Boron-Incorporating Silicon Nanocrystals Embedded in SiO2: Absence of Free Carriers vs. B-Induced Defects

Boron (B) doping of silicon nanocrystals requires the incorporation of a B-atom on a lattice site of the quantum dot and its ionization at room temperature. In case of successful B-doping the majority carriers (holes) should quench the photoluminescence of Si nanocrystals via non-radiative Auger recombination. In addition, the holes should allow for a non-transient electrical current. However, on the bottom end of the nanoscale, both substitutional incorporation and ionization are subject to significant increase in their respective energies due to confinement and size effects. Nevertheless, successful B-doping of Si nanocrystals was reported for certain structural conditions. Here, we investigate B-doping for small, well-dispersed Si nanocrystals with low and moderate B-concentrations. While small amounts of B-atoms are incorporated into these nanocrystals, they hardly affect their optical or electrical properties. If the B-concentration exceeds ~1 at%, the luminescence quantum yield is significantly quenched, whereas electrical measurements do not reveal free carriers. This observation suggests a photoluminescence quenching mechanism based on B-induced defect states. By means of density functional theory calculations, we prove that B creates multiple states in the bandgap of Si and SiO2. We conclude that non-percolated ultra-small Si nanocrystals cannot be efficiently B-doped.


Results and Discussion
Incorporation of B in Si nanocrystals. Si NCs embedded in SiO 2 are fabricated by annealing (1100 °C, 1 h) B-doped Si-rich oxide (SRO:B)/SiO 2 SLs deposited by plasma-enhanced chemical vapour deposition (PECVD). Small amounts of B 2 H 6 are added to the SiH 4 , O 2 , Ar gas mixture to introduce B in SRO, while the SiO 2 barrier layers are not doped. In order to determine the initial B-concentration, we use molecular Cs + secondary ion mass spectrometry (MCs + -SIMS) 23 . As shown in Fig. 1a we achieve B-concentrations in SRO in the range of 0.15 to 1.54 at%. Besides, the Si-and O-concentration is not completely unaffected by adding B 2 H 6 , though all SiO xstoichiometries remain in the x ≈ 1.0 ± 0.2 range, which is well below the percolation threshold. Hao et al. argued for their sputtered samples that a reduced Si-content occurs due to additional oxygen from the B 2 O 3 target 4 , which is certainly not the explanation for the PECVD process used here. However, the hydrogen from B 2 H 6 might influence the plasma chemistry and thereby the Si/O ratio. On the other hand, transient effects during the ultralow rate deposition (<1 Å/s) might occur that change the B-concentration. Figure 1b depicts the B-concentration in 50 nm SRO:B films that were not deposited in one step but from sequential thin films as used in the actual SRO:B/SiO 2 SLs. We observe a slightly lower B-concentration compared to the thick one-step film with maximum B 2 H 6 flux (1.54 at%, Fig. 1a): 1.24 at% for 25 × 2 nm, 1.34 at% for 15 × 3.5 nm, 1.38 at% for 10 × 5 nm. On average, the transient deposition effects reduce the B-concentration in SLs to ~86% compared to the thick one-step film. In the following, we will therefore refer to the nominal B-concentrations as [B min ] = 0.13 at%, [B med ] = 0.47 at%, and [B max ] = 1.32 at%. The second data set in Fig. 1b shows the B-concentration of the same SRO:B stack after annealing at 1100 °C. Both curves are hardly distinguishable, i.e., the MCs + -SIMS-measurement is not influenced by the structural changes the SRO undergoes during annealing and the B-concentration does not change during annealing. The latter is in agreement with the very low diffusion coefficient of B in SiO 2 4 . Whereas SIMS can quantify the B-concentration in the SRO from which the Si NCs are formed by annealing, we use atom probe tomography (APT) to detect individual B-atoms with sub-nm spatial resolution in the Si NC/ SiO 2 system. In Fig. 1c the proxigram (proximity histogram 24 ) of a Si NC SL sample with 4.5 nm SRO:B max using 60% Si iso-concentration surfaces