Double-sided van der Waals epitaxy of topological insulators

across an atomically thin membrane

Joon Young Park1,2, Young Jae Shin1, Jeacheol Shin2, Jehyun Kim2, Janghyun Jo3, Hyobin
Yoo1,4, Danial Haei1,5, Chohee Hyun6, Jiyoung Yun2, Robert M. Huber7,8, Arijit Gupta7,8, Kenji
Watanabe9, Takashi Taniguchi10, Wan Kyu Park8, Hyeon Suk Shin11,12, Miyoung Kim3, Dohun
Kim2, Gyu-Chul Yi2,*, and Philip Kim1,*
1
Department of Physics, Harvard University, Cambridge, MA 02138, USA
2
Department of Physics and Astronomy, and Institute of Applied Physics, Seoul National University, Seoul 08826,
Republic of Korea
3
Department of Materials Science and Engineering, and Research Institute of Advanced Materials, Seoul
National University, Seoul 08826, Republic of Korea
4
Department of Physics, Sogang University, Seoul 04107, Republic of Korea
5
Center for Nanoscale Systems, Harvard University, Cambridge, MA 02138, USA
6
Department of Chemistry, and Low Dimensional Carbon Materials Center, Ulsan National Institute of Science
and Technology (UNIST), Ulsan 44919, Republic of Korea
7
Department of Physics, Florida State University, Florida, 32306, USA
8
National High Magnetic Field Laboratory, Florida State University, FL 32310, USA
9
Research Center for Electronic and Optical Materials, National Institute for Materials Science, 1-1 Namiki,
Tsukuba 305-0044, Japan
10
Research Center for Materials Nanoarchitectonics, National Institute for Materials Science, 1-1 Namiki,
Tsukuba 305-0044, Japan
11
Department of Energy Science and Department of Chemistry, Sungkyunkwan University (SKKU), Suwon 16419,
Republic of Korea
12
Center for 2D Quantum Heterostructures, Institute of Basic Science (IBS), Sungkyunkwan University (SKKU),
Suwon 16419, Republic of Korea
*
e-mail: gcyi@snu.ac.kr; pkim@physics.harvard.edu

Atomically thin van der Waals (vdW) films provide a novel material platform for epitaxial
growth of quantum heterostructures. However, unlike the remote epitaxial growth of
three-dimensional bulk crystals, the growth of two-dimensional (2D) material
heterostructures across atomic layers has been limited due to the weak vdW interaction.
Here, we report the double-sided epitaxy of vdW layered materials through atomic
membranes. We grow vdW topological insulators (TIs) Sb2Te3 and Bi2Se3 by molecular
beam epitaxy on both surfaces of atomically thin graphene or hBN, which serve as
suspended 2D vdW "substrate" layers. Both homo- and hetero- double-sided vdW TI
tunnel junctions are fabricated, with the atomically thin hBN acting as a crystal-
momentum-conserving tunnelling barrier with abrupt and epitaxial interface. By
performing field-angle dependent magneto-tunnelling spectroscopy on these devices, we
reveal the energy-momentum-spin resonant tunnelling of massless Dirac electrons
between helical Landau levels developed in the topological surface states at the interface.

1
 Two-dimensional (2D) van der Waals (vdW) heterostructures composed of materials
with various electronic properties serve as novel platforms to explore a variety of interesting
physical properties and device applications1,2. One of the most intriguing systems is a vertical
heterostructure consisting of two materials separated by an atomically thin insulating barrier,
allowing vertical tunnelling transport through vdW interfaces3-5. Such a heterostructure with
high crystal quality and epitaxial alignment is of particular interest because crystal momentum
parallel to the interface can be conserved, offering a unique platform for investigating the
intrinsic selection rules of the resonant tunnelling process6-8. The surface states of three-
dimensional (3D) vdW topological insulators (TIs) of (Bi, Sb)2(Te, Se)3 compounds9, when
integrated into vdW tunnel junctions, can provide exciting opportunities to discover novel
quantum phenomena occurring at the atomically sharp epitaxial interface between
topologically dissimilar atomic layers10-15.
Multilayered TI/normal insulator vdW heterostructures have been constructed by
sequential direct growth using molecular beam epitaxy (MBE)10,16. In this conventional MBE
growth, the choice of the barrier materials is constrained by chemical compatibility and lattice
constant matching. Furthermore, the common occurrence of step edges during the growth
process significantly reduces the effectiveness of forming a smooth vdW heterointerface.
VdW epitaxy on both surfaces of an atomically thin, single-crystalline suspended vdW
membrane opens a new way to construct TI epitaxial heterostructures with atomically sharp
interfaces. This novel synthesis route presented in this work distinguished from the recently
demonstrated remote epitaxy of 3D/2D/3D hybrid systems. In remote epitaxial growth, the
electrostatic field of the substrate (lower 3D layer) penetrates the 2D vdW interlayer (Fig.
1a)17,18. Due to its strong dependence on the ionic character or polarity of the 3D substrate,
remote epitaxy establishes the long-range epitaxial relationship between 3D/3D bulks,
insensitive to the crystallographic alignment at the 3D/2D local interfaces. On the contrary, in
the newly proposed double-sided vdW epitaxy, as shown in Fig. 1b, one can utilize the short-
range local vdW interaction at the interfaces, structurally connecting the top and bottom
epilayers to the substrate (middle) layer. Moreover, unlike remote epitaxy where a stronger
coupling leads to a stronger correlation of the electronic structure across the heterostructure,
the electrical coupling can be tuned independently by the thickness and electronic properties
of the middle layer. To date, while 3D/2D/3D quasi-vdW epitaxial double heterostructures have
been demonstrated in the form of graphene sandwiched by semiconductor nanostructures with
reconstructed dangling bonds of polar facets at the interface19,20, all-2D double-sided vdW

2
Fig. 1. Double-sided vdW epitaxy of TI/hBN(graphene)/TI vertical heterostructures. a,b, Schematics
illustrating the growth mechanism of remote epitaxy (a) and double-sided vdW epitaxy (b). In remote epitaxy, the
middle 2D vdW layer serves as an atomically thin spacer to mediate the long-range interaction between 3D crystals.
In the new concept of double-sided vdW epitaxy, the middle vdW layer serves as an atomically thin growth
template where the top and bottom layers are epitaxially connected by the short-range vdW interaction. c–e,
Schematics of the double-sided vdW epitaxy procedure, where TIs are grown on the bottom and top surfaces of
atomically thin vdW substrates suspended on perforated SiNx membranes. f, Photograph of an array of SiNx
membrane windows compatible with MBE, TEM, and microfabrication. Inset: High-contrast optical microscope
image of an array of holes fabricated in the SiNx membrane covered by an atomically thin hBN crystal. Scale bar:
10 m. g–h, AFM topographic images (20 nm colour scale) of 2 ML hBN suspended over one of the holes in the
SiNx membrane taken before (g) and after (h) the first growth of Bi2Se3 on the back side, and after the second
growth of Bi2Se3 on the front side (i).

