IOP PUBLISHING NANOTECHNOLOGY

Nanotechnology 23 (2012) 415604 (7pp) doi:10.1088/0957-4484/23/41/415604

Spark erosion: a high production ratemethod for producing Bi0.5Sb1.5Te3nanoparticles with enhancedthermoelectric performance

P K Nguyen1, K H Lee2, J Moon1,3, S I Kim2, K A Ahn2, L H Chen3,S M Lee2, R K Chen3, S Jin1,3 and A E Berkowitz1,4

1 Materials Science and Engineering, UC San Diego, CA 92093, USA2 Advanced Materials Research Center, Samsung Advanced Institute of Technology, Yongin 446-712,Korea3 Mechanical and Aerospace Engineering, UC San Diego, CA 92093, USA4 Physics Department and Center for Magnetic Recording Research, UC San Diego, CA 92093, USA

E-mail: aberk@ucsd.edu

Received 23 July 2012, in final form 27 August 2012Published 26 September 2012Online at stacks.iop.org/Nano/23/415604

AbstractWe report a new ‘spark erosion’ technique for producing high-quality thermoelectricnanoparticles at a remarkably high rate and with enhanced thermoelectric properties. Thetechnique was utilized to synthesize p-type Bi0.5Sb1.5Te3 nanoparticles with a production rateas high as 135 g h−1, using a relatively small laboratory apparatus and low energyconsumption. The compacted nanocomposite samples made from these nanoparticles exhibit awell-defined, 20–50 nm size nanograin microstructure, and show an enhanced figure of merit,ZT , of 1.36 at 360 K. Such a technique is essential for providing inexpensive, oxidation-freenanoparticles which are required for the fabrication of high performance thermoelectricdevices for power generation from waste heat, and for refrigeration.

S Online supplementary data available from stacks.iop.org/Nano/23/415604/mmedia

(Some figures may appear in colour only in the online journal)

1. Introduction

The strong global demand for renewable energy sources andfor solid state, refrigerant-free cooling devices has generatedwidespread effort to produce cost-effective and efficientthermoelectric materials [1–3]. Thermoelectric materials candirectly convert waste heat from manufacturing processes,converting heating systems, automobile exhausts, solar andgeothermal sources, etc, into electrical power by the Seebeckeffect, in which temperature differences produce electricalpower. Conversely, the Peltier effect uses electricity forrefrigeration [4, 5]. At present, both of these phenomena haveonly limited usage for these applications. The principal reasonis the very low conversion efficiency of thermoelectric devices

as compared to the corresponding mechanical systems. Adimensionless figure of merit, ZT , measures the efficiency ofthermoelectric materials, with

ZT =s2σT

κ=

s2σT

κL + κe(1)

where T is the absolute temperature, S the Seebeck coefficient,σ the electrical conductivity and κ the thermal conductivitythat is composed of κL, the lattice thermal conductivity, andκe, the electronic thermal conductivity [4, 5]. In conventionalbulk semiconductor materials, the parameters S, σ and κeare interdependent, which makes the independent control ofthese quantities very challenging. As a result, the best bulkthermoelectric materials have possessed ZT ≈ 1 for more

10957-4484/12/415604+07$33.00 c© 2012 IOP Publishing Ltd Printed in the UK & the USA

Nanotechnology 23 (2012) 415604 P K Nguyen et al

than 40 years. At room temperature, the present commercialthermoelectric products for cooling use zone-refined bulkalloys in the (Bi,Sb)2Te3 and Bi2(Se,Te)3 families with ZTof ∼1.0 [5].

During the last decade, the field of thermoelectrics hasprogressed enormously as evidenced by numerous reportsshowing enhanced ZT in nanostructured materials [2, 3], suchas superlattices of Bi2Te3/Sb2Te3 [6] and PbSeTe/PbTe [7],nanowires [8, 9], and in bulk materials with complexstructures [10], such as lead antimony silver telluride(LAST) [11] and skutterudites [12]. It is now recognizedthat the enhanced ZT values in these nano- or complex-structured materials are mainly due to the reduced latticethermal conductivity produced by stronger phonon–boundaryscattering in nanostructures [2, 3, 13]. This approach has beenshown to be quite generic provided that the characteristicnanostructure size is smaller than the phonon mean free pathin the corresponding bulk material, which is usually dictatedby Umklapp scattering and alloy scattering.

