaks and complete absorption, respectively. Figure 2 plots the absorption spectrum of the purified nanotubes dispersed in D 2 O. The peak at ∼1.1 μm corresponds to eh 11 excitonic transitions of 0.75À1.15 nm inner tubes, overlapping with the eh 22 of 1.5À1.9 nm outer tubes. 66,69 The reference spectrum of D 2 O is also presented, highlighting that the eh 22 absorption peaks of the outer wall of DWNTs cannot be resolved above 1800 nm due to strong optical absorption from D 2 O. Figure 3 plots the PLE map of the purified nanotube sample (∼90% DWNTs, ∼8% SWNTs, and ∼2% TWNTs). The chiralities are assigned according to ref 69. The PLE map shows strong emissions from the diameter range of ∼0.7À1.15 nm. This corresponds to that of the inner tubes as derived by TEM (Figure 1e). Tan et al. 64,65 reported ∼2À3 meV red-shift in eh 11 and eh 22 of small SWNT bundles, which formed aggregations over a 2-month period after they were dispersed in waterÀ SDBS solution. Here, we observe a ∼5À7 meV red-shift in eh 11 and eh 22 of all the nanotube species compared to the aforementioned aggregated SWNTs. We attribute such red spectral shifts, even larger than those of the aggregated nanotubes, 64,65 to the dielectric screening 75 of the inner tubes by the outer tubes in DWNTs. 62 Such large red-shift could also be due to bundle formation of the individual s-SWNTs (i.e., not the inner tubes in DWNTs, but a % of the 8% SWNTs)   ARTICLE present in the sample. However, we do not observe significant evidence of bundle formation through strong optical signatures of EET between s-nanotubes of similar diameter. In line with our TEM observation, this indicates that the population of individual s-SWNTs present in the sample is very low. We argue that the (eh 22 ,eh 11 ) emission from small tubes comes from the inner s-nanotubes of DWNTs which, despite bundling of the DWNTs, have weak EET due to larger physical spacing between themselves (>0.7 nm as opposed to ∼0.34 nm in standard SWNT bundles discussed previously 64,65,76 ). This is because the EET process between s-nanotubes in small bundles occurs via Förster resonance energy transfer, whose efficiency is dependent on the inverse sixth power of the physical distance between the donorÀacceptor couple. 77 Note that our observation and explanation support PL from inner tubes as observed in previous reports 62,78,79 but contrast reports that the PL emission from the inner tubes in DWNTs is strongly quenched 80 by the outer tubes by up to 4 orders of magnitude. 81 We only detect very weak or no emissions above the 1375 nm range, from tubes with d t ∼1.2À1.3 nm. Indeed, tubes at this d t range constitute only a very small % of the overall population, as evident from the TEM (Figure 1e,f) and absorption spectra of nanotubes in the dispersion ( Figure 2). As discussed later, the nanotubeÀpolymer composite ( Figure 6) also shows weak absorption (from eh 11 excitonic transitions) in the 1400À 1600 nm range, that is, for d t ∼1.2À1.3 nm. The outer tubes of the DWNTs with a mean d t ∼1.8 nm are expected to emit in the ∼1800À2000 nm range from their eh 11 69,85 it is, in principle, possible to derive the SWNT chirality. 86,87 A variety of C 1 and C 2 values were proposed for this relation. 83,84,87,88 Here, we use C 1 = 228.8 cm À1 and C 2 = 2.4 cm À1 from ref 89, derived by plotting the experimental Pos(RBM) to d t relationship for CVDgrown DWNTs. However, if one is interested in estimation of the band gap, the precise choice of constants is less critical, as the difference in the calculated diameter from the actual value is small. The typical Raman spectrum of nanotubes in the 1500À1600 cm À1 region consists of the G þ and G À bands. In s-SWNTs, they originate from the longitudinal (LO) and tangential (TO) modes, respectively, derived from the splitting of the E 2g phonon of graphene. 90À92 The positions of the G þ and G À peaks, Pos(G þ ), Pos(G À ), are diameter-dependent, and the separation between them increases with decreasing diameter. 91 In m-SWNTs, a wide, low frequency G À is a fingerprint of m-SWNTs. On the other hand, the absence of such features does not necessarily imply that only s-SWNTs are present but could just signify that m-SWNTs are off-resonance.
