This journal is©The Royal Society of Chemistry 2016 Energy Environ. Sci.

Cite this:DOI: 10.1039/c6ee00322b

Electronegative guests in CoSb3†

Bo Duan,‡ab Jiong Yang,‡ac James R. Salvador,d Yang He,e Bo Zhao,a Shanyu Wang,a

Ping Wei,ab Fumio S. Ohuchi,a Wenqing Zhang,*c Raphael P. Hermann,f

Olivier Gourdon,g Scott X. Mao,e Yingwen Cheng,h Chongmin Wang,i Jun Liu,h

Pengcheng Zhai,b Xinfeng Tang,b Qingjie Zhang*b and Jihui Yang*a

Introducing guests into a host framework to form a so called inclusion compound can be used to

design materials with new and fascinating functionalities. The vast majority of inclusion compounds have

electropositive guests with neutral or negatively charged frameworks. Here, we show a series of

electronegative guest filled skutterudites with inverse polarity. The strong covalent guest–host interactions

observed for the electronegative group VIA guests, i.e., S and Se, feature a unique localized ‘‘cluster vibration’’

which significantly influences the lattice dynamics, together with the point-defect scattering caused by

element substitutions, resulting in very low lattice thermal conductivity values. The findings of electronegative

guests provide a new perspective for guest-filling in skutterudites, and the covalent filler/lattice interactions

lead to an unusual lattice dynamics phenomenon which can be used for designing high-efficiency

thermoelectric materials and novel functional inclusion compounds with open structures.

Broader contextInclusion compounds belong to a large class of materials that possess a rich spectrum of fundamental science and applications. In inclusion compounds, theelectronegative guest with cationic framework combinations are rarely reported. Therefore, conclusive links between structure and property remain elusive.A case in point are the filled CoSb3, which are typical inclusion compounds with two angstrom-sized-cages per unit cell. In this work, the electronegative guestsare successfully filled into CoSb3 under equilibrium conditions. It is shown that the electronegativity difference between Sb and the guest directly determinesthe nature of guest–host bonding in skutterudites and hence their physical properties. Particularly, the strong covalent guest–host interactions observed for thegroup VIA guests, i.e., S and Se, feature a unique localized ‘‘cluster vibration’’ which appreciably influences the lattice dynamics, resulting in very low latticethermal conductivity. Superior thermoelectric performance can be achieved, together with the advantages such as low preparation cost, easy scale-up, and highthermal stability, making them promising candidates for intermediate temperature power generation applications. Our work has filled a void in theunderstanding of guest–host dynamics based on a simple electronegativity rule and should be applicable to other inclusion compounds.

Introduction

An inclusion compound is a chemical complex in which a hostframework can accommodate guest components into its hollowstructures (channel-, layer-, cage-like, etc.). These types of ‘‘openstructured’’ compounds incorporating inorganic or organicguests exhibit fascinating properties and functionalities thatare of interest for many applications, such as electrochemicalenergy storage, thermoelectricity, high performance dielectrics,drug delivery, etc.1–4 Generally, the functionality of an inclusioncompound is primarily dominated by the chemical nature of‘‘host’’ and ‘‘guest’’ and more specifically their interactions orbonding. The guest–host interactions are commonly determinedby many physical parameters of the host and guest elements,such as their electronegativity, atomic size, coordination number,etc. The relative polarity of the guest–host combination isespecially important, though receives little attention, and could

a Materials Science and Engineering Department, University of Washington, Seattle,

WA 98195, USA. E-mail: jihuiy@uw.edub State Key Laboratory of Advanced Technology for Materials Synthesis and

Processing, Wuhan University of Technology, Wuhan 430070, China.

E-mail: zhangqj@whut.edu.cnc Materials Genome Institute, Shanghai University, Shanghai 200444, China.

E-mail: wqzhang@mail.sic.ac.cnd Chemical and Materials Systems Lab, General Motors R&D Center, Warren,

MI 48090, USAe Department of Mechanical Engineering and Materials Science,

University of Pittsburgh, Pittsburgh, PA 15261, USAf Materials Science and Technology Division, Oak Ridge National Laboratory,

Oak Ridge, TN 37831, USAg Research and Development, ZS Pharma, Coppell, TX 75019, USAh Energy and Environment Directorate, Pacific Northwest National Laboratory,

Richland, WA 99352, USAi Environmental Molecular Sciences Laboratory,

Pacific Northwest National Laboratory, Richland, WA 99352, USA

† Electronic supplementary information (ESI) available: Fig. S1–S12 andTables S1–S3. See DOI: 10.1039/c6ee00322b‡ Authors contributed equally.

Received 1st February 2016,Accepted 19th April 2016

DOI: 10.1039/c6ee00322b

www.rsc.org/ees

Energy &EnvironmentalScience

PAPER

Publ

ishe

d on

19

Apr

il 20

16. D

ownl

oade

d by

Uni

vers

ity o

f W

ashi

ngto

n on

27/

04/2

016

15:0

2:01

.

View Article OnlineView Journal

Energy Environ. Sci. This journal is©The Royal Society of Chemistry 2016

largely influence their interactions and thus physical properties ofthe resulting compound. Notably, the overwhelming majority ofhost frameworks are neutral or negatively charged and trapelectropositive (EP) guests, such as the alkali, alkaline earth, andrare earth metals, to form non-covalent interactions (hydrogenbonds, ionic bonds, van der Waals interactions, etc.). Thecationic-trapping inclusion compounds with inverse polarityare rarely observed, presumably due to the peculiarities asso-ciated with most potential electronegative (EN) guests, e.g. thelarge ionic radii and larger formation energies, etc.4

Unusual and intriguing guest–host interactions and physicalproperties, however, can be derived from the inverse polarityassociated with the EN guests and cationic hosts, e.g., the singlecovalent guest–host (Se–Ge) bonds and the superstructure inGe46�xPxSe8�y,

5 the strong dielectric anisotropy in (Hg11P4)(GaCl4)4

and (Hg3AsS)(GaCl4),6 and the large ferroelectric saturationpolarization in 1D-helical assemblies with perchlorate or nitrateanions,7 etc. Nevertheless, very limited studies have been performedon cationic inclusion compounds and therefore conclusive findingsregarding the structure–property relationships remain elusive.

