0 Bewertungen0% fanden dieses Dokument nützlich (0 Abstimmungen)

76 Ansichten12 SeitenMar 22, 2011

© Attribution Non-Commercial (BY-NC)

PDF, TXT oder online auf Scribd lesen

Attribution Non-Commercial (BY-NC)

Als PDF, TXT **herunterladen** oder online auf Scribd lesen

0 Bewertungen0% fanden dieses Dokument nützlich (0 Abstimmungen)

76 Ansichten12 SeitenAttribution Non-Commercial (BY-NC)

Als PDF, TXT **herunterladen** oder online auf Scribd lesen

Sie sind auf Seite 1von 12

JOURNAL OF APPLIED PHYSICS 108, 121101 2010

APPLIED PHYSICS REVIEWS–FOCUSED REVIEW

J. Martin,

1, a

T. Tritt, ^{2} and C. Uher ^{3}

^{1} Materials Science and Engineering Laboratory, National Institute of Standards and Technology,

Gaithersburg, Maryland 20899, USA

2

Department of Physics and Astronomy, Clemson University, Clemson, South Carolina 29634, USA ^{3} Department of Physics, University of Michigan, Ann Arbor, Michigan 48109, USA

Received 18 May 2010; accepted 27 September 2010; published online 22 December 2010

We present an overview of the challenges and practices of thermoelectric metrology on bulk materials at high temperature 300 to 1300 K . The Seebeck coefﬁcient, when combined with thermal and electrical conductivity, is an essential property measurement for evaluating the potential performance of novel thermoelectric materials. However, there is some question as to which measurement technique s provides the most accurate determination of the Seebeck coefﬁcient at high temperature. This has led to the implementation of nonideal practices that have further complicated the conﬁrmation of reported high ZT materials. To ensure meaningful interlaboratory comparison of data, thermoelectric measurements must be reliable, accurate, and consistent. This article will summarize and compare the relevant measurement techniques and apparatus designs required to effectively manage uncertainty, while also providing a reference resource of previous advances in high temperature thermoelectric metrology. © 2010 American Institute of Physics. doi:10.1063/1.3503505

TABLE OF CONTENTS

I. INTRODUCTION

A. Thomas Seebeck and thermoelectricity

B. TE materials and applications

II. METHODS OF MEASURING THE SEEBECK COEFFICIENT

A. The Seebeck coefﬁcient

B. Arrangement and general requirements

C. The integral method

D. The differential method and its conditions

1. The steady-state condition

2. The quasi-steady-state condition

3. The transient or ac condition

E. The absolute TE scale

III. METHODS OF MEASURING

IV. INSTRUMENTATION

.

V. SUMMARY.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

1

1

2

3

3

3

4

5

5

6

7

8

8

9

10

11

I. INTRODUCTION A. Thomas Seebeck and thermoelectricity

The earliest thermoelectric TE “measurement” was re- ported at a session of the Berlin Academy of Sciences on August 16, 1821, by Thomas Johann Seebeck b. 1770, d. 1831 , a graduate of Göttingen University in medicine and practical surgery. ^{1} Seebeck formed a bimetallic loop of cop-

^{a} ^{} Electronic mail: joshua.martin@nist.gov.

per and bismuth wires, then upon heating one end observed the deﬂection of a magnetic needle, describing this observa- tion as “the magnetic polarization of metals and ores pro- duced by a temperature difference”—thermomagnetism Fig.

1 . ^{2} ^{,}^{3} This term was actually coined in 1823 after subsequent investigations by von Yelin, ^{4} who observed a similar effect when heating open circuit single metal elements. It was Ye- lin’s intent to distinguish his observations from those of See- beck. However, the term became ubiquitous in describing the range of phenomena. Seebeck concluded the effect was ex- plicitly magnetic, objecting to contemporaries who suggested it was a pure extension of Hans Christian Ørsted’s electro- magnetic phenomenon. In March of 1823, Ørsted himself redirected the conceptualization of those experiments: “See- beck has discovered that one can establish an electric circuit in metals without the interposition of any liquid. One estab-

FIG. 1. Drawing depicting Seebeck’s experimental instrumentation.From Adolphe Ganot, Traité élémentaire de physique expérimentale et appliquée et de meteorologie: illustré de 568 belles gravures sure bois intercalées dans le texte Paris: L’auteur-éditeur, 1859 , p. 707.

0021-8979/2010/108 12 /121101/12/$30.00

108 , 121101-1

© 2010 American Institute of Physics

Downloaded 21 Mar 2011 to 139.143.5.160. Redistribution subject to AIP license or copyright; see http://jap.aip.org/about/rights_and_permissions

121101-2

Martin, Tritt, and Uher

FIG. 2. Color online The Seebeck effect for two dissimilar materials A and B with the interfaces at temperatures T _{1} and T _{2} , such that T _{1} T _{2} and a proportional voltage is generated.

lishes the current in this circuit by disturbing the equilibrium of temperature.” Ørsted thus proposed a new terminology: “it will from now on doubtless be necessary to distinguish this new class of electric circuits by an appropriate term; and as such I propose the expression TE circuit.” ^{5} By the 1840s, it was tacitly assumed Seebeck discovered thermoelectricity, although serendipitously. His ﬁrst paper on the subject ^{1} de- tailed “bismuthlike” behavior in many ores including PbS a lead chalcogenide and the skutterudite Co,Ni As _{3}_{−}_{x} , mate- rials that are still heavily investigated today. TE phenomena enable the solid-state interconversion of thermal and electrical energy. This conversion occurs through two primary phenomena: the Seebeck and the Peltier effect. The Seebeck effect describes the electric potential emergent across a conductor within a thermal gradient. The value of this ratio yields the Seebeck coefﬁcient

S _{a}_{b} = lim

T → 0

V ab T

,

1

where V _{a}_{b} is the electric potential across the interface of two dissimilar conductors within a thermal gradient T , as illustrated in Fig. 2 . According to this deﬁnition, the mea- sured Seebeck coefﬁcient S _{a}_{b} requires the correction S _{a}_{b} = S _{b} − S _{a} , where S _{b} is the second conductor’s contribution, to obtain S _{a} , the sample’s Seebeck coefﬁcient. The terms “ther- mopower” or “TE power,” while technically imprecise, are commonly substituted for the Seebeck coefﬁcient. This latter term was originally recommended by the joint AIEE-IRE Task Group on TE Devices in the late 1950s for consistency with the Peltier and Thomson coefﬁcients. ^{6} ^{,} ^{7} In the corre- sponding Peltier effect, passing an electric current through the Ohmic interface of two dissimilar conductors results in the liberation or the absorption of heat at the junction. ^{8} This arises from the difference in Fermi energies E _{F} beyond the Ohmic interface, whereby heat is absorbed liberated as an electron passes from a lower higher E _{F} than that of the connecting material. It is, therefore, dependent on the current direction. The rate of thermal exchange at each junction is given by

Q _{P} = S _{a}_{b} IT = _{a}_{b} I ,

2

where I is the current through the junction at temperature T and _{a}_{b} is the Peltier coefﬁcient of the junction. Methods of measuring the Peltier coefﬁcient are described in references. ^{9} ^{,} ^{1}^{0}

J. Appl. Phys. 108 , 121101 2010

B. TE materials and applications

While the low conversion efﬁciency of TE devices cur- rently limits their widespread commercial practicality to niche applications, they remain an integral component in NASA’s radioisotope TE generators for deep space power generation, small-scale waste-heat recovery devices, tem- perature measurement, and in electronic refrigeration. In ad- dition, TE devices are environmentally friendly, require minimal maintenance no moving parts , and reliably offer quiet and compact operation. Higher efﬁciency TE materials would enable the practi- cal solid-state conversion of thermal and electrical energy. The dimensionless ﬁgure of merit, ZT = S ^{2} T / , deﬁnes the effectiveness of these TE materials, where S is the Seebeck coefﬁcient, the electrical conductivity, T the absolute tem- perature, and the total thermal conductivity = _{L} + _{e} ; the lattice and electronic contributions, respectively . ^{1}^{1} ^{–} ^{1}^{7} Opti- mizing these interdependent physical parameters to achieve acceptable efﬁciencies where the ZT 1 requires materials exhibiting a unique combination of properties. In addition, these materials must exhibit an optimal ZT within speciﬁc target temperatures. Several approaches have been developed to identify these potential TE materials for high temperature energy conversion applications. ^{1}^{4} ^{–} ^{1}^{7} One approach, proposed by Slack, ^{1}^{8} suggests a phonon-glass electron-single-crystal PGEC . An ideal PGEC material would possess thermal properties similar to an amorphous material low thermal conductivity with anomalous temperature dependence and electrical properties similar to a good single crystal semicon- ductor. This approach has proven successful in both the skutterudite ^{1}^{9}^{–} ^{2}^{2} and clathrate ^{2}^{3}^{–} ^{2}^{5} crystal structures, with power factor S ^{2} optimization provided by adjusting the carrier concentration. The recent progress in identifying several higher efﬁ- ciency ZT 2 TE materials can be attributed to nanoscale enhancement. Many of these materials demonstrate increased S ^{2} and decreased thermal conductivity due to the phenom- enological properties of nanometer length scales, including quantum conﬁnement effects increased density of states , enhanced phonon scattering, and charge carrier ﬁltering. Candidate systems include thin ﬁlms, heterostructures, super- lattices, nanowires, granular nanocomposites, and bulk ma- terials with self-formed nanoinclusions. ^{2}^{6} ^{–} ^{3}^{5} The continued development of novel TE materials in the last two decades has given rise to interdisciplinary ap- proaches, where solid-state chemistry, materials, and theoret- ical physics form a collaborative framework with measure- ment science. This has led to the rapid incorporation of new concepts and directions in materials synthesis and engineer- ing. To fully evaluate the structure-property relationships and to illuminate the underlying physics of these new systems requires thorough characterization of the electrical and ther- mal transport properties. Unfortunately, many of the mea- surement techniques developed in the early part of the 20th century and during the 1950s and 1960s are at risk of being overlooked by this new generation of researchers. We note there are excellent general reviews on TE

