Interdigitated back-contacted crystalline silicon solar cells with low-temperature dopant-free selective contacts

In the field of crystalline silicon solar cells, great efforts are being devoted to the development of selective contacts in search of a fully low-temperature and dopant-free fabrication process compatible with high photovoltaic conversion efficiencies. For high-efficiency devices, selective contacts have to simultaneously combine high conductivity with excellent passivating properties. With this objective, a thin passivating extra layer of a-Si:H or SiO2 is usually introduced between the conducting layer and the silicon substrate. In this work, we present an interdigitated back-contacted (IBC) silicon based solar cell that avoids the use of either thermal SiO2 or a-Si:H interlayers achieving a dopant-free, ITO-free and very low thermal budget fabrication process. In this work, we propose a new electron transport layer using ultrathin Al2O3/TiO2 stacks deposited by atomic layer deposition at 100 C covered with a thermally evaporated Mg capping film. A specific contact resistance of 2.5 mU cm has been measured together with surface recombination velocities below 40 cm s . This electron-selective contact is combined with a thermally evaporated V2Ox-based hole selective contact to form the rear scheme of an IBC structure with a 3 3 cm active area as a proof-of-concept resulting in efficiencies beyond 19%. This approach sheds light on potential technological simplification and cost reduction in crystalline silicon solar cells.


Introduction
Highly selective contacts to one type of carrier, i.e. passivated contacts, are necessary for high-efficiency solar cells. The function of such contacts is to block totally (ideally) or partially (in practice) one type of carrier, electrons or holes, while the other one is collected. 1 Nowadays, the photovoltaic industry is mainly dominated by crystalline silicon (c-Si) solar cells, in which contact selectivity is usually achieved by doping the wafer surfaces with phosphorus (n + ) and boron (p + ) by means of high temperature oven-based diffusions, increasing either electron or hole conductivity. In this way, p + nn + and n + pp + doping schemes are a common way of achieving contact selectivity with n or p-type substrates, respectively. However, doping needs to be performed at high temperatures, requiring complex and energy consuming processes and lengthy cleaning protocols to avoid possible degradations of bulk lifetime, increasing the number of steps involved in the fabrication process and consequently the cost.
In order to replace those high temperature diffusions, several approaches have been studied. The well-known silicon heterojunction (SHJ) structure using doped and intrinsic hydrogenated amorphous silicon (a-Si:H) lms is probably the best known and combined with an interdigitated back-contacted (IBC) cell architecture holds the world efficiency record on c-Si substrates. 2 Another low-temperature technology proposed is laser-doping where the p + and n + regions are formed by laser-processing of dielectric lms. 3,4 Nevertheless, both options use toxic and ammable gases as dopant precursors, which is not desirable from the point of view of process complexity.
A possible alternative arises from the heritage of thin lm and organic photovoltaics, where novel materials are routinely used as semipermeable membranes to selectively extract carriers from a semiconductor light absorber. 5 In this way, alternative materials that overcome conventional limitations and difficulties have been explored using the so-called DASH approach (Dopant Free Asymmetric Heterocontacts); 6 materials such as organic polymers (PEDOT:PSS or P3HT), 7,8 alkaline salts (LiF x , MgF x or Cs 2 CO 3 ), 6,9,10 transition metal oxides (TMOs) 11 and multilayer TMO-metal-TMO 12 have been successfully applied in c-Si solar cells as hole and electron transport layers (HTL and ETL respectively). This new paradigm broadens the technological possibilities and multiple options are being actively explored to achieve the best trade-off between efficiency and simplicity. 13 In this context TMOs are becoming very attractive because of the combination of wide bandgaps (>3 eV) and consequently high transparency with low temperature deposition techniques, e.g. thermal evaporation, spin-on, spray-on and sputtering. High work-function TMOs ($7 eV) such as molybdenum oxide (MoO x ), tungsten oxide (WO x ) and vanadium oxide (VO x ) are preferred as hole-selective contacts, 14 while low work-function compounds ($4 eV) such as magnesium oxide (MgO x ), 15 lithium uoride (LiF x ) 6 and titanium oxide (TiO x ) 16 are being investigated as electron-selective contacts.
