Letter doi:10.

1038/nature25987

Massive Dirac fermions in a ferromagnetic kagome

metal
Linda Ye1*, Mingu Kang1*, Junwei Liu1†, Felix von Cube2†, Christina R. Wicker1, Takehito Suzuki1, Chris Jozwiak3,
Aaron Bostwick3, Eli Rotenberg3, David C. Bell2,4, Liang Fu1, Riccardo Comin1 & Joseph G. Checkelsky1

The kagome lattice is a two-dimensional network of corner-sharing of kagome insulators4, experimental realization of the former has been
triangles1 that is known to host exotic quantum magnetic states2–4. challenging, in part owing to the relative rarity of kagome materials.
Theoretical work has predicted that kagome lattices may also host An approach to realizing metallic kagome networks in the hexagonal
Dirac electronic states5 that could lead to topological6 and Chern7 transition-metal stannides AxSny (A = Mn, Fe or Co; x:y = 3:1, 3:2 or
insulating phases, but these states have so far not been detected in 1:1) has been reported17. As shown in Fig. 1g for A = Fe (studied here),
experiments. Here we study the d-electron kagome metal Fe3Sn2, starting from a single layer of a hexagonal close-packed structure of
which is designed to support bulk massive Dirac fermions in iron atoms, a kagome net emerges naturally by replacing a 2 × 2 sub-
the presence of ferromagnetic order. We observe a temperature- lattice (dashed cell) with tin atoms, resulting in an Fe3Sn plane with an
independent intrinsic anomalous Hall conductivity that persists underlying iron kagome lattice.
above room temperature, which is suggestive of prominent Berry Here we study the bilayer kagome compound Fe3Sn2 (space group
curvature from the time-reversal-symmetry-breaking electronic R3m ; hexagonal lattice constants a = 5.338 Å and c = 19.789 Å)—a
bands of the kagome plane. Using angle-resolved photoemission structural variation of FexSny that includes a stanene layer sandwiched
spectroscopy, we observe a pair of quasi-two-dimensional Dirac between Fe3Sn bilayers (Fig. 1h). In Fig. 1h we also show a correspond-
cones near the Fermi level with a mass gap of 30 millielectronvolts, ing transmission electron microscopy image of a (1010) cross-section
which correspond to massive Dirac fermions that generate Berry- of a single crystal of Fe3Sn2, which reveals the Fe3Sn and stanene layers.
curvature-induced Hall conductivity. We show that this behaviour is Previous studies17 have identified Fe3Sn2 as an unusual magnetic con-
a consequence of the underlying symmetry properties of the bilayer ductor with a high Curie temperature of TC = 670 K. Although attention
kagome lattice in the ferromagnetic state and the atomic spin–orbit was originally focused on the zero-field spin structure18, recent studies
coupling. This work provides evidence for a ferromagnetic kagome have focused on the formation of skyrmion bubbles19 and a substantial
metal and an example of emergent topological electronic properties anomalous Hall effect20,21 at finite field. The latter is particularly inter-
in a correlated electron system. Our results provide insight into the esting in comparison with the structurally related antiferromagnets
recent discoveries of exotic electronic behaviour in kagome-lattice Mn3Sn and Mn3Ge, which were recently reported to have a large
antiferromagnets8–10 and may enable lattice-model realizations of room-temperature anomalous Hall response8,9 and possible Weyl
fractional topological quantum states11,12. fermion states10.
The kagome lattice (Fig. 1a) is a network with trihexagonal (3.6)2 Measurements of magnetization M as a function of magnetic induc-
Archimedes tiling that has been studied extensively in the context of tion B along the c axis (Fig. 2a) demonstrate that the system is a soft
frustration-induced quantum-spin-liquid phases2–4. In terms of elec- ferromagnet, with the saturation field and saturation magnetization Ms
tronic structure, simple tight-binding models on kagome lattices have depending mildly on temperature T. The saturation magnetization Ms
long been known to yield unusual features, including dispersionless reaches approximately 1.9μB per iron atom at low temperature T (where
bands and Dirac points (Fig. 1b); the Dirac points appear in a man- μB is the Bohr magneton; Fig. 2a, inset). The crystals exhibit high
ner similar to those in hexagonal graphene lattices13. Although such metallicity, with the residual resistivity ratio of ρ(300 K)/ρ(2 K) ≈ 25
features have not previously been observed in experiments, theo- (Extended Data Fig. 1) allowing characterization by electrical transport.
retical interest has persisted and lead to several further predictions. The transverse resistivity in the kagome plane ρyx(B) (Fig. 2b) strongly
Of particular interest are kagome networks in which time reversal reflects M(B)—a characteristic of the anomalous Hall effect20,21. In fer-
symmetry is broken via ferromagnetism (Fig. 1c)5,7,11, which has the romagnetic conductors it is conventional16 to express ρyx(B) in terms
effect of splitting the spin-degenerate Dirac bands (Fig. 1d). Further of contributions from the ordinary (Lorentz-force) Hall coefficient R0
inclusion of spin–orbit coupling (Fig. 1e) yields various gapped phases and the anomalous Hall coefficient Rs: ρyx = R0B + RsM; as shown in
(Fig. 1f) with integer7 or fractional11,12 topological invariants (Chern the inset of Fig. 2b, R0 depends mildly on T (corresponding to 6 × 1021
numbers). When the chemical potential is within the Dirac gap, the electrons per cm3 at low T), whereas Rs is much larger but decreases
intrinsic anomalous Hall effect, which results from the integration of with decreasing T.
Berry curvature over the Brillouin zone, is quantized and in principle To elucidate the role of the bilayer kagome lattice further, we examine
affords detection of the Chern number of the wavefunction of the sys- the associated Hall conductivities in the kagome plane. The contribu-
tem14,15. More generally, with multiple bands and arbitrary positions tions to the total Hall conductivity σxy = σ Nxy + σ Axy , where the super-
of the chemical potential, such measurements can detect the Berry script ‘N’ (‘A’) denotes the normal (anomalous) component, can be
curvature that is concentrated by massive Dirac bands16. separated by using the field linearity of σ Nxy in the low-Hall-angle limit
Despite a broad theoretical understanding of electronic Berry- (Methods). Although σ Axy is known to have contributions of both intrin-
phase effects in ferromagnetic kagome models and extensive studies sic (Berry curvature) and extrinsic (scattering) origin16, it has recently
1
Department of Physics, Massachusetts Institute of Technology, Cambridge, Massachusetts 02139, USA. 2Harvard John A. Paulson School of Engineering and Applied Sciences, Harvard University,
Cambridge, Massachusetts 02138, USA. 3Advanced Light Source, E. O. Lawrence Berkeley National Laboratory, Berkeley, California 94720, USA. 4Center for Nanoscale Systems, Harvard University,
Cambridge, Massachusetts 02138, USA. †Present addresses: Department of Physics, Hong Kong UST, Clear Water Bay, Hong Kong, China (J.L.); Hitachi High-Technologies Europe GmbH, Krefeld,
Germany (F.v.C.).
*These authors contributed equally to this work.