Thus, a large number of excitation wavelengths are necessary for a complete characterization of nanotubes. 82,88 In particular, we note that ref 93 reported that tubes with up to 100 meV off-resonance from the excitation wavelength can be detected. It is important to note that tubes in resonance with the same laser energy can also have a different diameter. Figure 4a plots the RBM region for the nanotube samples. Note that the RBM detection range is limited by the cutoff of the notch and edge filters at 140, 160, 150, 130, and 150 cm À1 for 457, 488, 514.5, 632.8, and 785 nm, respectively. Thus, we can detect tubes with diameter up to 1.8 nm at 632.8 nm, while we cannot detect tubes with diameter >1.45 nm at 488 nm.

ARTICLE
For each excitation wavelength, we use Lorentzians to fit the RBMs from 20 different measurements to derive the statistics presented in Figure 5aÀe. The RBM spectra do not reveal a cluster distribution of inner and outer peaks around two well-defined diameters. Rather, they show a broad distribution, spanning the entire range from 140 to 400 cm À1 . This heterogeneous distribution was also observed in other Raman characterization of CVD-grown DWNTs. 89,94 The counts in Figure 5aÀe represent how many times the nanotubes of a particular diameter were observed due to resonance with the excitation wavelength. Keeping this in mind, the RBM results are in agreement with optical absorption and PLE measurements. Figure 4b,c plots the Raman spectra in the G and 2D region obtained from the nanotube powders, respectively. Very weak D band contributions are also observed in the G region, indicating a small number of defects. 95,96 The G þ and G À peaks are fitted with Lorentzians. The diameter dependence of G þ and G À peaks can be used to determine the diameter distribution of the nanotubes. 91 This gives an outer tube diameter range of 1.4À1.8 nm. On the other hand, inner tubes have a diameter distribution in the 0.6À1.0 nm range. The estimation for the inner tubes agrees well with the results of RBMs and the G region. For the case of the outer tubes, we cannot compare with the RBM data due to the cutoff of the notch/edge filter. However, the data agree with the Raman analysis of the G band and TEM, where a distribution of ∼1.7 nm is estimated for the outer tubes. The 2D bands in Figure 4c, under different excitations, show a spec-tral profile with multiple Lorentzian peaks, similar to other reports on DWNTs. 97 Therefore, we expect strong absorption bands at ∼1.1 and ∼2 μm from this nanotube sample. This enables us to maximize the change in absorption under strong optical irradiation, 15 making DWNTs ideal SA materials at these wavelengths.
We use the purified nanotubes to prepare nanotubeÀ polymer composites to take the fabrication and integration advantage of polymer photonics into various lightwave systems 15 as well as the optical properties of the constituent ∼90% DWNTs. First, the nanotube sample is ultrasonically dispersed in water using SDBS surfactant, centrifuged to remove the large insoluble particles, mixed with aqueous poly(vinyl alcohol) (PVA) solution, and sonicated again to obtain a homogeneous and stable dispersion free of aggregations. We use water as the solvent and SDBS as the surfactant to obtain higher concentration of isolated nanotubes or small bundles 98 than possible with nonaqueous solvents. 99À101 PVA is used for its solvent compatibility. Slow evaporation of water at room temperature produces a freestanding, ∼50 μm thick nanotubeÀPVA composite (∼90% DWNTs, 8% SWNTs, and 2% TWNTs). Figure 6 shows the optical absorption of the nanotubeÀpolymer composite. This has two absorption bands centered at ∼1.1 and ∼2 μm with a peak width of ∼250 and ∼350 nm, respectively. This corresponds to distinct diameter range d t ∼0.75À1.15 nm (eh 11 at ∼1.1 μm) and ∼1.5À1.9 nm (eh 11 at ∼2 μm and eh 22 at ∼1.1 μm), 69 respectively. This matches the inner and outer diameter distributions of the DWNTs. The 8%

ARTICLE
SWNTs present in the sample have a diameter range similar to that of the inner wall of DWNTs. Thus, eh 11 absorption of these SWNTs also contributes to the strong ∼1.1 μm absorption peak from the eh 11 of inner and eh 22 of the outer walls of DWNTs. The inner walls of the 2% TWNTs have diameter ∼2 nm as observed by TEM. We do not observe any significant eh 11 or eh 22 absorption peaks from the inner walls of the TWNTs. Note that although the eh 22 excitonic transitions of s-SWNTs can extend the operation range of SWNTs to shorter wavelengths, 102 the eh 22 f eh 11 relaxation is over 1 order of magnitude smaller than eh 11 f ground state relaxation. 103 This increases the I sat at eh 22 transition by over 1 order of magnitude compared to that at the eh 11 transition, making the SA device difficult to saturate at the wavelengths corresponding to eh 22 . 102 We argue that the presence of inner walls in DWNTs and strong contribution from their eh 11 transition at 1.1 μm contribution ensures I sat comparable to that of the SWNTs. Similarly, MWNTs, as discussed before, also have high saturation intensity, 49,50 which limits their SA applicability.
