Nitrogen plasma passivation of GaAs nanowires resolved by temperature dependent photoluminescence

We demonstrate a significant improvement in the optical performance of GaAs nanowires achieved using a mixed nitrogen-hydrogen plasma which passivates surface states and reduces the rate of nonradiative recombination. This has been confirmed by time-resolved photoluminescence measurements. At room temperature, the intensity and lifetime of radiative recombination in the plasma-treated nanowires was several times greater than that of the as-grown GaAs nanowires. Low-temperature measurements corroborated these findings, revealing a dramatic increase in photoluminescence by two orders of magnitude. Photoelectron spectroscopy of plasma passivated nanowires demonstrated a yearlong stability achieved through the replacement of surface oxygen with nitrogen. Furthermore, the process removed the As0 defects observed on non-passivated nanowires which are known to impair devices. The results validate plasma as a nitridation technique suitable for nanoscale GaAs crystals. As a simple ex situ procedure with modest temperature and vacuum requirements, it represents an easy method for incorporating GaAs nanostructures into optoelectronic devices.


Introduction
Nanostructured GaAs has a unique niche in next-generation technologies. Whereas its superior optoelectronic properties have been exploited in bulk for decades [1], nanoscaling now promises flexible integration schemes [2,3] and novel properties [4,5] along with reduced material cost and size [6,7]. Though shrinking its footprint has unlocked an elemental smorgasbord, the combination of its increased surface-to-volume ratio and lack of a passivating native oxide still presents a significant challenge [8].
The standard method of passivating GaAs and other semiconductor surfaces has been to replace their native oxides with a wide bandgap overlayer [9,10]. Ideally, this saturates dangling bonds, removes oxide-induced defects, electrically confines charge carriers, and chemically protects from the environment. GaN is one of the most sought-after candidates for this because of its large bandgap, chemical and thermal stability, resistance to oxidation, and potential for further device incorporation, unlike AlGaAs and As(Ga)S which are known to degrade in ambient [9,11]. In recent decades, nitridation of GaAs substrates has been demonstrated using a variety of precursors and techniques (e.g. hydrazine [12,13], ammonia [14]; plasma [15][16][17], thermal [18], epitaxy [19,20]). Though GaAs/GaN surface conversion and increased photoluminescence have been demonstrated on bulk materials [21] [22] and GaNAs shells [23], reports on GaAs nanostructures have been scarce [24], even requiring annealing to generate a nitride layer [25]. We demonstrate ex situ passivation of GaAs Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI. nanostructures using a simple, mixed nitrogen-hydrogen plasma without relying on volatile wet-chemistry, dissociation-high temperatures, or ultra-high vacuum. We correlate photoelectron spectroscopy, which shows that Ga-and As-oxides are replaced by N, and time-resolved photoluminescence, indicating the removal of multiple trap states which quench radiative recombination.
The efficacy of mixed-plasma nitridation is explained by two key features: the inclusion of hydrogen and the use of radicals. Conceptually, GaAs nitridation proceeds via the following reaction: For ex situ and ambient applications, the Gibbs free energies [26] of As 0 , As-and Ga-oxides favor reaction with atomic hydrogen [27]. Hydrogen radicals are critical therefore, not only for removing surface oxides which may form outside UHV, but also residual As 0 which is known to impair device performance by introducing midgap states [28][29][30]. Importantly, these nitridation and cleaning reactions are only thermodynamically favorable for monatomic N and H species, which are a product of plasma generation. Finally, the use of radical nitrogen and hydrogen avoids damage by ion bombardment [21,31]. The result is a protective and stable surface conversion only a few monolayers thick.
In this work, we expand the application of plasma nitridation to GaAs nanostructures. Using simple techniques and minimal processing, arrays containing millions of vertically grown GaAs nanowires (NWs) were passivated and characterized. We explain the nanoscale passivation with results from time-resolved photoluminescence, which characterizes the behavior of excited charges within the NWs, and x-ray photoelectron spectroscopy, which reveals the atomic layers responsible for this behavior. By demonstrating GaAs/GaN core-shell NW passivation with defect removal and long-term chemical stability, we introduce a flexible plasma nitridation suitable for nanofabrication.

