1. Introduction

Fabrication of nanocrystals has gathered a lot of attention over the recent years due to the size dependent and hence tunable properties that can vary considerably from those of the bulk material. For semiconductor nanocrystals this allows adjustment of absorption and emission properties due to the change in the electron and the hole band-edge states and a blueshift of the band gap of the material when the particle size is smaller than the Bohr exciton radius of the material.

Gallium arsenide (GaAs), which is the base material for many optoelectronic devices, has a band gap of 1.42eV at 300K and the diameter of its Bohr exciton is about 19 nm [1]. This allows tuning of the band gap of the GaAs nanocrystals across the visible spectrum, which makes nanocrystalline GaAs an interesting material for various applications. Partly due to the high quantum efficiency achieved with compounds of zinc, cadmium, sulfur and selenide and especially with their core/shell-structures, free-standing GaAs nanocrystals have not been studied as thoroughly as the II/VI semiconductors. Properties of the GaAs nanocrystals and quantum dots have been investigated within oxide and polymer matrices and as epitaxially fabricated heterostructures. However, synthesis of stable, free-standing GaAs nanocrystals, exhibiting luminescence at clearly shorter wavelengths than bulk GaAs, has been challenging.

GaAs nanocrystals have been produced through chemical methods from GaCl₃ and As(SiMe₃)₃ in quinoline [2,3], from GaCl₃ and (Na/K)₃As in toluene, monoglyme and diglyme [4], from GaCl₃ and As(NMe₂)₃ in 4-ethylpyridine [5] and from a single organometallic precursor in hexadecylamine [6]. The reports include observations of blueshifted band gap absorption [2–7] and visible photoluminescence [3–7]. These methods allow a good size control and produce a relatively narrow particle-size distribution, which is typical for chemical synthesis methods. However, these processes are fairly complex, requiring careful fabrication of precursors and require several process steps. The fabrication process can also take several days to finish.

Pulsed lasers have been used to fabricate nanocrystals by ablation of GaAs targets in low-pressure gas atmosphere. Thereafter, the nanocrystals were transferred to ethanol by nitrogen flow [8]. In these crystals, visible luminescence was observed at a low temperature of 2 K. A straightforward technique to fabricate nanocrystal solutions is pulsed laser ablation in liquids (PLAL). In this technique the GaAs target is suspended in a solvent and ablated with several laser pulses [9,10]. Lalayan reported visible luminescence from the samples fabricated using this technique, but the samples were not thoroughly characterized to confirm the source of the luminescence [9]. Ganeev et. al. reported gallium-rich (Ga:As-ratio 1.4:1) nanoparticles with increased extinction towards ultra-violet wavelength range, but no clear band-edge absorption was observed and photoluminescence was not studied [10].

Pure GaAs surface tends to oxidize, which can be detrimental for the properties of the nanocrystals. The thickness of the oxide can be comparable to the particle size and in some cases the particle can be completely oxidized. Another issue is the high ratio of surface atoms in nanocrystals, which introduces defects that change the electronic and optical properties of the particles. Particularly, in the case of photoluminescence, the (surface) defects can enhance non-radiative transitions which reduce the photoluminescence. To reduce the harmful oxidation and to passivate the possible defects Traub et. al. demonstrated a process for oxide etching and subsequent surface passivation of solution suspended GaAs nanoparticles [11]. They reported improved luminescence intensity, especially after thermal annealing, but the produced nanoparticles were too large to exhibit change in the band gap due to quantum confinement.

In this paper, we report on a study of GaAs nanocrystals prepared with PLAL with simultaneous chemical surface passivation. The passivation is accomplished by using ammonium sulfide, a surface passivation agent widely studied for passivation of bulk GaAs [12], allowing a single-step fabrication of surface passivated nanocrystals. The nanocrystals exhibit visible photoluminescence and the prepared samples show no weakening of their luminescence during several months. We also elucidate the chemical structure of the prepared GaAs nanocrystals as a function of thermal annealing.

