Hydrogen and nitrogen codoping of anatase TiO2 for efficiency enhancement in organic solar cells

TiO2 has high chemical stability, strong catalytic activity and is an electron transport material in organic solar cells. However, the presence of trap states near the band edges of TiO2 arising from defects at grain boundaries significantly affects the efficiency of organic solar cells. To become an efficient electron transport material for organic photovoltaics and related devices, such as perovskite solar cells and photocatalytic devices, it is important to tailor its band edges via doping. Nitrogen p-type doping has attracted considerable attention in enhancing the photocatalytic efficiency of TiO2 under visible light irradiation while hydrogen n-type doping increases its electron conductivity. DFT calculations in TiO2 provide evidence that nitrogen and hydrogen can be incorporated in interstitial sites and possibly form NiHi, NiHO and NTiHi defects. The experimental results indicate that NiHi defects are most likely formed and these defects do not introduce deep level states. Furthermore, we show that the efficiency of P3HT:IC60BA-based organic photovoltaic devices is enhanced when using hydrogen-doping and nitrogen/hydrogen codoping of TiO2, both boosting the material n-type conductivity, with maximum power conversion efficiency reaching values of 6.51% and 6.58%, respectively, which are much higher than those of the cells with the as-deposited (4.87%) and nitrogen-doped TiO2 (4.46%).


Results and Discussion
Theoretical results and discussion. TiO 2 has three polymorphs namely rutile, anatase and brookite. Both rutile and anatase are tetragonal with space groups P4 2 /mnm and I4/amd respectively 33,34 . The lattice parameters of anatase TiO 2 are calculated at a = 3.806 Å and c = 9.724 Å in excellent agreement with the experimental neutron diffraction results (a = 3.782 Å and c = 9.502 Å) 33 and theoretical results (a = 3.729-3.801 Å and c = 9.480-9.818 Å) 34 .
Although anatase and rutile have the same chemical composition, they differ in the coordination environments (i.e. chemical bonding) and this leads to different ionization potentials and chemical affinities 35 . Interestingly, in a recent report Luttrell et al 36 . studied epitaxial TiO 2 films to consider why anatase is a better photocatalyst as compared to rutile. This study determined that charge carriers that were excited deeper in the bulk contributed more to surface reaction in anatase as compared to rutile 36 . Considering this, the present study will focus on H and N defects in bulk anatase discussing wherever appropriate, recent DFT results concerning the anatase TiO 2 surface 37 . It is calculated using DFT that atomic hydrogen can be a substitutional defect occupying oxygen sites (referred to as H O ). This is in agreement with the present experimental results which identified only Ti 4+ states (as discussed in the section referred to the synthesis of TiO 2 materials) and previous DFT studies 38 . Additionally, we considered interstitial hydrogen (H i ) and developed an algorithm to track the minimum energy interstitial sites using an extensive search. We calculate that H i resides 0.99 Å from the nearest oxygen atom ( Fig. 1(a)). This is in agreement with previous DFT results 39 . Next, the electronic density of states (DOS) for anatase TiO 2 was calculated in order ito understand the electronic structure and nature of the band edge wavefunctions ( Fig. 2(a)). The valence band edge of the material is dominated by O 2p, and the conduction band edge is formed from Ti 3d orbitals. The band gap is calculated to be 3.1 eV in agreement with the experimentally measured one (3.2 eV). Adding H i atoms to form multi-hydrogen clusters is energetically unfavourable (for example in the H i H i pair the atoms repel each other with 1.14 eV). DOS reveal that the introduction of H leads only to a small band gap narrowing (refer to Fig. 2(b)) consistently with the experimental results shown below.
The introduction of nitrogen was considered at both a Ti site (N Ti ) or as an interstitial (N i , refer to Fig. 1(b)). For N i a density of new states (refer to Fig. 2(c)) is observed above the valence band maximum, VBM (i.e. p-type doping); this is attributed to the N 2p states lying just above the O 2p states.
