Conducting glasses as new potential thermoelectric materials : the Cu-GeTe case

Recent approaches to improve performance of bulk thermoelectric (TE) materials show that they should have complex structures, include inclusions and impurities, possess mass fluctuations, disorder and be based on heavy elements. Glasses can possess these properties. In order to identify glasses with interesting TE potential, attention should be focused on small gap semiconducting or semimetallic glasses. Chalcogenide glasses with Cux+yGe20−xTe80−y (0 ≤ x ≤ 20; 0 ≤ y ≤ 10) compositions were prepared by melt spinning. Their powder X-ray diffraction analyses point to a short-range order analogous to Ge20Te80, with Cu atoms most likely replacing Ge atoms in the GeTe4 structural unit. It also indicates, together with the differential scanning calorimetry results, a reduction in the glass stability with the increase in Cu concentration. The enhancement of Cu content dramatically reduces (five orders of magnitude) the electrical resistivity, while keeping the Seebeck coefficients at large values (∼400 μV K−1). As a consequence, a huge increase in the power factor is observed, up to a maximum value of 60 μW K−2 m−1 for the Cu27.5Ge2.5Te70 glass at T = 300 K. Ge20Te80 has extremely low lattice thermal conductivity values (∼0.1 W K−1 m−1 at 300 K), which points to relatively high values for the figure of merit ZT for this family of glasses, and indicates Cux+yGe20−xTe80−y based glasses as good candidates for obtaining high performance thermoelectric materials.


Introduction
The research of new environment friendly energy sources and optimization of energy use have become top priority to most of modern societies.Thermoelectric materials, which are able to directly convert thermal into electrical energy (using the Seebeck effect) and, reversibly, electrical into thermal energy (using the Peltier effect), have seen renewed interest due to their potential to provide a sustainable energy solution and optimization.Moreover, the absence of greenhouse effect substances, as halogenated cooling agents, and the lack of moving parts make thermoelectric devices highly attractive and reliable (as an example, NASA Voyager systems employ thermoelectric generators working uninterruptedly for more than 30 years).
The actual commercial thermoelectric devices use materials based on PbTe, Si-Ge and, in particular, Bi 2 Te 3 for near room temperature applications.However, they present low efficiencies (<10%), being crucial to discover new compounds or materials that could lead to devices with improved efficiencies.
The efficiency of a thermoelectric device is mainly controlled by: (i) the adimensional figure of merit, ZT, which is given by ZT =  2 T/ (where T represents the absolute temperature,  is the Seebeck coefficient, and  and  represent the electrical and thermal conductivities, respectively) and only depends on the material; (ii) the temperature difference, T, between the hot and cold junctions.The maximum value of the temperature difference depends on the characteristics of the constituent materials, in particular their thermal and chemical stability, and on the available heat and cooler sources.ZT maximization can be done via both the maximization of the numerator,  2  (power factor), and the minimization of the denominator, .  and  depend on the charge carriers concentration, a maximum on  2  being observed for concentrations of ~10 18 -10 21 carriers/cm 3 , which corresponds to low gap semiconductors or semimetals. can be considered as a sum of two different contributions,  =  e + L , where  e represents the electronic contribution and  L is the contribution from the lattice vibrations (phonons).As  e is related with  via the Wiedemann-Franz law,  e = LT (L is the Lorentz factor), ZT maximization implies the minimization of  L .
At the beginning of the 90´s Slack presented a new concept, "Phonon Glass Electron Crystal" (PGEC) [1], which proposes the research of compounds that conduct electricity as a crystalline material and heat as a glass.This concept has lead to a better understanding of the mechanisms that affect the phonons propagation without changing significantly the electrical charge propagation and to the development of general rules to increase the thermoelectric systems performance, the most important ones being [2]: (i) use of compounds with complex crystal structures; (ii) presence of heavy atoms weakly bounded to the structures; (iii) existence of inclusions and/or impurities; (iv) formation of solid solutions; (v) existence of a large number of grain boundaries.It must be noted that these rules are not exclusive and several can exist simultaneously in the same material.As a consequence, the PGEC concept, together with the use of modern synthesis techniques, has led to the discovery of new materials with improved thermoelectric characteristics, as the skutterudites, clathrates, half-Heusler phases or low dimensional systems [3].
However, the efficiency increase of the new bulk materials is still limited to ~50% [3] and the cost of low dimensional systems is very high.Therefore, it is highly desirable to identify and study new thermoelectric systems.
The careful analysis of the main general rules to increase the thermoelectric systems performance points to conducting glasses as one of the best potential materials.Indeed, this type of materials follows almost all the main rules: they have extremely complex structures, with a certain degree of order only at small distances, can have heavy atoms weakly bounded to the structure and present mass fluctuations, easily allowing high concentrations of inclusions and impurities.In this work is described a first study on the possibility of using conducting glasses for thermoelectric applications (up to the authors best knowledge, before the present work only a theoretical paper was dedicated to this subject [4]).
Preliminary experimental results already resulted on a national patent [5].
As was stressed before, to identify glasses (or any other type of materials) with improved thermoelectric performances it is necessary to center the studies on low gap semiconductor and semimetal glasses.Several glasses with semiconducting properties and containing heavy elements have already been reported in the literature, most of them based on pnictides and chalcogenides [6,7].This pioneer work was centered on Ge 20 Te 80 -based glasses: the original Ge 20 Te 80 glass is mainly formed by heavy atoms; is based only on two elements; has been described as easy to prepare [8] and it is reported as having a high Seebeck coefficient [9]; it has been also described as having small electrical conductivity values [9], but doping it with silver or cooper dramatically increases its electrical conductivity [10,11].For all these reasons we decided to study the Cu x+y Ge 20-x Te 80-y glasses as starting test materials and here is presented their synthesis, X-ray diffraction characterization, differential thermal analysis, thermal conductivity measurements and electrical transport (electrical resistivity and Seebeck coefficient) studies.

