Controlling the Nanostructure of Bismuth Telluride by Selective Chemical Vapour Deposition from a Single Source Precursor

Substrate preparation: Sputtered SiO2 and TiN substrates and photolithographically patterned TiN/SiO2 substrates were prepared as described previously.14 The E-beam lithography was carried out using a JEOL JBX-9300FS Electron beam Lithography System with a double-layered positive resist PMMA495. During exposure, the gun current was set to 4 nA with an accelerating voltage of 100 kV. The etching was performed by a RIE80+ with CHF3 and Ar and the etch rate was found to be 0.37 nm s-1.

Thermoelectric (TE) materials are widely regarded to have the potential to revolutionise electrical power generation and cooling by dramatically reducing the inefficiency of current methods and thus reducing global dependency on fossil fuels. Current coal, natural gas, oil, and nuclear power generation processes are typically only $40% efficient and it is estimated that if a further 1% of their primary energy could be recovered, $190 TW h of electricity would be generated annually within the EU, with a market value of >V19bn and signicant associated reductions in CO 2 emissions. 1 However, improving thermoelectric efficiency of current materials is a key barrier to wider adoption of this emerging technology.
Bismuth telluride (Bi 2 Te 3 ), a layered semiconductor with a narrow band gap of 0.16 eV, 2 and its alloys are commonly used in commercial bulk thermoelectric (TE) devices as they have among the best room temperature thermoelectric properties of known bulk materials. 3 While solid state TE technology has the potential to deliver sustainable and highly durable energy generation, it is currently used only in niche applications. This is due to the low efficiency ($12% for commercially available devices) and relatively high cost of manufacturing of the current generation of TE materials. However, it has been demonstrated that nanostructuring of TE materials can lead to signicant increases in efficiency, due to both quantum connement effects and reductions in lattice thermal conductivity, i.e. decoupling between the electron scattering (electrical conductivity) and phonon scattering (thermal conductivity). 4 Thus modern TE theory predicts that the efficiency of a TE device can be increased by a factor of ca. 3 if the diameter can be decreased in size to that at which quantum connement and interface scattering effects occur. 5 It has also been established that preferred orientation of the nanocrystalline Bi 2 Te 3 such that heat ow is in the h1 1 0i plane maximises its TE properties. 6 Additionally, there is considerable interest in methods to deposit individual single crystals of Bi 2 E 3 (E ¼ Te, Se) with specic orientations as topological insulators, containing protected surface quantum conduction states. 7 Such materials have potential applications in quantum computing. Importantly, connement of these crystals on the nanoscale gives the best surface to volume ratio and optimisation of this effect. 8 For both of these applications there is therefore signicant motivation to develop methods capable of selectively depositing individual, high quality, nanocrystalline materials such as Bi 2 Te 3 onto surfaces in predetermined positions and with a high degree of orientational control.
Electrodeposition has been the subject of intense research activity as it is currently the only potential method of achieving such control over Bi 2 Te 3 deposition. 9 Chemical vapour deposition (CVD) is an attractive alternative process and is widely used in the industrial manufacture of semiconductor thin lms, due to its simple and scalable nature. 10 Bi 2 Te 3 thin lms have previously been deposited on large surfaces by dual source CVD using trialkylbismuth and dialkyltelluride gases as precursors. 11,12 The use of molecular, single source precursors for CVD can give advantages in the control of lm stoichiometry and morphology, as well as ease of handling. 13 We recently reported the rst use of a metal telluroether precursor in low pressure (LP) CVD of Ga 2 Te 3 , 14 and have also demonstrated the growth of 2D micron-scale arrays of metal selenide semiconductor materials by LPCVD from selenoether complexes, exploiting substrate selectivity to achieve area selective deposition. 15, 16 Huang and co-workers have also identied a strong substrate inuence in the vapour deposition of layered chalcogenide nanoplates. 17 However, no single source precursors for the CVD of Bi 2 Te 3 are currently known, although deposition of Sb 2 Te 3 nanoplates has been demonstrated from [Sb{(Te P i Pr 2 ) 2 N} 3 ]. 18 Here we report the deposition of individual nanocrystals of Bi 2 Te 3 via a one-step LPCVD method that allows positional control on the nanometre scale. By exploiting the selectivity of the deposition onto TiN surfaces over SiO 2 , arrays of microcrystalline Bi 2 Te 3 thin lms were deposited into TiN wells in lithographically patterned SiO 2 /TiN substrates, with no deposition on the surrounding SiO 2 . In larger (>500 nm) diameter wells, lms were generally polycrystalline, with a high degree of h0 0 1i preferred orientation of the hexagonal crystallites (i.e. the c-axis aligns perpendicular to the TiN surface). In contrast, for wells of 100 to 500 nm diameter, individual nanocrystals are produced; SEM shows that these have the opposite orientation, with each crystallite contacting through a crystal edge to the TiN surface with the c-axis parallel to the substrate surface, hence exhibiting h1 1 0i preferred orientation (Fig. 1).