Filled skutterudites are typical inorganic inclusion com-pounds with two angstrom-sized-cages per unit cell. Traditionalfilled skutterudites with EP guests and anionic host, have beenextensively studied for intermediate temperature thermoelectricpower generation applications.8,9 Stable cationic filled skutteruditeswith EN guests are rarely studied. To this point, the onlyexample of an EN guest in skutterudite is iodine synthesizedby non-equilibrium methods.10,11 Filled skutterudites are stabi-lized by forming strong guest–host chemical bonds. This lowers

the formation energies for the antimony skutterudite phases(DHs) to levels below those of competing secondary phases.Here DH is defined as DH = (ERyCo4Sb12

� yER � ECo4Sb12)/y,

corresponding to the reaction

yR + Co4Sb12 - RyCo4Sb12, (1)

where E is the total energy for the respective chemical species.According to Shi et al.,12 only those metallic elements with lowenough electronegativity xR w.r.t. xSb (Dx = xR � xSb o �0.8),form filled skutterudites. The alkali, alkaline earth, and rareearth metals satisfy this criterion13–20 and have sufficientlynegative formation energies as shown in Fig. 1a. Very recently,group IIIA elements Ga and In were reported to be stabilizedby self-compensation,21–23 but still being EP guests. Herein,EN guest filled skutterudites are successfully prepared underequilibrium conditions. The results demonstrate that the electro-negativity difference between Sb and the guest directly determinesthe nature of guest–host bonding and hence their physical pro-perties. Particularly, the strong covalent guest–host interactionsobserved for S and Se, feature a unique localized ‘‘clustervibration’’ which appreciably influences the lattice dynamics,resulting in very low lattice thermal conductivity (kL).

Results and discussionEnergetics and experimental validation

The electronegativity of Sb is B2.05 (Pauling scale), a value thatis rather moderate among all elements. To explore the possibility

Fig. 1 Formation energies and lattice thermal conductivity (kL) of EN guest filled skutterudites. (a) EP and EN guests in CoSb3. The left-hand portion(xR� xSb o 0) shows the reported EP guests.13,14,21 The right-hand portion (xR� xSb 4 0) shows the EN guests studied in this work. The red stack columnsindicate formation enthalpy of EN guests without compensation, and the blue ones indicate those of EN guests with compensation. (b) ABF-STEM imageof Se-filled Se0.17�mCo4Sb11.31Te0.53Sem along the [100] direction. (c) Detailed atomic arrangement, the pink balls represent guest atoms, the dark blueballs represent Co atoms, the light blue balls represent Sb (or doped) atoms. (d) An intensity profile along the box marked in (b) to display the intensityvariation including the guest only atomic columns. (e) Room temperature experimental kL as a function of filling fraction y for EN guests filled systems,the EP guests15–19 filled systems are shown for comparison. The solid lines are guides for the eye. The inset compares low-temperature kL of EN andEP20 guests filled skutterudites.

Paper Energy & Environmental Science

Publ

ishe

d on

19

Apr

il 20

16. D

ownl

oade

d by

Uni

vers

ity o

f W

ashi

ngto

n on

27/

04/2

016

15:0

2:01

. View Article Online

This journal is©The Royal Society of Chemistry 2016 Energy Environ. Sci.

of EN guests, we performed formation energy calculations forthe groups VIA and VIIA elements. As shown in the right-handportion of Fig. 1a, group VIIA elements, such as Cl and Br, havelarge negative DHs, close to those of alkali guests.14 Group VIAelements S and Se, are unstable as fillers alone in CoSb3;however, under the condition of charge compensation such assubstituting a small amount of Te at the Sb sites (24g Wyckoffsite), the sign of DHs changes suggesting that these should bethermodynamically stable formulations. Charge compensationcan also greatly enhance the stability of the group VIIA guests.These formation energy variations are comparable to those of EPguests in CoSb3.13 As will be discussed later, the nature of guest–host chemical bonding interactions primarily depends on Dx(Fig. 1a), with larger Dx tending to be more ionic. In short, ourtheoretical work strongly suggests the possibility of an unusuallystable EN filling in CoSb3. Among EN guests, Se is different fromS, Cl, or Br due to its dual-site feature in skutterudites, and theactual Se filling fraction is lower than the measured Se content.The exact ratio between two sites (filling and doping) depends onthe Fermi level and is difficult to identify experimentally.

The annular bright field scanning transmission electronmicroscopy (ABF-STEM) technique is known to give better contrastfor the light elements, and the atom positions can be directlyvisualized as dark contrast regions. Fig. 1b shows the atomicresolution ABF-STEM image of Se-filled Se0.17�mCo4Sb11.31Te0.53Sem

along the [100] zone axis. In the enlarged unit-cell image shownin Fig. 1c, the lighter dark dot (marked by the pink arrow) locatedat the guest position can be clearly seen, in good agreement withthe structure model of filled RyCo4Sb12. A line-profile (reversedfor visual convenience) (Fig. 1d) extracted from the ABF signalshows the distinct intensity corresponding to the void sites. Thisprovides solid evidence for the presence of guests, and the guestsshould be the Se atoms since Te is a well-defined dopant at theSb-site in CoSb3.24–26 The occupancy of Br at the void sites is alsoconfirmed by the ABF-STEM images of the Br-filled sample,see Fig. S1 (ESI†). Further substantiating the structural modelproposed by the ABF-STEM investigation, X-ray and neutronpowder diffraction were also performed and analysed via theRietveld method (ESI,† Fig. S2–S4). We find that all samplescrystallize in the body centered cubic space group Im%3, andexcellent fits to the experimental data can be obtained bymodeling the system with fillers at the 2a site in an identicalmanner to EP-filled skutterudites. The refined quantities ofneutron diffraction are in good agreement with X-ray diffrac-tion data. A full description of the refinement can be found inESI,† Tables S1–S3. To demonstrate the thermodynamic stabilityof EN guests filled skutterudites, we compare the thermoelectricproperties before and after a 15 days anneal at 875 K and findthem nearly unchanged (ESI,† Fig. S5).

