UV plasmonic properties of colloidal liquid-metal eutectic gallium-indium alloy nanoparticles

Nanoparticles made of non-noble metals such as gallium have recently attracted significant attention due to promising applications in UV plasmonics. To date, experiments have mostly focused on solid and liquid pure gallium particles immobilized on solid substrates. However, for many applications, colloidal liquid-metal nanoparticle solutions are vital. Here, we experimentally demonstrate strong UV plasmonic resonances of eutectic gallium-indium (EGaIn) liquid-metal alloy nanoparticles suspended in ethanol. We rationalise experimental results through a theoretical model based on Mie theory. Our results contribute to the understanding of UV plasmon resonances in colloidal liquid-metal EGaIn nanoparticle suspensions. They will also enable further research into emerging applications of UV plasmonics in biomedical imaging, sensing, stretchable electronics, photoacoustics, and electrochemistry.

Gold and silver have dominated plasmonics in the past decade and remain highly relevant to the field 1 . However, many other plasmonic materials exist, which have significant advantages in several emerging applications. Such plasmonic materials include aluminium 2 , copper 3 , platinum and palladium 4 , as well as ferromagnetic metals and their alloys such as cobalt, nickel, iron and Permalloy 5 . Aluminium for example, shows strong plasmon resonances from the visible to the ultraviolet (UV) spectral region. It is abundant in nature, used industrially on a large scale and therefore inexpensive 2 . Copper enables ultralow-loss plasmonic waveguides that can outperform their counterparts made of gold and is also compatible with both silicon photonics and silicon microelectronics 3 . Platinum and palladium are essential for plasmonic optical gas sensing 6 . The plasmonic properties of ferromagnetic metals can conveniently be controlled via external magnetic fields, which is the basis of the field of magneto-plasmonics 5 . Furthermore, this allows combining plasmonics with magnonics and spintronics 7 .
The interest in plasmonic metals other than gold and silver is also rooted in their ability to support plasmon resonances in the UV spectral range. Important examples are gallium [8][9][10][11][12][13][14][15][16] , indium 9,10 , tin 9,10 , thallium 9,10 , antimony 14 , lead 10 , bismuth 14,17 , magnesium 18 , and rhodium 10,19,20 , as well as some of their alloys 21 . Nanoparticles (NPs) made of these metals are promising candidates for several specific applications in biomedical imaging and sensing. For example, many biomolecules such as DNA strongly absorb light in the UV spectral range. Plasmon-enhanced UV spectroscopy may provide a means to observe dynamic biochemical processes and characterise the chemical properties of these molecules in situ. Moreover, UV-plasmonic NPs have been predicted to find applications in biomedicine as well as in label-free DNA and single molecule sensing [22][23][24][25] . In cardiovascular science, plasmonic NPs also play an important role in the imaging of atherosclerotic plaques and the study of their pathophysiology with light at different wavelengths including the UV spectral range [26][27][28] .
Among UV-plasmonic NPs known today, those made of pure gallium have attracted particular attention 8,9,12,13,16,25,29 . Gallium is an environmentally stable liquid metal at near room temperature and it has been a key element in both electronic and optoelectronic devices since the 1960s 30 . However, despite being well-understood, gallium as well as its alloys continue to be a topic of active research revealing new and unique properties of this material [31][32][33][34] . Unlike gold and silver, gallium has a Drude-like dielectric permittivity function extending from the UV range through the visible and, mostly in the liquid state, into the infrared spectral region 13,21 . Due to their