Asymmetric conductivity in these materials is achieved either by band alignment or through a strong induced bandbending. In order to take advantage of the band structure of the silicon/contact interface while simultaneously decreasing surface recombination, it is important to avoid Fermi-level pinning by reducing the density of states at the silicon interface. 17 In most cases, this is achieved by plasma enhanced chemical vapour deposition (PECVD) growth of a thin lm of intrinsic a-Si:H between the TMO and the c-Si surface. 6 Two drawbacks overshadow this approach, the use of ammable gases and the intrinsic difficulty of obtaining good Si/a-Si:H interfaces. Another possibility is to take advantage of the in situ redox reaction during thermal evaporation of the TMO material which may result in spontaneous growth of a SiO x lm. 18,19 Finally, some authors, 20 although introducing a high temperature step, have grown a thin tunnel SiO 2 layer between the TMO and the silicon surface.
Regarding electron-selective collector materials, TiO 2 -based contacts are an interesting alternative. Titania was rstly applied in c-Si photovoltaics as an antireection coating, 21 being an alternative to silicon-nitride or silicon oxide lms. Recently, TiO 2 lms have begun to be used as a hole-blocking layer directly making contact with c-Si solar cells. However, poor contact passivation showing effective surface recombination velocities (S eff ) of around 300 cm s À1 has limited the obtained efficiency to $11%. 22 Interface recombination can be drastically reduced by growing an extremely thin thermal SiO 2 layer at the interface between silicon and TiO 2 , where its thickness is the result of a trade-off between contact resistance and passivation. Using this approach for the preparation of the electron-selective contact along with conventional high temperature boron diffusion for the preparation of the holeselective contact, a remarkable efficiency of 21.6% has been recently obtained for a double-side contacted solar cell. 20 However, the high value of the contact resistance (>20 mU cm 2 ) prevents the use of this approach in IBC structures where the electron contact area is only a fraction of the total rear surface.
The replacement of this passivation interlayer, e.g. thermal SiO 2 or a-Si:H, by other lms, which could be conformably deposited at low temperature with good thickness control, is an important research topic that may drastically simplify the fabrication process and reduce its cost. In this work, we focus on electron-selective contacts made of atomic layer deposited (ALD) TiO 2 lms combined with an ALD ultrathin Al 2 O 3 interlayer. The Al 2 O 3 is introduced between the TiO 2 and the silicon surface aiming to reduce the defect density at the interface. In addition to Al 2 O 3 and TiO 2 thicknesses, the inuence of different metal capping materials (Ni, Al and Mg) has been studied to optimize the selectivity. This Al 2 O 3 /TiO 2 bilayer structure, or a repetition of it in nanolaminates, has been explored in the past for optical or gate dielectric applications. 23,24 The passivation properties of this stack have been also reported, 25 but it has not been previously applied and optimized as an electron-selective layer on c-Si solar cells, i.e. simultaneously targeting low surface recombination and contact resistance.
The best contact conguration consisting of a Mg-coated Al 2 O 3 /TiO 2 stack is applied as an electron-selective contact on n-type c-Si solar cells. This ETL approach has not been reported in the literature on nished photovoltaic devices. In this paper, combining this new ETL concept together with a previously explored HTL stack (Ni-coated V 2 O x ), 19 we report on highly efficient low-temperature and dopant-free n-type IBC c-Si solar cells with conversion efficiencies up to 19.1% on 9 cm 2 devices. It is important to stress that both selective contacts (ETL and HTL) are fabricated at temperatures lower than 100 C reducing the overall thermal budget of the fabrication process. Additionally the low resistivity values achieved using the ETL proposed herein allow us to circumvent the use of Indium Tin Oxide (ITO), further simplifying the fabrication process and eliminating the use of scarce materials.

