JOURNAL OF RAMAN SPECTROSCOPYJ. Raman Spectrosc. 2003; 34: 663–667Published online in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/jrs.1046

Tip-enhanced Raman microscopy: practicalities andlimitations

D. Richards,∗ R. G. Milner, F. Huang and F. Festy

Department of Physics, King’s College London, Strand, London WC2R 2LS, UK

Received 1 December 2002; Accepted 3 February 2003

The feasibility of apertureless scanning near-field Raman microscopy, exploiting the local enhancementin Raman scattering in the vicinity of a silver or gold tip, was investigated. Using the finite difference timedomain method we calculated the enhancement of electric field strength, and hence Raman scattering,achieved through the resonant excitation of local modes in the tip. By modelling the frequency-dependentdielectric response of the metal tip we were able to highlight the resonant nature of the tip- enhancementand determine the excitation wavelength required for the strongest electric field enhancement, and henceRaman scattering intensity, which occurs for the excitation of modes localized at the tip apex. It isdemonstrated that a peak Raman enhancement of 107-fold should be achievable with <5 nm spatialresolution. We show that surface-enhanced Raman scattering from carbon contamination on a silver orgold tip can be significant. However, we find for a tip of radius of curvature 20 nm that the Ramanenhancement should decay totally within 20 nm from the tip. Hence withdrawal of the tip by this distanceshould lead to the disappearance of the tip-enhanced signal, leaving only that from carbon contaminationon the tip itself and the intrinsic signal from the sample. Copyright 2003 John Wiley & Sons, Ltd.

KEYWORDS: surface-enhanced Raman scattering; SERS; near-field optical microscopy; SNOM; NSOM; surface plasmons

INTRODUCTION

Surface-enhanced Raman scattering (SERS), the giantenhancement in Raman scattering from molecules in closeproximity to small particles or rough surfaces of noble metalssuch as silver or gold, has been put to remarkable effect inrecent years for the investigation of single molecules and car-bon nanotubes. In these recent experiments, which employedsilver nanoparticles and fractal colloidal clusters as the SERS-active media, enhancements of up to 1014-fold in the Ramansignal have been claimed.1 – 3 A metal tip in close proximityto the sample surface should in principle give rise to similareffects4 —tip-enhanced Raman scattering (TERS)—holdinggreat promise for scanning probe Raman microscopy with¾10 nm spatial resolution. Indeed, experiments of this kindhave recently been claimed to provide an initial demon-stration of such a tip-enhanced Raman scanning near-fieldoptical microscopy (SNOM).5,6

The dominant contribution towards the giant enhance-ments in Raman cross-section observed for SERS results is

ŁCorrespondence to: D. Richards, Department of Physics, King’sCollege London, Strand, London WC2R 2LS, UK.E-mail: david.r.richards@kcl.ac.ukContract/grant sponsor: EPSRC.Contract/grant sponsor: Royal Society.

thought to arise from large enhancements in electric fieldstrength as the external electromagnetic field couples to sur-face plasmons of metal particles with dimensions less thanthe wavelength of light, and strong fields associated withthe plasmon resonance build up at the particle surface. Inthe same way, TERS is expected to result from the reso-nant excitation of localized surface plasmons on a metal tip.7

To understand fully the potential and limitations of TERSrequires three-dimensional modelling of a realistic metal tipheld in close proximity to a dielectric substrate—the geom-etry pertinent to experiments of this kind—and taking intoaccount the frequency-dependent response of the tip. Therehave been a number of theoretical studies of scattering byfree metal spheres and, more recently, of irregular metalnanowires.8 Other work has considered the scattering ofmetal tips with a fixed (i.e. frequency independent) dielectricconstant.9 However, it is important to employ a frequency-dependent dielectric response for the metal tip, to account forthe frequency dependence, with geometric shape, of the elec-tromagnetic resonances in the tip, well known for metallicnanoparticles.8,10

Using the finite difference time domain (FDTD) method,we have recently performed such a theoretical investigationof the scattering of light by a Drude metal tip, investigatingthe effects of the size of the tip, the tip–sample separation, the

Copyright 2003 John Wiley & Sons, Ltd.