2. Experimental methods

2.1 Laser ablation

Laser ablation of solid GaAs targets was carried out in quartz and acrylic cuvettes using a high-repetition rate fiber laser (λ ≈1060 nm, pulse width 20 ps, repetition rate 1 MHz, pulse energy ≈1.6 μJ, FWHM of the focused beam ≈15 μm). Targets were submerged in ethanol, de-ionized water or water solutions containing ammonium sulfide. The laser beam was moved on the target surface with a 2D mirror scanner with a speed of 2 m/s to avoid crater formation and possible heat induced problems associated with a high number of overlapping subsequent laser pulses [13].

2.2 Materials

The GaAs ablation target was an epitaxial grade, n-doped wafer manufactured by AXT. The dopant was silicon with atomic concentration of approximately 1.5·10¹⁸ atoms cm⁻³. De-ionized water with resistivity of 18 MΩ cm⁻¹ was obtained from the Millipore Milli-Q-system. Ethanol (99.5%) was manufactured by Altia. Ammonium sulfide in a 20% water solution was obtained from Merck and it was used as received.

2.3 Sample preparation

The prepared nanoparticles were deposited on silicon substrates by drop casting. The substrate was placed on a hot-plate at a temperature of 170°C and drops of the nanoparticle solution were cast on the substrate until a total volume of 0.2 ml of the solution was consumed. Samples were then split in half; the first half was thermally annealed, while the second half was used as a reference.

Thermal annealing was performed in a nitrogen atmosphere. Annealing parameters were optimized to maximize the photoluminescence intensity of the samples while minimizing the changes to the shape of the luminescence spectra. A batch of 10 samples, prepared in ethanol, was used for the optimization procedure. Treatment at temperatures of 300°C and below caused no change in the photoluminescence of the samples when excited by laser pulses at 355 nm. The largest increase of the luminescence was observed at the temperature of 450°C, but the gain was accompanied by a considerable redshift in the wavelength of the maximum intensity. Parameters for the annealing experiment of the actual samples (375°C for 120 seconds) were selected as a compromise to maximize the PL with a minimal redshift.

2.4 Electron microscopy

To perform transmission electron microscopy (TEM) experiments, the nanoparticle samples dried on the silicon substrate were transferred to TEM-grids. This was achieved by first scratching the nanocrystals from the surface of the silicon substrate with a scalpel, then placing the grid on top of the sample and finally dropping a single drop of ethanol to facilitate the transfer of the crystals to the grid. After drying, the grids were examined with a high-resolution TEM (Jeol JEM-2200FS). The same device was also used for electron diffraction measurements and for energy dispersive spectrometry (EDS). Additional inspection by scanning electron microscopy (SEM) was performed with a Carl Zeiss Ultra-55.

2.5 X-ray photoelectron spectroscopy (XPS)

The photoemission spectra were measured using a spectrometer (Perkin-Elmer PHI 5400) with a monochromatized Al Kα x-ray source that was operated at 14 kV. The analyzer pass energy was 18 eV and the energy step was 0.1 eV. In spectral analysis (fitting), the Voigt function was used after Shirley's background subtraction. The minimum number of components, as deduced from the pure line shape was included in fittings. For results reported here, no sample charging was observed as the As 2p and Ga 2p emission components from the GaAs nanocrystal were located systematically at 1324-1325 eV and 1117-1118 eV binding energies, respectively.

2.6 Raman spectroscopy

Raman spectroscopy was carried out with a spectrometer (Andor Shamrock 303) and a cooled CCD-camera (Newton 940P). The excitation laser was a 532nm wavelength Cobolt Samba with a beam FWHM of 0.7 mm. The sample was illuminated with a collimated laser beam in an approximately 30 degree angle with respect to the sample surface. In order to avoid sample heating the beam was not focused. Indeed, varying the laser power induced no peak-shifts, confirming that all observed deviations from the bulk Raman spectra are due to the properties of the sample. The scattered light was collected with a microscope objective along the normal of the surface of the sample and Rayleigh scattered light was filtered out with a Semrock Razoredge filter.