121101-3

Martin, Tritt, and Uher

FIG. 3. Color online Ideal energy band diagrams representing electronic conduction for a metal and for n- and p-type semiconductors. The average energy of the carriers is illustrated by the brackets.

measurements. ^{3}^{6} ^{–} ^{4}^{0} However, there are few reports speciﬁ- cally on Seebeck coefﬁcient measurements, and fewer for those at high temperature. The Seebeck coefﬁcient is an essential physical property in evaluating the potential performance of TE materials, as it is highly sensitive to the electronic structure. However, the deceptive simplicity of measuring the Seebeck coefﬁcient has often promoted poor apparatus design and the implemen- tation of inadequate measurement techniques, especially at high temperature, where nonstandard practices have compli- cated the conﬁrmation of reported high ZT materials. There- fore, this article will summarize and compare the methods of measuring the Seebeck coefﬁcient in an effort to stimulate a mindful approach to apparatus design and instrumentation. To ensure meaningful interlaboratory comparison of data, these measurements must be reliable, accurate, and consis- tent. The following sections will summarize the relevant measurement techniques, arrangement, and instrumentation required to effectively manage uncertainty, while also pro- viding a reference resource albeit nondeﬁnitive of previous advances in high temperature Seebeck coefﬁcient metrology. Detailed apparatus descriptions will be left to the original literary references. The bulk measurement of resistivity will also be introduced, as this characterization commonly ac- companies Seebeck coefﬁcient measurements.

II. METHODS OF MEASURING THE SEEBECK

COEFFICIENT

A. The Seebeck coefﬁcient

The Seebeck coefﬁcient is essentially a measure of the transport entropy per particle and is proportional to the av- erage energy per carrier, relative to the Fermi energy E _{F} , divided by charge per carrier e and temperature T Refs. 15, 16, 41, and 42

S

1 _{e}_{T} E − E _{F} .

3

Metals typically exhibit smaller Seebeck coefﬁcients since the charge carriers participating in electrical transport are those with energies k _{B} T , resulting in a small energy per carrier. Semiconductors typically exhibit larger Seebeck co- efﬁcients since carriers excited from dopant states within the gap into the conduction band have a greater average energy per carrier Fig. 3 . From a physical perspective, the Seebeck

J. Appl. Phys. 108 , 121101 2010

coefﬁcient generally comprises two components: one diffu- sive, arising from the spatial redistribution of the charge car- riers, and another arising from the interaction of the charge carriers with the nonequilibrium portion of the phonon dis- tribution. For metals in a thermal gradient, the contrasting thermal diffusion of the charge carriers in each material relative to their Ohmic interface establishes a voltage difference. This electric potential provides the equilibrium necessary to pre- vent further net charge transfer, resulting in the exclusive transport of kinetic energy. ^{4}^{3}^{–} ^{4}^{5} Thus, measurement of the Seebeck voltage requires the open circuit condition J = 0 where J is the electrical current density , such that the junc- tions are isothermal and the electrochemical potential is continuous. ^{4}^{3} In the Harman method or Z meter , the Ohmic voltage is measured under both isothermal and adiabatic con- ditions, such that J 0 and the temperature difference forms due to the induced Peltier effect at the contacts. Here, the Seebeck coefﬁcient is obtained indirectly from the voltage data under both conditions, provided the temperature differ- ence is known. ^{4}^{6} ^{–} ^{4}^{9} In n-type p-type semiconductors, the electric potential establishes in the opposite direction same direction of the thermal gradient resulting in a negative positive Seebeck coefﬁcient. This convention ensures agreement between the Seebeck coefﬁcient and the sign of the carriers. However, the Seebeck coefﬁcient is not explic- itly indicative of the majority carrier, and consequently, mea- surements of n carrier concentration and R _{H} Hall coefﬁ- cient are required to fully evaluate the electronic

_{s}_{t}_{r}_{u}_{c}_{t}_{u}_{r}_{e}_{.} 15, 36, 38, 50 , 51

B. Arrangement and general requirements

While the Seebeck coefﬁcient is not geometry dependent for isotropic materials , the spatial arrangement of the probes may affect the result. The two primary arrangements developed in part from early thermal conductivity measurements, ^{5}^{2} as both quantities S and require the mea- surement of T under an applied axial heat ﬂux. By measur- ing the emergent electric potential, the Seebeck coefﬁcient is readily calculated. In the axial-ﬂow technique two-probe , the temperature difference and the electric potential are mea- sured on the probes which are in direct contact with the ends of the specimen Fig. 4 . This is the arrangement preferred by Goldsmid and Tritt for improved thermal and electrical contact. ^{3}^{8}^{,} ^{5}^{3}^{,} ^{5}^{4} However, many Seebeck coefﬁcient apparatus also concurrently measure resistivity, requiring additional voltage contacts away from the ends of the sample. In the potentiometric arrangement or four-probe , the temperature difference and the voltage difference are measured at two points on the sample or inserted within the sample equidis- tant from the hot and cold sinks and on the axis parallel to the thermal gradient. To maintain accuracy, the diameter of each temperature/voltage probe must be much smaller than the effective distance between them. There is some debate as to which method provides the most accurate determination of temperature and voltage dif- ference at the same points. These assertions are often stated

121101-4

Martin, Tritt, and Uher

FIG. 4. Color online Diagram comparing the potentiometric four-probe and the axial-ﬂow two-probe arrangements, where T _{1} T _{2} .

arbitrarily with little comparative evidence to substantiate either arrangement. It would therefore be prudent to develop an apparatus capable of in situ comparison. An apparatus developed by Bowers in 1959 Ref. 6 averaged the results obtained by both of these arrangements. The Seebeck coef- ﬁcients were consistent within 10% up to 800 °C. However, there was no estimation made of the overall measurement uncertainty, so this result may be regarded as preliminary. Wood ^{5}^{5} also compared the results of thermocouples pressed on the ends of a sample with those obtained by inserting them in holes drilled in the sample. For the temperature and the material evaluated, the results were consistent within the measurement uncertainty. It would be useful to evaluate these arrangements over a broad temperature range and for a variety of materials. Considering the diversity of research materials in both geometry and contact resistance, the poten- tiometric arrangement may beneﬁt longer, lower cross- sectional area specimens, while the axial-ﬂow arrangement may beneﬁt shorter or disk-shaped specimens, provided the thermal resistance of the sample is larger than the thermal resistance of the contact interface. Thus, a versatile and ﬂex- ible apparatus would not only enable comparative measure- ments on materials with different contact resistances, thermal conductivities, and heat capacities, but also expand the prac- tical range of sample geometries. The primary requirements for “good” Seebeck coefﬁ- cient measurements are: 1 the spatially synchronous mea- surement of voltage and temperature, that is, the voltage and temperature must be measured at the same location and at the same time; 2 probes in very good thermal and electrical contact with the specimen; and 3 the acquisition of low voltages microvoltages with minimal extraneous contribu- tions. Since at high temperature thermocouples are used to determine the temperature difference, where one pair of identical wires probe the TE voltage, measurement of the Seebeck coefﬁcient is essentially a practice in low voltage measurements. The various methods are best deﬁned by the behavior of state variable T and by the gradient heating tech- nique. However, a literature survey is challenging due to

J. Appl. Phys. 108 , 121101 2010

FIG. 5. Color online Graphical illustrations of the integral and the differ- ential Seebeck coefﬁcient measurement methods.

inconsistencies in the use of terms—e.g., steady-state, a lack of consensus on time scale deﬁnition, poor estimations of uncertainty, previous lack of Seebeck coefﬁcient standards, and minimal detail in published high temperature apparatus descriptions, when they are published. For material a whose junctions are Ohmic and isother- mal with material b , both of which are chemically and physi- cally homogeneous and isotropic, the electric potential is de- ﬁned as ^{4}^{1}

V _{a}_{b} T _{1} , T _{2} = _{} T T ^{2} S _{a}_{b} T dT = _{} T T 1 ^{2} S _{b} T − S _{a} T dT , 4

1

where S _{a} T is the absolute Seebeck coefﬁcient of the sample being measured and S _{b} T is the known Seebeck coefﬁcient of the reference leads. It is further assumed that materials a and b are those in which the empirical Law of Magnus ^{5}^{6} is valid, such that V _{a}_{b} is explicitly a function of T _{1} , T _{2} , and is independent of the temperature distribution between the in- terfaces. This is invalid for chemically and/or physically in- homogeneous materials. Despite some experimental efforts, notably by Benedicks, ^{5}^{7} to prove otherwise, the Magnus Law has been veriﬁed for extremely homogeneous metals. ^{5}^{8} How- ever, both theoretical ^{5}^{9} and experimental ^{6}^{0} examination sug- gest steep thermal gradients in nondegenerate semiconduc- tors may generate a nonequillibrium n T and therefore inﬂuence V _{a}_{b} . See Domenicali ^{4}^{3} for a complete discussion on anisotropic transport coefﬁcients and volume and surface heating effects on the Seebeck coefﬁcient.