6 3 8 | N A T U R E | V O L 5 5 5 | 2 9 m ar c h 2 0 1 8
© 2018 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
 Letter RESEARCH

a c e To examine the origin of this Hall response further, we measured

the electronic structure of Fe3Sn2 by using angle-resolved photoemis-
I
–2I sion spectroscopy (ARPES). In Fig. 3a, b we show the experimen-
I tal Fermi surface and energy–momentum dispersion, respectively,
K′
M of the electronic bands along high-symmetry directions parallel to
Γ K the kagome planes, measured at T = 20 K (see also Extended Data
Fig. 2). A rich spectrum of electronic excitations with hexagonal
b d f symmetry is observed, consistent with the metallicity and crystal-
lographic structure described above. More notably, linearly dispers-
Broken ing Dirac cones are observed at the corner points K and K′ of the
time-reversal Spin–orbit Brillouin zone. This spectrum, which is reminiscent of the electronic
symmetry coupling
structure of graphene24, is the long-sought realization of kagome-
derived Dirac fermions5. These Dirac-like bands are shown in detail
in the high-resolution energy–momentum section of the ARPES data
K, K′ K, K′ K, K′
across the K point in Fig. 3c (data are collected along the blue dashed
line in Fig. 3a and then symmetrized in momentum about K); two Dirac
g h
cones, separated in energy but centred at K, are resolved. Hereafter, we
focus on these bands and their role in generating Berry curvature.
The two-fold Dirac cones can also be identified in constant-energy
Fe3Sn contours (Fig. 3d). At the Fermi energy EF (Fig. 3d, top layer), a pair of
Stannene Dirac cones forms two electron pockets centred at K: a circular inner
Fe
pocket and a trigonally warped outer pocket. Moving down from EF
Sn
each pocket shrinks, forming apparent Dirac points at binding energies
c of 70 meV (Fig. 3d, second layer) and 180 meV (Fig. 3d, bottom layer).
b At the midpoint energy (125 meV), the two Dirac cones cross and,
c
a 1 nm within our experimental resolution, form a ring of Dirac points in the
a b x–y momentum plane. The experimental electronic structure near the
-
[1010]
K point is therefore characterized by two energy-split (ΔE = 110 meV)
Figure 1 | The kagome structure and Fe3Sn2. a, b, Structure of the interpenetrating Dirac cones. This splitting is a natural consequence
kagome lattice (a) and the associated Dirac point in the nearest-neighbour of the bilayer kagome structure, similarly to the case of multilayer
tight-binding model (b), with the Brillouin zone shown in the inset. The
band is degenerate, as denoted with red and blue spins. c, d, Ferromagnetic
graphene24, whereas the exchange splitting due to magnetic order
kagome lattice with broken time-reversal symmetry (moments in blue) (c) is expected25 to be much larger (in excess of 2 eV). Photon-energy-
and the associated spin-polarized Dirac band with coupling between the dependent ARPES (Extended Data Fig. 4) reveals negligible variation
magnetization and spin (d). e, f, Spin–orbit-coupled ferromagnetic in the Dirac bands as a function of out-of-plane momentum kz, indi-
kagome lattice with Berry phase φ accrued via hopping (e) and the cating quasi-two-dimensional (quasi-2D) bands confined to the iron
corresponding gapped Dirac spectrum (f). g, The Fe3Sn kagome plane in kagome bilayer.
Fe3Sn2, with the kagome network shown in red. h, Transmission electron Having established the Dirac fermiology of Fe3Sn2, we focus on
microscopy cross-section of Fe3Sn2 and the corresponding Fe3Sn and the role of spin–orbit coupling and the possible mass acquisition
stanene layers viewed from the [1010] direction. of the Dirac bands. Inspection of the raw ARPES data reveals that
the spectral intensity at the Dirac point is suppressed substantially
been demonstrated that the insensitivity of the latter to thermal excita- (Extended Data Fig. 3d), which is more clearly visualized in the second
tions allows the parameterization σ Axy = f (σxx ,0 )σ 2xx + σ int
xy , where f(σxx,0) derivative of the ARPES map (Fig. 3e). Analysis of the energy distri-
is a function of the residual conductivity σxx,0, σxx is the conductivity bution curves displayed in Fig. 3f reveals a break between the upper
22
and σ int
xy is the intrinsic anomalous Hall conductivity . Because σ xy
int
and lower branches of the Dirac cone, which signals the opening of an
does not depend on the scattering rate, in a system with substantial energy gap Δ. A quantitative analysis performed by fitting the energy
2
Berry curvature, σ int A
xy is then the remnant σ xy that is observed as σ xx → 0 distribution curves with Gaussian peaks returns Δ = 30 ± 5 meV
(Fig. 2c, top inset). Figure 2c demonstrates that σ Axy(T ) remains rela- (Methods). This value is similar to that predicted previously for spin–
tively unchanged from this remnant value at high temperature, until orbit-coupled 3d transition metals in kagome lattices7, but smaller
T ≈ 100 K at which an upturn concomitant with increasing σxx(T) is than that observed in magnetically doped topological insulators (about
observed. This upturn is indicative of the onset of an extrinsic response, 50 meV)26,27 and in hydrogen-decorated graphene (at least 0.5 eV)28.
which is expected16 with the longer relaxation time in this range of σxx; The emergence of massive Dirac fermions in Fe3Sn2 can be under-
the subsequent σ 2xx scaling of the additional σ Axy (Fig. 2c, top inset) is stood as a combination of ferromagnetic splitting and spin–orbit cou-
also consistent with an extrinsic origin22,23. The scattering-rate- pling in the underlying kagome geometry. Motivated by the weak kz
independent value of σ Axy at high T persists, varying by about 10% down dispersion observed in ARPES, we consider a stacked system of kagome
to T = 2 K (158 ± 16 Ω−1 cm−1), which corresponds to approximately layers. In Fig. 4a we show a perfect Fe3Sn kagome layer and the corre-
0.27e2/h per kagome bilayer, where e is the electronic charge and h is sponding Brillouin zone. The kagome layer has two-fold and three-fold
the Planck constant. We identify this contribution as σ int xy (Fig. 2c), with rotational symmetries (C2x and C3z, respectively) that leave the K and
behaviour akin to that expected from a massive Dirac band23. K′ points invariant and thus form point group D3. In the absence of
These observations point to a substantial Berry-curvature contri- spin–orbit coupling, the two-fold-degenerate crossing (Dirac) points
bution to the transport response in Fe3Sn2 in a geometry that samples at K and K′ belong to a two-fold irreducible representation (E) and
the kagome planes (Fig. 2c, middle inset). We also measured the are therefore protected. As illustrated in Fig. 4b, a Dirac crossing can
Hall response perpendicular to the kagome plane σzx (Fig. 2c, lower be observed at K in a tight-binding model HK for nearest-neighbour
inset). This out-of-plane signal is much smaller (Fig. 2c), with hopping on the kagome sites:
the ratio σ Azx/σ Axy being less than 10% at the highest T, indicating a
large relative enhancement of the Berry curvature in the kagome HK = ∑ tc †i c j (1)
plane. 〈ij 〉