Over the last few years, in search for a more scalableroute for making thermoelectric nanostructures, a newbulk-nanostructuring approach has been developed and hasachieved very notable success [3, 13]. This new approach isbased on the production of bulk thermoelectric compositesfrom powders containing nanoparticles and/or a high densityof grain boundaries. The key idea of the nanocompositesis to utilize the high density of grain boundaries to scatterphonons without significantly adversely affecting the chargetransport. An additional economic advantage of consolidatingnano-powders is the possibility of preparing net shapeswithout machining. The fabrication of various nanocompositematerials and their thermoelectric properties have beenreported [14–33]. Among these reports, the productionmethods of the starting thermoelectric powders containingnanoscale features can be generally categorized into twoapproaches: top-down [14, 16, 18, 20–23, 32, 33] andbottom-up [24–31].

In demonstrating a top-down method, Poudel et al re-ported the fabrication of nanocomposite p-type Bi0.5Sb1.5Te3by ball milling the bulk alloys for several hours to nanometerdimensions in inert environments and hot-pressing thesenano-powders to produce the sintered compacts [18]. Theyachieved a high ZT of 1.4 at 100 ◦C. The same groupalso reported a similar ball milling–hot pressing approachfor a variety of thermoelectric materials, including n-typeBi2Te2.7Se0.3 [34], n- and p-type SiGe [16, 33] and n-typeSi [23], among others. Another recently developed approach ismelt-spinning ribbons of the appropriate alloy, hand-grindingthem and compacting them by spark plasma sintering(SPS) [20, 21]. A similarly high ZT of ∼1.5 at 360 Kwas observed in these (Bi,Sb)2Te3 nanocomposites. Thecomprehensive characterization [20, 35] of the samplesprepared from the ball milling and melt-spinning techniquesdemonstrated that the low κL and the enhanced ZT were dueto the nanocrystalline domains present.

Besides these top-down approaches, there have been agrowing number of reports showing bottom-up routes usedto produce particles for nanocomposites. Such methods are

deemed attractive because of the tunability and uniformity ofparticle sizes, which are potentially important for additionalenhancement by achieving the optimal size for quantumconfinement effects. Several bottom-up methods have beenemployed to synthesize nanostructured thermoelectrics [6–9,24, 25], but since a reasonable yield is a prerequisitefor achieving bulk samples, the approaches for fabricatingnanocomposites are limited to solution processes [26–31].For example, Dirmyer et al reported the synthesis of bismuthtelluride nanoparticles by a solution process where the particlesize was controlled by the chain length of the cappingcompound, process time and temperature [26]. However, amain challenge to these ligand-based approaches has beenthe effective removal of this capping compound, whichotherwise impedes electrical transport. While recent workon bottom-up methods has focused on ligand removal [28]or being ligand-free [29], such techniques have yet todemonstrate a ZT higher than 0.5 at 25 ◦C, presumably dueto the hindered charge transport or perhaps particle surfaceoxidation occurring during solution processing. In addition,these processes typically require centrifugation, overnightpumping, or other tedious means of particle collection.

Despite the very important successes of the variousnano-bulk composite approaches, both the top-down andbottom-up methods have disadvantages. Although the top-down approach yields intriguing ZT , the yield of nano-powders for fabricating compacted samples is still relativelylow, and in some cases the powder fabrication process itselfcan be very energy-consuming, which implies high cost.Meanwhile, the potentially low cost and scalable bottom-upapproaches address this issue, but have yet to demonstratea ZT comparable to bulk materials. Therefore, a newtechnique capable of delivering both advantages would beextremely attractive. Such a technique would produce cleanthermoelectric nanostructures at a significantly improvedproduction rate, which would also lead to nanostructured bulkcomposites that demonstrate the high ZT achieved by the‘top-down’ approaches.

