Nano Energy 35 (2017) 321–330

Contents lists available at ScienceDirect

Nano Energy
journal homepage: www.elsevier.com/locate/nanoen

Full paper

Direct observation of vast off-stoichiometric defects in single crystalline MARK

SnSe
Di Wua,1, Lijun Wub,1, Dongsheng Hea,1, Li-Dong Zhaoc, Wei Lia, Minghui Wua, Min Jind,
⁎
Jingtao Xud, Jun Jiangd, Li Huanga, Yimei Zhub, Mercouri G. Kanatzidise, Jiaqing Hea,
a
Shenzhen Key Laboratory of Thermoelectric Materials, Department of Physics, Southern University of Science and Technology, Shenzhen 518055, China
b
Condensed Matter Physics and Materials Science Department, Brookhaven National Laboratory, Upton, NY 11973, USA
c
School of Materials Science and Engineering,Beihang University, Beijing 100191, China
d
Ningbo Institute of Materials Technology and Engineering(NIMTE), Chinese Academy of Science(CAS), Ningbo 315201, China
e
Department of Chemistry, Northwestern University, Evanston, IL 60208, USA

A R T I C L E I N F O A BS T RAC T

Keywords: Single crystalline tin selenide (SnSe) recently emerged as a very promising thermoelectric material for waste
Thermoelectricity heat harvesting and thermoelectric cooling, due to its record high figure of merit ZT in mediate temperature
Transmission electron microscopy range. The most striking feature of SnSe lies in its extremely low lattice thermal conductivity as ascribed to the
Interstitial defects anisotropic and highly distorted Sn-Se bonds as well as the giant bond anharmonicity by previous studies, yet no
Off-stoichiometry
theoretical models so far can give a quantitative explanation to such low a lattice thermal conductivity. In this
Lattice thermal conductivity
work, we presented direct observation of an astonishingly vast number of off-stoichiometric Sn vacancies and Se
SnSe
interstitials, using sophisticated aberration corrected scanning transmission electron microscope; and credited
the previously reported ultralow thermal conductivity of the SnSe single crystalline samples partly to their off-
stoichiometric feature. To further validate the conclusion, we also synthesized stoichiometric SnSe single
crystalline samples, and illustrated that the lattice thermal conductivity is deed much higher as compared with
the off-stoichiometric single crystals. The scattering efficiency of individual point defect on heat-carrying
phonons was then discussed in the state-of-art Debye-Callaway model.

1. Introduction symmetric crystallographic structure(space group Cmcm, #63), where

Thermoelectricity, a physical phenomenon that converts heat
directly to electricity or vice versa, can provide added efficiencies to
ameliorate the fierce conflict between rising energy consumption and
exhausting fossil fuels [1,2]. The performance of a specific thermo-
electric material is evaluated by a dimensionless figure of merit (ZT),
which is defined as ZT=(S2σ/κ)T, where S, σ and κ are the Seebeck
coefficient, electrical conductivity and thermal conductivity, respec-
tively [3–5]. Among the various promising thermoelectric materials
working in the mediate-temperature range, SnSe has attracted strong
interest since few years ago, not only because a peak ZT value of 2.6 [6]
at 923 K was reported in its pristine single crystalline form, but also a
broad ZT plateau over 1 was achieved when the single crystal was
properly hole doped [7,8]. SnSe has a layered orthorhombic crystal
structure at room temperature (space group Pnma, #62) with two-
atom-thick slabs (along b-c plane) of strong Sn-Se bonds stacking
along a axis; as temperature increases to ~800 K, it converts to a more
superior thermoelectric performance was achieved.
One of the most striking features of single crystalline SnSe is the
ultralow lattice thermal conductivity in a direction along which the Sn-
Se slabs are stacked. It has been argued [6] that the highly distorted
SnSe7 polyhedral coordination and sterically accommodated Sn2+ lone-
pairs cause giant bond anharmonicity, which was deemed the main
reason for the extremely low lattice thermal conductivity. Further
studies via inelastic neutron scattering (INS) and first-principles
simulations by Delaire et al. [9] provided strong evidence that
electronic structure instability of the SnSe lattice, associated with a
transverse optical phonon mode softening and broadening, causes
giant ionic-potential anharmonicity and intense phonon scattering.
Despite these reasonable qualitative physical pictures and analysis, no
quantitative account of the ultralow thermal conductivity in SnSe single
crystals has been given so far. Theoretical simulations based on first-
principles calculations and phonon Boltzmann Transport Equation
(BTE) always showed significantly higher lattice thermal conductivities

⁎
Corresponding author.
E-mail address: he.jq@sustc.edu.cn (J. He).
1
These authors contributed equally to this work.

