RES EARCH

ELECTRON MICROSCOPY supplementary materials (fig. S1). Compared

Single-atom vibrational spectroscopy in the scanning

with a conventional on-axis geometry, where the
EELS aperture is centered on the BF disc, this

transmission electron microscope

off-axis or dark-field EELS geometry marked-
ly suppresses the relative contributions of elec-
trons having undergone elastic and delocalized
F. S. Hage1, G. Radtke2*, D. M. Kepaptsoglou1,3, M. Lazzeri2, Q. M. Ramasse1,4* phonon scattering, favoring instead highly
localized impact phonon scattering (23). This
Single-atom impurities and other atomic-scale defects can notably alter the local vibrational responses approach makes it possible to record atomic-
of solids and, ultimately, their macroscopic properties. Using high-resolution electron energy-loss resolution phonon scattering maps of nanometer-
spectroscopy in the electron microscope, we show that a single substitutional silicon impurity in thick flakes of hexagonal boron nitride (22)
graphene induces a characteristic, localized modification of the vibrational response. Extensive ab initio or of single-layer graphene (fig. S2), where
calculations reveal that the measured spectroscopic signature arises from defect-induced pseudo- the off-axis geometry is key because the on-axis
localized phonon modes—that is, resonant states resulting from the hybridization of the defect modes EELS phonon response of graphene is vanish-
and the bulk continuum—with energies that can be directly matched to the experiments. This finding ingly small (24). Note that the large beam con-
realizes the promise of vibrational spectroscopy in the electron microscope with single-atom sensitivity vergence that is necessary for an atomic-sized
and has broad implications across the fields of physics, chemistry, and materials science. probe results in spectral integration over a
range of momentum transfer in the sample

