Communication www.advmat.de 3D Ordering at the Liquid–Solid Polar Interface of Nanowires Mahdi Zamani, Giulio Imbalzano, Nicolas Tappy, Duncan T. L. Alexander, Sara Martí-Sánchez, Lea Ghisalberti, Quentin M. Ramasse, Martin Friedl, Gözde Tütüncüoglu, Luca Francaviglia, Sebastien Bienvenue, Cécile Hébert, Jordi Arbiol, Michele Ceriotti,* and Anna Fontcuberta i Morral* Understanding the nature of interfaces is The nature of the liquid–solid interface determines the characteristics of a key to determining and engineering their variety of physical phenomena, including catalysis, electrochemistry, lubrica- properties. However, while liquid–solid tion, and crystal growth. Most of the established models for crystal growth interfaces are fundamental to many appli- are based on macroscopic thermodynamics, neglecting the atomistic nature cations and interactions, relatively little is known about them at the atomic level. of the liquid–solid interface. Here, experimental observations and molecular Here, we use vapor–liquid–solid (VLS) dynamics simulations are employed to identify the 3D nature of an atomic- grown nanowires (NWs) as a model system scale ordering of liquid Ga in contact with solid GaAs in a nanowire growth to identify and interpret the atomistic configuration. An interplay between the liquid ordering and the formation of nature of a liquid–solid interface in three a new bilayer is revealed, which, contrary to the established theories, sug- dimensions. VLS is a widely used method of crystal growth relying on the precipitation gests that the preference for a certain polarity and polytypism is influenced by of a solid from a supersaturated nanodro- the atomic structure of the interface. The conclusions of this work open new plet.[1] Recent in situ investigations have avenues for the understanding of crystal growth, as well as other processes revealed that precipitation/growth proceeds and systems involving a liquid–solid interface. in a layer-by-layer fashion, similarly to thin film epitaxy.[2,3] Growth of NWs can occur M. Zamani, N. Tappy, L. Ghisalberti, Dr. M. Friedl, Dr. G. Tütüncüoglu, Dr. D. T. L. Alexander Dr. L. Francaviglia, Prof. A. Fontcuberta i Morral Interdisciplinary Centre for Electron Microscopy (CIME) Laboratory of Semiconductor Materials École Polytechnique Fédérale de Lausanne Institute of Materials EPFL Faculty of Engineering Lausanne 1015, Switzerland École Polytechnique Fédérale de Lausanne Dr. S. Martí-Sánchez, Prof. J. Arbiol EPFL Catalan Institute of Nanoscience and Nanotechnology (ICN2) Lausanne 1015, Switzerland CSIC and BIST E-mail: anna.fontcuberta-morral@epfl.ch Campus UAB, Bellaterra, Barcelona, Catalonia 08193, Spain G. Imbalzano, S. Bienvenue, Prof. M. Ceriotti Prof. Q. M. Ramasse Laboratory of Computational Science and Modeling SuperSTEM Laboratory Institute of Materials SciTech Daresbury Campus Faculty of Engineering Keckwick Lane, Daresbury WA4 4AD, UK École Polytechnique Fédérale de Lausanne Prof. Q. M. Ramasse EPFL School of Chemical and Process Engineering and School of Physics Lausanne 1015, Switzerland and Astronomy E-mail: michele.ceriotti@epfl.ch University of Leeds Dr. D. T. L. Alexander, Prof. C. Hébert Leeds LS2 9JT, UK Electron Spectrometry and Microscopy Laboratory Prof. C. Hébert Institute of Physics Institute of Materials Faculty of Basic Sciences Faculty of Engineering École Polytechnique Fédérale de Lausanne École Polytechnique Fédérale de Lausanne EPFL EPFL Lausanne 1015, Switzerland Lausanne 1015, Switzerland Prof. J. Arbiol The ORCID identification number(s) for the author(s) of this article ICREA can be found under https://doi.org/10.1002/adma.202001030. Pg. Lluís Companys 23 Barcelona, Catalonia 08010, Spain © 2020 The Authors. Published by Wiley-VCH GmbH. This is an open ac- Prof. A. Fontcuberta i Morral cess article under the terms of the Creative Commons Attribution License, Institute of Physics which permits use, distribution and reproduction in any medium, provided Faculty of Basic Sciences the original work is properly cited. École Polytechnique Fédérale de Lausanne EPFL DOI: 10.1002/adma.202001030 Lausanne 1015, Switzerland Adv. Mater. 2020, 32, 2001030 2001030 (1 of 8) © 2020 The Authors. Published by Wiley-VCH GmbH www.advancedsciencenews.com www.advmat.de in different growth directions, crystal structures, and polari- In this work, we combine scanning transmission electron ties,[4–6] currently understood by relating the prevalence of these microscopy (STEM), electron energy loss spectroscopy (EELS), properties to the wetting characteristics of the liquid on the tip of atomic simulations using machine learning potentials and mole- the NWs,[5,7] where the role of the liquid droplet contact angle is cular dynamics (MD) to demonstrate, for the first time, the polarity- widely discussed but remains controversial.