Results and Discussion
EGaIn NPs were synthesized via ultrasonication. This technique is particularly advantageous for our optical analysis because it allows producing NPs with diameters of about 25 nm to 200 nm, which have theoretically been predicted to support strong localized UV-plasmon resonances 10 . Other fabrication techniques, such as microfluidic flow focusing 37 , produce much larger EGaIn particles (diameters of ~50 μm) that cannot support any localized UV plasmon mode. Bottom-up methods that are often used to produce metal NPs are difficult to employ with gallium alloys due to its large oxidation potential 25 .
EGaIn (200 mg) was added to absolute ethanol (10 mL) and sonicated using a probe sonicator (Qsonica, Q700) for 10 minutes at 20 W in a 20 mL vial to create EGaIn particles. Larger particles precipitated for 2 hours and the supernatant was used for further experiments. An exemplary scanning transmission electron microscopy image of the resulting particles as well as electron dispersive X-ray spectroscopy maps of the same particles are shown in the Supplementary Information Fig. S1. These images demonstrate the eutectic nature of the synthesized nanoparticles. The particles were colloidally stable for several weeks before aggregates formed, leading to precipitation. The NP suspension was diluted (1:50 in ethanol) and drop cast onto a holey carbon TEM grid for particle size analysis and diluted 10 times in ethanol for optical analysis. Figure 1A shows a representative TEM image of the synthesized EGaIn NPs. Based on the analysis of 264 NPs, we find that the sample contains a broad distribution of NP sizes peaking at 100 nm (Fig. 1B). We find more than 20 NPs in each bin (20 nm width) in the size range from 30 nm to 170 nm. A broad NP size distribution in our sample is due to stochastic nature of the sonication process. Finally, we determine (by using both TEM and ellipsometry) that all NPs have a gallium oxide shell (h) of ~3 nm thickness 38 . Significantly, removing the shell would cause the NPs to coalesce and become unstable, which is highly undesirable.
We performed UV-visible absorption spectroscopy experiments of the EGaIn NPs suspended in ethanol to measure the optical absorption spectra of the NPs. The NP dispersion was placed in a UV quartz cuvette for extended UV transparency. Extinction and absorption spectra (Agilent Technologies, USA, Cary 7000) were obtained using an integrating sphere for the NP dispersion and the solvent Fig. 2A. Ethanol starts to absorb UV www.nature.com/scientificreports www.nature.com/scientificreports/ light below ~250 nm and strongly absorbs light below 205 nm, which is generally used as a lower cut-off wavelength for ethanol 39 . The ethanol contribution to the spectrum was subtracted from the NP spectra (Fig. 2B).
The extinction and absorption spectra show two dominant peaks: one broad peak centered at ~275 nm and a second one at ~213 nm. Due to the close proximity of the latter peak to the cut-off wavelength of ethanol, both its absolute and relative amplitude and spectral position are potentially affected by the onset of strong ethanol absorption. By subtracting the absorption spectrum from the extinction spectrum, we find that the NPs also strongly scatter light in the region of the 213 nm peak, while the spectrum is clearly dominated by absorption above 220 nm. Another broad scattering peak is present in the visible centred at around 400 nm (see Fig. 2B, inset). This confirms that, similar to liquid pure-gallium NPs 13 , EGaIn NPs dispersed in ethanol also support plasmon resonances in the visible range.
We used Mie theory 40,41 to explain the physical origin of the resonance peaks in Fig. 2, and calculate the light scattering and absorption properties of single EGaIn NPs in ethanol. In general, Mie theory allows for the determination of the particles' scattering (C sca ) and absorption (C abs ) cross sections, as well as their extinction (C ext = C sca + C abs ) cross sections 40,41 . In the following, the light absorption is analysed in terms of a particle's absorption efficiency defined as Q abs = C abs /G 41 , where G = πR 2 is the geometrical cross-section of a sphere. Note that Q abs can have values greater than one, implying that the NP can absorb photons outside its geometrical cross-sectional area G 40,41 .
We first calculate the optical properties for single liquid gallium NPs with diameters d of 50 nm, 100 nm, 150 nm and 200 nm, all of which are present in the NP ensemble (Fig. 1). Then, we weight the resulting spectra by the proportion of NPs in the ensemble, which allows us to visually estimate the relative contribution of each spectrum (Fig. 3E). In all calculations, we take into account the experimentally measured gallium oxide shell thickness (h = 3 nm, Fig. 3A) and assume that the NPs are embedded in a constant dielectric material of refractive index n EtOH = 1.478, which is the refractive index of ethanol at 185 nm 42 . We also compared the result obtained with the Mie theory with that obtained with a 3D finite-difference time-domain (FDTD) method for a single NP. As shown in the Supplementary Information Fig. S2, the results obtained with both methods are in acceptable agreement, even though the FDTD method fails to reproduce some features seen in the spectrum produced by the exact Mie theory.
It is instructive to discuss the dielectric permittivity function of EGaIn. We find that in most studies on gallium and gallium alloys the surface oxide layer is not removed before measurements of the dielectric permittivity function are conducted 21 . When the oxide was removed to get the 'pure' metal properties, it was demonstrated that gallium and EGaIn had different optical properties in the absence of the oxide, but showed very similar properties in the presence of the oxide 21 . This can be explained by the fact that EGaIn becomes surface-enriched with indium in the absence of the oxide 21,42 .
In our simulations, we therefore use the fact that in the presence of the oxide layer, the optical properties of EGaIn and pure liquid gallium are similar 21,43 . In the UV-visible spectral region, the dielectric permittivity function of pure liquid gallium can accurately be represented by a simple Drude-like model 13,21,43 . Thus, as experimental data for the dielectric permittivity of EGaIn (ε EGaIn ) in the UV spectral region are not available, we employ the Drude model to extrapolate experimental values of ε EGaIn obtained for the visible-to-NIR (400 nm to 1000 nm) spectral region 21 . The refractive index of the oxide layer is n ox = 2 13 . www.nature.com/scientificreports www.nature.com/scientificreports/ Figure 3B shows the normalized field intensity distribution I/I 0 (=|E| 2 /|E 0 | 2 ) at a wavelength of 213 nm, where I and I 0 denote the calculated and incident field, respectively. Here, a close to 15-fold field intensity enhancement is reached at the surface of the NP, while for 275 nm excitation this enhancement is only 4-fold (Fig. 3C). The radial field intensity cross-section in Fig. 3C also shows that the field intensity is more strongly confined to the NP surface at 213 nm compared to 275 nm. From studies of conceptually similar elongated plasmonic NPs 44,45 , it is known that a stronger field confinement in the near-field zone of the NP leads to a lower radiative broadening of the resonance peak detected in the far-field zone. Indeed, in good agreement with our experimental observations, the theoretical resonance peak at 213 nm is significantly narrower than the peak at 275 nm (Fig. 3D). Figure 3C also reveals that the dielectric properties of the thin oxide layer result in an additional obstacle for the enhanced field to penetrate the metal surface of the NP, which is a well-known effect demonstrated in systems with a technologically important artificial thin dielectric layer 46,47 . For a 100 nm particle, the experimentally observed and calculated absorption peak positions are in excellent quantitative agreement and located at 213 nm and 275 nm (Fig. 3D). The peak at 213 nm shows a narrow spectral width of 9 nm (experiment) and 11 nm (theory). The broader peak at 275 nm has a linewidth of 92 nm (experiment) and 26 nm (theory). The number and relative intensity of resonance peaks in the spectral region between 150 nm and 250 nm varies for the investigated NP sizes in the size range of 50-200 nm (Fig. 3E). While the calculated spectrum for a 50 nm particle shows only one pronounced peak below 200 nm (experimentally not accessible here), the spectrum of a 200 nm particle exhibits three far less pronounced peaks. Importantly, all particle sizes show a significant peak at 275 nm in agreement with experiments.
Finally, we employ an effective medium model to simulate the collective optical response of the ensemble of NPs. We first weight the calculated spectra of the single liquid-metal NPs by the proportion of NPs in the ensemble (Fig. 1B) and then we add them up to obtain an effective response spectrum. In the resulting curve www.nature.com/scientificreports www.nature.com/scientificreports/ (Fig. 3D), the peak at 275 nm corresponds to the same peak seen in both experimental and theoretical spectra of the 100 nm NP. However, the 213 nm peak appears to merge with the 195 nm higher-order mode peak originating from the spectrum of the 50 nm NP (Fig. 3E). This result is consistent with expectations from an effective medium approach where resonance peaks may become become broader due to contributions all NPs in the ensemble.
Strong absorption of light by ethanol at wavelengths below 205 nm and the peculiarities of the field intensity distribution of higher-order plasmon modes (Fig. 3C) helps to explain why the effective medium approach fails to reproduce the 213 nm peak observed in experiment. The absorption spectrum was measured by using an integrating sphere for the NP dispersion. The 195 nm peak of the 50 nm NP is a higher-order plasmon mode and therefore its contribution to the experimental spectrum was small. Indeed, a high amplitude and narrow linewidth of the 195 nm peak imply that light is tightly confined to the NP and therefore is not emitted to the far-field zone where integrating sphere measurement are performed but it is mostly absorbed by ethanol. This process is not taken into account by the effective medium approach based on theoretically calculated near-field zone optical properties of NPs. However, the prediction of this model for the 275 nm peak confirms that in the 250-350 nm spectral range the optical response of the NP ensemble is dominated by the spectrum of single 100 nm NPs that are most numerous in our samples.