In this study, we demonstrate, for the first time,a powerful new technique for the fabrication of qualitythermoelectric powders with nano-sized particulates and grainsizes at a remarkably high synthesis rate. The process isbased on spark erosion, which has been developed to producefine powders of metals, alloys and compounds in the micronsize range [36, 37]. In this work, the spark erosion methodhas been optimized for the production of nanoparticlesof thermoelectric materials. The resultant nanostructuredcompacts using spark eroded Bi0.5Sb1.5Te3 show a highZT of 1.36 at 360 K, close to the ZT values of similaralloys prepared by other top-down techniques describedabove [18, 20]. However, spark erosion offers very significantadvantages over these other methods, particularly with respectto processing efficiency, rate of nano-powder production andscale-up potential. Furthermore, since spark erosion requiresno crucible, there is no contamination issue, and oxygen-freeprocessing conditions are readily achieved.

2

Nanotechnology 23 (2012) 415604 P K Nguyen et al

2. Methodology

2.1. The spark erosion process

In spark erosion, a charged capacitor is attached to electrodesof any conductive starting material. When the electric fieldacross these two electrodes, separated by a sufficiently smallgap, is larger than the dielectric breakdown field, the capacitordischarges producing a spark (micro-plasma) between thepieces involved. This micro-plasma, consisting of electronsand positive ions, is very hot, of order 10 000 K [36]. Thekinetic energies of the faster electrons and slower ions aredeposited on localized regions where the spark was initiated,superheating them to boiling. When the spark collapses,vaporized alloy and molten droplets are violently ejectedfrom the boiling regions and propelled through the plasmaregion into the dielectric liquid where they are very rapidlyquenched. This yields the production of nanoparticles used inthe reported thermoelectric nanocomposites.

In order to achieve a high production rate, the ‘shaker-pot’ version of the spark erosion technique was utilized, asdepicted in figure 1(a) [36]. Here, a 10 cm diameter sparkerosion cell is mounted in a double-walled, vacuum-jacketedglass container that holds the dielectric liquid. Two electrodesof the alloy of interest are mounted in the cell and connectedto the pulsed power source. ‘Charge’ pieces of the samealloy, ∼2 cm in diameter in irregular shapes, fill theperforated Delrin support, making contact with the electrodes.Figures 1(b) and (c) are photographs of the ‘shaker-pot’ andthe 10 cm cell, respectively. The glass container is vibratedso that contacts among the electrodes and charge pieces aremade and broken frequently. Thus, electrical contact betweenthe electrodes is made randomly, as are the gaps among thecharge pieces across which the sparks are generated.

Since it is essential to prevent any oxidation of theBi0.5Sb1.5Te3 nanoparticles, we chose liquid nitrogen as thedielectric. We found several key advantages to working withcryogenic liquids. The vaporizing gas and low temperaturewere sufficient to prevent any oxidation while sparking. Afterthe sparking process, the removal of the cryogenic liquidby vaporization is perhaps the simplest way of collectingnanoparticles, which would be more difficult in any otherliquid medium. The shaker-pot was enclosed by a specializedcap designed with a vent, power leads and secure access forintroducing the liquid nitrogen and new charge.

When a run was completed, the cell was removed fromthe shaker-pot, and most of the nitrogen was boiled off. Thepot was then put into the load-lock of the glove box, where theremainder of the nitrogen was pumped off. The dry powderwas then moved into the glove box with an argon atmospherepurified to ∼1 ppm oxygen. The samples were sieved witha vibrating sieve assembly to <53 µm to remove largeparticles and pieces chipped off the electrodes and charge.The oxygen content of powders handled in this manner wasextremely small, 0.44± 0.02 at.%, as determined by chemicalcomposition analysis using inert gas fusion—ASTM standardmethod.

Figure 1. The spark erosion process. (a) Sketch of the apparatusand process features. (b) The ‘shaker-pot’ container for the cell anddielectric liquid. (c) The cell, 10 cm in diameter, showing theelectrode connections to the pulsed power source, and the perforatedcharge support.

2.2. Materials

In this paper, the starting materials for spark erosion werezone-refined. High purity (>99.999%) Bi, Sb and Te granuleswere weighed according to the composition of Bi0.5Sb1.5Te3and loaded into a quartz tube of 20 mm in diameter. The tubewas vacuum sealed under 10−4 Torr, and the contents weremelted and homogeneously mixed in a rocking furnace for10 h at 1073 K, then quenched to room temperature. For thezone melting, the quenched ingot was directionally grown ata rate of 1 K min−1.