Experimental Section
Samples with (x+y)Cu:(20-x)Ge:(80-y)Te general nominal composition (Figure 1) were prepared from the stoechiometric amounts of the elements (Cu, Goodfellow, >99.99%;Ge, Johnson Matthey, 99.999%; Te, Alfa Aesar, >99.999%).The elemental mixtures were sealed into quartz ampoules under vacuum (10 -5 mbar) and heated at 850ºC for five periods of 10 minutes each.In between these periods, the ampoules were removed from the furnace, shaked and turned upside down (while the products are still melted) in order to achieve a good homogeneity.After the last period, the 20Ge:80Te composition sample was quenched into ice water, the other samples being removed from the furnace with no extra procedure.Pieces of all samples were used as starting materials to prepare the ribbons.The pieces were put into a quartz tube with a 0.5 mm diameter nozzle, which was inserted in the stainless steel chamber containing the melt spinning system.The pieces were then melted under the protection of an argon atmosphere and injected with a pressure of 1 bar of Ar onto a copper roller rotating with a linear speed of 1.6 m/s.The quality of the samples was checked via X-ray diffraction and differential thermal analysis measurements, together with optical microscope observations.The X-ray powder diffraction measurements were made on powders representative of the totality of the sample, at room temperature and under a dehumidified air atmosphere.The scans were performed using a Philips X'Pert diffractometer (Bragg-Brentano assembly) with a monochromatized Cu Kα radiation, a 2-range of 10º-70º, a step width of 0.03º and 30 s of counting time per 2 step.The glass transition (T g ), crystallization (T c ) and melting (T m ) temperatures of the materials were measured by differential scanning calorimetry (DSC), by using a DuPont 910 system, under an argon flux atmosphere, from 25ºC to 360ºC and with a constant heating rate of 10ºC/min.The measurements were made using 30-40 mg of material, which was put inside an aluminum sample container; an empty aluminum container was used as reference.
Electrical resistivity and Seebeck coefficient measurements were performed in the ~60-300 K temperature range on purely glassy pieces, with a needle-like shape of ~2mm x 0.25 mm 2 , removed from each sample.A previously described cell [12], attached to the cold stage of a closed cycle cryostat, was used.The resistivity was measured by a four-probe method using the DC technique with a Keithley 220 current source and a Keithley 619 electrometer.The Seebeck coefficient was measured by a slow ac technique (ca. 10 -2 Hz), the voltage across the sample and gold leads being measured with a Keythley 181 nanovoltmeter.The oscillating thermal gradient was kept below 1 K and was measured by a Au-0.005 at % Fe versus chromel thermocouple.The absolute Seebeck coefficient of the sample was obtained after correction for the absolute Seebeck coefficient of the gold leads (99.99% pure gold) by using the data of Huebener [13].
The thermal conductivity of the Cu 20 Te 80 glass was measured by a standard four-contact slow ac method, relative to a constantan wire.The method used was very similar to the one previously described [14], only this time the sample and the constantan wire were thermally connected to the copper cell with the help of screws and copper plates, and glued together with silver paste.The two constantan-chromel 12 m diameter thermocouples were glued with GE varnish, one to the sample and the other to the constantan, and the voltages measured with two Keythley 181 nanovoltmeters.The temperature gradients used were as small as possible (typically <5 K) and the results were calculated using data taken from ref. [15].