2.7 Photoluminescence measurements (PL)

Room temperature photoluminescence with excitation wavelengths of 355nm and 532nm was measured using the same Raman spectrometry setup. The 355nm excitation source was a diode-pumped solid-state laser (Ekspla NL202). The low-temperature measurements were done using a scanning monochromator (Oriel DK480) and a photomultiplier tube connected to a lock-in amplifier. The laser emitting at 532nm was used for excitation in the low-temperature measurements. The FWHM of the focused excitation beam in the photoluminescence measurements was approximately 100 μm.

2.8 Zetapotential measurements

The stability of the solutions was studied using a Malvern Zetasizer Nano instrument to measure the zeta-potential of the nanoparticle-solutions.

3. Results

3.1 Nanoparticles in liquids

Nanoparticle solutions produced by PLAL in ethanol were reddish-brown in color and stable for several months. On the contrary, the samples prepared in water were much lighter in color with visible clusters. Within 24 hours of preparation the samples in water were completely clear with clusters in the bottom of the container.

Laser ablation in a water solution of ammonium sulfide produced solutions with visual appearance and stability very similar to samples prepared in ethanol when the molarity of ammonium sulfide was between 0.1 mmol/l and 10 mmol/l. As the molarity of ammonium sulfide is increased above 10 mmol/l, increasingly lighter color solutions are produced. The color of the solutions keeps getting lighter during the following days until the liquid is completely clear. We interpret, that instead of stabilization, excess sulfur leads to formation of sulfide compounds that are subsequently oxidized. This interpretation is supported by the analysis of the samples prepared with ammonium sulfide molarity of 100 mmol/l. The Raman measurements show the distinct spectra of arsenolite (As₂O₃), which is the oxidation product of arsenic sulfide. The SEM-images also show micrometer-sized octahedra which are typical for arsenolite.

The Zeta-potential of the samples prepared in ethanol, 0.1 mmol/l and 1 mmol/l ammonium sulfide was measured to be −41 mV, −20 mV and −48 mV, respectively. Measurement of samples prepared in water was impossible due to rapid agglomeration.

3.2 Transmission electron microscopy

The TEM analysis shows that the samples consisted of crystalline nanoparticles and additional amorphous material.

The TEM samples were prepared from nanoparticles that were deposited on silicon substrates and therefore the primary nanocrystals have mostly agglomerated into larger clusters with sizes from 20 nm up to over one hundred nanometers. This prohibits proper statistical analysis of the size distribution of the primary nanocrystals. The diameter of the observed primary crystallites varied between 2 and 10 nm, the typical diameter being between 3 and 6 nm. The primary nanocrystals are polyhedron shaped, but on the average their shape is close to a spherical form. Therefore, a relatively good estimation of the nanoparticle properties can be achieved by models developed for spherical nanoparticles. Few rod-like crystals were also observed.

Electron diffraction and Fourier-transforms of the HRTEM images confirm that the crystals have zinc-blende structure with visible (111), (220), (311) and (422) diffraction peaks. The lattice constant was calculated to be 0.555 nm, which is lower than that of bulk GaAs (0.565 nm).

To analyze the influence of ammonium sulfide, we performed an EDS analysis on samples prepared in ethanol and 1mmol/l ammonium sulfide. Analysis was performed on agglomerates that included both nanocrystals and amorphous material. In ethanol the material is gallium rich, with 2 to 2.5 times more gallium than arsenic and with 30% to 35% oxygen. The RTA-treatment reduces the oxygen content to 25% and slightly increases the relative amount of arsenic, which is to be expected if amorphous material is removed and GaAs nanocrystals remain mostly unaffected by the treatment, as suggested by the Raman measurements. Samples prepared in ammonium sulfide have gallium to arsenic ratio between 1.2 and 1.6 with 33% oxygen and 2% sulfur. Again the RTA treatment improves the overall gallium to arsenic ratio and nearly halves the amount of oxygen and sulfur in the sample.

