Xiaojing Houab,
Shaoping Chen*a,
Zhengliang Dub,
Xianglian Liub and
Jiaolin Cui*b
aMaterials Science and Engineering College, Taiyuan University of Technology, Taiyuan 030024, China. E-mail: sxchenshaoping@163.com
bSchool of Materials, Ningbo University of Technology, Ningbo 315016, China. E-mail: cuijiaolin@163.com
First published on 26th November 2015
Here we present InSe-based alloys InSeSnx (x = 0–0.02) with improved thermoelectric performance upon Sn's preferential occupation on In lattice sites. This improvement is attributed to the enhancement in carrier concentration (n) and reduction in lattice thermal conductivity (κL). However, the enhancement in n is limited due to the presence of the intermediate band in the middle of the bandgap, which acts as an annihilation center for electrons and holes. The reduction in κL is caused by increased phonon scattering on the newly-created defect SnIn+. As a result, we attain the highest ZT value of 0.23 at x = 0.01@830 K, which is about 2.9 times that of virgin InSe.
Although many approaches to enhancing ZT have been developed, including the enhancement of Seebeck coefficients (by modifying the band structure,1 band convergence,2 quantum confinement effects3 and electron energy barrier filtering4), and reduction of lattice thermal conductivity (by nanostructuring5 and all-scale hierarchical architecturing6), there still remains necessity to explore new materials, especially wide gap semiconductors, which are potentially competent TE candidates.
In recent years indium selenides have been paid increasing attention due to their intrinsic structural characteristics, including phases, crystal structures and structural imperfection.7 Among them In4Se3 – (ZT = 1.48@705 K,8a and 1.11@723 K8b) and In2Se3-based (ZT = 1.23@916 K)9 alloys have already presented their potential TE performance. InSe, which is one of indium selenides, crystallizes in a layered structure where each layer contains sublayers closely packed with two In and two Se, in a stacking sequence Se–In–In–Se, whereas the bonding between two adjacent layers is of the weak Van der Waals type. Therefore, the materials of InSe family usually present anisotropic characteristics. Although there are some defects acting as donors in not purposely-doped InSe, such as interstitial In atom (Ini)10 and Se vacancy (Vse),11 the virgin n-type InSe crystal gives much lower carrier concentration (n) at room temperature (RT) (1014 to 1015 cm−3)10,12 than good TE candidates do, which usually have n values as high as 1019 to 1020 cm−3.13 Hence it is strongly necessary to enhance the carrier concentration in order to improve the TE performance of InSe based alloys.
Some investigations10,11 suggest that element Sn is the most suitable donor impurity in InSe as it permits to reduce the resistivity of the material without strongly affecting the electron mobility (μ) through forming shallow donor levels, and some indicate that only a small part of Sn gets into the lattice while the rest remains as interlayer precipitate planes forming large planar defects.11 However, Gridin etc.14 propose that there might exist some deep impurity levels in InSe upon Sn(Zn)-doping. Such deep impurity levels could provide annihilation centres for electrons and holes, and thereby reduce the carrier concentration. Besides, the impurities from the same column (such as Sn and Pb) in InSe act quite differently, the first as a donor and the second as an acceptor.11 Therefore, the action of Sn in InSe is still unclear and needs to be probed. In order to improve the TE performance of InSe-based materials, we should figure out the conducting mechanism of both carrier and phonons when impurity, such as Sn, acts.
In this work, we have measured carrier concentrations and calculated band structures upon proper incorporations of Sn into different lattice sites in InSe, through which we have been able to determine the preferential occupation of Sn in the InSe lattice. The incorporation of Sn, which creates an active donor defect SnIn+, increases carrier concentration (n) and phonon scattering, and thereby improves TE performance drastically.
The absorption coefficient measurements were carried out using a Perkin-Elmer Lambda 950 UV-VIS-NIR spectrophotometer, and absorption spectra for the powders were recorded between the visible and infrared regions (200–1500 nm).15
When calculating the band structures, PW91/GGA functional, implemented in the VASP software package, was used, and the 5s25p2, 5s25p1 and 4s24p4 were each treated as valence states of Sn, In and Se. The plane-wave basis set cutoff is 450 eV and a k mesh of 7 × 7 × 2. The Monkhost–Pack mesh 49 sampling with 1 × 1 × 21 k-points in a string Brillouin zone (x, y, z directions, respectively) is employed. Basically, there are three arrangements of Sn occupation, which are on the interstitial In (Ini), In (SnIn) and Se (SnSe) lattice sites, each corresponding to the chemical formulas of In16Se16Sn, In15Se16Sn and In16Se15Sn (equivalent to the Sn content x = ∼0.033).
