Structural and optical characterization of nanoalloys mixing gold or silver with aluminium or indium: evolution under various reactive environments †

In this study, the atomic and chemical structure and the optical response of A x B 1 (cid:1) x bimetallic nanoparticles (BNPs) combining gold or silver (A) with aluminium or indium (B) were investigated at various stoichiometries in order to examine if stable alloyed phases could exist and promote the emergence of localized surface plasmon resonance (LSPR) in the UV range. The structure and morphology of BNPs of a few nanometres, produced by laser vaporization, were analysed by transmission electron microscopy (TEM) and optical absorption measurements were performed on matrix-embedded BNPs. Information about the oxidation state of the BNPs can be inferred from a comparison between experimental optical spectra and Mie calculations in the dipolar approximation. The BNPs ’ internal structures were further investigated by additional characterization techniques. Firstly, in situ X-ray photoelectron spectroscopy provided information about the chemical state of the constituent elements and their evolution with time. Secondly, synchrotron-based X-ray scattering techniques were performed on Ag – Al BNPs in a wide-angle con ﬁ guration under grazing incidence, giving complementary information about structural and morphological heterogeneities in the BNPs. Finally, the restructuring of the partially oxidized Au 0.33 Al 0.67 BNPs annealed in a reducing atmosphere was also attempted by environmental TEM. The complementary techniques of characterization show that silver-based Ag – In and Ag – Al BNPs form metallic silver-rich alloyed cores surrounded by an indium or aluminium oxide shell. The initial LSPR is in the UV range for both systems, but the di ﬀ erence in the kinetics of oxidation between indium and aluminium involves less blue-shifted LSPR for Ag – Al BNPs. In the case of gold-based BNPs, we show evidence of ordered nanoalloys just after air exposure and the appearance of gold and indium (or aluminium)

more detail, either in the bulk phase [18][19][20][21][22] or at the nanoscale. [23][24][25][26] For instance, AuIn 2 nanoalloys were obtained by Boyen et al. by the atomic evaporation of indium on self-assembled naked gold NPs of a rather uniform size, ranging from 0.8 to 2.9 nm. 23 In situ X-ray photoelectron spectroscopy showed a shi in the Au-4f photoelectron spectra aer indium atomic deposition, except for 1.4 nm, corresponding to the very stable cluster Au 55 , leading the authors to conclude that alloying into AuIn 2 had occurred for the other sizes. The oxidation of Au-In alloy nanoparticles of 2 to 20 nm in diameter was also investigated by Sutter et al. 24 Bimetallic Au-In alloy nanoparticles were formed by the sequential room temperature evaporation of controlled amounts of indium and gold, before being exposed to air. Instead of a crystalline In 2 O 3 outer shell, as previously observed in Ag-In BNPs, 15 such oxidation leads to an amorphous shell of mixed Au-In oxide. This shell remains stable at high temperatures and the surface, enriched with Au, is capable of adsorbing CO and O 2 to convert them to CO 2 . The characterization of low-Au and high-Au content BNPs showed that Au-In cores consist of the ordered alloyed structures AuIn 2 and Au 3 In, respectively. The LSPR of Au-In BNPs was also observed at intermediate states during the formation process of oxidized Au-In BNPs. 25 Au-In alloy BNPs were rst obtained by combining gold NPs with indium and an LSPR emerges as a shoulder in the UV range below 350 nm. The BNPs then transform into intermetallic AuIn 2 cores surrounded by a shell of amorphous indium oxide (AuIn 2 @InO x ), before forming segregated Au-In 2 O 3 heterodimers. The corresponding LSPR is then red-shied because of indium oxidation.
The existence of stable ordered alloyed phases in Au-Al and Au-In BNPs, with encouraging results regarding the optical properties of AuIn 2 , led us to rst investigate BNPs of the following compositions, Au 0.50 Al 0.50 , Au 0.33 Al 0.67 , and Au 0. 33 In 0.67 , in the hope of retaining the ordered alloyed phase at the nanoscale. For Ag x -Al 1Àx and Ag x -In 1Àx BNPs, three compositions (x ¼ 0.25, 0.5 and 0.75) were studied. The rst results we obtained for these last systems are detailed in ref. 16 and complementary new results are reported in this manuscript, in particular results of in situ X-ray photoelectron spectroscopy and synchrotronbased X-ray scattering experiments. Indeed, the analysis of the BNP structure prior to air exposure was not possible on our setup until recently. Fortunately, it is now possible to perform in situ X-ray photoelectron spectroscopy (XPS) and analyze the chemical state of the metal atoms within the BNPs just aer deposition in a chamber maintained under ultra-high vacuum. As aluminium oxidizes more rapidly than indium, we rst chose to characterize Al-based BNPs (Ag-Al and Au-Al). On the other hand, as high-resolution transmission electron microscopy (HR-TEM) characterizations performed on Al-based BNPs do not allow pure metallic phases to be distinguished from alloyed phases (Au, Ag and Al have very close lattice parameters), synchrotron-based X-ray scattering spectroscopy has been performed on Al-based BNPs. Ag-Al BNPs were thus characterized by the grazing incidence wide-angle X-ray scattering technique to determine whether alloyed phases exist within the BNPs. We also present the results of TEM and optical characterizations of the four systems (Ag-In, Ag-Al, Au-In and Au-In), by more specically focusing on Au-Al and Au-In BNPs and referring when necessary to previous results published in ref. 16 for Ag-In and Ag-Al BNPs.
