Thermoelectric power generation: from new materials to devices

Abstract

Thermoelectric technology offers the opportunity of direct conversion between heat and electricity, and new and exciting materials that can enable this technology to deliver higher efficiencies have been developed in recent years. This mini-review covers the most promising advances in thermoelectric materials as they pertain to their potential in being implemented in devices and modules with an emphasis on thermoelectric power generation. Classified into three groups in terms of their operating temperature, the thermoelectric materials that are most likely to be used in future devices are briefly discussed. We summarize the state-of-the-art thermoelectric modules/devices, among which nanostructured PbTe modules are particularly highlighted. At the end, key issues and the possible strategies that can help thermoelectric power generation technology move forward are considered.

This article is part of a discussion meeting issue ‘Energy materials for a low carbon future’.

1. Introduction

Thermoelectric materials have drawn tremendous attention in the past two decades because they can enable devices that can harvest waste heat and convert it to electrical power thereby promising to improve the efficiency of fuel utilization [1]. The efficiency of a thermoelectric material is defined by the dimensionless figure of merit ZT = S²σT/κ, where S is the Seebeck coefficient, σ is the electrical conductivity, T is the operation temperature and κ is thermal conductivity that is composed of carrier contribution (κ_car) and lattice contribution (κ_lat). The efficiency of a thermoelectric device (η), however, depends not on ZT maximum at a given temperature but on the average ZT (ZT_ave) value over a wide temperature range as η = [(T_H − T_C)/T_H] [(1 + ZT_ave)^1/2 − 1]/[(1 + ZT_ave)^1/2 + T_C/T_H], where T_H and T_C are the hot side and cold side temperature, respectively. Thus, for practical device applications, the ZT_ave is the critical quantity that must be maximized over the entire operation temperature range, rather than maximum ZT (ZT_max). In addition to requiring high ZTs and low cost, the materials must have enough mechanical strength for robust device fabrication and high thermal stability for long-term service [2].

There have been several important breakthroughs in increasing the ZT of thermoelectric materials since 2000, as summarized in table 1 [3–26]. The ZT enhancement originates either from increase of power factors (PF = S²σ) [17,27,28] or decrease of κ_lat through microstructure engineering [6,9,29–31]. Several materials have now been reported to display high ZT values over 2 [9,19,25,26,30,32–37]. However, only a few of them have been demonstrated in thermoelectric devices or modules [38].

Table 1.

Record ZTs of some typical thermoelectric materials since the year 2000.

material	ZTmax at T	ZTave (TC − TH)	p/n	reference
Bi83.5Sb16.5	0.52 (80 K)	0.47 (50–100 K)	n	[3]
CsBi4Te6	0.8 (225 K)	0.44 (100–250 K)	p	[4]
Bi2Te2.79Se0.21	1.2 (357 K)	1.02 (300–450 K)	n	[5]
Bi0.5Sb1.5Te3 + x%Te	1.86 (320 K)	1.42 (300–450 K)	p	[6]
MgAg0.97Sb0.99	1.3 (533 K)	1.1 (300–550 K)	p	[7]
PbTe0.998I0.002-3%Sb	1.8 (773 K)	0.97 (300–773 K)	n	[8]
Pb0.98Na0.02Te-8%SrTe	2.5 (923 K)	1.67 (300–900 K)	p	[9]
PbSe0.998Br0.002-2%Cu2Se	1.8 (723 K)	1.1 (300–823 K)	n	[10]
Pb0.98Na0.02Se-2%HgSe	1.7 (973 K)	0.63 (400–900 K)	p	[11]
PbS-4.4%Ag	1.7 (850 K)	0.93 (300–850 K)	n	[12]
Pb0.975Na0.025S-3%CdS	1.3 (923 K)	0.55 (300–923 K)	p	[13]
β-Zn4Sb3	1.4 (750 K)	0.81 (300–750 K)	p	[14]
Ba0.08La0.05Yb0.04Co4Sb12	1.7 (850 K)	1.06 (300–850 K)	n	[15]
DD0.7Fe2.7Co1.3Sb11.8Sn0.2	1.45 (823 K)	0.82 (423–823 K)	p	[16]
Mg2.15Sb0.006Si0.28Sn0.71	1.3 (700 K)	0.93 (300–773 K)	n	[17]
Mg2Li0.025Si0.4Sn0.6	0.7 (650 K)	0.41 (300–750 K)	p	[18]
SnSe1−xBrx	2.8 (773 K)	1.29 (300–773 K)	n	[19]
SnSe	2.6 (923 K)	1.37 (600–923 K)	p	[20]
(Si95Ge5)0.65(Si70Ge30P3)35	1.3 (1173 K)	0.7 (300–1173 K)	n	[21]
Si80Ge20Bx	0.6 (1100 K)	0.55 (373–1300 K)	p	[22]
Zr0.2Hf0.8NiSn0.985Sb0.015	1.1 (1000 K)	0.66 (300–1100 K)	n	[23]
Nb0.88Hf0.12FeSb	1.5 (1200 K)	1.0 (500–1200 K)	p	[24]
Cu2Se-x%CNT	2.4 (1000 K)	1.5 (450–1000 K)	p	[25]
Cu1.97S	1.7 (1000 K)	0.77 (300–1000 K)	p	[26]

This article focuses on the potential of the new thermoelectric materials with extraordinary performance in being implemented in thermoelectric modules specifically for power generation (organic materials and their devices fall out of scope because efficiencies are still too low [39]). Unless the promise to build robust thermoelectric modules is ultimately realized, thermoelectric materials will remain in the basic research domain of this important energy conversion technology. The article is organized as follows: first, we briefly discuss the most important recent developments in high-performance thermoelectric materials, particularly those with good stability, low cost and scalable potential; second, we discuss the novel device fabrication technologies made of these materials; finally, we consider future challenges and possible strategies for further enhancing ZTs and device performance. We note that this article is not a comprehensive review of the literature but is a selected snapshot of the module-related field reflecting the authors' individual interests in developing new materials for modules.

2. Thermoelectric materials

(a). Low-temperature materials (300–500 K)

Since its discovery in the 1950s, Bi₂Te₃ and its alloys with ZT values of approximately 0.9–1.0 around room temperature have been widely commercialized but mainly for cooling applications [40]. Thermoelectric modules made of conventional Bi₂Te₃-based materials have been commercialized for decades. Recent researches mainly focus on their soldering and welding technologies [41,42] as well as fabrication of flexible Bi₂Te₃ thermoelectric devices on organic substrates [43]. However, because these modules are based on conventional material we are not going to give a detailed survey of Bi₂Te₃ thermoelectric devices in this article.