To further investigate the distribution of elements in the samples, we measured the EDS in the center (Fig. 1 , region A) and on the edge (Fig. 1, region B) of a nanoparticle cluster for a sample prepared in 1 mmol/l of ammonium sulfide. The atomic percentages are listed in Table 1 . The amorphous edge (region B) of the cluster is gallium rich with almost twice as much gallium as arsenic. The elemental composition suggests that the amorphous material attached to the nanocrystals is a mix of oxides and sulfides. The nanocrystals in the center (region A) are close to stoichiometric GaAs with gallium to arsenic ratio of 1.1 to 1. The amount of sulfur relative to oxygen is much higher than on the amorphous edge. Considering that the measurement of the center also includes some signal from the amorphous top layer, the results suggest that the surface of stoichiometric GaAs nanoparticles is covered with a compound with Ga:O:S ratio close to 1:1:1.

 

Fig. 1 HRTEM image of a large cluster of nanocrystals. The individual primary nanocrystals can be observed as separate crystalline regions within the cluster. The typical crystallite size is from 3 to 6nm. The EDS analysis was performed from regions A and B.

Download Full Size | PDF

Table 1. EDS Analysis of the Nanoparticle Cluster Shown in Fig. 1

View Table

3.3 Raman spectroscopy

The peaks of both TO and LO phonons of GaAs can be observed in the Raman spectra of all of the samples. The peaks show considerable shifts towards lower energies and asymmetrical broadening, which is typical of GaAs nanocrystals [14].

For samples prepared in ethanol (Fig. 2 ) we observe Raman-peaks corresponding to the LO and TO phonons at wavenumbers of approximately 286 cm⁻¹ and 265 cm⁻¹ correspondingly. In addition, we see a large shoulder on the low-energy side. A reasonably good fit for this feature is achieved with two Gaussian peaks centered at 245 cm⁻¹ and 205 cm⁻¹, which are attributed to amorphous GaAs [15,16] and amorphous arsenic [17–19] respectively. Also arsenic and gallium oxides have Raman features in this range, but other peaks related to these compounds are not observed. The amorphous components obscure the line shapes of the phonons related to crystalline GaAs. In the spectra of the annealed samples, the intensity of the amorphous components is significantly reduced (Fig. 3 ) revealing more precise shape of the crystalline phonon peaks. The LO phonon is observed at 286 cm⁻¹ and the TO phonon at 265 cm⁻¹.

 

Fig. 2 Raman spectrum of GaAs nanocrystals prepared in ethanol. The fitted spectra is dominated by Gaussian peaks related to amorphous GaAs and amorphous arsenic. The black line is the measured spectrum, the red dotted line is the overall fit and the green dashed lines represent the individual components; from left to right: amorphous As, amorphous GaAs, and crystalline GaAs TO and LO phonon modes.

Download Full Size | PDF

 

Fig. 3 Raman spectrum of a GaAs sample that was prepared in ethanol after thermal annealing. The black line is the measured spectrum; the dotted red line is the overall fit. The blue dashed lines are the TO and LO phonons of crystalline GaAs fitted with the Gaussian confinement model. The green lines represent the individual amorphous components: amorphous As (205 cm⁻¹ and 252 cm⁻¹) and amorphous GaAs (245 cm⁻¹). The size of the nanocrystals obtained from the fit is 4.0 nm.

Download Full Size | PDF

We used the Gaussian confinement model [20,21] to estimate the size of the nanoparticles from the Raman spectrum. For a full derivation of the model we refer to the original article by Richter et al. [20], a brief explanation can be found in the paper by Arora et al. [21]. The phonon dispersion curves were obtained from neutron scattering experiments performed at 12 K [22] and adjusted to room temperature using the temperature dependence measured by Chang et al. [23] The original model by Richter et al. [19] used α = 2 to set the phonon amplitude at the nanoparticle boundary to 1/e, however the best fits to the experimental data have often been obtained by setting α = 8π² as proposed by Campbell and Fauchet [24].

The best agreement with the experimental results, in very good agreement with the TEM measurements, was achieved for nanoparticle diameter of 4.0 nm, using α = 8π² and including additional Gaussian peaks to model the contribution of amorphous GaAs and amorphous arsenic (Fig. 3). The broad peak centered at 247 cm⁻¹ is attributed to amorphous GaAs and the peaks at 252 cm⁻¹ and 204cm⁻¹ to amorphous As. Calculations with α = 2 and α = 9.67, from the bond-polarizability model [25], produced relatively poor fits.