B. General considerations on systems mixing gold or silver with aluminium or indium Let us rst consider the two silver-based systems: Ag-In and Ag-Al. Silver and aluminum crystallize in a face-centered cubic structure (FCC) with very close lattice parameters (4.09Å and 4.05Å, respectively), while indium crystallizes in a tetragonal centred structure (TC) with lattice parameters of 3.24Å and 4.95Å. Therefore, silver and indium can be clearly identied by HR-TEM, whereas silver and aluminium are not distinguishable. Their phase diagrams show that both Ag-In and Ag-Al systems exhibit strong miscibility, especially for high proportions of silver, for which substitute alloys (intermetallic compounds) are observed. In the case of Ag-In, the alloyed phases can be distinguished from pure silver and indium phases. For Ag-Al, the substitute phases retain an FCC structure and the alloyed phase is easily observed by HR-TEM; the z phase, with a stoichiometry close to Ag 2 Al, displays lattice distances d hkl close to those of pure silver and pure aluminium. The other one (m phase) exhibits very low diffracted intensities, making it difficult to identify by HR-TEM.
Concerning the gold-based systems, Au-In and Au-Al, we can underline that the crystallographic properties of gold and silver are similar, since gold also displays an FCC structure, with a lattice parameter of 4.08Å. Therefore, gold and aluminium cannot be distinguished by HR-TEM either. Both mixed systems Au-In and Au-Al are miscible and their phase diagrams are similar to those of Ag-In and Ag-Al, respectively. Many dened compounds and various alloyed phases exist, especially at high gold concentrations, similar to Ag-In or Ag-Al. Table 1 shows the crystallographic characteristics of some of these alloys (a similar table is given in ref. 16 for In, Ag and Al). A remarkable feature of noble metal/trivalent metal systems involving gold is the existence of an equi-atomic compound (AuAl and AuIn, but also AuGa). Worthy of note are the dened compounds AuIn 2 and AuAl 2 , known as "blue gold" and "purple gold", respectively, which both crystallize as metallic salts (CaF 2 structure), which can be explained by the stronger electronegativity of gold compared to silver. This structure makes them very brittle. In micro-electronics, purple gold may form in junctions and is known as the "purple plague", because of its brittle nature. 27 From a crystallographic point of view, the AuIn 2 and AuAl 2 alloys have characteristic interplanar distances d 111 ¼ 3.46Å for AuAl 2 and d 111 ¼ 3.75Å and d 200 ¼ 3.25Å for AuIn 2 , which should allow the unambiguous identication of these materials if they are present and well oriented in HR-TEM. For Au-In, there are several other intermetallic compounds, denoted as b, 3 and g, having hexagonal, orthorhombic and cubic structures, respectively, with indium atomic compositions of 20%, 25% and 31%, respectively.
The comparison of the surface energies of each component, as well as lattice mismatches, also yields valuable information about the possible alloyed structures at the nanoscale. Firstly, the difference in the surface energies (1.24 J m À2 , 1.50 J m À2 , 0.70 J m À2 and 1.14 J m À2 for silver, gold, indium and aluminium, respectively) are in favour of indium and, to a lesser extent, aluminium surface enrichment in nanoalloys mixing gold or silver with aluminium or indium. Secondly, the important lattice mismatch between silver (or gold) and indium must also favor indium surface enrichment. And since the surface energy of gold Table 1 Data for In, Au, Al, Au-In and Au-Al (Wigner-Seitz radii (r s ), surface energies (E s ), LSPR wavelengths (l LSPR ) and energies (E LSPR ) of a sphere in a porous silica matrix (3 m z 2.17)) in the dipolar approximation and crystallographic properties of pure metals, Au-In and Au-Al alloys and oxides. All the corresponding references are indicated in the table. BCT, HCP, SC, FCC stand for body-centred tetragonal, hexagonal-closed packed, simple cubic and face-centred cubic lattices, respectively is larger than that of silver, the aluminium surface enrichment in BNPs should be more pronounced in the case of Au-Al compared to Ag-Al. Concerning BNP reactivity towards oxidation, aluminium, indium and, to a lesser extent, silver oxidize upon air exposure and spontaneously form an oxide shell around metallic cores in NPs due to the lower surface energies of the oxides compared to those of the metals. Aluminium is the fastest oxidizing material, while silver is the slowest. Aluminium oxidation has been studied at the nanoscale, showing the formation of a passivating layer which can be of the same order as the size of the BNPs we investigated. 37

C. Experimental methods