Small aggregation of CNTs (<1 μm), in general, is considered beneficial for saturable absorption. Indeed, such level of bundling improves carrier relaxation times by up to an order of magnitude 104 by providing multiple relaxation pathways for the excited carriers. 105 However, these bundles/aggregates are beyond the standard optical microscope resolution limit. Thus, typical optical microscopy is usually used to identify samples with large (>1 μm) aggregates, which are more likely to exhibit large scattering (i.e., nonsaturable) losses. 15 Optical microscopy ( Figure 6, inset) reveals no large nanotube aggregations or defects in the composite, thus avoiding such losses. 106 The polymeric film-based device is integrated by taking a ∼2 mm 2 nanotubeÀpolymer composite, sandwiching it between a fiber pigtailed fiber connector with physical contact (FC/PC) with index matching gel at both the fiber ends. The index matching gel is used to reduce the insertion losses. The SA device is used to demonstrate mode-locking at 1, 1.5, and ∼2.0 μm. The fundamental working principle for the nanotube-based SA devices is explained in refs 15 and 42. Figure 7a shows the setup for 1 μm. A 2 m Yb-doped fiber-based optical amplifier is used to provide gain for lasing. This is a commercial IPG amplifier unit with 25 dB gain at the signal wavelength. A polarization controller is employed for mode-locking optimization. A fused 20/80 coupler is utilized as the output coupler. The 20% port is used for the measurements. All of the cavity fiber has positive group velocity dispersion (GVD) (β 2 ≈ 18 ps 2 km À1 ). A circulator and chirped fiber Bragg grating are inserted inside the ring cavity to set the overall GVD negative, to facilitate soliton-like pulse shaping through the interplay of group velocity dispersion and self-phase modulation. 109 The total cavity length is ∼12.5 m.
Stable mode-locking at 1 μm is achieved. No pulses are observed after removing the composite from the cavity, confirming that mode-locking is initiated by the DWNT composite. Figure 7b plots a typical second harmonic generation autocorrelation trace, which is well fitted by a sech 2 temporal profile. This gives a full width at half-maximum (FWHM) pulse duration of 4.85 ps. Such broad pulse is mainly caused by large overall negative GVD. 109 The output spectral data are shown in Figure 7c. The output spectrum peaks at 1066 nm with FWHM = 0.284 nm. The sidebands are typical of soliton operation; due to intracavity periodical perturbations, 108 the asymmetry can be attributed to the significant third-order dispersion introduced by the chirped fiber Bragg grating. 108,109 The timeÀ bandwidth product (TBP) of the output pulses is 0.36. The deviation from TBP = 0.31 expected for transformlimited sech 2 pulses 109 indicates the presence of minor chirping of the output pulses. 109 The output repetition rate is ∼16.37 MHz. Typical output power is 0.12 mW, with single pulse energy of 7.3 pJ. These results are comparable to those obtained with SWNTs at 1 μm. 23,24 For 1.5 μm operation, we use an erbium-doped fiber (EDF) as the gain medium, pumped by a 980 nm diode laser via a wavelength division multiplexer (WDM). An isolator is employed after the gain fiber to ensure unidirectional operation. A polarization controller is used for mode-locking optimization. The 20% tap of the 20/80 coupler is used as output for measurement. The setup is presented in Figure 8a. The length of EDF is 0.8 m. The gain fiber has a ∼6/125 μm core/cladding geometry and has an absorption of ∼40 dB/m at a wavelength of 1531 nm. The total cavity length is 7.3 m. The output pulse duration is around 532 fs, assuming sech 2 pulse profile; see Figure 8b. The TBP of the output pulses at 1.5 μm is 0.42. The deviation from TBP = 0.31 is due to the minor chirping of the output pulses. In this case, the peak lasing wavelength is 1559 nm, with ARTICLE FWHM = 6.5 nm. Typical soliton sidebands are also observed. The sideband intensity difference is mainly because of higher EDF gain at ∼1545 nm compared to that at ∼1570 nm. 107,108 For ∼2.0 μm operation, we construct a ring laser cavity using a 3.5 m long Tm-doped silica fiber as the gain medium. The gain fiber is a single clad thulium-doped fiber from Nufern. It has a 9 μm/125 μm  ARTICLE core/cladding geometry and has an absorption of ∼10 dB/m at 1560 nm. An amplified 1560 nm diode laser through a WDM pumps the gain fiber. A coupler is used to couple 50% light back into the cavity while guiding 50% of the incident power to the output port of the laser. The mode-locker device is on one end, connected to the wide-band coupler and on the other end, spliced to an isolator to ensure unidirectional light propagation (Figure 9a). A dichroic mirror is used to eliminate the residual pump power at 1560 nm. At ∼350 mW pump power, we achieve an output power of ∼1.2 mW. The autocorrelation trace is shown in Figure 9b, which closely follows a sech 2 shape. The mode-locked optical spectrum in Figure 9c exhibits a center wavelength of 1883 nm with a FWHM spectral width of ∼3.2 nm. Typical sidebands are due to periodic perturbations from cavity components. The corresponding TBP is ∼0.44, which again indicates the presence of chirping. The repetition rate for the output pulse train is 17.82 MHz, as determined by the cavity length of ∼10.8 m.