Results and discussion
Surface contact passivation has been evaluated by extracting both the effective surface recombination velocity and the implicit open-circuit voltage (iV oc ) from effective lifetime (s eff ) measurements using the quasi-steady-state photoconductance decay technique (QSS-PC). 26 Passivation test samples were fabricated on n-type oat zone h100i c-Si substrates. One side was almost ideally passivated by means of a thick Al 2 O 3 lm (50 nm), 19,27 allowing us to neglect the corresponding recombination in the effective lifetime measurements. The other side was covered with the Al 2 O 3 /TiO 2 stack under investigation with different lm thicknesses. Fig. 1 shows the iV oc values as a function of the number of Al 2 O 3 and TiO 2 ALD cycles, i.e. lm thickness. Firstly, setting the number of Al 2 O 3 ALD cycles to 3, 20 cycles of TiO 2 seem to be the best choice from a passivation point of view, resulting in s eff and iV oc values of 660 ms (see ESI Fig. S1 †) and 683 mV, respectively. Both increasing and decreasing the number of TiO 2 cycles lead to a poorer surface passivation quality. This trend has already been reported by X. Yang et al. 28 Secondly, when setting the number of TiO 2 ALD cycles to 20, iV oc increases monotonically with the number of Al 2 O 3 cycles achieving an iV oc value of 685 mV (S eff of 33 cm s À1 ) for 12 ALD cycles. Information about the S eff extraction can be found in the ESI. † Thus, we conclude that an Al 2 O 3 interlayer between TiO 2 and c-Si improves the passivation, increasing the iV oc from 667 to 687 mV as is also shown from an effective lifetime point of view in the ESI (Fig. S1 †). Notice that no annealing is required to activate the passivation of the Al 2 O 3 / TiO 2 stacks. It is also important to stress that surface passivation is stable up to a temperature of 250 C as is shown in the ESI (Fig. S2 †), which is enough to perform the following technological steps and a hypothetical encapsulationlamination at 150 C.
The excellent surface passivation provided by the ALD Al 2 O 3 lms is due to both chemical and physical passivation mechanisms. 29 In other words, the silicon surface passivation when covered with Al 2 O 3 originates on one hand from the low surface state density and simultaneously from the electric eld effect due to the inherent charge in these lms. In annealed Al 2 O 3 lms this charge is negative. However, it has been reported that relatively thick dielectric Al 2 O 3 /TiO 2 stacks ($20 nm) exhibit a net positive charge even aer an annealing stage at 400 C. 30 Nevertheless, the details of the origin of surface passivation in ultra-thin conductive Al 2 O 3 /TiO 2 structures and the passivation behaviour with annealing temperature are still unclear and further research in this eld is required.
Although Al 2 O 3 /TiO 2 stacks provide excellent surface passivation on n-type silicon substrates, good electron conduction properties are mandatory for electron-selective contacts. To check the contact quality and the inuence of the metal capping on its electrical behaviour, test samples were fabricated on n-type c-Si substrates. The same Al 2 O 3 /TiO 2 stacks were deposited on both sides of bare c-Si substrates. Then, one side was totally covered with the capping metal (Mg, Al or Ni) with a thickness of around 15 nm and nished with an extra 150 nm thick Al layer, whereas the other side was selectively metalized with the same electrode materials using a shadow mask in order to dene dot features with different diameters (see the sketch in the inset of Fig. 2). The 150 nm thick Al layers  are deposited without breaking the vacuum and serve to facilitate the measurements.
Semi-log I-V curves for a reference Al 2 O 3 /TiO 2 stack (6/20 ALD cycles) with the different metal layers are shown in Fig. 2a. For a direct comparison, the I-V measurements were performed on dot features with the same pad diameter (4 mm) for all capping metals. The best behaviour corresponds to the sample in contact with Mg/Al achieving a low resistance ohmic curve with currents two orders of magnitude higher than that of the sample in contact with only Al. In contrast, the worst result is obtained for the sample in contact with Ni/Al, which results in a non-ohmic curve (see the asymmetric current level). Therefore, metal capping plays an important role in the functioning of the ETL structure. Note that depending on the metal material, the work function (4) changes from a low to a high value (4 Mg $ 3.6 eV, 4 Al $ 4.3 eV and 4 Ni $ 5.1 eV). The electrical contact quality improves with diminishing metal work function as expected in metal-semiconductor contacts on c-Si(n) substrates, i.e. the lower the work function value, the lower the contact barrier height, and the higher the majority carrier conduction through the silicon/contact interface. A similar behaviour has also been reported with a-Si:H passivated contacts using either Mg/Al or Al metal capping. 