Methods
We used vertical GaAs NW arrays optimized for photovoltaics as a model system to study nitride passivation. NWs were grown using the vapor-liquid-solid method from evaporated Au on highly p-doped (111)B GaAs:Zn wafers purchased from AXT, Inc. This followed a two-temperature, three-step, metal-organic vapor phase epitaxy process using trimethylgallium and arsine. We refer to Dagyte et al [32] for complete growth information and transmission electron microscopy of virtually identical NWs. Nanoimprint lithography, metal evaporation and lift off was used to create a hexagonal pattern with 500 nm pitch, 160 nm diameter for growth of arrays of NWs having 1.1 um height. The NWs were grown nominally intrinsic, without intentional doping. Following observations made by Jiang et al [33], some arrays underwent a post-growth annealing step by raising temperature to 750°C for 1 min under arsine. This is known to reshape NW sidewalls, confirmed by electron microscopy in figure 1, from mixed to non-polar facets, affecting both surface defect concentration and surface chemistry.
The ex situ nitridation treatment was inspired by Losurdo et al [22] and contains three steps. First, NW samples in ambient were deoxidized using a standard 30 s, 1 M HCl recipe. Similar procedures have previously been shown to efficiently remove oxide from GaAs NW surfaces [34] and even reduce roughness on GaAs substrates [35]. After this wet etch, samples were directly brought to vacuum (P = 0.05 Torr) in a Fiji F200 inductively coupled plasma source and heated to 250°C for a 3 s, H 2 plasma preclean. Though not strictly necessary for nitridation, these preparatory steps are essential for ex situ surface conversion since they expose the bare NW surface by removing water, native oxides and surface contaminants [36][37][38]. Finally, a 97% N 2 -3% H 2 plasma (molar flow rate) was used at 350°C for 10 s to convert the surface from GaAs to GaN. As explained in the introduction, hydrogen radicals are essential as they remove surface arsenic which is freed during nitridation. To eliminate ion damage, we generated plasmas outside the reaction chamber. In this remote configuration, it is radicals and other neutral species, not plasma, interacting with the sample. Nitridation parameters can be found in Supplementary datafigure S1.
Following nitridation, time-resolved photoluminescence (TRPL) measurements were recorded at 300 K and 100 K. Photoluminescence has long been used to characterize recombination in optoelectronic materials, giving important feedback on fabrication and performance. Resolving emission in time reveals charge dynamics, providing intricate details on material properties. Here, a Millennia Pro E3383 CW laser was used to pump a frequency doubled Tsunami 3960 pulsed laser, generating 100 fs, 400 nm pulses at 80 MHz. Above-bandgap excitation has the advantage of probing higher energy states which can be overlooked electronics but are relevant to photovoltaics. Our standardized setup was used to minimize absorption by the substrate [39]. This included a TE-polarized, 30 μm × 30 μm laser spot of gaussian profile incident at near the Brewster's angle of the substrate, exciting ∼10 5 NWs with 3 × 10 12 photons·pulse −1 ·cm −2 . A Chromex spectroscope and Hamamatsu C6860 streak camera were used to time-resolve sample emission. Fitting of the instrument response function, trapping rates, amplitude and global chirp was performed using a program supplied by Pascher Instruments AB using the Nelder-Mead-Simplex method.
Finally, representative samples were examined using synchrotron-based x-ray photoelectron spectroscopy (XPS) at the HIPPIE beamline of the MAX IV Laboratory in Lund, Sweden. Since photoelectron spectroscopy provides surface elemental and chemical information, core-level spectra were recorded for Ga 3d, As 3d, Si 2p and N 1 s to confirm oxide removal and nitrogen incorporation. Excitation at 502 eV ensured sensitivity for the detection of surface nitrogen at around 400 eV. Including a check for the long-term stability of the treatment, the three investigated samples were: untreated NWs, NWs treated by plasma approximately one year previously, and identical NWs treated one day before XPS was performed. NWs were removed from their growth substrates and deposited on silicon for ease of analysis. Binding energies are not calibrated but referenced to previously reported values [40,41]. Core-level spectra were fitted with a Voigt line shape with a Shirley background removed. Details are reported in table S1.

Results and discussions
Using a streak camera, we were able to track radiative recombination from the NWs directly after pulsed excitation. The photoluminescence at peak-emission measurements in figure 2(a) shows how the surface treatment quite literally turns on GaAs above room temperature. A substrate without NWs (red) is included as a reference, but with a different doping, and orders of magnitude less surface area, a direct comparison is limited. At 300 K, the as-grown NWs (gray) barely emit over the noise of the detector. Non-radiative processes appear to dominate emission observed at this temperature. Relative to the as-grown NWs, however, the plasma passivated NWs (blue) have much stronger emission, a 3x and 8x intensity increase in luminescence is observed for nonannealed and annealed samples, respectively (table 1). The increased luminescence from plasma-treated NWs indicates an increase in the number of mobile electrons near the conduction band, E g 300 K = 1.42 eV (872 nm). We ascribe this increase to the plasma treatment's reduction of defect states which traps electrons. To help explain this, we show the photoluminescence resolved in time.