is shown. In contrast to typical proxigrams for P-doped Si NCs 22,25,26 we do not observe a strong pile-up of dopants exactly at the Si/SiO 2 interface and a stable considerable level of dopants in the NC-volume. It was shown before for sputtered B-doped Si NC samples 22,25 that the proxigram exhibits a constant B-level in the SiO 2 matrix and a concentration-drop towards the inner NC volume. Here, we observe a slightly increasing B-concentration in the oxide matrix towards the NC-interface and a considerable B incorporation in the near-surface volume of the NC-interior, but almost no boron in the NC-cores. This behaviour is consistent with the segregation coefficient of ~0.3 for B in bulk-Si/SiO 2 27 , which is based on the lower solubility of B in Si than in SiO 2 . On the other hand, nanocrystals are subject to self-purification 28 , so that the reason for the evanescent B-concentrations in the NC-core might not only be related to insufficient solubility. In this picture, the small B-concentration increase in the Si/SiO 2 interface region could be the result of two counter-acting forces during annealing: Self-purification tries to eject B-atoms from the NC into the matrix but the solubility limit of B in SiO 2 is reached for values at around 1 at% and the only region left to deposit B is the interfacial SiO x transition shell 29,30 . Furthermore, it was theoretically calculated that B in silanol (OH) terminated Si NCs has an energetically most favourable site just a few Å under the surface 31 . For B in fully SiO 2 -embedded Si NCs theoretical results even suggest an energetical preference for real surface incorporation 32 . Besides, our samples represent a heterogeneous semiconductor-dielectric system with inherently different evaporation field strengths. At a given field the Si precipitates evaporate faster than the surrounding SiO 2 which causes local magnification effects (LME) 33, 34 that project some of the atoms in the exterior of the NCs towards the interior. Accordingly, we measure 15-20 at% of oxygen in the Si NC-volume. The B-concentration is also subject to this projection artefact, so that the true values in the near-surface region of the NC-volume will be lower than shown in Fig. 1c. The last factor that influences the measurement is the limited detection efficiency of APT (here 57%). To some extent, both effects compensate each other and in absence of any measurement technology with higher precision, we stick in the following to the as-measured data. In Fig. 1d the amount of B-atoms per NC as function of NC-size is shown. As observed before for P-dopants 35 there is a clear trend for larger NCs to incorporate more B-atoms, as indicated by the linear fit (dashed line). The inset of Fig. 1d shows the frequency of NCs that incorporate n B-atoms. Interestingly, more than 50% of the NCs in the sample grown from SRO:B max with 1.32 at% B do not contain a single B-atom. Out of Scientific RepoRts | 7: 8337 | DOI:10.1038/s41598-017-08814-0 the NCs that incorporate one or more B-atoms, ~90% have B-concentrations above the solubility limit, i.e., they are B-supersaturated, which increases the probability for interstitial rather than substitutional B-incorporation. In contrast, less than 20% of P-incorporating NCs exceed the solubility limit 35,36 .
Altogether, the APT analysis of the sample doped with the highest B 2 H 6 flux allows to conclude that only half of the NCs actually incorporate B-atoms and, according to the proxigram, the B-atoms are located very close to the surface of the NCs. This observation constitutes for boron a high probability to be not fully Si-coordinated but partially connected to O in the SiO x transition shell. A significantly higher exothermal formation energy of the B-O bond over the B-Si bond 37 corroborates this statement. By definition, B-atoms in this configuration cannot become dopants. Furthermore, a near-surface location of the dopant is the worst scenario in terms of dopant charge screening within the semiconductor, which gives rise to enhanced dielectric confinement and high ionization energies 38 .