The concept of nanosizing was applied in the 2000s to this system and marginally boosted ZTs to just over 1 [29,44–49]. The pioneering work was done by the Chen and Tang groups in late 2000s, where ball milling [29] and melt spinning techniques [47] were used to create nanostructuring inside the bulk materials. These nanosized grains were found to be effective in scattering heat-carrying acoustic phonons and reducing thermal conductivity. Hot deformation [5,49] developed by the Zhao group was also shown to lower thermal conductivity of bismuth telluride-based materials. In addition to reducing κ_lat, researchers also tried to enhance ZT values of Bi₂Te₃-based alloys by increasing S²σ through band structure engineering. For instance, Sn was found to form resonant impurity states in the valence band of Bi₂Te₃, thereby largely enhancing S of the host material at cryogenic temperatures through resonant scattering [50,51]. The well-accepted best ZT values reported for p- and n-type Bi₂Te₃-based thermoelectric alloys are approximately 1.5 [29,47] and 1.2 [5] (both at approx. 400 K), respectively. It is worth mentioning that Kim et al. [6] in 2015 reported a peak ZT of approximately 1.86 at 320 K in p-type Bi₂Te₃, which was a purported record high. However, a very recent study clearly suggests that such an abnormal ‘too good to be true’ ZT might originate from an incorrect combination of values: (i) thermal conductivity measured along the pressing direction of anisotropy, (ii) charge transport measured in the direction perpendicular to the anisotropic direction [52].

Although improved thermoelectric performance was reported in Bi₂Te₃ alloys in a laboratory scale [5,6,44,45,47,49], the ability to scale up of these materials, a key factor for industry production, remains an open question. Recently, Zheng et al. prepared large p-type Bi₂Te₃-based pellets of 30, 40 and 60 mm in diameter by melt spinning followed by plasma-activated sintering techniques (PAS), and examined their uniformity, thermoelectric properties as well as mechanical strength [53]. It was shown that, influenced by an ellipsoidal-shape-distributed temperature field during the sintering process, the as-prepared large pellet is inhomogeneous because of compositional fluctuation, figure 1a. However, by appropriately elongating the sintering time (figure 1b–d) or designing graphite dies (figure 1e–g), the temperature distribution homogeneity is largely improved during the sintering process, and the pressed pellets' uniformity is therefore greatly enhanced. The large Bi₂Te₃-based pellets show a peak ZT of 1.15 at 373 K, which is approximately 20% improvement over the commercial zone-melted Bi₂Te₃ ingots (figure 1h). Moreover, the compressive and bending strengths of the pressed pellets are significantly improved compared to zone-melted ingots, figure 1i. This work highlighted the possibility of fabricating large mechanically robust new generation of Bi₂Te₃-based materials with high thermoelectric performance at temperatures less than 500 K.

Figure 1.

Sample homogeneity demonstration in Bi₂Te₃ bulk thermoelectric materials. The distribution of Seebeck coefficients over the cross section of pressed pellets with the diameter of 30 mm and height of 12 mm sintered for the time of (a) 5 min, (b) 10 min, (c) 20 min, (d) 30 min and for the graphite dies with an outer diameter of (e) 60 mm, (f) 90 mm, (g) 120 mm. (h) A comparison of ZT values measured on different samples. Square symbols are for samples prepared by melt spinning and spark plasma sintering (MS-SPS); [53] solid curves are for samples prepared by conventional melting followed by spark plasma sintering (M-SPS); the dashed line is for zone-melted (ZM) sample. $\mathrm{\perp }\mathrm{p}$ and $\parallel \mathrm{p}$ represent the directions perpendicular and parallel to the sintering pressure, respectively. (i) The compressive and bending strength of Φ30 mm diameter MS-SPS samples with sintering duration of 5 min and 20 min, respectively. The data for the ZM sample are also included for comparison. Adapted from Zheng [53]. (Online version in colour.)

α-MgAgSb is an interesting new entry in the field of thermoelectrics with a narrow band gap (approx. 0.1 eV at 300 K), which may be a candidate for making device modules despite some undesirable phase transitions above room temperature [54,55]. It has been reported that ball milling combined with hot pressing technology can be employed to synthesize this material and [56] allows the precise control and fine-tuning of chemical stoichiometry. MgAgSb features complex crystal structures [54] and low sound velocities (1276 m s⁻¹) [57], which are supposed to contribute to its intrinsically low thermal conductivity (less than 0.7 Wm⁻¹K⁻¹ at T > 300 K) [56]. Moreover, it has a high valence band degeneracy of 8 and a large density-of-states effective masses (m* = 0.6 m_e, m_e is the inertial mass of free electrons), [58] enabling high power factors over 20 µW cm⁻¹K⁻² above 300 K [56]. Altogether, a peak ZT of approximately 1.2–1.4 at 550 K was reported for this material with p-type conduction [56,59,60]. More importantly, a realistic conversion efficiency of approximately 8.5% for a unileg under a temperature difference of 225 K has been experimentally demonstrated [61]. Even so, the main problem with the MgAgSb-based thermoelectric materials is the presence of silver metal which may make it cost prohibitive to develop an affordable technology.

(b). Mid-temperature materials (500–900 K)

In terms of ZT lead chalcogenides are the highest performing thermoelectric materials operating in the mid-temperature range and have a long history of powering several spacecraft launched by NASA [8–11,13,27,30–34,37,38]. They crystalize in a simple cubic rock-salt structure and possess intrinsically low thermal conductivity, which is ascribed to many factors including the unique ‘off-centring’ behaviour of Pb atoms [62]. Among them PbTe is the most studied and advanced in thermoelectric figure of merit with ZTs exceeding 2 for p-type PbTe [9,30,32,34] and 1.5 for n-type systems [33,63,64]. A maximum conversion efficiency of approximately 8.8% at a temperature difference of 570 K has already been demonstrated in the nanostructured PbTe-based module to be described below [38]. When segmented with Bi₂Te₃, the efficiency improved to approximately 11% under a temperature difference of 590 K, a record high value for PbTe-based systems [38]. We will discuss more on this PbTe device later in this article.

In consideration of the high cost of Te, there have been incentives to develop high performance in PbSe or PbS as well since they too have low thermal conductivity (1.7 Wm⁻¹K⁻¹ and 2.6 Wm⁻¹K⁻¹ at 300 K, respectively) and similar crystal and electronic structures [65]. Lee et al. [66] reported that Sb-doped PbSe shows a ZT of approximately 1.5 at 800 K, while Bi-doped material has a ZT of only approximately 0.9 at 920 K [66]. Very recently, Luo et al. [67] reported n-type GeSe-alloyed PbSe which attains a peak ZT approximately 1.54 at 773 K. This extraordinary ZT matches that of PbTe materials and stems from its ultralow lattice thermal conductivity of approximately 0.36 Wm⁻¹ K⁻¹ at 573 K, which approaches the theoretical limit of amorphous PbSe. First-principles calculations unravel that the alloyed Ge atoms prefer to stay at off-centre lattice positions (discordant atoms), inducing low-frequency modes and contributing to low κ_lat [67].