Fig. 1e shows the room temperature kLs for EN guest filledskutterudites. Both Cl- and Br-filled RyCo4Sb12 exhibit highkLs B 7 W m�1 K�1, due to their light atomic masses and largeionic radii analogous to the alkali metals15,27 and their very lowfilling fractions (y B 0.04). The low measured filling fractionsmay be attributed to the rapid evaporation of filler precursorsin the synthesis processes due to their lower melting and

boiling points, i.e., SbCl3, melting point 347 K, boiling point500 K; SbBr3, melting point 370 K, boiling point 561 K.Te-substitution is an effective way to reduce the lattice thermalconductivity of CoSb3 from 8–10 W m�1 K�1 (e.g., 8.9 W m�1 K�1 forCoSb2.99 in this work, 10 W m�1 K�1 for CoSb3

28) to 3–4 W m�1 K�1

(e.g., 3.9 W m�1 K�1 for Co4Sb11.46Te0.43 in this work, 3.4 W m�1 K�1

for Co4Sb11.5Te0.525) at room temperature, owing to the point

defect and possibly electron–phonon scatterings.26,29 However,it is impossible to further lower the kL value even with moreTe content (between 0.7 and 11), e.g., 3.9 W m�1 K�1 forCo4Sb11.3Te0.7,30 3.3 W m�1 K�1 for Co4Sb11.2Ge0.2Te0.8,31

3.5 W m�1 K�1 for Co4Sb10.4Ge0.5Te1.1.32 The kL B y for ENguest filled skutterudites with Te compensation clearly followstwo sets of trends. Cl- and Br-filled RyCo4Sb12�zTez showlimited kL reductions with respect to Co4Sb12�zTez. However,S- and Se-filled ones have very low kL values. Although theguests are lightweight themselves, the room temperature kL can bedecreased to B1.5 W m�1 K�1 for S0.26Co4Sb11.11Te0.73, which ismuch lower than those of unfilled counterparts Co4Sb12�zTez,alkali or alkaline earth filled skutterudites,15,16,27 and evencomparable with those of nanostructured YbyCo4Sb12.19

Furthermore, the S-filled skutterudites possess an anomaloustemperature dependence of kL as shown in the inset of Fig. 1e.Distinct from the Br-, Ba-, Yb-filled, and guest-free skutterudites,all of which have low temperature maxima in their kL values, theS-filled skutterudites lack this feature and instead have virtuallytemperature independent kL from 40 to 220 K that are at muchlower values by comparison. This low temperature behaviorand the very low kLs of group VIA guest filled RyCo4Sb12�zTez

can be attributed to the unusual guest–host bonding and theresulting unique lattice dynamics.

The origin of low lattice thermal conductivity

The electronegativity difference between Sb and the guest directlydetermines the nature of the chemical bonding in the cageof filled skutterudites.13 Since Dx for the VIIA guests are large,e.g. B0.9 for Br and B1.1 for Cl, they predominantly form ionicbonds with the host as shown by the charge density difference inFig. 2a for the Br-filled sample. Guest Br are located at the centerof the void, similarly to most previously reported EP guests inCoSb3. Owing to their small masses and large ionic radii, Cl andBr only lead to high vibrational frequencies B100 cm�1 for Br(Fig. 3) and B130 cm�1 for Cl (ESI,† Fig. S6). Their phonondispersion relations, in the low frequency regions, are very closeto those of unfilled skutterudites (ESI,† Fig. S6). These featureslimit kL reduction for Cl- and Br-filled samples, in both p-typeRyCo4Sb12 and n-type RyCo4Sb12�zTez systems as shown in Fig. 1eand ESI,† Fig. S7.

The group VIA guests S and Se form strong polar covalentbonds as shown in Fig. 2b due to their comparable electro-negativity values to Sb (Dx B 0.5). The Sb is ‘‘drawn’’ towards S,with a bond length of 2.57 Å. This value is slightly larger than thesum of covalent radii (B2.45 Å) of S and Sb, and much shorterthan the usual ionic guest–Sb distance (B3.40 Å) observed inCoSb3-based skutterudites. The charge density accumulatesprimarily between S and Sb, with higher density closer to

Energy & Environmental Science Paper

Publ

ishe

d on

19

Apr

il 20

16. D

ownl

oade

d by

Uni

vers

ity o

f W

ashi

ngto

n on

27/

04/2

016

15:0

2:01

. View Article Online

Energy Environ. Sci. This journal is©The Royal Society of Chemistry 2016

S due to its larger electronegativity. According to our Bader chargecalculations, S gains 1.04 electrons from the nearest neighboringSb.33,34 Se also forms the same polar covalent bonds with the hostlattice Sb, with a longer bond length (2.80 Å). In order to verify thetransient bonding between Sb and S predicted by the calcula-tions, the difference Fourier density maps (Fig. 2c) were deter-mined using the neutron powder diffraction refinements (ESI,†Table S3 as references). These maps clearly indicate that at 10 Kpart of the Sb density is displaced towards the S, with somemissing on-site density at the Sb–S distance of 3.32 Å, and amaximum at a distance of 2.67 Å in excellent agreement withthe calculations. There is no experimental evidence for displace-ment of S. Note that because these data are obtained frompowder diffraction all equivalent crystallographic sites exhibitthe same behavior. The appearance of a maximum in the Fourierdensity map at a distance 2.67 Å indicates that in a time averagedpicture a small amount of Sb exhibits a modified bonding with S.The neutron diffraction results strongly corroborate our theo-retical predictions of the short covalent bonds with Sb drawntowards S.