2 9 m ar c h 2 0 1 8 | V O L 5 5 5 | N A T U R E | 6 3 9
© 2018 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
 RESEARCH Letter

a c 1,200 1.2
2 1,200
B || c
1,000
2K

(Ω–1 cm–1)
800
1 300 K
M ( PB per iron atom)
1,000 600 1.0
2

Ms ( PB per iron atom)

400

Vxy
A
0
200

1 0
800 0 10 20 30 40 50 0.8
–1 2
Vxx (108 Ω–2 cm–2)
0

Vxx (105 Ω–1 cm–1)

0 200 400

V (Ω–1 cm–1)
T (K) Bz
–2 Ey
b 15 600 Ix 0.6
z
B || c 2K
50 K y
10 100 K
x
150 K Ez
200 K By
250 K Ix
5 300 K 400 0.4
350 K
400 K
Uyx (μΩ cm)

A
0 Vxy
int
0 20 Vxy
R0 (μΩcm T–1)

Rs (+Ω cm T–1)

A
–5 200 Vzx 0.2
–0.1 10

–10 –0.2 0
0 200 400
T (K)
–15 0 0
–3 –2 –1 0 1 2 3 0 100 200 300 400
B (T) T (K)
Figure 2 | Anomalous Hall response of Fe3Sn2. a, Magnetization M of σ Axy (red, left axis) and σ Azx (black, left axis) in the x–y and z–x planes,
Fe3Sn2 along the c axis as a function of magnetic induction B at room respectively, along with the longitudinal conductivity σxx (blue, right axis)
temperature (T = 300 K; orange) and low temperature (T = 2 K; black). The and estimated intrinsic Hall conductivity σ int xy (orange, left axis). The
inset shows the saturation magnetization Ms (measured at 2 T) as a measurement configurations for σxy (top) and σzx (bottom) are shown in
function of temperature T. b, Hall resistivity ρyx as a function of B. The the lower inset; I represents the charge current. The upper inset shows σ Axy
inset shows the ordinary and anomalous Hall coefficients R0 (black) and Rs plotted against σ 2xx ; the solid and dashed lines are the scaling curves
(purple), respectively, as a function of T. c, Anomalous Hall conductivities (see text).

where 〈ij〉 indexes nearest-neighbour pairs, t is the hopping integral To connect with the Hall response, we construct a k · p Hamiltonian
and cj (c †i ) is the fermion annihilation (creation) operator, which is near K and K′ for the dual Dirac fermions and fit to the ARPES data
taken to be spin-polarized owing to exchange. The kagome bilayers in (Fig. 4e, inset; Methods). We then calculate the contribution of the
Fe3Sn2 (Fig. 4c) are tiled by triangles of two different bond lengths, massive Dirac bands to the Hall response by integrating the Berry
2.59 Å and 2.75 Å, as indicated by the red and blue shading. However, curvature of the filled states, which yields σ calc
xy = (0.31 ± 0.05)e /h at
2

the combined unit of these kagome layers and the intervening stanene EF for a kagome bilayer (Fig. 4e), comparable to the observed value
layer preserves the {C2x, C3z} symmetry of the perfect kagome lattice of σ int
xy = (0.27 ± 0.03)e /h per bilayer. Remarkably, despite the
2
and the Dirac points are thus protected by crystal symmetry in the simplicity of our model, the action of the quasi-2D massive Dirac
absence of spin–orbit coupling. The additional layer degree of freedom fermions at K and K′ largely accounts for the observed Hall response
further enriches the electronic structure. In particular, the ABA layer with the crystal viewed as a parallel network of bilayer kagome planes.
stacking of the structure in Fig. 4c gives rise to bonding–antibonding However, there are limitations; for example, for a 2D model there is
splitting29, as seen in a simple tight-binding model with this additional no contribution to the out-of-plane Hall response. We suggest instead
hopping (Fig. 4d). that this out-of-plane response originates from the three-dimensional
We next introduce Kane–Mele-type spin–orbit coupling HSOI to the (3D) network of tin atoms and the associated kz-dispersive bands
tight-binding model HK, with near Γ (Methods). The relative smallness of σ Azx is then consistent
with a minor contribution of tin-atom-derived bands to the overall
HSOI = i ∑ λ ij (c †i ↑c j ↑ − c †i ↓c j ↓) (2) Berry curvature. More generally, the model evidences the role of the
〈ij 〉
concentration of Berry curvature in the quasi-2D massive Dirac
where λij represents the effect of spin–orbit coupling and ↑ and bands, which have small EF comparable to the spin–orbit coupling
↓ denote the spin quantum number30. Writing λij = λ(Eij × Rij) · s, strength16. The robustness of the Hall response observed here is
where λ is the spin–orbit coupling constant, E is the electric field comparable to that of the Hall response that is driven by chiral
on the hopping path, R is the hopping vector and s represents the antiferromagnetic order in Mn3Sn (ref. 8) and Mn3Ge (ref. 9);
electron spin, for spin-polarized bands near K and K′ with non-zero ­however, instead of originating from 3D Weyl nodes10, the Hall
z polarization sz, HSOI effectively reduces to the Haldane term31. response observed here is driven by quasi-2D Dirac fermions in a
Accordingly, for the single-layer case (Fig. 4b), when EF is positioned ferromagnetic kagome network interleaved with stanene layers.
in the Dirac gap, the system enters a Chern insulating phase with By combining electrical transport measurements, ARPES and
quantized anomalous Hall conductance7,31. theoretical analysis, this study provides a comprehensive proof of