C. The integral method

There are two primary techniques used to measure the relative Seebeck coefﬁcient: the integral and the differential Fig. 5 . In the integral method or large T , one end of the specimen is maintained at a ﬁxed temperature T _{1} while the opposite end is varied through T _{2} = T _{1} + T , the temperature range of interest. ^{3}^{9}^{,} ^{4}^{1}^{,} ^{6}^{1}^{,} ^{6}^{2} An appropriate analytic approxima- tion is selected and applied to the entire data set V _{a}_{b} T _{1}_{,} T _{2} , then differentiated with respect to T _{2} , where Eq. 4 becomes

S _{a}_{b} T _{2} = S _{b} T _{2} − S _{a} T _{2} = ^{d}^{V} ^{a}^{b} ^{} ^{T} ^{1} ^{,} ^{T} ^{2} ^{} . dT _{2}

5

The ﬁtting method must comprise the data set with minimal oscillations, as small random and biased errors will be am- pliﬁed in the derivative. To this end, various curve ﬁtting techniques have been explored. One common analytical rep-

121101-5

Martin, Tritt, and Uher

resentation is to continuously approximate the data by a se- ries of orthogonal coefﬁcients with the order selected by an F test. ^{6}^{2} The mathematical oscillations can be further mini- mized by approximating the derivative in Eq. 5 directly and evaluating the derivative at sliding partial intervals where the standard deviation is a minimum. Kettler ^{6}^{3} used this proce- dure to calibrate Chromel and Constantan thermocouple ref- erence wires with a pure Pb reference. The integral method developed naturally from the prac- tices of early TE experiments, where Seebeck’s techniques were easily extended to higher temperatures. For example, in 1865 Stefan ^{6}^{4} measured the TE potential of some chalco- genides galenites and cuprous oxide ore couples by heating one interface with an alcohol ﬂame. Additional examples of these early experiments can be found in Ref. 65. The next notable advance to this form of measurement was the work of Tait, a friend and colleague of Thomson Lord Kelvin . It was Thomson who ﬁrst devised the theo- retical relations to relate the three fundamental TE phenomena. ^{6}^{6} In effort to establish a graphical representation of these relationships, Thomson suggested a chart consisting of parallel columns, each ordering by temperature the rela- tive position of a number of metals. Lines drawn between successive columns indicate how the relative position of each material changes with temperature. Expanding on this suggestion, Tait then developed his well known TE diagram, in which the abscissa indicate absolute temperatures, and the TE voltage is given by the difference of the ordinates at a speciﬁc temperature of the lines for any two metals. There- fore, a drawing made through experimental measurements could graphically indicate the relative magnitude and direc- tion of the TE electric potential of the metals, their tempera- ture dependence, as well as representations of the speciﬁc heat of electricity and the magnitude of the Peltier effect. ^{6}^{7} Inspired by Tait’s analysis on the temperature dependent behavior of TE voltage, Dewar and Fleming ^{6}^{8} compiled an extensive experimental tabulation of the temperature depen- dence of other couples 200 and 100 °C . These include Pt, Au, Pd, Ag, Cu, Zn, Cd, Mg, Sn, Fe, Ni, Steel, Mn–Steel, Manganin, Bi, Sb, Al, and C, relative to Pb. These data en- abled measurements throughout the next several decades as a comparative reference for other semiconductors and oxides at higher temperatures. One early implementation of the integral technique at higher temperature was given by Wick in 1907. ^{6}^{9} The See- beck coefﬁcient of silicon rods 30 cm long and 3 mm in diameter were measured from 190 to 350 °C relative to copper. In an effort to establish good electrical and thermal interfaces, contacts to copper and constantan wires compris- ing the thermocouple were made to each sample end by the formation of an electroplated copper bridge. Copper wires were then soldered to each of these wires and the terminals submerged in an isothermal mercury bath for connection to a potentiometer. One end of the rod was maintained isothermal at 0 °C by inserting it within a metal tube submerged in a Dewar bulb ﬁlled with a freezing mixture ice, “liquid air,” or liquid oxygen were available at the time . The other end was enclosed within an iron wire coil wrapped around an asbestos covered porcelain tube. Corrections for nonisother-

J. Appl. Phys. 108 , 121101 2010

mal conditions were made by comparing the TE voltage of the copper–silicon and the copper-constantan interfaces. The resulting curves were ﬁt and differentiated in 20° intervals. In a modern embodiment, Wood ^{6}^{2} developed an appara- tus to measure the Seebeck coefﬁcient up to 1200 K. The sample is pressed between a molybdenum heater and a water cooled copper plate, where springs external to the furnace maintain the contacting pressure. The temperature of each end is measured by a W–Nb thermocouple not welded em- bedded within an alumina tube for the cold side and a sap- phire button for the hot junction. Measurements are con- ducted under an inert atmosphere 10 ^{−}^{5} Pa . The Seebeck coefﬁent is obtained by differentiating an analytical approxi- mation of the measured voltages. An additional detailed ap- paratus description of the integral method can be found in Ref. 70. The integral method succeeds in approximating TE op- erating conditions and in minimizing the inﬂuence of voltage offsets, due to the large thermal gradients and subsequently larger voltage signals. It is, therefore, most useful for longer samples, wires, metallic ribbons, semimetals, and some semidegenerate semiconductors. However, it is difﬁcult to maintain T _{1} isothermal throughout the large T at high tem- peratures, requiring additional corrections, and even more difﬁcult to obtain a satisfactory ﬁt for complex S _{a}_{b} T , where the Seebeck coefﬁcient may have many more components than simple carrier diffusion. Furthermore, there are no ob- jective criteria to evaluate the accuracy of the obtained de- rivative.

D. The differential method and its conditions

For the reasons described above, the differential method or small T comprises the majority of high temperature Seebeck coefﬁcient characterizations. In this technique, a small thermal gradient T is applied across a specimen maintained at the mean temperature of interest T _{o} = T _{1} + T _{2} / 2, where T _{1} = T _{o} − T / 2 and T _{2} = T _{o} + T / 2. ^{3}^{9} ^{,} ^{4}^{1} By ex- panding the Seebeck coefﬁcient S _{a}_{b} T in a Taylor series with center T _{o} and integrating, Eq. 4 becomes

V ab T

= S _{a}_{b} T _{o} + _{}

n

=1

1

d ^{2} ^{n} S _{a}_{b} T T

2 n

dT _{0}

2

2 n + 1 !

2 n

6

Therefore, in the differential technique, the Seebeck coefﬁ- cient is given exclusively by the ratio of the electric potential and the temperature difference, provided T / T _{o} 1 and S _{a}_{b} / S _{a}_{b} 1, when V _{a}_{b} T _{o} and the latter term in Eq. 6 can be neglected. In addition, the linearity in S _{a}_{b} T observed for typical metals cannot be expected in semiconductors, where larger thermal gradients would effectively average the struc- ture in S _{a}_{b} T and reduce measurement resolution in T . The following sections describe the different embodiments of the differential method.

= S _{a}_{b} T _{o} + S _{a}_{b} T _{o} .