In codoped anatase it should be expected that N and H defects can interact. Here we have performed an extensive investigation of the different possibilities that the interstitial and/or substitutional hydrogen and nitrogen defects can associate. Figure 1(c-e) represents the N i H i , N i H O and N Ti H i defect pairs. The N i H i defect pair is bound (−0.33 eV) when the two interstitials are close at N-H = 1.05 Å (Fig. 1(c)) as compared to when the N-H distance is 8.07 Å. The N i H O is calculated to have a binding energy of −3.77 eV (energy difference between configurations being at N-H = 0.93 Å and N-H = 7.96 Å), however, its concentration will be limited by the number of available oxygen vacancies (as there as required for H O to form). Therefore, although the majority of oxygen vacancies should be occupied by atomic hydrogen 18,32 , the concentration of N i H O will depend upon the growth conditions. Finally, the N Ti H i defect is bound with −1.45 eV (energy difference between configurations being at N-H = 2.88 Å and N-H = 10.87 Å). In essence, N Ti defect will capture migrating H i , but again this defect will require Ti vacancies to form in the first place.
The DOS reveal that the N i H O and N Ti H i defects (Fig. 2(e-f)) introduce levels at about 1.05 eV and 0.75 eV below the conduction band minimum, CBM. More importantly, N i H O defect introduces a pronounced density of states above the VBM resulting in undesirable p-type doping. These deep levels, in essence, may act as recombination centers of photo-excited electron-hole pairs and consequently can lower the efficiency of OSCs using TiO 2 ETLs. Therefore, H,N codoped TiO 2 should be grown under O-rich conditions (to provide a stoichiometric TiO 2 as the starting material) to avoid the formation of oxygen vacancies and thus N i H O defects. anatase TiO 2 films were converted into H-doped (referred to as TiO 2 :H), N-doped (referred to as TiO 2 :N), or codoped (referred to as TiO 2 :H,N) TiO 2 films by annealing at 550 °C in N 2 , H 2 , or in forming gas, respectively. In Fourier transform infra-red (FTIR) transmittance spectra (refer to Fig. S1, Supporting Information) taken in the above films exhibiting the same thickness of ~40 nm (refer to Fig. S2), apart from a broad band at the lower wavenumbers which is due to Ti-O bonds, the spectral features due to defect species are negligible; however, hydrogen and forming gas annealing seems to increase the intensity of the broad peak at 3000-3600 cm −1 in the region were stretching of hydroxyl group ν(OH) appears which is an indication for the insertion of hydrogen within the oxide lattice (i.e., H doping). This peak is decreased after annealing in nitrogen. From the analysis of FTIR spectra the incorporation of hydrogen dopants within the material after annealing in hydrogen and forming gas can be concluded; however, the content of H dopants cannot be estimated from those spectra since the contribution of water in the intensity of the peak around 3500 cm −1 cannot be excluded. Notably, the incorporation of hydrogen within the material causes some alteration in its crystallinity, especially in the hydrogen annealed sample (refer to Fig. S3). Moreover, TiO 2 samples exhibit very small band gap narrowing with H doping (from 3.20 to 3.14 eV) while the band gap narrowing is more pronounced with N doping (3.10 eV) (Fig. S4). N and H codoping only has a minor effect on the band gap value (3.13 eV). These results are in agreement with those obtained by Pan et al 12 .
X-ray photoelectron spectroscopy (XPS) analysis of the surface elementary composition of the formulated materials confirms that N was detected in both samples deposited in environment containing N (Fig. S5). This observation indicates that N doping takes place after annealing in nitrogen and forming gas. Only Ti 4+ states were identified in the starting material (Fig. S6), revealing that the as-deposited TiO 2 film is highly stoichiometric 40 . In addition, Ti 4+ states were also detected in the annealed samples (Fig. S6). Since we do not find localized Ti 3+ states in our annealed samples, it can be concluded that on the sample surface there are no other Ti states. Figure S7 represents the O1s spectra of TiO 2 samples. The peak at binding energy 530.4 ± 0.1 eV (fittings are not included here) corresponds to O-Ti bonds in TiO 2 . A second component is always present which is broader at binding energy 531.6 ± 0.1 eV due to Ti-OH bonds. The broadening in the O 1 s peaks of nitrogen containing samples could be due to the combination of hydroxyl groups and N-O-Ti bonds present in these samples.
The surface electronic structure of different TiO 2 materials was evaluated by ultra-violet photoelectron spectroscopy (UPS) measurements shown in Fig. 3, which reveal that the valence band of TiO 2 samples remain almost unaffected.