6 4 0 | N A T U R E | V O L 5 5 5 | 2 9 m ar c h 2 0 1 8
© 2018 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
 Letter RESEARCH

a c d e
0.5 0
K
EF
K
ky (Å–1)

0 Γ M
First Dirac point
−0.5

Energy, E – EF (meV)
−100 f
EF − 70 30 meV
−1.0 −0.5 0 0.5 1.0
kx (Å–1) Dirac circle
b 0.1
−200 0

Intensity
0
EF − 125
Energy, E – EF (eV)

−0.1
K
−0.2 Second Dirac point
−0.3
−300
−0.4 0.2 Å–1 EF − 180
100
−0.5 ky (Å–1)
kx (Å–1) −0.2 −0.1 0
Γ K M Γ −0.3 K 0.3
k (Å–1) k (Å–1) Energy, E – EF (eV)

Figure 3 | Massive Dirac fermion at the zone corner of Fe3Sn2. a first Dirac point (second layer), a Dirac circle (third layer) and a second
a, Experimentally obtained Fermi surface of Fe3Sn2. The hexagonal Dirac point (bottom layer) can be clearly detected from the maps. The
Brillouin zone (red dashed lines) and high-symmetry points (black dots) corresponding energies and Brillouin zone contours are marked with
are marked. b, Experimentally determined band dispersion of Fe3Sn2 along coloured dashed lines in c and d. e, f, The second-derivative plot (e) and
the high-symmetry directions. c, High-resolution ARPES data, measured the stack of energy distribution curves (f) across the Dirac points. Both
along the blue dashed line in a and then symmetrized with respect to K. panels share the momentum range and direction with c. The red double-
The complete band dispersion is shown in Extended Data Fig. 3 (with a headed arrow marks the discontinuity between the upper and lower
modified colour scale). d, Constant-energy maps at binding energies of branches of the Dirac cone. Coloured markers indicate the fitted band
0 meV, 70 meV, 125 meV and 180 meV. Two electron pockets (top layer), energies. All data were obtained using 92-eV photons.

principle for engineering band-structure singularities and topolog- a 2D ‘Chern gap’—a time-reversal-symmetry-broken topologically
ical phenomena in correlated systems. In particular, we realize lat- non-trivial phase that is intrinsic to stoichiometric materials and has
tice-driven6,7,30,31 topological 3d electronic bands, which we suggest a dominant contribution to the electrical response at temperatures of
exhibit the defining properties of a ferromagnetic kagome metal. up to 300 K and above. To isolate these bands, as a step towards reali­
Viewed in isolation, the bands near K can be considered to exhibit zing high-temperature dissipationless modes15, we propose finding

a c e 0
C3z
C3z Fe
Sn

C2x
–100
K′ M C2x
Γ K
EF (meV)

b d
t t –200

K, K′
0

–100
E (meV)
E

–300
–200

–300
–0.1 0.0 0.1
k (Å–1)
–400
Γ K M Γ Γ K M Γ 0 0.5 1.0
k k Vxy (e2/h)

Figure 4 | Tight binding and hall conductivity of a kagome bilayer. hopping 0.3t, with (red) and without (blue) spin–orbit coupling. The
a, Two-fold (C2x) and three-fold (C3z) rotation-axis symmetry operations inset shows a magnified view of the double Dirac structure near K. The
of a single Fe3Sn kagome layer. b, Tight-binding band model of the single- spin–orbit coupling strength λ = 0.05t for b and d. e, Anomalous Hall
layer kagome lattice with (red) and without (blue) spin–orbit coupling. conductivity σxy as a function of Fermi energy EF from the k · p model
The inset shows a magnified view of the (avoided) crossing near K. c, C2x (black solid curve; see text). The red and blue dashed curves represent the
and C3z symmetries of the ‘breathing’ kagome (illustrated by the red and contributions from the upper and lower Dirac bands, respectively. The
blue inequivalent bonds) and stanene layers. d, Tight-binding band model inset shows the fit (solid lines) of a massive Dirac dispersion to the ARPES
of the double-layer kagome lattice with in-plane hopping t and inter-plane results (circles) near K.

2 9 m ar c h 2 0 1 8 | V O L 5 5 5 | N A T U R E | 6 4 1
© 2018 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
 RESEARCH Letter