[4,8] So far, the liquid– dependent 3D nature of the liquid ordering at the solid interface. solid boundary is considered as a clear-cut, binary interface, with The study is performed in the frame of the VLS growth of no regard for its structure. Modeling has mostly reasoned in NWs. We elucidate the interplay between the ordering and the terms of macroscopic parameters, including contact angle, the formation of a new bilayer, the preference for a certain polarity surface energies at the solid–vapor and liquid–solid interfaces and the effect of ordering on defect formation. While our study and the chemical potentials.[9,10] The atomistic nature of the concentrates on the Ga-assisted growth of GaAs, the results can participating parts has rarely been considered. extend to other relevant cases such as Au-assisted NW growth, The idea that a solid surface can induce local order to an adja- bulk crystal growth from the melt, and even liquid–solid cent liquid phase has previously been demonstrated in simula- interaction in the context of biological environments. tions of interfaces between solids and aqueous solutions, as well We start by providing experimental evidence of liquid as liquid metals in contact with their solid compounds.[11–13] In ordering at the interface with the solid in a VLS-grown NW addition, there have been some experimental attempts to study using aberration-corrected STEM. Our study focuses on the such ordering.[14–16] In particular, grazing incident X-ray diffrac- Ga(l)–GaAs(s) interface of A and B polar GaAs NWs obtained tion has been used to study the ordering of liquid AuSi alloy,[17,18] by the Ga-assisted method.[20,21] A schematic drawing of the and in situ X-ray diffraction has been employed to study the process is provided in Figure 1A. A Ga droplet preferentially liquid ordering at the interface between P-terminated crystalline gathers As4 molecules, which dissolve in the droplet. The chem- InP substrate and liquid InAu phase.[19] While this second study ical bonding between As and Ga results in the growth of GaAs was targeted at understanding VLS-NW growth, due to the geo- at the bottom of the droplet.[20] Along the [111] direction, GaAs metrical difficulties for probing NWs using X-rays, it was limited is organized as indivisible bi-layers of As–Ga pairs, known to investigating an interface in which a bulk liquid completely as dumbbells. Due to their different electron affinity and the covers a bulk substrate, as opposed to the measurement of a relative position with respect to the (111) surface, each bilayer nanoscale droplet in contact with the crystalline NW. It further exhibits an intrinsic electric field, called polarity. Positive (A) applied an assumption of a fixed spacing between the ordered and negative (B) polarities correspond to termination by Ga and liquid layers. In addition, the composition of the liquid phase As, respectively. was tuned to match that observed in the gold-catalyzed growth Representative aberration-corrected high-angle annular dark- of InP NWs; therefore, it is not clear if the results can apply to field (HAADF) STEM images of the interface for two NWs the self-catalyzed growth of III–V NWs. Using electron micro- observed along the [11 0] zone axis of zinc-blende (ZB) and copy, Kaplan and co-workers studied the longitudinal ordering of [112 0] of wurtzite (WZ), with A and B-polarities, are shown in liquid Al at the interface with Al2O3, which can extend up to five Figure 1B,C. We mark the As and Ga atoms in blue and red, layers depending on the facets.[14] However, thus far, the effect of respectively. Polarity determination for these NWs is offered such ordering has been ignored in the VLS growth. in Figure S1 of the Supporting Information. The A-polar NW Figure 1. Microscopic view of the Ga(l)–GaAs(s) interface. A) Graphic representation of Ga-assisted growth of GaAs NWs: As4 arrives on the Ga droplet and is dissolved. The Ga atoms at the interface with the GaAs are ordered following the underlying crystalline structure. B,C) HAADF-STEM images of the liquid–solid interface of the A- and B-polar NWs observed along [110] ZB and [112 0] WZ zone axes, respectively. The positions of the As and Ga atoms are indicated by the blue and red disks, respectively, as deduced from HAADF intensities. The stacking for the A-polar GaAs is ABCABC, cor- responding to the zinc-blende crystal structure, whereas the stacking in the tip of the B-polar NW is ABAB, which is indicative of the wurtzite crystal structure. The right-side panels of (B) and (C) show intensity profiles from the STEM images integrated along the direction parallel to the NW surfaces. The peaks in the intensity correspond to the visually apparent ordered layers in the STEM images, which are in addition highlighted by the overlap- ping transparent red rectangles. The pink dashed line indicates the position of the interface. The liquid on the B-polar surface exhibits a longer range ordering. The white scale bar on the top left panels of (B) and (C) is 0.5 nm. Adv. Mater. 