There have been great progresses in p-type PbSe materials as well. For example, CdSe or HgSe alloying has been used to converge the two valence bands of PbSe for higher Seebeck coefficients [11,68]. MSe (M = Ca, Sr and Ba) as second phases enhance the ZT values of PbSe by reducing the thermal conductivity without harming the electrical conductivity too much [69]. The best ZT reported so far for p-type PbSe is approximately 1.7 around 900 K [11]. As both high-performance n- and p-type PbSe materials can now be made, they might be on their way to becoming competitive with PbTe for mid-temperature thermoelectric power generation, and if they succeed this will be a great development considering their lower cost and higher melting point [70].

Figure 2a,b shows the ZT values and ZT_ave of the best performing lead chalcogenides [8–13] reported so far. Undoubtedly, both p- and n-type PbTe have the highest performance over the entire temperature range of interest. All n-type lead chalcogenides show ZT_ave values close to or greater than unity within their respective working temperature range, which suggests a great promise for real application. Although p-type PbSe and PbTe display ZT_max > 1.5 at elevated temperatures, their lower ZT_ave respect to the n-type counterparts needs to be improved in the future studies.

Figure 2.

Thermoelectric figure of merit for the best performing p-type and n-type lead chalcogenides [8–13]: (a) ZT and (b) ZT_ave. (Online version in colour.)

Although lead chalcogenides hold the greatest promise for use in the temperature range of 600–900 K, notable materials in this range are also β-Zn₄Sb₃, skutterudites (also known as CoSb₃), magnesium silicides and single crystal SnSe [14,15,17–20,35,71–73].

β-Zn₄Sb₃ has long be viewed as a high-performance material with an outstanding ZT > 1.2, the material, however, has been problematic [74]. The main issue that prevents its use in devices is its thermal instability [71]. A recent study by Lin et al. [14] revealed that Zn atoms migrate from crystalline sites to interstitial positions after 425 K and gradually decompose the compound into Zn and ZnSb. However, above 565 K the material recovers its stability and any damage caused by the phase decomposition is repaired. They also demonstrated an excellent ZT of 1.4 at 750 K where β-Zn₄Sb₃ is thermodynamically stable [14]. Despite this fascinating behaviour, it seems that β-Zn₄Sb₃ has poor prospects for being implemented in thermoelectric devices.

Skutterudites have been some of the most promising materials for devices for over two decades, especially for n-type systems because of their larger power factors and good mechanical properties [75,76]. The cubic structure of skutterudite has void cages that can be filled by guest atoms which act point defects and rattlers that cause strong phonon scattering thereby significantly reducing κ_lat [72,77]. By carefully combining multiple species of filling guest atoms, one can introduce two or more localized rattling modes with different frequencies into the skutterudite void to scatter a wide range of phonons to reach very low thermal conductivities [78]. For example, Ba-La-Yb triple-filled skutterudites show a much-reduced κ_lat to their amorphous limit and outstanding thermoelectric performance (ZT = 1.7 at 850 K) [15]. High efficiency was also demonstrated in thermoelectric modules made of skutterudite compounds [79,80], which will be detailed later in this article.

Magnesium silicides/stannides made of light and cheap elements also show excellent thermoelectric performance [17,18]. It has been reported that n-type Mg₂Si-Mg₂Sn solid solutions prepared by a two-step solid-state reaction method display the greatest ZT of approximately 1.3 (700 K) at 70 mol.% Mg₂Sn, as a result of composition induced conduction band convergence [17]. However, it is very difficult to scale up the preparation of magnesium silicides by high-temperature melting or solid-state reaction technique, because of the large difference in melting temperature between Mg and Si. Moreover, due to the high vapour pressure, Mg can easily evaporate during high-temperature reactions, which leads to phase instability and degraded thermoelectric performance of Mg₂Si in service. Recently, ball milling as an easy-to-scale-up technology has been successfully employed to synthesize single phase magnesium silicides [81,82]. This synthesis route also avoids the Mg evaporation problem. Other than materials, a lot of device work has been done on the basis of magnesium silicides as well [83–87]. For example, Sakamoto et al. fabricated Sb-doped Mg₂Si thermoelectric device with Ni as electrodes using a PAS technique, which has no notable deterioration even after ageing for 1000 h [83]. Even so, the progress of Mg₂Si-based thermoelectric device has been stagnant because of the low performance of its p-type counterpart (ZT approx. 0.7) [18].

SnSe as an emerging compound has received increasing attention in thermoelectric community since 2014 [20]. It has many merits for advanced thermoelectric materials, for example, strong anharmonic chemical bonding for low thermal conductivity, [20] multiple electronic bands for high Seebeck coefficients [35] and ‘3D charge and 2D phonon transport’ behaviour in the out-of-plane direction of its single crystal ingot for high n-type electrical conductivity [19]. These give peak ZTs of approximately 2 and approximately 2.8 (both at 773 K) for single crystals of p- and n-type SnSe, respectively [19,35]. Despite very high thermoelectric performance found in SnSe single crystals, their easy-to-cleave nature limits their processability during cutting and machining. Researchers have been focusing intensely on polycrystalline SnSe, which is more mechanically robust. However, polycrystalline SnSe prepared and handled in air displays significantly higher thermal conductivity and therefore much lower ZTs than their single crystalline form [88–90]. This is surprising because usually, polycrystals have more grains and defects than single crystals, which should contribute to lower thermal conductivities. This was attributed to surface oxidation forming SnO₂ which has a very high thermal conductivity [91]. A recent study by Chung and colleagues confirms that the underperformance of polycrystalline SnSe is mainly due to the presence of tin oxides on the surface of the sample [92]. They used an H₂ reduction process to remove the majority of tin oxide films that ubiquitously cover SnSe grains and eventually unveiled the ultralow thermal conductivity intrinsic in this material. This study resolves the thermal conductivity discrepancy between single and poly-crystalline SnSe, and attains a ZT approximately 2.5 for polycrystalline SnSe which nearly matches the record-high value of the single crystals. We believe SnSe might revolutionize thermoelectric power generation technology if its device can be demonstrated, but to date such reports are few and will be discussed later [93,94].

(c). High-temperature materials (greater than 900 K)

SiGe alloys were some of the prototypical high-temperature thermoelectric materials to be studied for more than half a century [21,22]. The n-type SiGe shows a promising ZT of 1.1 at approximately 1200 K [21]. However, the relatively poor thermoelectric performance of the p-type counterpart (peak ZT = approx. 0.6) and the high cost of Ge limit the large-scale commercial application of SiGe alloys [22]. The old Si/Ge-based devices are a legacy system and they are outside the scope of this article.