http://dx.doi.org/10.1016/j.nanoen.2017.04.004
Received 25 January 2017; Received in revised form 1 April 2017; Accepted 3 April 2017
Available online 06 April 2017
2211-2855/ © 2017 Elsevier Ltd. All rights reserved.
D. Wu et al.
than experimental values [6,8,10,11], even with higher-order intera-
tomic interactions and bond anharmonicity already included in these
simulations. Another pending issue in SnSe is that the typical experi-
mental thermal conductivity values of polycrystalline samples are
larger than those of single crystals [12–14], which is beyond rule of
thumb for phonon scattering considering the effect of grain boundary
scattering. Although a recent neutron diffraction study demonstrated
that an ultralow thermal conductivity value < 0.2 W m−1 K−1 at room
temperature could be obtained in off-stoichiometric polycrystalline
SnSe [15,16], no further experimental or theoretical evidences have
been given yet to support these conclusions. This potential controversy
regarding the thermal conductivity of SnSe and its physical nature
motivated us to investigate the underlying mechanisms in more details.
As for the experimental lattice thermal conductivity values, one might
argue that the strong tendency of the fragile SnSe crystals to cleave
along the bc plane leaves micro-gaps in the specimens which can
complicate specimen preparations and the accuracy of some thermal
conductivity measurements; besides, oxidation was also reported to
affect the exact value of lattice thermal conductivity significantly [17].
In this work, our recent findings via aberration corrected scanning
transmission electron microscopy (STEM) revealed another possibility,
since we have directly observed in our SnSe and Sn0.985Na0.015Se single
crystals [6,8] an astonishingly large number of point defects, i.e., Se
interstitials and Sn vacancies. Electron probe micro-analysis also
validated severe off-stoichiometry in them. In contrast, in the almost
stoichiometric single crystal synthesized by our collaborators (at
NIMTE, using horizontal gradient freezing method), no such point
defects were seen under STEM. The strong scattering of heat-carrying
phonons by these defects can then be used to resolve the puzzles
associated with the previously reported ultralow lattice thermal con-
ductivity in single crystalline SnSe and Sn0.985Na0.015Se samples.
2. Results and discussions
2.1. Discrepancy between experimental and simulated lattice thermal
conductivity in SnSe single crystal
First, we present the calculated thermal conductivity of perfect SnSe
single crystal along a, b and c directions, respectively, as shown in
Fig. 1(a), where the mass density has already been corrected according
to a recent comment by Wei et al. [18]. The calculations were obtained
via feeding the second-order and third-order interatomic force con-
stant (IFC) derived from density functional theory (DFT) into the
Boltzmann transport equation (BTE), where the contributions of all
phonon modes are well taken into account (details in Supplementary
information). Our simulations are in decent agreement with a pre-
viously published work by Huang et al. [11]. It is also remarkable that
the cut-off radius for third-order IFC used in our anharmonic calcula-
tions is as large as 6.5Å, to allow adequate atomic interaction range as
suggested by Delaire et al. [9]. To avoid any unnecessary complications
related to the phase transition, our simulations and discussion are
limited to the low-temperature Pnma phase (below 750 K). Hereafter,
we denote these calculated results as κ_BTE, to distinguish them from
a later simulation using the Callaway model (κ_CM). Indeed, obvious
discrepancies are found between the calculated and experimental
values, Fig. 1(b), under perfect SnSe single crystal presumption;
specifically, even though the simulated results do indicate a very low
thermal conductivity they are still more than two times as large as the
experimental ones, along all the three principle directions. We also
present the extremely low lattice thermal conductivity values of hole
doped Sn0.985Na0.015Se [8] single crystal in this figure. We credit the
small difference in lattice thermal conductivity values between pristine
and hole doped samples to the NaSn substitutional point defects.
To explore what caused the discrepancies between BTE calculations
and experiments, we conducted chemical composition analysis in
single crystalline samples with nominal composition SnSe and
322
Nano Energy 35 (2017) 321–330
Sn0.985Na0.015Se using the electron probe micro-analysis (EPMA),
which works similarly to a scanning electron microprobe but with a
significantly higher resolution for chemical composition determination
(~100 ppm). EPMA results indicate the Sn:Se atomic ratio is severely
off-stoichiometric, and that the numbers are almost unchanged regard-
less of probing spots on the same sample or from sample to sample,
revealing a consistent off-stoichiometric feature (Table 1). This very
interesting off-stoichiometric feature might come from the segregation
of Sn2+ and Se2- at the melting state due to gravity during the crystal
growing process with vertical Bridgman method. Besides, solidified Sn
drops were found adhering to the inner surface of the evacuated quartz
tubes at the end of synthesis process. A further study under scanning
transmission electron microscope (STEM) disclosed that a large
amount of Se atoms exist at the interlayer interstitial positions, and
that a number of Sn vacancies are also present to accommodate the off-
stoichiometric Sn-Se ratio in both SnSe and Sn0.985Na0.0015Se single
crystals. The existence of interstitial point defects is schematically
shown in Fig. 1(c) and to be discussed in the follow-up section. Density
functional theory (DFT) calculations suggest that Se interstitials and Sn
vacancies are energetically favorable, Fig. S6. Considering Se inter-
stitials and Sn vacancies are strong phonon scattering sources for
phonons with small mean free paths (MFPs) such as in SnSe, they are
thus implemented into a modified Debye-Callaway model to explain the
experimentally observed ultralow lattice thermal conductivity in SnSe
single crystals, Fig. 1(d). The simulated lattice thermal conductivity of
SnSe involving off-stoichiometry, κ_CM_os, is way lower than that of
stoichiometric SnSe (κ_CM_s). Similar simulated results for
Sn0.985Na0.015Se single crystal are given in Fig. S7(b),with extra
scattering from NaSn point defects taken into account.
2.2. TEM discovery of vast interstitial and vacancy point defects
In this section, we present in details the aberration corrected atomic
TEM imaging of the lattice structure of SnSe. Considering the similarity
in lattice structure between pristine and hole doped SnSe single
crystals, the discussion associated with TEM images in the main text
will be restricted to hole doped Sn0.985Na0.015Se; extra TEM images for
pristine SnSe can be found in SI. Fig. 2(a)-(c) show the high angle
annular dark field-scanning transmission electron microscope
(HAADF-STEM) images of single crystalline Sn0.985Na0.015Se viewed
along a, b and c axes, respectively; multi-slice simulations, atomic
models and corresponding diffraction patterns are displayed alongside.
Due to the Z contrast, Sn and Se atomic columns can be directly
distinguished by their brightness, as can be seen along b and c
directions that two-atom-thick Sn-Se slabs are stacked in staggered
fashion. As for a direction, Sn and Se columns are overlapped so that
they are not distinguishable by the electron microscope. The observed
atomic array is exactly the same as the reported room temperature
crystallographic space group Pnma, confirming a layered orthorhombic
crystal structure in Sn0.985Na0.015Se. Moreover, simultaneously ac-
quired annular bright filed-transmission electron microscope (ABF-
STEM) images along c direction not only confirm the alternatively
stacking of Sn-Se slabs, but also unexpectedly reveal a vast number of
interstitial atoms, Fig. 2(d). Dissimilar with HAADF which has a
strong Z-contrast, ABF has been demonstrated extremely useful in
identifying atoms with weak scattering power [19,20]. That explains
why interstitial atoms can be seen clearly in ABF mode rather than
HAADF mode. Interstitial atoms were also observed in pristine SnSe
single crystals, Fig. S1.
2.3. No interstitial and vacancy point defects & higher lattice
thermal conductivity in stoichiometric SnSe samples
As we mentioned earlier, a variety of reported lattice thermal
conductivity of our SnSe single crystals [6,8] might have caused
potential controversies in the field of thermoelectrics. Therefore, we
D. Wu et al.
Fig. 1. The controversy regarding with first-principles simulated results and experimental data. (a) Calculated lattice thermal conductivities of perfect SnSe lattice via BTE along a, b
and c directions, our results agree well with a previous work by Huang et al.; (b) experimental lattice thermal conductivities in single crystalline SnSe and Sn0.985Na0.015Se, these values
are way below the calculated ones in (a) for perfect SnSe single crystal; (c) Schematically view of Se interstitials along b and c directions; (d) Callaway model calculated lattice thermal
conductivities of stoichiometric SnSeκ_CM_sare found way larger than experimental values, while that of off- stoichiometric SnSeκ_CM_os agrees well with the experimental values, in
all three directions.
Table 1
Electron Probe X-Ray Microanalysis (EPMA) results for different single crystalline SnSe
samples, and polycrystalline SnSe as well. SC and PC denote for single crystal and
polycrystal, respectively. The ratios are referring to atomic ones for Sn:Se.
SnSe (SC, NW) Sn0.985Na0.015Se (SC,NW) SnSe (SC, SnSe (PC,
NIMTE) SUSTC)
83.5:100 84.5:100 98.9:100 98.7:100
(at NIMTE, CAS, using horizontal gradient freezing method) synthe-
sized another batch of SnSe single crystal and found the lattice thermal
conductivity values are much higher both in a direction and along b-c
plane, Fig. 3(a). ABF-STEM images of this chunk of single crystal along
c axis, however, exhibit neat Sn-Se lattice without noticeable interstitial
defects, Fig. 3(c). Further EPMA results reveal that this batch of
sample is almost perfectly stoichiometric, Table 1. Noticeably, the
experimental lattice thermal conductivities of this chemically stoichio-
metric sample match very well with the calculated results based on
perfect Sn-Se lattice, as the inset of Fig. 3(a) shows. These findings
reinforce the validity of our above argument that off-stoichiometric
point defects cause significant phonon scattering and result in the
much lower lattice thermal conductivity values of off-stoichiometric
samples than those of stoichiometric SnSe sample. Furthermore, we
also synthesized almost stoichiometric SnSe polycrystals, and com-
pared their thermal conductivity values with stoichiometric SnSe single
crystals, Fig. 3(b). It can be seen that the lattice thermal conductivities
of SnSe polycrystals stay lower than that of single crystals in both
directions. In this manner, the rule-of-thumb that polycrystal normally
exhibits a lower lattice thermal conductivity than corresponding single
crystal as the result of grain boundary scattering remains valid.
323
Nano Energy 35 (2017) 321–330
2.4. Estimate the number density of point defects
To estimate the number density of interstitial atoms in SnSe single
crystals, we adopted a quantitative image analysis as follows. First, the
HAADF image along c direction is chosen and fitted with multiple two
dimensional Gaussians, Fig. 4(a)-(b), because Sn and Se columns are
most separated from each other in this direction. With the intensity of
Sn and Se columns extracted via this technique, the intensity histo-
gram containing the information of the three dimensional structure of
the atomic columns can be derived. For very thin samples, kinematic
scattering dominates the HAADF-STEM intensity, the value of which is
then proportional to the number of atoms within the atomic column. In
this case, the intensity histogram should be a discrete distribution
function, with each peak representing n (an integer number) atoms
within an atomic column [21]. Unfortunately the intensity can be
smeared out by experimental noise or instability of the scanning
electron beam [22]; therefore, a model-based analysis of the histogram
is needed to estimate the number n of each peak [23]. Fig. 4(c) shows
the intensity histogram of Sn atomic columns. By calculating the
integrated complete likelihood (ICL), the number of Gaussian distribu-
tions within the histogram could be identified [24]. In this case, there
are 3 Gaussian distributions, which are shown with blue dashed line in
Fig. 4(c). Coincidently, by fitting the histogram centres using a linear
function and carefully adjusting the integer number of atoms each peak
represents, the straight line can pass extremely closely through the
origin (see the inset in Fig. 4(c)). Using the same method, the intensity
contribution of single Se atom in the HAADF-STEM image as well as
the atomic thickness of the specimen can be extracted (Fig. S3).
Independently, the atomic thickness of Se column can also be obtained.
As the HAADF detector generates Z contrast [25], where the intensity
is generally proportional to Zα (the widely accepted α value lies
between 1.2 and 1.7 [26]). Here, considering that the intensity ratio
of Sn and Se atoms is 1.8, and that their atomic number ratio is 50/34,
D. Wu et al. Nano Energy 35 (2017) 321–330