664 D. Richards et al.

frequency and polarization of the incident light and the angleof incidence.11 By taking into account both the frequencydependence of the dielectric response of the metal tip and thefull scattering geometry, we were able to identify the plasmamodes providing large field enhancement for differentgeometries, separating the bulk and localized excitations inthe tip, highlighting the importance of the interplay betweenthe geometry and frequency response of the scatteringsystem. In particular, we found that the electromagneticmodes of a metal tip most suited for exploitation in TERS arelocalized modes at the apex of the tip, potentially producinglarge enhancements with resolutions below the radius ofcurvature of the tip. Similar FDTD calculations have alsobeen performed more recently by Krug et al.12

In this paper, from such calculations of the electromag-netic response of metallic tips, we provide a quantitativeassessment of the feasibility of a scanning near-field Ramanmicroscopy exploiting the SERS effect. In particular, we con-sider the importance of the resonant excitation of local fieldson the tip. We also highlight the problem associated withcarbon contamination of tips and indicate, from the resultsof our theoretical work, how this may be overcome.

THEORY: THE FINITE DIFFERENCE TIMEDOMAIN METHOD

We have calculated the scattering of electromagnetic radia-tion from a metal tip using the FDTD method.13,14 In thesecalculations, we have described the tip as a metal cone ofsemi-angle 30° with a spherical apex of radius 20 nm, asindicated in Fig. 1. We used a Drude frequency-dependentdielectric response for the metal tip:

ε�ω� D 1 � ω2p

ω�ω C i���1�

Figure 1. Cross-section through the tip and sample, in theplane of incidence and along the tip axis, illustratingschematically the scattering geometry for the present FDTDcalculations. The tip is modelled as a cone with spherical apexof radius r D 20 nm, held at distance d D 2 or 20 nm from thesurface of a glass substrate. The incident Gaussian derivativepulse propagates at angle of incidence � normal to the air/glassboundary, with the polarization in the plane of incidence.

taking a plasma energy ωp D 3.8 eV, close to that of the bulkplasmon of silver. For the purposes of the FDTD, Eqn (1) forthe Drude response can be re-cast in the form of a Debyedielectric function.13 In fact, to model accurately silver orgold would require the addition of Lorentz poles in themetallic dielectric response function. However, this wouldlead to even greater demands on computer memory andtime for these already computationally intensive calculationsand so the simpler Drude form was employed as beingrepresentative of the frequency response of these metals. Inparticular, a model incorporating the Drude response allowsthe description of plasma resonances, of central importanceto SERS and, hence, TERS.

The FDTD method looks at the time evolution of theelectromagnetic fields in a system subjected to a plane-waveincident field or pulse. We note that when a system issubjected to a wave of fixed frequency, ω, the FDTD methodis not stable when the dielectric constant goes negative, thatis, when ω < ωp, yet this is precisely the frequency rangeover which nanoparticle systems exhibit plasma resonances.In such cases, a full frequency-dependent calculation needsto be performed to ensure stability. Correspondingly, in ourcalculations, the system was illuminated with a Gaussianderivative pulse, to provide a probe of the electromagneticresponse at all frequencies;11,13 the fields are stored at alltime steps as the pulse passes through the problem spaceand the scattered fields decay, with a Fourier transformthen determined at each point to provide the frequency-dependent response of the system. The spectral range of theGaussian derivative pulse was chosen to provide a goodoverlap with the spectral range of interest.

Calculations were performed for tip–sample separationsof 2 and 20 nm, with the substrate taken to be glass, with arefractive index of 1.5. The tip was illuminated from belowwith a range of incident angles � to the normal of theair/glass interface, from � D 0 to 40°, corresponding to therange of incident angles present in illumination using a lensof numerical aperture NA D 0.95. The electric field waspolarized in the plane of incidence. Thus, for illuminationincident at � D 40°, which is close to the critical angle, thepulse is incident on the tip in free space with the directionof propagation at just 15° to the air/glass interface, with themain component of the electric field along the tip axis.

The problem space was discretized into 60 ð 60 ð 60 Yeecells with a 2 nm cell size. Cells on the metal/air interfacewere weighted so as to describe a ‘fuzzy’ boundary andsecond-order Mur boundary conditions were used to mini-mize reflections from the boundaries.13 However, as the Murboundary conditions do not take into account frequency-dependent materials, a dielectric layer of cells was insertedbetween the metal and the boundary, with a dielectric con-stant chosen to minimize reflections at this boundary overthe frequency range of interest. The calculations were per-formed with time steps 10 times smaller than the Courantstability criterion and the simulations were run for 20 fs,

Copyright 2003 John Wiley & Sons, Ltd. J. Raman Spectrosc. 2003; 34: 663–667

Tip-enhanced Raman microscopy 665

giving a frequency resolution of 0.21 eV. Following the sim-ulation, the Fourier transform of the temporal evolution ofthe total fields was taken for each cell in the computationalvolume and normalized to the spectrum of the incidentpulse, to provide a description of the frequency-dependentresponse of the system. The Drude damping parameter �in Eqn (1) was taken to be 0.4 eV to ensure that the spectralresolution of the calculation was sufficient for the calculatedspectral response; in fact, a realistic value of � would be lessthan this for silver, leading to sharper and larger plasmaresonances.