(1) Sample synthesis The A x B 1Àx (A ¼ Ag or Au, B ¼ In or Al) BNPs of a few nanometres in diameter were produced as a beam from an alloyed target of a given composition in a high vacuum chamber with a laser vaporisation source. The laser beam of a pulsed frequency doubled Nd:YAG laser (20 Hz, 532 nm, 10 ns pulse duration) was focused at the surface of a movable target (3 mm diameter rod) and created a hot plasma quenched by a continuous helium carrier gas ow in a 2 cm 3 cylindrical chamber (about 20 mbar inner pressure), inducing the growth of BNPs of various sizes. The BNP-gas mixture was ejected in a secondary vacuum chamber (5 Â 10 À7 mbar residual pressure) through a 0.6 mm diameter conical nozzle in the form of a supersonic beam. The emerging charged BNP beam could be deected with a static quadrupole deviator and the BNPs were thus size-selected before deposition in a high or ultra-high vacuum chamber (see details in ref. 38 and 39). Their average size could be determined by time-of-ight mass spectrometry (TOF-MS) with an uncertainty of AE0.1 nm. The BNPs could be deposited or co-deposited simultaneously with a transparent amorphous silica matrix (200-300 nm thickness) evaporated by an electron gun on various substrates (TEM grids, nano-chips from the company DENSsolutions for environmental TEM, silicon for GIWAXS and XPS analysis or a fused silica slide for optical measurements). Co-deposition was performed by ensuring that BNPs remained in a low BNP concentration in the matrix, thus preventing coalescence and allowing a comparison of their optical absorption with calculations using Mie theory in the dipolar approximation. For GIWAXS, the size-selected BNPs were deposited under ultra-high vacuum (UHV) and then protected by a thin layer of amorphous carbon (a few nanometres) before exposure to air. For in situ XPS measurements, the samples, composed of uncoated BNPs deposited on an ultra-thin layer of carbon over silicon under ultra-high vacuum (between 5 Â 10 À10 and 10 À9 mbar), were translated in another ultra-high vacuum chamber for analysis. The XPS measurements were launched a few tens of minutes aer BNP fabrication.
Transmission electron microscopy (TEM) experiments were performed, either with a FEI TITAN environmental (80-300 kV) transmission electron microscope operating at 300 kV in high vacuum and corrected for spherical aberrations of the objective lens (Cs corrector) or with a JEOL 2100F transmission electron microscope operating at 200 kV. The environmental mode was operated with the FEI TITAN environmental TEM microscope by direct gas injection (up to 20 mbar) near the sample. The substrates consisted in this last case of a thin lm of amorphous carbon or Si 3 N 4 deposited on a nano-chip from DENSsolutions specically designed for the wildre sample holder, allowing very rapid heating up to 1300 C. Energy dispersive X-ray spectroscopy (XEDS, also abbreviated as EDX) was performed using an Oxford Instruments X-Max 80 mm 2 SDD detector and the relative atomic percentages were estimated using the acquisition soware's standardless quantication method.
In situ X-ray photoelectron spectroscopy (XPS) was performed with a VG CLAM IV spectrometer using Al K a radiation for AgAl and Mg K a radiation for AuAl 2 . Wide scans and high-resolution spectra were collected with a constant pass energy of 50 and 20 eV and a step size of 1 and 0.1 eV, respectively. More details are given in the ESI. † Grazing incidence wide angle X-ray spectroscopy (GIWAXS) measurements were carried out at the European Synchrotron Radiation Facility (ESRF) in Grenoble, France at the BM02-D2AM beamline. A photon energy of 11.91 keV was used with an angle of incidence xed at 0.11 , which was able to reduce/optimize the analysis depth to a few nanometres without uorescence emission. GIWAXS signals were collected with a WOS-XPAD 2D-detector, calibrated in terms of detection uniformity, spatially and energetically, and allowing a q reciprocal space vector step of detection of 0.005 nm À1 . The 1D experimental scattered intensity was extracted as a function of the in-plane angle with respect to the substrate surface (dening the momentum transfer q). 40,41 Then, the spectrum was compared with a simulated pattern obtained from a calculated model cluster on the basis of the Debye equation 40,42 and considering the set-up geometry using a dedicated home-made soware. Preliminarily, a reference pattern from a sample without particles but with a cover-layer on the substrate region was measured in the same conditions, and was subtracted from the deposited nanoparticle sample pattern. The model clusters were obtained from geometrical construction with several levels of disorder of Ag-Al atom positions or Monte Carlo (MC) simulations for Ag only, using a semi-empirical tight-binding potential. A weighted sum of the intensities from several sizes or structures was used to t the WAXS patterns considering the size distribution coming from TEM results or TOF-MS. 40,41 (2) Optical spectroscopy The optical transmission spectra of silica-embedded BNPs deposited on fused silica were measured with a double beam PerkinElmer spectrophotometer using linearly polarized light (transverse magnetic). The sample was mounted at Brewster incidence in order to prevent Fabry-Perot interferences within the sample lm. The silica matrix was shown to be slightly porous and its dielectric function was previously measured by ellipsometry; its average value was 3 m ¼ 2.17, corresponding to an average optical index n m ¼ 1.47. The experimental values of 3 m (l) are used in the following for calculating the absorption spectra.

D. Results and discussion
(1) TEM characterization of BNPs mixing gold or silver with aluminium or indium The BNPs, deposited with or without the silica matrix on ultrathin amorphous carbon aer size selection, were rst characterized by transmission electron microscopy techniques.