The covalent bonds between S (Se) and neighboring Sb causeunique lattice dynamic features. The two strongly bonded atoms,S and Sb, vibrate as a unit (see Video S1, ESI†), and introduce‘‘cluster vibration’’ optical modes at 35–50 cm�1 (Fig. 3). Since oneend of the cluster is on the framework, the vibrations of the clusterare expected to be very effective at impeding phonon transportthrough the host by the acoustic-optical interactions. These opticalmodes caused by the ‘‘cluster vibration’’ have noticeably lowerfrequencies than those of Ba-filled CoSb3 (Fig. 3), and comparableto those of Yb-filled CoSb3 (ESI,† Fig. S6), whose low-frequency(33–50 cm�1) optical branches mainly originate from the rattlingmotions of heavy Yb guest ion. EN covalent filling provides a novelapproach to introduce low-frequency localized optical modes, andsignificantly reduce kL of CoSb3, using light and earth-abundantfiller elements. Motivated by the unusual lattice dynamics ofEN guest filled skutterudites, we performed phonon calcula-tions for the EP guest clathrate Rb8Al6Si40 and the EN guestanalog, Se8P16Si30, since Rb and Se are comparable in mass

(ESI,† Fig. S8). As expected, the Se-filled Se8P16Si30 showsoptical phonons with much lower frequencies than those inRb8Al6Si40, leading to stronger acoustic-optical interaction.A much lower kL in EN filled clathrates is thus expected.Therefore covalent filling and the resulting ‘‘cluster vibration’’feature induced by the large local structural distortion, shouldbe a new general principle for significant kL reduction in cagedcompounds.

At elevated temperatures, the specific features of the ‘‘clustervibration’’ might change due to the enhanced thermal vibrationalenergy. Our ab initio molecular dynamics (AIMD) simulation (ESI,†Fig. S9) implies multiple S–Sb covalent bonds, each with a shortperiod time, and statistically leading to non-spherical trajectoriesof all neighboring Sb atoms towards S. This is analogous to thetunneling phenomenon, observed in Eu8Ga16Ge30,35 except that inour case the fillers are ‘‘static’’ while the pairing Sb atoms in thenearest neighbor positions are rapidly switching to form the S–Sbpairs and possibly tunneling at the lowest temperature. Ourneutron diffraction results at 295 K show that a ring of enhanceddensity surrounds an on-site minimum rather than a specific short

Fig. 2 Chemical bonding and Fourier difference maps in filled skutterudites with EN guests. (a and b) Calculated charge density difference at Sb four-ringplane in Br0.063Co4Sb12 (a) and S0.063Co4Sb11.5Te0.5 (b). (c) Fourier difference maps F(obs) � F(calc) obtained from the neutron powder diffractionrefinements on S0.26Co4Sb11.11Te0.73 at 10 K (left) and 295 K (right). The represented plane corresponds to integrated intensity between z = 0.49 and 0.51.In this plane S is located at x = y = 0.5 and Sb at x = 0.335 and y = 0.158 and equivalent positions mirrored by the x = 0.5 and y = 0.5 planes.

Fig. 3 Phonon spectra in filled skutterudites with EN guests. Phonon dis-persion in Br0.5Co4Sb11Te1, S0.5Co4Sb10.5Te1.5, and Ba0.5Co4Sb12, projectedonto the corresponding guests, vertical bars represent the strength ofphonon characters from certain atoms.

Paper Energy & Environmental Science

Publ

ishe

d on

19

Apr

il 20

16. D

ownl

oade

d by

Uni

vers

ity o

f W

ashi

ngto

n on

27/

04/2

016

15:0

2:01

. View Article Online

This journal is©The Royal Society of Chemistry 2016 Energy Environ. Sci.

S–Sb bond at 10 K (Fig. 2c), indicative of changed vibrationalpatterns induced by the higher temperature. This is consistentwith our AIMD results of the statistically averaged S–Sb distancesat room temperature. The unique vibrational features of VIAelements filled skutterudites are responsible for the low andplateaued temperature dependence of kL in these compounds,and other point defects (e.g., Te-substitution, Sb-vacancy)should also contribute to the overall kL reduction, though pointdefects alone would not lower the room temperature kL tobelow 2 W m�1 K�1.

Chemical state of EN guests in filled skutterudites

The EN guests in skutterudites are expected to accept electronsfrom the host lattice to complete their octets. Fig. 4a shows the

total and guest-projected density of states (DOS) for S-filledskutterudite. The zero energy point corresponds to the Fermilevel. The projected DOS for the S guest is well below its Fermilevels, as for the Br guest in Br0.04Co4Sb11.84 (ESI,† Fig. S10). Theseresults are also supported by our XPS measurements. As shown inFig. 4b, the binding energy of the S 2s peak in S0.26Co4Sb11.11Te0.73

is clearly lower than that of the elemental S, indicative of anegative charge on S. A negative charge could also be verifiedvia XPS for Br in Br0.04Co4Sb11.84 (ESI,† Fig. S10). The resultsdemonstrate that Br and S guests are acceptors in CoSb3,in spite of their different types of chemical bonding. Indeed, ourelectrical transport measurements of Br- and Cl-filled RyCo4Sb12

indicates typical p-type conduction over the entire temperaturerange investigated (300–850 K) (ESI,† Fig. S7). In the n-type systems,

Fig. 4 Chemical state of the S guest in filled skutterudites. (a) Total and projected (on S guest) DOS for S0.063Co4Sb11.5Te0.5. The Fermi level is at the zero energypoint. (b) X-ray photoelectron spectroscopy (XPS) of S 2s core levels for S0.26Co4Sb11.11Te0.73. The data of elemental S 2s core levels are plotted for comparison.