epitaxy has not yet been realized. In this Article, we present the double-sided vdW-MBE
growth of 2D/2D/2D structures where two epitaxially aligned TIs are separated by atomically
thin crystalline barriers with abrupt interfaces, which has enabled us to observe the resonant
tunnelling between Landau levels (LLs) developed in the topological surface states (TSSs) at
the vdW heterointerface where the energy, momentum, and spin of the helical Dirac fermions
are conserved during the tunnelling process.
The method developed to fabricate the double-sided epitaxial vdW heterostructures is
schematically illustrated in Fig. 1c–e. Ultrathin hBN layers (2–5 monolayers; MLs) or
monolayer graphene are suspended over hole arrays pre-patterned in SiNx membranes to
expose both surfaces (Fig. 1f and Extended Data Fig.1). Subsequently, Bi2Se3 (n-type) and
Sb2Te3 (p-type) thin films are heteroepitaxially grown on each side of the suspended layer using
ultra-high vacuum (UHV) MBE (see Methods). The suspended vdW layer serves as a
crystalline substrate for the epitaxy of the TI films. In particular, the atomically thin hBN layers
can serve as a wide bandgap insulating barrier for tunnelling spectroscopy between the two TIs
whose interface is sealed under UHV. Since the suspended vdW heterostructures are electron-
transmitting and mechanically robust, this templated growth approach enables us to directly
use plan-view TEM and atomic force microscopy (AFM) to study their structural properties

3
(Fig. 1g–i and Fig. 2) and the microfabrication processes to form electrical contacts to the
junctions21-24.
The microstructural properties of the TI/hBN(graphene)/TI heterostructures can be
investigated by TEM. We first demonstrate that bottom (b-) and top (t-) Bi2Se3 films are
epitaxially aligned through a suspended monolayer graphene, a single-atom-thick vdW crystal
substrate. Figure 2a shows the plan-view selected area electron diffraction (SAED) image of
the resulting heterostructure exhibiting the preferential epitaxial relationship of
1010 b-Bi2 Se3 ∥ 1010 graphene ∥ 1010 t-Bi2 Se3 . The weak Bragg peaks from Bi2Se3 with
small deviation angles can be attributed to the wrinkles formed in the suspended vdW layers
(inset of Fig. 2a and ref.23) and to grains with a small in-plane misorientation angle, which is
likely due to the weak vdW interactions at the interface22,25-27.
The same methodology can be applied to insulating hBN substrates, which is of
particular interest for vertical tunnel junctions. The cross-sectional HR-TEM image in Fig. 2b
shows the quintuple layer (QL) structures of the Bi2Se3 thin films, and the atomically clean and
abrupt interfaces between b-Bi2Se3, 2 ML hBN, and t-Bi2Se3 (see also Extended Data Fig. 2).
Their epitaxial relationship can be confirmed by comparing the fast Fourier transform (FFT)
of the representative areas (yellow and orange boxes) with the interfacial region (red box),
where one can find matching structures of b- and t-Bi2Se3 (white arrows) relative to the hBN
atomic registry (red arrows).
Next, we present the growth of dissimilar TI materials—namely Bi2Se3 and Sb2Te3—
across a suspended hBN, further demonstrating the power of our double-sided vdW epitaxy.
Figure 2c shows the plan-view HR-TEM image of the Sb2Te3/hBN/Bi2Se3 heterostructure,
taken near the edge of a hole in the SiNx membrane so that there is a region where only Bi2Se3
is grown on hBN (first growth on the top side) without the coverage of Sb2Te3 (second growth
on the bottom side). While the moiré patterns arise from the lattice mismatch between Sb2Te3
and Bi2Se3 (~2.8%) (refs.28,29) and their small in-plane misorientations (Fig. 2d), the FFT
patterns of the selected areas of the image (Fig. 2d,e) identify the epitaxial alignment between
Bi2Se3 and hBN as well as the epitaxial alignment between the Bi2Se3 and Sb2Te3 films through
the hBN substrate: 1010 Bi2 Se3 ∥ 1010 hBN ∥ 1010 Sb2 Te3 . The epitaxial relationship in our
bidirectional MBE on suspended hBN and graphene is consistent with that of the TI thin films
grown on a single surface of hBN and graphene substrates (Extended Data Fig. 3 and ref.22,27),
suggesting that the epitaxial alignment is mainly due to the interaction with the suspended vdW
substrate layer.
4
Fig. 2. Structural properties of TI/hBN(graphene)/TI double-sided vdW epitaxial heterostructures. a, Plan-
view SAED pattern of Bi2Se3/monolayer graphene/Bi2Se3. Inset: Plan-view bright-field (BF) TEM image of the
heterostructure. The dashed yellow circle and the red circle indicate the suspended area (a hole in the SiNx
membrane) and the area in which the SAED is recorded, respectively. Scale bar: 200 nm. b, Cross-sectional HR-
TEM image of Bi2Se3/2ML hBN/Bi2Se3 taken along the zone axis 1210 . Insets: FFT patterns of the areas
corresponding to the yellow, red, and orange boxes in the main panel. c, Plan-view HR-TEM image of Sb2Te3/few-
layer hBN/Bi2Se3 taken around the boundary of Sb2Te3 (dashed yellow line). d,e FFTs of the regions marked by
the red (d) and orange (e) boxes in c. f, Cross-sectional HR-TEM image of Bi2Se3/2ML hBN/Sb2Te3 taken along
the zone axis 1100 . Inset: Cross-sectional STEM image overlaid with EDS map. g,h FFTs of the regions marked
by cyan (g) and orange (h) boxes in f.