Once the electric field distribution has been determinedfor a given excitation photon energy, the local enhancementin Raman scattering for a molecule at any given positionis determined to be the fourth power of the electric fieldenhancement:

ITERS

I0D

∣∣∣∣

Elocal

E0

∣∣∣∣

4

�2�

where I0 and E0 are the Raman cross-section and electricfield strength in the absence of the tip and ITERS is the Ramancross-section at a point for which the presence of the tiphas enhanced the electric field to Elocal. This approximationexploits the fact that the tip acts as an antenna for boththe incident and scattered light. A more detailed analysiswould require an ab initio evaluation the of Raman intensitiesfor a molecule in the vicinity of the tip, as performed byCorni and Tomasi for metal particle aggregates.15 The fourth-power dependence of the Raman signal on the electric fieldenhancement means that even modest enhancement factorslead to huge increases in the Raman cross-section.

THE IMPORTANCE OF RESONANCE

The excitation of localized plasmons in silver and goldnanostructures is a resonant phenomenon. As such, it isimportant to match the incident photon energy to that of theresonance to obtain the maximum possible tip enhancement.Theoretical and near-field investigations of fractal silver

colloidal clusters have demonstrated the presence of a veryinhomogeneous field distribution over the silver clusters,including extremely large electromagnetic fields in ‘hotspots,’16,17 providing the key effect for extremely largeSERS enhancement factors. So, for SERS using colloidalaggregates, a resonance will occur at some localized pointin the aggregate for any given excitation wavelength. Incontrast, tip-enhanced Raman scattering requires that theexcitation is tuned to match the tip resonance, or that the tipis designed for a given wavelength, for strong enhancementto occur precisely at the tip apex.

Indeed, from our calculations we have found that a strongand highly localized electric field enhancement beneath thetip occurred on resonance with a plasma excitation localizedat the tip. Figure 2 shows the electric field distributionfor such a resonance, which occurred at a photon energyof 2.1 eV (excitation wavelength 590 nm), calculated for atip-sample separation of 2 nm and an angle of incidence� D 40°. Indeed, this resonance frequency is very closeto ωp/

p3, the frequency of the Mie plasmon on a small

sphere.10 The electric field enhancement in the plane of thesample [Fig. 2(a)] and in the plane of polarization alongthe tip axis [Fig. 2(b)] are shown. From Fig. 2(a), we cansee that there is a maximum electric field enhancement of53-fold in the plane of the sample (for a 2 nm tip–sampleseparation), corresponding to a local Raman enhancementof 8 ð 106. The enhancement is strongly dependent on theangle of illumination � or, rather, the presence of a strongcomponent of polarisation along the tip axis. We havefound that as � is reduced from the critical angle, theenhancement reduces rapidly, going to zero for � D 0°.Note also that the largest enhancements observed withSERS from colloidal aggregates probably occur for moleculeslocated between silver nanoparticles, i.e. in the small gapbetween two highly curved surfaces, where the electric fieldenhancements should be greatest. For TERS there is onlyone metal surface and so the enhancements are likely to besignificantly lower.

Figure 2. Enhancement in the magnitude of the electric field, for a tip–sample separation d D 2 nm, excited at photon energy2.1 eV (wavelength 590 nm), with illumination at � D 40° to the normal. The Raman enhancement at any given point is proportional tothe fourth power of the electric field enhancement, jElocal/E0j. (a) Enhancement in the plane of the sample. (b) Cross-section throughthe tip, along the tip axis (cf. Fig. 1); a localized mode is strongly confined at the tip apex.

Copyright 2003 John Wiley & Sons, Ltd. J. Raman Spectrosc. 2003; 34: 663–667

666 D. Richards et al.

The surrounding dielectric environment also influencesthe frequency for resonant tip enhancement. For calculationsperformed with a tip–sample separation of 20 nm (still with a20 nm radius tip), we found that the excitation photon energyfor maximum field enhancement at the tip apex shifted to1.9 eV (excitation wavelength 650 nm). The slightly lowerpeak enhancement observed for this configuration couldresult from not being precisely on resonance, as the spectralresolution of our calculations is only 0.2 eV.

