LETTERdoi:10.1038/nature09664

Atom-by-atom spectroscopy at graphene edgeKazu Suenaga1 & Masanori Koshino1

The properties of many nanoscale devices are sensitive to localatomic configurations, and so elemental identification and elec-tronic state analysis at the scale of individual atoms is becomingincreasingly important. For example, graphene is regarded as apromising candidate for future devices, and the electronic propertiesof nanodevices constructed from this material are in large partgoverned by the edge structures1. The atomic configurations at gra-phene boundaries have been investigated by transmission electronmicroscopy and scanning tunnelling microscopy2–4, but the elec-tronic properties of these edge states have not yet been determinedwith atomic resolution. Whereas simple elemental analysis at thelevel of single atoms can now be achieved by means of annular darkfield imaging5 or electron energy-loss spectroscopy6,7, obtainingfine-structure spectroscopic information about individual lightatoms such as those of carbon has been hampered by a combinationof extremely weak signals and specimen damage by the electronbeam. Here we overcome these difficulties to demonstrate site-specific single-atom spectroscopy at a graphene boundary, enablingdirect investigation of the electronic and bonding structures of theedge atoms—in particular, discrimination of single-, double- andtriple-coordinated carbon atoms is achieved with atomic resolution.By demonstrating how rich chemical information can be obtainedfrom single atoms through energy-loss near-edge fine-structure ana-lysis8, our results should open the way to exploring the local elec-tronic structures of various nanodevices and individual molecules.

A low-voltage scanning transmission electron microscope (STEM)was used for the single-atom spectroscopy9. Flakes were cleaved fromthe synthetic highly oriented pyrolytic graphite (HOPG) and put ontothe microgrids for energy-loss near-edge fine structure (ELNES)

analysis. STEM annular dark field (ADF) images indicate that the gra-phene flakes have open and active edges3 and that the edges are steadilyetched by the incident electron beam when the probe-scanning isrepeated at the same region (Supplementary Fig. 1). The acceleratingvoltage used here (60 kV) is below the critical energy predicted forsevere knock-on damage10 and therefore the carbon atoms in bulk aremostly stable. Only the edge atoms are mobile during the observation,as indicated by the wiggling contrast frequently observed at the edgeregions. The fast Fourier transformation of an ADF image of few-layergraphene shows that the spatial resolution of the experimental set-up isbetter than 0.106 nm (inset to Supplementary Fig. 1a) and so the hexa-gonal network of carbon atoms, separated by about 0.14 nm, is clearlyvisible in a monolayer region (Supplementary Fig. 1b). A probe of thesame size and brightness was used for the following ELNES analysis.

Figure 1a shows a typical ADF image of the edge region of a singlegraphene layer. The hexagonal network of carbon atoms in bulk is visibleon the right-hand side of the image and the vacuum region appears inblack on the left-hand side. The possible carbon atom positions derivedfrom the local intensity maxima of ADF signals are marked by yellowcircles after an image-smoothing process in Fig. 1b. There is strong wigglecontrast at the edge regions and some of the atom positions cannot becompletely identified. We note that some of the hexagonal networks areimperfect and considerably reconstructed at the edge region.

The typical ELNES spectra of carbon K (1s)-edge are displayed withtheir corresponding atomic positions in Fig. 1c. Figure 1d shows threecharacteristic carbon K-edge fine structures extracted using sequentialelectron energy-loss spectroscopy (EELS) with probe-scanning (knownas the spectrum-image mode)11. The spectrum in green was recorded atan atomic position in bulk (indicated by a green circle and arrow in

1Nanotube Research Center, National Institute of Advanced Industrial Science and Technology (AIST), AIST Central 5, Tsukuba 305-8565, Japan.

CC

D c

ount

s (a

.u.)