Fig. 2. STEM evidence of interstitial point defects in SnSe lattice. (a)-(c) HAADF-STEM images of Sn0.985Na0.015Se single crystal viewed along a, b and c axes, respectively; left bottom:
multi-slice simulations; right bottom: atomic models; right panel: electron diffraction; (d) ABF-STEM image of single crystalline Sn0.985Na0.015Se viewed along c axis, where yellows
indicate the interstitial atoms.

Fig. 3. (a) Stoichiometric SnSe single crystal (NIMTE) exhibits much higher lattice thermal conductivities than our off-stoichiometric SnSe sample; meanwhile, the lattice thermal
conductivity values of stoichiometric single crystal sample match very well first-principles calculations (κ_BTE) based on perfect Sn-Se lattice (inset). (b) Stoichiometric polycrystalline
SnSe exhibits lower lattice thermal conductivity than stoichiometric single crystal. (c) ABF-STEM image of stoichiometric SnSe single crystal along c axis exhibits neat Sn-Se lattice;
almost no interstitials can be seen.

one can thus estimate anα value of ~1.5. This turns out to be a strong images with the simulated results. We built a three-dimensional atomic
validation for the model- analysis based on Z contrast. SnSe structure based on the local thickness measurement stated above
After obtaining the atomic thickness of the observed sample, we can and doped it with Sn and Se atoms as interstitials separately. 1 or 2
identify the type of interstitial atoms by comparing the experimental atoms have been implemented in our SnSe lattice model and the

324
D. Wu et al.
Fig. 4. Model-analysis of Se interstitial defects and Sn vacancies. (a) Experimental HAADF-STEM images of Sn0.985Na0.015Se viewed c axis; (b) corresponding image fitting using two-
dimensional Gaussians; (c) the intensity histogram for Sn atomic columns, where the inset shows the linear fit for peak centres assuming the first peak represents 4 atoms; (d)
experimental (left) and simulated (right) HAADF-STEM images of a small area for Sn0.985Na0.015Se; (f) line profiles of the image intensity indicated by the blue lines in (d); (e)
experimental (left) and simulated (right) ABF-STEM images of the same area in (d); (g) line profiles as indicated by blue in (e); (h) a line scan for Sn atom column, the arrow indicates a
possible Sn vacancy.
generalized contrast in refer to their atomic column is to be estimated.
By comparing both experimental HAADF-STEM intensities and ABF
intensity with simulations, we find that only Se atoms are able to
generate the contrast close to experimental images for both HAADF-
STEM and ABF images, Fig. 4(d)-(g). This deduction is consistent
with our DFT calculations (to be discussed later). In addition, we were
also able to estimate the number density of interstitial atoms, which is
~0.5–2 nm−3 according to the simulations. The vacancy features were
also observed ubiquitously, Fig. 4h, with an estimated number density
~ 0.13–0.68 nm−3; but one shall be aware that this number could be
overestimated by counting in possible surface vacancies.
Using the estimated concentration of Se interstitials obtained from
the STEM image analysis, one can then approximate the number of Sn
vacancies, considering these two point defects are the dominating
defects as suggested by our density functional theory calculations. The
calculated defect formation energies in bulk SnSe (Pnma phase) as a
function of the Fermi level (EF) are shown in Fig. S6. It can be seen
that the acceptor-like Sn vacancies (VSn) and interstitial Se (ISe) have
lower formation energies relative to the other types of defects, and thus
contribute to the p-type conducting in SnSe. We considered two
Frenkel pairs (VSn+ISn, VSe+ISe) and a compound defect
(VSn+ISe). We find that when the interstitial atom is located near
the vacancy, it would move back to the vacancy site in the Frenkel pair
cases; otherwise the defects may cause local distortion. Under Se-rich
conditions, as a Se interstitial atom is seated next to a Sn vacancy, the
VSn+ISe compound defect has lower formation energy than the
Frenkel pair. For Na doping, Na would spontaneously substitute Sn,
as evidenced by the negative formation energy of NaSn. This strongly
off-stoichiometric feature also exists in Mg2(Si, Sn) [27,28] and
AgSbTe2 [29,30] systems, without severely altering the carrier con-
centration via the possible self-doping mechanism. In what follows, the
individual role each point defect plays in phonon scattering processes
will be evaluated and analyzed.
325
Nano Energy 35 (2017) 321–330
2.5. Evaluate the scattering effect of off-stoichiometric point defects
on phonon scattering
Upon introducing a point defect into the lattice, additional phonon
scattering will take place due to mass difference, elastic constant/force
bond difference, and corresponding strain field fluctuation. Utilizing
substitutional point defects to reduce the lattice thermal conductivity
has been well understood and is an effective strategy of enhancing the
performances of various thermoelectric systems [31–35], as guided by
the theory developed by Klemens [36] and Abeles [37].
Unlike substitutional point defects, vacancies, scattering by virtue
of both missing atoms and missing linkages, are much stronger phonon
scatterers; therefore, it is improper to treat a vacancy as the limiting
case of a substitutional impurity. It was also argued that the distortion
of lattice around a vacancy is of little consequence to phonon scatter-
ing, and that the treatment of a vacancy can then be approximated to
the sum of a missing atom and missing linkages with its nearest
neighbouring atoms [38,39] (see Eq. 6). One also needs to be aware
that the scattering enhancement near the intrinsic resonance frequency
of a vacancy is insignificant since that frequency is way higher than the
high frequency limit (Debye frequency ωD ~kB θD /ℏ ) of phonons [39].
Distinctively, the treatment for point defects occupying interstitial
positions can be differentiated into two cases. In the case that the
binding forces between interstitial impurities and surrounding matrix
are weak, the scattering is treated as phonons interacting with
individual rattling oscillators, e.g., rattle scattering in CoSb3 [40,41]
and Clathrates [42–44], resonant scattering of NO2- in KCl lattice [45]
and Si/Ge self-interstitial defects [46,47]. Since the binding strength is
not strong, the intrinsic frequency of the oscillator ω0 (~ K / M , K is the
binding force, M is the mass of the interstitial atom) lies well within the
Debye frequency of the matrix material, thus causing extremely strong
interaction with phonons in the close proximity of resonant frequency
ω0. In the other case when the binding forces are strong, no relevant
discussions about the scattering effects of interstitial atoms on phonons
can be found in the literature. One possible reason might be the very
D. Wu et al. Nano Energy 35 (2017) 321–330