Boron is known to reduce the viscosity of SiO 2 at high temperatures. Therefore, the structural integrity of the SL stacking order during annealing cannot be taken for granted. Energy-filtered transmission electron microscopy (EFTEM) cross-section images, as shown in Fig. 2, demonstrate that the SL structure is affected by high B-concentrations: The layer-wise arrangement of NCs diminishes for concentrations >1 at% B. This is related to the lower viscosity of SiO 2 :B during the annealing (1100 °C), i.e., the SiO 2 barrier layers which separate the NC-layers are softened by B. For comparison, borosilicate glasses (8-13% B 2 O 3 ) are well known to soften already at ~820 °C. It can be anticipated from the EFTEM images that B-concentrations ≫1 at% cannot be used without losing NC-size control mediated by the SL. Hence, the insufficient B-incorporation into NCs (cf. Fig. 1d) cannot be improved by adding increasing amounts of B, if at the same time small NC sizes and size distributions are to be maintained.

Luminescence of B-incorporating Si NCs.
The PL-intensity of Si NCs with boron is typically reported to be lower than for undoped reference samples, even for very small B-concentrations of ≤0.1 at% 1-5, 7, 8 . So far, only Puthen-Veettil et al. observed a ~40% PL-intensity increase for their lowest B-sputter target power, accompanied by a decreased dangling bond (DB) defect signal measured by electron spin resonance 6 . This behaviour is typical for phosphorus dopants (cf. ref. 39 and references therein) and generally attributed to DB-defect passivation by P. In the case of boron, this mechanism might be present as well.
In Fig. 3a the PL spectra of the Si NC samples with varying B-concentrations are shown together with the same sample set passivated at 450 °C for 1 h in pure H 2 . Boron does not significantly influence the PL peak positions and we attribute the differences, at least partially, to the small changes in the excess-Si content of the samples (cf. Fig. 1a). All H 2 -passivated samples are slightly red-shifted in their PL by 10-15 nm, which is related to the well-known preferential emission enhancement of larger NCs within the ensemble (larger surface area implicates Figure 2. Influence of B on the SiO 2 viscosity. EFTEM cross-section images of 1100 °C-annealed 10-bilayer 4-nm-SRO:B/2-nm-SiO 2 SLs, filtered around the Si-plasmon loss energy. The layered structure is well preserved for the lowest and the medium B-concentration but the NC-stacking order is less clearly pronounced for the highest B-concentration. Obviously, the threshold for SL-preservation and thereby for NC size-control is slightly above 1 at% B. higher probability to have a non-radiative Si/SiO 2 interface defect) [40][41][42] . As shown in the peak analysis in Fig. 3b, the PL-increase by H 2 is on average around 60%, irrespective of B-doping. It is evident on first sight, that small and medium B-concentrations have only little impact on the PL-intensity, while the PL of the B max -samples is quenched by a factor of 10. Like in ref. 6, we observe a 30-60% increase in PL-intensity for the lowest B-concentration B min . Interestingly, the B med -samples have almost the same PL-intensity as the undoped reference samples, i.e., for 0.47 at% B the PL-enhancing and the PL-quenching effect seem to compensate each other. This is a surprising result since a simple linear extrapolation of the APT data measured for the B max -sample reveals that the B med -sample contains at least ~1/3 B-incorporating NCs. In case the B-atoms do provide free carriers, a significant loss of PL-intensity should be visible when only ~2/3 PL-active NCs are left. In addition, the PL-peak should blue-shift since the largest NCs within each sample have the highest B-concentrations (cf. Fig. 1d). Instead, we observe very similar peaks for B med and the B-free reference, which is hardly explainable by active B-dopants. Furthermore, the sudden drop in PL-intensity when increasing the B-concentration from 0.47 to 1.32 at% (B med to B max ) indicates something like a B-threshold that is required to activate the B-induced PL-quenching -a parameter incompatible with conventional B-doping. We derived above (cf. inset of Fig. 1d) that roughly half of the NCs in the B max sample are B-free and therefore potentially PL-active. The circumstance that this ~50% fraction emits only ~10% of the luminescence compared to the undoped sample is a clear indication for a B-induced non-radiative defect centre. Another argument against Auger recombination with B-induced free charge carriers arises from the inadequately small PL-blueshift of the B max -samples that does not reflect the 1 order of magnitude intensity quenching. We note that the nature of the B-induced defect centre cannot be a simple DB-type defect since the PL-quenching is nearly unaffected by H 2 -passivation.