D

280 290 300Energy loss (eV)

310

a c

d

b

S π*

Figure 1 | Graphene edge spectroscopy. a, ADFimage of single graphene layer at the edge region.No image-processing has been done. Atomicpositions are marked by circles in a smoothedimage (b). Scale bars, 0.5 nm. d, ELNES of carbon K(1s) spectra taken at the colour-coded atomsindicated in b. Green, blue and red spectracorrespond to the normal sp2 carbon atom, adouble-coordinated atom and a single-coordinatedatom, respectively. These different states of atomiccoordination are marked by coloured arrows ina and b and illustrated in c. CCD, charge-coupleddevice.

1 0 8 8 | N A T U R E | V O L 4 6 8 | 2 3 / 3 0 D E C E M B E R 2 0 1 0

Macmillan Publishers Limited. All rights reserved©2010

Fig. 1b) as a reference. This spectrum exhibits the features of typical sp2

coordinated carbon atoms, such as the sharpp*peak around 286 eV andthe exciton peak of s* at 292 eV. These features are in good agreementwith the previously reported spectra recorded from a bulk graphitespecimen12. The spectrum in blue was recorded from an edge atomlocated at the border of the hexagonal network with two-coordination,as illustrated in Fig. 1c. Remarkably, this spectrum has a extra peakaround 282.6 6 0.2 eV (labelled D in Fig. 1d), with the p* peak havingreduced intensity. Also the exciton peak intensity is considerably reducedand broadened compared to the bulk spectrum (marked by open circles).

The spectrum in red shows similar features, also with weaker p*peak and broadeneds* peak. Its extra peak occurs at a different energyposition of 283.6 6 0.2 eV (labelled S in Fig. 1d). It is extremely difficultto assign the atomic position completely for this red spectrum becausethe spectrum disappears quickly and is not fully reproducible. Theedge region of the specimen tends to be strongly damaged and theedge morphology frequently changes after recording the spectrumimage. Therefore, we can reasonably infer that this energy state mustbe somehow damage-related. One of the possible models for this edgestructure is the Klein edge13,14. The edge atom indicated in red in Fig. 1bis indeed single-bonded to its neighbour. The structure should be veryunstable under the incident electron beam and so it may also explainthe wiggling contrast often observed at the graphene edge.

These spectral features involving peaks D and S have not previouslybeen reported, to our knowledge. No fingerprinting method, compar-ing against the reference spectra of the existing polymorphic carbon, isable to explain them. We therefore performed ELNES simulations tocorrelate the experimental features with different atomic configura-tions (Fig. 2). The p* peak shift to the lower energy is well reproducedfor the edge atoms in the Klein, zigzag and armchair edge configura-tions (Fig. 2a, b and c), in comparison with the bulk carbon atom(Fig. 2d). The diminished excitonic effect can be confirmed for theKlein edge (Fig. 2a). The peak shift around 2 eV is well reproduced forthe zigzag edge (Fig. 2b). In the spectrum of the armchair edge a sharppeak between p* and s* is expected (Fig. 2c).

To show an atom-by-atom spectroscopy, we also performed EELS inthe spectrum-line mode across a graphene edge. The probe scannedacross the protruded carbon atom—the Klein edge—from the vacuumto the bulk region along the dotted line in Fig. 3a. A series of 100 spectrawere sequentially recorded by scanning the electron probe with a con-stant step of about 0.02 nm. The total acquisition time was as small as50 s. The illustrated model in Fig. 3b shows that eight carbon atoms wereinvestigated in the spectrum line. Figure 3c shows a profile of ADFsignals (in red) that was simultaneously recorded with the ELNES spec-tra. It shows good agreement with the simulated profile (in blue) show-ing eight maxima sequentially corresponding to the eight carbon atoms.Although the experimental profile is rather scattered owing to specimeninstability or a possible inclination of the specimen to the incidentelectron beam, which should produce a slight asymmetry in the profileof the carbon doublets, we can deduce the carbon atomic positionsreasonably well from the line profile and extract the ELNES spectracorresponding to each atom. Figure 3d shows ELNES fine structuresobtained in this way, with the corresponding atoms numbered inFig. 3c (each spectrum presented consists of four spectra in total).