Results and Discussion
The Cu x+y Ge 20-x Te 80-y glasses show X-ray diffraction patterns comparable to Ge 20 Te 80 (Figure 2), pointing to analogous short-range order in this type of materials.The structure of Ge x Te 100-x (10 ≤ x ≤ 25) glasses has been extensively investigated due to their possible technological applications on optical data storage devices.The most recent X-ray diffraction studies, using high-energy synchrotron radiation, agree well with previous neutron diffraction works [16] and describe it as composed by GeTe 4 tetrahedral structural units, which are bridged by Te-Te bonds [17,18].Cu x+y Ge 20-x Te 80-y glasses are much less studied, but X-ray diffraction investigations made on the Cu 0.08 Ge 0.18 Te 0.74 material also point out the existence of GeTe 4 tetrahedral structural units, together with other tetrahedral units centered on the copper atoms, such as CuGeTe 3 and CuTe 4 [19] (in fact, CuTe 4 tetrahedral units also exists in the CuTe compound [20] that crystallizes in its own structure -type, which supports too the possibility of having this type of unit in the Cu x+y Ge 20-x Te 80-y glasses).Therefore, in Cu x+y Ge 20-x Te 80-y glasses, the increase of the copper content, simultaneously with a large germanium and a smaller tellurium decrease (87.5% and 12.5%, respectively, in the case of higher copper concentration glass, Cu 27.5 Ge 2.5 Te 70 ), can be mainly seen as copper replacing germanium atoms in the GeTe 4 structural unit, together with the formation of some new CuTe 4 structural units.
Optical microscope observations of the samples show that they generally have a glassy aspect.
However, those with higher copper concentrations frequently present regions where crystallization already appears (Figure 3).This is most probably due to: (i) a decrease of the glass stabilization with the increase of copper content; (ii) the existence of inhomogeneities on the cooling rate during the melt spinning process.These facts are also evidenced on the X-ray diffraction measurements, where many of the higher copper concentration samples already show small crystallization peaks, which can be indexed as Te and Cu 1.33 Ge 0.54 Te 2 (Figure 2).The extreme composition 30Cu:70Te, albeit still presenting some disorder, is formed by Te and CuTe, and therefore its physical properties have not been studied.DSC measurements show a single glass transition for all compositions, albeit copper samples present more than one exothermal peak, pointing to a sequential crystallization of more than one phase (Figure 4).This can be understood by the different atomic-scale structures of the different Cu x+y Ge 20-x Te 80-y glasses, with the Te segregation followed by the Cu 1.33 Ge 0.54 Te 2 crystallization, similarly to what was observed in Ge 20-x Te 80-x (x = 5, 10) glasses (Te segregation followed by GeTe crystallization) [21] and points to the existence of tellurium atoms not directly bonded to germanium or copper atoms (Te atoms connecting ATe 4 (A = Ge, Cu) tetrahedra).A significant decrease of the glass transition (T g ), crystallization (T C ) and melting (T m ) temperatures with the increase of copper concentration is observed, confirming the reduction of the glass stability, already indicated by the existence of crystalline regions.
A maximum decrease of ~50ºC is observed for the higher copper concentration glass, Cu 27.5 Ge 2.5 Te 70 , when compared with Ge 20 Te 80 , indicating a limited temperature range of applicability of Cu x+y Ge 20- x Te 80-y glasses as thermoelectric materials.
The variation of the room temperature resistivity of the Cu x+y Ge 20-x Te 80-y glasses as a function of composition is plotted in Figure 5.A large drop of five orders of magnitude, from ~3x10 8 to 2.6x10 3 m, is observed on the resistivity, the lower values corresponding to glasses with higher copper and lower germanium concentrations.This drop is much higher than those previously observed for the Ag x GeTe 4.7 (0≤x≤1.4)and Cu x Ge 15 Te 85-x (0≤x≤9) glasses [10,11], probably due to the successful preparation of glasses with higher copper concentrations in the present work and to the different main type of atomic substitution.The electrical conductivity of chalcogenide glasses has been ascribed to depend on three major factors: (i) the bond strengths, (ii) the network connectivity and (iii) the density [22].In the Cu x+y Ge 20-x Te 80-y glasses the main factor should be the first one (change in the bond strengths).Indeed, in the Cu x+y Ge 20-x Te 80-y prepared glasses it is expected that copper mainly replaces germanium in the GeTe 4 structural unit, which neither changes the network connectivity nor appreciably decreases the glass density (the density change produced by the lower atomic mass of copper should be compensated by its lower atomic volume, when compared with germanium).The formation of new CuTe 4 structural units could significantly change the network connectivity, but its increase is expected to cause a larger splitting between  (bonding) and * (antibonding) orbitals, and, consequently, should increase the electrical resistivity [22], which is not observed.On the other hand, the Cu-Te bond dissociation energy (230.5±14.6 kj mol-1 [23]) is significantly smaller than the Ge-Te dissociation energy (396.7±3.3 kJ mol-1 [23]), and therefore, a strong increase of the electrical conductivity is expected due to this factor [22].
All the Cu x+y Ge 20-x Te 80-y glasses show a semiconducting behavior, with the electrical resistivity increasing with the decreasing temperature (Figure 6).The temperature dependence of the electrical resistivity, (T), obeys the (T) = (0) exp(E a /kT) relation, where T represents the temperature, E a is the activation energy for the electronic conduction (half of the energy gap) and k represents the Boltzmann constant.A significant variation of the high temperature activation energy can be observed with the change of composition (Table 1), the higher and the lower values (470 meV and 126 meV) being observed for the un-substituted Ge 20 Te 80 and with upper copper concentration glass, Cu 27.5 Ge 2.5 Te 70 , respectively.When it was possible to measure a large enough temperature range, it was observed that the activation energy starts to slightly decrease below a certain temperature.The magnitude of the high temperature activation energies confirms the narrow-gap semiconducting character of the studied glasses, being similar to those observed on the best classical thermoelectric materials, PbTe (~100-250 meV), Si-Ge (~350-550 meV) and Bi 2 Te 3 (~75-140 meV).
Albeit the fact that room temperature resistivity values of the Cu x+y Ge 20-x Te 80-y glasses are consistent with those expected for narrow-band-gap semiconductors, the best ones are still one order of magnitude higher than those observed on new materials with good thermoelectric properties, such as the skutterudites or half-Heusler phases [24,25].
The variation of the room temperature Seebeck coefficient of the Cu x+y Ge 20-x Te 80-y glasses as a function of composition is plotted in Figure 7.Although the resistivity decreases by five orders of magnitude, the increase of copper concentration has a much limited effect on the Seebeck coefficient, with all glasses presenting very large values.A decrease of ~40% from the original Ge 20 Te 80 value (980 V K -1 ) is first observed after a small copper introduction in the composition, but further increase of copper to higher concentrations just slightly decreases this value, which stabilizes at ~400 V K -1 .
The variation of the Seebeck coefficient of the Cu x+y Ge 20-x Te 80-y glasses as a function of temperature is plotted in Figure 8.The absolute values of the Seebeck coefficients are always positive, indicating dominant p-type conduction for all the Cu x+y Ge 20-x Te 80-y glasses.In glasses with higher copper concentrations, the Seebeck coefficient is only slightly dependent on the temperature, increasing with the decreasing temperature, and their values ranging from 400 to 500 V K -1 ; the low copper concentration glasses show a higher increase of the Seebeck coefficient with the decreasing temperature.
The combination of a very large decrease of the electrical resistivity, together with the stabilization of the Seebeck coefficients at high values, has as a consequence a dramatic increase (of five orders of magnitude) of the power factor with the increasing copper concentration (Figure 9).An exponential increase of the power factor with increasing temperature is observed, the maximum value of 60 W/K 2 m being obtained for the Cu 27.5 Ge 2.5 Te 70 composition at T = 300 K, the higher temperature measured.