Conclusions
We have investigated the UV plasmonic properties of eutectic gallium-indium (EGaIn) liquid-metal NPs suspended in ethanol and calculated the particles' optical properties by using Mie theory. We experimentally demonstrate that the NPs exhibit UV plasmonic resonances. In agreement with our theoretical model for a 100 nm particle and an effective medium approach, we show that the two main resonances are located at 213 nm and 275 nm. These UV plasmonic properties of liquid-metal NPs are of significant interest due to their potential applications in biomedical imaging, sensing and medicine. The results presented in this work also further contribute to the concept of reconfigurable liquid-metal plasmonics 35 , where plasmon resonances of liquid-metal NPs can be tuned by deforming the shape of the NP using, for example, ultrasound. In this case, the NPs must be in solution to enable an efficient propagation of ultrasound from the source to the NPs. Importantly, although the plasmonic properties of liquid-metal NPs made of pure gallium are similar 25 , the lower melting point of EGaIn (~15.5 °C) compared to pure Ga (~30 °C) for example make it more suitable for applications at room temperature. Finally, an increasingly important role of liquid metal EGaIn NPs in stretchable electronic devices and mechanics 34 opens up opportunities to further integrate plasmonics with electronics, optomechanics and electrochemistry 36 at the nanoscale.

Methods
EGaIn nanoparticles were synthesized using sonication. Briefly, EGaIn (200 mg) was added to absolute ethanol (10 mL) and sonicated using a probe sonicator (Qsonica, Q700) for 10 minutes at 20 W in a 20 mL vial to create EGaIn particles. Larger particles were allowed to precipitate for 2 hours and the supernatant was used for experiments. Extinction and absorption spectra (Agilent Technologies, USA, Cary 7000) were obtained using an integrating sphere. A UV quartz cuvette was used for extended UV transparency.
Standard Mie theory 40,41 and finite-difference time-domain (FDTD) methods 48 were used to calculate the absorption spectra of EGaIn NPs. In all simulations, the standard Drude model 48 was employed to fit experimental values of the dielectric permittivity of EGaIn 21 . See Supplementary Information for more details regarding FDTD simulations.

Data Availability
The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request.