Fig. 5. Evaluation on the scattering effects of off-stoichiometric point defects on phonon transport. (a) Schematic phonon scattering by interstitial Se atoms in SnSe viewed along b axis,
these interstitials can effectively filter high-frequency phonons and less likely to scatter low-frequency ones; (b) the relaxation times of U+N processes, NaSnsubstitutional point defects
and off-stoichiometric defects (Sn vacancies and Se interstitials) along a axis are presented; inset is a close-up comparison between the three scattering terms at high frequency region;
(c) spectral lattice thermal conductivity reduces at high frequency region upon the introduction of NaSn defects and off-stoichiometric point defects; (d) cumulative lattice thermal
conductivity with phonon wavelength. It is estimated that the contribution of high-frequency phonons (wavelength λ < 1 nm) to lattice thermal conductivity is reduced from 49% to 43%
and 24%, in the case of stoichiometric SnSe (U+N), stoichiometric Sn0.985Na0.015Se (NaSn substitution) and off- stoichiometric Sn0.985Na0.015Se, respectively. All discussions in (b)(c)(d)
are restricted to room temperature and a axis.

high resonant frequency ω0 restricts their scattering efficiency on the mechanism in anisotropic SnSe single crystals. We consider the
comparably lower frequency phonons which dominate the thermal intrinsic Umklapp and Normal processes, together with three kinds
transport. Another reason might lie in the fact that the number density of point defect scattering, i.e., Sn vacancies, Se interstitials and Na-Sn
of such tight-binding interstitial impurities is usually very low owing to substitutional alloy scattering; other scattering mechanisms, such as
the high formation energy. Although they do generate strong lattice precipitate, boundary and dislocation scatterings, are not taken into
distortion to the local surroundings, they can hardly impede phonon count in this single crystalline case. In the relaxation time approxima-
transport effectively at a very small number density. Nevertheless, in tion, the lattice thermal conductivity κlat can be written as:
the case of our SnSe samples, since a massive number of Se atoms
kB ⎛ kB T ⎞
3
appear to sterically occupy the interstitial location, Fig. 1(c), we can
θD / T ς 4e ς
κlat = ⎜
2π v ⎝ ℏ ⎠
2
⎟ ∫0 τc
(e ς − 1)2
dς
(1)
proceed to re-assess their contributions to phonon scatterings. Our
arguments are as follows. First, unlike substitutional point defects, the
Individual scattering mechanism contributes to the total relaxation
force constants between surrounding atoms (of Se interstitials) and
time τc via Matthiessen's rule:
their nearest neighbours are not supposed to have obvious changes
since the bonds themselves are not modified. Second, mass fluctuation τc = (τU−1 + τN−1 + τv−1 + τi−1 + τs−1)
−1
(2)
is not considered because the interstitial atoms do not change the
original lattice permutations. Ultimately, we attribute the scattering Umklapp process [56].
from interstitial Se atoms solely to the strain field fluctuation and ⎛ θ ⎞
ℏγ 2 2
propose that they serve as effective scattering centres of high-frequency τU−1 ≈ ω T exp ⎜ − D ⎟
Mv 2θD ⎝ 3T ⎠ (3)
phonons, Fig. 5(a), as discussed below.
The phonon scattering mechanisms by NaSn substitutional defects, Normal process [48].
Sn vacancies and Se interstitials are implemented into a modified
Callaway model discussed elsewhere [37,48,49] and below. The τN−1 ≈ βτU−1 (4)
calculations were not conducted in the parameters-fed Boltzmann
Substitutional alloy scattering [37].
transport equation, because it is impractical to build an off-stoichio-
metric SnSe supercell which can give accurate high order force ϖ4δ 3
τs−1 = x (1 − x )[(ΔM / M )2 + ε (Δδ / δ )2 ]
constants and other lattice vibration information. The lattice thermal 4πv 3 (5)
conductivity calculations of the pristine SeSe single crystal along a, b
Vacancy scattering [39,57].
and c directions using the Callaway model (κ_CM) were tuned to
match with BTE results at first, Fig. S7a. ω4 δ 3 ⎡ M ⎤2
Note that Callaway model is usually applied in the isotropic case; τv−1 = y (1 − y) ⎢ − v − 2⎥
4πv 3 ⎣ M ⎦ (6)
here, we adopt this method only to evaluate the role of each scattering
Interstitial atom scattering [36,41,45].