2.3. Sample measurements and preparation

From the zone-refined ingot, disks (10 mm in diameter and1 mm in thickness) for thermal diffusivity measurement and

3

Nanotechnology 23 (2012) 415604 P K Nguyen et al

bars (2 mm × 2 mm × 8 mm) for carrier transport propertyand Seebeck coefficient measurement were cut in the planeand the axial directions of the ingot, respectively.

For sintered compacts of spark eroded nanoparticles,disk-shaped bulk samples (10 mm in diameter and 13 mmin thickness) were fabricated from this powder by SPS under90 MPa and at 450 ◦C for 1 min in a vacuum. The sampledensities for these compounds were about 97% of theoreticaldensities. Disks for thermal diffusivity measurement andbars for carrier transport property and Seebeck coefficientmeasurement were cut in the planes parallel and perpendicularto the press direction, respectively (so that all three parameterswere measured in planes parallel to the press direction; seefigure S2 in the supporting information section available atstacks.iop.org/Nano/23/415604/mmedia).

For powder x-ray diffraction, samples were preparedon a glass slide in air and bound by a mixture ofDevcon R© Duco Cement and acetone. X-ray diffractionpatterns were measured in a Rigaku Geiger-flex unit with Coradiation in the Bragg–Brentano mode.

The electrical conductivity and Seebeck coefficientswere measured from 300 to 520 K by a four-point probemethod using a thermoelectric measurement system (ZEM-3,ULVAC, Japan). The thermal conductivity values (κ = ρsCpλ)were calculated from measurements taken separately; sampledensity (ρs), heat capacity (Cp) via the thermal relaxationmethod and thermal diffusivity (λ) were measured undervacuum by the laser-flash method (TC-9000, ULVAC, Japan).

3. Results and discussion

3.1. Spark erosion yield

The spark erosion nanoparticle synthesis rate was approxi-mately 135 g h−1 for clean non-oxidized Bi0.5Sb1.5Te3 alloy.This high rate, along with high ZT as shown below, isremarkable considering the small laboratory cell dimensionof only 10 cm in diameter (∼1 l in volume), indicating thatscale-up to many tons/month with either larger cell diameteror parallel operation of multiple units is quite feasible.Remarkably, the energy consumption of the spark erosionprocess is estimated to be <2.0 kWh kg−1 of Bi0.5Sb1.5Te3nanoparticles, based on the electrical power used by the pulsepower mode operation.

The TEM micrograph in figure 2(a) shows sometypical nanoparticles, illustrating their mostly sphericalnature and nanoscale dimensions. Within these mostlyspherical nanoparticles are smaller crystalline ‘sub-grains’.The spark eroded Bi0.5Sb1.5Te3 nanoparticles appear to bewell crystallized, as shown in figure 2(b). The lattice imagesin some of the nanoparticles in the TEM microstructure offigure 2(b) show their single crystal microstructure, whilesome of the larger particles are polycrystalline with ananograin structure.

Figure 2(c) shows nanoparticles clustered around largerspherical particles. These larger particles have a finegrain structure due to rapid quenching. The Bi0.5Sb1.5Te3nanoparticles synthesized by the spark erosion technique

Figure 2. Spark eroded Bi0.5Sb1.5Te3 nanoparticles andnanostructures. (a) Bright field TEM of mostly sphericalnanoparticles, showing smaller crystallized regions within thenanoparticles. (b) TEM micrograph of typical Bi0.5Sb1.5Te3nanoparticles. The lattice images indicate their single- orpolycrystalline nature. (c) SEM of mostly 20–30 nm nanoparticlesand several larger spherical particles. (d) A cluster of nanoparticles.

exhibit relatively uniform diameters in the narrow rangeof ∼10–50 nm as determined by extensive TEM analysis(see figure S1 in the supporting information available atstacks.iop.org/Nano/23/415604/mmedia) with the averageparticle diameter being ∼25 nm. Nevertheless, even furtherenhancement of ZT would result from increasing the fractionof smaller nanoparticles in figures 2(a) and (d). This is theobjective of some of our current research.