The valence band of the TiO 2 sample consists of a broad peak at about 6 eV and a narrow peak at about 8 eV, which correspond to π (non-bonding) and σ (bonding) Ti3d-O 2p orbitals 41 . An additional small peak, located at 10.6 eV, is observed in all spectra and it is more pronounced in the hydrogenated samples. This feature suggests the presence of an adsorbate such as -OH on the surface and it is correlated with the 3σ of such molecules. However, this feature is more pronounced in the spectra of samples annealed either in hydrogen or in forming gas, an indication of H doping in these cases. The VBM of TiO 2 is located at ~3.2 eV below the Fermi level while its work function was found equal to 4.1 eV. As its band gap is estimated to be 3.2 eV the bottom of its conduction band locates at ~4.1 eV indicating the n-type character of the TiO 2 film. The VBM of the doped TiO 2 films slightly shifts to 3.0 eV before the Fermi level while for the TiO 2 :H and TiO 2 :N,H samples there is an increase in the W F value by 0.3 eV which is due to the adsorbed -OH species on the surface. The work function of N containing samples is increased by 0.4 eV in comparison with the TiO 2 sample due to the p-type doping character of N. The results provide evidence that treatment of TiO 2 with hydrogen and, especially, with nitrogen results in the creation of gap states above the VBM and, concomitantly, in a decrease of the effective band gap of the semiconductor 42 .
H intercalation within a metal oxide such as TiO 2 is expected to increase its n-type conductivity as well as to passivate surface defects 43,44 . To support this argument we next measured the current-voltage characteristics of devices with the structure glass/FTO/TiO 2 /Al ( Fig. 4(a)). The thickness of TiO 2 was measured at 40 nm in all cases. The significant increase in the slopes of I-V curves upon H doping and N,H codoping of TiO 2 witnesses the improved n-type conductivity of the latter. On the contrary, reduction of n-type conductivity is concluded from the reduced slope of the I-V curve of the TiO 2 :N embedding diode.
The photoluminescence (PL) emission spectrum of TiO 2 results from the recombination of free carriers; therefore, these spectra are used to understand the effect of surface traps on photo-generated electrons and holes in TiO 2 samples 45 . The PL spectra of as-deposited, N and H doped and codoped TiO 2 samples in the wavelength range of 350-700 nm with the excitation at 325 nm are shown in Fig. 4(b). The shape of the emission spectra is very similar consisting of two broad peaks located at about 400-450 nm and at 450-600 nm, respectively. These two peaks result from surface defects of the TiO 2 samples. The PL peaks intensity of the H doped and N,H codoped TiO 2 decreases as compared to the as-deposited TiO 2 which is an indication of lower recombination rate of electrons and holes due to effective passivation of the surface (and bulk) defects of these samples upon H doping. Note that our doped TiO 2 samples retained their electrical and optical characteristics presented above when stored in air for a long period of time (data not shown).

Application of H and N doped and codoped TiO 2 in organic solar cells.
The configuration of the inverted polymer solar cells using TiO 2 electron transport layers is illustrated in Fig. 5(a), where the chemical structures of organic semiconductors (P3HT and IC 60 BA) used in this study are also shown. Figures 5(b),(c) represent the J-V characteristics under full 1.5 AM simulated solar illumination and in dark, respectively, of devices using photoactive blends based on P3HT:IC 60 BA with as-deposited, N and H doped and codoped TiO 2 ETLs, respectively. The corresponding electrical output parameters of those devices are summarized in Table 1. The external quantum efficiency (EQE) characteristics of the same devices are presented in Fig. 5(d). For the reference P3HT:IC 60 BA-based device with the as-deposited TiO 2 layer a PCE of 4.87% is obtained while N p-type doping of TiO 2 results in a PCE of  4.46%, which is 8% lower as compared to the reference cell. On the contrary, H doping and N and H codoping result in PCEs up to 6.51% and 6.58%, respectively, representing a 34% and 35% improvement as compared to the control device. From the photovoltaic parameters presented in Table 1 becomes evident that by using TiO 2 :H and, especially, TiO 2 :N,H ETLs a simultaneous improvement in the open-circuit voltage (V oc ), short-circuit current density (J sc ) and fill factor (FF) of the devices is obtained. In addition, the dark J-V characteristics (shown here in semi-logarithmic scale, Fig. 5(c)) are improved in the case of TiO 2 :H and TiO 2 :N,H, compared to the reference and, especially, to TiO 2 :N-based one. Importantly, the reverse saturation current density decreases significantly in the case of hydrogen containing