Half-Heusler (HH) alloys are another group of materials which have gained popularity as new candidates for high-temperature thermoelectric power generation because of their excellent electrical properties [24,95–99], very good mechanical robustness [100] and high thermal stability [101]. They belong to a large family of ternary intermetallics with a general formula of ABX, where A is usually the most electropositive rare earth or early transition metal, B is a less electropositive transition metal and X is a main group element [102]. Their physical properties are largely determined by the valence electron count (VEC) [103]. High thermoelectric performance is generally achieved in the semiconducting HH phases with VEC = 18 [23,97,99,104–112], while VEC > 18 usually leads to metallic conduction behaviour [113].

The HH alloys have seen great advances in ZT of both n- and p-type samples, figure 3a,b. ZrNiSn and ZrCoSb are the two most important n-type compositions [95,104,107,109–111,114,117,118,123–125]. ZrNiSn, for example, exhibits a high power factor [23,110,117,126] which originates from a large Seebeck coefficient because of a large density-of-states conduction band effective mass (2.8 m_e at 300 K) [110]. Sb doping at Sn sites is often used to tune the carrier density to an optimal level (n_opt = approx. 3–4 × 10²⁰ cm⁻³), leading to a ZT of approximately 0.8 at 873 K [127]. The main shortcoming of HH materials is their high lattice thermal conductivity and enormous effort has gone into finding ways to decrease it [23]. Grain size refinement for example by ball milling or using in situ second phase nanostructuring similar to the PbTe strategies mentioned above have been used to decrease κ_lat of ZrNiSn alloys [104,106,119]. However, this reduction in κ_lat is often offset by the loss of carrier mobility because of the comparable mean free paths of electrons and phonons (approx. 1 nm) in the ZrNiSn system, making the net enhancement of ZT marginal [131]. Recent studies suggest that isoelectric substitution of Zr by heavier Hf effectively may reduce κ_lat of ZrNiSn compounds via strong point defect scattering, without obviously deteriorating their electrical properties [95,123,126]. The highest ZT of approximately 1.1 at 1000 K and a ZT_ave of 0.58 (300–973 K) has been obtained for Zr_0.2Hf_0.8NiSn_0.985Sb_0.015 [95].

Figure 3.

Temperature-dependent ZT values for typical (a) n-type [23,96,104,105,110,113–116] and (b) p-type HH alloys [24,99,108,117–122]. (c) Evolution of the peak ZT values for some typical HH alloys in 20 years [24,95,97–99,104,106–112,114,115,117,123–129]. Adapted from Yu [130]. (d) Maximum power output and conversion efficiency as a function of hot side temperature for the thermoelectric device made of n-type ZrNiSn and p-type FeNbSb-based HH materials. The dash line is theoretically calculated conversion efficiency of the module by assuming that no electrical or thermal contact resistances exist. Adapted from Fu [24]. (Online version in colour.)

The thermoelectric properties of p-type HH alloys have attracted significant effort in the last 5 years because ZT in these systems is even lower than in the n-type systems (figure 3c). For instance, the peak ZT of NbFeSb has been reported to increase rapidly from only approximately 0.4 in 2011 [108] to a record value of approximately 1.5 in 2015 [24] by alloying with Ti [112]. However, limited by the low solubility of Ti, the optimal carrier concentration (n_opt) could not be reached [98]. Two approaches were attempted to solve this problem: (i) since n_opt is proportional to m*, substituting Nb partially by V could lower m* (due to the less spatially localized 4d orbital of Nb than the 3d orbital of V) and therefore decrease n_opt [97]; (ii) looking for a more efficient dopant than Ti, for example Hf. This is not only good for electrical properties enhancement, but provides more opportunities for thermal conductivity reduction because the heavier Hf doping results in stronger mass fluctuation and reinforced phonon scattering [24,107,110,123,126]. As a result, an outstanding ZT of approximately 1.5 at 1200 K together with a promising ZT_ave of approximately 1 (500–1200 K) is achieved in the composition Nb_0.88Hf_0.12FeSb [24]. More importantly, an eight-couple prototype thermoelectric module made of n-type ZrNiSn and p-type NbFeSb with a high conversion efficiency of 6.2% has been demonstrated experimentally at a temperature difference of approximately 650 K [24]. We will discuss this in detail later. In addition, some facile and cost-efficient fabrication methods (self-propagating high-temperature synthesis, thermal combustion, laser) have been developed to prepare HH alloys [132–134], which point to a reasonable path to scalability [24,110,123,127]. All these advances clearly demonstrate the great potential of HH alloys for high-temperature thermoelectric power generation.

There are other thermoelectric materials that have been explored in the past several years for relevant to high temperatures, for example copper chalcogenides [25,26]. These materials have rigid anion sub-lattices but contain too much copper whose state in the lattice is ‘liquid-like’. This leads to extremely low thermal conductivities but also to destructive electroplating of copper metal prolonged under bias [135]. Although high ZTs over 1.5 have been seen in copper chalcogenides [25,26,135], their stability to high electrical currents or large temperature gradients is poor because of the highly mobile copper ions [136,137]. Recently, efforts have been described aimed to address this issue. By constructing a series of electronically conducting, but ion-blocking barriers to reset the chemical potential of such mixed ionic–electronic conductors seems to help keep the system below the threshold (approx. 0.11 V) for decomposition [138]. Even so, there is still a long way to go before the copper-rich chalcogenides are brought into real use.

Despite the above-mentioned extensive efforts to develop high-ZT materials, and the dramatic progress achieved [1,139–148], there have been few efforts to move the field to the next level and closer to commercialization, which is module development employing the new higher performance materials. Module fabrication using newly developed high-ZT materials is not a trivial pursuit but is needed for achieving practical thermoelectric power generation and to justify continued research efforts in the field of thermoelectrics.

3. Thermoelectric devices

(a). Design of thermoelectric devices and modules

Below we describe the most important developments and efforts in using these very new materials described above in fabrication of functional power generation devices and modules. The maximum conversion efficiency of a thermoelectric device for power generation (η_max) theoretically defined using two terms, Carnot efficiency (T_h − T_c)/T_h and the average (device) ZT of the temperature drop (ZT_ave) [149–151]. The actual efficiency obtained is normally lower than the calculated value because of parasitic ohmic losses at the electrode interfaces and heat losses in various parts in a module.

Figure 4a shows a schematic of a thermoelectric module made from thermocouple of p- and n-type thermoelectric legs. The thermocouples are alternately joined by interconnect electrodes in electrical series but in a thermally parallel connection (figure 4b). The interconnect metallic electrodes need to have very low electrical contact resistance to extract effectively electric power from legs and high thermal conductivity to supply effectively thermal power to legs. An effective way to enhance the conversion efficiency is a stacking of different materials/legs (segment type) or stacking of different modules (cascade type). A segmented module is fabricated by stacking materials/legs with high ZT at high temperature on top of materials/legs with high ZT at low temperature (figure 4c). In a cascaded module, at least two different modules are used. One is a module with high efficiency at higher temperature and is stacked on top of a second module with high efficiency at lower temperature (figure 4d).