In order to exclude a major uncertainty of standard PL-spectroscopy, i.e., the sensitivity of the PL-intensity on the number of luminescent NCs and their absorption cross-section in each sample, we measured the luminescence quantum yield (QY) 43 . By measuring the ratio of emitted and absorbed photons, these ambiguities are ruled out and the true luminescence efficiency of each sample is revealed. The QY-data of H 2 -passivated NC samples as function of B-concentration is given in Fig. 4. The absolute QY value of up to ~43% for the undoped reference sample is (to our knowledge) the highest ever reported for matrix-embedded Si NCs. Recently, a record efficiency of 35% was published 44 , which is clearly outperformed here. Higher QY of 60-70% was only achieved for organically passivated, free-standing Si NCs 45 . Due to the low initial fraction of defective, dark NCs in our samples, we have a highly sensitive system to observe the influence of non-radiative centres introduced by B-doping. In direct comparison to Fig. 3, we see that the QY does not show the same trends like the PL-intensity. Neither the PL-intensity enhancing effect of 0.13 at% (B min ) nor the almost identical PL peaks of the reference and the 0.47 at% (B med ) samples (Fig. 3a, green and black lines) are reproduced in the QY. Only the strong quenching at 1.32 at% (B max ) is clearly visible. It is likely that structural effects of B on the NC formation or crystallization and the slightly different excess-Si concentrations are the reason for the PL-behaviour observed above. We note here that, generally, luminescence QY and not just PL-intensity should be analysed to draw conclusions about the impact of impurities on Si NC samples since this method circumvents disturbing structural side effects. Whereas no direct APT-data is available about B-incorporation in NCs grown from low or medium B-doped SRO, we can presume that the fraction of B-incorporating NCs of such samples correlates to the small loss of QY. Therefore, we cannot identify the origin of QY-loss (Auger recombination with B-induced free carriers vs. B-induced non-radiative defects). On the other hand, the ~50% B-free NCs in the B max -sample emit PL with only 18% of the efficiency compared to the intrinsic reference sample. This is a clear indication that at least partially a defect-related luminescence quenching is present. Electrical properties of B-incorporating Si NCs. Since the luminescence measurements done here cannot distinguish between non-radiative recombination with free carriers or defects, we investigate the current-voltage (I-V) properties of B-incorporating Si NCs in metal-oxide-semiconductor (MOS) capacitors. In the first place, only transient displacement currents are measured by means of NC-superlattice stacks sandwiched between 10 nm thick SiO 2 injection blocking layers 46 . At low electric fields (E) this blocking-MOS device allows for the observation of a current peak, if dopant atoms are ionized by the E-field and release free carriers that subsequently accumulate under the gate blocking oxide 17,35,46 . Figure 5a shows the current densities (J) of the B-incorporating Si NC samples together with the intrinsic NC sample and a B-doped SiO 2 (0.6 at%) thin film of equal thickness. All J-E curves are hardly distinguishable and there is no indication for a J-peak. The SiO 2 :B reference sample proves that the B-atoms in the silicon oxide matrix between the Si NCs does not contribute to any charge carrier generation. The slightly higher current density of the B med -sample might be again related to the marginally higher excess-Si content 47 .
In Fig. 5b the current density of a blocking-MOS capacitor as function of time at E = 0.2 MV/cm is plotted, i.e., at a typical E-field where P-dopants would cause up to two orders of magnitude higher J 35,46 . Here, we observe very similar J-t curves for all samples, irrespective of B-doping. After a few seconds, the transient displacement currents reach the noisy sub-100 pA cm −2 level where the internal charge redistribution and dielectric relaxation is completed. By integrating over time, the total amount of carriers is calculated (cf. refs. 17, 35 and 46), which typically reaches values on the order of magnitude of 10 17 cm −3 . As shown in the inset of Fig. 5b, our samples seem to have carrier densities in the 10 15 cm −3 range with, oddly enough, the B-free reference having the highest value. However, given the very short time scale (≤1 second) on which the I-t curves differ significantly before reaching the noise level, indicates that just dielectric relaxation rather than charge redistribution takes place. We therefore deduce that the carrier densities (Fig. 5b inset) are just an artefact caused by dielectric relaxation and can thus be considered as an upper limit.