C
plane. To achieve a signal-to-noise ratio suf-
hanges in the normal mode frequencies response of materials at a spatial resolution ficient for resolving the phonon loss spectrum
of dynamical systems that arise from that is superior to that of other experimental fine structure, the electron beam is scanned
the presence of impurities have been techniques (9, 10). Tip-enhanced Raman spec- repeatedly over a small window, tightly de-
studied since the 19th century, which has troscopy (TERS) (11) and inelastic electron fined around the impurity of interest, while
resulted in the set of classical theorems tunneling spectroscopy (IETS) (12, 13) provide the spectrum intensity is accumulated (25).
now referred to as the Rayleigh theorems (1, 2). high spatial and energy resolution alternatives, Figure 1B shows a dark-field EEL spectrum
However, the modern theory of defect modes but they are strictly limited to surface exper- from a single Si atom impurity in graphene
in crystals was established in the 1940s with iments and, therefore, present challenges for a (labeled Si) alongside that acquired from a
the pioneering work of Lifschitz (3). Many range of applications. Vibrational STEM-EELS, comparably sized region of pristine graphene
studies followed, mainly based on optical spec- on the other hand, takes advantage of versatile (labeled C), located only a few atoms away
troscopies (4), which identified two types of probe-forming optics to offer ground-breaking from the Si impurity. The relative positions of
nontrivial defect-induced modes known as lo- capabilities: nanometer-scale thermometry the two scanned regions are indicated by red
calized and resonant modes. Resonant modes (14), mapping of bulk and surface-phonon- (Si) and blue (C) boxes on the high-angle an-
are also called quasi- or pseudo-localized modes polariton modes (15), establishing phonon dis- nular dark-field (HAADF) image in Fig. 1C. A
because, despite being spatially extended, they persion diagrams from nano-objects (16), and close-up of the probed Si atom (Fig. 1D) and
involve a large-amplitude vibration of the im- site-specific isotopic labeling in molecular the corresponding fine structure of the Si L2,3
purity itself. Defect modes can control ma- aggregates (17). These reports highlight the ionization edge (fig. S3C) confirm that the
terials’ properties such as electric and heat complementarity of STEM-EELS with con- brighter-contrast Si atom is trivalently substi-
transport or, more generally, processes that ventional vibrational spectroscopies whose tuted into the graphene lattice. Asymmetric
are affected by the scattering of electrons or energy resolutions remain unmatched. How- annular dark-field (aADF, thus denoted be-
phonons. This can be exploited, for example, ever, the ultimate promise of vibrational STEM- cause of the off-axis geometry) movies were
to suppress heat propagation in thermoelec- EELS is the ability to reach the single-atom recorded during spectrum acquisition to mon-
trics using rattler modes (5), to tune the super- or molecular level, in the same way that modern itor possible beam-induced structure modifi-
conductivity in two-dimensional films (6), or to microscopes have enabled electronic structure cations, while ensuring that the probed atom
affect the optoelectronic properties of conduct- analysis (18), plasmonic (19) and UV-optical re- remained centered within the scanned region.
ing polymers (7). Although the existence of an sponse fingerprinting (20), and energy-dispersive Averaged aADF movies are shown as insets in
atomically localized spectroscopic signature of x-ray spectroscopy (21) from single atoms. Atom- Fig. 1B, with individual frames shown in the
single-atom defects has long been discussed (8), ically resolved phonon maps of bulk systems supplementary materials.
conventional vibrational spectroscopies typi- are preliminary steps in this direction (22). The Si and C spectra in Fig. 1B are normal-
cally average information over much larger In this work, we use STEM-EELS to mea- ized to the maximum of their respective zero-
length scales. sure the localized vibrational signature of a loss peaks (ZLPs). As a result, the tails of the
Vibrational electron energy-loss spectroscopy single trivalent substitutional Si atom in single- ZLPs closely overlap immediately before the
(EELS) in the scanning transmission electron layer graphene (Si@Gr). From ab initio simu- first observable loss features, which allows for
microscope (STEM) has recently emerged as lations, we attribute the measured atomic-scale a straightforward visual comparison of rela-
a powerful means of probing the vibrational spectroscopic response to scattering by pseudo- tive changes in energy loss caused by inelastic
localized vibrational modes arising from a scattering by phonons. Any change in spec-
1
SuperSTEM Laboratory, SciTech Daresbury Campus,
resonance between the Si impurity–specific trum intensity above the coinciding ZLP tails
Daresbury WA4 4AD, UK. 2Sorbonne Université, Muséum modes and the bulk continuum. should be representative of differences in rela-
National d'Histoire Naturelle, UMR CNRS 7590, Institut de Figure 1A illustrates how electron beam tive phonon scattering probability. The fine
Minéralogie, de Physique des Matériaux et de Cosmochimie,
deflectors are adjusted to displace the EEL structure in the phonon energy range of the
75005 Paris, France. 3York Nanocentre and Department
of Physics, University of York, Heslington, York YO10 5DD, spectrometer entrance aperture by 69 mrad two recorded spectra is strikingly different.
UK. 4School of Chemical and Process Engineering and (or an 8.87-Å−1 momentum transfer) with re- Although the C spectrum is consistent with
School of Physics and Astronomy, University of Leeds, spect to the bright field (BF) disc so that that of nondoped bulk graphene (24), the Si
Leeds LS2 9JT, UK.
*Corresponding author. Email: qmramasse@superstem.org these no longer overlap. Further details of the spectrum comprises phonon loss features at
(Q.M.R.); guillaume.radtke@sorbonne-universite.fr (G.R.) experimental geometry are provided in the different energies.

Hage et al., Science 367, 1124–1127 (2020) 6 March 2020 1 of 4

RES EARCH | REPOR T

Fig. 1. Experimental geometry and vibrational STEM-EEL spectrum of a Si spectra, respectively, were acquired. (D) HAADF close-up of the (bright) trivalent
impurity in graphene. (A) Beam deflectors shift the BF disc away from the EEL Si impurity. (E) Detail of the normalized Si and C EEL spectra shown in (B) and
spectrometer entrance aperture (Ap.) in the diffraction plane. (B) Normalized the difference spectrum. L, longitudinal; T, transverse; A, acoustic; O, optical.
vibrational EEL spectra of a substitutional Si impurity and of defect-free (F) Comparison of the calculated differential PPDOS (broadened to match the
graphene. Insets show aADF images of the repeatedly scanned sample regions. experimental resolution) and the experimental difference spectrum. The blue and
Smoothed spectra (thin lines) are superimposed on the raw data (shading red shaded areas highlight energy ranges where the contributions of the Si
around the lines). (C) HAADF overview of the experimental region. Red and blue impurity and its three nearest neighbors, or that of bulk graphene, are
boxes indicate the positions of the sub-scan regions from which the Si and C comparatively stronger. au, arbitrary units.