1. The steady-state condition

Differential methods can be categorized into three con- ditions: steady-state dc , quasi-steady-state qdc , and tran-

121101-6

Martin, Tritt, and Uher

FIG. 6. Early Seebeck coefﬁcient apparatus described by Bidwell Ref. 72 .Reprinted with permission from C. C. Bidwell, Phys. Rev. 19, 447 1922 . Copyright 1922 by the American Physical Society.

sient ac , assuming the observation time scale. Here, obser- vation time is deﬁned as the interval required to acquire one voltage measurement assuming T _{1} , T _{2} , and V _{a}_{b} are mea- sured simultaneously and is therefore instrumentation de- pendent. Perhaps the earliest implementation of the differential steady-state method is that of Wold in 1916, ^{7}^{1} who measured simultaneously the Seebeck coefﬁcient, the Nernst effect, and the Hall effect in tellurium. The temperature differences between 2 to 10 °C were measured using platinum/ platinum–rhodium thermocouples. Bidwell ^{7}^{2} then measured the Seebeck coefﬁcient and resistivity of germanium rods 2.4 cm long between 191 and 675 °C Fig. 6 . Platinum/ platinum–rhodium 10% thermocouples measured the tem- perature differences in grooves at the ends of the sample and were pressed on by nickel spring-loaded graphite rods. Dis- tance adjustments within the heating proﬁle of a vertical chromel wire furnace varied the thermal gradient between 5 to 60 °C. Middleton and Scanlon ^{7}^{3} also describe a similar apparatus with attention to thermocouple contacts. Other de- tailed apparatus descriptions can be found in Refs. 74 and

75.

Under steady-state conditions, differential methods often calculate the Seebeck coefﬁcient from the linear ﬁt of mul- tiple electric potential/temperature difference data points rather than one. This practice avoids the assumption that the curves intersect the ordinate V _{a}_{b} =0, T = 0 , effectively eliminating the offset voltages arising from thermocouple in- homogeneities and nonequilibrium contact interfaces the en- tire series of interfaces between thermocouple wires and the voltmeter . However, the selected temperature differences must be acutely incremental, such that the total range satis- ﬁes the assumption of linearity in S _{a}_{b} T . In addition, the

temperature for S _{a}_{b}

must represent the average for the

collection of steady-state T _{o} . In 1956, Weiss ^{7}^{6} implemented the multiple gradient method to eliminate offset voltages. The sample InAs was clamped between graphite rods, us- ing nickel/chromium thermocouples to measure thermal gra- dients between 5 and 20 °C between room temperature and 1000 °C. The resulting slope of each line was used to calcu- late the Seebeck coefﬁcient, estimated to be accurate within 5 V / C. Two recent articles, one by Burkov ^{7}^{7} Fig. 7 , and the other by Zhou and Uher ^{7}^{8} Fig. 8 , provide detailed descriptions of the apparatus and instrumentation suggested for high temperature implementation of the differential steady-state method.

^{T} avg

J. Appl. Phys. 108 , 121101 2010

FIG. 7. Seebeck coefﬁcient apparatus described by Burkov Ref. 77 .Re- printed with permission from A. T. Burkov, A. Heinrich, P. P. Konstantinov, T. Nakama, and K. Yagasaki, Meas. Sci. Technol. 12, 264 2001 . Copyright 2001 by the Institute of Physics.

2. The quasi-steady-state condition

Although the steady-state condition succeeds in reducing the contribution of offset voltages, the time burden to prop- erly stabilize each increment of T may be impractical and inefﬁcient. To enable rapid measurements of the Seebeck co- efﬁcient the quasi-steady-state condition was introduced in 1961 by Testardi and by Ivory in 1962. ^{7}^{9}^{–} ^{8}^{1} This condition simultaneously measures multiple electric potential/

FIG. 8. Seebeck coefﬁcient apparatus described by Zhou and Uher Ref. 78 .Reprinted with permission from Z. Zhou and C. Uher, Rev. Sci. Instrum. 76, 023901 2005 . Copyright 2005, American Institute of Physics.

121101-7

Martin, Tritt, and Uher

temperature difference data points but employs a continu- ously increasing heat ﬂux rather than multiple static, steady- state T ’s. However, to properly implement these dynamic or gradient sweep techniques requires multiple high imped- ance nanovoltmeters voltage meters with uncertainty 100 nV . Voltage channel switchers, while economical, should not be used to obtain data when continuously ramping the thermal gradient. During the ﬁnite time required to switch and measure the next voltage channel there will likely be sufﬁcient thermal drift to distort the temperature-voltage correspondence, essentially smearing the Seebeck coefﬁ- cient. This error is proportional to the thermal and voltage drift during the time interval between each channel acquisi- tion, usually compounded over three voltage measurements per data point provided a differential thermocouple is not used . One solution to minimize these distortions is to ﬁt the time dependence of the temperature gradient and interpolate

the values corresponding in time to the electric potentials. ^{8}^{2} Furthermore, the physical switching of voltage channels may cause brief transient signals or thermal offsets from the re- lays. Additional time must then be introduced before acquir- ing a voltage reading to obviate these effects. There is also a question of how to select the appropriate time rate of change

in T . Unfortunately, this is rarely stated in the literature

making comparison of data quality a challenge. One modern embodiment of the differential technique

under the quasisteady-state condition is given by Wood. ^{5}^{5} The apparatus measures the Seebeck coefﬁcient up to 1900

K by compressing a sample between two fused-quartz light

pipes. W–Nb thermocouples are inserted into drilled holes near the ends of each light pipe Fig. 9 . A dynamic tempera- ture impulse 5 K is generated by illuminating the light pipes using 600 W tungsten ﬁlaments at each end. These are inversely toggled in such a way as to maintain a constant average sample temperature. This apparatus was further modiﬁed to concurrently measure the Seebeck coefﬁcient and electrical resistivity up to 1300 K. ^{8}^{3} In another example, Ponnambalam and Tritt ^{8}^{4} developed an apparatus to measure resistivity and the Seebeck coefﬁ- cient between 300 and 1000 K in the axial ﬂow arrangement. Samples are compressed between silver blocks spring-loaded external to the furnace for consistency Fig. 10 . Type K thermocouples silver epoxied to these silver blocks measure the gradient sweep 3 to 5 °C and platinum wires measure the electric potential. The base temperature of the furnace is also increasing at a steady rate rather than stabilized at each temperature of interest.

3. The transient „or ac … condition

To eliminate the rigorous thermal stability demands of steady-state conditions, Freeman and Bass ^{8}^{5} and Hellenthal and Ostholt ^{8}^{6} introduced the transient condition in 1970. This condition is characterized by a sinusoidal temperature differ- ence T sin t , where T is between 10 to 500 mK and / 2 is between 0.1 to 60 Hz. Lock-in ampliﬁers then con- tinuously extract the corresponding voltage and temperature amplitudes to obtain the Seebeck coefﬁcient. By modulating the temperature difference, extraneous voltages are elimi-

J. Appl. Phys. 108 , 121101 2010

FIG. 9. Seebeck coefﬁcient apparatus described by Wood Ref. 55 .Re- printed with permission from C. Wood, D. Zoltan, and G. Stapfer, Rev. Sci. Instrum. 56 , 719 1985 . Copyright 1985, American Institute of Physics.

nated more rapidly and using smaller T values as compared to those under steady-state conditions. These features can sharpen the structural resolution in S _{a}_{b} T .

Nonsteady-state conditions are more sensitive to the thermal diffusivity, heat capacity, mass, and geometry of the materials measured, requiring adjustments to the positioning of the thermocouples, the sample thickness, and the fre- quency of sinusoidal variation. The frequency should scale

with the material’s thermal diffusivity or the sample length

thermocouple position adjusted to ensure proper attenua-

tion of the T toward the cold end. ^{8}^{7} However, the values of

T can trace a slight frequency dependence above 1 Hz. In

addition, the magnitude of T is greatest at lower frequen-

cies, where the signal-to-noise ratio yields a more accurate

measure of the Seebeck coefﬁcient. To avoid frequency de- pendence in the Seebeck coefﬁcient, the sample thickness d must be signiﬁcantly smaller than the thermal diffusion length = D / f ^{1} ^{/} ^{2} , where D is the thermal diffusivity of

the material and f is the frequency, to satisfy the relation d / 1. This promotes a homogenous, one-dimensional temperature distribution. If these criteria are not met, the Seebeck coefﬁcient will increase with frequency and differ from results obtained in the dc method, especially for lower

thermal diffusivity materials. ^{8}^{8} The required thickness 0.1

mm to 0.2 mm may be impractical for many brittle TE

materials. In addition, there are no objective criteria in which to choose the appropriate frequency.

121101-8

Martin, Tritt, and Uher

FIG. 10. Color online Seebeck coefﬁcient apparatus described by Ponnam- balam and Tritt Ref. 84 .Reprinted with permission from V. Ponnambalam, S. Lindsey, N. S. Hickman, and T. M. Tritt, Rev. of Sci. Instrum. 77, 073904 2006 . Copyright 2006, American Institute of Physics.