The delocalization effect at the carbon K-edge (,290 eV) with anincident electron probe of 60 kV is estimated as 0.20–0.25 nm in classicaltheory8 and as ,0.12 nm at 300 kV more recently15. Therefore the EELSsignals, if combined with the probe size (,0.1 nm), may not be com-pletely localized at the single atoms on which the probe is exactlypositioned. However the series of ELNES spectra in Fig. 3d stronglysuggest that site-specific spectroscopy is indeed possible with atomicresolution at the graphene edge. The spectrum from atom 1 clearlyshows peak S at 283.6 eV (indicated by a dotted circle), which is relatedto the Klein edge, but spectrum 2 does not show peak S (it may be aminor feature). Spectrum 5 shows a small trace of peak D, which canreasonably be explained by a possible introduction of the bond-breakage

during the probe-scanning across the atom. Spectrum 8 from an atom1.5 nm away from the edge shows normal sp2 features with the sharp p*and excitonic s* peaks, which is very close to the bulk spectrum12.

We performed intensity mapping of peaks D and S to estimate thedelocalization effects further. A number of experiments, involving oneset of spectrum-image and seven sets of spectrum-line on the grapheneedges, are summarized in Supplementary Figs 3, 4 and 5. Resultsconfirm that single-atom spectroscopy at specific sites of the grapheneedge is indeed feasible with the reduced delocalization effect.

We found no trace of oxygen at the investigated edges. This maycontradict a generally accepted concept in which the graphene edgecan be terminated by -OH or -COOH groups and the edge carbonatoms cannot be bared1. In this experiment, in situ etching with con-tinuous removal of the carbon edge atoms in vacuum always takesplace and therefore the edge structures are always kept fresh.

From this study, we have picked up some practical informationabout the graphene edge engineering. The open edges involve bothsingle- and double-coordinated carbon atoms but their specific edge

280 290 300Energy (eV)

Uno

cup

ied

par

tial d

ensi

ty o

f sta

tes

for

2p c

arb

on

a

b

c

d π*

×1/2

Figure 2 | ELNES simulations for three graphene edge structures. CarbonK-edge spectra simulated for the Klein edge (a), zigzag edge (b), armchair edge(c) and bulk (three-coordinated) atom (d). A core–hole was introduced bypartially removing a 1s electron from the carbon atoms (indicated by pinkshading) to estimate the relative peak shift of the spectra. The reduced excitonpeak found experimentally is well reproduced. The simulated ELNES from thezigzag and armchair edges show at least a qualitative match with experiments,although the absolute value for the energy shift cannot be fully confirmed.

LETTER RESEARCH

2 3 / 3 0 D E C E M B E R 2 0 1 0 | V O L 4 6 8 | N A T U R E | 1 0 8 9

Macmillan Publishers Limited. All rights reserved©2010

states are completely localized at the atomic level. Even for triple-coordinated carbon atoms, slight electronic structure modification,as indicated by the restricted excitonic effect (or the reduced s* peak),may exist near the edge region but it vanishes after 1.5 nm from theedge front. The properties of graphene nanoribbons with smallerwidths might be governed by the edge effects16.

It is very surprising that the EELS signal delocalization has turned outnot to be very important for atom-by-atom spectroscopy in the presentexperiment. The EELS signal delocalization should be substantiallydecreased when a lower accelerating voltage is used for the incidentelectron probe8. The delocalization effect with a 30–60 kV incidentprobe is only a fraction of that for the normal STEM operation voltageat 200–300 kV. Lowering the accelerating voltage of the electron micro-scope is therefore very beneficial, reducing the delocalization effect inaddition to contrast enhancement and damage reduction.