simplified structures with a single kagome layer per unit cell or pur- 20. Kida, T. et al. The giant anomalous Hall effect in the ferromagnet
suing 2D morphologies via thin films or exfoliation of related materi- Fe3Sn2—a frustrated kagome metal. J. Phys. Condens. Matter 23, 112205
(2011).
als with van der Waals bonding. Furthermore, the frustrated hopping 21. Wang, Q., Sun, S., Zhang, X., Pang, F. & Lei, H. Anomalous Hall effect in a
network of kagome lattices is predicted to support electronic bands ferromagnetic Fe3Sn2 single crystal with a geometrically frustrated Fe bilayer
(as in Fig. 4d) with non-zero Chern number and enhanced electronic kagome lattice. Phys. Rev. B 94, 075135 (2016).
22. Tian, Y., Ye, L. & Jin, X. Proper scaling of the anomalous Hall effect. Phys. Rev.
correlation5. Searching for these bands by chemically doping Fe3Sn2, Lett. 103, 087206 (2009).
by using alternative structures and stoichiometries in AxSny or by using 23. Shitade, A. & Nagaosa, N. Anomalous Hall effect in ferromagnetic
other spin–orbit-coupled 4d and 5d kagome metals are key to realizing metals: role of phonons at finite temperature. J. Phys. Soc. Jpn 81, 083704
(2012).
new, magnetically driven fractionalized phases of matter11,12. 24. Kim, K. S. et al. Coexisting massive and massless Dirac fermions in symmetry-
broken bilayer graphene. Nat. Mater. 12, 887–892 (2013).
Online Content Methods, along with any additional Extended Data display items and
25. Sales, B. C., Saparov, B., McGuire, M. A., Singh, D. J. & Parker, D. S.
Source Data, are available in the online version of the paper; references unique to
Ferromagnetism of Fe3Sn and alloys. Sci. Rep. 4, 7024 (2014).
these sections appear only in the online paper.
26. Chen, Y. L. et al. Massive Dirac fermion on the surface of a magnetically doped
topological insulator. Science 329, 659–662 (2010).
received 26 September 2017; accepted 19 January 2018.
27. Xu, S. Y. et al. Hedgehog spin texture and Berry’s phase tuning in a magnetic
Published online 19 March 2018. topological insulator. Nat. Phys. 8, 616–622 (2012).
28. Balog, R. et al. Bandgap opening in graphene induced by patterned hydrogen
1. O’Keeffe, M. & Hyde, B. G. Crystal Structures. I. Patterns and Symmetry Ch. 5 adsorption. Nat. Mater. 9, 315–319 (2010).
(Mineralogical Society of America, 1996). 29. Ishii, Y., Harima, H., Okamoto, Y., Yamaura, J. & Hiroi, Z. YCr6Ge6 as a candidate
2. Sachdev, S. Kagome- and triangular-lattice Heisenberg antiferromagnets: compound for a kagome metal. J. Phys. Soc. Jpn 82, 023705 (2013).
ordering from quantum fluctuations and quantum-disordered ground states 30. Kane, C. L. & Mele, E. J. Quantum spin Hall effect in graphene. Phys. Rev. Lett.
with unconfined bosonic spinons. Phys. Rev. B 45, 12377–12396 (1992). 95, 226801 (2005).
3. Han, T.-H. et al. Fractionalized excitations in the spin-liquid state of a 31. Haldane, F. D. M. Model for a quantum Hall effect without Landau levels:
kagome-lattice antiferromagnet. Nature 492, 406–410 (2012). condensed-matter realization of the “parity anomaly”. Phys. Rev. Lett. 61,
4. Zhou, Y., Kanoda, K. & Ng, T.-K. Quantum spin liquid states. Rev. Mod. Phys. 89, 2015–2018 (1988).
025003 (2017).
5. Mazin, I. I. et al. Theoretical prediction of a strongly correlated Dirac metal.
Acknowledgements We are grateful to X.-G. Wen and E. Tang for discussions.
Nat. Commun. 5, 4261 (2014).
This research was funded in part by the Gordon and Betty Moore Foundation
6. Guo, H.-M. & Franz, M. Topological insulator on the kagome lattice. Phys. Rev. B
EPiQS Initiative, grant GBMF3848 to J.G.C. and NSF grant DMR-1554891. L.Y.,
80, 113102 (2009).
J.L. and F.v.C. acknowledge support by the STC Center for Integrated Quantum
7. Xu, G., Lian, B. & Zhang, S.-C. Intrinsic quantum anomalous Hall effect in the
Materials, NSF grant number DMR-1231319. L.Y. acknowledges support by the
kagome lattice Cs2LiMn3F12. Phys. Rev. Lett. 115, 186802 (2015).
Tsinghua Education Foundation. M.K. acknowledges a Samsung Scholarship
8. Nakatsuji, S., Kiyohara, N. & Higo, T. Large anomalous Hall effect in a non-
from the Samsung Foundation of Culture. This research used resources of the
collinear antiferromagnet at room temperature. Nature 527, 212–215 (2015).
Advanced Light Source, which is a DOE Office of Science User Facility under
9. Nayak, A. K. et al. Large anomalous Hall effect driven by a nonvanishing Berry
contract number DE-AC02-05CH11231. A portion of this work was performed
curvature in the noncolinear antiferromagnet Mn3Ge. Sci. Adv. 2, e1501870 (2016).
at the National High Magnetic Field Laboratory, which is supported by NSF
10. Kuroda, K. et al. Evidence for magnetic Weyl fermions in a correlated metal.
cooperative agreement number DMR-1157490, the State of Florida and the US
Nat. Mater. 16, 1090–1095 (2017).
Department of Energy.
11. Tang, E. & Wen, X.-G. High-temperature fractional quantum Hall states.
Phys. Rev. Lett. 106, 236802 (2011).
12. Bergholtz, E. J., Liu, Z., Trescher, M., Moessner, R. & Udagawa, M. Topology and Author Contributions L.Y., T.S. and C.R.W. grew the single crystals. L.Y.
interactions in a frustrated slab: tuning from Weyl semimetals to C > 1 characterized the materials, performed the transport and magnetic
fractional Chern insulators. Phys. Rev. Lett. 114, 016806 (2015). measurements and analysed the resultant data. M.K., C.J., A.B. and E.R.
13. Wallace, P. R. The band theory of graphite. Phys. Rev. 71, 622–634 (1947). performed the ARPES experiment and analysed the resultant data. J.L. and L.Y.
14. Thouless, D. J., Kohmoto, M., Nightingale, M. P. & den Nijs, M. Quantized Hall performed the theoretical calculations. F.v.C. and D.C.B. performed the electron
conductance in a two-dimensional periodic potential. Phys. Rev. Lett. 49, microscopy study. All authors contributed to writing the manuscript. L.F., R.C.
405–408 (1982). and J.G.C. supervised the project.
15. Chang, C.-Z. et al. Experimental observation of the quantum anomalous Hall
effect in a magnetic topological insulator. Science 340, 167–170 (2013).
Author Information Reprints and permissions information is available at
16. Nagaosa, N., Sinova, J., Onoda, S., MacDonald, A. H. & Ong, N. P. Anomalous
www.nature.com/reprints. The authors declare no competing financial
Hall effect. Rev. Mod. Phys. 82, 1539–1592 (2010).
interests. Readers are welcome to comment on the online version of the paper.
17. Giefers, H. & Nicol, M. High pressure X-ray diffraction study of all FeSn
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional
intermetallic compounds and one FeSn solid solution. J. Alloys Compd. 422,
claims in published maps and institutional affiliations. Correspondence and
132–144 (2006).
requests for materials should be addressed to R.C. (rcomin@mit.edu) or
18. Le Caër, G., Malaman, B. & Roques, B. Mössbauer effect study of Fe3Sn2.
J.G.C. (checkelsky@mit.edu).
J. Phys. F 8, 323–336 (1978).
19. Hou, Z. et al. Observation of various and spontaneous magnetic Skyrmionic
bubbles at room temperature in a frustrated kagome magnet with uniaxial Reviewer Information Nature thanks E. Bergholtz, B. Lake and O. Rader for their
magnetic anisotropy. Adv. Mater. 29, 1701144 (2017). contribution to the peer review of this work.