The calculated spectrum agrees with the measurements surprisingly well considering that the size distribution of our samples is relatively wide. The nanoparticle diameter has a significant impact on the shape of the calculated spectrum in the size range observed in the TEM measurements. The broad size distribution was expected to produce considerably broadened Raman peaks and a greater mismatch between the fit and the experimental data. However, the results are very similar to the ones calculated for monodisperse nanocrystals [13]. This can be partly explained by enhancement of the Raman signal by resonant excitation. Using a laser with a wavelength which is resonant with a transition in the studied material can increase the Raman signal by several orders of magnitude. In the particular case of 532 nm excitation wavelength, the signal from nanocrystals with transitions near the energy of 2.33 eV is enhanced. Nanocrystal size with corresponding band gap energy falls within the size range observed with TEM as discussed later and thus resonance enhancement could explain the observed strong Raman peaks.

For the samples prepared in the solutions of ammonium sulfide, the peak of the crystalline GaAs LO phonon is observed at 285 cm⁻¹ and the peak of the TO phonon at 264 cm⁻¹ (Fig. 4 ). Two additional broad features are observed around 230 cm⁻¹ and 345 cm⁻¹. The broad peak around 230 cm⁻¹ is in part attributed to amorphous gallium arsenide and arsenic that are also observed for the samples prepared in ethanol. The shape and the location of the peak is, however, slightly different suggesting formation of gallium and arsenic sulfides. The feature around 345 cm⁻¹ is attributed to arsenic sulfide As₂S₃ (orpiment) [26], which is most likely in the amorphous form. This is suggested by the broadness of the peak and the absence of crystalline sulfides in the TEM analysis. Gallium sulfide, GaS, has Raman peaks near 345 cm⁻¹ as well [27], but gallium sulfide has a high melting point of 965 °C and the feature almost completely vanishes during RTA at 375 °C. Therefore the observed feature is most likely due to arsenic sulfide which has considerably lower melting point near 300 °C.

 

Fig. 4 Raman spectrum of the samples prepared in the ammonium sulfide solution. In addition to the peaks of crystalline GaAs, a Raman signal from arsenic sulfides is observed.

Download Full Size | PDF

The RTA treatment decreases the signal from the sulfides and amorphous components, but at the same time increases the photoluminescence making detailed analysis of the peaks corresponding to GaAs difficult.

3.4 XPS

The core-level photoemission spectra in Figs. 5 and 6 show many As- and Ga-oxide related components in addition to the GaAs components. Due to the nature of the XPS measurements the signal arises mostly from the oxidized surface parts of the nanocrystal agglomerates (Fig. 1, region B) and from the amorphous material between the agglomerates. The As₂S₃ component is also observed, and its intensity clearly decreases during the RTA, being in good agreement with the above Raman results. Furthermore, it is observed that the relative intensity of the GaAs emission decreases. The accuracy of the analysis does not allow separation of emission from GaAs and amorphous As, and thus this apparent decrease of GaAs photoemission is in line with the disappearance of the amorphous GaAs- and amorphous As-components in the Raman spectra.

 

Fig. 5 As 2p photoemissions from the nanocrystal samples prepared with ammonium sulfide. The origins of different photoemission components are proposed.

Download Full Size | PDF

 

Fig. 6 Ga 2p photoemissions from the nanocrystal samples prepared with ammonium sulfide. The origins of different photoemission components are proposed.

Download Full Size | PDF

Before the RTA the samples prepared with the mild sulfur content appear to include also metallic Ga. Furthermore, the As XPS spectra reveal that the increased sulfur amount in the preparation decreases the formation of the highest oxidation state, As₂O₅, and enhances the As₂O₃ formation during the RTA. The increased sulfur concentration also decreases the amount of oxidized Ga. The signal corresponding to GaAs and amorphous As in the As 2p spectrum is exceptionally strong in the sample prepared in 5 mmol/l (NH₄)₂S which is reduced during the RTA suggesting that the sample has a considerable amount of amorphous material.

It is worth noticing that the possible emission from the lowest Ga oxidation state, Ga₂O, overlaps with the GaAs emission. These XPS results are discussed later along with the PL results.