3.2. Compacted Bi0.5Sb1.5Te3 nanocomposites

Figure 3 shows TEM micrographs of sections of the SPScompacts made from the spark eroded powder. Figure 3(a)

4

Nanotechnology 23 (2012) 415604 P K Nguyen et al

Figure 3. TEM micrographs of sections of the SPS compacts madefrom the spark eroded Bi0.5Sb1.5Te3 nanoparticles.

shows that the mostly spherical nanoparticles are convertedby the SPS into flat-sided parallelepiped grains. It isnoteworthy that the sintering has not increased the grainsize significantly over the original nanoparticles’ dimensions.The higher magnification TEM in figure 3(b) indicates thatmany coherent grain boundaries are present in the sinteredcompacts.

Figure 4 shows the x-ray diffraction patterns of thestarting bulk ingot, the <53 µm spark eroded powder andthe SPS compact made from the powder. Only lines ofBi0.5Sb1.5Te3 are present (the shoulder on the 33◦ line ofthe powder is not present in the compact, and remains anunexplained feature). The powder and compact patterns showrandom orientations. Random orientation of the crystallites’axes is necessary to ensure isotropy in thermoelectric devices,a very desirable feature.

Figure 4. X-ray diffraction patterns with Co radiation of the bulkstarting ingot, the <53 µm spark eroded powder and the SPScompact made from that powder, compared to the standardBi0.5Sb1.5Te3 XRD pattern shown at the top.

The transport properties of the starting bulk alloy and theSPS compact are shown in figure 5. We note that the propertiesof a sample of the zone-refined starting material may bestrongly influenced by its structural anisotropy. To examinethat issue, Bi0.5Sb1.5Te3 ingots were milled at 200 rpm for10 h in a N2 atmosphere using a planetary ball mill. The milledpowder was sieved to obtain <45 µm diameter particles. Thispowder was then compacted under the same conditions asthe spark eroded powder, and the transport properties weremeasured on disks and rods of the same sizes as used for thestarting material and the SPS compacts of the spark erodedpowder. The properties were very similar to those of thesample of starting material.

As shown in figure 5, the electrical conductivity of thespark eroded samples is close to that of the starting bulk ingots(figure 5(a)), while the Seebeck coefficient is lower below(higher above) 400 K than that of the ingot (figure 5(b)).Overall, the power factor (S2σ ) values are comparable to thoseof the ingot below 360 K, and slightly higher above 360 K.The electrical transport properties of our spark eroded samplesare very similar to those of the Bi0.5Sb1.5Te3 nanocompositereported by Poudel et al [18], showing the comparable qualityof nanoparticles fabricated by spark erosion, e.g, with minimaloxidation.

The thermal conductivity of the spark eroded samplesis significantly lower than that of the ingot (figure 5(d)),which leads to a substantial increase in ZT over the entiretemperature range of 300–520 K (figure 5(f)). The peak ZTof 1.36 at 360 K is significantly higher than that of ∼1.0 at300 K for the bulk ingot, and is close to the highest reportedZT values in Bi0.5Sb1.5Te3 nanocomposites fabricated byball milling as well as melt-spinning (1.4 at 100 ◦C byPoudel et al [18] and ∼1.5 at 360 K by Xie et al [20]).Above 360 K, the ZT of the spark eroded samples decreaseswith temperature as a result of the decreasing electricalconductivity and Seebeck coefficient as well as the increasingthermal conductivity. The electrical conductivity decreasesbecause of enhanced carrier-phonon scattering at highertemperature. Furthermore, the total thermal conductivity isincreased due to bipolar diffusion of carriers above 360 K,which is caused by the thermal excitation of minority carriers

5

Nanotechnology 23 (2012) 415604 P K Nguyen et al

Figure 5. Thermoelectric properties of the starting bulk alloy and the spark eroded and SPS samples, as indicated by their respectivemarkers. (a) Electrical conductivity, (b) Seebeck coefficient, (c) power factor, (d) total thermal conductivity, (e) lattice thermal conductivity,(f) ZT value.

(electrons in this case) in narrow band-gap semiconductorssuch as Bi0.5Sb1.5Te3 [10]. As shown in figure 5(b), theonset of bipolar thermal diffusion also leads to a reducedSeebeck coefficient because of the coexistence of electronsand holes. A similar temperature dependence of the transportproperties due to bipolar diffusion has also been observed innanostructured Bi0.5Sb1.5Te3 [18, 20].