326
D. Wu et al.
Aω 2 2ω4δ 3 ⎡ Δδ ⎤
2
τi−1 = + z (1 − z ) ⎢3.2⋅γ⋅ ⎥
(ω 2 2 2
− ω0 ) πv 3 ⎣ δ ⎦ (7)
where the kB is the Boltzmann constant, ℏ is reduced Plank constant, v
is the sound (phonon-group) velocity along each principle axis at the
long wave limit, θD is the axial Debye temperature considering the
contribution from both acoustic and optical vibration modes (since the
optical mode softening and broadening plays a significant role in heat
conduction in SnSe [9]), and ς is defined as hω/kBT, γ is the Grüneisen
parameter, M is the average molar mass of one atom;β is a fitting
parameter for Normal process; δ and Δδ are the dimension of a host
atom and dimension change due to the introduction of a point defect,
x(~0.015/2) is the molar ratio of NaSn substitutional point defects, ΔM
is the mass difference between impurity and host atoms, ε(~45) is a
phenomenological factor as a function of Grüneisen parameter; Mv is
the molar mass of the missing atom (vacancy), y(~0.03) is the molar
ratio of Sn vacancies; A is a fitting parameter containing the informa-
tion of interstitial defects concentration and corresponding binding
force with surrounding matrix atoms, ω0 is the intrinsic resonant
frequency of interstitial defects, which is proportional to ~ K / M where
K is the effective force constant, finally, z(~0.08) denote the molar ratio
of interstitial point defects. However, in a self-interstitial case such as
SnSe, since the coupling between an interstitial atom and surrounding
lattice atoms is so strong that the characteristic frequency ω0 is even
larger than the Debye frequency ωD [58]; therefore, this frequency ω0
is localized and its scattering effects on phonons are very limited. We
hereby chose to ignore the contribution from resonant scattering. One
also need to be aware that what we discussed above, especially, the
interstitial scattering term is deducted merely by simple analysis; a
more rigorous estimation of interstitial atom scattering might depend
on the exact local lattice symmetry and the coupling matrix which
probably can be derived via solving the Green's-function technique
[46]. The parameters used during the simulations are listed in Tables
S1, S2 and S3.
Upon taking into account the off-stoichiometric point defect
scattering terms, the simulated curves κ_CM_os, thick solid lines in
Fig. 1(d) and S7(b), exhibit decent agreement with experimental
values as reported [6,8]. As shown in Fig. 5(b), point defect scattering
dominates phonon transport at high frequency region, while Umklapp
and Normal processes play a crucial role at the low frequency end. This
scenario can also be understood via the frequency dependent spectral
lattice thermal conductivity κs, which is defined as
ωD ωD
κ = ∫ κs (ω) dω = ∫ Cp (ω) v 2τ (ω) dω , Fig. 5(c). Considering the
0 0
specific heat Cp is ~ω2 in Debye approximation, and that the reciprocal
relaxation time for Umklapp and Normal processesτUN−1 is ~ω2 too,
combined with the group velocity is almost frequency independent
(~ω0), one can easily infer that κs~constant in perfect crystals where
only Umklapp and Normal processes are considered. Whereas, as point
defects are introduced, κs can be significantly reduced at the high
frequency end. One can see clearly from the plot that the overall lattice
thermal conductivity is reduced by more than half due to the off-
stoichiometric Sn and Se defects. We address the extremely strong
scattering of off-stoichiometric defects rather than NaSn defects on
high-frequency phonons, by evaluating the cumulative lattice thermal
conductivity in SnSe, where NaSn and off-stoichiometric point defects
are gradually introduced into the structure, Fig. 5(d). While high-
frequency phonons of wavelength < 1 nm contribute ~49% to the
thermal conduction in perfect SnSe, ~43% in the presence of NaSn
defects, they only offer ~24% in the case of off-stoichiometric
Sn0.985Na0.015Se.
Notably, the substitutional NaSn defects do not cause a significant
reduction of lattice thermal conductivity [50], unlike interstitial point
defects and vacancies. This suggests vacancies and interstitials are
much stronger phonon scattering sources than substitutional defects,
in the case of SnSe.
327
Nano Energy 35 (2017) 321–330
3. Conclusions
For the first time, we directly observed a high number density of
interstitial atoms in single crystals of nominal compositions of SnSe
and Sn0.985Na0.015Se under aberration corrected scanning transmission
electron microscope. These interstitials were then confirmed by ABF-
STEM image modeling to be Se atoms. EPMA composition detection
was adopted to further validate the off-stoichiometric feature of this
batch of SnSe and Sn0.985Na0.015Se single crystals. In contract, no
detectable interstitials were observed in stoichiometric single crystals
provided by our collaborators (NIMTE); meanwhile, this sample
exhibits a much higher lattice thermal conductivity which is close to
the calculated ones for perfect Sn-Se lattice. These findings reveal that
our previously reported ultralow lattice thermal conductivities in SnSe
single crystals can be partly ascribed to their off-stoichiometric feature.
This work could be helpful in clarifying the associated controversy of
understanding the thermoelectric performance of SnSe.
4. Method
4.1. Sample synthesis
4.1.1. Vertical Bridgman method (at Northwestern University, USA)
Ingots (~20 g) with nominal compositions of SnSe and
Sn0.985Na0.015Se were synthesized by mixing proper ratios of high-
purity (99.999%) starting materials of Sn, Se and Na in carbon-coated
fused silica tubes. The tubes were evacuated to a pressure of ~10−4Torr,
flame-sealed, slowly heated to 1223 K in 10 h, soaked at this tempera-
ture for 6 h and subsequently furnace cooled to room temperature. The
obtained ingots were crushed into powders and charged into a conical
carbon-coated silica tube, evacuated and flame-sealed. The conical
silica tube was then placed inside another bigger fused silica tube,
which was evacuated and flame-sealed again. The outer tube is used to
prevent the crystal from oxidation by air because the inner tube can
often break owing to the high temperature phase transition and
considerable difference of thermal expansion between the crystal and
silica. This is why thick wall tubes are recommended. The flame-sealed
double-silica-tube was heated in vertical Bridgman crystal growth
furnace by heating to 1223 K over 10 h, and then cooling from
1223 K to room temperature with the sample moving at a rate of
2 mm h−1. Eventually, SnSe crystals with dimensions of 13 (diame-
ter)×20 mm (length) were obtained. All the manipulations and pre-
parative steps were carried out in a purified N2-atmosphere glove box
with O2 and H2O level < 0.1 ppm.
4.1.2. Horizontal gradient freezing method (at NIMTE, CAS, China)
Single crystalline SnSe was grown by the horizontal gradient
freezing method using a 17-zone furnace. About 300 g of stoichiometric
Sn and Se raw materials (99.99%) were placed in a quartz tube, which
was then sealed in a larger quartz crucible. The crucible was put
between 8# and 11# zone of the furnace. All 17 zones were heated to
1223 K and kept there for 24 h. Afterwards, temperature was carefully
decreased from 17# zone in order to form a temperature gradient
~3 K/cm across the 17 zones. The whole process was carefully
controlled during the growing period of SnSe single crystal.
4.1.3. Solid solution method for SnSe poly-crystals
High-purity Sn, Se raw materials of 5 g were weighed at a strict
stoichiometric ratio, and then sealed in a quartz tube ( < 10−4 Pa). The
quartz tube was then slowly heated up to 400 °C in 4 h, soaked at this
temperature for another 4 h. The temperature was then lifted up to
950 °C in 4 h and kept there for 6 h. After the tube was furnace cooled
to room temperature, the obtained ingot was then pulverized into
powders for SPS sintering. The sintering process was conducted at a
uniaxial pressure of 50 MPa and 450 °C.
D. Wu et al. Nano Energy 35 (2017) 321–330