3.5 Photoluminescence

After drop casting on silicon substrate, the samples prepared in ethanol show weak photoluminescence when excited with a wavelength of 355 nm (Fig. 7A ). The spectrum is very broad and seems to have two underlying peaks centered at 530 nm and 650 nm. The RTA treatment with optimized parameters increases the intensity of the luminescence approximately by an order of magnitude (Fig. 7B). For this sample the enhancement is stronger near the peak at 530 nm creating an apparent blueshift for the total spectrum.

 

Fig. 7 Photoluminescence of the samples under 355 nm excitation. (A) As-deposited samples. (B) Samples after thermal annealing for 120 s at 375 °C. Luminescence spectrum of the unannealed sample prepared in the 1mmol/l (NH₄)₂S solution is included as a reference.

Download Full Size | PDF

The samples prepared in pure water did not show luminescence in our measurements. However, addition of small amounts of ammonium sulfide produced samples with more intense photoluminescence than the samples prepared in ethanol. With 0.1 mmol/l molarity, the spectrum has the two features also observed in the case of ethanol; the biggest difference being that the shorter wavelength peak is dominant even before the RTA and the center is at a slightly shorter wavelength of 520 nm (Fig. 7A). Increasing the molarity further enhances the luminescence intensity and especially the peak at 520 nm until at molarities of 5 mmol/l and above the peak starts to slowly redshift and the maximum intensity starts to fall. With 1 mmol/l molarity of ammonium sulfide the intensity of the luminescence is roughly an order of magnitude stronger than for nanocrystals prepared in pure ethanol. Interestingly, a further gain of an order of magnitude is observed after the RTA treatment (Fig. 7B). The RTA treatment also induces a redshift, of the peak wavelength to approximately 550 nm. The luminescence of the RTA-treated samples is also visible to the naked eye.

The significant improvement of the photoluminescence intensity from the crystals prepared with (NH₄)₂S and after the RTA treatment made the luminescence measurable with excitation at 532 nm. This enabled us to measure the temperature dependence of the luminescence from 30K to 290K (Fig. 8 ). As expected, the intensity of the photoluminescence increases for all of the samples as the temperature is lowered. The temperature dependence of the integrated intensity, I_PL, was fitted with a simple model to estimate the activation energy, E_A, of the thermally activated non-radiative processes related to quenching of the luminescence at higher temperatures. The intensity can be written as:

 

Fig. 8 Temperature dependent photoluminescence measurements excited with laser at wavelength of 532 nm. The samples produced in ethanol did not produce measurable luminescence before the RTA treatment. The values on the y-axis are included as a scale reference and are not directly comparable between different graphs.

Download Full Size | PDF

For the RTA-treated samples prepared in ethanol the intensity is enhanced by a factor of 2 at low temperatures. The data and the fitted curve are plotted in Fig. 8A. A good agreement is achieved using two competing non-radiative decay mechanisms with activation energies of 2.5 meV and 50 meV. Unannealed samples did not have a strong enough PL signal to allow the measurement.

The measured temperature behavior and the fitted curves for samples prepared in 1mmol/l ammonium sulfide solution before and after annealing are plotted in Fig. 8C. Before annealing, the best fit is achieved using two terms with activation energies of 25 meV and 100 meV. The annealing seems to remove the low-energy decay route and a good fit for the RTA-treated sample is achieved using only one term yielding activation energy of 100 meV.

For samples prepared in 0.1 mmol/l of ammonium sulfide solution after the RTA-treatment, a good fit is achieved with two energies of 32 meV and 96 meV (Fig. 8B). This suggests that only partial passivation has been achieved, which agrees well with the lower overall intensity when compared with the samples prepared in molarity of 1 mmol/l. The unannealed sample behaves in a slightly more complex way showing a step-like behavior. A moderate fit is achieved with activation energies of 26 meV and 100 meV. However, a far better fit is achieved when using a sum of two independent emission terms, both following Eq. (1) with activation energies of 97 meV and 106 meV. This suggests that the emission might actually derive from two different sources with different band-edge electron and hole states which could be related to different size nanoparticles or due to the incomplete surface passivation creating two different populations of differently behaving nanoparticles.