31 In Fig. 2b the impact of the Al 2 O 3 interlayer thickness ranging from 0 to 24 ALD cycles on the electrical contact quality is shown. In all cases, the thickness of the TiO 2 layer was xed at 20 ALD cycles and Mg/Al was used as a capping electrode. As we can see in this gure, for voltages from À1 to 1 V the characteristics are clearly ohmic for 0, 3, and even 6 ALD cycles. Nevertheless, this electrical behaviour changes to a poor ohmic contact for 12 ALD cycles, and exhibits a pseudo-rectifying I-V curve for the the case of 24 ALD cycles. This fact is in agreement with the dielectric nature of ALD Al 2 O 3 lms, which must nally block the electron-extraction, i.e. the tunnel conduction mechanism vanishes. Consequently, Al 2 O 3 lms prepared with 6 cycles ($0.5 nm) yield the optimum trade-off between good passivation (thick lms) and good electrical contact performance (thin lms). Fig. 2c shows the I-V curves for different dot-like contact diameters with the best stack conguration, Al 2 O 3 (6 cycles)/ TiO 2 (20 cycles)/Mg. The specic contact resistance (r c ) of the stack can be extracted by tting the measured total resistance (R T ) for the different dot diameters using 3D simulations (Silvaco ATLAS TCAD). 32 The results can be seen in the inset of Fig. 2c. Excellent r c values of 2.5 AE 0.5 mU cm 2 are obtained for our electron-selective contacts. Cumulative 5 min annealings have been performed showing stable results both for the specic contact resistance and for the iV oc up to 200 C (see ESI Fig. S2 †). This result improves r c by more than one order of magnitude compared to SiO 2 /TiO 2 stacks on n-type substrates. 20 It is important to point out that small r c values are crucial for manufacturing non-full contact high efficiency solar cells, in which the ETL coverage area fraction (f ETL ), dened as the ratio between ETL and total cell area, is typically in the 1-10% range. Considering that the specic series resistance (r s ) is limited only by the ETL contact quality, the total series resistance could be roughly calculated as r s y r c Â 100/f ETL (%), yielding for the worst case (f ETL ¼ 1%) values below $0.3 U cm 2 in nal devices.
Finally, a single gure of merit quantifying contact quality is the so-called contact selectivity (S 10 ) dened as S 10 ¼ log(V th /(r c Â J 0c )), 33 where V th is the thermal voltage (25.69 mV at 25 C) and J 0c is the contact recombination current density. This parameter is an alternative way to characterize surface recombination and can be calculated from the more common S eff concept (see the ESI † where a detailed explanation of this calculation can be found). Good contacts are characterized by simultaneously low r c and J 0c values. Thus, the higher the S 10 value, the better the contact behaviour. Table 1 compares the main gures of merit reported in the literature for different ETLs. As can be observed, although the use of a thin a-Si:H(i) or alternatively a thermal SiO 2 lm as a passivating interlayer provides excellent surface passivation, these approaches usually exhibit poor conductivity, i.e. high contact resistance values. The ETL stack reported herein achieves a remarkable S 10 value of 13.9, which is the best result for electron collector layers without using any PECVD or thermally grown passivation interlayer.
In order to investigate the surface chemical composition of the reference Al 2 O 3 /TiO 2 stack (6/20 ALD cycles), both uncoated and Mg-coated samples were analysed by the X-ray photoelectron spectroscopy (XPS) technique. In Fig. 3 the XPS analysis Table 1 Summary of ETL strategies applied on n-type c-Si substrates and their relevant contact quality parameters. In the lower part of the a Reported values related to contact passivation have been measured before LiF x deposition. b Extrapolated using the relationship J 0c ¼ S eff Â qn i 2 / N D from current S eff values, where N D , q and n i are the substrate doping, elementary charge and intrinsic concentration (n i ¼ 8.56 Â 10 9 cm À3 at T ¼ 25 C), respectively; see the ESI. c Measured in asymmetric test samples (one side ideally passivated). results of Si 2p, Ti 2p, and O 1s core levels for both samples are compared. In both cases, two peaks appear in the Si 2p spectrum (Fig. 3a). These peaks, which correspond to bulk Si and silicon oxide (Si-O) states, can be deconvoluted and tted using Gaussian-Lorentzian curves centred at $99.0 and $102.0 eV binding energies, respectively. Since a HF dip was performed just before the ALD stack growth, oxidation of the Si surface might occur during the rst water pulse in the Al 2 O 3 deposition   The Ti 2p 3/2 core level spectrum is shown in Fig. 3b. For the uncoated sample, a single peak centred at a binding energy of $458.7 eV is enough to t the curve. However, an extra peak centred at $457.1 eV appears for the Mg-coated sample. The peak at higher binding energy corresponds to a fully oxidized TiO 2 , while the lower binding energy peak is assigned to a titanium rich TiO x layer. 24 From the relative area of the Ti-O peak with respect to the Ti 2p spectrum, an approximate TiO x content of 26% was calculated, indicating that only a fraction of the TiO 2 layer is under-stoichiometric. The TMO work function changes strongly with the lm stoichiometry, 34 i.e. oxygen-decient layers exhibit lower work function values and a semi-metallic behaviour. In our devices, these effects might enhance the electron selectivity because of a proper band alignment and an improvement of layer conductivity. This oxygen deciency could be attributed to a reaction of the thermally evaporated Mg with the TiO 2 , forming a magnesium oxide layer (MgO x ). This hypothesis is corroborated by the Mg 2p spectrum, which exhibits a peak centred at a binding energy of $50.8 eV corresponding to an Mg-O compound (ESI Fig. S3a †).