Figure 4.

Schematic of (a) a thermoelectric module made from thermocouple of p- and n-type thermoelectric legs. Thermocouple configuration: (b) single stage, (c) segment type and (d) cascade type. (Online version in colour.)

In fabricating good thermoelectric modules, the most important factor is to realize stable low resistance electrical contacts (especially on the hot side) between the interconnect electrodes and the thermoelectric materials [140,152,153]. A module must be exposed to high temperature, which can result in chemical reaction and/or atomic diffusion at interface between the interconnect electrodes and the thermoelectric materials to instability and degradation. These phenomena lead to the gradual development of high-resistance secondary phases at the interface and/or degeneration of the thermoelectric materials, reducing conversion efficiency of a module. Therefore, the formation of excellent diffusion barriers between interconnect electrodes and the thermoelectric materials is needed to prevent the chemical reaction and/or atomic diffusion on the sides of the modules (figure 4b)). This is a challenging task because a fundamental understanding of how to achieve such barriers while at the same time achieving minimal contact resistance is not yet available and basic research on the subject is lagging. Part of the reason for scarce basic research activity on this subject in the past decades, we believe, is the broad perception that this is essentially a device engineering problem and can be solved empirically in a straightforward fashion, an attitude which up to now has clearly been proven incorrect.

The high-temperature operation imposes additional challenges in the joining methods that can be used between materials, diffusion barriers and interconnect electrodes. Good match in the coefficient of thermal expansion between these materials is also needed to reduce thermal mechanical stress [140,152,153]. Therefore, to achieve full benefit from the high-performance materials now available, we must learn how to create low electrical contacts while taking into account the nature of the diffusion barrier, the joining method and the coefficient of thermal expansion. This is a real fundamental materials research problem with many unanswered questions, which needs the support of funding agencies engaged in supporting basic research.

A reduction in heat loses is critical for the development of efficient modules operating at high temperatures (especially greater than 573 K). Heat leaks through parasitic radiation from the hot side to the cold side in the open space between thermoelectric legs, radiation losses from the side walls of legs, poor thermal contacts between the hot source and the hot side of the modules all reduce conversion efficiency.

Table 2 lists η_max values for the state-of-the-art modules fabricated with newly developed high-ZT materials. η_max > 7% has been achieved in single-stage modules of nanostructured PbTe [38,154], Sb₂Te₃/Bi₂Te₃ [155], skutterudite [79,80,157,163] and silicides [155]. For example, η_max of approximately 7% was achieved in skutterudite-based modules with Mo diffusion barrier for T_h = 872 K and T_c = 314 K [157]. The η_max was enhanced in segmented and cascaded modules [38,154,155,159–161]. For example, η_max of silicide-module increases from approximately 8% in single stage T_h = 823 K and T_c = 303 K to approximately 12.1% in cascaded module with Sb₂Te₃/Bi₂Te₃ for T_h = 853 K and T_c = 303 K [155]. Recently, ‘potential’ η_max have been reported in unileg only composed of p-type leg or n-type leg, including HH, MgAgSb, Mg₃Sb₂ and Cu₂₆Nb₂Ge₆S₃₂ [61,164–167]. For MgAgSb, the p-type unileg with Ag contact pad barrier shows η_max of approximately 8.5% for T_h = 518 K and T_c = 293 K. In this case, we do not have real devices or modules but a value predicted from single leg measurements.

Table 2.

Maximum conversion efficiency (η_max) of the state-of-the-art module with newly developed high-performance materials.

Th (K)	Tc (K)	module configuration	p-type	n-type	diffusion barrier	ηmax (%)	reference
873	303	single stage	nanostructured PbTe	PbTe	p- and n-type: Co–Fe	8.8	[38]
873	283	single stage	nanostructured PbTe	PbTe	p-type: Fe	8.5	[154]
					n-type: Co–Fe
553	303	single stage	Sb2Te3	Bi2Te3		7.8	[155]
873	323	single stage	skutterudite	skutterudite	p- and n-type: Co–Fe–Ni	8	[156]
773	313	single stage	skutterudite	skutterudite	p- and n-type: Mo	7	[157]
873	296	single stage	skutterudite	skutterudite		8.4	[79]
872	314	single stage	skutterudite	skutterudite		9.3	[80]
820	293	single stage	half-Heusler	half-Heusler		*power density: approximately 3.2 W cm−2	[158]
991	336	single stage	half-Heusler	half-Heusler		6.2	[24]
823	303	single stage	higher manganese silicide	Mg2Si		8	[155]
673	303	single stage	clathrate	clathrate		4.8	[159]
674	368	segment type	Sb2Te3/Ag0.9Pb9Sn9Sb0.6Te20 (LASTT)	Bi2Te3/Ag0.86Pb19+xSbTe20 (LAST)		6.56	[160]
873	283	segment type	Bi2Te3/nanostructured PbTe	Bi2Te3/PbTe	p-nanostructured PbTe and n-PbTe: Co–Fe	11	[38]
873	283	cascade type	Bi2Te3/nanostructured PbTe	Bi2Te3/PbTe	p-nanostructured PbTe: Fe	12	[154]
					n-PbTe: Co–Fe
863	363	segment type	Sb2Te3/Zn4Sb3/skutterudite	Bi2Te3/skutterudite		10	[161]
849	308	segment type	Sb2Te3/skutterudite	Bi2Te3/skutterudite	p- and n-skutterudite: Ti-Al	12	[162]
753 K	303	segment type	Sb2Te3/Zn4Sb3	Bi2Te3/skutterudite		7.5	[159]
853	303	cascade type	Sb2Te3/higher manganese silicide	Bi2Te3-Bi2Se3/Mg2Si		12.1	[155]
673	303	segment type	Sb2Te3/clathrate	Bi2Te3/clathrate		7.5	[159]
823	343	single stage (unicouple)	skutterudite	skutterudite	p-type: Co2Si	9.1	[163]
					n-type: CoSi2
823	323	single stage (unileg)	half-Heusler	—		9	[164]
973	317	single stage (unileg)	half-Heusler	—		11.4	[165]
518	293	single stage (unileg)	MgAgSb	—	*Ag contact layer	8.5	[61]
773	373	single stage (unileg)	Mg3Sb2	—	Fe	10.6	[166]
570	297	single stage (unileg)	Cu26Nb2Ge6S32		Au	3.3	[167]

(b). Modules based on nanostructured PbTe materials

The nanostructured PbTe-based materials have begun to make the transition from mainly materials development to module fabrication and device assembly, and simulation [38,154]. As listed in table 2, high η_max was demonstrated in nanostructured PbTe-based modules. Here, we review recent progress to bridge the technology gap between materials development and module fabrication in nanostructured PbTe.