The behavior of the samples prepared in 5 mmol/l solution of ammonium sulfide is more complex before annealing (Fig. 8D). The integrated luminescence intensity starts to oscillate as a function of temperature at temperatures below 150 K. Due to these oscillations, the relevance of the fitting parameters of Eq. (1) becomes highly questionable. We cannot explain this behavior, but it is likely that the mechanism for luminescence quenching is more complicated for higher sulfur concentrations. After the RTA treatment a good fit is achieved also for this sample with activation energies of 5 meV and 81 meV.

4. Discussion

Laser ablation of the target induces formation of plasma. The expansion of the plasma is restricted by the surrounding liquid which causes nucleation of the evaporated material to form nanoparticles [28]. In inert liquids this tends to lead to electrically charged nanoparticles repel each other and therefore form very stable solutions. In the case of GaAs in water and solutions of ammonium sulfide the liquid is clearly not inert and oxygen and sulfur can react with the formed GaAs nanoparticles leading to the observed thin oxide and sulfide shells. The reactive species in the solvent can also react directly with gallium and arsenic ions at the liquid-plasma boundary, which can lead to direct formation of gallium and arsenic oxides and sulfides. We suspect that this is the origin of the amorphous material observed in the TEM and Raman analysis.

The highest stability of the nanoparticle colloids is achieved in the intermediate molarity range between 0.1 mmol/l and 10 mmol/l of ammonium sulfide in water. The TEM and EDS analyses show that the most stable particles have a GaAs core and the surface is a gallium rich compound with both sulfur and oxygen. Therefore the stability is achieved by a combination of oxygen and sulfur and not by sulfur passivation alone. Particles produced in ethanol are stable and have a thin surface oxide layer, but it is unclear whether this oxide is produced in the liquid or after deposition in ambient air. High molarity of ammonium sulfide solution leads to destabilization and change in particle composition over time. All of these results support the proposed model for the PLAL in which the newly formed, very hot nanoparticles are reactive, but after cooling down the particles are stable unless the liquid is chemically reactive.

It is noteworthy that the most stable particles also exhibit the highest luminescence intensity. This supports the hypothesis that the stabilization is related to successful surface passivation. The samples were very stable after deposition on silicon and the measured Raman and photoluminescence spectra were unchanged for at least six months. This is unexpected, considering that sulfur passivation of GaAs wafers has been shown to deteriorate within days [29,30]. The origin of the high stability in our samples might arise from the fact that the deposited nanocrystals have formed clusters where the passivated surfaces are actually interfaces between nanocrystals and therefore protected. It is also possible that the passivation layer, being a compound with both sulfur and oxygen, is more stable in the case of nanoparticles than sulfur alone.

Due to the properties of the samples, statistical size distribution measurements were not possible. However, the different estimates for the average nanocrystal size agree relatively well. Based on the TEM analysis, the typical nanocrystals are between 3 and 6 nm in diameter and the diameter obtained from the fitted Raman spectra is 4 nm. The photoluminescence maximum is near to the photon energy of 2.34 eV. Using the commonly applied effective mass approximation [1] a corresponding particle diameter of 2.3 nm can be calculated, however it is well known that this approximation does not work well for very small nanocrystals and yields too small size estimates. Calculations by Diaz et al. [31] based on an empirical sp³d⁵s* nearest neighbor tight-binding model developed by Jancu et al. [32] gives an estimated nanocrystal diameter of 3.4 nm which is in better agreement with the other size approximations. This is also in excellent agreement with the speculated resonance enhancement during Raman measurements.

A speculative comparison might be done between the observed lattice constant and previous findings on pressure-dependent behavior of the band gap: the change in the GaAs bulk lattice constant of 0.565 nm to 0.555 nm would require 4 GPa hydrostatic pressure [33] that would cause the band gap energy to increase by approximately 0.4 eV [34]. The tight-binding model in [32] did not explicitly allow lattice relaxations and used bulk values instead, therefore the additional blueshift caused by the decreased lattice constant should be considered as an additional effect. Therefore the HRTEM observation of a smaller lattice constant would imply that the nanocrystals have larger band gap than would be implied by their size related quantum confinement alone. On the other hand, the change of the lattice constant to smaller values could also be expected to shift the Raman peaks towards higher wave numbers as observed for epitaxial GaAs films in diamond anvil experiments [35]. However, this is not observed in our nanoparticles.