2020, 32, 2001030 2001030 (2 of 8) © 2020 The Authors. Published by Wiley-VCH GmbH www.advancedsciencenews.com www.advmat.de Figure 2. EELS analysis of the liquid–solid interface on an A-polar NW. A) Atomic-resolution HAADF image acquired simultaneously with the EELS map. The interface is indicated by the dashed line and the scale bar is 5 nm. The inset shows a magnified view of the interface, where the ordered liquid region is visible. B) Variation in the plasmon peak position close to the interface. C,D) Comparison of EELS measurements (solid lines) with ICA model (dashed curves), and ICA decomposition corresponding to ordered liquid (OL), liquid phase (L), and solid (S) phase. Spectra are extracted from the ZB GaAs (P1), interface (P2), and liquid Ga (P3) regions of the map indicated in panel (A). The color plot and all curves show normalized spectra averaged on horizontal lines of the map. exhibits a pure ZB structure, whereas the B-polar NW exhibits Figure 2A shows a HAADF-STEM image acquired simultane- a mixed-phase structure, finishing with WZ. We attribute this ously with the EELS signal from a spectrum image of an A-polar to the change in conditions upon growth termination and the GaAs NW in contact with a Ga droplet. Figure 2B depicts the known tendency toward polytypic growth along this polarity.[5] corresponding spectral evolution integrated across a region of Thanks to their intensity dependence on the atomic number the interface, represented by the blue area shown in Figure 2A. as ≈Z1.7–2, HAADF images provide an easily interpretable The bulk plasmon excitation shifts smoothly from a broad and chemically sensitive representation of the structure. As peak in the GaAs to a sharper peak in the liquid Ga, over a expected, the solid shows strong peaks corresponding to the spatial distance of ≈10 nm. Intensity oscillations in the GaAs atomic columns, whereas, further from the interface, the liquid corresponding to the atomic planes are clearly observable due shows a uniform intensity because of its disordered nature. to signal convolution with an elastic scattering contribution. However, close to the interface, the liquid shows intensity fluc- Reference spectra extracted away from the interface at posi- tuations corresponding to an ordering. For the A-polar interface, tions deep in the Ga and GaAs phases (marked P1 and P3 this is limited to one additional layer, while several layers are in Figure 2A), and a spectrum extracted from the interface observed for the B-polar interface. This is confirmed in the inte- (at position P2 in Figure 2A) are presented in Figure 2C,D, grated HAADF intensity profiles shown in the right of panels of respectively. The spectrum at P2 might be assumed to be a linear Figure 1B,C and we attribute it to a longer-range ordering on top combination of the contributions of the liquid and solid phases. of the B-polar interface. The first layer seems to be more clearly However, this cannot correctly describe the P2 spectrum: a structured, with further layers becoming gradually more amor- non-negligible residual is found, as presented in Figure S2 phous. This is in contrast to the findings by Krogstrup et al.,[19] of the Supporting Information. We interpret this residue as in which the first three layers show similarly strong ordering, the interface contribution. To quantitatively study this interface with an abrupt reduction in ordering for the fourth layer. One contribution, we use independent component analysis (ICA), should note that, while high-resolution transmission electron which identifies a finite number of spectral components that microscopy has been employed before to visualize the ordering can reconstruct the spectrum image.[22–24] Although care must of a liquid in contact with a crystalline solid,[14] aberration-cor- be taken when interpreting the components obtained from rected HAADF-STEM offers a more directly interpretable image ICA, it has been successfully applied to unmix independent contrast, free of both thickness/focus related contrast inversions components in low loss EELS data.[22] and delocalization effects. This derives from the incoherent Once scaled in intensity, the first two components in our ICA nature of this imaging mode. analysis, shown in Figure 2C and labeled S and L, match spectra We further analyze the liquid–solid interface using EELS P1 and P3 well and can be safely interpreted as representing the hyperspectral mapping to probe the bulk plasmon response solid GaAs and liquid Ga components. A third ICA component is around the interface. Being related to the valence/free elec- also found, which is interpreted as the “ordered liquid” (OL) con- tron density and characteristics of a material, this technique tribution, dominating the P2 interface signal, see Figure 2D. A can discriminate between different chemical/structural phases. reconstruction of the entire dataset using this three-component Compared to analysis by core-loss EELS or energy dispersive model shows a negligible residual, demonstrated in Figure S2 X-ray spectroscopy, the required electron beam dose is orders of the Supporting Information, confirming that these three of magnitude lower, allowing us to apply the technique without components are sufficient to describe the data. This is further destroying the fragile structure of the interface. emphasized from an analysis of the scree plot in Figure S2 of Low-loss EELS spectrum images of the interface were the Supporting Information, a common tool in multivariate recorded using an atomically sized probe. As an example, statistical analysis used to identify the number of significant Adv. Mater. 