A variety of techniques have been employed to initiate the periodic heat pulse. Freeman and Bass ^{8}^{5} used a tungsten lamp chopped at 21 Hz. Using this technique, they were able to obtain temperature differences as low as 0.03 K, measured with aluminum/gold thermocouples. Goto et al. ^{8}^{8} used a semiconductor laser to periodically heat one end of the sample, and measure the temperature difference with R-type thermocouples attached with silver paste. Kawai ^{8}^{9} and Zhen-Jun ^{9}^{0} also describe a laser modulated ac technique. However, light modulated heating should not be applied to lightly-doped semiconductors, as photo-TE effects are likely to affect the result. Chen ^{9}^{1} describes a slightly modiﬁed ac measurement technique, in which two resistive heaters gen- erate orthogonal sinusoidal temperature waves. This tech- nique has been modiﬁed with square wave pulses in opposite phase for a toggled or bipolar heating. ^{9}^{2} The T produced by each heater must be identical, rather than the power, because of sample differences in heat capacity and thermal conduc- tivity. Maldonado employed a similar technique but with only a single heat current pulsed with a square wave. ^{9}^{3} This technique has been extended to higher temperatures 475 K by Hogan. ^{9}^{4} Fujimoto ^{9}^{5} also describes a Peltier heating based transient technique.

E. The absolute TE scale

The measured value of a TE couple is proportional to the difference between the Seebeck coefﬁcient for the material of interest and that for the reference wires Eq. 1 . There- fore, the Seebeck coefﬁcient of the reference material must be determined in a separate experiment throughout the tem-

J. Appl. Phys. 108 , 121101 2010

perature range of interest. The most common reference ma- terials at high temperature are platinum and copper. There are two techniques to obtain the absolute Seebeck coefﬁ- cient. In some low temperature measurements, it is possible to use a superconducting material as the reference, since S = 0 in the Meissner state. ^{9}^{6} While limited to T 120 K and below, this technique can also provide calibration for Cu or other reference materials at higher temperatures. However, at high temperature an additional technique is required. The Seebeck coefﬁcient can be calculated for pure materials by measuring the Thomson heat _{T} or Thomson coefﬁcient directly and using the relation

S T = _{} 0

^{T} ^{} T

_{T}

dT ,

7

where T is the temperature. This technique established the absolute scales of thermoelectricity for common reference materials in TE measurements. Early measurements of the Thomson effect of copper were conducted by Borelius ^{9}^{7} and Lander. ^{9}^{8} Combined with these results, higher temperature measurements by Nyström ^{9}^{9} extended the absolute TE scale for Cu and Pt up to 1023 K. Cusack and Kendall ^{1}^{0}^{0} further extended this range to 1300 K for Cu and 2000 K for Pt by compiling the avail- able data. However, the most accurate reference data has been obtained more recently by Roberts for Pb between 0 and 550 K, ^{1}^{0}^{1} ^{,} ^{1}^{0}^{2} for W between 273 and 1800 K, ^{1}^{0}^{3} and for Pt between 273 and 1600 K. ^{1}^{0}^{3} Roberts estimates the uncer- tainty at the highest temperatures for Pt at 0.2 V / K. Burkov ^{7}^{7} provides an empirical interpolation function for the Seebeck coefﬁcient of Pt between 70 and 1500 K, by cor- recting low temperature data ^{1}^{0}^{4} according to the TE scale of Roberts ^{1}^{0}^{1} for consistency with the higher temperature data by Roberts. ^{1}^{0}^{3} Bentley ^{1}^{0}^{5} also provides similar absolute TE data for Pt between 5 and 2041 K but using different low temperature data than Burkov. ^{7}^{7}

III. METHODS OF MEASURING RESISTIVITY

Many of the apparatus described above concurrently measure resistivity and the Seebeck coefﬁcient, therefore, a brief discussion on electrical resistivity measurements is ben- eﬁcial to apparatus design and instrumentation. Resistivity measurements in TE materials present unique challenges due to enhanced Seebeck voltage contributions and Peltier effects at the probe contacts. These can be largely overcome through four-probe bipolar or ac resistance bridge techniques with adequate considerations. In the four-probe method potentio- metric , one pair of lead wires sources current through a bulk parallelopiped and a separate pair measures the correspond- ing voltage drop. This effectively eliminates discrete voltage contributions from measurement wires and sample contacts. Combined with dimensional measurements, the resistivity is given by

= ^{V}

I

A

l o

,

8

where V is the measured voltage drop, I the current sourced through the sample, A the cross-sectional area, and l _{o} the

121101-9

Martin, Tritt, and Uher

effective distance between the centers of circular voltage contacts midpoint to midpoint . ^{1}^{0}^{6} Positioning these voltage contacts away from the planar current contacts according to l − l _{o} 2 w where l is the sample length and w is the sample thickness promotes a homogeneous current density where the voltage is measured. These contacts must also be small to avoid perturbations in the sample’s electric ﬁeld. The geo- metric factor A / l _{o} is the most dramatic source of uncer- tainty in the resistivity measurement, therefore, extreme care should be given to parallel sample cutting using a wire saw and polishing in addition to characterizing the geometry. As mentioned above, the measurement of resistivity in TE materials presents unique difﬁculties. Finite thermal gra- dients arising from Joule heating or the Seebeck measure- ment superimpose a TE voltage V = S T onto the resistive voltage drop. Alternating current polarity and averaging the subsequent voltage measurements with no measurement po- larity change eliminates these Seebeck voltage contributions

_{V} IR _{=} V I ^{+} + S T − V I ^{−} + S T

2

,

9

as well as inhomogeneous current ﬂow. In addition, the cur- rent through the sample is often calculated by measuring the voltage drop across a precision resistor. Both of these tech- niques have been common practice since at least Bidwell’s 1922 TE characterization of germanium. ^{6}^{5} A more challenging problem arises from the Peltier ef- fect. The passage of current through the metal/semiconductor measurement interfaces result in the liberation of heat at one current contact and the absorption at the other. Depending upon the direction of the current ﬂow, attempts to measure the resistivity contribute to standing temperature gradients across the sample and consequently increase the apparent Seebeck voltage. Unfortunately, the reversal of current re- verses the direction of both the temperature gradient and its corresponding Seebeck voltage, nullifying any beneﬁt in- curred by averaging the two readings. While the resistive voltage can be measured nearly instantaneously, the imposed Peltier thermal gradient requires a ﬁnite time to propagate. The time for thermal energy to diffuse from a contact to distance x is proportional to x ^{2} / D , where D is the thermal diffusivity. ^{8}^{7} Experimentally, ^{1}^{0}^{7} this time has a maximal du- ration of 1.1 to 1.3 s for common TE materials. Therefore, to negate the Peltier effect requires fast switching of current polarity. These requirements have recently led to the use of ac resistance bridges. However, similar results can be ob- tained using nanovoltmeters and low ripple current sources. Colinear four-probe and van der Pauw techniques for thin disks and ﬁlms are covered in the literature. ^{3}^{8} ^{,}^{4}^{0} ^{,}^{1}^{0}^{8} ^{,} ^{1}^{0}^{9}

IV. INSTRUMENTATION

To aid in the veriﬁcation in reported high ZT materials and to effectively interpret the physics between comparative sample data, researchers must provide an estimation of the measurement uncertainty. Any measurement lacking this es- timate is essentially incomplete. Uncertainty is the limit of measurement accuracy characteristic to the apparatus and the contributing processes. This estimate can be traceable to Na-

J. Appl. Phys. 108 , 121101 2010

tional Institute of Standards and Technology or an interna- tional standard when possible. Systematic uncertainties are generally reduced or eliminated by design while some ran- dom errors may be eliminated by statistical methods. ^{1}^{1}^{0} The physical parameters that comprise the ZT values are gener- ally measured separately, requiring a ﬁt of the temperature dependence of either the electrical or thermal properties to ensure correspondence in temperature. However, the contri- bution of this temperature correspondence error to the total uncertainty in ZT is rarely discussed or estimated. In addi- tion, the Seebeck coefﬁcient should be measured along the same direction as the resistivity and thermal conductivity, and on the same sample if possible. This ﬁnal section will further discuss some of the sources of error previously de- scribed, while also providing a practical guide to designing a modern high temperature Seebeck coefﬁcient apparatus. In addition to choosing an appropriate measurement technique, one must consider the instrumentation required to effectively manage uncertainty. This includes the voltmeters, the ther- mocouples, the probe design, and the cumulative series of interface connections between the sample and the voltmeter. Since at high temperature thermocouples are used to measure the temperature difference, measurement of the See- beck coefﬁcient is essentially a practice of low voltage mea- surements. Therefore, the voltmeter must be sensitive to the range of voltages characteristic to the selected thermo- couples, be able to detect small changes in the signal, and to complete the measurement within the time scale appropriate for the measurement technique. When selecting a voltmeter, one common mistake is to confuse instrument resolution with the uncertainty. The resolution of a voltmeter is deﬁned as the ratio of the smallest count to the largest count on the display. Physically, resolution represents the smallest change in the signal that can produce a change in the display. A 5-1/2 digit display has the values 0-199,999 0-9+ half digit of 0 or 1 , with a resolution of 1/200,000, or 0.0005%. However, this does not include every source of error in the meter, and is, therefore, not representative of the total uncertainty. For example, the stated uncertainty of a standard commercial voltmeter, given in ppm of reading+ ppm of range , is 50+ 6. The ﬁrst term, 50 ppm of reading 50 / 10 ^{6} is negli- gible, but for the lowest range 200 mV range , the second term is 6 / 10 ^{6} 0.2 V or 1.2 V. Therefore, measuring a small T of 1 K with an R-type thermocouple near room temperature would have an uncertainty approaching 20%. Often the resolution of this voltmeter, 0.01 V, is substi- tuted for the uncertainty, giving a vastly different represen- tation of the instrument’s error. Other sources of voltmeter uncertainty include: temperature dependence of the meter, internal thermal offsets, channel relay self-heating, time drift, noise/rf interference, and improper zeroing of the channels. Details of how to regulate or eliminate these parameters can be found in Ref. 111 . The thermocouple is probably the most overlooked com- ponent of any measurement apparatus. It is generally treated as a “black box” device, where the thermocouple is expected to behave according to “calibration,” where the temperature/ voltage correspondence is identical to an established stan- dard. However, there are a variety of other sources of error