Due to the difficulty in obtaining Cu x+y Ge 20-x Te 80-y glasses with an appropriate shape, it was decided to measure the thermal conductivity only on Ge 20 Te 80 and to use the obtained values for the other compositions, after correcting them with the help of the Wiedemann-Franz law.The measured thermal conductivity of Ge 20 Te 80 is extremely low, with a value of ~0.1 W/Km at 300 K. Due to the high difference between the thermal conductivities of the sample and the constantan wire, used as reference, the heat radiation losses are not negligible, especially at high temperatures, (but unfortunately not easy to estimate), so the true value of Ge 20 Te 80 thermal conductivity is certainly smaller than 0.1 W/Km at 300 K.The very low thermal conductivity value of the Ge 20 Te 80 glass is in agreement with measurements made on other chalcogenide glasses, where very low thermal conductivities were also observed [26].These low values are most probably mutually consequence of the high disorder and of the high atomic weights of the constituents.The germanium replacement by copper is not expected to greatly change the lattice thermal conductivity of the glass, as any effect of the mean atomic weight decrease should be compensated by the lower Cu-Te bond strength, which should increase the rattling.
The relatively high value of the electrical resistivity of Cu x+y Ge 20-x Te 80-y glasses at room temperature (Table 1) results on a low electronic contribution to their thermal conductivity: in Cu 27.5 Ge 2.5 Te 70 , the glass with the higher electrical conductivity value, this contribution is only ~3x10 -4 W/Km and therefore can be neglected.For the Cu 27.5 Ge 2.5 Te 70 glass, which has the higher power factor at 300 K, and considering a 0.1 W/Km value of the thermal conductivity at room temperature, a figure of merit value of ZT = 0.19 is obtained.Albeit not reaching the values of the best actual thermoelectric materials at 300 K, it is a relatively high ZT and definitely puts conducting glasses as a new class of materials candidate for obtaining high performance thermoelectric materials.
A deeper understanding of the nature of Cu x+y Ge 20-x Te 80-y glasses can give some hints on the way to optimize their properties.The Seebeck coefficient of a semiconducting chalcogenide glass can be expressed as [27]  = ±k/e (E  /kT + A) where e represents the electronic charge, E  is the activation energy for the thermoelectric power, A is a constant that depends on the mechanism of the electrical transport, and the positive and negative signs represent the p-and n-type conduction mechanism, respectively.The observed positive Seebeck coefficient at high temperatures is very common in chalcogenide glasses, being consistent with an intrinsic conduction: if we suppose that we have an intrinsic semiconductor, the positive Seebeck coefficient is a consequence of the holes mobility being much higher than the electrons mobility in that temperature region.The activation energies obtained by fitting the data using this expression [27] at high temperatures, E  , are considerably smaller than the activation energies obtained from the resistivity data (Table 1).This large difference observed in the activation energies, E a and E  , points a conduction occurring predominantly in band tails, with more density of states in the valence band, when compared to the conduction band.The difference E hop = E a -E  is the hopping energy for holes or small polarons, also shown on Table 1, which is not very surprising since these materials are amorphous semiconductors, with no long range order.The values obtained for the hopping energies, between 0.08V and 0.25V, are typical for chalcogenide glasses [27].In some samples it is very clear that the Seebeck coefficient presents a smooth maximum around 155 K, which is probably due to the transition to a further localized variable range hopping conduction regime at lower temperatures [27].The predicted temperature dependence,  =  o exp(A/T 1/4 ), for the variable range hopping regime is not clearly observed at low temperatures, most probably because the change in regime occurs very smoothly and the temperature range of measurements was not wide enough.
The likely intrinsic semiconducting nature of the Cu x+y Ge 20-x Te 80-y glasses opens the possibility of further change their composition to optimize the electrical properties.Moreover, it should also allow the addition of glass stabilizing agents without degrading their electrical properties, in order to increase the glass transition temperatures and consequently their maximum temperature of application as thermoelectric materials and ZT values.