2020, 32, 2001030 2001030 (3 of 8) © 2020 The Authors. Published by Wiley-VCH GmbH www.advancedsciencenews.com www.advmat.de sources of information in a dataset, which shows that the first three components represent the majority of variations in the spectrum image.[24] Considering the confined width of the OL region between GaAs and Ga, combined with the delocalized nature of the plasmon oscillations, the low-loss EELS response of the ordered liquid region cannot be isolated by a direct meas- urement. While the effect of an ordered liquid region on the overall low-loss EEL spectra of liquid–solid interfaces has been shown previously,[25] our use of ICA isolates its response for the first time. A similar analysis of B-polar WZ interfaces is dem- onstrated in the Supporting Information, leading to an ordered liquid spectral contribution similar to the A-polar component in the range around the plasmon peak. From these results, it is possible to determine a “bulk plasmon” energy for the OL of 14.6 eV (compared to 13.7 and 15.9 eV for Ga and GaAs), with a full width at half maximum (FWHM) of 2.35 eV (compared with 1.95 and 5.3 eV for Ga and GaAs, respectively). Thus, the peak position and FWHM of the plasmon for the OL sit between those of the liquid Ga and solid GaAs. Because of the single, Lorentzian-shaped bulk plasmon peak for each component, the response is consistent with that given by the Drude model, in which the peak position and its FWHM can be linked to the free electron density, n, and its damping constant, Γ.[26] The results indicate that both n and Γ decrease from the solid GaAs through the ordered liquid to the liquid Ga. The extent of damping depends on the lattice and band structure. For instance, damping can be caused by the transfer of energy from the plasmon reso- nance to single-electron transitions (i.e., creation of e−–h+ pairs). Conversely, the closer the electronic nature of a material approxi- mates a free electron gas, the smaller the damping constant is. It Figure 3. Comparison of intensity/mean atom density profiles from is therefore logical that the solid semiconducting GaAs NW has experimental STEM and MD simulations. The ZB A-polar and WZ B-polar a larger Γ than the metallic liquid Ga, since the lattice and band interfaces are demonstrated in (A) and (B), respectively. The dashed pink line indicates the interface position. structure of the former will increase the probability of energy transfer to single electron transitions. A similar difference is seen between semiconducting Si and metals such as Al. from experimental images for A- and B-polar cases, respec- From this analysis, one can conclude that the ordered tively. For any given polarity, simulations do not show signifi- liquid has a distinct electronic nature, which can intuitively cant differences in the liquid ordering with respect to the ZB be correlated with its semistructured nature. Considering or WZ solid phases. The experimental curves are obtained by this and given the fact that the ordered liquid cannot exist as projecting the intensity profiles from Figure 1B,C along the axis a bulk phase outside of the interface, we propose that it can normal to the surface. In the solid, we obtain regularly ordered be considered as an interfacial complexion.[27–29] The existence peaks at the positions of the dumbbells. The liquid also exhibits of complexion could have profound consequences for crystal some peaks in the density profile, characteristic of atomic-level growth. To understand this, the 3D nature of the ordering and ordering. The range of the ordering is different for the two the influence of the polarity on its structure are key aspects, polarities. As noted earlier, in STEM images, while the B-polar which we shall address next. order is observed for four layers, it does not extend beyond We now investigate the nature of the ordering using MD the first layer for the A-polar case. Similar to the experimental simulations. For this purpose, we use the 2D projection of the observations, the simulations show that the ordering gradu- structure in STEM imaging to validate the MD results. ally diminishes when the distance from the crystalline phase Given the size and time scales needed for these simulations, is increased, which is in contrast to the conclusions given by brute-force ab initio MD is prohibitively expensive. To circum- Algra et al.