121101-10

Martin, Tritt, and Uher

that are erroneously eliminated through an assumption of their absence. The actual process of negating these errors is called the “standard practice of thermocouples.” Bentley ^{1}^{0}^{5} provides a through discussion of thermocouple theory, error reduction, and methods to estimate the total uncertainty. Other references also discuss the theory and use of thermocouples. ^{1}^{1}^{2} ^{–} ^{1}^{1}^{5} By deﬁnition, a thermocouple is a relative measure of temperature. The thermocouple transfers measurement of temperature from a difﬁcult location to a more practical one, the reference cold junction. This junction is an isothermal region serving as the interface between the thermocouple wires and the copper wires to the voltmeters. Thermocouple extension wires are invariably inhomogeneous and should rarely be used, since they are made from lower purity alloys to mimic the TE properties of more expensive and pure ther- mocouple wire. ^{1}^{0}^{5} All of the connections between the refer- ence junction and the voltmeter should be Cu-to-Cu to re- duce the emergence of TE voltages at the interfaces, using oxygen free copper or the more pure Ohno continuous cast wire. Traditionally, the reference junction is an ice-water bath at 0 °C, such that the temperature is stable to within 0.01 °C. Where this is impractical, the temperature of the reference junction can be measured directly using a precision platinum resistance thermometer. Another option is to use a cold junction compensation circuit that provides a calibrated voltage correction to the thermocouple voltage. However, this is less reliable with lower accuracy and should be avoided where high accuracy is desired. The contributing magnitude of this error depends on the variation in S T around 0 °C for the speciﬁc thermocouple. Real thermocouples are not chemically or physically ho- mogenous, especially since one wire is usually an alloy. Physical inhomogeneity includes variation in crystallite size and macroscopic strains cold-working . This error repre- sents the variation in signal output dependent on the longi- tudinal temperature proﬁle. That is, if the immersion depth of the thermocouple is varied while the tip and the reference junction are constant, the change in voltage would reﬂect this inhomogeneity 0.02% for Pt-based thermocouples . Physical strain can often be removed by a proper annealing procedure. ^{1}^{1}^{5} This depends on the thermocouple type. For high temperature measurements up to 1300 K, the recom- mended types are platinum-base R, S or single element thermocouples, which are not commercially available. Single element thermocouples such as Pt–Au, have minimal effects from lattice ordering or selective volatilization and oxida- tion. Due to the higher purity materials and homogeneity, accuracy can approach 0.01 °C with a drift 0.1 °C after 1000 h. ^{1}^{0}^{5} The wire diameter must minimize heat sinking without signiﬁcantly increasing contamination susceptibility or drift. Common diameters in TE measurements are 0.125 to 0.25 mm. In addition, the thermocouple tips must be shielded from thermal radiation and the entire length embed- ded in a ceramic sheathing to avoid strain and contamination. Another important consideration is the probe material. The probe provides the sample interface for electric current resistivity measurements and for the hot and cold sinks. Selection criteria for probe materials include: resistivity

J. Appl. Phys. 108 , 121101 2010

low , thermal conductivity high , coefﬁcient of thermal ex- pansion low and matched to the ceramic supports , heat ca- pacity low for ac or ballistic cycles; high for enhanced ther- mal stability , maximum operating temperature, high temperature strength, thermal shock resistance for the probe supports, diffusion/interface/chemical reaction properties with candidate TE materials, oxidation resistance, machin- ability, and cost. Common probe materials are W excellent physical properties but difﬁcult to machine , Mo can be used in an oxidizing atmosphere , Pt good physical proper- ties but very expensive , Ni a ferromagnet with a curie tem- perature near 631 K, near which the resistivity and yield strength can change by a factor of 5 , ^{1}^{1}^{6} Ag prone to diffusion/interaction with TE materials , Cu lower maxi- mum operating temperature , Al lower operating tempera- ture , and graphite resistive . One ﬁnal consideration is contacts. These include both electrical and thermal contacts to the sample and also to the instrumentation. Seebeck coefﬁcient measurements are fur- ther complicated at high temperature by the possibility of chemical reactions within the sample and/or with the sample probes or thermocouples. These reactions may give rise to systematic voltage offsets. It is, therefore, prudent to repeat the measurement cycle, especially when conﬁrming mea- surements on high ZT materials. There are many references that describe the challenges of forming Ohmic contacts be- tween metal and semiconductor interfaces. ^{1}^{1}^{7} ^{,} ^{1}^{1}^{8} One reliable and consistent method of forming good electrical and ther- mal contacts at high temperature for a variety of materials has been pressure contacts. ^{7}^{7}^{–} ^{8}^{4} Often low pressure, pure he- lium gas is used as an intermediary to enhance thermal con- tact between the thermocouple and the sample. However, there are reports suggesting low pressure gases such as he- lium and nitrogen may affect the thermal contact and thus the measured Seebeck coefﬁcient value. ^{1}^{1}^{9} ^{,} ^{1}^{2}^{0} Due to the ever evolving nature of TE materials research, the method and nature of forming good contacts must continually be evalu- ated and tested. To this end, the apparatus and methodologies implemented must be compared with materials previously measured and/or to well characterized standards. ^{1}^{2}^{1}^{,} ^{1}^{2}^{2}

V. SUMMARY

The continued development of new TE materials for high temperature power conversion applications requires re- liable and accurate characterization of the electrical and ther- mal transport properties. The Seebeck coefﬁcient is an essen- tial physical property in evaluating the potential performance of these TE materials, as it is sensitive to the electronic struc- ture. However, the deceptive simplicity of measuring the Seebeck coefﬁcient has led to the implementation of non- ideal practices that have further complicated the conﬁrmation of reported high ZT materials. Therefore, researchers must practice caution in order to accurately measure the Seebeck coefﬁcient. While it may be experimentally uncertain which arrangement provides the most accurate determination of temperature and voltage difference at the same location, or which method and condition provides the most accurate de- termination of the Seebeck coefﬁcient at high temperature,

121101-11

Martin, Tritt, and Uher

the primary requirements remain the same: 1 the voltage and temperature must be measured at the same location and at the same time, 2 the probes must be in very good thermal and electrical contact with the specimen, and 3 low volt- ages must be acquired with minimal extraneous contribu- tions. In addition, the implementation must avoid distortion of the temperature-voltage correspondence and must use ap- propriate temperature differences such that the range satisﬁes the assumption of linearity in S _{a}_{b} T . It would be useful to evaluate these arrangements and methodologies over a broad temperature range and for a variety of materials and contact interfaces. Thus, a versatile and ﬂexible apparatus would not only enable comparative measurements on materials with different contact resistances, thermal conductivities, and heat capacities, but also expand the practical range of sample ge- ometries.

ACKNOWLEDGMENTS

J. M. would like to thank Keith Martin and Stacy Bruss in the NIST Library for their assistance in obtaining early reference material. The work of C. U. is supported as part of the Center for Solar and Thermal Energy Conversion, an Energy Frontier Research Center funded by the U. S. Depart- ment of Energy, Ofﬁce of Basic Energy Sciences under Award No. DE-SC00000957. The work at Clemson Univer- sity is supported by DOE/EPSCoR Implementation Grant No. DE-FG02–04ER-46139 and the SC EPSCoR cost shar- ing program.