Figure 1E shows, in greater detail, the pho- ability induced by the presence of the single plane perpendicular to the electron beam
non loss region of the spectra. The C spec- Si atom impurity. Virtually identical results trajectory in our experiments, only the com-
trum exhibits two distinct loss peaks at 85 meV (detailed in the supplementary materials) were ponents of the phonon polarization that are
(685 cm−1) and 170 meV (1371 cm−1). Following obtained from complementary measurements parallel to the graphene plane are relevant. A
(24), we attribute these peaks to scatter- carried out in a different area of the sample. tentative comparison to the experimental dif-
ing by transverse (T) or longitudinal (L), These experimental results lead to the remark- ference spectrum is then provided by combin-
acoustic (A) or optical (O) modes in graphene, able conclusion that the single Si atom impurity ing the PPDOS projected on the Si atom, nSi;
respectively (the graphene-phonon dispersion in graphene possesses a characteristic vibra- the PPDOS projected on its three C neighbors,
diagram is presented for reference in fig. S10). tional signature localized at the atomic scale. nC1; and the bulk phonon DOS per atom, nbulk:
Spectral contributions of out-of-plane phonon To gain insights into the physics associated ~ ðwÞ ¼ ½nSi ðwÞ þ 3nC1 ðwÞ  4nbulk ðwÞ=4. This
n
modes are expected to be negligible, as the in- with these results, we have calculated, within differential PPDOS reflects the experimen-
cident electron beam is normal to the graphene the framework of density functional theory tal spectrum averaging over the scanning
plane. Despite stemming from a position only (DFT) (26) and using periodic boundary con- window, which is expected to include con-
a few atoms away, the Si spectrum shows a ditions, the vibrational spectrum of a large tributions from the impurity’s neighboring
remarkably different phonon fine structure 96×96 supercell of graphene (96 unit cells by C atoms. The resulting differential PPDOS is
comprising a prominent loss peak at about 96 unit cells) containing one substitutional shown in Fig. 1F, after broadening to match
55 meV (443 cm−1) and weaker structures at Si atom. The structure of the defect and com- the experimental resolution. It predicts all the
125 and 150 meV (1008 and 1209 cm−1). To putational details are presented in the sup- main features of the experimental difference
enhance the differences between the spectra, plementary materials. As discussed therein, spectrum, including a single peak at ~55 meV
we subtracted the C from the Si spectrum, and the important features observed in the vibra- (443 cm−1), two overlapping peaks at 125
the resulting difference spectrum is shown in tional EEL spectra of graphene can be safely and 150 meV (1008 and 1209 cm−1), and dips
Fig. 1, E and F. This has the additional benefit interpreted in terms of the phonon density centered around 100 and 180 meV (807 and
of effectively subtracting the elastic scattering of states (DOS) of the bulk. The local be- 1452 cm−1).
ZLP tail (making the reasonable assumption havior of the DOS can be quantified by the The physical origin of these spectral fea-
that the tail contribution, before any expected projectedX phonon DOS (PPDOS), defined as tures can be understood by considering the in-
loss contribution, is similar between spectra) nk ðwÞ ¼ jek j2 dðw  wn Þ, where k denotes
n n
dividual in-plane PPDOS employed to construct
without possible errors associated with com- a specific atom, wn and en are the phonon an- the differential PPDOS, ñ, and the PPDOS of
mon background removal techniques, as dis- gular frequency and normalized polarization, C atoms located at increasing distances away
cussed in the supplementary materials (fig. and the sum is carried over all the phonon from the Si impurity (Fig. 2A). The Si PPDOS
S4). Thus, the difference can be interpreted as modes, n, of the supercell. Because the momen- is dominated by an intense peak at 55 meV,
a relative change in phonon scattering prob- tum transfer occurs predominantly in the closely matching the low-energy experimental