The high interfacial quality of the heterostructure is also confirmed by the cross-
sectional HR-TEM (Fig. 2f). Since the strong and dense in-plane chemical bonding in the hBN
layers provides a completely impermeable atomic membrane, we do not expect any
interdiffusion of atoms through the hBN substrate during the growth of the second layers on
the opposite side of the surface after the first layer growth. The inset of Fig. 2f shows cross-
sectional chemical composition analysis using scanning tunnelling electron microscopy
(STEM) equipped with energy-dispersive X-ray spectroscopy (EDS). The signals of Sb and Te
are observed only above the hBN layer while those of Bi and Se are detected only below the
hBN layer, demonstrating the clear spatial separation of Sb2Te3 and Bi2Se3 across the two-
atom-thick hBN barrier. The epitaxial relationship across the hBN substrate can be again
confirmed by comparing the FFT patterns of Bi2Se3 and Sb2Te3. As shown in Fig. 2g,h, the
112𝑙 and 224𝑙 spots of Bi2Se3 (white arrows in Fig. 2g) align well with the
5
corresponding spots of Sb2Te3 (white arrows in Fig. 2h).
With the hBN barrier layer, our double-sided vdW epitaxy provides TI/hBN/TI
tunnelling devices, enabling the study of electron tunnelling properties of helical Dirac
fermions through the tunnelling barrier. To perform tunnelling spectroscopy, individual
electrical contacts to an array of suspended junctions are fabricated on the top side while they
share a common bottom contact (Fig. 3a; see also Extended Data Fig. 4 and Methods). A DC
bias voltage V is applied to the top electrode with respect to the bottom electrode, and the
tunnelling current I can be measured across the junction. Although we have fabricated
tunnelling devices on Bi2Se3/hBN/Bi2Se3 as well (results in Extended Data Fig. 5), in the
remainder of this paper, we focus on the Bi2Se3 (bottom)/hBN/Sb2Te3 (top) heterostructure,
whose configuration yields n-TI/insulator/p-TI junctions.
The Bi2Se3/hBN/Sb2Te3 heterojunction at zero bias is expected to form a type-III band
alignment with a broken gap due to the large difference in their electron affinities and relatively
small bulk band gaps (Fig. 3b)30,31. Accordingly, charge transfer and band bending result in
carrier accumulation near the junction. Away from the heterointerface, the Fermi level EF of
Bi2Se3 (Sb2Te3) films is located near the edge of the bulk conduction (valence) band due to the
unintentional n-type (p-type) doping22,31,32. The TSSs at the interface are hence electrically
contacted by the bulk states.
Figure 3c shows J(V), the current density J as a function of the applied bias voltage V,
measured in devices with different hBN barriers ranging from 2 to 5 ML. The measured
differential conductance at zero bias, normalized by the junction area 𝐺 𝜕𝐼/𝜕𝑉 𝑉 0 ,
scales exponentially with the thickness of the insulating hBN barrier (inset of Fig. 3c),
suggesting that electron transport in our devices is dominated by tunnelling through hBN
barriers33,34. For a negative bias voltage, |J(V)| monotonically increases rapidly with increasing
|V|. However, for the positive bias side, J(V) increases non-monotonically as V increases,
exhibiting negative differential conductance-like characteristics at a certain bias voltage range.
These features are consistent with the characteristics of tunnel diodes formed by p–n junctions
with broken bandgap alignments35.
The detailed tunnelling process in the junction can be described in the differential
conductance 𝐺 𝑉 𝜕𝐼/𝜕𝑉 curve for the device with a 2 ML hBN barrier (Fig. 3d).
Schematic band alignments at the heterointerface under different bias voltages are illustrated
together, as labeled by (i) – (iv). The differential conductance exhibits a strongly reduced value
between bias voltage regions (i) and (iv), indicating that the Fermi level of n-Bi2Se3 (𝐸 )

6
Fig. 3. Tunnelling spectroscopy in Bi2Se3/hBN/Sb2Te3 junctions at zero magnetic field. a, False-coloured low-
magnification cross-sectional BF TEM image of a vertical tunnel junction device in two-terminal measurement
configuration. While this image shows Bi2Se3/hBN/Bi2Se3, the device structure is identical for all measured
junctions, including Bi2Se3/hBN/Sb2Te3, except for the hBN barrier thickness and the junction area. b, Schematic
band alignment of the Bi2Se3/hBN/Sb2Te3 junction at zero bias voltage, depicted in energy – out-of-plane axis
plane. c, Tunnelling current density as a function of bias voltage for the devices with different thicknesses of hBN
barrier. V is applied to Sb2Te3 (top), and the current drained through Bi2Se3 (bottom), I, is measured. Inset: Zero-
bias differential conductance per unit area as a function of the number of hBN layers. d, Differential conductance
of the Bi2Se3/2ML hBN/Sb2Te3 device as a function of V. (i) – (iv): Schematic band alignments at the tunnelling
interface under different bias conditions, plotted in the energy – in-plane momentum plane. TSSs are marked by
red and blue colours, indicating their spin-momentum locking textures. The tunnelling conductance is determined
by the position of the bulk conduction band (BCB) and the bulk valence band (BVB) relative to the 𝐸 and
𝐸 . Devices with 2 and 3 ML hBN barriers are measured at temperature T = 1.8 K and those with 4 ML and 5
ML barriers are measured at T = 1.5 K.

crosses the bulk band gap of p-Sb2Te3 as V is increased in this region. As the Fermi level of p-
Sb2Te3 (𝐸 ) moves out of the BCB of n-Bi2Se3 and into its bulk band gap ((ii) and (iii)), the
G(V) exhibits a peak at V = 225 mV. This local maximum can be attributed to the enhanced
tunnelling density of states (DOS) when the bulk band edges align in the presence of band
bending36-38. At V ≲ 40 mV and V ≳ 350 mV, oscillatory spectral features appear due to the
bulk quantum well states (QWSs)38-40.