(a) Ag x In 1Àx and Ag x Al 1Àx BNPs. Concerning the Ag-In system, the TEM analyses of Ag x In 1Àx BNPs for x ¼ 0.25 and 0.50 showed that they adopt, aer air exposure, a core@shell (Ag-In@In 2 O 3 ) structure consisting of a silver-rich alloyed core (z phase, Ag 2 In or a-Ag FCC structure) surrounded by an indium oxide In 2 O 3 shell. For silver-rich BNPs (x ¼ 0.75), one instead observes a monophased silverrich alloy. 16 The STEM image of an assembly of Ag 0.75 In 0.25 BNPs (x ¼ 0.75) is shown in Fig. 1b, for which no shell is visible, neither in the HR-TEM nor HR-STEM mode. As an illustration, an HR-TEM image of a decahedral BNP is displayed in the inset of Fig. 1b, where the presence of an indium oxide shell is not obvious. If the BNPs are slightly oxidized for x ¼ 0.75, the shell thickness is probably too small to be observed by electron microscopy. Ag x Al 1Àx BNPs (x ¼ 0.25, 0.50, 0.75) display the same tendency, with core@shell structures for higher proportions of aluminium (x ¼ 0.50 and 0.25) consisting a priori of an aluminium oxide shell. The structure of the core cannot be determined by HRTEM because both constituents have very similar lattice parameters. 16 However, aluminum does not seem to be entirely incorporated in the shells, since the cores are larger than they would have been if they were exclusively made of silver. Indeed, the comparison between the size of the cores (deduced from the size measured by mass spectrometry) and the size distribution of the cores deduced from STEM image analysis suggests that they still contain some aluminium. 16 For larger proportions of silver (Ag 0.75 Al 0.25 ), the BNPs do not exhibit any aluminium oxide shell (see the STEM image of an assembly of BNPs in Fig. 1a), and the average size deduced from STEM image analysis is in agreement with the size measured by mass spectrometry. 16 There is no evidence of any aluminium oxide shell in the HRTEM image either (inset of Fig. 1a), and families of planes (hkl) of the BNP can correspond to an FCC structure of aluminium or silver or a mixing of silver and aluminium.
We can conclude from these TEM observations that both Ag x In 1Àx and Ag x Al 1Àx contain a substantial amount of indium (or aluminium) in the metallic state. For Ag x In 1Àx BNPs, silver-rich alloyed phases are visible, while the type of alloys (intermetallic, ordered phases or segregated alloys) in Ag x Al 1Àx BNPs cannot be inferred from TEM analysis. Additional characterization techniques are required to conclude how silver and aluminium are alloyed within the BNPs (see Part D.2.b).
(b) Au-In x and Au x Al 1Àx BNPs (2 < x < 3 for Au-In x and x ¼ 0.5 or 0.33 for Au x Al 1Àx ). We study Au-Al and Au-In BNPs of compositions corresponding to stable ordered phases in the bulk, namely purple gold AuAl 2 , AuAl and blue gold AuIn 2 produced with the laser vaporization source from a target of the corresponding compositions (Au 0.33 Al 0.67 , Au 0.5 Al 0.5 and Au 0.33 In 0.67 from Goodfellow Inc.). In the case of Au 0.33 In 0.67 , there is doubt regarding the real composition of the target as the rod does not display the expected blue colour and EDX analyses performed on single BNPs and assemblies of BNPs give an average composition of Au 0.25 In 0.75 , in disagreement with the composition given by the manufacturer. Therefore the composition for Au-In BNPs will be noted hereaer as AuIn x , with 2 < x < 3. Fig. 1c and d display STEM images of assemblies of size-selected Au 0.33 Al 0.67 and AuIn x co-deposited in a silica matrix. Similar observations were also realised on Au 0.5 Al 0.5 BNPs (see Fig. SI1 †). In all cases, we observe core@shell structures with a strong contrast, suggesting the formation of gold-rich cores with shells composed of oxides (Al 2 O 3 or In 2 O 3 ). In the case of Au 0.5 Al 0.5 BNPs, the average core diameter, deduced from STEM image analysis (see the size distribution in Fig. SI1 †) is larger than the core diameter deduced from TOF-MS and assuming a pure gold core (3.3 nm from STEM vs. 3.0 nm from mass spectrometry, see Fig. SI1 †). This may be a sign that, as observed in Ag-Al BNPs, some of the aluminium is metallic inside the cores. Nevertheless, as in the case of Ag-Al BNPs, it is not possible to know how aluminium is alloyed with gold (alloying at the atomic level or segregated). HR-STEM analysis does not allow the discrimination between pure gold or alloyed phases in the cores either. Indeed, the HR-STEM or HR-TEM images, such as the one shown in the le inset of Fig. 1c, display FFC phases with a lattice parameter corresponding to gold or aluminium. We also performed EDX analysis on a single BNP (right inset in Fig. 1c). The averaging of the spectra associated with the pixels of the shell indicates that the shell is exclusively composed of aluminium (or alumina). At the core level, we nd both peaks associated with gold and aluminium, but note that this signal comprises counts from the core and the shell below and above the core. This result by itself thus cannot conrm that some aluminium is alloyed with gold at the atomic level inside the cores. For AuIn x BNPs, the core@shell structure is more visible in HR-TEM (see the inset in Fig. 1d) because of the higher density of In 2 O 3 compared to that of Al 2 O 3 , but we have not yet been able to identify alloyed Au-In structures within the cores. Additional experiments are therefore required.