[19] for InP in contact with liquid InAu. This ten- vent this, we have trained a neural network potential (NNP)[30,31] dency correlates well with the simulations, which show that based on a relatively small number of reference density func- the in-plane order of the A-polar case is more diffuse than that tional theory (DFT) calculations.[32,33] Then, we have used this of B-polar, as demonstrated in Figure S6 of the Supporting machine-learning model to drive the dynamics.[31,34] MD details Information. Note that an exact correlation of the projected in addition to the calculations and validation of the NN based linear density obtained from the simulations with the HAADF- on a multiple-timestep integration[35,36] are discussed in the STEM integrated intensities is not expected: although the latter Supporting Information. scales approximately with the former, a full quantum mechan- Figure 3A,B compare the projected linear density obtained ical image simulation using the atomistic model derived across the simulation cell with intensity line profiles derived from MD would be necessary for a quantitative comparison. Adv. Mater. 2020, 32, 2001030 2001030 (4 of 8) © 2020 The Authors. Published by Wiley-VCH GmbH www.advancedsciencenews.com www.advmat.de The key observation, however, is that there is a quantitative the ensemble for a B-polar WZ solid in contact with liquid Ga. match between the MD predictions of the spacing between Figure 4B provides top and side views for the A- and B-polar ordered layers with the experimental observations. This is solids. In both cases, we find ordering within the plane, how- demonstrated in Table S3 of the Supporting Information, in ever with significant differences in the arrangement of Ga which the experimental observations and MD predictions of atoms as a function of the polarity. the spacing between ordered layers are compared for a B-polar On the B-polar surface, Ga is adsorbed right on top of the interface. We therefore conclude that MD based on the NNP terminal As. This is consistent with the large electronegativity correctly predicts the nature of the liquid ordering. difference between As and Ga that is likely to induce strong The 3D representations from the simulations indicate that electrostatic interactions. In-plane ordering is also present at the the order in the liquid extends fully within the plane parallel A-polar surface. Here, instead, Ga atoms in the liquid pair with to the interface, as illustrated in Figure S6 of the Supporting the terminal Ga atoms in the solid, consistent with the dimeri- Information, consistent with the fact that this order is visible zation tendency of Ga. As expected, at the higher temperature, in HAADF images. One interesting observation is that contrary the range of the order has decreased relative to observations at to what is assumed by Algra et al.,[19] the spacings between con- 300 K, with simulations indicating the presence of at most one secutive ordered liquid layers are not identical. As clear from ordered layer on top of the solid. Most of the qualitative features Table S3 of the Supporting Information, these layers gradually of the liquid ordering, however, are preserved between the two become closer to each other as the distance from the crystalline temperatures, suggesting that experiments performed at 300 K solid is increased. We further find that the order at the liquid– can provide insights into the relevant mechanisms for growth solid interface is dominated by the tendency of Ga atoms to at higher temperatures. form Ga–Ga pairs. Ga atoms form pairs with the solid when We move now to the microscopic picture of GaAs growth. the surface is Ga-terminated, while it forms Ga–Ga pairs in Any new layer of GaAs forms new Ga–As pairs. Ga and As the liquid when the solid is As-terminated. This tendency for atoms are shown in red and blue in Figure 4, while the posi- dimerization is well-known in liquid Ga, which is often referred tions leading to ZB and WZ configurations are indicated by to as a molecular metal.[37] We should note that besides this squares and hexagons, respectively. The inclined dashed lines remarkable ordering at the liquid–solid interface, the liquid provide a guide to define the atomic positions that lead to the remains a liquid as the position of the atoms is dynamic and relevant crystalline structures in Figure 4B,C. In the A-polar thus varies with time. In Figure SI9 of the Supporting Informa- case, As must displace and occupy the position of ordered Ga tion, we provide a comparison of the Ga–Ga radial pair distribu- in the liquid, with which it may form a dumbbell. The Ga tion functions of the solid, liquid, and interface phases as an atoms have two choices, as indicated in Figure 4 by the illustration. square and hexagon. Depending on the position selected, ZB The simulations shown previously were performed at 300 K, (ABCABC stacking) or a twin (ABA stacking) is formed. These corresponding to the temperature used to acquire the STEM two positions are not equivalent in terms of their first and