7
 The topological nature of the coupled surface states in our vdW epitaxial structure can
be revealed by the magneto-tunneling transport study. When magnetic field perpendicular to
the junction, B⟂, is applied, LLs form in the TSSs, which can modulate the tunnel conductance
associated with the TSSs40-43. Figure 4a shows the differential conductance spectra
𝐺 𝑉, 𝐵⟂ 𝜕𝐼/𝜕𝑉 𝑉, 𝐵⟂ of the Sb2Te3/2ML hBN/Bi2Se3 device at different B⟂. As B⟂
increases, the tunnelling conductance spectrum exhibits additional oscillatory features. A more
systematic analysis can be performed by subtracting off the zero magnetic field conductance:
𝐺 𝑉, 𝐵⟂ 𝜕𝐼/𝜕𝑉 𝑉, 𝐵⟂ 𝜕𝐼/𝜕𝑉 𝑉, 𝐵⟂ 0 . Figure 4b shows G as a function of V and
B⟂. We observe that G(V) oscillations exhibit larger amplitudes at higher B⟂. These oscillatory
features can be categorized into three groups: oscillations with peak and valley positions (I)
that are independent of B⟂; (II) that vary with B⟂ and converges to V ~135 mV as B⟂ vanishes;
and (III) that vary with B⟂ and converges to a large bias voltage (V > 450 mV) as we extrapolate
the position of the origin.
To further investigate the nature of G oscillation, we tilt the magnetic field, B, and
provide a component of the magnetic field parallel to the junction, B∥. Figure 4b–e shows G(V,
B) at different B⟂ and B∥. We find that the oscillatory feature (I) depends only on the total
magnetic field B, indicating a 3D isotropic nature, while features (II) and (III) depend only on
B⟂, suggesting a 2D origin. We attribute the feature (I) to the bulk QWSs due to its angle-
insensitivity and the bias voltage ranges in which the peaks appear (see Extended Data Fig. 6).
The 2D oscillatory features (II) and (III) can be related to the resonant tunnelling
associated with LL formation in the TSSs. In particular, the observed unequal spacing, which
increases with increasing B⟂, suggests that these LLs originate from the linearly dispersed
Dirac spectrum of the TSSs. In a perpendicular magnetic field, the Dirac dispersion of the TSSs
can lead to sharply quantized LLs at energy 𝐸 𝐸 sgn 𝑛 𝑣 2𝑒ℏ|𝑛|𝐵 , where 𝐸 is
the energy of Dirac point (DP), 𝑣 is the Fermi velocity of the TSSs, 𝑒 is the elementary
charge, ℏ is the reduced Planck constant, and 𝑛 0, 1, 2, … is the LL index, ignoring the
Zeeman energy40-43. As the bias voltage increases, V > 20 mV, the resonant tunnelling between
the BCB of Bi2Se3 and the LLs developed in the TSS of Sb2Te3 becomes possible, and indeed,
the oscillatory features (II) in the magneto-tunnelling spectra Fig. 4b–d follow 𝑒 𝑉 𝑉
sgn 𝑛 𝑣 2𝑒ℏ|𝑛|𝐵 , centered around V0 ≈ 135 mV. We identify the peaks of these oscillations,
Vn, and assign the LL indices n as shown in the upper part of Fig. 4b, marked by grey fonts (see
Extended Data Fig. 7). We then plot the positions of the peaks in energy eVn against

8
Fig. 4. Magnetic field-dependent tunnelling conductance of the Bi2Se3/2ML hBN/Sb2Te3 device. a, B⟂-
dependence of tunnelling conductance. Spectra under different fields, in 0.5 T increments, are vertically offset by
4 S for clarity. b–e, Colour scale maps showing the field-modulated component of the conductance G(V, B) at
different field angles θ defined with respect to the out-of-plane direction. Perpendicular and in-plane components
of the field are indicated by the left and right ticks, respectively. Vertical axes of b–d are adjusted to have the same
B⟂ scale, while those of b,e have the same |B| scale. The black dashed (dash-dotted) horizontal line indicates the
maximum B⟂ in c (d). f, LL peak positions at different B⟂ plotted versus sgn(n)(|n|B⟂)1/2. The lines are linear fits
to the data. The lower (upper) inset is the schematic representation of the band alignment that causes the LL peaks
in the tunnelling conductance to appear at lower (higher) bias voltages. The spin helicity of the TSSs is expressed
in red and blue colours. Numbers denote LL indices. All measurements are performed at T = 1.8 K.

sgn 𝑛 |𝑛|𝐵 as shown in the lower part of Fig. 4f. Remarkably, all data points for n > 0
(electron LLs) and n < 0 (hole LLs) separately form a straight line and converge at B⟂ 0
when extrapolated, within 1 mV which is our measurement resolution. The slope of each linear
fit line provides the estimation of the 𝑣 of the corresponding TSS: we obtain 𝑣 ≈ 4.1 × 105
m/s for n < 0 LLs (red line) and 5.6 × 105 m/s for n > 0 LLs (blue line), and the corresponding
bands are assigned to the valence and conduction band of TSSs in Sb2Te3, respectively.
There are, however, two exceptional data sets that do not fall into the LL schemes
discussed above. First, the peak of the oscillation corresponding to n = 0, started at V0 ≈ 135

9
mV at zero magnetic field, exhibits a weak linear dispersion as B⟂ increases, with an estimated
g-factor of ~29. Applying an external magnetic field to the TSS breaks the time-reversal
symmetry to open a gap, causing a Zeeman energy shift, most significantly affecting the zeroth
LL44. The experimentally measured g-factors, however, vary widely among the experiments43-
45
. Second, there exists another copy of n = 0 and 1 LL peaks, marked as 0’ and 1’ respectively,
emerging at lower bias voltages with smaller intensities. Although their origin is unclear, these
modulations do not split from or merge with the main LL peaks.
We now turn our attention to the high bias oscillations (III) above. We identify their
peak positions Vn* and assign the corresponding LL indices (Extended Data Fig. 4), marked by
magenta fonts in the upper part of Fig. 4b. We then perform a similar analysis as for (II) and
obtain the corresponding LL origin V0* ≈ 647 mV and velocity 𝑣 ≈ 1.1 × 106 m/s (magenta
line in Fig. 4f). At this higher bias voltage regime, both of the 𝐸 and 𝐸 are outside of
the BVB and BCB of the counterpart, respectively, and the resonant tunnelling can occur
between the LLs developed in the TSSs of Bi2Se3 and Sb2Te3 (see the upper inset schematic in
Fig. 4f). In order for such tunnelling to happen, the quantized energy, En, and angular
momentum of LLs, 𝑘 2𝑒|𝑛|𝐵 /ℏ , as well as the spin-momentum locking helicity of
TSSs should be matched. Considering the two TSSs at the tunnelling interface have the
opposite spin-momentum locking helicity due to their opposite surface normal directions, the
only allowed inter-LL resonant tunnelling is the inter-band tunnelling between an occupied –
nth LL of Bi2Se3 to an unoccupied nth LL of Si2Te3 satisfying 𝑒𝑉 ∗ 𝐸
sgn 𝑛 𝑣 2𝑒ℏ|𝑛|𝐵 𝐸 sgn 𝑛 𝑣 2𝑒ℏ|𝑛|𝐵 , where 𝑣 and 𝐸 are the
Fermi velocity and the DP energy of the Bi2Se3 (Sb2Te3) TSS measured with respect to the
𝐸 , respectively. This relationship results in a straight line in the plot of eVn* versus

