Strong Plasmonic Enhancement of Photovoltage in Graphene
T. J. Echtermeyer, L. Britnell, P. K. Jasnos, A. Lombardo, R. V. Gorbachev, A. N. Grigorenko, A. K. Geim, A. C. Ferrari, K. S. Novoselov
aa r X i v : . [ c ond - m a t . m t r l - s c i ] J u l Strong Plasmonic Enhancement of Photovoltage in Graphene
T. J. Echtermeyer , L. Britnell , P. K. Jasnos , A. Lombardo , R. V. Gorbachev ,A. N. Grigorenko , A. K. Geim , A. C. Ferrari , and K. S. Novoselov Department of Engineering, University of Cambridge,Cambridge CB3 0FA, UK School of Physics & Astronomy, University of Manchester, Oxford Road, Manchester M13 9PL, UK and Centre for Mesoscience & Nanotechnology, University of Manchester, Oxford Road, Manchester, M13 9PL, UK
Amongst the wide spectrum of potential applications of graphene[1, 2], ranging from transistorsand chemical-sensors to nanoelectromechanical devices and composites, the field of photonics andoptoelectronics is believed to be one of the most promising[3–7]. Indeed, graphene’s suitabilityfor high-speed photodetection was demonstrated in an optical communication link operating at 10Gbit/s[8]. However, the low responsivity of graphene-based photodetectors compared to traditionalIII-V based ones[8] is a potential drawback. Here we show that, by combining graphene withplasmonic nanostructures, the efficiency of graphene-based photodectors can be increased by up to20 times, due to field concentration in the area of a p-n junction. Additionally, wavelength andpolarization selectivity can be achieved employing nanostructures of different geometries.
Graphene-based photodetectors have excellent charac-teristics in terms of quantum efficiency and reaction time,due to the very large room-temperature mobility andhigh Fermi velocity of charge carriers in this material[8–10]. Although the exact mechanism for light to currentconversion is still debated[11, 12], a p-n junction is usu-ally required to separate the photo-generated electron-hole pairs. Such junctions are often created close to thecontacts, due to the difference in work-function of metaland graphene[13, 14]. Whatever the photocurrent gener-ation mechanism, all such devices suffer from the threefollowing problems:(i) low light absorption of graphene(2.3% of normal incident light[15, 16]; (ii) difficulty ofextracting photoelectrons (only a small area of the p-njunction contributes to current generation); (iii) absenceof a photocurrent for the condition of uniform flood illu-mination on both contacts of the device. Unless the con-tacts are made of different materials, the voltage/currentproduced at both contacts will be of opposite polarity forsymmetry reasons, resulting in zero net signal[8, 9, 11].One possible way of overcoming these restrictions is toutilize plasmonic nanostructures placed nearby the con-tacts. Incident light, absorbed by such nanostructures,can be efficiently converted into plasmonic oscillations,which lead to a dramatic enhancement of the local elec-tric field. One might consider this process as generationof evanescent photons that exist only in the near-fieldregion[17–19]. Such field enhancement, exactly in thearea of the p-n junction formed in graphene, can resultin a significant performance improvement of graphene-based photodetectors. The role of the plasmonic nanos-tructures is therefore to guide the incident electromag-netic energy directly to the region of the p-n junction.Here we demonstrate that the efficiency of such devicescan be 20 times larger than traditional ones[8–10].We used graphene prepared by micromechanical ex-foliation of graphite[20, 21]. The single layer natureof our flakes was confirmed by a combination of opti-cal contrast[23–25], Raman spectroscopy[26] and Quan- tum Hall Effect[27, 28] measurements. Ti/Au (3nm Ti,80nm Au) contacts were formed by e-beam lithography,e-beam evaporation and lift-off. Fig.1a shows the layoutof the resulting devices. Various nanostructures were fab-ricated close to one of the macroscopic contacts of such 2-terminal devices (examples are shown in Figs.1b-d). Thelayout and composition of the structures are chosen toproduce strong light absorption in the visible range, andare similar to what we previously designed to achieve aplasmonic blackbody, resulting in almost complete ab-sorption of incident visible light[29]. We employed sev-eral designs, but here we will mainly concentrate on oneparticular structure (grating with 110nm finger width;300nm pitch Fig.1b), giving the best performance.The local photovoltage and photocurrent response ofour devices is measured by coupling several lasers to amicroscope, and scanning the position of the illumina-tion spot. A Nanovoltmeter Keithley 2182A is used torecord the photovoltage at the device terminals with anadditional Keithley 2400 Sourcemeter, allowing controlof the gate voltage. 