To conclude on gold-based Au-Al and Au-In BNPs, the TEM analyses show that the BNPs rapidly oxidized upon exposure to air, but we can reasonably assume that a part of aluminium (or indium) remains in the metallic state within the BNPs. However, no ordered alloyed phases were identied, probably because of demixing during oxidation. In order to further explore how both metals mix and if the BNPs are oxidized before air exposure, additional characterization techniques have been employed.
(2) Complementary characterizations (a) In situ XPS spectroscopy on Ag-Al and Au-Al BNPs. In situ X-ray photoelectron spectroscopy was performed in order to infer the oxidation state of the as-prepared BNPs deposited under ultra-high vacuum (UHV). As aluminium is more sensitive to oxidation than indium, XPS measurements were rst performed on aluminium-based BNPs (Ag 0.5 Al 0.5 and Au 0.33 Al 0.67 ) maintained under ultra-high vacuum. The spectra for the Al 2p level display a broad shouldered peak resulting from two underlying peaks at 73.1 eV and 75 eV, which correspond to metallic and oxidized aluminium, respectively (see Fig. SI3 and SI4 †). The analysis of their relative intensities allows the proportion of metallic aluminium compared to that of oxidized aluminium to be determined. For Au 0.33 Al 0.67 BNPs, the proportion of metallic aluminium is 83% twenty-ve minutes aer deposition (see Fig. SI3 †). This means that the BNPs initially contain a high proportion of metallic aluminium, but it is not possible to know if aluminium and gold are alloyed at the atomic scale or segregated with the BNPs. Aer ve hours and despite the UHV conditions (5 Â 10 À10 mbar), the proportion of metallic aluminium decreased to 67%, highlighting the high reactivity of aluminium at the nanoscale. In the case of Ag 0.5 Al 0.5 BNPs, the proportion of metallic aluminium is a little bit lower (55%) and decreased to 32% aer 9 hours under UHV conditions (see Fig. SI4 †). The 3d 5/2 peak of silver corresponds to metallic silver and does not show any sign of an oxidationinduced shi over the same period (see Fig. SI5 †).
XPS measurements lead us to conclude that the initially prepared Au-Al BNPs and Ag-Al BNPs contain metallic aluminium, but oxidize over time, even under UHV conditions, highlighting the high sensitivity of aluminium to a reactive environment. This shows that matrix protection of the BNPs is crucial to prevent or at least to slow down aluminium oxidation. In the near future, XPS measurements will also be performed on indium-based BNPs (Ag-In and Au-In), but since indium oxidizes slower than aluminium, the initial proportion of metallic indium within BNPs is expected to be higher than in Au-Al or Ag-Al BNPs.
(b) Grazing incidence wide angle X-ray scattering (GIWAXS) on Ag-Al BNPs. Grazing incidence wide-angle X-ray scattering was performed ex situ on sizeselected Ag-Al BNPs deposited under UHV and then protected by a thin layer (several nanometres) of amorphous carbon before exposure to air. Fig. 2a-d show four samples of size-selected Ag-based BNPs of a similar initial average size range (between 5 to 6.5 nm). For the pure Ag sample, the experimental spectrum is in very good agreement, with a distribution of crystalline twinned structures and/or decahedral multiple-twinned structures. [42][43][44] The lattice parameter is close to the bulk value without any contraction due to size reduction. It then decreases as the content of Al increases in Ag x Al 1Àx BNPs (x ¼ 0.25, 0.50, 0.75), as shown in Fig. 2f. This decrease is higher than a simple variation due to the Al incorporation in the lattice following Vegard's law, as shown in Fig. 2f. For example, the measured contraction is 2% for the Ag 0.25 Al 0.75 sample, while Vegard's law predicts only 0.7%. In addition, as the Al content increases, the spectrum exhibits higher structural disorder (broader peaks) and additional contributions in the range 40-50 nm À1 (see the black arrows in Fig. 2) which do not correspond to a metallic cubic phase, but can be assigned to aluminium oxide. Those peaks are all the more important as the proportion of aluminium is high, showing that the carbon coating does not completely protect the BNPs from oxidation. Furthermore, the behavior, showing a contraction of the atomic nearest-neighbour distances, is in agreement with a compressive stress applied on the BNPs due to the formation of an aluminium oxide shell. In a conned system of metallic BNPs in a carbon matrix, as the Al content increases, the oxide shell volume increases (i.e. a volume expansion with respect to the Al metallic volume) and the metallic core volume decreases, which induces a higher level of stress.
Finally, Fig. 2e shows the GIWAXS spectrum of the Ag 0.75 Al 0.25 sample aer annealing. The lattice parameter is close to that of the pure silver NPs (a ¼ 4.09Å), which is the signature of a stress relaxation and a total segregation of Al atoms, but not in a shell conguration.