Fig. 5 Thermoelectric transport properties of EN guest filled skutterudites. (a–d) Temperature dependence of electrical conductivity s (a), Seebeck coefficienta (b), lattice thermal conductivity kL (c), and dimensionless figure of merit ZT (d). The error bars denote the experimental measurement uncertainties.

Energy & Environmental Science Paper

Publ

ishe

d on

19

Apr

il 20

16. D

ownl

oade

d by

Uni

vers

ity o

f W

ashi

ngto

n on

27/

04/2

016

15:0

2:01

. View Article Online

Energy Environ. Sci. This journal is©The Royal Society of Chemistry 2016

the carrier concentration (nH) and electrical conductivity (s) arelowered by filling of the EN guests, as shown in Fig. 5a andTable 1. The nH decreases from 5.6� 1020 cm�3 for Co4Sb11.46Te0.43

to 3.7 � 1020 cm�3 for Br0.16Co4Sb11.34Te0.52, and further to1.9 � 1020 cm�3 for S0.19Co4Sb11.42Te0.52.

Thermoelectric transport properties

Thermoelectric transport properties between 300 and 850 K areshown in Fig. 5. With increasing temperature, all EN guest filledsamples display similar variation for s and |a| as comparedto the Te-doped sample, with s decreasing and |a| increasingmonotonically, indicative of typical degenerate semiconductingbehavior. Filling with EN guests lowers the s of n-type samplesowing to the decreased nH (Table 1), with group VIA guestsproviding two holes while VIIA guests provide one; however,EN guests exert a negligible influence on the band structure,thus a is not modified and is similar to the traditional EP guestfilled skutterudites (ESI,† Fig. S11),9,36 indicating that rigidband approximation is valid.

The S and Se containing samples with covalent interactionsbetween the guest and host have largely reduced kL values over thewhole temperature range (Fig. 5c) as compared with unfilled sampleand samples with ionic interactions (the Br-filled). The lowest valueof B0.8 W m�1 K�1 is obtained at B750 K for S0.26Co4Sb11.11Te0.73,a B60% reduction compared to Co4Sb11.46Te0.43 at the corres-ponding temperature. As a result of the significant reductionin kL, S-filled samples exhibit high ZT values, but with muchlower nH and mH (Table 1 and ESI,† Fig. S11) as compared withoptimal values for typical EP guest filled n-type skutterudites.12,37 Amaximum ZT of B1.5 at 850 K is achieved for S0.26Co4Sb11.11Te0.73,which is comparable to the best n-type skutterudites withmultiple EP guests. The average ZT value between 550 and 850 Kof S0.26Co4Sb11.11Te0.73 reaches 1.3, making it a very promisingcandidate for intermediate temperature power generation applica-tions, particularly when considering other advantages such as lowmaterials cost, easy scale-up (relatively short preparation time,lower temperature, and without protective atmosphere for thestarting materials, etc.), and high thermal stability. Moreover,n-type filled systems with EN guests need charge compensa-tion, which provide many different possible formulations to beexplored. ZT values can potentially be further improved by

co-doping with different electron dopants at the Co and/orSb sites, such as Ni (Fig. 5 and Table 1), Pd, Pt, or by multiplyfilling with additional cation or anion guests. P-type Cl- andBr-filled samples show low ZT values since the hole transport isdominated by the light valence band of Sb (ESI,† Fig. S7).38

Concluding remarks

These EN elements have surprisingly been overlooked for manyyears. A possible reason might be that either charge compen-sation, for S and Se, or proper precursors, for Cl and Br, areneeded to obtain stable filled phases. In addition, since Te hasbeen well documented to be a dopant on the Sb sites, otherelements in the same column of the periodic table, such asS or Se, are intuitively excluded as fillers. Our study clearlyshows that the formation of filled skutterudites requiresstrong guest–host interaction; however, it does not have aprerequisite for the polarity of guests’ charge state. Based onthis concept and related discussions, the peculiar thermo-electric transport properties of the EN elements (S or Se) containingskutterudites,39,40 specially, the low kL, reduced nH, etc., are nowcomprehensible.

In summary, EN guests in inclusion compounds present greatopportunities of possessing unusual physical properties, suchas demonstrated in CoSb3-based skutterudites. These uniquematerials that possess interesting and potentially useful propertiescertainly demand further investigation.

MethodsSample preparation

Starting materials were Co (99.995%), Ni (99.99%), Sb (99.999%),Te (99.999%), S (99.9%), Se (99.9%), SbCl3 (99%), and SbBr3

(99%) powders. The stoichiometric amounts of powders werehand mixed in an agate mortar, cold pressed and sealed intoquartz tubes under vacuum. The tubes with cold-pressed pelletswere heated to 953 K for 60 h. The obtained materials were thencompacted using spark plasma sintering (SPS) at 920 K for 5 minunder 40 MPa. This synthesis technique is obviously more time-and cost-efficient than the traditional melting–annealing–sinteringtechnique for preparing skutterudites with electropositive guests

Table 1 Summary of compositions and physical properties of EN guest filled skutterudites. The nominal composition, actual composition (EPMA), latticeparameter (a), carrier concentration (nH), Hall mobility (mH), and kL at 300 K for EN guest filled samples. The +/� sign in the nH column indicates the sign ofthe majority carrier

Nominal composition EPMA a (Å) nH (1020 cm�3) mH (cm2 V�1 s�1) kL (W m�1 K�1)