The TRPL in figure 2(b) shows the exponential decay of array emission following pulsed excitation. Visual inspection and single-exponential fitting show that plasma nitridation of as-grown NWs extends emission lifetime via removal of non-radiative states. Emission is a measure of excited charge concentration, and the lifetime of electrons from excitation to emission is inversely related to non-radiative defects and traps. Following excitation, the number of electron-hole pairs, N, can be modelled by the equation trap t and g are the rate constants of charge trapping and bi-molecular recombination. Though as-grown emission at 300 K (gray) was too weak to be well-fit, plasma emission (blue) was first-order exponential in character, N N e , k 0 t = and appears linear on a semi-logarithmic plot of I versus t. In line with previous observations [42], this indicates that trapping, not bi-molecular recombination, characterizes the kinetics. Thus, 1 trap t rates are extracted using first-order exponential fitting, and we observe that plasma nitridation induces a clear change. trap t lifetimes are between 6 and 8 ps for the passivated samples and nearly unresolvable by the detector in the asgrown samples (instrument response ∼4.75 ps). This increased emissive lifetime exhibited by GaAs/GaN arrays is a shift toward bulk-like behavior and a clear indication that non-radiative trap-states have been removed. Removal of trap states is essential for the optimization and performance of devices relying on the separation and extraction of charges.
A similar optical enhancement was observed at lower temperature (100 K). Visible in figure 3(a), our setup exploits the bandgap narrowing effect of dopants on semiconductors [43,44]. This 20 nm redshift of the p-doped substrate conveniently provides discrimination from the NWs (dashed lines), and enables in situ, full array characterization without removal from the growth substrate. This shift corresponds well with theory and corroborates the manufacturer doping specification of n ∼10 19 cm −3 . A temperature dependent blueshift in NW emission to 826 nm was also observed and corresponds to that predicted by the Varshni equation [45], shown in figure S2. Unlike after plasma passivation, as-grown array emission (gray) appears to be dominated by its substrate. This is observed as a peak at 846 nm. Substrate emission, as discussed in previous reports, is usually negligible when covered by arrays in this configuration due to the strong absorption properties of the NWs [6,32,46]. However, due to the large surface area of nanostructures, without suitable passivation nearly all photogenerated charges in the NWs become trapped and/or recombine nonradiatively. In the case of as-grown NWs, this allows the very limited absorption of the substrate to substantially contribute to overall emission. This is not the case after plasma nitridation however, which greatly increased mobile charge concentration and therefore photoemission in NWs from both non-annealed and annealed samples. Peak emission was more than an order of magnitude greater in the passivated arrays as compared to as-grown NWs at low temperature, 17x and 13x for non-annealed and annealed, respectively. From the time-resolved kinetics in figure 3(b), we see that the emission lifetime is several times longer in the plasma treated NWs, 5x and 4x for non-annealed and annealed, respectively. Clearly, as photoelectron spectroscopy reveals below, passivation of the surface by removing the oxide and converting GaAs to GaN considerably removes nonradiative surface states.  Compared with the results obtained at 300 K, the optical characteristics of both NW types are improved by lowering the temperature to 100 K, at which an increase in emission lifetime by a factor of 2-3 is observed for all samples. Temperature dependent reaction processes can be characterized by an Arrhenius equation, k Ae Ea K B T =k with an associated rate constant, k , 1 trap = t and activation energy, E , a where K B is Boltzmann's constant, T is temperature in kelvin and A is a constant. We consider these temperature-dependent changes in photoluminescence thermally activated electron trapping and find activation energies around 12-13 meV. Results are collated in table 1. Similar observations have been made in GaAs/AlGaAs NW arrangements [42,47], with activation energies around 17 meV. Because the energy of activation for this trapping is between K T B at 100 and 300 K, it is expected to have a significant influence on electron recombination.