(c) Environmental TEM. The structural evolution of oxidized silicaembedded Au 0.33 Al 0.67 BNPs under varying H 2 partial pressures and substrate temperatures was investigated in real-time by environmental transmission microscopy (E-TEM). The objective was to reduce aluminium oxide, as previously done for indium oxide, where oxidized Ag 0.25 In 0.75 BNPs were found to be transformed into silver-rich single-phased Ag-In BNPs. 15 Size-selected BNPs of 4.2 nm in diameter were co-deposited in a silica matrix, identical to the samples used for optical spectroscopy (see below). In the environmental TEM mode, the H 2 Fig. 3 Sequence of TEM images extracted from the real-time tracking of two Au 0.33 Al 0.67 BNPs co-deposited in a silica matrix at 1100 C under 10 mbar of H 2 . (a) Initially, at time t 0 , the BNPs denoted 1 and 2 are in a core-shell and Janus configuration, respectively. After 7 seconds (b), BNP 1 changes from a core@shell structure to a Janus structure, while the core of BNP 2 begins to disappear. After one minute (c), both cores have melted and diffused. Worthy of note is that only the BNPs under the electron beam undergo a structural change.
pressure and the temperature can be independently controlled within a few seconds.
The starting point is 2 mbar of H 2 at room temperature. At 200 C, with this pressure, no change is observed. The pressure of H 2 is then increased to 10 mbar. At 250 C, core@shell to Janus transitions are observed when the BNPs are exposed to the electron beam. Everywhere else, the BNPs remain core@shell. The temperature is then increased by steps of 50 C. At 650 C, the BNPs become more and more spherical. The temperature is increased up to 1100 C without observing any reduction of the alumina shell. However, at this temperature, coreshell to Janus transitions, followed by the melting and diffusion of gold, are apparent, as can be seen in Fig. 3. Again, this effect is due to the electron beam. Elsewhere, the BNPs remain in a core@shell conguration.
Finally, the reduction of the aluminium oxide was not observed even at a very high temperature. Aluminium oxide, known to be very stable in the bulk (see Ellingham diagrams) seems to be also stable at the nanoscale and extremely difficult to reduce. We observe here the striking difference in behaviour between aluminium oxide and indium oxide, the latter being much easier to reduce. 15 (3) Optical studies (a) Optical spectra of BNPs mixing gold or silver with aluminium or indium just aer air exposure. The optical absorption spectra of BNPs embedded in a silica matrix, presented in Fig. 4, were recorded between 10 and 30 minutes aer exposure to air.
Silver-based Ag-In and Ag-Al BNPs. Non-size-selected Ag x In 1Àx BNPs of about 4 nm in diameter display broad LSPR in between those of pure indium and silver NPs (Fig. 4a). They are further in the UV range the higher the proportion of indium (from 401 nm down to 312 nm from x ¼ 0.75 to x ¼ 0.25). Comparison with theoretical simulations (Mie theory in the dipolar approximation) shows that these LSPR spectral positions are consistent with the existence of some metallic indium within the BNPs (i.e. BNPs consisting of silver-rich indium cores surrounded by an indium oxide shell). 16 This suggests that the partially oxidized BNPs retain metallic indium for the three compositions, but even though the silica matrix effectively slows down the oxidation, it does not inhibit it entirely. Moreover, the silica-embedded BNPs appeared not to be stable with prolonged air-exposure and the LSPR shis towards lower energies because of gradual indium oxidation. 16 Notably, in the case of Ag 0.75 In 0.25 BNPs, if any shell of indium oxide could be observed by TEM aer air-exposure, the evolution of the optical spectra showed evidence of a progressive indium oxidation (broadening and damping), with the indium shell probably being too thin to be observed by TEM.
Size-selected Ag x Al 1Àx BNPs embedded in a silica matrix show better dened LSPR compared to that of Ag x In 1Àx BNPs, which may be related to the size selection (spectra in Fig. 4b). Nevertheless, the LSPR is less blue-shied compared to the initial ones of Ag x In 1Àx BNPs, the maximum ranging from 414 nm down to 382 nm for a decreasing proportion of silver (from x ¼ 0.75 to x ¼ 0.25). Comparison with theoretical calculations permits us to attribute this reduced blue-shi for Ag x Al 1Àx to a higher degree of oxidation from the rst absorption measurements. 16 This means that the proportion of metallic aluminium is probably lower than that of indium just aer air-exposure, and that the oxidation of aluminium is faster than that of indium. Nevertheless, the absorption spectra do not evolve signicantly with time; the BNPs, aer being rapidly partially oxidized, are probably protected by the surrounding alumina shell. 16 So nally, both Ag-In and Ag-Al BNPs, with prolonged air-exposure, are certainly oxidized in a similar proportion, but their cores still contain metallic indium or aluminium.