Finally, B-incorporating Si NC superlattices in normal MOS-capacitors, i.e., without thick SiO 2 injection barriers, are investigated to find out whether the presence of B-dopants increases the charge carrier transport. The results for non-blocking MOS-capacitors in accumulation are shown in Fig. 5c for n-type and in Fig. 5d for p-type Si substrates. Apparently, neither increased electron-nor hole-conductivity is present for times ≥10 s, when the initial transient currents faded and steady-state conditions are reached. Once again this is in contrast to P-doped Si NCs 48 . We want to emphasise that little information can be derived from initial time interval at t < 1s where dielectric relaxation and internal charge redistribution takes place. Although the two samples with higher B-concentrations seem to have larger initial currents than the other two samples, there is no systematic trend with B-concentration and in particular no difference between electron and hole transport.
Given the absence of a J-peak (Fig. 5a) and enhanced current transport (Fig. 5c,d) for B-incorporating Si NC samples, we conclude that there is no significant amount of ionisable B-acceptors. Hence, there are no free carriers in the B-incorporating Si NCs from thermal ionization at room temperature. Furthermore, free carriers cannot be generated by field-ionization. The latter aspect deviates substantially from phosphorus, where the small fraction of NC-incorporated P-atoms that reside on a substitutional site does provide charge carriers, if ionized by an external electrical field 35 . Consequently, the few B-atoms incorporated in small (~4 nm) and non-percolated Si NCs are predominantly located on interstitial sites. We note that similar measurement on percolated NC-networks revealed a J-peak and thereby a small fraction of substitutional B-atoms is anticipated for larger nano-Si volumes 17 . As argued above, it is also possible that B-atoms in very small NCs resides very close to the surface, where they are not completely Si-coordinated or subject to a huge dielectric confinement 38 due to the adjacent SiO 2 matrix. The preferred near-surface position of B in Si NCs is in accord with APT-data ( Fig. 1c and  refs. 22 and 25) and theoretical calculations of H-terminated 9 , OH-terminated 31 and SiO 2 -embedded Si NCs 32 . Concerning the B-ionization energy, which we cannot derive from our data (no J-peak), the work of Lechner et al. indicates for B-concentrations of 3 × 10 16 cm −3 and 2 × 10 18 cm −3 in ~20 nm free-standing Si NCs (equivalent to ~0.1 and ~8 B-atoms per NC) values of 420 and 280 meV, respectively 49 . Thus, even Si NCs with sizes too large for significant quantum confinement have considerable B-ionization energies, which underlines the crucial role of dielectric confinement, or respectively, the influence of hole-trapping DB-defects, as argued by the same authors.
B-induced states in the Si NC/SiO 2 system. As shown above, the majority of B-atoms reside in the SiO 2 matrix or in the suboxide transition shell surrounding the NCs. This raises the question how boron in these configurations changes the electronic environment of the NCs. We use density functional theory (DFT) to calculate the density of states (DOS) of SiO 2 :B (Fig. 6a) and SiO 0.9 :B (Fig. 6b) together with the respective intrinsic approximants. The B-free oxides do not contain any states in or near the fundamental Si NC gap (as expected). However, for SiO 2 :B we observe a state that is only in the vicinity of the highest occupied states (HOS; equivalent to the valence band edge (VBE) in bulk) for very small NCs of 1.5 nm in diameter (Fig. 6a). For larger Si NCs this state is outside of the fundamental NC-gap. At the NC-interface, simulated here with SiO 0.9 :B, the unoccupied state shown in Fig. 6b is shifted to a position slightly above mid-gap for bulk and nanocrystalline Si. At this energetic position, it can efficiently capture the electron of a photo-excited exciton, which causes non-radiative recombination similar to a Si-DB defect.