457, 488, 514, 633 and 785nm lightfrom multi-wavelength Ar + , He-Ne and solid-state in-frared lasers is coupled to the sample via a Leica DMLM microscope and a 100x ultra-long working distanceobjective, with a ∼ µ m spot size. A PI piezoelectricstage translates the sample with respect to the laser spotin the x/y-directions, with 200nm steps, resulting in posi-tion dependent recording of the generated photovoltage.Measurements are done at room-temperature in ambientatmosphere. This allows us to measure the photovolt-age dependence on intensity, wavelength and polarizationof the incoming light, as well as the gate voltage. Thelaser power on the samples is kept ∼ µ W. At this power,the photovoltage signal is larger than any thermopower-related signal (verified by changing the incident power).This laser intensity is also low enough not to give anyobservable overheating of the samples (which ensures wework in the linear regime). Raman spectra are also col-lected, by coupling the light scattered from the sample
FIG. 1: a) SEM micrograph of one of our samples (in artificial colors). Purple: SiO (300nm); bluish: graphene; yellow:Ti/Au electrodes. Scale bar 20 µ m. b-d) Blow-up of contacts with various tested plasmonic nanostructures (in artificial colors).Longitudinal (L) and transverse (TR) incident light polarizations are indicated. Scale bars 1 µ m. to a Renishaw Raman spectrometer.Our devices have field effect mobility ∼ /V sat room temperature (Fig.2a). They show uninten-tional p-doping of up to 5 × cm − (confirmed both byelectronic transport[30] and Raman measurements[31],(Fig.2c,d), probably due to water adsorption[30]. Thecontacts provide local weak p-type doping[14], again con-firmed by the Raman data, Fig.2c,d. The photovoltagegenerated on the non-structured, flat part of the contact(FC), is positive for electron doping, and negative forhole-doping, as a consequence of the formation of p-n orp − -p + junctions, see Fig.2b.The photovoltage generated on the structured part ofthe contacts (SC) is significantly higher than that on theFC. The enhancement is more than one order of magni-tude for the p-n junction (Fig.2b). However, the photo-voltage generated on the SC has remarkably different be- havior than on the FC. It is positive for all the gate volt-ages, monotonically decreasing for higher hole-doping,Fig.2b. We do not have a complete understanding of thisphenomenon, but we speculate that the most probablereason is the complex distribution of the optical electricfield around the SC, allowing us to probe different partsof the p-n or p − -p + junctions (which also have very com-plex shapes due to simultaneous screening and doping bythe metal contacts) in comparison with the FC.The doping profile is confirmed by a Raman line-scanacross the contacts, carried out at zero gate voltage,Fig.2c. Fig.2d plots the ratios of the areas of 2D andG peaks, A(2D)/A(G), and the full width at half maxi-mum of the G peak, FWHM(G). Far away from the con-tacts the Raman parameters correspond to ∼ × cm − p-doping[31, 32] (Fig.2d), consistent with the transportgate-voltage measurements (Fig.2a). A(2D)/A(G) sig- FIG. 2: Resistance, photovoltage and enhancement as function of the gate voltage. a) Resistance as function of the gate voltage.b) Photovoltage for illumination close to the flat part of the contact (FC, blue), close to the structured part of the contact (SC,red) and enhancement (purple) as function of the gate voltage. Illumination wavelength: 514nm. c) Raman spectra recordedon graphene at different distances from SC. d) FWHM(G) and A(2D)/A(G) as a function of position. nificantly increases when moving close to the contacts,accompanied by a FWHM(G) increase. This impliesthe sample becomes less p-doped, with the area aroundcontacts being only lightly p-doped, up to about few10 cm − [31, 32]. In the vicinity of the SC, Fig.2d showsthat both FWHM(G) and A(2D)/A(G) exhibit a non-monotonous behavior, resulting in local maxima. Thiscan be explained by the interplay between the inhomo-geneous doping and strong amplification of the Ramansignal around metallic nanostructures[18].To demonstrate the plasmonic nature of the enhance-ment, we mapped the photovoltaic response for differ- ent polarizations and excitation wavelengths for normallight incidence (Fig. 3). This allowed us to directly com-pare the signal produced when shining light on the FCand SC. Wavelengths covering the visible to near-infraredrange (457, 488, 514, 