Co4Sb11.4Te0.6 Co4Sb11.46Te0.43 9.050 �5.6 20 3.9S0.2Co4Sb11.4Te0.6 S0.19Co4Sb11.42Te0.52 9.048 �1.9 27 2.1S0.27Co4Sb11.2Te0.8 S0.26Co4Sb11.11Te0.73 9.054 �3.1 22 1.5Se0.2Co4Sb11.4Te0.6 Se0.17�mCo4Sb11.31Te0.53Sem 9.052 �3.2 21 2.4Cl0.2Co4Sb11.4Te0.6 Cl0.18Co4Sb11.36Te0.51 9.050 �3.4 25 3.4Br0.2Co4Sb11.4Te0.6 Br0.16Co4Sb11.34Te0.52 9.055 �3.7 23 3.4Co3.4Ni0.6Sb12 Co3.4Ni0.59Sb12.01 9.051 �24 5 4.3S0.2Co3.4Ni0.6Sb12 S0.18Co3.4Ni0.58Sb11.94 9.050 �16 4 2.2CoSb3 CoSb2.99 9.034 +0.03 1.4 � 103 8.9Cl0.2Co4Sb12 Cl0.04Co4Sb11.86 9.035 +0.53 400 6.8Br0.2Co4Sb12 Br0.04Co4Sb11.84 9.037 +0.59 330 6.9

Paper Energy & Environmental Science

Publ

ishe

d on

19

Apr

il 20

16. D

ownl

oade

d by

Uni

vers

ity o

f W

ashi

ngto

n on

27/

04/2

016

15:0

2:01

. View Article Online

This journal is©The Royal Society of Chemistry 2016 Energy Environ. Sci.

(melting at 1273–1473 K for 10–40 h, and annealing at 873–1073 Kfor 120–340 h),15–18 primarily due to the absence of air-sensitiveguests and high melting-point intermediate phases.

Characterization and transport property measurements

SPS’d samples were mechanically polished to less than 50 micron,then mounted on f3 copper grids and Ar-ion milled in a GatanPrecision Ion Polishing System (PIPS) to electron transparencyfor TEM observation. Annular bright field (ABF) images wereacquired using FEI probe corrected Titan 80-300 S/TEM operat-ing at 300 keV. The actual composition of the target phase wasestablished by electron probe micro-analyzer (EPMA, JXA-8230).The phase compositions were characterized by using X-raydiffraction (XRD, Bruker: D8 Advance) with Cu Ka radiation.Neutron diffraction data for 5 g of S-filled skutterudite powderwas acquired on the Powgen time-of-flight instrument of theSpallation Neutron Source, Oak Ridge at 10 K and 295 K. Datareduction with Mantid utilized calibration with diamond powder,background subtraction, and vanadium normalization. TheRietveld refinements were carried out using Jana200641 in the5.2 ms to 60 ms time-of-flight range, corresponding to ad-spacing range of 0.23 to 2.5 Å. Room temperature Hallcoefficients were measured on a home-built Hall systemequipped with a 1.5 T electromagnet, with a four-probe configu-ration. The carrier concentration (nH) and Hall mobility (mH)were estimated from the measured Hall coefficient (RH) andelectrical conductivity by the relation nH = 1/eRH and mH = sRH,respectively, where e is the electron charge. Charge states forthe guests were determined by X-ray photoemission spectro-scopy (XPS) on a PHI 5000 VersaProbe system using monochro-matic Al Ka X-rays. All high resolution scans are taken with23.5 eV passing energy and step size of 0.025 eV. A low-energyelectron flood gun coupled with low-energy Ar ion neutralizerwas used while taking XPS spectra on insulating samples forcharge neutralization. The electrical conductivity (s) and Seebeckcoefficient (a) were measured simultaneously by the standardfour-probe method (ZEM-3, Ulvac-Riko) in He atmosphere. Thelow temperature transport properties were collected from 4 to300 K on rectangular bar-shaped samples (3� 3� 6 mm) using aQuantum Design physical property measurement system (PPMS)within the thermal transport option. The high-temperaturethermal conductivity (300–850 K) was calculated from themeasured thermal diffusivity (l), specific heat (Cp), and density(d) according to the relationship k = Cpld. Here the thermaldiffusivity was measured by the laser flash method using aNetzsch LFA-457 system. The specific heat was measured byNetzsch DSC 404F1 using sapphire as the reference. The densitywas measured by the Archimedes method. The electronic thermalconductivity was estimated based on the Wiedemann–Franz law,ke = LsT, where L is the Lorenz constant and taken as 2.0 �10�8 V2 K�2. The lattice thermal conductivity was deter-mined by subtracting the electronic component from the totalthermal conductivity. The overall measurement errors in theelectrical conductivity, Seebeck coefficient, thermal conducti-vity and ZT were estimated to be about �5%, �3%, �7%, and�12% respectively.

Computational method

First-principles calculations were performed with the Viennaab initio simulation package (VASP).42 Generalized gradientapproximation (GGA) functional43 and projected augmentedwave (PAW) methods44,45 were used. The formation energy, chargedensity, and electronic structure calculations for EN-element filledskutterudites were carried out based on a Co64Sb192 supercell inorder to mimic the experimentally achievable compositions.Lattice dynamics for various compounds was investigated withthe frozen phonon method46 which was implemented in thePhonopy package.47 The calculations were based on a Co8Sb24

conventional cell to better visualize the influences from theguests. We constructed a 2 � 2 � 2 supercell of fully relaxedconventional units for each compound, and calculated theHellmann–Feynman forces for the supercell with small displace-ments (1 pm for each nonequivalent atom). We set the conver-gence criteria to be 10�5 eV Å�1 for structural relaxation of theunit cell, and 10�7 eV for static calculations of displaced supercellto ensure the accuracy of phonon results. Ab initio moleculardynamic calculation was performed by the VASP code with thecanonical ensemble. A supercell with 264 atoms was used. Thesimulation temperature was set to be 300 K, and the totalsimulation time was longer than 30 ps.