For those B-atoms that are incorporated into the Si NC core, two configurations exist: interstitial and substitutional. In Fig. 6c the DOS of a 1.5 nm NC with an interstitial B-atom is shown to contain a B-induced occupied state near the VBE of bulk-Si. For any NC that is subject to bandgap widening by quantum confinement, this state is energetically located slightly above the HOS, so that its electron could immediately recombine with a photo-generated hole. Therefore, interstitial-B is also regarded as a PL-quenching centre. The DOS of the substitutional B-configuration is shown in Fig. 6d. There is an unoccupied state slightly above the HOS of a NC, which would make it a hole-generating acceptor state, if it is thermally ionisable and, moreover, if this configuration exists in sufficient concentrations. For the first condition, this state would have to be as close to the HOS as in the case of bulk-Si, i.e., ~45 meV. Otherwise, the exponential dependence of the ionization probability on the ionization energy decreases the free carrier density dramatically. The substitutional incorporation of a B-atom into the lattice of a small Si NC requires a rather high formation energies 31, 32, 50 of beyond ~1 eV. This energy has to be provided by the 1100 °C-annealing that forms the NCs, which is equivalent to a thermal energy of only ~0.12 eV. Hence, substitutional B is very unlikely to occur in small, non-percolated NCs, which is in accord with the absence of any electrical activity (cf. Fig. 5). Although differences between the DOS shown here for boron and that calculated for phosphorus in the same approximants 51 exist, the main conclusion is the same: Interstitial dopants and dopants in the suboxide transition shell are the most likely candidates to create defect states in the bandgap, while majority charge carrier generation from substitutional dopants is impeded by high ionization energies.
Whereas a number of B-induced states in the Si/SiO 2 system with energetic vicinity to the fundamental NC-gap are present, we cannot unambiguously identify which B-induced state is the dominant PL-quencher. However, contributions of Auger recombination with B-induced free-carriers appear to be virtually nil.
Scientific RepoRts | 7: 8337 | DOI:10.1038/s41598-017-08814-0 Conclusion B-atoms were shown to be incorporated in small (~4 nm) non-percolated Si NCs, though at a lower concentration and with lower probability than e.g. P-dopants. In addition, B prefers a near-surface location in Si NCs. PL-QY measurements reveal that B quenches the luminescence of NCs with a nearly linear dependence on the B-concentration. In contrast, no free-carriers are measured electrically and even field-ionization is not capable to generate significant amounts of carriers. In other words, B-doping has no measurable impact on the I-V and I-t behaviour of Si NC/SiO 2 MOS capacitors. We conclude that the vast majority of B-atoms are incorporated on interstitial lattice sites, which is supported by its high substitutional formation energy. The electrical inactivity of B in small NCs and the apparent absence of B-atoms on Si lattice sites render B-induced defect states to be the main origin of luminescence quenching. DFT results suggest that especially interstitial-B and B in the suboxide transition shell surrounding the NCs act as non-radiative centres.
The absence of successful B-doping, as shown here, is restricted to small and individual, i.e. non-percolated, Si NCs. It was shown in the literature, that percolated nano-Si networks do exhibit some electrical activity 17 , which supports the presence of some field-ionisable, substitutional B-atoms in such samples. However, considering Si NCs as a model system to study B-doping in the limit of few-nm Si-nanovolumes, spatially-isolated and mainly spherical structures are mandatory. Another restriction for this model system arises from the B-concentration, which has to remain within the doping-range (~1 at% at max) and must not exceed the semiconductor-metal transition threshold (quasi-metallic properties such as plasmon resonances) or the threshold for Si-B alloy formation. Specifically, for size-and shape-controlled Si NCs the fabrication via SiO x /SiO 2 superlattices is a convenient approach, unless exceeding B-concentrations deteriorate the stacking order via a reduction in viscosity of SiO 2 :B (as shown by EFTEM for ≥1.3 at% B). All these factors implicate that there is no technological solution to enable B to become an efficient impurity dopant for ultra-small Si nanovolumes. An alternative approach to facilitate p-type behaviour in such structures is the generation of Al-induced acceptor states in Si-adjacent SiO 2 that capture electrons from the Si valence band, which leaves holes as free carriers behind 52 .