sgn 𝑛 |𝑛|𝐵 as shown in the upper part of Fig. 4f, where the slope of this line corresponds
to 𝑣 𝑣 , which is approximately twice as high as 𝑣 for TSSs Bi2Se3 and of
Sb2Te332,40,41,43,46, in agreement with our observation. Thus, the energy-momentum-spin
resonance between two sets of LLs manifests as the sum of the 𝑣 ’s of Bi2Se3 and Sb2Te3.
Finally, we discuss the effect of the in-plane magnetic field B∥ on the resonant
tunnelling between LLs. As shown in Fig. 4c,d, the G peaks corresponding to inter-LL
tunnelling (III) are rapidly suppressed with increasing B∥ (e.g., compare the LL peaks in the
region highlighted by the white dashed box in Fig. 4c with the corresponding B⟂ and V ranges
in Fig. 4b,d). The application of B∥ provides an in-plane momentum gain to the tunnelling

10
electrons6,8, which suppresses inter-band inter-LL resonant tunnelling, where the energy,
angular momentum, and spin of the electrons must be conserved. Note that the tunnelling from
bulk band to LL (II) is not suppressed by the presence of finite B∥ (Fig. 4b–d), since the electron
momentum is not quantized in the BCB of Bi2Se3 from which electrons tunnel.
The realization of two different types of TI materials epitaxially coupled via an
ultrathin crystalline barrier, with atomically sharp and clean interfaces and without chemical
intermixing, has allowed us to systematically investigate the selection rules of electron
tunnelling between the TSSs. Our results show that the two TSSs can be brought together to a
sub-nm distance, which is much shorter than the thickness limit of a single 3D TI slab thanks
to the wide bandgap of hBN39,40,47,48. This system can serve as a platform to explore interesting
physics involving coupled TSSs, such as the proposed topological exciton condensation14,15.
The double-sided vdW epitaxial growth technique can be further applied to various
combinations of vdW materials including superconductors and ferromagnets, potentially
leading to emergent phenomena and novel device applications.

11
References
1. Geim, A. K. & Grigorieva, I. V. Van der Waals heterostructures. Nature 499, 419-425 (2013).
2. Novoselov, K. S., Mishchenko, A., Carvalho, A. & Castro Neto, A. H. 2D materials and van der
Waals heterostructures. Science 353, aac9439 (2016).
3. Bretheau, L. et al. Tunnelling spectroscopy of Andreev states in graphene. Nat. Phys. 13, 756-760
(2017).
4. Jung, S. et al. Direct probing of the electronic structures of single-layer and bilayer graphene with
a hexagonal boron nitride tunneling barrier. Nano Lett. 17, 206-213 (2017).
5. Song, T. et al. Giant tunneling magnetoresistance in spin-filter van der Waals heterostructures.
Science 360, 1214-1218 (2018).
6. Mishchenko, A. et al. Twist-controlled resonant tunnelling in graphene/boron nitride/graphene
heterostructures. Nat. Nanotechnol. 9, 808-813 (2014).
7. Greenaway, M. T. et al. Resonant tunnelling between the chiral Landau states of twisted graphene
lattices. Nat. Phys. 11, 1057-1062 (2015).
8. Wallbank, J. R. et al. Tuning the valley and chiral quantum state of Dirac electrons in van der Waals
heterostructures. Science 353, 575-579 (2016).
9. Heremans, J. P., Cava, R. J. & Samarth, N. Tetradymites as thermoelectrics and topological
insulators. Nat. Rev. Mater. 2, 17049 (2017).
10. Shibayev, P. P. et al. Engineering topological superlattices and phase diagrams. Nano Lett. 19, 716-
721 (2019).
11. Zhao, Y.-F. et al. Tuning the Chern number in quantum anomalous Hall insulators. Nature 588, 419-
423 (2020).
12. Deng, P. et al. Topological surface state annihilation and creation in SnTe/Crx(BiSb)2–xTe3
heterostructures. Nano Lett. 22, 5735-5741 (2022).
13. Kou, L. et al. Graphene-based topological insulator with an intrinsic bulk band gap above room
temperature. Nano Lett. 13, 6251-6255 (2013).
14. Seradjeh, B., Moore, J. E. & Franz, M. Exciton condensation and charge fractionalization in a
topological Insulator film. Phys. Rev. Lett. 103, 066402 (2009).
15. Wang, Z., Hao, N., Fu, Z.-G. & Zhang, P. Excitonic condensation for the surface states of
topological insulator bilayers. New J. Phys. 14, 063010 (2012).
16. Wang, Z. Y. et al. Superlattices of Bi2Se3/In2Se3: Growth characteristics and structural properties.
Appl. Phys. Lett. 99, 023112 (2011).
17. Kim, Y. et al. Remote epitaxy through graphene enables two-dimensional material-based layer
transfer. Nature 544, 340-343 (2017).
18. Kong, W. et al. Polarity governs atomic interaction through two-dimensional materials. Nat. Mater.
17, 999-1004 (2018).
19. Hong, Y. J. et al. Van der Waals Epitaxial Double Heterostructure: InAs/Single-Layer
Graphene/InAs. Adv. Mater. 25, 6847-6853 (2013).
20. Tchoe, Y. et al. Vertical monolithic integration of wide- and narrow-bandgap semiconductor
nanostructures on graphene films. NPG Asia Mater. 13, 33 (2021).
21. Tsen, A. W. et al. Tailoring electrical transport across grain boundaries in polycrystalline graphene.
Science 336, 1143-1146 (2012).
22. Park, J. Y. et al. Molecular beam epitaxial growth and electronic transport properties of high quality
topological insulator Bi2Se3 thin films on hexagonal boron nitride. 2D Mater. 3, 035029 (2016).
23. Lapano, J. et al. Van der Waals epitaxy growth of Bi2Se3 on a freestanding monolayer graphene
membrane: Implications for layered materials and heterostructures. ACS Appl. Nano Mater. 4, 7607-
7613 (2021).
24. Ko, K. et al. Operando electron microscopy investigation of polar domain dynamics in twisted van
der Waals homobilayers. Nat. Mater. 22, 992-998 (2023).