images. To make the link with the growth process, we per- second nearest neighbors configuration. In the ABC stacking, formed the same simulations at the growth temperature (900 K). Ga is found at the middle of a projected hexagon, while in the Figure 4 depicts the ordering of the interface at 900 K for both ABAB it is at a vertex. Further calculations beyond the scope polarities where the spatial variation in the Ga atomic density of this work should corroborate if there is a position that is is indicated by red isosurfaces. Figure 4A depicts a 3D view of more energetically favorable. Still, to form a new bilayer, As Figure 4. Simulations of the Ga(l)–GaAs(s) interface at 900 K. Isocontours of the Ga atom density, showing ordering at the liquid–Ga/GaAs inter- face—corresponding to density ρ = 0.11 Bohr−3 (opaque) and ρ = 0.06 Bohr−3 (translucent). As and Ga atoms are drawn in blue and red, respectively. Objects further from the viewer are represented with less contrast as a depth cue. A) 3D view of WZ GaAs terminated with B-polarity; B) views along the [111] (left) and [1 10] ZB/[112 0] WZ (right) directions; in the case of A-polar ZB and B-polar WZ (top and bottom, respectively). Similar isocontours presented for simulations at 300 K in the Supporting Information. We indicate the positioning of the As and Ga atoms that are required to create a new Ga–As bilayer, taking the isocontours and the underlying structure as the canvas. Adv. Mater. 2020, 32, 2001030 2001030 (5 of 8) © 2020 The Authors. Published by Wiley-VCH GmbH www.advancedsciencenews.com www.advmat.de atoms should first displace the ordered Ga layer on top of the SmartAlign plug-in for DigitalMicrograph.[39] Consecutive frames the NW, and this should increase the formation barrier. This were acquired at a 90° rotations using a low beam current (<40 pA). The electron probe convergence semiangle and HAADF detector inner is consistent with the difficulty in synthesizing A-polar GaAs collection semiangles were 20 and 50.5 mrad. Fischione photomultiplier NWs,[5,21] suggesting that the slower process may help the tube was used as the HAADF detector. Pixel dwell time of 0.5 µs and growing layer achieve the more thermodynamically stable pixel step size ≈6 pm were employed. The images were captured with ABC stacking. 2k × 2k size. The associated low probability of twinning also reduces the EELS Hyperspectral Imaging: The data were acquired with the same probability of WZ formation, as both are closely connected.[38] microscope, HT and probe convergence semiangle, using a Gatan GIF Experimentally, the A-polar GaAs NWs exhibit mostly a ZB Quantum ERS spectrometer with Ultrascan 2k × 2k detector. In order to avoid damaging the interface, the probe current was set to 75 pA structure with an absence of transverse twins.[5,21] and a scan step size of ≈0.9 Å was applied. The energy dispersion was During the formation of a new bilayer for the B-polar sur- set to 0.05 eV per ch, while the zero-loss peak FWHM was 1.05 eV. face, the Ga atoms adsorbed on top of the As-terminated sur- A spectrometer entrance aperture of 2.5 mm, corresponding to a face can stay at their position. The incoming As atoms occupy collection semiangle of 47 mrad was employed. For the measurements empty positions in the second row. The fact that the growth on the A-polar NW discussed in Figure 2 and in the Supporting can proceed without displacing Ga atoms is consistent with Information, per spectrum pixel time was 0.25 ms, while this value was 0.5 ms for the measurement on B-polar NW reported in the Supporting the observation of a more facile growth for this polarity. The Information. For all the EELS data, detector was binned by 130× on higher growth rate can also partly explain the higher propensity nondispersive axis and it was operating in high speed mode. “High to introduce stacking defects and polytypism in B-polar GaAs quality dark correction” was applied. Processing the data started with NWs.[5] These results could explain why in polar semiconduc- removing spectral spikes from cosmic rays, centering the zero-loss tors, growth in a certain polarity is preferred and how polarity peak to compensate for HT fluctuations and removing it, removing determines the tendency for polytypism. plural scattering and the plasmon peak replica. Then, second-order ICA was performed using the DigitalMicrograph plugin described by Lucas In conclusion, we have demonstrated that interfaces between et al.,[24] details of which are presented in the Supporting Information. a solid GaAs NW and the liquid Ga droplet exhibit a 3D order Molecular Dynamics: To investigate the ordering at the interface at that depends on the polarity of the solid and the temperature. the two (111) surfaces, an orthorhombic supercell composed of a central This order can be explained by the electrostatic interac- solid GaAs section (144 atoms, corresponding to 6 layers of 24 atoms), tions within the liquid and the solid termination. Detailed in contact with