TiO 2 -based devices. These results suggest that H doping and N,H codoping of TiO 2 reduces reverse leakage and shunt current while it also facilitates electron extraction probably due to an increase of the built-in field and/or due to enhanced n-type conductivity of the metal oxide. Indeed, upon O-H bond formation, an electron is donated from the Ti-O(H)-Ti species, which results in n-type doping of the TiO 2 thus improving its n-type conductivity. Both reduction of the reverse saturation current and increase of the turn-on voltage contribute to the increase of the V oc of the H-doped TiO 2 -based solar cells. In addition, the devices with the TiO 2 :H and TiO 2 :N,H layers exhibit significantly reduced series and larger shunt resistances (refer to Table 1) verifying the overall higher quality of the corresponding diodes and explaining the enhanced J sc and FF obtained in these devices. The large improvement in J sc is further supported by the external quantum efficiency (EQE) measurements (refer to Fig. 5(d)). In addition, the EQE spectra of the devices were integrated from 300 to 800 nm and J sc(EQE) values were calculated from the integration and listed in Table 1. Notably, these values are quite consistent with the J sc obtained from J-V characteristics (especially for the devices with the H doped and N,H codoped TiO 2 ) under illumination indicating the reliability of the measured J sc.
Similar efficiency enhancement was obtained in organic solar cells using blends of P3HT donor with [6,6]-phenyl-C 70 butyric acid methyl ester (PC 70 BM) acceptor (Fig. S8) thus proving the universality of the present approach. From these results, it becomes evident that H doping of TiO 2 , especially in the presence of nitrogen defects (i.e. N,H codoping), is an effective way to significantly improve the operational characteristics and overall performance of OSCs using TiO 2 ETLs while a simple nitrogen doping may negatively affect the device performance. According to our theoretical predictions and experimental results one should argue that, the presence of interstitial nitrogen (N i ) within the material may be beneficial for the incorporation of the non-bonded hydrogen atoms via the formation of N i H i bonds thus suppressing the p-type character of N doping while simultaneously increasing the amount of H dopants in TiO 2 .
To shed more light on the mechanisms responsible for enhanced device performance when using H doped and N,H codoped TiO 2 ETLs we next constructed the energy level diagram (based on UPS and absorption measurements) of various interfaces before contact (considering vacuum level alignment) at the cathode side of the device. From Fig. 6(a) it becomes evident that in all cases the lowest unoccupied molecular orbital (LUMO) of IC 60 BA acceptor lies above the conduction band minimum of TiO 2 which indicates that electron extraction occurs via transport of electrons through the conduction band of TiO 2 . In the n-type hydrogen-doped oxides (TiO 2 :H and TiO 2 :N,H) the increase of electron conductivity highly facilitates electron extraction. However, p-type doping of TiO 2 :N sample causes the filling of localized states lying above the VBM as predicted by DFT calculations. Figure 6(b) shows that these localized states are probably aligned with the highest occupied molecular orbital (HOMO) of P3HT. As a result, after photoexcitation of P3HT through light absorption a high hole recombination rate is expected at the TiO 2 :N/P3HT interface which significantly suppresses the device photocurrent. Note that surface states of TiO 2 arising from oxygen vacancies have also been proven to act as recombination centers thus deteriorating the device current and V oc 30 . Passivation of such defect states via hydrogen doping and codoping is expected to be beneficial for electron extraction/collection and overall device performance. Based on the above, one should also conclude that N i H i defects are rather formed in the codoped oxide because N i H O and N Ti H i are predicted to introduce deep level states, exactly as N i does, which are expected to deteriorate the device performance which was not observed in our devices embedding N,H codoped TiO 2 .
To further elucidate the effect of N and H doping and codoping of TiO 2 on the optoelectronic properties of the P3HT:IC 60 BA-based OSCs, we examined the charge transport properties by measuring electron-only devices with the structure: FTO/TiO 2 /P3HT:IC 60 BA/Al. The J-V characteristics are shown in semi-logarithmic plots in Fig. 7(a) and reveal a substantial increase in electron current after the insertion of H dopants in TiO 2 . This improvement in electron current density can be attributed to the significant passivation of surface defect states as well as to increase of n-type conductivity of TiO 2 upon H doping and is further improved in the case of N,H codoping.
