REPORTS ated a consistent amount of light in the B cavity for each measurement, avoiding difficulty of inserting light from a se fiber with a constant insertion efficiency Losses ranged from 1 to 8 dB mm-I for wavelengths between 450 and 550 nm, de- 514 the density of scattering centers. These val ues are higher than those reported recently for subwavelength silica waveguides(14) mainly because of substrate-induced radia Wavelength(nm) tion loss and. in some cases. the existence of minor scattering crystal steps and terrac es along the nanoribbon surface. We note however. that the losses here are much etter than what is required for integrate planar photonic over sub-micrometer distance The nanoribbons are of sufficient length and strength to be pushed, bent, and shaped with the use of a commercial micromanip ulator under an optical microscope. Free- lastically curved into loops with radii as Fig 3. Nanoribbon short-pass filters and shape manipulation. (A)Room-temperature PL spectra of small as 5 um, which is remarkable for a ranging from 465 to 580 nm. Cross-sectional dimensions of the 465, 492, 514, 527, and 580 nm On appropriately chosen surfaces, single filters are 310 nm by 100 nm(0.031 um2), 280 nm by 120 nm(0.034 um), 350 nm by 115 nm ribbons are easily fashioned into a variety (0056m2).and elp a single nanoribbon(315 um by 355 nm by 110 nm) as the pump spot was scanned away from the forces to prevent elastic recoil (Fig. 3C) ollection area. The unguided PL curve was obtained at a pump-probe separation of 50 um Larg parations result in a progressive loss of the long wavelengths. C) An SEM image of a simple structive to the ribbon cavities. In practice, hape, demonstrating the high level of by the micromanipulator. This this manipulation method is applicable to shape was created from a single straight ribbon(dimensions of 400 nm by 115 nm)that was cut nanostructures that are free to move and visible and e)o anoribbon aspect ratio- 5200), showing the minimal effect of curvature on waveguiding. (D)a lower size limit, short nanowires (40 nm by 3 captured after an S turn was completed. Blue light is guided around both 1 um radii curves. An SEM um)and even large nanocrystals. Although an image(inset)resolves the bent geometry. The scattering center on the bend is because of a step inherently slow serial process, it is faster and edge rather than physical contact. more versatile than similar approaches using for instance, scanning probes(22)or in situ waveguiding was rare; green, common; and The approximate size of a nanoribbon SEM manipulation (23). We can create net- blue, ubiquitous. For a given dielectric ma- can be inferred from the color of its guided works of nanoribbon waveguides and build terial and cavity geometry and wavelength, PL; large ribbons are white, whereas small functioning optoelectronic components by there exists a critical diameter below which ribbons are blue. When a ribbon of average assembling individual nanowire elements all higher order optical modes are cut off size is pumped nearer to one end, it shines one at a time and waveguiding becomes increasingly dif- blue at the far end and green at the near Manipulation also makes it possible to ficult to sustain. If a nanoribbon is treated end, demonstrating the higher radiation investigate the shape-dependent waveguid as a cylinder of SnO, embedded in air, we losses for longer wavelengths. This effect ing of single nanoribbon cavities For ex- find cutoff diameters for higher order trans- makes nanoribbons excellent short-pass fil- ample, we fashioned a tight s turn in one verse modes of about 270, 220, and 180 nm ters with tunable cutoffs based on path end of a long, thin ribbon(dimensions of for the 652-, 532, and 442-nm light used in length. We have identified ribbon filters 785 um by 275 nm by 150 nm) to illustrate our experiment(20). Although this approx- spanning the 465- to 580-nm region that the robust nature of optical steering in these imation simplifies the cavity shape and ig- feature steep cutoff edges and virtually zero structures(Fig 3, D and E). Although loss- nores substrate coupling and other effects, transmission of blocked wavelengths( Fig. es around the bends could not be measured these values are in reasonable agreement 3, A and B) in this case, they were small enough to only with scanning electron microscope (SEM) We quantified the wavelength-depen- minimally reduce light output from the end measurements of the blue and green dent loss of long, straight ribbons with the of the ribbon. In general, twists and bend waveguide sizes. Most of the ribbons in our use of near-field scanning optical micros- with radii of curvature as small as l um do samples are too thin to propagate red light copy (NSOM). Ribbons were pumped with not disrupt the ability of these subwave over distances greater than 100 um. How- a tightly focused laser beam (3.8 eV) at length waveguides to channel light across ever, sufficiently large ribbons guide wave- different points along their lengths relative hundreds of micrometers lengths across the visible spectrum, acting to a NSOM collection tip held stationary ending a nanoribbon, even slightly, can as subwavelength red-green-blue optical fi- over one of their ends Directly exciting the dramatically change the mode structure of its bers(Fig. 2, D to F) semiconductor waveguide in this way cre- output light (fig. S1). This is most likely .sciencemag. org SCIENCE VOL 305 27 AUGUST 2004 1271waveguiding was rare; green, common; and blue, ubiquitous. For a given dielectric ma￾terial and cavity geometry and wavelength, there exists a critical diameter below which all higher order optical modes are cut off and waveguiding becomes increasingly dif￾ficult to sustain. If a nanoribbon is treated as a cylinder of SnO2 embedded in air, we find cutoff diameters for higher order trans￾verse modes