^{1} E. Velmre, Proc. Estonian Acad. Sci. Eng. 13, 276 2007 . ^{2} T. J. Seebeck, Magnetische Polarisation der Metalle und Erze durch Temperatur-Differenz, Abhandlungen der physikalischen Klasse der Königlichen Akademie der Wissenschafften zu Berlin, Aus den Jahren 1822 und 1823, pp. 265–373 1825 . Extracts from four lectures delivered at the Academy of Sciences in Berlin on August 16, 1821, October 18, 25 1821, and February 11 1822. ^{3} K. L. Caneva 2001 notes the term “thermomagnetism” was not in fact present in any of Seebeck’s publications but was mentioned in subsequent notes. ^{4} J. C. von Yelin, Ann. Phys. 73, 415 1823 . ^{5} K. L. Caneva, The Form and Function of Scientiﬁc Discoveries, Dibner Library Lecture, Vol. viii-32 Smithsonian Institution Libraries, Washing-

ton, DC, 2001 , and references therein. ^{6} R. Bowers, R. W. Ure, Jr., J. E. Bauerle, and A. J. Cornish, J. Appl. Phys. 30, 930 1959 . ^{7} H. J. Goldsmid, Br. J. Appl. Phys. 11, 209 1960 . ^{8} J. C. Peltier, Ann. Chim. Phys. 56, 371 1834 . ^{9} A. J. Woodall, Proc. Phys. Soc. London 47, 615 1935 . ^{1}^{0} A. E. Caswell, Phys. Rev. 33 , 379 1911 . ^{1}^{1} A. F. Ioffe, Semiconductor Thermoelements and Thermoelectric Cooling Infosearch Ltd., London, 1957 . ^{1}^{2} H. J. Goldsmid, Electronic Refrigeration Pion Limited, London, 1986 . ^{1}^{3} C. Wood, Rep. Prog. Phys. 51, 459 1988 . ^{1}^{4} G. D. Mahan, Solid State Phys. 51 , 81 1998 . ^{1}^{5} G. S. Nolas, J. Sharp, and H. J. Goldsmid, Thermoelectrics: Basic Prin- ciples and New Materials Developments Springer, New York, 2001 . ^{1}^{6} Thermoelectrics Handbook, edited by D. M. Rowe CRC, Boca Raton, FL,

1995

.

^{1}^{7} Chemistry, Physics, and Materials Science of Thermoelectric Materials:

Beyond Bismuth Telluride, edited by M. G. Kanatzidis, S. D. Mahanti, and T. P. Hogan Plenum, New York, 2003 . ^{1}^{8} G. A. Slack, in CRC Handbook of Thermoelectrics, edited by B. D. M. Rowe CRC, Boca Raton, 1995 , p. 407. ^{1}^{9} G. S. Nolas, D. T. Morelli, and T. M. Tritt, Annu. Rev. Mater. Res. 29, 89 1999 . ^{2}^{0} C. Uher, in Semiconductors and Semimetals, edited by T. M. Tritt Aca-

J. Appl. Phys. 108 , 121101 2010

demic, New York, 2000 , Vol. 69, p. 139.

G. S. Nolas, M. Kaeser, R. T. Littleton, and T. M. Tritt, Appl. Phys. Lett.

77, 1855 2000 .

G.

G.

G.

Semimetals, edited by T. M. Tritt Academic, New York, 2000 , Vol. 69, p.

255, and references therein.

G. S. Nolas, in Thermoelectrics Handbook: Macro to Nano-Structured

Materials, edited by D. M. Rowe CRC, Boca Raton, FL, 2005 , p. 33.

V. L. Kuznetsov, L. A. Kuznetsova, A. E. Kaliazin, and D. M. Rowe, J.

Appl. Phys. 87, 7871 2000 .

Ren, J. P. Fleurial, and P. Gogna, Adv. Mater. Weinheim, Ger. 19 , 1043

2007 .

L.

R.

London 413, 597 2001 .

T. C. Harman, P. J. Taylor, M. P. Walsh, and B. E. LaForge, Science 297 ,

2229 2002 .

T. Koga, S. B. Cronin, M. S. Dresselhaus, J. L. Liu, and K. L. Wang, Appl. Phys. Lett. 77, 1490 2000 .

J. P. Heremans, C. M. Thrush, and D. T. Morelli, J. Appl. Phys. 98 , 063703 2005 .

K. F. Hsu, S. Loo, F. Guo, W. Chen, J. S. Dyck, C. Uher, T. Hogan, E. K.

Polychroniadis, and M. G. Kanatzidis, Science 303, 818 2004 .

M.

Z.

Direct Thermal-to-Electric Energy Conversion , edited by J. Yang, T. P. Hogan, R. Funahashi, and G. S. Nolas, Materials Research Society, Pitts- burgh, 2006 , Vol. 886, p. 3, p. 3; K. Kishimoto and T. Koyanagi, J. Appl. Phys. 92, 2544 2002 .

S. Dresselhaus, G. Chen, M. Y. Tang, R. G. Yang, H. Lee, D. Z. Wang,

F. Ren, J. P. Fleurial, and P. Gogna, Materials and Technologies for

Venkatasubramanian, E. Siivola, T. Colpitts, and B. O’Quinn, Nature

A. Lamberton, Jr., S. Bhattacharya, R. T. Littleton IV, T. M. Tritt, and

^{2}^{1}

^{2}^{2}

^{2}^{3}

^{2}^{4}

^{2}^{5}

^{2}^{7}

^{2}^{8}

^{3}^{0}

^{3}^{2}

^{3}^{3}

S. Nolas, Appl. Phys. Lett. 80 , 598 2002 .

S. Nolas, G. A. Slack, and S. B. Schujman, in Semiconductors and

D. Hicks and M. S. Dresselhaus, Phys. Rev. B 47, 16631 1993 .

^{3}^{5}

^{3}^{6}

^{3}^{7}

^{3}^{8}

^{3}^{9}

^{4}^{0}

^{4}^{1}

^{4}^{2}

^{4}^{3}

^{4}^{4}

^{4}^{5}

^{4}^{6}

^{4}^{7}

^{4}^{8}

^{5}^{0}

J. P. Heremans, C. M. Thrush, and D. T. Morelli, Phys. Rev. B 70 , 115334 2004 .

J. Martin, L. Wang, L. Chen, and G. S. Nolas, Phys. Rev. B 79, 115311 2009 .

C.

T.

M. Tritt, in Thermoelectric Materials-New Directions and Approaches,

MRS Symposia Proceedings No. 478, edited by T. M. Tritt, M. Kanazidis,

G. Mahan, and H. B. Lyons, Jr. Materials Research Society, Pittsburgh,

1997 , p. 25.

T. M. Tritt and V. Browning, Semicond. Semimetals 69, 25 2001 .

A. T. Burkov, Thermoelectrics Handbook: Macro to Nano CRC, Boca

Raton, FL, 2005 , p. 22.

T. M. Tritt, in Thermoelectrics Handbook: Macro to Nano CRC, Boca

Raton, FL, 2005 , p. 23.

T. Heikes and R. W. Ure, Thermoelectricity: Science and Engineering

Interscience, New York, 1961 , p. 311.

Yu. I. Ravich, B. A. Eﬁmova, and I. A. Smirnov, Semiconducting Lead Chalcogenides Plenum, New York, 1970 , p. 149.

C.

C.

P. M. Chaikin, in Organic Superconductivity , edited by V. Z. Kresin and

W.

T.

T.

1959 .

A. E. Bowley, L. E. J. Cowles, G. J. Williams, and H. J. Goldsmid, J. Sci.

Instrum. 38, 433 1961 .

A.

E.

don, 1960 .

H. Putley, The Hall Effect and Related Phenomena Butterworth, Lon-

Uher, Nav. Res. Rev. XLVIII , 44 1996 .

A. Domenicali, Rev. Mod. Phys. 26, 237 1954 .

Herring, Phys. Rev. 96, 1163 1954 .

A. Little Plenum, New York, 1990 , p. 101.

C. Harman, J. Appl. Phys. 29, 1373 1958 .

C. Harman, J. H. Cahn, and M. J. Logan, J. Appl. Phys. 30 , 1351

W. Penn, J. Sci. Instrum. 41 , 626 1964 .

^{5}^{1}

^{5}^{2}

^{5}^{3}

^{5}^{4}

^{5}^{5}

^{5}^{6}

^{5}^{7}

^{5}^{8}

^{5}^{9}

^{6}^{0}

C. Hurd, The Hall Effect in Metals and Alloys, Plenum, New York, 1972 .

See for example, de Nobel, Physica 17, 551 1951 .

A. L. Pope, R. T. Littleton IV, and T. M. Tritt, Rev. Sci. Instrum. 72, 3129

2001 .

J. Kopp and G. A. Slack, Cryogenics 11, 22 1971 .

C.

G.

C.

N.

J. Tauc, Photo and Thermoelectric Effects in Semiconductors Pergamon, New York, 1962 , p. 179.

Z. Trousil, Czech. J. Phys. 6, 96 1956 .

Wood, D. Zoltan, and G. Stapfer, Rev. Sci. Instrum. 56, 719 1985 .

Magnus, Pogg. Ann. 83, 469 1851 .

Benekics, Ergeb. Exakten Naturwiss. 8, 26 1929 .

Fuschillo, Proc. R. Soc. London, Ser. B 65, 896 1952 .