In addition, as shown in Fig. 7(b), where a log-linear plot of V oc as a function of the light intensity is presented, the slope values of V oc were decreased from 1.14 to 1.07 and 1.05 k B T/q upon H doping and N,H codoping, where k B is Boltzmann's constant, T is the absolute temperature, and q is the elementary charge. On the contrary, the slope increases to 1.21 k B T/q upon N doping of TiO 2 ETL. This confirms that the insertion of H dopants into the ETL J sc (mA cm −2 ) J sc(EQE) (mA cm −2 ) V oc (V) FF PCE(%) R s (Ω cm 2 ) R sh (Ω cm 2 )  46 . Apart from the increased efficiency the high long term stability under ambient air of organic solar cells is highly desirable and was verified in the case of H doped and N,H codoped TiO 2 embedding devices (refer to Fig. 8). The improved stability of the doped samples can be partly explained by the fact that H dopants terminate dangling bonds preventing adsorption of oxygen and moisture on the TiO 2 surface 47 . Note that the devices were intentionally un-encapsulated so that they were exposed to ambient conditions (moisture and oxygen) throughout the aging study.
Finally, to evaluate our approach in cases where lower thermal budget is necessary, we fabricated P3HT:IC 60 BA-based OSCs using TiO 2 ETLs post-annealed at 300 °C for 30 min to obtain H, N, or N,H codoping (refer to Fig. S9). It is observed that annealing in hydrogen containing environment induces significant enhancement in the device performance as compared to the reference cell while nitrogen annealing of TiO 2 causes a small reduction in the device photocurrent and FF. Those results represent an indication that the intercalation of significant amount of H dopants in TiO 2 is an effective way for boosting its electron transport capability for application in organic photovoltaics and related cells. Note that, our group has recently demonstrated the beneficial effect of H doping of ZnO (annealed at 300 °C in hydrogen environment) for organic solar cells application 32 . Based on the present results, an extensive study regarding N, H doping and N,H codoping of ZnO and influence on organic optoelectronics seems necessary and is currently carried out.
In the present study, H and N doped and codoped anatase TiO 2 was systematically investigated using a range of experimental techniques and DFT calculations. DFT calculations reveal that the N i H i , N i H O and N Ti H i defects are bound, although the latter two defects require vacancies and this may hinder their formation. Conversely, H i do not bind together to form extended H clusters. It is therefore expected that in codoped anatase there will be

Methods
Titanium Oxide Layer Preparation. FTO-coated glass (Pilkington TEC 15, <15 Ohms/sq) was cleaned by sonication in detergent solution (Hellmanex III, Hellma Analytics), water and ethanol, followed by treatment in an oxygen plasma for 5 min. A solution of 13 µL concentrated aqueous HCl in 5 mL of dry isopropanol was slowly added to a stirred solution of titanium isopropoxide (711 mg, 2.5 mmol) in 5 mL of dry isopropanol. The cleaned substrates were spin-coated at 2000 rpm with this titania precursor solution and immediately placed on a hotplate at 150 °C. Subsequently, the samples were calcinated at 500 °C for 45 min (1 h ramp). N and H doped and codoped TiO 2 were obtained after annealing in nitrogen, hydrogen and forming gas environment at 550 °C for 1 hour. The annealing was made in a home-made furnace equipped with a quartz chamber with a graphite susceptor on which samples were placed. The susceptor was radiatively heated by three tungsten-halogen lamps of 1000 W each. The temperature of the susceptor (and therefore of the samples on it) was controlled with an automatic temperature control system, which was receiving feedback from a thermocouple positioned in a hole on the graphite and was controlling the power of the lamps. After loading samples, the chamber was evacuated down to 2 10 −2 Torr. Then a nitrogen stream was allowed to flow through it to maintain a pressure of 0.1 Torr and the temperature was raised to the desired point (550 °C). After the preset temperature was reached, the chamber was evacuated again down to 2 10 −2 Torr and a stream of the doping gas (hydrogen or forming gas or pure nitrogen) was introduced in it at a pressure of 1 Torr. At the end of the annealing the chamber was evacuated from the doping gas, the heating lamps were turned off and the samples were left to cool down to 70 °C under 0.1 Torr of nitrogen.