The Hall coefficient (Rh) measurements at RT, were performed on rectangular samples in size of 2 × 2 × 7 mm3, were conducted on a Physical Property Measurement System (PPMS, Model-9) using a four probe configuration with a magnetic field sweeping between ±2.0 T. The Hall mobility (μ) and carrier concentrations (n) were subsequently calculated based on the relations μ = |Rh|σ and n = 1/(Rhe) respectively, where e is the electron charge. The current and Hall voltage leads were fine copper wires, and the contacts were made of silver paste.
Fig. 1 (a) The X-ray diffraction patterns of InSeSnx (x = 0, 0.005, 0.01, 0.02) powders, (b) lattice constants (a, c, V) as a function of x value. |
The absorption coefficient spectra A(hν) are shown in Fig. 2(a), where A is the absorption coefficient and hν photon energy. In the spectra A(hν) we have observed two absorption thresholds (one is around 1.24 eV and the other on high energy side 1.50 eV), instead of three edges observed by Balkanski.18 The presence of two edges, which is likely originated from the 3d orbital splitting of the Se atoms,19 results in two sets of bandgap values (Eg). The E′g value, which fits the first edge (∼1.50 eV), reduces slightly from 1.01 eV to 0.95 eV (Fig. 2(b)); while the second E′′g value, which fits the second one (∼1.24 eV), tends to reduce from 1.10 eV to 1.03 eV as x value increases (Fig. 2(c)). Although both the bandgaps bears limited relevance to the composition, the measured Eg value of pure InSe (1.01–1.10 eV) is about ∼0.30 eV lower than those reported (∼1.35 eV (ref. 18a) and ∼1.20 eV for Sn-containing InSe,18b where the Fermi level is assumed to lie in the top of the valence band maximum).
In order to have a thorough understanding of the incorporation of Sn in InSe lattice, we have analyzed the oxidation states of In, Sn and Se using In 3d5/2, Sn 3d5/2 and Se 3d XPS spectra in InSeSnx (x = 0, 0.005, 0.01, 0.02), as shown in Fig. S2.† The average binding energy (BE) values with uncertainties estimated at ∼±0.01 eV are listed in Table S2,† in which we have observed that the BE values of In 3d (444.44–444.73 eV) (Fig. S2(a)†) are roughly equal to those of In3+ in CuInSe2 (444.70 eV)20 and 444.9 eV in In(OH)3.21 Therefore, the obtained spectra can be assigned to In3+ cation. Each of the Se 3d spectra is clearly split into a resolved doublet (3d5/2 and 3d3/2 levels), with the BE values of Se 3d5/2 around 53.85 eV and those of Se 3d3/2 54.52 eV, as shown in Fig. S2(b).† These values are in good agreement with those of anionic Se in Zn1−xMnxCr2Se4 (ref. 22) and Se2− (GeSe: ∼54.5 eV; Bi2Se3: ∼54.2 eV),23 which confirms the existence of anionic Se2−. Besides, there are no visible shifts of the doublet peak as Sn content increases. The BE values of Sn 3d, which are determined by XPS peaks, are roughly equal to ∼486.0 eV, close to that of Sn4+ (SnO2: 486.0–486.3 eV),24 though these peaks are unconspicuous and a little difficult to identify due to limited Sn content in InSe, shown in Fig. S2(c).†
When Sn4+ exists in the InSe crystal, there are basically three ways of arrangements in InSe lattice sites, that is, on the interstitial In (Ini), In and Se lattice sites. To determine the exact occupation site of Sn in InSe lattices, we have calculated the band structures upon Sn occupations in different sites. The results are shown in Fig. 3. The band structure of virgin InSe is calculated for comparison (Fig. 3(a)), in which case a direct band gap is observed with Eg = ∼0.51 eV and the Fermi level (Fr) is just located at the top of the valence band maximum (VBM). When Sn occupies interstitial In sites, the bandgap narrows by lifting both the Fr and VBM, and lets Eg = ∼0 eV, (see Fig. 3(b)), which indicates semimetal behaviour of the material. When Sn presumably occupies In lattice sites, the Fr is lifted to the conduction band with the bandgap Eg = ∼0.88 eV (Fig. 3(c)). Besides, there appears an intermediate band in the middle of the bandgap, which is approximately 0.34 eV above the valence band edge. This deep level can be due to an isolated substitutional donor, which undergoes a shallow-to-deep transition under certain conditions of hydrostatic pressure.25 The formation of the intermediate band might result from the creation of the In–Sn pockets via the incorporation of Sn in crystal lattice.14a However, this impurity level could act as an annihilation centre for electrons and holes and it does no help to the carrier concentration enhancement. Fig. 3(d) presents the band structure with Sn occupying Se lattice sites, where we have observed the lowering of the Fr towards the valence band, indicating p-type conducting behaviour. However, the VBM approaches to conducting band minimum (CBM), which leads to Eg = ∼0 eV.