Conclusion
New chalcogenide glasses, with Cu x+y Ge 20-x Te 80-y (0≤x≤20; 0≤y≤10) compositions, have been prepared by melt spinning and used to test the possibility of obtaining conducting glasses for thermoelectric applications.Their short-range order is most likely analogous to Ge 20 Te 80 glasses, being based on CuTe 4 and GeTe 4 structural units.The addition of copper reduces the glass stability, but increases their thermoelectric properties, with a consequent huge increase on the power factor at T = 300 K, from 3.3x10 -3 W/K 2 m up to 60 W/K 2 m for the glasses with the extreme Ge 20 Te 80 and Cu 27.5 Ge 2.5 Te 70 compositions, respectively.The exceptionally low thermal conductivity measured on the Ge 20 Te 80 glass, together with the type of structural replacement and the low electrical contribution for the thermal conductivity point to a similar behavior on all series, definitely indicating the Cu x+y Ge 20- x Te 80-y -based glasses as having good potential for high performance thermoelectric materials.Their probable intrinsic semiconducting nature opens the possibility of further improvements, and clearly indicates conducting glasses as very promising thermoelectric materials.

Figure 3 .
Figure 3. Photo of a Cu 22.5 Ge 2.5 Te 75 glass, including some crystallized part.

Figure 4 .
Figure 4. DSC measurements versus temperature showing glass transition (T g ) and crystallization

Figure 5 .
Figure 5. Variation of the room temperature resistivity of the Cu x+y Ge 20-x Te 80-y glasses as function of

Figure 6 .
Figure 6.Electrical resistivity versus temperature in the Cu-Ge-Te glass system.

Figure 7 .
Figure 7. Variation of the room temperature Seebeck coefficient of the Cu x+y Ge 20-x Te 80-y glasses as

Figure 8 .
Figure 8. Variation of the Seebeck coefficient of the Cu x+y Ge 20-x Te 80-y glasses versus temperature.

Figure 9 .
Figure 9. Variation of the room temperature power factor of the Cu x+y Ge 20-x Te 80-y glasses as function of Figure 2

Table 1 .
Electrical transport properties ( 300K , E  ,  300K , E  , E Hopp and power factor parameters) of the Ge 20 Te 80 -based glasses