Radio frequency (rf) spectrum measurements can be used to monitor the mode-locking stability. 110 The typical rf spectra at 1, 1.5, and 2 μm have a peak at an output repetition rate corresponding to the cavity round trip time ( Figure 10). This confirms reliable continuous-wave mode-locking. 111 For the rf spectrum at 1066, 1559, and 1883 nm, the peak to pedestal extinction ratios are ∼60, 86, and 77 dB, respectively (>10 6 contrast). This confirms a stable output pulse. 111

CONCLUSIONS
Our demonstration of wide-band mode-locking optical pulses at 1, 1.5, and 2 μm using DWNTÀpolymer  ARTICLE composites shows that they could potentially be useful for wide-band ultrafast photonic applications. In particular, building on our experiments, few-wall (e.g., <5) nanotubes (FWNTs) with broader absorption profile and I sat lower than large diameter FWNTs may become very attractive for wide-band mid-infrared applications.

EXPERIMENTAL METHODS
Transmission Electron Microscopy. For TEM analysis, dispersions are drop-cast onto a lacey carbon support grid (400 mesh). Water is readily removed by an absorbent underneath, leaving DWNTs on the carbon grid. TEM images are taken using a JEM-3000F FEGTEM at 300 kV.
Raman and Optical Absorption Spectroscopy. Raman spectroscopy measurements are carried out using a Renishaw inVia spectrometer using a 50Â objective. A PerkinElmer Lambda 950 spectrophotometer is used for the visÀNIR absorption measurements with 1 nm step.
Preparation of DWNT Dispersions for Photoluminescence Excitation Spectroscopy. Purified DWNTs (0.01 wt %) are ultrasonicated with 0.2 wt % of SDBS in 10 mL of D 2 O inside a sealed glass test tube in a bath sonicator (Bioruptor; Diagenode) at 270 W, 20 kHz for 4 h. The homogeneous dispersion is then centrifuged in a TH-641 rotor using a sorvall WX 100 ultracentrifuge at ∼200 000g for 2 h and the top 30% decanted for PLE characterization.
Photoluminescence Excitation Spectroscopy. The PLE maps are taken from nanotube dispersions in a HORIBA Jobin Yvon excitationÀemission spectrofluorometer (Fluorolog-3) equipped with a xenon lamp excitation source and a liquid-nitrogen-cooled InGaAs detector (Symphony solo). The PLE maps are measured at right angle scattering by scanning the excitation wavelength from 300 to 900 nm with 5 nm steps and 60 s exposure for ∼850 to ∼1450 nm emission range. The entrance and exit slit widths are 14 and 30 nm, respectively.
Preparation of DWNTÀPVA Composites. Purified DWNTs (0.04 wt %) grown by CCVD method 71 are ultrasonicated in 10 mL of DI water with 1 wt % of SDBS using a tip sonicator (Branson 450A, 20 kHz) at ∼100 W output power and 10°C temperature for 4 h. The homogeneous dispersion is centrifuged at ∼100 000g for 30 min in a TH-641 rotor using a sorvall WX100 ultracentrifuge. The top 60% of the dispersion is then decanted. Then, 4 mL of this dispersion is mixed with a 1 wt % aqueous solution of 120 mg of PVA and is ultrasonicated again for 30 min. The homogeneous mixture has a viscosity of 1.6 mPa 3 s at 25°C, measured using a TA Instruments Discovery HR-1 rheometer. This low viscosity mixture is then drop-cast and the solvent slowly evaporated at room temperature in a desiccator, resulting in a ∼50 μm freestanding DWNTÀPVA composite.
Characterization of Mode-Locked Pulses. The pump and average output power are monitored by a power meter. The pulse duration and spectrum are recorded using a second harmonic generation (SHG) intensity autocorrelator and an optical spectrum analyzer, respectively. To study the operation stability, we measure the rf spectrum using a photodetector connected to a spectrum analyzer.