Gold-based Au-In and Au-Al BNPs. The optical absorption spectra of sizeselected Au 0.33 Al 0.67 , Au 0.50 Al 0.50 and AuIn x (2 < x < 3) silica-embedded BNPs are displayed in Fig. 4c aer a few tens of minutes of exposure to air. The silicaembedded AuIn x (2 < x < 3) BNPs show an LSPR in the UV range (at about 254 nm), with the absorption increasing towards the UV range, similar to the one experimentally measured on slightly oxidized AuIn 2 BNPs. 25 For Au 0.33 Al 0.67 BNPs, we observe only an increase of the absorption towards the UV range, but no clear resonance can be discerned, except for a shoulder at 210 nm, while an LSPR at around 511 nm emerges for Au 0.50 Al 0.50 BNPs. These initial experimental spectra are compared with simulated ones in Fig. 5, in which the dielectric functions of Au, In, Al, In 2 O 3 and Al 2 O 3 are taken from measurements in the bulk phase. [45][46][47][48][49] In the case of AuIn x BNPs, the experimental absorption spectrum is in relatively good agreement with the optical absorption of a core@shell sphere AuIn 2 @In 2 O 3 (overall composition: Au 0.25 In 0.75 ) made of an alloyed core of AuIn 2 surrounded by a shell of indium oxide (Fig. 5c and d). In these calculations, the dielectric function of AuIn 2 is the dielectric function calculated by Keast et al. 21 We also obtain a rather good agreement with the optical absorption of an alloyed sphere of AuIn 2 with the same dielectric function. 21 Conversely, the optical absorption of a homogeneous sphere of Au 0.33 In 0.67 using the average dielectric function of both constituents Au and In gives a well-dened resonance, which does not really correspond to the experimental spectrum. The latter shows a gradual increasing of the absorption superimposed on the resonance, which may be correlated to IB transitions, but also to the onset of an increasing absorption by the silica matrix. The optical response of a core@shell sphere Au@In with the composition Au 0.33 In 0.67 is similar to that of a homogenous sphere of Au 0.33 In 0.67 using the weighted average functions of Au and In.
Finally, these measurements do not allow the real composition of the target to be determined. Nevertheless, the experimental-theory comparison seems to indicate that the BNPs are initially partly composed of Au 0.33 In 0.67 . However, we cannot give denite conclusions on how gold and indium initially mix (alloyed at the atomic level or segregated). Aging of the sample aer a few months leads to a strong red-shi of the LSPR (see the spectrum at t 0 + 173 days in Fig. 5c), resulting in a broad resonance centred at 555 nm, in agreement with the expected LSPR of a core@shell sphere of Au@In 2 O 3 . This means that a complete demixing of gold and indium takes place over time with the formation of an indium oxide shell surrounding a gold core, in agreement with TEM observations for which no ordered Au-In phases were found.
For Au 0.50 Al 0.50 BNPs, the initial optical spectrum exhibits a strong resonance centred at 511 nm, corresponding to the LSPR of partially oxidized goldaluminium and suggesting that aluminium oxidation occurs rapidly, aer less than a few tens of minutes (time required to record the rst spectrum aer venting the sample). The experimental absorption spectrum agrees with the theoretical absorption cross-section of a multi-shell sphere of Au@Al@Al 2 O 3 with a volume oxidation rate of aluminium of 88%. As the proportion of oxidized aluminium is initially high, the spectrum does not strongly change over time; the aluminium oxide shell thus formed can prevent further oxidation. We can also conclude that we did not succeed in making silica-embedded BNPs with the ordered phase Au 0.50 Al 0.50 .
For Au 0.33 Al 0.67 , we noted that the sample just aer synthesis was grey and the corresponding absorption spectrum shows an increase of the absorption towards the UV range with a shoulder at about 210 nm that might be related to the emergence of LSPR (see Fig. 5b). No sign of strong oxidation appears in the optical response just aer air exposure. For comparison, the simulated spectrum of an Au 0.33 Al 0.67 sphere with the dielectric function taken as the weighted average of those of gold and aluminium also displays an increase of the absorption towards the UV range, but with a clear LSPR (a similar response can be obtained with a core@shell Au@Al with the composition Au 0.33 Al 0.67 ). Aer a few days, the colour of the sample became slightly pink and the absorption spectrum shows a well-dened LSPR at 522 nm. This value corresponds to the LSPR peak position of a multi-shell Au@Al@Al 2 O 3 embedded in silica. This means that Au 0.33 Al 0.67 BNPs initially contain metallic aluminium, but they are rapidly oxidized, because of the poor protection of silica against oxidation. Note that the spectrum aer aging is less red-shied than the one of AuIn x because the optical index of alumina is lower than that of Al 2 O 3 .
To conclude, optical absorption measurements on silver-based BNPs embedded in a silica matrix have shown that both Ag-In and Ag-Al systems retain metallic indium (or aluminium). Ag-In BNPs display, just aer air-exposure, an LSPR deeper in the UV range as the proportion of indium increases, while Ag-Al BNPs show a less blue-shied LSPR due to a more rapid oxidation. This more rapid oxidation of aluminium is also visible in gold-based BNPs. Au-In BNPs, just aer air exposure, have the LSPR furthest in the UV range among the four systems (Ag-In, Ag-Al, Au-In and Au-Al) investigated. Nevertheless, aer prolonged aging, the Ag-Al system is the only one keeping an LSPR still in the UV range, probably because of the formation of an oxide shell, which protects the remaining metallic aluminium within the cores. The initial LSPR of AuIn x BNPs at around 254 nm may be attributed to the formation of blue gold (AuIn 2 alloy), but its rapid red-shi over time is the sign of a rapid demixing of both elements and indium oxidation. For Au 0.33 Al 0.67 silica-embedded BNPs, even if no LSPR appears in the far UV (it may be masked by the onset of an increasing absorption by the silica and we have to keep in mind that the LSPR of aluminium is deeper in the UV than that of indium), we do not observe any sign of oxidation just aer synthesis. This suggests that the purple gold (AuAl 2 alloy) is formed at the nanoscale and it is only aer several days that the demixing of both elements, accompanied by aluminium oxidation, leads to the emergence of the LSPR of gold. For Au 0.50 Al 0.50 BNPs, gold and aluminium demix on a shorter time scale, showing that the alloy AuAl, if formed, is less stable against oxidation at the nanoscale when BNPs are embedded in silica.