The high quality of the Bi 2 Te 3 deposited by LPCVD from the new reagent was established from thin lms grown onto larger SiO 2 and TiN surfaces under a range of conditions. Both the thin lms and the arrays of Bi 2 Te 3 were characterised by X-ray diffraction (XRD), scanning electron microscopy (SEM), energy dispersive X-ray spectroscopy (EDX), Raman spectroscopy and wavelength dispersive X-ray spectroscopy (WDX) to establish composition, purity, crystal structure and morphology. Hall effect and Seebeck coefficient measurements were also performed to evaluate the thermoelectric performance.
The new molecular precursor [BiCl 3 (Te n Bu 2 ) 3 ], a very rare example of a bismuth telluroether complex, was prepared and characterised as described. ‡ It is a dark red oil which is highly moisture sensitive and mildly thermally sensitive, though it is stable for long periods stored under N 2 at À18 C. Thermogravimetric analysis (TGA) (Fig. S1, ESI †) was undertaken as a guide to the LPCVD conditions. SEM images of Bi 2 Te 3 lms grown by LPCVD onto SiO 2 or TiN from ca. 50 mg of [BiCl 3 (Te n Bu 2 ) 3 ] at 500 C, 0.05 mm Hg, show continuous lms of pseudo-hexagonal platelets, most of which lie at on the surface of the substrate (Fig. S2, ESI †).
Hall effect measurements conducted on polycrystalline lms deposited onto insulating SiO 2 substrates show that for lms of thickness ca. 1.0 mm, the resistivity was (5.65 AE 0.02) Â 10 À4 U$cm. The Bi 2 Te 3 is n-type with a carrier concentration of 1.95 Â 10 20 cm À3 , and a mobility of 56.6 cm 2 V À1 s À1 . Seebeck effect measurements were performed on the same Bi 2 Te 3 sample in order to evaluate the potential performance for TE applications. The mean Seebeck coefficient was À109 mV K À1 , consistent with lms of n-type conductivity. These data are comparable with values reported previously for thin lms of Bi 2 Te 3 grown by other thin lm technologies, such as MOCVD, 12 molecular beam epitaxy (MBE), 19 co-sputtering 20 and electrodeposition. 21 Importantly, their transport properties, carrier concentrations and Hall mobility values are also close to the range of optimum values that are required for thermoelectric applications. 22 Micro-and nano-patterning and selectivity A similar LPCVD method was employed using patterned SiO 2 / TiN substrates, which have SiO 2 surfaces containing arrays of photolithographically etched wells (1 mm deep, 1-100 mm diameter) giving access to a TiN surface exposed at their base. Under carefully controlled conditions Bi 2 Te 3 was deposited into these TiN wells with excellent substrate selectivity, the wells being lled with crystals, but with no deposition being observed on the surrounding SiO 2 capping layer (Fig. 2). An important parameter is the quantity of precursor employed: an excess caused overlling of the holes and was accompanied by some deposition onto the SiO 2 layer, whereas too little precursor resulted in incomplete lling. However, careful control of the quantity of reagent allows a continuous lm to be deposited within each well. Lowering the temperature of deposition from 500 to 450 C reduced the crystallite size although, as expected, the time required for complete deposition was increased. Lowering the temperature further (to 400 C) resulted in very little material being deposited.
To investigate the applicability of this technique toward nanostructuring of Bi 2 Te 3 materials, it was necessary to decrease the dimensions of the wells further. e-Beam lithography was used to introduce nanowells between 100 and 500 nm in diameter on the patterned TiN/SiO 2 substrate; the thickness of the SiO 2 layer was reduced to 200 nm to maintain a reasonable aspect ratio. LPCVD onto these substrates at 450 C resulted in selective growth of a single nanocrystal into each of the smaller wells. The crystals lie at in the base of the wells with diameter of $1 mm (Fig. 3a), whereas in those of 100-200 nm diameter the individual crystallite size is larger than the wells in which they sit (generally 200-500 nm across and less than 100 nm thick). SEM images (Fig. 1) show that almost all of the crystals in these smallest wells stand on end, apparently contacting through a crystal edge to the TiN surface within the well, but occupying a larger footprint above the substrate. The orientation of crystallites in #200 nm wells is ca. 90 to those in the $1 mm wells. In the intermediate 500 nm wells both behaviours were observed (Fig. 3b and c). This can be correlated with the size of the nanocrystal relative to the diameter of the well, suggesting that the change to h1 1 0i preferred orientation reduces the less favourable interactions of the larger Bi 2 Te 3 crystals with the SiO 2 wallssee Fig. 3b.