121101-12

Martin, Tritt, and Uher

^{6}^{1} R. M. White, Sandia Laboratories Report No. SC-RR-68-636, October

1968.

^{6}^{2} C. Wood, A. Chmielewski, and D. Zoltan, Rev. Sci. Instrum. 59, 951

1988 . ^{6}^{3} W. H. Kettler, T. Wernhardt, and M. Rosenberg, Rev. Sci. Instrum. 57,

3053 1986 .

^{6}^{5} C. C. Bidwell, Phys. Rev. 3 , 204 1914 . ^{6}^{6} W. Thomson, Proc. R. Soc. Edinburgh 1851, 91.

^{6}^{7} P. G. Tait, Nature London 8 , 86 1873 , from the Rede Lecture delivered in the Senate House, Cambridge, May 23, 1873. ^{6}^{8} J. Dewar and J. A. Fleming, Philos. Mag. 40, 95 1895 . ^{6}^{9} F. G. Wick, Phys. Rev. 25, 382 1907 . ^{7}^{0} R. J. Gripshover, J. B. VanZytveld, and J. Bass, Phys. Rev. 163, 598 1967 . ^{7}^{1} P. I. Wold, Phys. Rev. 7 , 169 1916 . ^{7}^{2} C. C. Bidwell, Phys. Rev. 19, 447 1922 . ^{7}^{3} A. E. Middleton and W. W. Scanlon, Phys. Rev. 92 , 219 1953 . ^{7}^{4} L. S. Phillips, J. Sci. Instrum. 42, 209 1965 . ^{7}^{5} T. Dasgupta and A. M. Umarji, Rev. Sci. Instrum. 76 , 094901 2005 . ^{7}^{6} V. H. Weiss, Z. Naturforsch. B 11a , 131 1956 . ^{7}^{7} A. T. Burkov, A. Heinrich, P. P. Konstantinov, T. Nakama, and K. Ya- gasaki, Meas. Sci. Technol. 12 , 264 2001 . ^{7}^{8} Z. Zhou and C. Uher, Rev. Sci. Instrum. 76, 023901 2005 . ^{7}^{9} L. R. Testardi and G. K. McConnell, Rev. Sci. Instrum. 32, 1067 1961 . ^{8}^{0} J. E. Ivory, Rev. Sci. Instrum. 33 , 992 1962 . ^{8}^{1} C. N. Berglund and R. C. Beairsto, Rev. Sci. Instrum. 38, 66 1967 . ^{8}^{2} G. R. Caskey, D. J. Sellmyer, and L. G. Rubin, Rev. Sci. Instrum. 40, 1280 1969 . ^{8}^{3} L. D. Zoltan, C. Wood, and J. P. Fleurial, Proceedings of the Ninth Inter- national Conference on Thermoelectrics, 19–20 March 1990. ^{8}^{4} V. Ponnambalam, S. Lindsey, N. S. Hickman, and T. M. Tritt, Rev. Sci. Instrum. 77, 073904 2006 . ^{8}^{5} R. H. Freeman and J. Bass, Rev. Sci. Instrum. 41, 1171 1970 . ^{8}^{6} W. Hellenthal and H. Ostholt, Z. Angew. Phys. 28, 313 1970 . ^{8}^{7} H. R. Meddins and J. E. Parrott, Br. J. Appl. Phys. 2, 691 1969 . ^{8}^{8} T. Goto, J. H. Li, T. Hirai, Y. Maeda, R. Kato, and A. Maesono, Int. J. Thermophys. 18, 569 1997 . ^{8}^{9} M. Kawai, Appl. Phys. Lett. 33, 9 1978 . ^{9}^{0} F. Zhen-Jun and Z. Dian-Lin, Chin. Phys. 14, 1533 2005 . ^{9}^{1} F. Chen, J. C. Cooley, W. L. Hults, and J. L. Smith, Rev. Sci. Instrum. 72,

4201 2001 .

^{9}^{2} R. Resel, E. Gratz, A. T. Burkov, T. Nakama, M. Higa, and K. Yagasaki, Rev. Sci. Instrum. 67 , 1970 1996 . ^{9}^{3} O. Maldonado, Cryogenics 32, 908 1992 . ^{9}^{4} T. Hogan, N. Ghelani, S. Loo, S. Sportouch, S.-J. Kim, D.-Y. Chung, and M. G. Kanatzidis, Proceedings of the International Conference on Ther- moelectrics, 1999, Vol. 39, pp. 756–760. ^{9}^{5} S. Fujimoto, H. Kaibe, S. Sano, and T. Kajitani, Jpn. J. Appl. Phys., Part 1

45, 8805 2006 .

J. Appl. Phys. 108 , 121101 2010

^{9}^{6} C. Uher, J. Appl. Phys. 62, 4636 1987 .

^{9}^{7} G. Borelius, W. H. Keesom, and C. H. Johansson, Commun. Phys. Lab. Univ. Leiden 196a 1928 .

^{9}^{8} J. J. Lander, Phys. Rev. 74, 479 1948 .

^{9}^{9} J. Nyström, Ark. Mat., Astron. Fys. 34a, N27 1948 .

^{1}^{0}^{0} N. Cusack and P. Kendall, Proc. Phys. Soc. 72 , 898 1958 .

^{1}^{0}^{1} R. B. Roberts, Philos. Mag. 36, 91 1977 .

^{1}^{0}^{2} R. B. Roberts, Philos. Mag. 43, 1125 1981 .

^{1}^{0}^{3} R. B. Roberts, F. Righini, and R. C. Compton, Philos. Mag. 52, 1147 1985 .

^{1}^{0}^{4} J. P. Moore and R. S. Graves, J. Appl. Phys. 44 , 1174 1973 .

^{1}^{0}^{5} R. Bentley, Theory and Practice of Thermoelectric Thermometry, Vol. 3 Springer-Verlag Singapore Pte. Ltd., Singapore, 1998 .

^{1}^{0}^{6} T. P. Hogan and Y. Sim Loo, and Fu Guo, in Low Temperature Electron- ics and Low Temperature Coﬁred Ceramic Based Electronic Devices, edited by C. L. Claeys, W. Wong-Ng, and K. M. Nair The Electrochemi- cal Society, Pennington, 2004 , p. 231.

^{1}^{0}^{7} I. A. Nishida, in CRC Hand Book of Thermoelectrics, edited by D. M. Rowe CRC, Boca Raton, FL, 1995 , Vol. 157.

^{1}^{0}^{8} F. M. Smits, Bell Syst. Tech. J. 37, 711 1958 .

^{1}^{0}^{9} L. J. van der Paw, Philips Res. Rep. 13, 1 1958 ; 16 , 187 1961 .

^{1}^{1}^{0} J. R. Taylor, An Introduction to Error Analysis: The Study of Uncertain- ties in Physical Measurements, 2nd ed. University Science Books, Sau- salito, CA, 1997 .

^{1}^{1}^{1} Low Level Measurements Handbook: Precision dc Current, Voltage, and Resistance Measurements, 6th ed. Keithley Instruments, Inc., Cleveland, OH, 2004 .

^{1}^{1}^{2} R. L. Powell, W. J. Hall, and J. G. Hust, Temperature: Its Measurement and Control in Science and Industry Instrument Society of America, Pittsburgh, PA, 1972 , Vol. 4, p. 1423.

^{1}^{1}^{3} D. Ripple, NIST Spec. Pub. 260-134.

^{1}^{1}^{4} G. K. White and P. J. Meeson, Experimental Techniques in Low- Temperature Physics Oxford University Press, New York, 1968 , p. 146.

^{1}^{1}^{6} See for example, E. Zaretsky, Appl. Phys. Lett. 92, 011913 2008 .

^{1}^{1}^{7} W. R. Runyan and T. J. Shaffer, Semiconductor Measurements and In- strumentation, 2nd ed. McGraw-Hill, New York, 1997 .

^{1}^{1}^{8} B. G. Streetman, Solid State Electronic Devices Prentice-Hall, Engle- wood Cliffs, NJ, 1995 .

^{1}^{1}^{9} C. Panagopoulos, T. Fukami, and T. Aomine, Jpn. J. Appl. Phys., Part 1 32, 4684 1993 .

^{1}^{2}^{0} A. B. Villaﬂor and L. H. de Luna, Jpn. J. Appl. Phys., Part 1 33, 4051 1994 .

^{1}^{2}^{1} E. Muller, C. Stiewe, D. M. Rowe, and S. G. K. Williams, in Thermo- electrics Handbook: Macro to Nano CRC, Boca Raton, FL, 2005 , p. 26.

^{1}^{2}^{2} Z. Q. J. Lu, N. D. Lowhorn, T. N. Tran, J. Martin, and G. S. Nolas, J. Res. Natl. Inst. Stand. Technol. 114, 37 2009 .

## Viel mehr als nur Dokumente.

Entdecken, was Scribd alles zu bieten hat, inklusive Bücher und Hörbücher von großen Verlagen.

Jederzeit kündbar.