Comparing figures 2(b) and 3(b), even more dramatic are the peak intensity increases at low temperature. For NW arrays differing only in surface treatment, we expect similar absorption and therefore initial concentration of photogenerated electron-hole pairs. We notice instead, however, a sharp difference between surface treatments' peak intensity prior to decay (10x and 28x for as-grown samples, and 54x and 46x for passivated samples). We must therefore consider the temperature dependency of the initial mobile charge concentrations at the bottom of the conduction band. Reduced intensity can be explained by a process which reduces electron concentration faster than the resolution of the streak camera. A non-radiative, sub-picosecond electron trapping that is much faster than state-filling would be observable as a change in the peak emission immediately following excitation. Since decreasing trap concentration has the effect of increasing the photogenerated charge concentration, with the help of time-resolved photoluminescence we conclude that the plasma treatment removes not only slow traps which is expressed via increased lifetime, but also fast traps, which quench initial emission intensity. The former occurs on a picosecond timescale, while the latter appears to be much faster (∼fs). With such observations and assumptions, we presume a fast-trapping surface state governed by E a fast ≈ 50 meV. Evidence that much of this trapping has been removed can be seen by looking at the intensities just after time-zero for plasma passivated NWs ( figure 3(b), blue) which are nearly restored to the pure-substrate reference level ( figure 3(b), red).
A final observation on photoluminescence is also discussed. Until now as-grown samples and plasma-nitride samples have been grouped together since qualitatively they exhibit similar trends regardless of post-growth annealing. In other words, with respect to passivating GaAs via plasma nitridation, NW annealing seems to be largely irrelevant and may be achieved at relatively low temperature and vacuum. The exception to their similarities was a notable spectral shift which was observed in the plasma nitride samples. This is most evident in the 5 nm blueshift from non-annealed (light blue) to annealed (dark blue) arrays in figure 3(a). The shift is slightly larger, 7 nm, at 300 K. As shown by SEM in figure 1, our post-growth, pre-nitridation annealing procedure visibly reshapes mixed-faceted NWs into lower surface energy {110}-faceted hexagonal NWs. In bulk configurations {110} surfaces are known to be defect-free within the bandgap. This ∼ 6 nm blueshift is only present in GaAs/GaN NWs and acts to restore the central emission wavelength of the NWs to the expected edgeto-edge position at 826 nm (1.50 eV) for bulk GaAs [48]. Low lying defects near the band edges are known to trap excitons and offer shallower recombination pathways (e.g. (D 0 , X) recombination [49]). Since post-growth annealing both increases emission intensity and restores the central wavelength to the full bandgap (figures 3(a), (b)), we mention this as an independent procedure which may also remove defects.
Since a critical component of material passivation is surface cleaning and protection from the environment, photoelectron spectroscopy was conducted to give a chemical characterization of the treatment. Photoelectron binding energies give us precisely such information about the chemical environment of the studied atoms. In4(a), we see a N 1 s signal present only from the samples with NWs exposed to nitrogen plasma (blue). Whereas as-grown NWs (black) exhibited a flat background signal across the N 1 s energy range, NWs which underwent plasma nitridation clearly show a peak at 397 eV which we assign to GaN. This is the case, notably, not just for the NW array subjected to plasma the day before measurement (light blue), but also for the sample passivated hundreds of days earlier (dark blue). This stability over time is explained by the strength of the GaN bond and is one of the major reasons why GaN is desirable as a passivating layer. Only qualitative observations concerning the presence or lack of nitrogen are appropriate here. Due to the difference in NW density deposited on the silicon substrates, and therefore atoms within the x-ray beam, quantitative analysis is not feasible. Instead, we look at relative changes in chemical composition to combine observations of selective nitrogen incorporation, along with significant and persistent As 0 , As-and Ga-oxide removal to demonstrate chemical passivation. That nitridation is responsible for the removal and suppression of oxide and arsenic defects is evidenced in figures 4(b) and (c). Unpassivated GaAs reacts with air at the surface, creating native oxide and defects. Deconvolutions of the normalized As 3d core-level spectra in figure 4(b) show how arsenic composition at the surface of the NWs varies between samples. A major change is the arsenic oxide component (red) visible at a binding energy of 44.0 eV. As reported previously [41], with a chemical shift of + 3.0 eV above the bulk GaAs component (41.0 eV), this is likely As 2 O 3 . This naturally occurring arsenic oxide which is prevalent on untreated NWs (40% of the signal) is almost completely removed (4%) from the NWs that underwent plasma treatment and were measured the next day ( figure 4(b), bottom). This 89% reduction of arsenic oxide persisted despite daylong exposure to oxygenrich ambient conditions. Significantly, plasma-treated NWs kept in air for over three hundred days also resisted significant reoxidation (17%), showing less than half the arsenic oxide found on the as-grown reference NWs. Clearly nitridation resulted in a stable surface replacement of arsenic oxide.