PbTe itself is a traditional thermoelectric material, and shows high ZT of approximately 1.1 in the temperature range of 600 K to 900 K. PbTe-based modules have been used decades ago in radioisotope generators for space missions [168]. Using a radioisotope heat source the nominal efficiency of these old modules was approximately 5.1% for T_h = 783 K and T_c = 366 K [169].

Because the ZT of PbTe-based materials has now been dramatically boosted through the new strategies discussed above [28,30,170–174], recently we achieved higher η_max in nanostrucuted Pb-based modules than the old PbTe modules [38,154]. The new modules were constructed from nanostructured Pb_0.940Mg_0.020Na_0.040Te or Pb_0.953Ge_0.007Na_0.040Te as p-type legs, and non-nanostructured PbTe_0.9960I_0.0040 or PbTe_0.9964I_0.0036 as n-type legs. As shown in figure 5, ZT values of approximately 1.8 at 810 K (approx. 1.9 at 805 K) and approximately 1.4 at 750 K (approx. 1.3 at 780 K) are easily achieved for scaled-up nanostructured p- and non-nanostructured n-type materials, respectively.

Figure 5.

Temperature dependence of the thermoelectric figure of merit ZT in nanostructured Pb_0.940Mg_0.020Na_0.040Te (p-type) [38], non-nanostructured PbTe_0.9960I_0.0040 (n-type) [38], nanostructured Pb_0.953Ge_0.007Na_0.040Te (p-type) [154] and non-nanostructured PbTe_0.9960I_0.0040 (n-type) [154]. (Online version in colour.)

The fabrication process used for the nanostructured PbTe-based module is shown in figure 6 [38,154]. First, p-type and n-type pucks were produced by pressure-assisted sintering of PbTe and diffusion barriers (figure 6a). For the new PbTe-based modules, Fe-, Ni and Nb-based alloys were investigated as diffusion barriers [38,175–179]. In newly developed high-performance module, Fe and 80% Co-20% Fe were used as a diffusion barrier for p-type Pb_0.953Ge_0.007Na_0.040Te and n-type PbTe_0.9964I_0.0036, respectively [154]. Then, the two pucks were polished together to uniformize these heights (figure 6b) and diced to into rectangular legs (figure 6c). The p- and n-type legs were alternately positioned onto the insulated metal substrate (cold side), where Cu patterns were printed onto the polymer film. In hot side, the legs were interconnected by the Cu electrodes. For the module of p-type Pb_0.953Ge_0.007Na_0.040Te and n-type PbTe_0.9964I_0.0036, Ag-based nanopaste and Pb–Sn-based solder were used to join the legs and Cu interconnecting electrodes at hot and cold sides, respectively. A photograph of single-stage thermoelectric module of p-type Pb_0.953Ge_0.007Na_0.040Te/n-type PbTe_0.9964I_0.0036 is shown in figure 6d.

Figure 6.

(a) Pucks of Pb_0.953Na_0.040Ge_0.007Te (p-type) and PbTe_0.9964I_0.0036 (n-type). The modules were prepared by (b) polishing, (c) cutting and assembling the components onto the insulated metal substrate with the dimension of 18 mm × 15 mm × 1 mm. (d) Single-stage nanostructured PbTe module (p-type: Pb_0.953Ge_0.007Na_0.040Te; n-type: PbTe_0.9964I_0.0036), (e) segmented Bi₂Te₃/nanostructured PbTe module (p-type: Pb_0.940Mg_0.020Na_0.040Te; n-type: PbTe_0.9960I_0.0040), and (f) cascaded Bi₂Te₃/nanostructured PbTe module (p-type: Pb_0.953Ge_0.007Na_0.040Te; n-type: PbTe_0.9964I_0.0036). (d,f) PbTe-based legs: 2.0 mm × 2.0 mm × approximately 4.3 mm. (e) PbTe-based legs: 2.0 mm × 2.0 mm × approximately 2.8 mm, and (e,f) Bi₂Te₃-based legs: 2.0 mm × 2.0 mm × 2.0 mm. (g) Home-built testing system for thermoelectric power generation module. Adapted from Jood [154] and Hu [38]. (Online version in colour.)

The power generation characteristics of the nanostructured PbTe-based module were measured for T_h of 873 K–573 K and for T_c of 283 K in vacuum (10⁻²–10⁻³ Pa), figure 7g [38,154,159]. In this system, the conversion efficiency η was calculated from P and Q_out:

$$\eta =\frac{P}{P+Q_{\text{out}}}.$$ 3.1

The terminal voltage (V), electrical power output (P), heat dissipated from the cold side of the module (Q_out) and conversion efficiency (η) of the PbTe-based module comprising eight p-n couples are shown as a function of electrical current I in figure 7. The measured maximum power output (P_max), open-circuit heat flow (Q_oc) and η_max are listed in table 3. The η_max approximately 8.5% for T_h = 873 K and T_c = 283 K was obtained in single-stage p-type Pb_0.953Ge_0.007Na_0.040Te/n-type PbTe_0.9964I_0.0036 (figure 6d).

Figure 7.

Power generation characteristics of the nanostructured PbTe-based module (p-type: Pb_0.953Ge_0.007Na_0.040Te; n-type: PbTe_0.9964I_0.0036). Measured (a) terminal voltage (V), (b) electrical power output (P), (c) heat dissipation from the cold side of the module (Q_out) and (d) conversion efficiency (η) as functions of electrical current (I). The hot side temperature (T_h) was changed from 873 K to 573 K, while the cold side temperature (T_c) was maintained at 283 K. Adapted from Jood [154]. (Online version in colour.)

Table 3.

Measured and simulated maximum power output (P_max), open-circuit heat flow (Q_oc) and maximum conversion efficiency (${\eta }_{\text{max}}$) in single-stage nanostructured PbTe module (p-type: Pb_0.953Ge_0.007Na_0.040Te; n-type: PbTe_0.9964I_0.0036) [154], segmented Bi₂Te₃/nanostructured PbTe module (p-type: Pb_0.940Mg_0.020Na_0.040Te; n-type: PbTe_0.9960I_0.0040) [38], and cascaded Bi₂Te₃/nanostructured PbTe module (p-type: Pb_0.953Ge_0.007Na_0.040Te; n-type: PbTe_0.9964I_0.0036) [154] for the hot side temperature (T_h) of 873 K and the cold side temperature (T_c) of 283 K.

module configuration		Th (K)	Tc (K)	Pmax (W)	Qoc (W)	${\mathit{\eta }}_{\mathbf{\text{max}}}$(%)
single stage	measured	873	283	2.23	24.0	8.5
	simulated			2.58	20.6	11.1
segment type	measured	873	283	2.34	15.8	11
	simulated			2.55	10.8	15.6
cascade type	measured	873	283	1.77	13.2	12

To leverage a wider temperature range of high ZT and particularly to increase the efficiency at the lower temperatures, segmented and cascaded modules can be fabricated using good low-temperature materials such as the Bi₂Te₃ alloys. In the segmented module, the nanostructured PbTe legs were stacked on top of low-temperature materials Bi₂Te₃ legs (figure 6e). For the cascaded module, the nanostructured PbTe module was stacked on top of another Bi₂Te₃ module (figure 6f). This led to a significant enhancement in η_max to approximately 11% for the segmented module (p-type: Pb_0.940Mg_0.020Na_0.040Te; n-type: PbTe_0.9960I_0.0040) and approximately 12% for the cascaded module (p-type: Pb_0.953Ge_0.007Na_0.040Te; n-type: PbTe_0.9964I_0.0036), respectively, for T_h = 873 K and T_c = 283 K (table 3). These η_max values are among the highest achieved in modules using the new state-of-the-art materials (table 2).