Samples | x | Rh (m3 C−1) | Carrier concentration, n [m−3] | Mobility, μ [m2 V−1 s−1] | σ [Ω−1 m−1] | Remarks |
---|---|---|---|---|---|---|
InSeSnx (C⊥) | 0 | −2.43 × 10−2 | 2.57 × 1020 | 1.68 × 10−3 | 0.06 | |
0.005 | −6.84 × 10−3 | 9.14 × 1020 | 1.43 × 10−2 | 2.09 | ||
0.01 | −9.82 × 10−3 | 6.36 × 1020 | 1.06 × 10−2 | 1.02 | ||
InSeSnx (C∥) | 0 | −5.45 × 10−2 | 1.15 × 1020 | 3.06 × 10−3 | 0.06 | |
0.005 | −8.64 × 10−3 | 7.23 × 1020 | 4.94 × 10−2 | 5.71 | ||
0.01 | −8.95 × 10−3 | 6.98 × 1020 | 1.38 × 10−3 | 0.15 | ||
Sn-doped InSe | 0.70 × 1022 to 5.70 × 1022 at RT | 5.62 × 10−2 to 11.0 × 10−2 at RT | Ref. 10 and 11 |
Fig. 4 is the plot of TE performance (C⊥) perpendicular to the pressing direction, where we observe that the absolute Seebeck coefficients (|α|), all of which decrease with the measuring temperature, bear little relevance to the chemical composition above 550 K. At ∼815 K the α value is ∼−440.0 (μV K−1), as shown in Fig. 4(a). The electrical conductivity (σ) increases slightly from 420.0 (Ω−1 m−1) at x = 0 to 511.0 (Ω−1 m−1) at x = 0.005 at 815 K, as shown in Fig. 4(b). Although the incorporation of Sn has a limited influence on the electrical property, it exerts a profound impact on the lattice thermal conductivity (κL), as shown in Fig. 4(c). First, we have observed the bipolar effect at high temperatures, as the κL value for the virgin InSe shows an increasing tendency with the increase of temperature above 600 K. Second, the κL value reduces drastically from 1.09 W K−1 m−1 at x = 0 to ∼0.44 W K−1 m−1 at x = 0.005–0.01@∼815 K. However, the κL value of the sample at x = 0.02, which is much higher, decreases from 4.72 W K−1 m−1 at RT to 2.38 W K−1 m−1 at ∼815 K. In addition, the lattice contributions takes more than ∼90% in total thermal conductivities, as shown in Fig. 4(c) and S3.†
Combining the three physical parameters (α, σ and κ) we attain the dimensionless figure of merits (ZT), which is shown in Fig. 4(d). The sample at x = 0.005 gives the highest ZT value (∼0.20) at 815 K, while the virgin InSe gives only 0.06. The highest ZT value for InSe based alloys is much lower than those of other In–Se based families, such as Zn-doped α-In2Se3 (ZT = 1.23@916 K)9 and In4Se3-based alloys (1.48@705 K,8a 1.40@733 K,26a 1.53@698 K (ref. 26b)), and also lower than that for one of the In1.3−xSnxSe alloys (x = 0.05) (ZT = 0.66@700 K).27 However, In1.3−xSnxSe are actually more In4Se3-than InSe-based alloys, since the chemical compositions of In1.3Se is closer to In4Se3 than to InSe, even though InSe and In4Se3 bear similar structures.