12
25. Liu, Y. et al. Tuning Dirac states by strain in the topological insulator Bi2Se3. Nat. Phys. 10,
Advanced Materials294-299 (2014).
26. Park, J. Y. et al. Bi2Se3 thin films heteroepitaxially grown on α−RuCl3. Phys. Rev. Mater. 4, 113404
(2020).
27. Guha, P. et al. Molecular beam epitaxial growth of Sb2Te3–Bi2Te3 lateral heterostructures. 2D Mater.
9, 025006 (2022).
28. Nakajima, S. The crystal structure of Bi2Te3−xSex. J. Phys. Chem. Solids 24, 479-485 (1963).
29. Anderson, T. L. & Krause, H. B. Refinement of the Sb2Te3 and Sb2Te2Se structures and their
relationship to nonstoichiometric Sb2Te3−ySey compounds. Acta Cryst. B 30, 1307-1310 (1974).
30. Takane, D. et al. Work function of bulk-insulating topological insulator Bi2–xSbxTe3–ySey. Appl. Phys.
Lett. 109, 091601 (2016).
31. Levy, I. et al. Designer topological insulator with enhanced gap and suppressed bulk conduction in
Bi2Se3/Sb2Te3 ultrashort-period superlattices. Nano Lett. 20, 3420-3426 (2020).
32. Zhang, J. et al. Band structure engineering in (Bi1−xSbx)2Te3 ternary topological insulators. Nat.
Commun. 2, 574 (2011).
33. Britnell, L. et al. Electron tunneling through ultrathin boron nitride crystalline barriers. Nano Lett.
12, 1707-1710 (2012).
34. Chandni, U., Watanabe, K., Taniguchi, T. & Eisenstein, J. P. Evidence for defect-mediated tunneling
in hexagonal boron nitride-based junctions. Nano Lett. 15, 7329-7333 (2015).
35. Yan, R. et al. Esaki diodes in van der Waals heterojunctions with broken-gap energy band alignment.
Nano Lett. 15, 5791-5798 (2015).
36. Cappelluti, E., Grimaldi, C. & Marsiglio, F. Topological change of the Fermi surface in low-density
Rashba gases: application to superconductivity. Phys. Rev. Lett. 98, 167002 (2007).
37. Sun, Y. J. et al. Van Hove singularities as a result of quantum confinement: The origin of intriguing
physical properties in Pb thin films. Nano Res. 3, 800-806 (2010).
38. Bahramy, M. S. et al. Emergent quantum confinement at topological insulator surfaces. Nat.
Commun. 3, 1159 (2012).
39. Zhang, Y. et al. Crossover of the three-dimensional topological insulator Bi2Se3 to the two-
dimensional limit. Nat. Phys. 6, 584-588 (2010).
40. Jiang, Y. et al. Landau quantization and the thickness limit of topological insulator thin films of
Sb2Te3. Phys. Rev. Lett. 108, 016401 (2012).
41. Cheng, P. et al. Landau quantization of topological surface states in Bi2Se3. Phys. Rev. Lett. 105,
076801 (2010).
42. Hanaguri, T., Igarashi, K., Kawamura, M., Takagi, H. & Sasagawa, T. Momentum-resolved Landau-
level spectroscopy of Dirac surface state in Bi2Se3. Phys. Rev. B 82, 081305 (2010).
43. Yoshimi, R. et al. Dirac electron states formed at the heterointerface between a topological insulator
and a conventional semiconductor. Nat. Mater. 13, 253-257 (2014).
44. Fu, Y.-S. et al. Observation of Zeeman effect in topological surface state with distinct material
dependence. Nat. Commun. 7, 10829 (2016).
45. Zhang, Z. et al. Zeeman effect of the topological surface states revealed by quantum oscillations up
to 91 Tesla. Phys. Rev. B 92, 235402 (2015).
46. Xia, Y. et al. Observation of a large-gap topological-insulator class with a single Dirac cone on the
surface. Nat. Phys. 5, 398-402 (2009).
47. Jin, H., Im, J., Song, J.-H. & Freeman, A. J. Multiple Dirac fermions from a topological insulator
and graphene superlattice. Phys. Rev. B 85, 045307 (2012).
48. Costa, M. et al. Controlling topological states in topological/normal insulator heterostructures. ACS
Omega 3, 15900-15906 (2018).

13
Methods
Preparation of suspended ultrathin hBN and graphene
For vertical tunnel junctions, atomically thin hBN single crystal flakes are prepared by
mechanical exfoliation on the oxygen-plasma treated surfaces of Si substrates with a 90 nm
SiO2 layer and identified by optical microscopy. The thickness and surface cleanliness of the
identified layers are further characterized by AFM. We choose 2–5 ML hBN for the tunnel
barrier because they provide reliably working tunnel junctions over a suitable current and bias
voltage range. The target flake is then picked up and transferred onto perforated SiNx
membranes using sticky polymer films, such as polycaprolactone (PCL)49 or polycarbonate
(PC), on polydimethylsiloxane (PDMS) stamps. Here, we employ 50 nm thick low-stress
silicon-rich nitride membranes with TEM-compatible silicon frames (Silson Ltd). Arrays of
holes with typical diameters of 0.4 – 1 m are fabricated in the SiNx membranes by standard
electron (e)-beam lithography and CF4/Ar reactive ion etching. The oxygen plasma treatment
of the SiNx membranes immediately prior to the drop-down process enhances the adhesion
between the vdW layers and the membranes, reducing the risk of wash-away of the flake or
wrinkles formation during the subsequent wet process. After the transfer, the sticky polymer is
removed by organic solvents: tetrahydrofuran (THF) for PCL and chloroform for PC. Finally,
the sample is annealed (typically at 350–400oC) under high vacuum to reduce the polymer
residues. For plan-view TEM study of the double-sided vdW epitaxial heterostructures, we also
employ few-layer hBN epitaxially grown on a sapphire substrate50 and monolayer graphene
grown on a Cu foil. These CVD-grown large-scale vdW substrate layers are suspended on the
holey SiNx membranes by the poly(methyl methacrylate) (PMMA)-assisted wet transfer
method. The suspended vdW layer undergoes another round of in-situ thermal cleaning in the
UHV-MBE chamber before the growth of TI thin films.