6 4 2 | N A T U R E | V O L 5 5 5 | 2 9 m ar c h 2 0 1 8
© 2018 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
 Letter RESEARCH

Methods We measured the anomalous Hall effect at temperature T = 300 K in multiple

Single-crystal growth. Single crystals of Fe3Sn2 were grown using an I2-catalysed samples. We find a consistently enhanced anomalous Hall conductivity in the
reaction. A stoichiometric ratio of Fe and Sn powders was sealed in a quartz tube kagome plane σ Axy relative to out of the kagome planeσ Azx . For samples C1, C4 and
with approximately 1% I2 by mass and kept in a horizontal three-zone furnace C5, the observed value of σ Axy is 163.6 Ω −1 cm −1 , 179.1 Ω −1 cm −1 and
with a temperature gradient from 750 °C to 650 °C for five weeks followed by 138.8 Ω−1 cm−1, respectively. For samples C2, C4, C5 and C6, the observed value
water quenching to stabilize the Fe3Sn2 phase. Hexagonal, plate-like crystals of sub- of σ Azx is 20.5 Ω−1 cm−1, 22.0 Ω−1 cm−1, 55.6 Ω−1 cm−1 and 53.7 Ω−1 cm−1, respec-
millimetre size (Extended Data Fig. 1a, inset) formed near the high-temperature tively. The T dependence for C1 and C4 is shown in Fig. 2.
region as has been reported previously32 for Fe3Ge. The hexagonal surfaces were Energy–momentum dispersion along high-symmetry directions. Extended Data
confirmed as (0001) kagome planes by using single-crystal X-ray diffraction. Fig. 3a, e shows the experimental Fermi surface of Fe3Sn2 obtained from different
Magnetization measurements. Magnetization measurements were performed orientations of the crystal axis with respect to the photoelectron emission plane.
using a commercial superconducting quantum interference device (SQUID) Extended Data Fig. 3b–d, f, g shows the experimental band dispersion of Fe3Sn2
magnetometer with the field oriented along the c axis and in the a–b plane. along high-symmetry directions. Despite the occurrence of a complex-matrix-el-
Demagnetization corrections were performed for all measurements. The ements effect near the K point, the structure of the two interpenetrating Dirac
measured saturation moment (1.9μB per Fe atom) is consistent with previous cones is clearly visible in all momentum directions probed in this experiment.
reports18,19,33. In Extended Data Fig. 3h, energy distribution curves at different Brillouin-zone
Transport measurements. Four probe transport measurements were performed corner points are shown. All energy distribution curves show a consistent two-peak
for longitudinal and Hall resistivity in a commercial cryostat with a superconduct- structure near the first Dirac point, signalling a gap opening. The velocity of the
ing magnet. High-field transport measurements in fields of up to 31 T were per- Dirac fermions vD is found to be isotropic in the kagome plane, with magnitude
formed in a He-3 cryostat at Cell-9 of the National High Magnetic Field Laboratory. vD = (1.76 ± 0.11) × 105 m s−1, comparable to that observed recently in iron pnic-
For measurements in the kagome plane, the field was applied along the [0001] tide35 and selenide36 superconductors, but lower than in graphene37,38, possibly
direction with current and voltages in the kagome plane. For Hall measurements reflecting the correlated character of the Fe-3d states.
perpendicular to the kagome plane, the magnetic field and current were applied Photon-energy-dependent ARPES. Photon-energy-dependent ARPES probes
orthogonally in the kagome plane and the out-of-plane voltage is measured. The the variation of the band structure along the momentum direction perpendicular
choice of coordination for in-plane and out-of-plane Hall measurements is such to the sample surface (the kz direction in our geometry)39. The experiment here
that, for the ordinary Hall effect of holes, σxy > 0 when the field is applied along was conducted by varying the photon energy from 55 eV to 120 eV. The ARPES
the +z direction and σzx > 0 (σxz < 0) when the field is applied along the +y spectral weight of the localized d electrons is suppressed at low photon energy,
direction. Electrical signals were detected using standard AC lock-in techniques so we report only the results from 90 eV to 120 eV. Considering the large c-axis
with a typical current density of 10 A cm−2. To correct for contact misalignment, lattice constant (c = 19.8 Å), the photon energy range presented here spans more
the measured longitudinal and transverse voltages were field-symmetrized and than three complete Brillouin zones. The dependence on photon energy varies
field-antisymmetrized, respectively. Demagnetization corrections were performed substantially between different bands. For example, the dispersion of the innermost
for all measurements. electron pocket at Γ (Extended Data Fig. 4b) depends critically on the photon
Scanning transmission electron microscopy (STEM). STEM experiments were energy (kz), whereas the dispersion of Dirac bands shows negligible dependence
conducted at a probe-corrected STEM (JEOL ARM) operated at an acceleration on kz. The latter finding demonstrates the pronounced 2D nature of the Dirac
voltage of 200 kV. Fe3Sn2 samples were prepared by a Helios focused-ion beam fermions that we observed. We ascribe the highly dispersive bands centred at Γ to
(FEI), operated at an acceleration voltage of 30 kV for the gallium beam during the 3D network of Sn atoms within the crystal structure.
lift-out and of 2 kV during polishing. Additional polishing was performed at 1 kV Tight-binding models. We use a simple 2D tight-binding model to capture the
and 0.5 kV with a NanoMill (Fischione). At both acceleration voltages, samples symmetry-protected Dirac nodes at K and K′ in the single-layer (Fig. 4a) and bilayer
were polished for 20 min on each side. (Fig. 4c) kagome lattice. Starting with the former, in the absence of spin–orbit cou-
Angle-resolved photoemission spectroscopy (ARPES). ARPES experiments were pling, the nearest-neighbour tight-binding model HK (equation (1)) yields Dirac
conducted at the Microscopic and Electronic Structure Observatory (MAESTRO) bands and a dispersionless band (Fig. 4b). For the bilayer kagome lattice, we include
at beamline 7 (main data) and at the meV Resolution Soft X-ray Inelastic Scattering additional hopping t0 = 0.3t between the vertically displaced sites on each kagome
Beamline (MERLIN) at beamline 4 (preliminary measurement) of the Advanced layer, which introduces a layer splitting of the Dirac states in energy (Fig. 4d)29.
Light Source. The two ARPES endstations are equipped with R4000 hemispherical We examine the effect of the spin–orbit interaction by adding a Kane–Mele-type
electron analysers (VG scienta, Sweden). Data in Fig. 3 and Extended Data Fig. 3 term30 HSOI (equation (2)) in the leading-order nearest-neighbour hopping to
were collected at 20 K with a photon energy of 92 eV, which maximizes the ARPES the tight-binding model HK with strength given by λ(Eij × Rij) · s. The magnetic
spectral weight of the Dirac bands. The photon-energy-dependent measurement moments in Fe3Sn2 have been proposed to be subjected to spin reorientation and
was conducted from 45 eV to 120 eV (Extended Data Fig. 4). Energy and momen- microscopic domain formation19,33; contributions to λij arise from orthogonality
tum resolutions were better than 15 meV and 0.01 Å−1, respectively. Fe3Sn2 samples of the hopping path and local electric-field and spin directions in each domain. In
were cleaved in the ultrahigh-vacuum chamber with a base pressure of better than the simplified hopping model, these contributions are represented by an in-plane
4 × 10−11 torr. All of the data were collected within 8 h after cleaving to minimize electric field and sz, with λij = 0.05t in Fig. 4b, d. Similar Hamiltonians40,41 have
the effect of sample degradation. Six different samples from various growth batches been used in the spin sector for insulating materials with kagome structures40–43.
were analysed to confirm the consistency of results. Model calculation of Hall conductivity. To connect with the Hall response, we
Longitudinal electrical transport. The resistivity in the kagome plane of sample first construct a k · p Hamiltonian near K and K′ for the dual Dirac fermions
C1 as a function of temperature, ρ(T), is shown in Extended Data Fig. 1a. A metallic observed in the ARPES spectra:
response is seen at all T, with a residual resistivity ratio of ρ(300 K)/ρ(2 K) = 25.
The magnetoresistance for magnetic induction B normal to and within the kagome HD = [ħv F(kxσy − k yσx)] ⊗ I + E 0τx + mσz (3)
plane is shown in Extended Data Fig. 1b, c, respectively. In both figures, the electri-
cal current I is perpendicular to the applied field. For B  c (Extended Data Fig. 1b), where σi (i = x, y, z) are the Pauli matrices of pseudospin for each Dirac band, E0
we observe a non-monotonic response below the saturation field (B ≤ 1.2 T), is the energy splitting of the Dirac bands described by the Pauli matrix τx, and
which may reflect a transition through a skyrmion bubble phase, as was reported m = Δ/2 is the Dirac mass. To obtain the band parameters, we fit the observed
recently19. This structure is absent for B ⊥ c (Extended Data Fig. 1c). In addition, dispersion E(k) to the massive Dirac model Ei±(k) = ± (ħkv D) 2 + (Δ/2) 2 + Ei0   ,
for both B  c and B ⊥ c, a negative linear magnetoresistance characteristic of mag- where Ei0 is the energy offset of the upper (i = 1) and lower (i = 2) Dirac bands
non suppression34 is observed at high T. from EF. As shown in the inset of Fig. 4e, a satisfactory fit is found with
Analysis of the Hall effect. In a ferromagnetic metal, the total Hall conductivity v D = (1.85 ± 0.15) × 10 5 m s −1, Δ = 32 ± 3 meV, E10 = −73 ± 5 meV and
is composed of contributions from the normal Hall effect (‘N’) induced by Lorentz E02 = −182 ± 5 meV. The Dirac band centred at 73 meV can be analysed reliably,
force and the anomalous Hall effect (‘A’): σxy = σ N A
xy + σ xy. In the limit of small Hall whereas matrix-element effects and the proximity of neighbouring bands interfere
2
angle (Θ H ≡ ρyx/ρxx  1), we have σxy ≈ B/(neρ xx), where n is the carrier density with the intensity distribution of the lower Dirac point. Applying the same set of
and e is the electronic charge. Given the relatively small magnetoresistance, we Fermi velocity and mass for the upper and lower Dirac bands, we then calculate
extract σ Nxy from a linear fit of σxy(B) for fields above saturation (1.7 T to 5 T), with the contribution of the massive Dirac bands to the Hall response by integrating the
the intercept returning σ Axy (Extended Data Fig. 2a). That the condition Θ H  1 is Berry curvature over the filled states described by equation (3) as follows.
satisfied is confirmed to high field, with Θ H  0.04 for fields up to 30 T (Extended The Hall conductivity σxy can be considered a geometric quantity that charac-
Data Fig. 2b). terizes the mapping from the kx–ky momentum plane to the two-component