4.2. Thermal conductivity characterization which is defined as follows, was utilized to estimate the component
number:
The obtained SnSe single crystals were cut into bars along different
directions with dimensions ~10×2.5×2.5 mm. Simultaneous measure- ICL (k ) = −2 log L (ψk ) + [2EN (Vi , ψk ) + 2k log n] (9)
ments of Seebeck coefficient and electrical conductivities were per-
Here, k is the number of Gaussian component within the intensity
formed using an Ulvac Riko ZEM-3 instrument under a helium
distribution, L (ψk) is the maximized likelihood function, EN (V,i ψk)is
atmosphere from room temperature to 773 K. Single crystals were also
the estimated entropy, n is the number of intensity data. For each
cut and polished into rectangular shapes with side length of ~6 mm
histogram, we have run the ICL for 100 times with random initial
and thickness of~2 mm for thermal diffusivity measurements. The
parameters within the data range. Through interpreting the results, one
rectangular samples were then coated with a thin layer of graphite to
not only can obtain the ICL value of k, but is also able to obtain the best
minimize errors from possible material emissivity. Thermal conductiv-
parameters ψk (distribution centres, variation) to maximize the ex-
ity was calculated viaκ=D·Cp·ρ, where the measurement of thermal
pectation.
diffusivity along each direction was consistent with those of electrical
With ICL as a function of k in hand, one can move forward to figure
properties, using the laser-flash method in a Netzsch LFA457 instru-
out how many Gaussian distribution within the model, and to interpret
ment (Northwestern University). These thermal diffusivity values were
the parameters for the Gaussian components. The center for each
reconfirmed in a Netzsch LFA427 (Beihang University), using the same
Gaussian component is taken as the intensity contribution of integer
method. The specific heat capacity (Cp) was indirectly derived by
number of atoms. As the imaged region is rather small, the thickness
comparing with a reference sample (Pyroceram 9606) in the range of
variation of the sample should also be very limited; therefore, the atom
300–773 K, and the mass density (ρ) was firstly estimated using the
number within an atomic column shall be continuous (for example 3, 4,
mass/volume ratio and then reconfirmed with a gas pycnometer
3, … or 7, 8, 9,…). In a very thin sample, the image intensity represents
(Micromeritics AccuPyc1340). The obtained density values were
the atom numbers along each atomic column, indicating that a linear
~6.0 g/cm3.
fitting of intensity-number should cross the origin (0, 0). Our data
shows good agreement with the argument above, if the integers are
4.3. Electron Probe X-Ray Micro-Analysis (EPMA) assumed to be x, x and x. Using this method, one can determine the
local thickness of sample at the atomic scale.
The composition analysis of single crystals of nominal composition
SnSe and Sn0.985Na0.015Se was conducted by a commercial analyzer
EPMA-1610 (Shimadzu, Japan). Three spots were probed for each
4.5.2. Image simulation
piece of sample to validate the results. Single crystalline samples were
Based on measurement of the thickness of the Sn and Se atomic
particularly polished at first, then cleaned in a supersonic cleaner for
columns, a perfect SnSe lattice was built. We have put 1–2 Sn and Se
15 min and dried in air before being examined under EPMA. The
atoms within the model. This model was imported in QSTEM (version
results were listed in Table 1.
2.30, qstem.org) to do the multi-slice image simulation. The experi-
mental parameters (such as convergence angle and collection angle)
4.4. Scanning Transmission Electron Microscopy (STEM)
were set according to the actual experimental conditions. The focus was
set on the interstitial atoms. A Gaussian blurring were applied to the
Specimens suitable for TEM observations were prepared following
simulated images before image comparison were performed. Both
the conventional routine: polishing, dimpling, and ion milling (with a
HAADF and ABF were calculated simultaneously and compared with
starting high-angle voltage ~4 kev, and an ending low-angle voltage
experimental images. The comparison is performed by taking the
~1 kev). STEM- HAADF (High-angle Angular Dark Field) and STEM-
neighbour column as a reference. The contrast and brightness of the
ABF (Angular Bright Field) images were taken on a JEOL ARM 200 F
experimental image and simulated image have been adjusted to fit the
of Brookhaven National Laboratory, USA.
reference peaks.

4.5. STEM image modeling

4.6. Density functional theory (DFT) calculations

The HAADF images were modelled as combination of Gaussian
peaks, which can be described as:
⎡ (x − x i ) 2 + ( y − y ) 2 ⎤
I (x, y) = I0 + ∑ Ai exp ⎢ − i
⎥
⎣ w 2
0 ⎦ (8)
i
whereI0 is the background intensity, Ai is the amplitude of each peak.
All of the Gaussian peaks shared the same width, w0. The experimental
images are pixels×pixels, corresponding to nm×nm. The image fitting
was carried out by a piece of self-developed MATLAB code. The
parameters were tuned to minimize the sum of squared residuals
between fitting and experimental images.
4.5.1. Thickness analysis
After image fitting, the amplitudes (Ai) of Sn and Se atomic
columns were extracted and analysis separately. Ideally, the histogram
of either Sn or Se atomic column amplitude should be a combination of
a series of discretely distributed delta function. Practically, the
histogram could be smeared out by the noise and thus each delta
function is broaden and appears to be a Gaussian distribution instead.
Firstly, one needs to figure out how many Gaussian components are
there within each histogram. Integrated complete likelihood (ICL),
The first-principles DFT calculations are performed using the
standard frozen-core projector augmented-wave (PAW) method, as
implemented in Vienna Ab initio Simulation Package (VASP). The
exchange-correlation interactions are treated in the generalized gra-
dient approximation (GGA) of Perdew-Burke-Ernzerh (PBE) [51–53].
The van der Waals density functional (vdW-DF) method is employed
using the optB86b-vdW functional to account for the non-local
correlation energies. The Kohn-Sham orbitals are expanded in plane
waves with an energy cutoff of 400 eV. The energy convergence
criterion is chosen to be 10−6 eV and the Hellman-Feynman forces
on all atoms are below 0.01 eV/Å. The optimized lattice parameters are
found to be a=11.617 Å, b=4.193 Å and c=4.443 Å, in good agreement
with the experimental values. In order to evaluate the formation
energy, a 2×4×4supercell with 256 atoms is used to model the
defect-free SnSe. A Monkhorst-Pack Γ-centered 2×3×3 k-point grid is
used for Brillouin zone sampling. Temperature dependent lattice
thermal conductivity can be calculated by summing the contributions
of all phonon modes denoted by the wave vector q and the dispersion
branch s based on the phonon Boltzmann Transport Equation (BTE) as
implemented in ShengBTE [54,55].