of about 270, 220, and 180 nm for the 652-, 532-, and 442-nm light used in our experiment (20). Although this approx￾imation simplifies the cavity shape and ig￾nores substrate coupling and other effects, these values are in reasonable agreement with scanning electron microscope (SEM) measurements of the blue and green waveguide sizes. Most of the ribbons in our samples are too thin to propagate red light over distances greater than 100 m. How￾ever, sufficiently large ribbons guide wave￾lengths across the visible spectrum, acting as subwavelength red-green-blue optical fi￾bers (Fig. 2, D to F). The approximate size of a nanoribbon can be inferred from the color of its guided PL; large ribbons are white, whereas small ribbons are blue. When a ribbon of average size is pumped nearer to one end, it shines blue at the far end and green at the near end, demonstrating the higher radiation losses for longer wavelengths. This effect makes nanoribbons excellent short-pass fil￾ters with tunable cutoffs based on path length. We have identified ribbon filters spanning the 465- to 580-nm region that feature steep cutoff edges and virtually zero transmission of blocked wavelengths (Fig. 3, A and B). We quantified the wavelength-depen￾dent loss of long, straight ribbons with the use of near-field scanning optical micros￾copy (NSOM). Ribbons were pumped with a tightly focused laser beam (3.8 eV) at different points along their lengths relative to a NSOM collection tip held stationary over one of their ends. Directly exciting the semiconductor waveguide in this way cre￾ated a consistent amount of light in the cavity for each measurement, avoiding the difficulty of inserting light from a second fiber with a constant insertion efficiency. Losses ranged from 1 to 8 dB mm1 for wavelengths between 450 and 550 nm, de￾pending on ribbon cross-sectional area and the density of scattering centers. These val￾ues are higher than those reported recently for subwavelength silica waveguides (14), mainly because of substrate-induced radia￾tion loss and, in some cases, the existence of minor scattering crystal steps and terrac￾es along the nanoribbon surface. We note, however, that the losses here are much better than what is required for integrated planar photonic applications, in which waveguide elements would transmit light over sub-micrometer distances. The nanoribbons are of sufficient length and strength to be pushed, bent, and shaped with the use of a commercial micromanip￾ulator under an optical microscope. Free￾standing ribbons can be repeatedly and elastically curved into loops with radii as small as 5 m, which is remarkable for a crystal that is brittle in its bulk form (21). On appropriately chosen surfaces, single ribbons are easily fashioned into a variety of shapes with the help of ribbon-substrate forces to prevent elastic recoil (Fig. 3C). Careful manipulation is normally nonde￾structive to the ribbon cavities. In practice, this manipulation method is applicable to nanostructures that are free to move and visible with dark-field microscopy, including, at the lower size limit, short nanowires (40 nm by 3 m) and even large nanocrystals. Although an inherently slow serial process, it is faster and more versatile than similar approaches using, for instance, scanning probes (22) or in situ SEM manipulation (23). We can create net￾works of nanoribbon waveguides and build functioning optoelectronic components by assembling individual nanowire elements one at a time. Manipulation also makes it possible to investigate the shape-dependent waveguid￾ing of single nanoribbon cavities. For ex￾ample, we fashioned a tight S turn in one end of a long, thin ribbon (dimensions of 785 m by 275 nm by 150 nm) to illustrate the robust nature of optical steering in these structures (Fig. 3, D and E). Although loss￾es around the bends could not be measured in this case, they were small enough to only minimally reduce light output from the end of the ribbon. In general, twists and bends with radii of curvature as small as 1 m do not disrupt the ability of these subwave￾length waveguides to channel light across hundreds of micrometers. Bending a nanoribbon, even slightly, can dramatically change the mode structure of its output light (fig. S1). This is most likely Fig. 3. Nanoribbon short-pass filters and shape manipulation. (A) Room-temperature PL spectra of five different nanoribbons, each 200 to 400 m long, with 50% intensity cutoff wavelengths ranging from 465 to 580 nm. Cross-sectional dimensions of the 465, 492, 514, 527, and 580 nm filters are 310 nm by 100 nm (0.031 m2 ), 280 nm by 120 nm (0.034 m2 ), 350 nm by 115 nm (0.040 m2 ), 250 nm by 225 nm (0.056 m2 ), and 375by 140 nm (0.052 m2 ), respectively. Spectra are normalized and offset for clarity. (B) A series of normalized emission spectra taken of a single nanoribbon (315 m by 355 nm by 110 nm) as the pump spot was scanned away from the collection area. The unguided PL curve was obtained at a pump-probe separation of 50 m. Larger separations result in a progressive loss of the long wavelengths. (C) An SEM image of a simple shape, demonstrating the high level of positional control afforded by the micromanipulator. This shape was created from a single straight ribbon (dimensions of 400 nm by 115nm) that was cut into two pieces and then assembled. (D and E) Optical images of the emission end of a long nanoribbon (aspect ratio  5200), showing the minimal effect of curvature on waveguiding. (D) A true-color photograph taken after crafting a single bend. (E) A black-and-white dark-field PL image captured after an S turn was completed. Blue light is guided around both 1 m radii curves. An SEM image (inset) resolves the bent geometry. The scattering center on the bend is because of a step edge rather than physical contact. R EPORTS www.sciencemag.org SCIENCE VOL 30527 AUGUST 2004 1271