Excellent selectivity was maintained even on the nanoscale, with very few crystals observed outside of the wells, although around 20-25% of the nanowells appear to be empty. This switching of the preferred orientation simply by reducing the dimensions of the recessed TiN regions is extremely unusual, and, coupled with the demonstrated ability to selectively grow crystalline Bi 2 Te 3 in predetermined areas, offers exciting prospects for increasing the TE efficiency of Bi 2 Te 3 using this new precursor and LPCVD method.   XRD Symmetric (q-2q) XRD data were collected for lms deposited onto SiO 2 and TiN at 500 C and into 40-100 mm TiN wells on a patterned substrate at 450 C. The XRD patterns identify trigonal (R 3m) Bi 2 Te 3 as the only phase present. The rened lattice parameters are in good agreement with literature data 23 and are presented in Table S1. † However, relative peak intensities differ signicantly compared to literature values for bulk Bi 2 Te 3 , indicating strongly preferred c-axis orientation of the crystallites. The identication of this preferred orientation is consistent with the SEM images. The degree of preferred orientation varies between lms deposited onto different substrates (Fig. 4), with the highest degree of orientation observed from microfocus XRD from individual lled wells in the micro-scale arrays (ESI †). These XRD patterns display only 0 0 l reections (Fig. 4a). Pole gure measurements were made for two reections on the same sample, over a larger region of lled wells. The pole gure taken with 2q ¼ 49.80 , corresponding to the 0 0 15 reection, exhibits a single, very sharp peak (FWHM $ 1 ) at the centre of the gure, with a ¼ 90 (Fig. 5a). The gure taken with 2q ¼ 27.67 , corresponding to the 0 1 5 reection, exhibits a narrow ring with a ¼ 32 (Fig. 5b). These results are consistent with a highly preferred crystallite orientation with the substrate perpendicular to the c-axis (calculated values of a in this case are 90.0 for any 0 0 l reection and 32.1 for the 0 1 5see Eqn (S1) †). It is probable that the high degree of orientation observed here is not inherent to the patterning, but rather that the attest crystallite orientation is observed in a thin but continuous lm, which are also the conditions that yield the highest degree of selectivity, as discussed above.
EDX spectroscopy of the lms and micro-scale arrays of Bi 2 Te 3 gave a consistent Bi : Te ratio of 2 : 3 (40.31% Bi, 59.69% Te), measured quantitatively against a reference sample of Bi 2 Te 3 (99.99%, Strem Chemicals) (Fig. 6). In the case of the single nanocrystals, EDX measurements also identied Bi and Te, but the low signal relative to that from the substrate precluded accurate determination of the Bi : Te ratio. Raman spectra on both the at lms and the nanocrystals identied them as Bi 2 Te 3 (ref. 24) (Fig. 7 and S3 †).
WDX analysis on the thin lms allowed better peak resolution, and conrmed the absence of any peaks in the region of Cl Ka (2.621 keV) or O Ka (0.525 keV) (Fig. S4a †). A small peak is observed at 0.28 keV which is likely to have contributions from C Ka (0.277 keV) and the weaker Bi N6-N5 peak (0.284 keV), however these peaks could not be resolved by WDX (Fig. S4b †). In order to determine the C content in these lms, combustion analysis was undertaken on samples of Bi 2 Te 3 that were removed from the substrate by scraping. The C atom content was below 0.5%.

Conclusions
We have demonstrated that regular arrays of high quality single Bi 2 Te 3 nanocrystals can be deposited by LPCVD from a single source bismuth chloride telluroether complex with a very high   degree of positional control, and that the preferred orientation of the nanocrystals is strongly governed by the dimensions of the underlying micro-or nano-patterned substrate. The ability to achieve such a high level of control over topology for Bi 2 Te 3 offers exciting prospects for developing this system for thermoelectric applications, and towards topological insulators. We expect that this approach could be extended to the synthesis of nanocrystalline arrays of other key semiconductor materials by the judicious choice of molecular precursors.