A 24 h stability test was performed using various temperature gradients in the single-stage module of p-type Pb_0.953Ge_0.007Na_0.040Te and n-type PbTe_0.9964I_0.0036 (figure 8) [154]. The module shows good durability in η_max over 24 h with the exception of at the largest temperature gradient (T_h = 873 K and T_c = 283 K). A slight decrease in η_max was observed after testing with largest temperature gradient of 590 K. This is due to a small increase in internal resistance, which probably was caused by the deterioration of the interfaces between the legs and electrodes and/or the degassing of Te as a result of the large temperature gradients applied over 24 h.

Figure 8.

A 24 h stability test of η_max as a function of time (hours) with T_h = 873, 773, 673 and 573 K and T_c = 283 K in single-stage nanostructured PbTe module (p-type: Pb_0.953Ge_0.007Na_0.040Te; n-type: PbTe_0.9964I_0.0036). Adapted from Jood [154]. (Online version in colour.)

We modelled the power generation characteristics of the nanostructured PbTe-based module using the commercial three-dimensional finite-element software (COMSOL Multiphysics with Heat Transfer Module) using a measured Seebeck coefficient, electrical resistivity and thermal conductivity of the materials. The simulated values of P_max, Q_oc and η_max for single-stage nanostructured PbTe module and segmented Bi₂Te₃/nanostructured PbTe module for T_h = 873 K and T_c = 283 K are listed in table 3 along with the experimentally measured values for comparison. The simulated η_max are 11.1% for single-stage module and 15.6% for cascaded module, which are approximately 30% and approximately 40% higher than the measured values. The reason for the lower measured values of η_max is the lower P_max and higher Q_oc values of the fabricated module. In both modules, the measured values of P_max (approx. 2.23 W for single stage and approx. 2.34 W for cascade type) were lower than the theoretically simulated ones (2.58 W for single stage and 2.55 W for cascade type). This difference arises from the parasitic electrical resistances at the material interfaces. The parasitic electrical resistances result in ohmic losses, leading to the lower power output. Further improvement of the electrical contact at material interfaces such as thermoelectric materials and interconnect electrodes will enable higher P_max and higher η_max. Moreover, in both modules, simulated values of Q_oc (approx. 20.6 W for single stage and approx. 10.8 W for cascade type) are lower than measured ones (approx. 24.0 W for single stage and approx. 15.8 W for cascade type). The deviation can be explained by radiative heat exchange from hot side to cold side in the module as a loss mechanism. The open area between the thermoelectric legs acts as a path for thermal radiation transfer. Further optimization of geometrical configuration of the module will reduce these radiation heat losses and enable lower Q_oc and higher η_max.

(c). Skutterudite-based module

Skutterudites have also demonstrated attractive performance in temperature range of 600–900 K. The fabrication process used for the segmented Bi₂Te₃/skutterudite (p-type: CeFe_3.85Mn_0.15Sb₁₂; n-type: Yb_0.3Co₄Sb₁₂) module is shown in figure 9a [162]. The p- and n-type samples were hot-pressed with diffusion barrier (Ti-Al) and contact layer (Ni). After polishing and cutting, interconnect electrodes (Mo-Cu) were joined at both hot and cold sides. The Ti-Al-based diffusion barrier yields a low specific electrical contact resistance of approximately 10 µΩ m² at the interface between Ti-Al and skutterudite. For the hot side, the electrode and Ni contact layer on top of the leg were joined by brazing with Cu-Ag-Zn. An Sn-based solder was used for joining Bi₂Te₃ and skutterudite. Glass fibres were filled into the gaps between legs to reduce thermal convection and radiation losses.

Figure 9.

(a) Fabrication process used for the segmented Bi₂Te₃/skutterudite module (p-type: CeFe_3.85Mn_0.15Sb₁₂; n-type: Yb_0.3Co₄Sb₁₂) and (b) the conversion efficiency (η) as functions of electrical current (I) [162]. Adapted from Zhang [162]. (Online version in colour.)

Salvador et al. have achieved approximately 7% conversion efficiency in a skutterudite-based module for T_h = 773 K and T_c = 313 K (table 2) [157,180]. In this module, Mm_0.30Fe_1.46Co_2.54Sb_12.05 (Mm stands for Misch metal which is an alloy of La, Ce, Pr and Nd) was used as a p-type material and Yb_0.09Ba_0.05La_0.05Co₄Sb₁₂ was used as an n-type material (figure 10a). Mo diffusion barriers were formed to prevent the Sb in the skutterudites from reacting with the braze and interconnect electrode for both the p- and n-type legs. Recently, η_max of a skutterudite-based module was improved through grain-boundary engineering. For example, η_max approximately 8.4% for T_h = 873 K and T_c = 296 K was measured in modules based on grain-boundary engineered skutterudite with graphene [79] and η_max approximately 9.3% for T_h = 872 K and T_c = 314 K was found in that of grain-boundary engineered skutterudite with carbon nanotube (table 2) [80].

Figure 10.

Skutterudite-based module and interface between skutterudites and diffusion barrier. (a) Single-stage module of p-₀Fe_1.46Co_2.54Sb_12.05 (Mm: Misch metal) and n-type Yb_0.09Ba_0.05La_0.05Co₄Sb₁₂ with Mo diffusion barrier [157,180]. (b) Single-stage module of p-type La_0.7Ba_0.1Ga_0.1Ti_0.1Fe₃CoSb₁₂ and n-type Yb_0.3Ca_0.1Al_0.1Ga_0.1In_0.1Co_3.75Fe_0.25Sb₁₂ with Co–Fe–Ni diffusion barrier [156,181]. (a) Adapted from Salvador [157,180], (b) Adapted from Guo [156]. (Online version in colour.)

Like the PbTe-module above, the η_max was dramatically enhanced by segmenting the skutterudite-based device with a Bi₂Te₃-based device on the cold side (figure 9b). This led to a η_max approximately 12% for T_h = 849 K and T_c = 363 K, figure 9b and table 2. In this module, CeFe_3.85Mn_0.15Sb₁₂, Yb_0.3Co₄Sb₁₂ and Ti-Al were used as p-type skutterudite, n-type skutterudite and diffusion barriers, respectively [162].