Device Fabrication. Organic solar cells were fabricated on 40 nm thick TiO 2 films (as-deposited or doped), which served as the electron transport layers, deposited on FTO coated glass substrates, as described above. The active layer consisted of a P3HT:IC 60 BA blend (17 mg ml −1 for P3HT, 17 mg ml −1 for IC 60 BA in 1,2-dichlorobenzene). After spin coating at 800 rpm for 30 sec the photoactive layers were left to dry for 30 min and then annealed at 150 °C for 15 min. Note that all depositions and thermal annealing treatments were carried out in the inert environment of an argon filled glove-box. Then, an approximately 20 nm-thick under-stoichiometric molybdenum oxide (MoO 3 ) layer was deposited on top of the active layer, using a previously reported method, to serve as the hole transport/extraction material 48,49 . The devices were completed with a 150 nm thick aluminium anode, deposited in a dedicated thermal evaporator at a pressure of 10 -6 Torr through a shadow-mask, which defined the device active area to be equal to 12.56 mm 2 . The devices were then measured in air at room temperature without additional encapsulation. P3HT was purchased from Rieke metals and IC 60 BA was purchased from Solenne.
Measurements and Instrumentation. XPS and UPS were recorded by Leybold EA-11 electron analyzer operating in constant energy mode at pass energy of 100 eV and at a constant retard ratio of 4 eV for XPS and UPS, respectively. All binding energies were referred to the C 1 s peak at 284.8 eV of surface adventitious carbon. The X-ray source for all measurements was an unmonochromatized Mg Kα line at 1253.6 eV (12 keV with 20 mA anode current). The valence band spectra of TiO 2 samples were evaluated after recording the UPS spectra of approximately 40 nm thick films deposited on an FTO substrate. For the UPS measurements, the He I (21.22 eV) excitation line was used. A negative bias of 12.22 V was applied to the samples during UPS measurements in order to separate secondary electrons originating from sample and spectrometer and to estimate the absolute work function value from the high BE cut-off region of the UPS spectra. The analyzer resolution is determined from the width of the Au Fermi edge to be 0.16 eV. Absorption measurements were taken using a Perkin Elmer Lambda 40 UV/Vis spectrophotometer. FTIR transmission spectra were obtained on a Bruker Tensor 27 spectrometer (at 4 cm −1 resolution, 128 scans). Film thicknesses were measured with an Ambios XP-2 profilometer. Photoluminescence measurements on TiO 2 were carried out using a Horiba Jobin-Yvon iHR320 Spectrometer with a He-Cd laser (325 nm) as excitation source. X-ray diffraction (XRD) measurements were performed using a Siemens D500 diffractometer with Cu-Ka radiation. Current density-voltage characteristics of the fabricated solar cells were measured with a Keithley 2400 source-measure unit. Cells were illuminated with a Xe lamp and an AM 1.5 G filter to simulate solar light illumination conditions with an intensity of 100 mW/cm 2 (1 sun), as recorded with a calibrated silicon photodiode. To accurately define the active area of all devices, we used aperture masks during the measurements with their area equal to those of the Al contacts (12.56 mm 2 ). EQE measurements were carried out using an Autolab PGSTAT-30 potentiostat, with a 300 W Xe lamp in combination with an Oriel 1/8 monochromator for dispersing the light in an area of 0.5 cm 2 . A Thorlabs silicon photodiode was used for the calibration of the spectra. All measurements were performed in air.
Computational methodology. The calculations were performed in thermally stable anatase TiO 2 using the plane wave DFT code CASTEP 50,51 . Exchange and correlation interactions were modelled with the corrected density functional of Perdew, Burke and Ernzerhof (PBE) 52 in the generalized gradient approximation (GGA), with ultrasoft pseudopotentials 53 . Spin polarized calculations with the inclusion of the Hubbard U contribution were used to account for the strong Coulombic interaction of the localised electrons of the 3d orbitals of Ti. The U value was set to 8.2 eV as this brings the band gap close to the experimental one in agreement to the study of Kiarii et al 54 . The plane wave basis set was set to a cut-off of 480 eV, in conjunction with a 7×7 ×7 Monkhorst-Pack (MP) 55 k-point grid and a 108-atomic site supercell. The calculations were under constant pressure conditions. The minimum energy interstitial sites and interstitial clusters were calculated by using an extensive search of all the possible combinations. We adopted and reported only the minimum energy configurations. The calculated DOS values are given in Fig. 2 with the Fermi level set at 0 eV.