liquid Ga (192 atoms) on both of its surfaces, totaling simulations of the interface, enabled by the combination of 336 atoms, was used. first-principles calculations and machine-learning, provide The initial lattice parameter for the solid part was set to the one obtained from DFT calculations, whereas the initial density of liquid an atomistic picture of the growth process from the liquid, Ga was set to that obtained with independent simulations in a smaller the likelihood for growth in a certain polarity and defect for- (96 atoms) box at the objective temperature. All the simulations were mation. This work lays the foundations for further atomistic run using i-PI[36] in combination with large-scale atomic/molecular studies of crystal growth processes. While our results have been massively parallel simulator,[34] and n2p2[31] to evaluate the NNP. obtained for the Ga(l)–GaAs(s) system, we believe they are of First, the system was equilibrated in the NσT ensemble, allowing the a general character. Similar ordering features likely play an cell degrees of freedom to change independently. After equilibration, production simulations were run in the NVT ensemble, using the important role in the crystal growth of other materials and in average lattice parameters, at the temperatures indicated in the text. more general solidification processes, raising new questions for The temperatures were controlled using a combination of a groups observing NW growth by in situ electron microscopy. generalized Langevin[40] and stochastic velocity rescaling[41] thermostats. This work also opens analogous perspectives in other fields Pressures, where applicable, were constrained using an anisotropic such as catalysis and (photo)electrochemistry. barostat.[42] Simulations were run with a timestep of 4 fs (at 300 K) and 2 fs (at 900 K), for a total of 10 ns. Neural Network and Ab Initio Calculations: A neural network potential Experimental Section of the Behler–Parrinello kind[30] was trained to reproduce the results obtained with DFT calculations. A total of 970 structures were used to fit NW Growth: GaAs NWs were grown on GaAs (100) substrates the potential, with a 90/10 split between training and test. At the end of as described by Zamani et al.[21] HF etching was used to remove the the training procedure, the average error in the test set was 2.4 meV per substrate native oxide. Silicon oxide layers were obtained by spin-coating atom for energies and 120 meV Å−1 for forces. a solution of hydrogen silsesquioxane and methyl isobutyl ketone. The DFT calculations were carried out with the plane-wave code The spin-coated layers were annealed at 300 °C for 10 min. The oxide Quantum-Espresso,[32] using the Perdew–Burke–Ernzerhof exchange- thicknesses used for the growth was around 3.3 nm. Two additional correlation functional,[43] and ultrasoft pseudopotentials[33] from the annealing steps at 150 and 300 °C, lasting 2 h each were performed SSSP library (version 0.7).[44] inside the MBE chamber. The growth conditions for the NWs were: a The wavefunction energy cutoff was set to 50 Ry for all calculations, V/III ratio of 1.9, equivalent 2D Ga-limited GaAs growth rate of 0.24 Å s−1, whereas the kinetic energy cutoff was set to 400 Ry. at a substrate temperature of 640 °C. Energy and forces for the structures used as training points were HAADF-STEM Imaging: The STEM samples were prepared by computed with a converged (<1 meV per atom absolute convergence) scraping as-grown samples onto Cu TEM grids with amorphous 3 × 3 × 1 Monkhorst–Pack k-point grid.[45] holey carbon films. The STEM imaging was performed using a double aberration-corrected FEI Titan Themis 60–300, operated at 300 kV high tension (HT). To minimize both electron-beam induced damage to the Supporting Information interface and scan distortions from, e.g., sample drift, the HAADF- STEM images were acquired as 24-frame image series, which were then Supporting Information is available from the Wiley Online Library or averaged while correcting for rigid and nonrigid displacements using from the author. Adv. Mater. 2020, 32, 2001030 2001030 (6 of 8) © 2020 The Authors. Published by Wiley-VCH GmbH www.advancedsciencenews.com www.advmat.de Acknowledgements [7] D. Jacobsson, F. Panciera, J. Tersoff, M. C. Reuter, S. Lehmann, S. Hofmann, K. A. Dick, F. M. Ross, Nature 2016, 531, 317. M.Z., N.T., and A.F.i.M. were supported by SNF Back-Up Schemes [8] L. Ghisalberti, H. Potts, M. Friedl, M. Zamani, L. Güniat, Consolidator Grant Nos. BSC-GI0_157705 and NCCR QSIT, G.I. and G. Tütüncüoglu, W. C. Carter, A. F. I. Morral, Nanotechnology 2019, M.C. were supported by the NCCR MARVEL, funded by the Swiss 30, 285604. National Science Foundation. ICN2 acknowledges funding from [9] P. Krogstrup, H. I. Jørgensen, E. Johnson, M. H. Madsen, Generalitat de Catalunya 2017 SGR 327 and the Spanish MINECO project C. B. Sørensen, A. F. I. Morral, M. Aagesen, J. Nygård, F. Glas, ENE2017-85087-C3. ICN2 was supported by the Severo Ochoa program J. Phys. D: Appl. Phys. 2013, 46, 313001. from Spanish MINECO (Grant No. SEV-2017-0706) and was funded by [10] V. G. Dubrovskii, Cryst. Growth Des. 2017, 17, 2544. the CERCA Programme/Generalitat de Catalunya. Part of the present [11] J. Q. Broughton, G. H. Gilmer, J. Chem. Phys. 1983, 79, 5095. work was performed in the framework of Universitat Autònoma de Barcelona Materials Science Ph.D. program. S.M.