The local enhancement could in fact be even greater thanthe predicted value of ¾107-fold, as we have included anartificially high damping energy � in our calculations toensure that any resonance was detected with the spectralresolution of our calculations. It is possible that a completedescription of the dielectric response of silver, further tothe Drude model employed here, could account for greaterenhancement in the local field strength. The full width athalf-maximum of the field enhancement of ¾10 nm leadsto an expected spatial resolution for Raman microscopy of5 nm.

We note that in the majority of experimental workpublished in this field 488 or 514 nm excitation wasused. However, our calculations indicate that such anexcitation wavelength is unlikely to give a strong resonantenhancement for a small silver tip (although the resonantfrequency does shift from ωp/

p3 with increasing radius

of curvature or for departures from a simple sphericalgeometry). We suggest that the correct choice of wavelengthis essential if this technique is to become practicallyimplementable.

DEPENDENCE ON TIP–SAMPLE SEPARATION

The high spatial resolution promised by TERS results fromthe highly localized tip plasmon resonance. As such, thedecay of the electric field enhancement away from the tip, andhence the decay in Raman enhancement with increasing tip-sample separation, are rapid. Figure 3 shows the calculatedenhancement in Raman intensity as a function of distancefrom the tip, along the tip axis, for resonant excitation of atip plasmon.

The calculation was performed for a tip–sample sep-aration of 20 nm and an angle of incidence � D 40°, andthe enhancements presented in Fig. 3 correspond to a reso-nant excitation photon energy of 1.9 eV. We can see that theenhancement is expected to have disappeared for a moleculelocated 20 nm from the tip apex (i.e. of the order of the radiusof curvature of the tip).

Therefore, like all near-field optical microscopies, theintensity at any point is very dependent on the tip–sampleseparation, with the consequent danger of topographicartifacts in the optical signal. On the other hand, thedemonstration of the rapid decay of signal strength withtip–sample separation will serve as a strong verification oftip enhancement.

1

10

100

1000

104

105

106

0 5 10 15 20

Ram

an E

nhan

cem

ent

Distance from Tip (nm)

Figure 3. Predicted Raman enhancement as a function ofdistance z away from the tip apex, along the axis of the tip. TheFDTD calculation was performed for resonant excitation atphoton energy 1.9 eV (wavelength 650 nm), with illumination at� D 40° to the normal and the glass substrate a distanced D 20 nm from the tip apex.

THE ‘TIP’ SIGNAL — THE PROBLEM OFCARBON CONTAMINATION

The majority of published reports on TERS so far haveprovided only an indication of the signal strength for a silveror gold tip in close proximity to a sample surface (eithertouching, as in the case of an AFM cantilever tip, or within afew nanometres for a tip held close to the sample using shear-force feedback), or removed far from the surface (usually afew microns) such that the tip is no longer within the focusof illumination, while the sample remains in focus. Thisdoes not, however, necessarily provide confirmation that thesignal observed derives from the sample and not from thetip itself.

Figure 4 shows a series of Raman spectra from silverand gold tips prepared by evaporating metal on to etchedglass-fibre tips. The spectrum for the silver tip was measuredwith 488 nm excitation and that for the gold tip with 647 nmexcitation; however, similar spectra are observed for a widerange of excitation wavelengths. The spectra change withtime and with position; the second spectrum in Fig. 4(a) wasmeasured from a thin film of silver evaporated on silicon,providing another example of typical spectra measured.We also observe similar spectra from scattering features onfreshly cut 99.9999% silver (Alfa Aesar Johnson Matthey).

The spectra observed are typical of those from carboncontamination on silver and gold surfaces and illustrate thestrength of the SERS effect. This rich spectrum is a well-known limitation to the use of SERS as an analytical tool,18 – 20

particularly for the study of any carbon-based material.Nevertheless, for SERS the molecules under investigationare usually absorbed on the silver or gold surface or colloid,giving very large enhancements, and the coverage of themolecules on the silver surface is significant compared

Copyright 2003 John Wiley & Sons, Ltd. J. Raman Spectrosc. 2003; 34: 663–667

Tip-enhanced Raman microscopy 667

500 1000 1500 2000

Wavenumber / cm-1

Inte

nsity

(a.

u.)