Finally, an analysis of the O 1s core level spectrum (Fig. 3c) corroborates the presence of the many chemical species mentioned above. First, Al-O and Ti-O species (with a characteristic binding energy of $531.5 eV and $530.0 eV respectively) are identied in both samples, which correspond to the TiO 2 and Al 2 O 3 layers. However, the Mg-coated sample shows a smaller Ti-O intensity, attenuated by the Mg on top, as well as a Mg-O peak at 530.4 eV, conrming the presence of MgO x in the TiO 2 /Mg interface. An interesting feature is the detection of carbon contaminants adsorbed to the surface aer airexposure, whose contribution is particularly high for the Mgcoated sample ($532.8 eV) and can be attributed to metal carbonate species, i.e. MgCO 3 . 35 Also, the Si-O component ($533.0 eV) is discernible for both samples.
In order to get a better insight into the inuence of Mg on both the chemistry of Al 2 O 3 /TiO 2 stacks and the electrical contact performance, we analysed their chemical composition using both energy dispersive X-ray spectroscopy (EDX) and electron energy loss spectroscopy (EELS) analysis. In Fig. 4a the high-angle annular dark eld (HAADF) scanning transmission electron microscope (STEM) image reveals details of the contact structure. Firstly, a $15 nm thick Mg layer is observed as well as a $2 nm thick Al 2 O 3 /TiO 2 stack. With the closer view shown in Fig. 4b, an Al 2 O 3 interlayer ($0.5 nm) was observed in the TiO 2 / c-Si interface capped with a $1.5 nm thick TiO 2 layer. Due to the strong oxygen affinity of Mg metal, magnesium atoms easily attract oxygen atoms from the TiO 2 interlayer forming an MgO x layer with a thickness of $2 nm. The existence of this extra layer is in agreement with the previous XPS measurements and conrmed by the EELS chemical composition cross scanning (see Fig. 4c). Interestingly, a similar behaviour was also observed when aluminium was evaporated directly onto TiO 2 layers forming in this case an AlO x layer (see ESI Fig. S4 †), in agreement with other reported studies. 20 However, if Mg is used as a capping metal electrode, the spontaneous growth of a MgO x lm may enhance the electron-selectivity of our contacts due to the already reported ETL nature of magnesium oxide. 15 To demonstrate the viability of Al 2 O 3 /TiO 2 /Mg stacks as electron-selective contacts for c-Si solar cells, we fabricated a proof-of-concept 3 Â 3 cm 2 IBC structure on n-type c-Si substrates. The inset of Fig. 5a depicts the cell structure. The front-textured surface was covered with a 50 nm thick Al 2 O 3 layer and further coated up to 75 nm with a close to stoichiometric silicon carbide layer (a-SiC x ) providing excellent passivation as well as antireection properties. 36 It is important to point out that the gap between the interdigitated back-contacts is also passivated with the same stack. Therefore all non-contact surfaces are covered with a passivating and light conning Al 2 O 3 /SiC x stack which redirects light towards the bulk of the absorber and photogenerated carriers towards the corresponding contact.
External quantum efficiency (EQE) measurements for the best device are shown in Fig. 5a. EQE values close to 97% in the visible wavelength range conrm an excellent front and rear surface passivation pointing out that electron-selectivity in our ETL regions is good enough to obtain high photovoltaic efficiencies. Short circuit current densities (J sc ) of about 40 mA cm À2 and efficiencies (h) in the 18-19% range independently of the ETL coverage (f ETL ) have been measured, suggesting that surface passivation in the ETL regions is good enough to avoid excessive electrical shadowing. 37 A similar high-efficiency structure has recently been reported combining a LiF x -based ETL and V 2 O x -metal-V 2 O x as the HTL in an IBC structure. 12 However, in that device a high temperature phosphorus diffusion stage is needed to form a front surface eld (FSF).