Double-sided vdW-MBE growth

First, unless otherwise stated, Bi2Se3 thin films are typically grown first on the bottom surface
of the suspended hBN or graphene through the holes in the SiNx membranes by loading the
chip upside-down on a substrate holder of the MBE. We employ the two-step growth process
developed for high-quality Bi2Se3 thin film growth on vdW-layered substrates22,26. The top
surface of the suspended vdW layer remains intact after bottom side growth, preserving its
atomically clean surface (see Fig. 1h). This growth result is attributed to the highly directional
molecular fluxes of the precursors which are hardly subject to scattering events during MBE

14
growth in a UHV environment. Next, the substrate is inverted ex-situ and loaded back to the
MBE chamber so that the top side faces the molecular fluxes. Another round of two-step MBE
growth is then performed to grow either Bi2Se3 or Sb2Te3 thin films on top of the
hBN(graphene)/Bi2Se3 heterostructure using a similar two-step growth process27. In this step,
care is taken to protect the pre-grown Bi2Se3 such as omitting the thermal cleaning process and
reducing the 2nd step growth temperature. The b-TI film has crystalline structures up to the
interface with hBN and exhibits similar structural quality to the t-TI film (Fig. 2b,f), indicating
that the hBN layer provides excellent protection of the bottom layer from the environment
during the ex-situ sample flipping process and subsequent growth of the top layer. The typical
thickness of the TI films is 20 QL for the top side and 10 QL for the bottom side.

Microstructural characterizations
The TEM lamella for the cross-sectional study is prepared by a focused ion beam system (FEI
Helios 650). Field-emission TEMs with an acceleration voltage of 200 kV (JEOL JEM-2100F
and FEI Tecnai F20) are used for HR imaging. An average background subtraction filter is
applied to remove noise in the HR-TEM images. The elemental distribution mapping is
performed with a silicon drift detector-based EDS system operating in the STEM mode and
analyzed with AZtec software (Oxford Instruments).

Device fabrications
Electrical contacts on both sides of the suspended TI/hBN/TI heterostructures, prepared on
SiNx membrane templates, are fabricated by the following steps. First, Ti/Au 5/200 nm metal
layer is deposited on the bottom side of the sample by e-beam evaporation. The thick metal
layer provides ohmic contacts to the bottom TI and mechanical support to the membrane for
the following fabrication steps. We note that the frames of our membrane chips are made of
high-resistivity silicon, which becomes insulating at low temperatures. The individual top
ohmic contacts to the array of junctions are fabricated by e-beam lithography followed by e-
beam evaporation of Ti/Au 5/55 nm. Using the metal contacts as etch masks, an Ar plasma
etching is carried out to remove the TI films on the uncovered area, electrically separating the
array of junctions. Finally, bonding pads and interconnection lines to the contacts are deposited
by another round of e-beam lithography and e-beam evaporation of Ti/Pd/Au 5/20/125 nm. The
typical junction area is in the range of 0.2 to 0.7 m2.

15
Magneto-tunnelling measurements
The current–voltage characteristics of the tunnelling devices are measured using a voltage
source (Yokogawa GS200 DC voltage source or Keithley Instruments 2400 SourceMeter) to
apply DC bias voltage and a low-noise current preamplifier (DL Instruments Model 1211)
connected to a digital multimeter (Agilent 34401A) for measuring the current. The differential
conductance is measured simultaneously with the current–voltage curve by mixing the DC bias
voltage with a small AC excitation voltage of 2 mV at 47.77 Hz using a lock-in amplifier
(SR830, Stanford Research Systems), a voltage divider, and a transformer. Temperature and
magnetic field are controlled by either Oxford Instruments Teslatron PT (base temperature Tbase
= 1.5 K, maximum magnetic field Bmax = 8 T) or Quantum Design Physical Property
Measurement System (Tbase = 1.8 K, Bmax = 14 T). The angle of magnetic fields with respect to
the tunnelling interface is controlled by changing the orientation of sample ex-situ. For each
field angle, tunnelling characteristics at zero magnetic field are cross-checked to confirm the
robustness of the tunnelling conductance against thermal cycling. The Bi2Se3/hBN/Bi2Se3
device presented in Extended Data Fig. 5 is measured at the National High Magnetic Field
Laboratory (Tbase = 0.3 K, Bmax = 31.5 T) with a rotator probe that can change the orientation
of sample in-situ. During the device fabrication, loading, and shipping processes, meticulous
measures are implemented to prevent electrostatic discharge.

Data availability
The data that support the findings of this study are presented in the Article and its Extended
Data. Further data are available from the corresponding authors upon reasonable request.

References
49. Son, S. et al. Strongly adhesive dry transfer technique for van der Waals heterostructure. 2D Mater.
7, 041005 (2020).
50. Jang, A. R. et al. Wafer-scale and wrinkle-free epitaxial growth of single-orientated multilayer
hexagonal boron nitride on sapphire. Nano Lett. 16, 3360-3366 (2016).

Acknowledgements
We thank J. Eom, Y. S. Kim, and C.-H. Lee for their help in the transfer of atomically thin vdW
crystals on the perforated SiNx membranes. The major part of this work was supported by the
Office of Naval Research (ONR) Multidisciplinary University Research Initiatives (MURI)

16
program (N00014-21-1-2377) and the Science Research Center (SRC) for Novel Epitaxial
Quantum Architectures (NRF-2021R1A5A1032996). J.S., J. K., and D.K. acknowledge
support from the National Research Foundation of Korea (NRF) grants funded by the Korean
Government (MSIT) (RS-2023-00283291, RS-2023-00207732, and No. 2023R1A2C2005809).
J.J. and M.K. acknowledge support from the NRF grant NRF-2022R1A2C3007807. H.S.S
acknowledges support from the Institute for Basic Science (IBS-R036-D1). A portion of the
user collaboration grant program (UCGP) project was performed at the National High
Magnetic Field Laboratory supported by the National Science Foundation through NSF/DMR-
1644779 and the State of Florida. K.W. and T.T. acknowledge support from the JSPS
KAKENHI (Grant Numbers 20H00354 and 23H02052) and World Premier International
Research Center Initiative (WPI), MEXT, Japan.

Author contributions
P.K., G.-C.Y., and J.Y.P. conceived the experiments. J.Y.P. performed the double-sided vdW
epitaxy and fabricated the tunnelling devices. J.Y.P. and G.-C.Y. analysed and optimized these
processes. J.Y.P. performed the magneto-tunnelling measurements together with J.S. and J.K..
J.Y.P., R.M.H., A.G., and W.K.P. carried out the measurement at the National High Magnetic
Field Laboratory. J.Y.P., J.S., J.K., D.K., and P.K. analysed the transport data. J.J. conducted
the TEM experiments. J.J., J.Y.P., G.-C.Y., and M.K. analysed the TEM data. J.Y.P. and H.Y.
prepared the hole-patterned SiNx membrane templates. Y.J.S., D.H.N., H.Y., and J.Y.P. carried
out the suspension of ultrathin vdW layers on the perforated SiNx membranes. C. H. and H.S.S.
provided the epitaxial few-layer hBN films and collaborated on discussions about the double-
sided vdW epitaxy on them. J.Y. grew the monolayer graphene. K.W. and T.T. provided the
bulk hBN single crystals. G.-C.Y. and P.K. jointly supervised the project. J.Y.P. and P.K. wrote
the manuscript with input from all authors.