(a)

500 1000 1500 2000

Wavenumber / cm-1

Inte

nsity

(a.

u.)

(b)

tip

film

Figure 4. Raman spectra from carbon contamination on(a) silver evaporated on to an etched glass-fibre tip and a thinfilm evaporated on to silicon, both excited at 488 nm, and(b) gold evaporated on to glass-fibre tips, exited at 647 nm.The form of the spectra changes with time and from point topoint on the evaporated metal surface. Similar spectra areobserved from the surface of freshly cut 99.9999% silver.

with that of the carbon contamination. However, in thecase of TERS, a large area of the metal tip is usuallyilluminated (¾1 µm) while the goal is the enhancement ofRaman scattering from only a very small area of the sample(<10 nm); correspondingly, the background carbon signalfrom the tip can become significant. Therefore, it is obviouslyimportant that any implementation of tip-enhanced RamanSNOM should ensure that only signals deriving from thesample are measured—performed most easily by takingadvantage of the disappearance of the TERS signal onremoval of the tip by just ¾20 nm, over which distancethe background tip signal should remain unchanged.

CONCLUSIONS

From calculations of the electromagnetic field distributions inthe vicinity of a Drude metal tip, calculated using the FDTDmethod, we have demonstrated that tip-enhanced scanningnear-field Raman microscopy should be possible with <5 nmspatial resolution for a 20 nm radius tip. However, ourcalculations have demonstrated the importance of resonantexcitation of local fields on the tip apex for the realizationof large and highly localized Raman enhancements. Thebackground signal from carbon contamination on a tip couldalso serve to complicate the implementation of tip-enhancedRaman microscopy, although our calculations indicate thatthe rapid decay of the Raman enhancement away fromthe tip should allow the separation of any background tip-contamination signal from the tip-enhanced signal from thesample.

REFERENCES1. Nie S, Emory SR. Science 1997; 275: 1102.2. Kneipp K, Kneipp H, Corio P, Brown SDM, Shafer K,

Motz J, Perelman LT, Hanlon EB, Marucci A, Dresselhaus G,Dresselhaus MS. Phys. Rev. Lett. 2000; 84: 3470.

3. Kneipp K, Kneipp H, Itzkan I, Dasari RR, Feld MS. J. Phys.:Condens. Matter 2002; 14: R597.

4. Wessel J. J. Opt. Soc. Am. B 1985; 2: 1538.5. Stockle RM, Suh YD, Deckert V, Zenobi R. Chem. Phys. Lett. 2000;

318: 131.6. Anderson MS. Appl. Phys. Lett. 2000; 76: 3130.7. Fischer UC, Pohl DW. Phys. Rev. Lett. 1989; 62: 458.8. Kottmann JP, Martin OJF, Smith DR, Schultz S. Chem. Phys. Lett.

2001; 341: 1.9. Martin YC, Hamann HF, Wickramasinghe HK. J. Appl. Phys.

2001; 89: 5774.10. Bohren CF, Huffman DR. Absorption and Scattering of Light by

Small Articles. Wiley: New York, 1983.11. Milner RG, Richards D. J. Microsc. 2001; 202: 66.12. Krug JT, Sanchez EJ, Xie XS. J. Chem. Phys. 2002; 116: 10 895.13. Kunz KS, Luebbers J. The Finite Difference Time Domain Method

for Electromagnetics. CRC Press: Boca Raton, FL, 1993.14. Taflove A. Computational Electrodynamics: the Finite-Difference

Time-Domain Method. Artech House: Boston, 1995.15. Corni S, Tomasi J. J. Chem. Phys. 2002; 116: 1156.16. Markel VA, Shalaev VM, Zhang P, Huyn W, Tay L, Haslett TL,

Moskovits M. Phys. Rev. B 1999; 59: 10 903.17. Gadenne P, Quelin X, Ducourtieux S, Gresillon S, Aigouy L,

Rivoal J-C, Shalaev V, Sarychev A. Physica B 2000; 279: 52.18. Tsang JC, Demuth JE, Sanda PN, Kirtley JR. Chem. Phys. Lett.

1980; 76: 44.19. Taylor CE, Garvey SD, Pemberton JE. Anal. Chem. 1996; 68:

2401.20. Norrod KL, Rowlen KL. Anal. Chem. 1998; 70: 4218.

Copyright 2003 John Wiley & Sons, Ltd. J. Raman Spectrosc. 2003; 34: 663–667