The enhanced ZT of the spark eroded Bi0.5Sb1.5Te3

nanoparticles after sintering is primarily due to the reducedthermal conductivity from 300 to 520 K, and is partially due tothe slightly increased power factor above 400 K. To elucidatethis, we calculated the lattice thermal conductivity (κL) ofboth samples (figure 5(e)) via the following relationship:κL = κ − κe, where the electronic thermal conductivity (κe)is estimated from the Wiedemann–Franz law, κe = LTσ ,

and the Lorenz number L0 = 2.0 × 10−8 V2 K−2, typicalfor a heavily doped semiconductor [21]. Figure 5(e) showsthat the κL in the spark eroded samples is reduced byabout 50% over the entire temperature range, indicatingstronger phonon scattering at nanograin interfaces, which isthe main reason for the enhanced ZT . It is worth notingthat the calculated κL of our spark eroded samples isvery similar to those of nanocomposite Bi–Sb–Te samplesreported in [18, 20] if the same Lorenz number is used (seefigure S3 in the supporting information section available atstacks.iop.org/Nano/23/415604/mmedia), indicating similarphonon and charge transport behavior, presumably due to thesimilar microstructures. Note that the calculated κL shownin figure 5(e) also includes the contribution from the bipolarthermal conduction [10, 20, 35], and may not represent

6

Nanotechnology 23 (2012) 415604 P K Nguyen et al

the true lattice thermal conductivity [35]. Poudel et al [18]modeled κL based on the Boltzmann equation and showedthat it actually decreases with temperature, and is reducedby a factor of two in nanocomposites when compared tobulk. In our spark eroded and bulk samples, the fact thatthe peaks of the Seebeck coefficient and the minima ofthe total thermal conductivity occur at similar temperatures(360–400 K) suggests that bipolar thermal and electricalconduction becomes important above 360 K [18, 20], andthe higher Seebeck coefficient in the spark eroded samplesindicates that the bipolar conduction is suppressed, which alsocontributes to the reduced κ and the enhanced ZT [20, 35].

4. Summary

We have developed a uniquely efficient, cost-effective methodfor the production of thermoelectric nanoparticles at a highproduction rate. These nanoparticles were spark plasmasintered into nanograined thermoelectric alloy with a ZTcomparable to the best reported values. The improved ZTis a result of the reduced lattice thermal conductivity, aswell as the suppressed high temperature bipolar contributionsto the electronic thermal transport. Spark erosion not onlyproduces these clean, oxidation-free, high-ZT thermoelectricnanoparticles directly at low energy consumption, but therate of production is very high, indicating that scale-upto many tons/month is quite feasible. Furthermore, it isanticipated that the use of this technique will be able to beextended to various other fields where nanoparticle productionis a challenge. With the inherent nanoparticle synthesisand nanograined structure using spark eroded thermoelectricalloys, the prospects for further improvements in performanceand production rate appear to be promising.

Acknowledgments

The authors acknowledge very useful discussions withDr David Smith at Arizona State University and Dr FredSpada at UC San Diego.

References

[1] Bell L E 2008 Science 321 1457–61[2] Vineis C J, Shakouri A, Majumdar A and

Kanatzidis M G 2010 Adv. Mater. 22 3970–80[3] Lan Y C, Minnich A J, Chen G and Ren Z F 2010 Adv. Funct.

Mater. 20 357–76[4] Goldsmid H J 1964 Thermoelectric Refrigeration (The

International Cryogenics Monograph Series) (New York:Plenum) p 240

[5] Rowe D M 1995 CRC Handbook of Thermoelectrics(Boca Raton, FL: CRC Press) p 701

[6] Venkatasubramanian R, Siivola E, Colpitts T andO’Quinn B 2001 Nature 413 597–602

[7] Harman T C, Taylor P J, Walsh M P and LaForge B E 2002Science 297 2229–32

[8] Boukai A I, Bunimovich Y, Tahir-Kheli J, Yu J K,Goddard W A and Heath J R 2008 Nature 451 168–71