Author contributions

B. D., J. R. S., S. W. and P. W. synthesized the samples andcarried out the thermoelectric properties measurements. JiongYang carried out the calculations. B. Z. and F. S. O. performedthe XPS measurements and analysis. Y. H., Y. C., C. W., S. X. M.and J. L. performed the TEM experiment and analysis. R. H. andO. G. performed the neutron diffraction measurements and analysis.J. R. S performed X-ray Rietveld analysis. B. D., Jiong Yang, J. Y.,W. Z., P. Z., X. T. and Q. Z. conceived the experiments and analyzedthe results. B. D., Jiong Yang, J. Y., W. Z. and Q. Z. wrote themanuscript and all authors participated in editing.

Acknowledgements

This work was supported by U.S. Department of Energy undercorporate agreement DE-EE005432, by GM, and by NationalScience Foundation under award number 1235535. R. H.acknowledges support by U.S. Department of Energy, Office ofScience, Basic Energy Sciences, Materials Sciences and Engi-neering Division for the neutron diffraction measurements andanalysis work. A portion of this research at ORNL’s SpallationNeutron Source was sponsored by the Scientific User FacilitiesDivision, Office of Basic Energy Sciences, U.S. Department ofEnergy. Melanie Kirkham and Ashfia Huq are acknowledgedfor assistance during neutron diffraction data acquisition.W. Z. acknowledges support by Natural Science Foundation ofChina under Grant No. 11234012 and No. 51572167, andsupport by Program of Shanghai Subject Chief Scientist(No. 16XD1401100). Jiong Yang acknowledges support by theProgram for Professor of Special Appointment (Eastern Scholar)

Energy & Environmental Science Paper

Publ

ishe

d on

19

Apr

il 20

16. D

ownl

oade

d by

Uni

vers

ity o

f W

ashi

ngto

n on

27/

04/2

016

15:0

2:01

. View Article Online

Energy Environ. Sci. This journal is©The Royal Society of Chemistry 2016

at Shanghai Institutions of Higher Learning. Q. Z. acknowl-edges support by National Basic Research Program of China(No. 2013CB632505), National Natural Science Foundation ofChina (No. 51302205), and China Postdoctoral Science Founda-tion (No. 2013M531752). The TEM work was conducted in theWilliam R. Wiley Environmental Molecular Sciences Labora-tory, a National Scientific User Facility sponsored by DOE’sOffice of Biological and Environmental Research and locatedat PNNL.

References

1 J. L. Atwood, J. E. D. Davies and D. D. MacNicol, InclusionCompounds: Physical Properties and Applications, AcademicPress, 1984.

2 E. Macia, Thermoelectric Materials: Advances and Applica-tions, CRC Press, 2015.

3 M. Christensen, A. B. Abrahamsen, N. B. Christensen,F. Juranyi, N. H. Andersen, K. Lefmann, J. Andreasson,C. R. H. Bahl and B. B. Iversen, Nat. Mater., 2008, 7,811–815.

4 M. F. Hawthorne and Z. Zheng, Acc. Chem. Res., 1997, 30,267–276.

5 M. A. Kirsanova, T. Mori, S. Maruyama, M. Matveeva,D. Batuk, A. M. Abakumov, A. V. Gerasimenko, A. V.Olenev, Y. Grin and A. V. Shevelkov, Inorg. Chem., 2013,52, 577–588.

6 X. M. Jiang, X. G. Li, Y. H. Fan, B. W. Liu, H. Y. Zeng andG. C. Guo, Dalton Trans., 2013, 42, 15475–15481.

7 A. K. Srivastava, B. Praveenkumar, I. K. Mahawar, P. Divya,S. Shalini and R. Boomishankar, Chem. Mater., 2014, 26,3811–3817.

8 L. E. Bell, Science, 2008, 321, 1457–1461.9 M. Rull-Bravo, A. Moure, J. F. Fernandez and M. Martin-

Gonzalez, RSC Adv., 2015, 5, 41653–41667.10 X. D. Li, B. Xu, L. Zhang, F. F. Duan, X. L. Yan, J. Q. Yang and

Y. J. Tian, J. Alloys Compd., 2014, 615, 177–180.11 H. Fukuoka and S. Yamanaka, Chem. Mater., 2010, 22,

47–51.12 X. Shi, J. Yang, J. R. Salvador, M. F. Chi, J. Y. Cho, H. Wang,

S. Q. Bai, J. H. Yang, W. Q. Zhang and L. D. Chen, J. Am.Chem. Soc., 2011, 133, 7837–7846.

13 X. Shi, W. Zhang, L. D. Chen and J. Yang, Phys. Rev. Lett.,2005, 95, 185503.

14 Z. G. Mei, J. Yang, Y. Z. Pei, W. Zhang and L. D. Chen,Phys. Rev. B: Condens. Matter Mater. Phys., 2008, 77,045202.

15 Y. Z. Pei, L. D. Chen, W. Zhang, X. Shi, S. Q. Bai, X. Y. Zhao,Z. G. Mei and X. Y. Li, Appl. Phys. Lett., 2006, 89, 221107.

16 L. D. Chen, T. Kawahara, X. F. Tang, T. Goto, T. Hirai,J. S. Dyck, W. Chen and C. Uher, J. Appl. Phys., 2001, 90,1864–1868.

17 X. Y. Zhao, X. Shi, L. D. Chen, W. Q. Zhang, S. Q. Bai,Y. Z. Pei, X. Y. Li and T. Goto, Appl. Phys. Lett., 2006, 89,092121.

18 G. S. Nolas, M. Kaeser, R. T. Littleton and T. M. Tritt,Appl. Phys. Lett., 2000, 77, 1855–1857.

19 T. Dahal, Q. Jie, G. Joshi, S. Chen, C. F. Guo, Y. C. Lan andZ. F. Ren, Acta Mater., 2014, 75, 316–321.

20 L. Guo, X. Xu, J. R. Salvador and G. P. Meisner, Appl. Phys.Lett., 2013, 102, 111905.

21 L. L. Xi, Y. T. Qiu, S. Zheng, X. Shi, J. Yang, L. D. Chen,D. J. Singh, J. H. Yang and W. Q. Zhang, Acta Mater., 2015,85, 112–121.