Surface quality is essential for electronic devices and a presence of arsenic in the +0 oxidation state ( figure 4(b), gray) is known to create electronic states within the bandgap which pin the fermi level and impair device performance [29,50]. From angle resolved XPS, this As 0 defect manifests as a high binding energy shoulder of As-Ga which broadens the spectra [40,51,52]. In the reference NWs we find such a component at + 1.3 eV above the binding energy of the bulk which is not resolvable in the plasma treated samples. The dashed lines centered on the 3d 5/2 components help guide the eye. Furthermore, As 0 may comprise a significant fraction of the material. Fitting estimates suggest that more than six percent of the arsenic signal emanating from the surface comes from this new atomic arrangement. Qualitatively, whereas the spin-orbit splitting of the bulk As- Figure 4. X-ray photoelectron spectroscopy of as-grown NWs (black), and NWs treated with nitrogen plasma approximately one day (light blue) and one year (dark blue) before measurement. Three core-levels, (a) N 1 s, (b) As 3d, and (c) Ga 3d, were recorded to monitor the oxidation and nitridation resulting from the plasma treatment.
Ga doublets (green) are visible in the raw spectra from the passivated samples ( figure 4(b), blue), we find ambient conditions introduce a broadening. Consistent with previous reports of GaAs substrate [22], this suggests that As 0 related defects may be permanently removed by the plasma treatment.
A similar effect is observed on the chemical environment of gallium atoms, which indicates a partial removal of oxide that is stable for at least one year. In figure 4(c), Ga 3d photoelectrons from GaAs are centered at 19.0 eV. Gallium oxides are observed as a high binding energy shoulder at 20.1 eV (Ga 2 O 3 and/or Ga 2 O). This native gallium oxide signal is large and clearly present on the unpassivated reference array (black) but diminished on the treated samples (blue). Finally, a low binding energy component at 18.1 eV which may be related to surface or elemental gallium was also observed. Remarkably, the passivated samples are virtually identical with respect to gallium observed, after nearly one year of exposure to ambient conditions. Whereas fitting suggests that in the reference sample almost half (44%) of the gallium is in an oxidized state, in the cleaned NW arrays having surface arsenic replaced by nitrogen, the amount of oxide is reduced by 18% and no compositional change was observed with time. Chemical stability is significant here. Once the As 0 defects generated during ambient oxidation have been removed, persistent trivalent oxides like Ga 2 O 3 (and As 2 O 3 ) may simply enhance passivation by acting as a wide bandgap diffusion barrier [53].
These changes in NW surface composition help explain the removal of trap states and increased photoluminescence observed at 300 K and 100 K. Nanostructures have orders of magnitude more surface area to volume than their macro counterparts. This is why surface treatments are especially relevant to nanoscale semiconductors like GaAs which lack a passivating native oxide. By converting surface GaAs to GaN in the appropriate way, parasitic defects like As 0 can be completely removed and the result is a surface much better suited for devices. Consistent with previous theory and similar reports on bulk materials [22,26], critical to attaining this is the use of hydrogen radicals which serve a two-fold purpose. First, pure H 2 cleans surface impurities and oxidation resulting from exposure to ambient. Second, including 3% H 2 during nitridation removes excess arsenic which would otherwise accumulate at the interface during a pure N 2 -plasma procedure.

Conclusions
To conclude, we demonstrate nitrogen plasma as a flexible tool for the surface conversion of nanostructured GaAs to GaN. Without a need for annealing, we demonstrate how an N 2 -H 2 mixed-plasma treatment significantly improves the optical performance of GaAs NWs configured for photovoltaics. Using time-resolved photoluminescence, we were able to characterize the treatment's removal of multiple trap states. This passivation resulted in greatly increased photoluminescence and emission lifetime at both 100 K and 300 K. Furthermore, we correlate this with an apparently permanent removal of As 0 defects and dramatic and persistent removal of As-and Ga-oxides via replacement with nitrogen. By demonstrating optical enhancement and yearlong chemical stability we show this ex situ passivation technique is relevant to nanoscale GaAs crystals.
Stable surface layers are critical for protecting materials during fabrication, experimentation and application which can involve changing temperature environments and exposure to air. III-V semiconductors like GaAs, especially at the nanoscale, have an urgent need for surface treatments compatible with ambient conditions and subsequent processing. From a device perspective, the need is even more pressing. Thermodynamic stability and lack of As 0 are essential prerequisites for GaAs electronics which plasma nitridation appears to meet. Though additional optical, electronic, and chemical characterization will be pursued in upcoming reports, we recognize these findings as a critical step toward understanding GaAs plasma nitridation at the nanoscale.