ELNES analysis from single atoms is highly desirable because the richinformation it supplies will become accessible from individual atoms atany local area. The ELNES fingerprinting method has been widely usedto determine the electronic/bonding states of unknown materials bycomparison with the reference spectra of known materials. For example,the chemical state of Ce31 or Ce41 in metallofullerene molecules hasbeen clearly discriminated at the single-atom level simply by measuringthe energy shift17. Here we have demonstrated the possibilities of ELNESspectra analysis beyond the simple fingerprinting method. Non-bulkatoms provide peculiar electronic structures and therefore their ENLESshould be completely new (or previously unknown) and cannot becompared with any existing reference. Further efforts should be madeto obtain the electronic state information from new ELNES spectra bycombining atomic resolution imaging with theoretical calculations.

METHODS SUMMARYSTEM-EELS experiments. A JEOL 2100F transmission electron microscope withthe DELTA corrector was operated at 60 kV (ref. 9). The energy resolution wasaround 0.4 eV. We used a probe of 0.1 nm diameter with 20 pA for experiments.For spectroscopy, we used GIF Quantum18, designed for low-voltage operations.The convergence angle for incident probe was set to 30 mrad, while the inner anglefor ADF imaging was around 45–50 mrad, which is equal to the EELS collectionangle. ELNES analysis was performed at each pixel while the incident probe

digitally scanned11. The spectrum-image mode, consisting of a two-dimensionalset of ELNES spectra, takes longer for total acquisition and easily leads to thedestruction of the specimen. Therefore we frequently used the spectrum-linemode, consisting of a one-dimensional set of ENLES spectra, in this study.Typical acquisition time is around 0.1 to 1.0 s for each spectrum. A spectrum lineconsists of 100 spectra, while an image spectrum consists of typically 12 3 12spectra (see also Supplementary Fig. 3).Specimen preparation. Commercially available synthetic HOPG (NT-MDTCompany) was used for experiments. Some of the flakes were cleaved usingScotch tapes and then transferred to transmission electron microscope microgridsfollowing the method developed by Meyer and co-workers19.ELNES simulations. The first-principles calculation based on DFT theory wasused to estimate energy levels and partial density of states on carbon atoms ofgraphene structures. In the discrete variance-Xa method, the energy levels andpartial density of states of unoccupied carbon 2p orbitals are estimated from theself-consistent charge calculation. To estimate the threshold energy of the carbonK-edge, the core–hole effect was considered by employing the transition-stateapproximation method, which configures a half-electron removed from the carbon1s orbital and added to an unoccupied orbital20,21. See also Supplementary Fig. 2.

Received 23 July; accepted 9 November 2010.

Published online 15 December 2010.

1. Kobayashi, Y., Fukui, K., Enoki, T. & Kusakabe, K. Edge state on hydrogen-terminated graphite edges investigated by scanning tunneling microscopy. Phys.Rev. B 73, 125415 (2006).

2. Gass, M. H. et al. Free-standing graphene at atomic resolution. Nature Nanotechnol.3, 676–681 (2008).

3. Liu, Z., Suenaga,K., Harris, P.& Iijima,S. Openand closededgesof graphene layers.Phys. Rev. Lett. 102, 015501 (2009).

4. Girit, C. O. et al. Graphene at the edge: stability and dynamics. Science 323,1705–1708 (2009).

5. Krivanek, O. L. et al. Atom-by-atom structural and chemical analysis by annulardark-field electron microscopy. Nature 464, 571–574 (2010).

6. Suenaga, K.et al. Element selective single atom imaging. Science 290, 2280–2282(2000).

7. Krivanek, O. L. et al. Gentle STEM: ADF imaging and EELS at low primary energies.Ultramicroscopy 110, 935–945 (2010).

8. Egerton,R. F.ElectronEnergy-LossSpectroscopy in theElectronMicroscope2ndedn,363–369 (Plenum, 1996).

9. Sasaki, T. et al. Performance of low-voltage STEM/TEM with delta corrector andcold field emission gun. J. Electron Microsc. 59, s7–s13 (2010).