Hage et al., Science 367, 1124–1127 (2020) 6 March 2020 2 of 4

RES EARCH | REPOR T

feature seen in Fig. 1E. This peak is followed

by a broad band with weaker structures at 105,
127, and 155 meV (847, 1024, and 1250 cm−1).
The absence of intense features in the bulk
graphene DOS at 55 meV implies that the
corresponding modes should possess a degree
of localization. By inspecting the PPDOS of
neighboring C atoms in Fig. 2, it is evident
that, although the first two C neighbor shells
coordinating the impurity still display traces
of the 55 meV peak, its contribution is weak.
The PPDOS of subsequent neighbors rapidly
tends toward the bulk signature, which is fully
retrieved after six shells. Corresponding atom-
ically resolved experimental spectra in Fig. 2C,
from a full-spectrum image over equivalent
C neighbor positions (fig. S8), exhibit an iden-
tical trend: The EELS signal reproduces the
main features observed in the in-plane PPDOS.
It is instructive to consider a calculation
performed on a smaller 13-atom fragment of
C3v symmetry centered on the impurity (fig.
S9), decoupled from the supercell by artifi-
cially setting the interatomic force constants
linking the fragment to the rest of the 96×96
supercell to zero. The fragment displays two
modes with E symmetry at 52 and 124 meV
Fig. 2. Localization of the vibrational signal. (A) Calculated in-plane component of the phonon DOS
(419 and 1000 cm−1), involving large in-plane dis-
projected on the Si and C atoms at increasing distances from the impurity. Overlaid light gray lines show the
placements of the Si atom either in phase (mode
bulk graphene phonon DOS per atom, fully recovered from atom 6. The curves are vertically shifted and
A) or out of phase (mode B) with the neigh-
smeared by a 2-meV FWHM Lorentzian for clarity. (B) Sketch of the position of the C atoms, labeled 1 to 6,
boring C atoms (Fig. 3). The resonances in the
and Si impurity (red sphere). (C) Background-subtracted experimental spectra acquired at equivalent atomic
full Si@Gr system, simulated by the 96×96
positions. Smoothed (black lines) and raw (gray dots) data are overlaid.
supercell, can thus be interpreted as a hybrid-
ization of these local impurity modes with the
Fig. 3. Localized com- vibrational continuum of the graphene bulk.
ponents of the Si vibra- The associated atomic displacements, in-
tions. (A) Gray histogram
A cluding those arising from the in-plane vibra-
shows the square of the tion of the Si atom, do not decay far from the
phonon eigenmodes of the defect; the full system presents a delocalized
96×96 supercell projected continuum, a concept quantified with the in-
on the in-plane Si atom verse participation ratio analysis shown in
component. Black line the supplementary materials. However, these
indicates the in-plane delocalized phonon modes possess an enhanced
component of the phonon component atomically localized on the im-
DOS projected on the Si purity. The power of EELS is the technique’s
atom (same as the Si ability to probe this quasi-localization, there-
PPDOS shown in Fig. 2) by revealing the paradoxical nature of defect-
obtained by broadening induced resonant modes. It is also notable
the gray histogram with a that the experimentally measured ~30 meV
2.0-meV FWHM Lorentzian. (242 cm−1) full-width at half-maximum (FWHM)
The red histogram is cal- of the impurity peak at ~55 meV (Fig. 1F)
culated from a 13-atom closely matches the intrinsic theory-predicted
fragment centered on the
impurity. Two dominant
B width of the resonant mode (Fig. 3A). The
experimental energy resolution is therefore
modes, denoted A and B, not limiting, and the EEL spectra faithfully
are observed. (B) Atomic capture the fine structure of the Si@Gr sys-
model of the 13-atom tem’s vibrational response.
fragment (the Si atom is Localized and resonant modes arising from
shown in red), with relative point defects have been widely discussed (8).
atomic displacements for The former are characterized by frequencies
modes A and B indicated lying out of the continuum of the unperturbed
as arrows with lengths crystal and atomic amplitudes dying off faster-
proportional to the than-exponentially with increasing distance
displacement amplitudes. from the defect (27). By contrast, the latter