633, 785nm, corresponding to Figs.3a,b,c,d,e,f, respectively) were used. Fig. 3 shows thatthe SC provides some level of enhancement for all wave-lengths used (the photovoltage on SC is always largerthan that on FC). The generated photovoltage is usu-ally maximum when the laser beam is positioned at thetips of the nanostructures. This is because, in this area,both large electron band bending (due to doping from the FIG. 3: Photovoltage (normalized to laser power) measuredon one of our nanostructured contacts (finger structure; fin-ger width 110nm, pitch 300nm, except for g) as a functionof the position of the illumination spot (spot size ∼ µ m,illumination via microscope x100 objective) for various exci-tation wavelengths. Gate voltage 90V. Scale (except for g):from 0V (blue) to 20V (red). Overlaid is a schematic posi-tion of the contact. a) 457nm, TR polarization. b) 488nm,TR polarization. c) 514nm, TR polarization. d) 633nm, TRpolarization. e) 785nm, TR polarization. f) 514nm, L polar-ization. g) Example of photovoltage measured on a samplewith an array of nanodots. 633nm, TR polarization. Scale:from -4V (blue) to 12V (red). h) Polarization dependent en-hancement at 514nm with 0 ◦ being TR polarization. Blacksquares: measured data; Red line: cos θ fit. FIG. 4: Photocurrent and maximum enhancement coefficientas a function of excitation wavelength for two of our fingerstructures of 300nm pitch. a) finger width 110 nm; b) 130nm.Insets: schematic representations of such structures. contacts[13, 14]) and strong enhancement of the opticalfield[17–20] are achieved. In-between the metal stripes,although the optical field enhancement is still produced,the band bending is significantly smaller due to screeningby the metal contacts.We observed enhancement of the photovoltage for allwavelengths, with maximum amplification of more than20 at the plasmonic resonance of our structure. Indeed,the strong spectral dependence of the photovoltaic en-hancement suggests the importance of the plasmonic res-onances in our nanostructures. The maximum enhance-ment for 110nm wide stripes (Figs.3,4) is observed at514nm (Fig.4a, which is useful for solar cell applica-tions, for instance). Depending on the SC dimensions,the resonance can be tailored to match any part of the FIG. 5: Numerical finite difference time domain simulations for different excitation wavelengths and polarizations. a) 514 nm,TR polarization. b) 514 nm, L polarization. c) 633 nm, TR polarization. d) 633 nm, L polarization. Scale bar 300 nm. spectrum, which might be important for applications intelecommunications. Indeed, for wider structures the res-onance shifts towards larger wavelengths (e.g. the 130nmwide stripes have maximum enhancement close to 633nm,Fig.4b). Such wavelength dependence rules out the pos-sibility that this enhancement is simply due to the ge-ometric enlargement of the junction area for the nanos-tructured contacts. We note that light interference inSiO could provide some dependence of the photovoltageon the wavelength of the excited light, and can be usedto enhance the signal even further[22–25, 33]. However,in our experiments the enhancement coefficient (Fig.4)does not depend on the optical properties of SiO andallows us to concentrate on the on the performance ofsuch plasmonic nanostructures.The SC photovoltage polarization dependence can befitted with a cos θ function, Fig.3h, where θ is the anglebetween the polarization and the long sides of the nanos-tructured ”fingers” (see Fig. 1). The transverse (TR)polarization gives much stronger enhancement than thelongitudinal (L), since the former couples resonantly tothe plasmonic modes across the nanostructured fingers,matching the plasmon wavelength[34]. The FC photo-voltage polarization dependence is much weaker (the dif-ference between TR and L does not exceed 30%). Westress that, even though the far-field polarization prop-erties of the metal stripes also shows cos θ dependence, they cannot generate any enhancement of photovoltagecompared to FC. Hence the observed large anisotropy inenhancement ratio comes from the near-fields generatedby plasmonic nanoresonators.We modeled the enhancement of the electric field withthe help of finite difference time-domain analysis usingthe High Frequency Structure Simulator (HFSS11)[35].The actual device geometry was utilized in the model,and the optical constants of gold, graphene and the sub-strate were taken from Ref.