Long-term stability test has been reported on p-type Ce_yFeCo₃Sb₁₂ and n-type Yb_0.3Co₄Sb₁₂ modules [152]. The η_max of this module was approximately 7.2% for T_h = 843 K and T_c = 338 K. The power output of the module was steady for 840 h and more for T_h of 773 K and T_c of 298 K. The module also shows good stability in the power output for the heat cycle test over 2400 times under T_h of 373 K– 823 K and T_c of 298 K. Moreover, the 8000 h stability test under T_h = 873 K and T_c = 353 K was performed on a module of p-type La_0.7Ba_0.1Ga_0.1Ti_0.1Fe₃CoSb₁₂ and n-type Yb_0.3Ca_0.1Al_0.1Ga_0.1In_0.1Co_3.75Fe_0.25Sb₁₂ with a Co–Fe–Ni diffusion barrier (figure 10b) [156,181]. The η_max of this module was approximately 8% for T_h = 843 K and T_c = 338 K. A power output initially decreased by approximately 6% in 300 h then remained constant. This module also shows good stability in the power output for heat cycle test over 500 times under T_h of 473 K– 873 K and T_c of 313 K. These life tests exhibit promising durability of skutterudite-based modules for practical use.

(d). Half-Heusler-based module

As mentioned above in §2c, η_max of approximately 6.2% has been demonstrated experimentally in an HH-based module of p-type FeNbSb and n-type ZrNiSn-based alloys for T_h = 991 K and T_c = 336 K (figure 3d and table 2) [24]. As shown in figure 3d, η_max of 11.3% is estimated for T_h = 991 K and T_c = 336 K, assuming no electrical and thermal contact resistances in the module. The difference between measured and estimated values in η_max is due to insufficient electrical contact and the large heat radiation and convection. In particular, the contact resistance was estimated to contribute to about 3.2% efficiency loss.

Long-term stability of HH-based module of p-type Zr_0.5Hf_0.5CoSb_0.8Sn_0.2 and n-type Zr_0.4Hf_0.6NiSn_0.98Sb_0.02 was investigated through heat cycle and heat shock tests [158]. Power output of the module with substrate size of 16 mm × 16 mm reached approximately 2.8 W (power density: approx. 3.2 W cm⁻²) for T_h = 820 K and T_c = 293 K (table 2). The power output was unchanged under several heat cycle tests (such as 130 cycles of hot side temperature change from 523 K to 823 K) and six shocks of the module by increasing the hot side temperature from 373 K to 873 K with a heating rate of 100 K min⁻¹.

Recently, the ZT value in p-type HH materials was improved and corresponding efficiency was demonstrated in these p-type unilegs. Figure 11 shows ZT value in p-type HH materials and η_max [164,165]. High ZT values of approximately 1.42 and approximately 1.52 at 973 K were found in ZrCoBi₀Sb_0.15Sn_0.20 and Ta_0.74V_0.1Ti_0.16FeSb (figure 11a), respectively, leading to higher conversion efficiency in unilegs. For example, η_max values of approximately 9% and approximately 11.4% have been reported in unileg of ZrCoBi_0.65Sb_0.15Sn_0.20 for T_h = 823 K and T_c = 323 K and Ta_0.74V_0.1Ti_0.16FeSb for T_h = 973 K and T_c = 317 K, respectively (figure 11b and table 2). The results indicate the p-type TaFeSb-based HHs are promising for thermoelectric power generation. For real application, development high-performance n-type HH legs is required.

Figure 11.

(a) ZT value in p-type half-Heuslers and (b) conversion efficiency η_max for these unilegs [24,109,118,165,182]. The efficiency was estimated for Ta_0.74V_0.1Ti_0.16FeSb (Theoretical prediction). Adapted from Zhu [165]. (Online version in colour.)

(e). Other modules

Zintl compounds are an important class of thermoelectrics [183,184]. One example is Yb₁₄MnSb₁₁, which shows high ZT approximately 1.0 at high temperature (1223 K) [184]. The module based on Yb₁₄MnSb₁₁ had been developed for radioisotope thermoelectric generators in space missions [185]. The Yb₁₄MnSb₁₁ leg demonstrated good durability in thermoelectric properties over the six month test at 1273 K and 1323 K under vacuum. The sublimation of Yb₁₄MnSb₁₁ was suppressed through formation sublimation barrier made of porous alumina layer. Stable contact with low resistance was achieved between the Mo and Yb₁₄MnSb₁₁.

As mentioned above (§2b), SnSe has recently emerged as a new class of thermoelectric material. The fabrication of a module based on polycrystalline SnSe has just begun. Ag/Ni bilayer and Ag/Co/Ti multi-layer were developed as diffusion barriers [93,186]. For example, no secondary phases and no cracks were observed between the Ag/Co/Ti multi-layer and the SnSe interface [186]. Moreover, the specific contact resistance between the Ag/Co/Ti multi-layer and the SnSe was observed to be approximately 1.53 mΩ cm² at room temperature. However, after heat treatment at 723 K for 20 h under Ar atmosphere the specific contact resistance increased to approximately 2.27 mΩ cm² at room temperature because of diffusion of Sn and Se into Ag layer. It is necessary to develop stable diffusion barriers for the fabrication of SnSe module.

4. Summary and future prospects

There has been tremendous progress in the science of thermoelectric materials, but more progress needs to be made when it comes to transitioning from materials to devices. This is crucial for commercialization and practical use. On the material side, many outstanding ZTs > 1.5 or even over 2 have been reported; however, one should keep in mind that the conversion efficiency is more related to a material's average ZT over the whole operating temperature. Moreover, in addition to thermoelectric performance, we also need to look into a material's mechanical strength and thermal stability, which are crucial in real applications but have been long ignored in most studies. Going forward, researchers should put more effort into fabricating novel thermoelectric devices with high efficiency from the new materials, including the design of new device structures and evaluation protocols of device performance/stability.

Data accessibility

This article has no additional data.

Competing interests

We declare we have no competing interests.

Funding

G.T. would like to acknowledge the recent financial support from the Natural Science Foundation of China (grant no. 11804261) and ‘the Fundamental Research Funds for the Central Universities (WUT: 2019IVA068 and 2019IVB049)’. The work at AIST was partially supported as part of the Development of Thermal Management Materials and Technology funded by the New Energy and Industrial Technology Development Organization (NEDO) and the International Joint Research Program for Innovative Energy Technology funded by the Ministry of Economy, Trade and Industry (METI). At Northwestern University, the work was supported primarily by the Department of Energy, Office of Science, Basic Energy Sciences under grant no. DE-SC0014520 (M.G.K.)