328
D. Wu et al.
Acknowledgements
This contribution was supported by the Natural Science Foundation
of Guangdong Province (Grant No. 2015A030308001), the leading
talents of Guangdong Province Program (Grant No. 00201517), and
the Science, Technology and Innovation Commission of Shenzhen
Municipality (Grant Nos. JCYJ201508311142508365,
KQTD2016022619565991 and KQCX2015033110182370). TEM work
at BNL was supported by the U.S. Department of Energy, Office of
Basic Energy Science, Division of Materials Science and Engineering,
under Contract No. DE-SC0012704. This work was also supported by
NSFC under Grant Nos. 11504160, 51571007 and 51632005, and
partly by the “Zhuoyue” Program from Beihang University and the
Recruitment Program for Young Professionals. At Northwestern
University this project was supported by the Department of Energy,
Office of Science Basic Energy Sciences grant DE-SC0014520. The
authors appreciate the help of Prof. Wenyu Zhao from Wuhan
University of Technology for the EPMA measurements.
Appendix A. Supplementary material
Supplementary data associated with this article can be found in the
online version at http://dx.doi.org/10.1016/j.nanoen.2017.04.004.
References
[1] A.D. LaLonde, Y. Pei, H. Wang, G.J. Snyder, Mater. Today 14 (2011) 526–532.
[2] J.R. Sootsman, D.Y. Chung, M.G. Kanatzidis, Angew. Chem. Int. Ed. Engl. 48
(2009) 8616–8639.
[3] M.S. Dresselhaus, G. Chen, M.Y. Tang, R.G. Yang, H. Lee, D.Z. Wang, Z.F. Ren,
J.P. Fleurial, P. Gogna, Adv. Mater. 19 (2007) 1043–1053.
[4] J.P. Heremans, M.S. Dresselhaus, L.E. Bell, D.T. Morelli, Nat. Nanotechnol. 8
(2013) 471–473.
[5] T.M. Tritt, M. Subramanian, MRS Bull. 31 (2006) 188–198.
[6] L.D. Zhao, S.H. Lo, Y. Zhang, H. Sun, G. Tan, C. Uher, C. Wolverton, V.P. Dravid,
M.G. Kanatzidis, Nature 508 (2014) 373–377.
[7] K. Peng, X. Lu, H. Zhan, S. Hui, X. Tang, G. Wang, J. Dai, C. Uher, G. Wang,
X. Zhou, Energy Environ. Sci. 9 (2016) 454–460.
[8] L.-D. Zhao, G. Tan, S. Hao, J. He, Y. Pei, H. Chi, H. Wang, S. Gong, H. Xu,
V.P. Dravid, Science 351 (2016) 141–144.
[9] C.W. Li, J. Hong, A.F. May, D. Bansal, S. Chi, T. Hong, G. Ehlers, O. Delaire, Nat.
Phys. 11 (2015) 1063–1069.
[10] J. Carrete, N. Mingo, S. Curtarolo, Appl. Phys. Lett. 105 (2014) 101907.
[11] R. Guo, X. Wang, Y. Kuang, B. Huang, Phys. Rev. B 92 (2015) 115202.
[12] S. Sassi, C. Candolfi, J.B. Vaney, V. Ohorodniichuk, P. Masschelein, A. Dauscher,
B. Lenoir, Appl. Phys. Lett. 104 (2014) 212105.
[13] Y. Fu, J. Xu, G. Liu, J. Yang, X. Tan, Z. Liu, H. Qin, H. Shao, H. Jiang, B. Liang,
J. Jiang, J. Mater. Chem. C 4 (2016) 1201–1207.
[14] C.-L. Chen, H. Wang, Y.-Y. Chen, T. Day, G.J. Snyder, J. Mater. Chem. A 2 (2014)
11171.
[15] F. Serrano-Sanchez, M. Gharsallah, N.M. Nemes, F.J. Mompean, J.L. Martinez,
J.A. Alonso, Appl. Phys. Lett. 106 (2015) 083902.
[16] T. Bera, A.V. Sanchela, C.V. Tomy, A.D. Thakur, arXiv:1601.00753v1 Cond. Mat.
Mater. Sci., 5 Jan, 2016.
[17] Y. Li, B. He, J.P. Heremans, J.-C. Zhao, J. Alloy. Compd. 669 (2016) 224–231.
[18] P.-C. Wei, S. Bhattacharya, J. He, S. Neeleshwar, R. Podila, Y.Y. Chen, A.M. Rao,
Nature 539 (2016) E1–E2.
[19] R. Ishikawa, E. Okunishi, H. Sawada, Y. Kondo, F. Hosokawa, E. Abe, Nat. Mater.
10 (2011) 278–281.
[20] Y.-J. Kim, R. Tao, R.F. Klie, D.N. Seidman, ACS Nano 7 (2012) 732–739.
[21] Z. Li, N. Young, M. Di Vece, S. Palomba, R. Palmer, A. Bleloch, B. Curley,
R. Johnston, J. Jiang, J. Yuan, Nature 451 (2008) 46–48.
[22] G. Van Tendeloo, S. Bals, S. Van Aert, J. Verbeeck, D. Van Dyck, Adv. Mater. 24
(2012) 5655–5675.
[23] S. Van Aert, K.J. Batenburg, M.D. Rossell, R. Erni, G. Van Tendeloo, Nature 470
(2011) 374–377.
[24] G. Martinez, A. Rosenauer, A. De Backer, J. Verbeeck, S. Van Aert, Ultramicroscopy
137 (2014) 12–19.
[25] S. Pennycook, Annu. Rev. Mater. Sci. 22 (1992) 171–195.
[26] Z. Wang, Z. Li, S. Park, A. Abdela, D. Tang, R. Palmer, Phys. Rev. B 84 (2011)
073408.
[27] G. Jiang, J. He, T. Zhu, C. Fu, X. Liu, L. Hu, X. Zhao, Adv. Funct. Mater. 24 (2014)
3776–3781.
[28] X. Liu, L. Xi, W. Qiu, J. Yang, T. Zhu, X. Zhao, W. Zhang, Adv. Electron. Mater. 2
(2016) 284.
[29] M.D. Nielsen, C.M. Jaworski, J.P. Heremans, AIP Adv. 5 (2015) 053602.
[30] C.E. Carlton, R. De Armas, J. Ma, A.F. May, O. Delaire, Y. Shao-Horn, J. Appl.
Phys. 115 (2014) 144903.
329
Nano Energy 35 (2017) 321–330
[31] C. Fu, S. Bai, Y. Liu, Y. Tang, L. Chen, X. Zhao, T. Zhu, Nat. Commun. 6 (2015)
8144.
[32] S.J. Poon, D. Wu, S. Zhu, W. Xie, T.M. Tritt, P. Thomas, R. Venkatasubramanian, J.
Mater. Res. 26 (2011) 2795–2802.
[33] C.B. Vining, W. Laskow, J.O. Hanson, R.R. Van der Beck, P.D. Gorsuch, J. Appl.
Phys. 69 (1991) 4333.
[34] W. Yim, E. Fitzke, F. Rosi, J. Mater. Sci. 1 (1966) 52–65.
[35] H. Wang, J. Wang, X. Cao, G.J. Snyder, J. Mater. Chem. A 2 (2014) 3169.
[36] P. Klemens, Proc. Phys. Soc. Sect. A 68 (1955) 1113.
[37] B. Abeles, Phys. Rev. 131 (1963) 1906.
[38] E.F. Steigmeier, B. Abeles, Phys. Rev. 136 (1964) A1149–A1155.
[39] P.G. Klemens, Phys. B: Condens. Matter 263 (1999) 102–104.
[40] B. Sales, D. Mandrus, R.K. Williams, Science 272 (1996) 1325–1328.
[41] G.S. Nolas, G. Fowler, J. Yang, J. Appl. Phys. 100 (2006) 043705.
[42] J.-H. Kim, N.L. Okamoto, K. Kishida, K. Tanaka, H. Inui, Acta Mater. 54 (2006)
2057–2062.
[43] S. Mano, T. Onimaru, S. Yamanaka, T. Takabatake, Phys. Rev. B 84 (2011) 214101.
[44] Y. Takasu, T. Hasegawa, N. Ogita, M. Udagawa, M.A. Avila, K. Suekuni,
T. Takabatake, Phys. Rev. Lett. 100 (2008) 165503.
[45] R. Pohl, Phys. Rev. Lett. 8 (1962) 481.
[46] D.K. Brice, Phys. Rev. 140 (1965) A1211.
[47] W.-K. Leung, R. Needs, G. Rajagopal, S. Itoh, S. Ihara, Phys. Rev. Lett. 83 (1999)
2351.
[48] J. He, S.N. Girard, M.G. Kanatzidis, V.P. Dravid, Adv. Funct. Mater. 20 (2010)
764–772.
[49] D.Z. Wu, Li-Dong, Jiaqing He, Adv. Mater. (2016).
[50] J. He, J. Androulakis, M.G. Kanatzidis, V.P. Dravid, Nano Lett. 12 (2011) 343–347.
[51] P.E. Blöchl, Phys. Rev. B 50 (1994) 17953.
[52] G. Kresse, J. Furthmüller, Phys. Rev. B 54 (1996) 11169–11186.
[53] J.P. Perdew, K. Burke, M. Ernzerhof, Phys. Rev. Lett. 77 (1996) 3865–3868.
[54] W. Li, J. Carrete, N.A. Katcho, N. Mingo, Comput. Phys. Commun. 185 (2014)
1747–1758.
[55] M. Omini, A. Sparavigna, Phys. B: Condens. Matter 212 (1995) 101–112.
[56] D. Morelli, J. Heremans, G. Slack, Phys. Rev. B 66 (2002).
[57] P. Klemens, Int. J. Thermophys. 22 (2001) 265–275.
[58] P.H. Dederichs, C. Lehmann, A. Scholz, Phys. Rev. Lett. 31 (1973) 1130–1132.
Dr. Di Wu is currently a research assistant professor in
Physics department of Southern University of Science and