Details related to the photovoltaic parameters of the fabricated solar cells are listed in Table 2. Open circuit voltages (V oc ) of up to 633 mV and ll factors (FF) of around 75% are achieved in our devices. Another interesting analysis is shown in Fig. 5b, where the illuminated current density-voltage (J-V) curve for the best IBC solar cell is compared with the corresponding pseudo-light J-V curve from Suns-V oc measurements. 38 These measurements allow the determination of a pseudo J-V curve that excludes the series-resistance contribution. The matching between these two curves suggests that series resistance is not the limiting factor in the efficiency, as is corroborated by similar values of ll factor and pseudo-FF (pFF) parameters. The specic series resistance calculated using the relationship r s y (1 À FF/pFF) Â V oc /J sc 39 yields values below 0.4 U cm 2 in all cases, conrming the high contact quality of our ETL stacks.
The relatively low FF can be attributed to a diode ideality factor (n) close to two as can be seen in the dark J-V curve shown in the inset of Fig. 5b. This high n value might be attributed to poor coverage at the edges of the strip-like regions of the V 2 O xbased HTL as is explained in detail in the ESI. † This drawback could be overcome in the future using the ALD technique to deposit both ETL and HTL selective contacts.

Conclusions
In this work, we apply atomic layer deposition of aluminium and titanium oxide (Al 2 O 3 /TiO 2 ) stacks to form electron-selective contacts for interdigitated back-contacted n-type c-Si solar cells.
The use of a very thin ALD Al 2 O 3 passivation interlayer ($0.5 nm) between silicon and the thin TiO 2 lm ($1.5 nm) improves contact surface passivation, yielding surface recombination velocities below 33 cm s À1 and implicit open circuit voltages of about 680 mV in our test devices. Additionally, the metal capping layer plays an important role in contact conductivity, Mg being the best choice of material, with specic contact resistance values as low as 2.5 mU cm 2 being reached. Chemical composition analysis (XPS and EELS) conrms spontaneous growth of a MgO x layer, leaving an oxygen-decient TiO x layer in our Mg-coated samples.
Finally, the optimum electron-selective contact conguration consisting of an Al 2 O 3 (6 ALD cycles)/TiO 2 (20 ALD cycles)/ Mg stack was applied to an interdigitated back-contacted solar cell structure using a V 2 O x /Ni stack as a hole transport layer. In this way, fully TMO-based (9 cm 2 area) solar cells reaching efficiencies up to 19.1% through a dopant-free and lowtemperature fabrication process have been fabricated. These results outline the potential of TMOs to improve the trade-off between technological complexity and efficiency in Si solar cells.

Experimental
All the structures were fabricated using high quality h100i planar oat zone (FZ) n-type c-Si wafers with a resistivity of 2 AE 1 U cm and a thickness of 280 AE 10 mm. The general trends of the IBC cell fabrication process have been described in a previous study. 19 In this work the Al 2 O 3 /TiO 2 /Mg stack replaces laser doped contacts to create ETL regions. Therefore, fully lowtemperature and dopant-free IBC solar cells based on TMOs have been fabricated. Photovoltaic parameters and illuminated J-V curves were measured under standard test conditions (1 kW m À2 , AM1.5G solar spectrum, 25 C) using an ORIEL 94021A (Newport) solar simulator. Light irradiance was properly calibrated by means of a pyranometer. The EQE curves were measured using a commercial instrument (QEX10, PV measurements) with a white light bias of 0.2 Suns and a beam spot of 2 Â 2 cm 2 centred within the active device area.
The passivation test samples were subjected to a cleaning sequence of RCA1/2 (ref. 40) and a diluted HF (1%) dip prior to deposition at 200 C on both sides of a 50 nm thick Al 2 O 3 layer by ALD (Savannah S200, Cambridge Nanotech). Trimethylaluminium (TMA) and water were used as the aluminium precursor and oxidant species, respectively. A subsequent 10 min annealing was performed in forming gas (H 2 /N 2 ) at