10. Zobelli, A., Gloter, A., Ewels, C. P., Seifert, G. & Colliex, C. Electron knock-on crosssection of carbon and boron nitride nanotubes. Phys. Rev. B 75, 245402 (2007).

11. Jeanguillaume, C. & Colliex, C. Spectrum-image: the next step in EELS digitalacquisition and processing. Ultramicroscopy 28, 252–257 (1989).

12. Garvie, L. A., Craven, A. J. & Brydson, R. Use of electron-energy loss near-edge finestructure in the study of minerals. Am. Mineral. 79, 411–425 (1994).

13. Klein, D. J. Graphitic polymer strips with edge states. Chem. Phys. Lett. 217,261–265 (1994).

14. Kusakabe, K. & Maruyama, M. Magnetic nanographite. Phys. Rev. B 67, 092406(2003).

15. Cosgriff, E. C., Oxley, M. P., Allen, L. J. & Pennycook, S. J. The spatial resolution ofimaging using core-loss spectroscopy in the scanning transmission electronmicroscope. Ultramicroscopy 102, 317–326 (2005).

16. Jiao, L. et al. Facile synthesis of high-quality graphene nanoribbons. NatureNanotechnol. 5, 321–325 (2010).

17. Suenaga, K. et al. Visualising and identifying single atoms using electron energy-loss spectroscopy with low accelerating voltage. Nature Chem. 1, 415–418 (2009).

18. Gubbens, A. et al. The GIF Quantum, a next generation post-column imagingenergy filter. Ultramicroscopy 110, 962–970 (2010).

19. Pacile, D., Meyer, J. C., Girit, C. O. & Zettl, A. The two-dimensional phase of boronnitride: few-atomic-layer sheets and suspended membranes. Appl. Phys. Lett. 92,133107 (2008).

20. Mizoguchi, T. et al. Core-hole effects on theoretical ELNES/NEXAFS of MgO. Phys.Rev. B 61, 2180–2187 (2000).

21. Koshino, M., Kurata, H. & Isoda, S. Study of structures at the boundary and defectsin organic thin films of perchlorocoronene by high-resolution and analyticaltransmission electron microscopy. Ultramicroscopy 110, 1465–1474 (2010).

Supplementary Information is linked to the online version of the paper atwww.nature.com/nature.

Acknowledgements This work is partially supported by the JST-CREST programme.We thank C. Jin for discussions and H. Kobayashi for specimen preparations.

Author Contributions K.S. designed and performed experiments. K.S. and M.K.analysed data. M.K. performed simulations. K.S. and M.K. co-wrote the paper.

Author Information Reprints and permissions information is available atwww.nature.com/reprints. The authors declare no competing financial interests.Readers are welcome to comment on the online version of this article atwww.nature.com/nature. Correspondence and requests for materials should beaddressed to K.S. (suenaga-kazu@aist.go.jp).

280

8

7

6

5

4

3

2

1

1 2 3 4 5 6 7 8

200

100

0 0.5 1.0

AD

F co

unts

(a.u

)

CC

D c

ount

s (a

.u)

1.5 2.0290 300

Energy loss (eV)Probe position (nm)

a

b

c

d

A B

A B

Figure 3 | Atom-by-atom spectroscopy across the Klein edge. a, ADF imageof graphene edge (no image-processing). The dotted arrow indicates where thespectrum-line was made (A to B). Scale bar, 0.5 nm. b, An atomic model of theinvestigated edge. c, Line-profile of the ADF counts (in red) recordedsimultaneously with the spectrum-line. For comparison with the simulatedADF counts (blue), the number of each atom is indicated (from 1 to 8). d, Thecarbon K-edge ELNES obtained from each atom across the Klein edge. Thesingle-coordinated carbon atom (numbered 1) clearly shows peak S.

RESEARCH LETTER

1 0 9 0 | N A T U R E | V O L 4 6 8 | 2 3 / 3 0 D E C E M B E R 2 0 1 0

Macmillan Publishers Limited. All rights reserved©2010