© 2018 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
 RESEARCH Letter

Dirac band structure44. For a general two-level Hamiltonian d(k) · σ, (where equation (3) and adding up the contributions from the upper and lower Dirac
σ = {σ x, σy, σz} and d(k) is a generic vector), the wave functions may be points to the anomalous Hall conductivity described by equations (4) and (5), we
represented as points on the unit Bloch sphere by ± d̂(k) = d(k)/ d(k) , where obtain the energy-dependent 2D Hall conductance σ Axy (Fig. 4e). The Hall conduc-
the ± denotes the two eigenstates of the Hamiltonian at a given k. σxy then takes tivity at the Fermi level evaluates to σxy(EF) = (0.31 ± 0.05)e2/h. The uncertainty
the form44 within this model arises from fitting the experimental band parameters near K;
developing models that use the complete electronic structure and Berry curvature
e2
σxy =
h
∬ dˆ ⋅ (∂x dˆ × ∂y dˆ)dkx dk y (4)
to compare to the experimental results is an important future direction.
Data availability. The data that support the findings of this study are available
filled states
from the corresponding authors on reasonable request.
where the integrand (the Berry curvature) can be seen as the Jacobian associated
with mapping the kx–ky plane to the unit sphere. Therefore, σxy is proportional to
the total area covered on the unit sphere by the filled states. Also, because the 32. Drijver, J. W., Sinnema, S. G. & van der Woude, F. Magnetic properties of
integrand in equation (4) is a pseudoscalar, the states represented by ±d̂ have hexagonal and cubic Fe3Ge. J. Phys. F 6, 2165–2177 (1976).
33. Fenner, L. A., Dee, A. A. & Wills, A. S. Non-collinearity and spin frustration in the
opposite contributions to σxy. itinerant kagome ferromagnet Fe3Sn2. J. Phys. Condens. Matter 21, 452202
The above formulation for σxy of a single Dirac fermion is illustrated in Extended (2009).
Data Fig. 5 for the gapless and gapped cases. For the former (Extended Data 34. Raquet, B., Viret, M., Sondergard, E., Cespedes, O. & Mamy, R. Electron-magnon
Fig. 5a), the wavefunction of the Dirac fermions is confined to the equator of the scattering and magnetic resistivity in 3d ferromagnets. Phys. Rev. B 66,
Bloch sphere and therefore σxy = 0. Extended Data Fig. 5b shows the case in which 024433 (2002).
35. Richard, P. et al. Observation of Dirac cone electronic dispersion in BaFe2As2.
the lower branch of the massive Dirac cone is filled. In this case, the occupied states
Phys. Rev. Lett. 104, 137001 (2010).
span the lower hemisphere and yield σxy = 0.5e2/h. When the upper branch of the 36. Tan, S. Y. et al. Observation of Dirac cone band dispersions in FeSe thin films by
gapped Dirac fermion is partially filled (Extended Data Fig. 5c), a contribution photoemission spectroscopy. Phys. Rev. B 93, 104513 (2016).
of opposite sign appears. The resulting σxy for a single Dirac fermion is therefore 37. Bostwick, A., Ohta, T., Seyller, T., Horn, K. & Rotenberg, E. Quasiparticle
dynamics in graphene. Nat. Phys. 3, 36–40 (2007).
 e2 S
 38. Nevius, M. S. et al. Semiconducting graphene from highly ordered substrate