22 Y. T. Qiu, L. L. Xi, X. Shi, P. F. Qiu, W. Q. Zhang, L. D. Chen,J. R. Salvador, J. Y. Cho, J. H. Yang, Y. C. Chien, S. W. Chen,Y. L. Tang and G. J. Snyder, Adv. Funct. Mater., 2013, 23,3194–3203.

23 Y. L. Tang, Y. T. Qiu, L. L. Xi, X. Shi, W. Q. Zhang,L. D. Chen, S. M. Tseng, S. W. Chen and G. J. Snyder, EnergyEnviron. Sci., 2014, 7, 812–819.

24 L. Bertini, K. Billquist, M. Christensen, C. Gatti,L. Holmgren, B. Iversen, E. Mueller, M. Muhammed,G. Noriega, A. Palmqvist, D. Platzek, D. M. Rowe,A. Saramat, C. Stiewe, M. Toprak, S. G. Williams,Y. Zhang, in Twenty-Second International Conference onThermoelectrics, Proceedings ICT’03, IEEE La Grande Motte,France, 2003, pp. 85–88.

25 X. Y. Li, L. D. Chen, J. F. Fan, W. B. Zhang, T. Kawahara andT. Hirai, J. Appl. Phys., 2005, 98, 083702.

26 J. W. Sharp, E. C. Jones, R. K. Williams, P. M. Martin andB. C. Sales, J. Appl. Phys., 1995, 78, 1013–1018.

27 Y. Z. Pei, J. Yang, L. D. Chen, W. Zhang, J. R. Salvador andJ. H. Yang, Appl. Phys. Lett., 2009, 95, 042101.

28 T. Caillat, A. Borshchevsky and J.-P. Fleurial, J. Appl. Phys.,1996, 80, 4442–4449.

29 B. L. Liao, B. Qiu, J. W. Zhou, S. Huberman, K. Esfarjani andG. Chen, Phys. Rev. Lett., 2015, 114, 115901.

30 B. Duan, P. C. Zhai, L. S. Liu, Q. J. Zhang and X. F. Ruan,J. Mater. Sci.: Mater. Electron., 2012, 23, 1817–1822.

31 X. L. Su, H. Li, G. Y. Wang, H. Chi, X. Y. Zhou, X. F. Tang,Q. J. Zhang and C. Uher, Chem. Mater., 2011, 23, 2948–2955.

32 X. L. Su, H. Li, Q. S. Guo, X. F. Tang, Q. J. Zhang and C. Uher,J. Electron. Mater., 2011, 40, 1286–1291.

33 O. Jepsen, A. Burkhardt and O. K. Andersen, The programTB-LMTO-ASA Version 4.7., Max-Planck-Institut fur Festkor-perforschung, Stuttgart, 1999.

34 I. Opahle, K. Koepernik and H. Eschrig, Phys. Rev. B:Condens. Matter Mater. Phys., 1999, 60, 14035.

35 R. P. Hermann, V. Keppens, P. Bonville, G. S. Nolas,F. Grandjean, G. J. Long, H. M. Christen, B. C.Chakoumakos, B. C. Sales and D. Mandrus, Phys. Rev. Lett.,2006, 97, 017401.

36 S. Q. Bai, Y. Z. Pei, L. D. Chen, W. Q. Zhang, X. Y. Zhao andJ. Yang, Acta Mater., 2009, 57, 3135–3139.

37 J. Yang, L. Xi, W. Zhang, L. D. Chen and J. H. Yang,J. Electron. Mater., 2009, 38, 1397–1401.

38 D. J. Singh and W. E. Pickett, Phys. Rev. B: Condens. MatterMater. Phys., 1994, 50, 11235–11238.

39 B. Duan, P. C. Zhai, L. S. Liu, Q. J. Zhang and X. F. Ruan,J. Solid State Chem., 2012, 193, 8–12.

Paper Energy & Environmental Science

Publ

ishe

d on

19

Apr

il 20

16. D

ownl

oade

d by

Uni

vers

ity o

f W

ashi

ngto

n on

27/

04/2

016

15:0

2:01

. View Article Online

This journal is©The Royal Society of Chemistry 2016 Energy Environ. Sci.

40 B. Duan, P. C. Zhai, L. S. Liu, G. Chen, P. Li and Q. J. Zhang,J. Electron. Mater., 2013, 42, 1454–1457.

41 V. Petricek, M. Dusek and L. Palatinus, Z. Kristallogr., 2014,229, 345–352.

42 G. Kresse and J. Furthmuller, Phys. Rev. B: Condens. MatterMater. Phys., 1996, 54, 11169–11186.

43 J. P. Perdew, K. Burke and M. Ernzerhof, Phys. Rev. Lett.,1997, 77, 3865.

44 G. Kresse and J. Joubert, Phys. Rev. B: Condens. Matter Mater.Phys., 1999, 59, 1758–1775.

45 P. E. Blochl, Phys. Rev. B: Condens. Matter Mater. Phys., 1994,50, 17953–17979.

46 R. M. Martin, Electronic Structure: Basic Theory and PracticalMethods, Cambridge University Press, 2004.

47 A. Togo, F. Oba and I. Tanaka, Phys. Rev. B: Condens. MatterMater. Phys., 2008, 78, 134106.

Energy & Environmental Science Paper

Publ

ishe

d on

19

Apr

il 20

16. D

ownl

oade

d by

Uni

vers

ity o

f W

ashi

ngto

n on

27/

04/2

016

15:0

2:01

. View Article Online