Similarly, the temperature dependences of the Seebeck coefficients (α) in C∥ are almost identical to those in C⊥ (Fig. 5(a)), and the electrical conductivities (σ) are only about 120.0 (Ω−1 m−1) lower for the corresponding Sn content (Fig. 5(b)) at 830 K. Furthermore, above 620 K the samples at x = 0.005 and 0.01 give lower lattice contributions κL than Sn-free InSe, whereas the sample at x = 0.02 gives much higher κL values at high temperatures, and at ∼830 K it gives 1.55 W K−1 m−1, which is about three times that of Sn-free InSe (Fig. 5(c)). Likewise, the lattice contributions (κL) in C∥ play a dominant role in carrying heat, as shown in (Fig. S4†). The highest ZT value in C∥ is 0.23 at x = 0.01@∼830 K (Fig. 5(d)), a little higher than that in C⊥.
The improvement in TE performance is closely related to the Sn incorporation in InSe. When Sn is assumed to occupy In lattice sites, we have observed a lift of Fr to the conduction band (Fig. 3(c)) through the band structure calculation, which indicates that the electrons can occupy those extended states that give rise to the degenerate behaviour of the electron concentration. However, the measurement reveals that the unexpected enhancement in the carrier concentration does not agree to our assumption (Table 1) because the highest n value is still in the order of magnitude ∼1021 m−3. Despite that the estimated bandgaps, which are usually lower than those from measurements because of Local Density Approximation (LDA)/GGA problem, all tend to narrow no matter where Sn occupies In or Se lattice sites. Nevertheless, the Eg values will vary widely depending on the occupation site of Sn (see Fig. 3). If we compare the Eg values from calculations with those from measurements, and also take the measured n values into account, we believe that Sn prefers the In to Se lattice sites. It is not only because the bandgap (0.88 eV) through calculation is closer to the measured one (0.95–1.03 eV), but we have also observed the formation of an intermediate band acting as annihilation centres for electrons and holes, which is exactly what leads to a limited enhancement in n. Moreover, Sn is least likely to occupy the interstitial In sites (Ini), because such an incorporation tends to yield semimetal behaviour of the materials.
The preferential occupation of Sn on the In sites is in agreement with the proposed configuration by Segura,11 who suggested that in the Sn-doped InSe the suitable configuration is one couple of Sn atoms substitute a couple of neighbour In atoms. Based on this configuration, an active donor defect SnIn+ forms and thereby distorts the lattice structure.
Although the lattice distortion increases carrier scattering factor (γ) and thereby exerts a positive effect on the Seebeck coefficients, such an effect is neutralized by a negative one caused by the enhancement in carrier concentration (n), which is why we have observed a subtle change in the α value over the measuring temperature range. On the other hand, the enhancement in carrier concentration (n) is limited because of the formation of the intermediate band. Therefore, we have only observed a subtle/minimal improvement in electrical conductivity. Besides, both the n and μ values reach the highest at the same Sn content (x = 0.005), which might be due to the fact that Sn occupation in the In sites facilitates the cationic interdiffusion, which promotes the transport of carriers.28 However, this argument needs clarification after further investigations. Also, the newly-created defect SnIn+ acts not only as scattering centers for carriers, but also as scattering centers for phonons, which thereby reduces the lattice contribution κL when Sn content increases to x = 0.005–0.01. However, in the highly Sn-doped InSe the electron concentration could reach a saturation at a certain value.10a This implies that there might exist other substitutional configurations of Sn in InSe which help form acceptor centers in the highly-doped samples, that is, a substitutional Sn atom on a Se site10a that creates species SnSe2−. This assumption is supported by the band structure calculation, which reveals that the Fermi level (Fr) is lowering to the valence band upon Sn occupation Se sites, as shown in Fig. 3(d). In this case, in the stacking sequence Se–In–In–Se in the layered InSe structure, there exist interactive donor–acceptor defect pairs (2SnIn+ and SnSe2−) (DADPs). The reaction between cations and anions in DADPs can lead to their partial annihilation, thus resulting in the reduction of the defect concentration and the number of species (SnIn+ and SnSe2−).29 Since the neutralization between the cationic Sn+ and the neighbor anionic Sn2− leads to the short range ordering tendency of the structure, therefore, the phonon scattering weakens and the lattice part κL increases drastically in the highly Sn-doped sample (x = 0.02).
Footnote |
† Electronic supplementary information (ESI) available. See DOI: 10.1039/c5ra23023c |
This journal is © The Royal Society of Chemistry 2015 |