Hage et al., Science 367, 1124–1127 (2020) 6 March 2020 3 of 4

RES EARCH | REPOR T

occur at frequencies lying within the allowed dimensional structures, although challenges 23. C. Dwyer et al., Phys. Rev. Lett. 117, 256101 (2016).
bands of the host. The recognition of reso- will arise from the complexity of the compu- 24. R. Senga et al., Nature 573, 247–250 (2019).
25. Q. M. Ramasse et al., Nano Lett. 13, 4989–4995 (2013).
nant modes was delayed by their peculiar tational work necessary to inform these ex- 26. P. Giannozzi et al., J. Phys. Condens. Matter 21, 395502
characteristics, where the vibrational ampli- periments. Nevertheless, this opens up a path (2009).
tude does not vanish far from the defect, ex- to further applications in solid-state science, 27. E. W. Montroll, R. B. Potts, Phys. Rev. 100, 525–543
(1955).
tending instead over the entire crystal (28). where the electron beam of the STEM can be 28. R. Brout, W. Visscher, Phys. Rev. Lett. 9, 54–55 (1962).
Furthermore, experimental observations of used to assemble functional devices atom by 29. M. T. Hotz et al., Microsc. Microanal. 24, 1132–1133 (2018).
these effects have, thus far, been limited to atom (30) and to spectroscopically probe the 30. T. Susi et al., 2D Mater. 4, 042004 (2017).
indirect fingerprints, often at the macroscopic resulting lattice dynamics and their coupling AC KNOWLED GME NTS
scale. Volgmann et al. (13) used scanning probe with other quasiparticles. The authors gratefully acknowledge R. Brydson, M. Bugnet, and
microscope (SPM)–IETS to detect a local energy- E. Prestat for useful discussions. Funding: SuperSTEM is the UK
dependent increase in phonon DOS on a Ag (100) RE FERENCES AND NOTES National Research Facility for Advanced Electron Microscopy,
supported by the Engineering and Physical Sciences Research
surface, which they attributed to a substitu- 1. W. R. Hamilton, Philos. Trans. R. Soc. 124, 247–308 (1834).
Council (EPSRC). This work was granted access to the
tional Cu atom. However, the surface nature 2. J. W. Strutt, The Theory of Sound (Cambridge Univ. Press,
High-Performance Computing resources of the Institut du
1887).
of these experiments and the lack of more 3. I. M. Lifšic, Nuovo Cim. 3, 716–734 (1956).
Développement et des Ressources en Informatique Scientifique
(IDRIS) under allocation 2019-A0060910820, attributed by the
direct visualization means precluded an un- 4. A. S. Barker, A. J. Sievers, Rev. Mod. Phys. 47, S1–S179 (1975).
Grand Equipement National de Calcul Intensif (GENCI). Author
ambiguous interpretation. 5. D. J. Voneshen et al., Nat. Mater. 12, 1028–1032 (2013).
contributions: Q.M.R. conceived the project. F.S.H., Q.M.R., and
6. C. Brun et al., Nat. Phys. 10, 444–450 (2014).
In contrast, the ability demonstrated in this 7. M. Shao et al., Nat. Commun. 5, 3180 (2014).
D.M.K. designed and carried out the experiments and interpreted
the data. G.R. and M.L. carried out and interpreted the theoretical
work to directly measure, at the atomic scale, 8. R. F. Wallis, Ed., Localized Excitations in Solids (Springer, 1968).
calculations. All authors contributed to the preparation of the
the localized component of the vibrational 9. O. L. Krivanek et al., Nature 514, 209–212 (2014).
manuscript. Competing interests: The authors declare no
10. T. Miyata et al., Microscopy 63, 377–382 (2014).
signature of a single impurity atom within a 11. P. Z. El-Khoury et al., Adv. Phys. X. 1, 35–54 (2016).
competing interests. Data and materials availability: All data
solid, and to match the observed spectral fine necessary for evaluating the conclusions of the paper are included
12. H. Gawronski, M. Mehlhorn, K. Morgenstern, Science 319,
in the main text and/or the supplementary materials. Data related
structure to theoretically predicted modes, 930–933 (2008).
to those presented here are available from the authors upon
13. K. Volgmann et al., Nat. Commun. 5, 5089 (2014).
realizes the potential of phonon spectroscopy 14. J. C. Idrobo et al., Phys. Rev. Lett. 120, 095901 (2018).
reasonable request.
in the STEM. The STEM-EELS technique, char- 15. M. J. Lagos, A. Trügler, U. Hohenester, P. E. Batson, Nature
acterized by single-atom defect sensitivity com- 543, 529–532 (2017). SUPPLEMENTARY MATERIALS
16. F. S. Hage et al., Sci. Adv. 4, eaar7495 (2018). science.sciencemag.org/content/367/6482/1124/suppl/DC1
bined with isotope selectivity (17) and the ability
17. J. A. Hachtel et al., Science 363, 525–528 (2019). Materials and Methods
to operate at cryogenic temperatures (29), now 18. K. Suenaga, M. Koshino, Nature 468, 1088–1090 (2010). Supplementary Text
enables potential experiments where a single 19. W. Zhou et al., Nature Nanotech. 7, 161–165 (2012). Figs. S1 to S12
functionalizing isotope is fingerprinted at the 20. F. S. Hage et al., ACS Nano 12, 1837–1848 (2018). References (31–43)
21. T. C. Lovejoy et al., Appl. Phys. Lett. 100, 154101 (2012).
atomic scale through its vibrational signature. 22. F. S. Hage, D. M. Kepaptsoglou, Q. M. Ramasse, L. J. Allen, 7 November 2019; accepted 3 February 2020
The approach should be applicable to three- Phys. Rev. Lett. 122, 016103 (2019). 10.1126/science.aba1136

Hage et al., Science 367, 1124–1127 (2020) 6 March 2020 4 of 4