Methods
Sample fabrication. Si NC/SiO 2 superlattices were fabricated by PECVD (using SiH 4 , O 2 and Ar 53 ) of alternating 4.5 nm SiO x≈1.0 and SiO 2 barrier layers on (100)-oriented Si wafers. For B-doping, 0.18-0.94 sccm of 10%-B 2 H 6 /SiH 4 were added, depending on the intended B-concentration. The samples for APT had 5 nm SiO 2 barriers and 30 bilayers, those for PL and PL-QY 4 nm barriers and 20 or 40 bilayers, respectively. All samples dedicated to I-V measurements had total thicknesses of ~100 nm with 2 nm SiO 2 barriers and 14 bilayers (injection-blocking MOS devices with 10 nm thick SiO 2 buffer and capping layers) or 17 bilayers (non-blocking MOS-capacitors), respectively. The substrates of the I-V samples were either n-type (P, 1-30 Ω cm) or p-type (B, 1-30 Ω cm). After deposition, all samples were annealed in a quartz glass tube furnace at 1100 °C for 1 h in high-purity N 2 and subsequently defect passivated at 450 °C for 1 h in pure H 2 . In order to fabricate MOS-capacitors for electrical characterization, aluminium contacts were thermally evaporated and photolithographically structured.
Sample characterization. MCs + -SIMS was measured using a Cameca IMS-4f with 3 keV Cs + ions. APT was measured with a Cameca-LEAP ™ 4000X Si with a pulsed UV laser (355 nm, 100 pJ, 250 kHz), a cooled specimen holder (~40 K) and a chamber pressure of 10 −12 -10 −11 Torr. For data reconstruction IVAS ™ (version 3.6.6) was used. APT specimen (i.e., needle-shaped tips attached onto the apex of a Mo support grid) were fabricated using a Zeiss Auriga focused ion beam scanning electron microscope (FIB-SEM). For EFTEM imaging a JEOL 2010 operated at 200 kV and equipped with a Gatan imaging filter (GIF-863 Tridiem) was used. The energy filter was set to 16 eV (Si plasmon loss) with a 3.5 eV window. PL was carried out using a LN 2 -cooled CCD camera attached to a single grating monochromator with excitation by a HeCd laser (325 nm, ~3 mW/cm²). The luminescence QY was determined on samples deposited on quartz glass and a setup based on an integrating sphere, for details see ref. 43. I-V and I-t were measured in accumulation using an Agilent B1500A semiconductor device analyser. The MOS-capacitors were contacted by W-needles in a Cascade M150 Prober located in a shielded dark box.

Hybrid density functional theory (h-DFT) calculations.
Calculations were carried out in real space with a molecular orbital basis set (MO-BS) and the h-DFT method described below, employing the Gaussian03 and Gaussian09 program packages 54,55 . Initially, the MO-BS wavefunction ensemble was tested and optimized for stability with respect to describing the energy minimum of the system (variational principle; stable = opt) with the B3LYP hybrid DF method 56, 57 using a 6-31 G(d) MO-BS 58 (B3LYP/6-31 G(d)). This MO wavefunction ensemble was then used for the structural optimisation of the approximant to arrive at its most stable configuration (maximum integral over all bond energies), again following the B3LYP/6-31 G(d) route. Using these optimized geometries, their electronic structure was calculated again by testing and optimizing the MO-BS wavefunction ensemble with the B3LYP/6-31 G(d) route. Root mean square (RMS) and peak force convergence limits were 8 meV/Å and 12 meV/Å, respectively. Tight convergence criteria were applied to the self-consistent field routine. During all calculations, no symmetry constraints were applied to MOs. Further accuracy evaluations can be found elsewhere 59,60 . Electronic DOS were calculated from MO eigenenergies, applying a Gaussian broadening of 0.2 eV.