E. Conclusion and perspectives
In this work, the optical and structural properties of nanoalloys combining gold or silver with aluminium or indium were investigated. Despite the presence of many alloyed phases in the bulk, it remains difficult to keep purely metallic nanoalloys stable against oxidation in the long term, even when the BNPs are embedded in a silica matrix. The BNPs, produced by laser vaporization, surfactant-free and initially in the metallic state, are indeed restructured upon airexposure or even under UHV in the case of Al-based BNPs because of the strong sensitivity of aluminium to oxidation. Nevertheless, the silver-based Ag-In and Ag-Al BNPs retain some metallic indium or aluminium even aer aging (with prolonged air-exposure), 15,16 while a demixing, accompanied by aluminium or indium oxidation, occurs in gold-based Au-In and Au-Al BNPs. Aer aging, the BNPs show a core@shell structure which generally retains some indium or aluminium inside the cores, but as the proportion of the non-oxidized trivalent metal remains poor, the corresponding LSPR is not as far in the UV range as desired.
However, in some cases, some nanoalloys at the atomic level are initially formed within the BNPs. For Ag-In BNPs, the initial spectral position of the LSPR is in agreement with a core@shell In-Ag@In structure with indium poorly alloyed cores, and even aer aging, the cores remain alloyed with a small proportion of indium. 16 For Ag-Al BNPs, the GIWAXS analysis of carbon-coated BNPs showed evidence of the incorporation of aluminium in the silver lattice according to Vegard's law, with, in addition, a compressive stress due to the formation of an aluminium oxide shell. This is mirrored in the optical response ,with an LSPR in the UV range at around 388 nm for aluminium-rich proportions. In this last case, the aluminium oxide formed around the cores probably protects metallic aluminium in the core from further oxidation. For gold-based BNPs, the initial optical responses of AuIn x and Au 0.37 Al 0.67 BNPs are also in favor of the formation of ordered alloys (blue gold AuIn 2 and purple gold AuAl 2 ), but the demixing, leading to core@shell structures, as observed by TEM and corroborated by the appearance of the gold LSPR, shows that they are not stable.
To keep such nanoalloys stable in air in the long term, new techniques of protection against oxidation have to be developed, such as, for instance, the use of other matrices. Another possibility is to regenerate the metallic character by annealing under reducing atmosphere, as has already been shown for silverbased Ag-In and Ag-Al BNPs. 16 Such experiments are in progress, but in the case of Au-Al, the reduction of aluminium in the environmental electron microscope seems to be difficult to achieve, even at very high temperatures. Identical annealing under reducing atmosphere of samples prepared either for optical characterization or for TEM observation should be performed in order to better understand the evolution of the optical response aer annealing in relation to the restructuring of the BNPs. Furthermore, as the oxide shell of alumina or indium oxide protects the cores from oxidation, the production of larger BNPs in order to get larger alloyed cores surrounded by an oxide shell, developing LSPR more in the UV range, could also be an alternative.
Concerning the structural characterization of the BNPs, anomalous GIWAXS and GISAXS (grazing incidence small angle X-ray scattering) measurements should bring complementary information about the internal structure of the BNPs. Environmental TEM and HR-STEM experiments will also be investigated in the near future and theoretical calculations will also be performed, which will be helpful to better understand the behaviour of these nanoalloys in the very small size range.
In conclusion, our results highlight the difficulties of alloying reactive trivalent metals with noble metals at the nanoscale while stabilizing them against oxidation. Despite these difficulties, we show that metallic mixed phases of varying stability can nevertheless be fabricated using a physical approach. Even though a more or less large fraction of the trivalent element is pulled to the BNP surface by oxidation, we could still show evidence of the presence of both aluminium and indium in the metallic state in the BNP core. We would, however, like to stress the importance of the complementary experimental techniques; it is only their combination that allowed us to unambiguously characterize these complex nanostructures.

Author contributions
The results presented in this publication come from a work initiated by E. Cottancin, M. Pellarin and M. Hillenkamp.
All the authors participated in this work, which also corresponds to the main part of the thesis work ofÉ. Camus. More specically, the samples were fabricated byÉ. Camus The manuscript was written by E. Cottancin andÉ. Camus and by P. Andreazza for the GIWAXS parts. The ESI, † containing XPS measurements, was written by M.
Hillenkamp and O. Boisron. The manuscript and ESI † were carefully and critically reviewed by all the authors.

Conflicts of interest
There are no conicts to declare.