[29]. Fig.5 shows the ampli-tude of the in-plane electric field around the nanostruc-tures for incident light wavelengths of 514nm (Fig.5a,b)and 633nm (Fig.5b,c), and TR and L polarizations. Theresults correlate well with our experimental data, seeFigs.3,4. Thus, the TR polarization for 514nm excitation(Fig 5a) gives very strong field enhancement on 110nmwide structures: a factor 5 in terms of field, which is afactor 25 in terms of power amplification, very similarto what we observe in our experiment, Fig 4. The en-hancement is much weaker for 633nm excitation, againin excellent agreement with our experiments. We note,however, that one cannot draw a direct quantitative com-parison between the calculated field enhancement andthe measured photovoltaic signals. Indeed, the gener-ated photovoltage depends on two factors: 1) the am-plitude of the local optical field and 2) the strength anddirection of the electronic band bending (built-in electricfield due to the p-n junctions). The field amplificationis strongly inhomogeneous, diverging near the contactedges, Fig.5a. This, together with the fact that the p-n junction profile might also be non-trivial, complicatesthe problem. However, the qualitative correspondencebetween the experimental results and the theoretical pre-dictions proves the viability of the concept of using fieldamplification by plasmonic nanostructures for light har-vesting in graphene-based photonic devices.In conclusions, light harvesting aided by plasmonicnanostructures helps to enhance the photovoltage signaland allows operation of such devices under flood illu-mination. Nanostructures with geometries resonant atdesired wavelengths can be utilized in graphene-basedphotodetectors for selective amplification, potentially al-lowing light filtering and detection, as well as polariza-tion determination in a single device at high operatingfrequencies. The frequency performance can be even im-proved in comparison with traditional devices, as the pas-monic structures add only negligible contribution to thecapacitance (fractions of fF), but can significantly reducethe contact resistance. We believe that further optimiza-tion of such plasmonic nanostructures (e.g. making useof coupled or cascaded plasmon resonances[35, 36]) mightlead to even greater photovoltage enhancement. ACKNOWLEDGEMENTS
The authors are grateful to the Royal Society (UK),Engineering and Physical Research Council (UK), Eu-ropean Research Council, EU grants Rodin and MarieCurie ITN-GENIUS (PITN-GA-2010-264694), a SAITGRO program, and Nokia Research Centre Cambridge,for financial support and to Elefterios Lidorikis for usefuldiscussions. [1] A.K. Geim, K.S. Novoselov, Nature Mater. , 183 (2007)[2] A.K. Geim, Science , 1530 (2009)[3] F. Bonaccorso, Z. Sun, T. Hasan, A.C. Ferrari, NaturePhoton. , 611 (2010) [4] P. Blake et. al., Nano Lett. , 1704 (2008)[5] X. Wang, L.J. Zhi, K. Mullen, Nano Lett. , 323 (2008)[6] M. Liu et. al., Nature , 64 (2011)[7] Z. Sun et. al., ACS Nano , 803 (2010)[8] T. Mueller, F.N.A. Xia, P. Avouris, Nature Photon. ,297 (2010)[9] E.J.H. Lee, K. Balasubramanian, R.T. Weitz, M.Burghard, K. Kern, Nature Nanotech. , 486 (2008)[10] F.N. Xia, T. Mueller, Y.M. Lin, A. Valdes-Garcia, P.Avouris, Nature Nanotech. , 839 (2009)[11] J. Park, Y.H. Ahn, C. Ruiz-Vargas, Nano Lett. , 1742(2009)[12] X.D. Xu, N.M. Gabor, J.S. Alden, A.M. van der Zande,P.L. McEuen, Nano Lett. , 562 (2010)[13] G. Giovanetti et. al., Phys. Rev. Lett. , 026803 (2008)[14] P. Blake et. al., Solid State Commun. , 1068 (2009)[15] R.R. Nair et. al., Science , 1308 (2008)[16] A.B. Kuzmenko, E. Van Heumen, F. Carbone, D. vander Marel, Phys. Rev. Lett. , 117401 (2008)[17] S.M. Nie, S.R. Emery, Science , 1102 (1997)[18] F. Schedin et. al., ACS Nano , 5617 (2010)[19] J. Lee, K.S. Novoselov, H.S. Shin, ACS Nano , 608(2011)[20] A.N. Grigorenko, N.W. Roberts, M.R. Dickinson, Y.Zhang, Nature Photon. , 608 (2008)[21] K.S. Novoselov et. al., Science , 666 (2004)[22] K.S. Novoselov et. al., P. Natl. Acad. Sci. , 10451(2005)[23] P. Blake et. al., Appl. Phys. Lett. , 063124 (2007)[24] D.S.L. Abergel, A. Russel, V.I. Fal’ko, Appl. Phys. Lett. , 063125 (2007)[25] C. Casiraghi et. al., Nano Lett. , 2711 (2007)[26] A.C. Ferrari et. al., Phys. Rev. Lett. , 187401 (2006)[27] K.S. Novoselov et. al., Nature , 197 (2005)[28] Y.B. Zhang, Y.W. Tan, H.L. Stormer, P. Kim, Nature , 1201 (2005)[29] V.G. Kravets, F. Schedin, A.N. Grigorenko, Phys. Rev.B. , 205405 (2008)[30] F. Schedin et. al., Nature Mater. , 652 (2007)[31] A. Das et. al., Nature Nanotech. , 210 (2008)[32] D.M. Basko, S. Piscanec, A.C. Ferrari, Phys. Rev. B ,165412 (2009)[33] J. Horng et. al., Phys. Rev. B , 165412 (2011)[34] S.A. Maier, H.A. Atwater, J. Appl. Phys. , 01101(2005)[35] V.G. Kravets, Phys. Rev. Lett. , 246806 (2010)[36] V.G. Kravets, F. Schedin, A.N. Grigorenko, Phys. Rev.Lett.101