 for S ≤ 2π interactions. Phys. Rev. Lett. 115, 136802 (2015).
σxy =  h 4π 39. Damascelli, A., Hussain, Z. & Shen, Z.-X. Angle-resolved photoemission studies
 2

 e 4π − S of the cuprate superconductors. Rev. Mod. Phys. 75, 473–541 (2003).
 for S ≥ 2π 40. Onose, Y. et al. Observation of the magnon Hall effect. Science 329, 297–299

 h 4π
 (2010).
where S is the total area of the filled states on the Bloch sphere; this is shown in 41. Chisnell, R. et al. Topological magnon bands in a kagome lattice ferromagnet.
Phys. Rev. Lett. 115, 147201 (2015).
Extended Data Fig. 5d as a function of energy E normalized by hvF (where vF is 42. Inami, T., Nishiyama, M., Maegawa, S. & Oka, Y. Magnetic structure of the
the Fermi velocity). kagome lattice antiferromagnet potassium jarosite KFe3(OH)6(SO4)2.
The above formulation is consistent with that expected from the semi-classical Phys. Rev. B 61, 12181–12186 (2000).
Boltzmann equation and Kubo formalism for a single massive Dirac fermion45: 43. Hiroi, Z. et al. Spin-1/2 kagome-like lattice in volborthite Cu3V2O7(OH)2 ∙ 2H2O.
J. Phys. Soc. Jpn 70, 3377–3384 (2001).
e2 Δ/2 44. Qi, X. L., Wu, Y.-S. & Zhang, S.-C. Topological quantization of the spin Hall effect
σxy = (5) in two-dimensional paramagnetic semiconductors. Phys. Rev. B 74, 085308
h (Δ/2) 2 + (ħv Fk F) 2 (2006).
45. Sinitsyn, N. A., MacDonald, A. H., Jungwirth, T., Dugaev, V. K. & Sinova, J.
where Δ is the size of the Dirac gap and kF is the Fermi wavevector. From Anomalous Hall effect in a two-dimensional Dirac band: the link between the
equation (5) and Extended Data Fig. 5c, we see that σxy is maximized to e2/(2h) Kubo-Streda formula and the semiclassical Boltzmann equation approach.
when EF is within the gap and drops outside the gap with a long tail46,47. Phys. Rev. B 75, 045315 (2007).
46. Haldane, F. D. M. Berry curvature on the Fermi surface: anomalous Hall effect
Importantly, the massive Dirac fermions at the K and K′ valleys are related by as a topological Fermi-liquid property. Phys. Rev. Lett. 93, 206602 (2004).
inversion symmetry and therefore contribute similarly to the Berry curvature; the 47. Fang, Z. et al. The anomalous Hall effect and magnetic monopoles in
contribution from equation (5) is therefore doubled. With the experimental fits to momentum space. Science 302, 92–95 (2003).

© 2018 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
 Letter RESEARCH

Extended Data Figure 1 | Metallic transport in Fe3Sn2. a, Resistivity b, c, Magnetoresistance (defined as MR = [ρxx(B) − ρxx(0)]/ρxx(0)) at
ρ as a function of temperature T in the kagome plane for Fe3Sn2 selected T with B applied perpendicular (b) or parallel (c) to the kagome
sample C1. The inset shows a photograph of Fe3Sn2 single crystals. plane and B ⊥ I (schematics of the configurations are shown as insets).

© 2018 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
RESEARCH Letter

Extended Data Figure 2 | Extracting anomalous Hall conductivity and linear fit to σ N
xy . The data at 2 K and 50 K have been scaled by the factors
high-field transport. a, In-plane Hall conductivity σxy as a function of shown for clarity. b, Magnetoresistance (main panel) and Hall effect (inset)
magnetic induction B at selected temperatures. Dashed lines represent the of Fe3Sn2 with applied magnetic field m0H  c up to 31 T.

© 2018 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
 Letter RESEARCH

Extended Data Figure 3 | Momentum and energy-dependent band in d shows the raw data of Fig. 3c (with the same energy and momentum
structure along high-symmetry directions. a, e, Fermi surface of Fe3Sn2 range), highlighting the spectral weight distribution near the Dirac
obtained from different experimental geometries. b–d, f, g, Band points. h, Energy distribution curves at different K points indicated in
dispersion of Fe3Sn2 along high-symmetry directions. The panels c, d, f and g. The curves are shifted along the vertical direction for clarity.
correspond to the momentum directions along the red (b), orange (c), The inset shows an example of Gaussian fits; the extracted gap size is
green (d), magenta (f) and purple (g) dotted lines in a and e. The inset Δ = 30 ± 5 meV.

© 2018 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
RESEARCH Letter

Extended Data Figure 4 | Photon-energy dependence of ARPES spectra.

ARPES intensity plot for Fe3Sn2 taken along the Γ–K direction as a
function of binding energy k and photon energy.

© 2018 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
 Letter RESEARCH

Extended Data Figure 5 | Berry curvature and Hall conductivity for a states for the gapless case (a) and the gapped case with EF in (b) and out of
massive Dirac fermion. a–c, Schematics of 2D Dirac fermions and the (c) the gap. d, Fermi energy EF dependence of σxy for the case of a single
corresponding Bloch-sphere representation of the wavefunction of filled massive Dirac fermion with gap Δ and Fermi velocity vF.

© 2018 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.