Competing interests
The authors declare no competing interests.

17
Extended data figures

Extended Data Fig. 1. Suspension of atomically thin hBN crystals on the perforated SiNx membrane. a–c,
Optical microscope images of ultrathin hBN layers: as exfoliated on the SiO2 surface (a), picked up by the PCL
polymer stamp (b), transferred on the SiNx membrane with premade holes (c).

18
Extended Data Fig. 2. Larger scale homogeneity of the Bi2Se3/hBN/Bi2Se3 interface. Cross-sectional HR-
TEM images of Bi2Se3/2ML hBN/Bi2Se3 taken along the zone axis [1210] exhibiting atomically sharp and clean
interfaces.

19
Extended Data Fig. 3. Epitaxial relationship of the TI thin films grown on a single surface of suspended
graphene substrates. a–c, Plan-view SAED patterns of the Bi2Se3 thin film grown on the top surface of suspended
graphite (a), the Bi2Se3 thin film grown on the bottom surface of suspended monolayer graphene (b), and the
Sb2Te3 thin film grown on the top surface of suspended monolayer graphene (c). Insets: Plan-view TEM BF
images. The white solid circles and the yellow dashed circles indicate the areas in which the SAED in the
corresponding main panels are taken and the suspended regions, respectively. The entire area displayed in the
inset is suspended for a. Scale bars in the insets: 1 m (a), 200 nm (b), and 200 nm (c). All data (a–c) exhibit the
preferential epitaxial relationship of 1010 Bi2Se3 or Sb2Te3 ∥ 1010 graphene or graphite , indicating that the epitaxial
alignment in the double-sided vdW epitaxy is owing to the local interaction at the interface with the suspended
vdW substrate layer.

20
Extended Data Fig. 4. Tunneling device fabrication processes. a, Optical microscope image of atomically thin
hBN layers suspended on the perforated SiNx membrane. Inset: High-contrast image of the main panel. Scale bar:
10 m. b, Optical microscope image of the sample in (a) after the double-sided vdW epitaxy of TI thin films (10
QL Bi2Se3 on the back side and 20 QL Sb2Te3 on the top side). c, Photograph (upper left) and schematic (lower
right) of the double-sided vdW-epitaxial heterojunction after the evaporation of the Ti/Au 5/200 nm common
bottom contact. d,e, Optical microscope images of the junction arrays: after the fabrication of Ti/Au 5/55 nm
individual top contacts followed by Ar plasma etching (d) and after the formation of bonding pads and leads
(Ti/Pd/Au 5/20/125 nm). f, Optical microscope image (upper left) and schematic structure (lower right) of the
finished device.

21
Extended Data Fig. 5. Magnetic field-dependent tunnelling conductance of a Bi2Se3/3 ML hBN/Bi2Se3 device.
a, Differential conductance 𝐺 𝑉 𝜕𝐼/𝜕𝑉 of the Bi2Se3/3ML hBN/Bi2Se3 device at zero magnetic field. b–d,
Colour scale maps showing the field-modulated component of the conductance G(V, B) at different field angles:
out-of-plane (b), 45 tilted (c), and in-plane (d) directions. B⟂ and B∥ components of the field are indicated by the
left and right ticks, respectively. The black dashed horizontal line in b indicates the maximum B⟂ in c. We observe
oscillatory features that develop at higher B⟂, with their peak and valley positions varying as a function of B⟂, in
addition to the field-angle-independent oscillatory features arising from the bulk QWSs. We are however unable
to conclusively identify conductance oscillations associated with the LL formation in TSSs, presumably because
the conductance is dominated by the bulk-to-bulk tunnelling as both the b- and t-Bi2Se3 films are degenerately n-
doped. The difference in thickness between the b-Bi2Se3 (10 QL) and t-Bi2Se3 (20 QL) is responsible for the
asymmetry in the periodicity of the QWS-related oscillations at negative and positive biases. All measurements
are performed at T = 0.3 K.

22
Extended Data Fig. 6. Oscillations in the tunnelling conductance associated with the bulk QWSs. Differential
conductance 𝐺 𝑉 𝜕𝐼/𝜕𝑉 of the Bi2Se3/2ML hBN/Sb2Te3 device at zero magnetic field (black) and B∥ = 14 T
(red). The inset shows 𝜕𝐺/𝜕𝑉 𝜕 𝐼/𝜕𝑉 as a function of V at different B∥, obtained by taking numerical
derivative of G(V). The black arrows indicate the positions of the peaks of the oscillatory feature (I) related to the
bulk QWSs, appearing in G(V, B) (Fig. 4b–e). One can find that they correspond to the valleys of the QWS
resonances in G(V, B = 0), which become smoothened out at higher B∥ as fine structures develop due to the
Zeeman effects. We also observe that the G(V, B = 0) peak at V = 225 mV, attributed to the alignment of bulk
band edges, is suppressed at higher B∥. All data is measured at T = 1.8 K.

23
Extended Data Fig. 7. LL peak identifications and LL index assignments. a, Field-modulated component of
⁄
the tunnelling conductance G(V, B) at θ = 0° and T = 1.8 K, the same data as Fig. 4b replotted with 𝐵 as the
vertical axis so that the LLs appear as linear lines. b, Peaks identified in a. The lime-coloured dots represent the
peak positions identified by MATLAB’s findpeaks function, a tool for locating local maxima, applied along the
vertical and horizontal axes, as well as in two diagonal directions. The transparency of the dots corresponds to the
prominence of the peaks the function returns. c, LL index assignments. The positions of the peaks assigned to
each LL index n are indicated by the corresponding colour. The green dots are the positions of the peaks not
associated with the LLs. The solid lines tracing the peaks corresponding to n < 0 (red) and n > 0 (blue) LLs of
Sb2Te3, and the inter-band inter-LL resonant tunneling between Bi2Se3 and Sb2Te3 (magenta) are reconstructed
from the slopes and intercepts of the linear fits in Fig. 4f. The dark grey curve is a fit to the n = 0 LL peaks shifting
linearly with respect to B⟂. The dotted curve and line trace the n = 0’ and 1’ LL peaks of unclear origin, respectively.

24