[9] Hochbaum A I, Chen R K, Delgado R D, Liang W J,Garnett E C, Najarian M, Majumdar A and Yang P D 2008Nature 451 163–7

[10] Snyder G J and Toberer E S 2008 Nature Mater. 7 105–14[11] Hsu K F, Loo S, Guo F, Chen W, Dyck J S, Uher C, Hogan T,

Polychroniadis E K and Kanatzidis M G 2004 Science303 818–21

[12] Morelli D T, Tritt T M and Nolas G S 1999 Annu. Rev. Mater.Sci. 29 89–116

[13] Dresselhaus M S, Chen G, Tang M Y, Yang R G, Lee H,Wang D Z, Ren Z F, Fleurial J P and Gogna P 2007 Adv.Mater. 19 1043–53

[14] Ebling D G, Jacquot A, Bottner H, Kirste L, Schmidt J andAguirre M J 2009 Electron. Mater. 38 1450–5

[15] He J Q, Sootsman J R, Girard S N, Zheng J C, Wen J G,Zhu Y M, Kanatzidis M G and Dravid V P 2010 J. Am.Chem. Soc. 132 8669–75

[16] Joshi G et al 2008 Nano Lett. 8 4670–4[17] Kim W, Zide J, Gossard A, Klenov D, Stemmer S,

Shakouri A and Majumdar A 2006 Phys. Rev. Lett.96 045901

[18] Poudel B et al 2008 Science 320 634–8[19] Tritt T M, Gothard N, Ji X and He J 2008 J. Appl. Phys.

103 054314[20] Xie W J et al 2010 Nano Lett. 10 3283–9[21] Xie W J, Yan Y G, Tang X F, Zhang Q J and Tritt T M 2009

J. Appl. Phys. 105 113713[22] Zhu G H et al 2009 Phys. Rev. Lett. 102 196803[23] Bux S K, Blair R G, Gogna P K, Lee H, Chen G,

Dresselhaus M S, Kaner R B and Fleurial J P 2009 Adv.Funct. Mater. 19 2445–52

[24] Wang R Y, Feser J P, Lee J S, Talapin D V, Segalman R andMajumdar A 2008 Nano Lett. 8 2283–8

[25] Zhou W, Zhao W, Lu Z, Zhu J, Fan S, Ma J, Hng H H andYan Q 2012 Nanoscale 4 3926

[26] Dirmyer M R, Martin J, Nolas G S, Sen A andBadding J V 2009 Small 5 933–7

[27] Scheele M, Oeschler N, Meier K, Kornowski A, Klinke C andWeller H 2009 Adv. Funct. Mater. 19 3476–83

[28] Scheele M, Oeschler N, Veremchuk I, Reinsberg K G,Kreuziger A M, Kornowski A, Broekaert J, Klinke C andWeller H 2010 ACS Nano 4 4283–91

[29] Zhang Y C, Wang H, Kraemer S, Shi Y F, Zhang F,Snedaker M, Ding K L, Moskovits M, Snyder G J andStuck G D 2011 ACS Nano 5 3158–65

[30] See K C, Feser J P, Chen C E, Majumdar A, Urban J J andSegalman R A 2010 Nano Lett. 10 4664–7

[31] Wang R Y, Feser J P, Gu X, Yu K M, Segalman R A,Majumdar A, Milliron D J and Urban J 2010 Chem. Mater.22 1943–5

[32] Ma Y, Hao Q, Poudel B, Lan Y C, Yu B, Wang D Z,Ren Z F and Chen G 2008 Nano Lett. 8 2580–4

[33] Wang X W et al 2008 Appl. Phys. Lett. 93 193121[34] Chen G, Yan X A, Poudel B, Ma Y, Liu W S, Joshi G,

Wang H, Lan Y C, Wang D Z and Ren Z F 2010 Nano Lett.10 3373–8

[35] Lan Y C, Poudel B, Ma Y, Wang D Z, Dresselhaus M S,Ren Z F and Chen G 2009 Nano Lett. 9 1419–22

[36] Berkowitz A E and Walter J L 1987 J. Mater. Res. 2 277–88[37] Carrey J, Radousky H B and Berkowitz A E 2004 J. Appl.

Phys. 95 823–9

7