Technology, China. He received his B.S. (Applied Physics)
degree from University of Science and Technology of China
in 2006, and Ph.D. (Physics) degree from University of
Virginia in 2013. He has been working on thermoelectric
materials since 2008, and his research interest mainly
includes: phonon/electron transport simulations, struc-
ture–property relation quantifications, strategy design,
and new materials explorations.
Dr. Lijun Wu, is a scientific staff in Condensed Matter
Physics and Materials Science Division, Brookhaven
National Laboratory, USA. He got his B.Sc, and M.Sc in
materials science at Shanghai Jiao Tong University, China;
Ph.D. in materials science at Hunan University, China.
Currently Dr. Wu's research interests focus on the devel-
opment and application of quantitative transmission elec-
tron microscopy techniques, including high resolution
transmission electron microscopy, scanning transmission
electron microscopy, quantitative electron diffraction and
electron energy loss spectroscopy, to solve materials pro-
blems in nanomaterials, thermoelectric materials, super-
conductors and electrode materials in Li-battery.
Dr. Dongsheng He is currently working at Southern
University of Science and Technology. He gained the B.S.
degree from Northwestern Polytechnical University in
2009, and Ph.D. degree from University of Birmingham
in 2014. His expertise lies at understanding the advanced
functional materials through TEM related techniques.
D. Wu et al.
Li-Dong Zhao is a full Professor of Materials Science and
Engineering at Beihang University, China. He received his
B.E. and M.E. degrees in Materials Science from Liaoning
Technical University and his Ph.D. degree in Materials
Science from University of Science and Technology Beijing,
China, in 2009. He was a postdoctoral research fellow in
the ICMMO at University of Paris-Sud from 2009 to 2011
and continued as a postdoctoral research fellow in the
Department of Chemistry at Northwestern University from
2011 to 2014. His research interests include the fabrication
and characterizations of layered structural thermoelectrics,
superconductors, and thermal barrier coatings.
Wei Li is currently a research assistant at Department of
Physics, Southern University of Science and Technology.
He has been working on first-principles calculations of
advanced functional materials for six years, mainly about
defect formation and electronic structures calculations. He
holds a B.E. from Liaocheng University and a M.E. from
Ningbo University, China.
Dr. Minghui Wu is a post doctorial at Physics depart-
ment of Southern University of Science and Technology. He
has been working on computation and design of materials
for several years. He holds a B.E at Sichuan University,
China, and M.E and Ph.D. from Shanghai Jiaotong
University, China.
Prof. Min Jin, Ningbo Institute of Materials Technology
and Engineering, Chinese Academy of Sciences. Min Jin is
a professor at Ningbo Institute of Materials Technology and
Engineering, CAS. He has been working on single crystal
growth for more than one decade. He holds a B.E. from
University of Geosciences,China, a Ph.D. from Shanghai
Institute of Ceramics, CAS.
Dr. Jingtao Xu, Ningbo Institute of Materials Technology
and Engineering, Chinese Academy of Sciences. Jingtao Xu
is an associate professor at Ningbo Institute of Materials
Technology and Engineering, CAS. His research interest
includes synthesizing and understanding the physics of
functional materials, such as thermoelectric materials,
superconductors, and II-VI semiconductors. He received
a B.S. and a Ph.D. both from University of Science and
Technology of China.
330
Nano Energy 35 (2017) 321–330
Prof. Jun Jiang, Ningbo Institute of Materials
Technology and Engineering, Chinese Academy of
Sciences. Jun Jiang is a professor at Ningbo Institute of
Materials Technology and Engineering, CAS. He has been
working on advanced thermoelectric materials for more
than one decade. He holds a B.E. degree and a M.E. degree
both from Wuhan University of Technology, China, a Ph.D.
from Shanghai Institute of Ceramics, CAS, and research
experience from Boston College, USA.
Dr. Li Huang is currently an assistant professor in
Physics department of Southern University of Science and
Technology. She obtained her Ph.D. degree from Fudan
University in 2006, and served as post-doctorial in Georgia
Institute of Technology and Ames National Lab from 2006
to 2012. Her researches mainly involve studying func-
tional/energy materials, metallic glasses, surface structure
and thermodynamics et al. using first-principles calcula-
tions.
Professor Yimei Zhu is Senior Physicist at Brookhaven
National Laboratory and Adjunct Professor at Columbia
University and Stony Brook University. He received his BS
from Shanghai Jiaotong University, and M.S. and Ph.D.
from Nagoya University. He joined BNL as Assistant
Scientist in 1988, rising through the rank to become
Tenured Senior Physicist in 2002. His research interests
include structure-property of strongly correlated electron
systems and energy materials and advanced electron
microscopy. He is Inaugural Fellow of MSA, Fellow of
APS and AAAS. Zhu has published more than 500 peer-
reviewed journal articles and delivered 300 invited talks at
international conferences.
Mercouri G. Kanatzidis is a Professor of Chemistry and
of Materials Science and Engineering at Northwestern
University in Evanston, IL. He also has a senior scientist
appointment at Argonne National Laboratory. His interests
include the design and synthesis of new materials with
emphasis on systems with highly unusual structural/phy-
sical characteristics or those capable of energy conversion,
energy detection, environmental remediation, and cataly-
sis. After obtaining his B.Sc. degree from Aristotle
University in Greece, he received his Ph.D. degree in
Chemistry from the University of Iowa and was a post-
doctoral research fellow at the University of Michigan and
Northwestern University.
Jiaqing He is a full professor at Southern University of
Science and Technology. He received his joint Ph.D. degree
in physics from both Juelich Research Center and Wuhan
University in 2004. He was a postdoctor at Brookhaven
National Laboratory (2004–2008), research associate
(2008–2010) and research assistant professor (2010–
2012) at Northwestern University, and a professor at
Xi’an Jiaotong University (2012–2013). His research inter-
ests include transmission electron microscopy, thermo-
electric materials, and structure and property relationship.