As a future work it would be interesting to carry out band gap and lattice constant measurements of nanoparticles with a very narrow size distribution in order to clarify the possible abovementioned issue related to the relationship between the lattice constant and PL wavelength in small nanocrystals and whether it should be taken into account in modeling.

The RTA treatment was observed to yield a significant increase in the PL intensity. This is attributed to several factors. Firstly, during the formation process, the nanoparticles are cooled down by the solvent from very high temperatures (thousands of K) in the plasma to room temperature very rapidly, which can leave plenty of crystal defects that are removed with annealing. Secondly, the passivating sulfur and oxygen layer can reorganize and improve the passivation. Thirdly, the amount of defects at the oxidized surface of the agglomerate, causing absorption, decreases during the RTA. It has been previously found related to PL of bulk GaAs that (i) Ga₂O₃ phase causes PL weakening defects, (ii) Ga₂O might even improve the oxidized GaAs surface, and (iii) As oxides are not harmful [36–38]. Our results are consistent with this picture. And finally, the amorphous material which possibly absorbs the emitted light is evaporated as shown by our Raman and XPS measurements.

The broad luminescence spectrum suggests that the nanocrystals have a relatively broad size distribution. The spectra for samples prepared in ethanol or in low ammonium sulfide molarities have two pronounced features, one centered at approximately 530 nm and another at 650 nm. This suggests that the size distribution is roughly bimodal with two separate and broad size distributions, which is typical to PLAL samples. The RTA treatment seems to enhance the luminescence from the smaller particles more efficiently. This is also the case when molarity of ammonium sulfide is increased, suggesting that the number of nanoparticles in the lower size-distribution mode is greater, but the smaller particles are also affected by defects more severely.

Roughly similar photoluminescence has been measured for porous GaAs [39,40]. This has also been attributed to quantum confinement in nanocrystals, although the evidence of existence of the nanocrystals in those cases has been ambiguous.

The wide application range of the nanocrystals include fluorescence labels in biodiagnostics [41], luminescent media to convert and broaden the emission of blue LEDs to solid state white light source [42] and optical limiting by utilization of nonlinear optical properties of the particles [43]. Our results could pave way for the GaAs nanocrystals as an alternative technology for generating various colors by pumping them with blue LEDs. On the other hand, solid state InGaN quantum wells have been intensively investigated and developed for the blue/green/yellow emitters for obtaining white light directly, and indeed great progress in this field has been made [42,44–46]. An interesting future direction towards electrically pumped light source could rely on our nanoparticles uniformly layered on silicon substrates. At present, the properties of the crystals cannot compete with the properties of solid InGaN/GaN technology.

One factor that can limit the applications of the nanocrystals demonstrated in this work is the broad size distribution. The distribution can be narrowed by careful optimization of the process parameters, by modifying the liquids by additional chemicals, e.g. cyclodextrins [47] prior to ablation and by further laser processing [48]. For size critical applications, further size-selective filtering and deposition techniques can be applied. Further work must also be done to improve the quantum efficiency of the nanocrystals. This could be done by further optimization of the surface passivation or by deposition of a suitable shell-layer to form core/shell-particles.

5. Conclusions

The combined information from the TEM, EDS, Raman and XPS analysis shows that we have successfully produced surface passivated GaAs-nanocrystals that form stable colloidal solutions. The as-prepared nanoparticle solutions are not luminescent. Nanocrystals deposited on silicon wafers show photoluminescence with a broad spectrum on visible wavelengths peaked near 530 nm. The luminescence shows no deterioration within six months. It is also evident that a considerable portion of the evaporated material reacts with the solvent and forms amorphous material. The PL intensity is considerably enhanced by the sulfur and oxygen containing surface passivation and by thermal annealing via the removal of non-radiative defects and possibly light-absorbing amorphous material.