-S. acknowledges funding [12] H. E. A. Huitema, M. J. Vlot, J. P. Van Der Eerden, J. Chem. Phys. from “Programa Internacional de Becas “la Caixa”-Severo Ochoa.” The 1999, 111, 4714. authors acknowledge support from CSIC Research Platform on Quantum [13] S. Toxvaerd, Faraday Symp. Chem. Soc. 1981, 16, 159. Technologies PTI-001. SuperSTEM is the UK National Research Facility for [14] S. H. Oh, Y. Kauffmann, C. Scheu, W. D. Kaplan, M. Rühle, Science Advanced Electron Microscopy, supported by the Engineering and Physical 2005, 310, 661. Sciences Research Council (EPSRC). [15] M. Gandman, Y. Kauffmann, W. D. Kaplan, Appl. Phys. Lett. 2015, 106, 051603. [16] Y. Kauffmann, S. H. Oh, C. T. Koch, A. Hashibon, C. Scheu, M. Rühle, W. D. Kaplan, Acta Mater. 2011, 59, 4378. Conflict of Interest [17] O. G. Shpyrko, R. Streitel, V. S. K. Balagurusamy, A. Y. Grigoriev, M. Deutsch, B. M. Ocko, M. Meron, B. Lin, P. S. Pershan, Science The authors declare no conflict of interest. 2006, 313, 77. [18] T. U. Schülli, R. Daudin, G. Renaud, A. Vaysset, O. Geaymond, A. Pasturel, Nature 2010, 464, 1174. [19] R. E. Algra, V. Vonk, D. Wermeille, W. J. Szweryn, M. A. Verheijen, Author Contributions W. J. P. Van Enckevort, A. A. C. Bode, W. L. Noorduin, E. Tancini, A. E. F. De Jong, E. P. A. M. Bakkers, E. Vlieg, Nano Lett. 2011, 11, 44. M.Z. and G.I. contributed equally to this work. The samples were grown by M.Z. The STEM images and EELS data were captured by D.T.L.A. [20] C. Colombo, D. Spirkoska, M. Frimmer, G. Abstreiter, Processing of STEM images was done by M.Z. M.Z. and N.T. performed A. Fontcuberta, I. Morral, Phys. Rev. B. 2008, 77, 155326. the analysis of EELS spectra and the application of ICA. D.T.L.A. and [21] M. Zamani, G. Tütüncüoglu, S. Martí-Sánchez, L. Francaviglia, C.H. helped with discussions on the application of ICA and its results. L. Güniat, L. Ghisalberti, H. Potts, M. Friedl, E. Markov, W. Kim, MD simulations were done by G.I. and S.B. under the supervision of J. B. Leran, V. G. Dubrovskii, J. Arbiol, A. Fontcuberta I Morral, M.C., M.Z., N.T., L.F., and G.I. made the figures. S.M.-S., Q.M.R., and Nanoscale 2018, 10, 17080. J.A. contributed to the discussions on the electron microscopy aspects [22] N. Bonnet, D. Nuzillard, Ultramicroscopy 2005, 102, 327. of this research and obtained complementary STEM images and EELS [23] F. de la Peña, M. H. Berger, J. F. Hochepied, F. Dynys, O. Stephan, spectra. M.Z. and G.I. made the Supporting Information section. The M. Walls, Ultramicroscopy 2011, 111, 169. manuscript was mainly written by A.F.i.M., N.T., and M.C. For discussion [24] G. Lucas, P. Burdet, M. Cantoni, C. Hébert, Micron 2013, 52–53, 49. on implications of results on NW growth, M.Z., M.F., L.G., G.T., L.F., [25] M. Gandman, Y. Kauffmann, W. D. Kaplan, Appl. Phys. Lett. 2015, G.I., M.C., and A.F.i.M. had input. G.T. and A.F.i.M. proposed and 106, 051603. conceptualized the idea. M.Z. and A.F.i.M. coordinated the collaboration [26] R. F. Egerton, Electron Energy-Loss Spectroscopy in the Electron Micro- between different groups. A.F.i.M. directed the work. All the authors have scope, Springer, USA 2011. contributed to the correction and proofreading. [27] W. D. Kaplan, D. Chatain, P. Wynblatt, W. C. Carter, J. Mater. Sci. 2013, 48, 5681. [28] M. Tang, W. C. Carter, R. M. Cannon, Phys. Rev. B: Condens. Matter Keywords Mater. Phys. 2006, 73, 024102. [29] S. J. Dillon, M. Tang, W. C. Carter, M. P. Harmer, Acta Mater. 2007, liquid ordering, liquid–solid interface, nanowires 55, 6208. [30] J. Behler, M. Parrinello, Phys. Rev. Lett. 2007, 98, 146401. Received: February 13, 2020 [31] A. Singraber, T. Morawietz, J. Behler, C. Dellago, J. Chem. Theory Revised: June 29, 2020 Comput. 2019, 15, 3075. Published online: August 6, 2020 [32] P. Giannozzi, O. Andreussi, T. Brumme, O. Bunau, M. Buongiorno Nardelli, M. Calandra, R. Car, C. Cavazzoni, D. Ceresoli, M. Cococcioni, N. Colonna, I. Carnimeo, A. Dal Corso, S. De Gironcoli, P. Delugas, R. A. Distasio, A. Ferretti, A. Floris, [1] R. S. Wagner, W. C. Ellis, Appl. Phys. Lett. 1964, 4, 89. G. Fratesi, G. Fugallo, R. Gebauer, U. Gerstmann, F. Giustino, [2] C. Y. Wen, J. Tersoff, M. C. Reuter, E. A. Stach, F. M. Ross, Phys. Rev. T. Gorni, J. Jia, M. Kawamura, H. Y. Ko, A. Kokalj, E. Kücükbenli, Lett. 2010, 105, 195502. M. Lazzeri, M. Marsili, N. Marzari, F. Mauri, N. L. Nguyen, [3] J. C. Harmand, G. Patriarche, F. Glas, F. Panciera, I. Florea, H. V. Nguyen, A. Otero-De-La-Roza, L. Paulatto, S. Poncé, D. Rocca, J. L. Maurice, L. Travers, Y. Ollivier, Phys. Rev. Lett. 2018, 121, 166101. R. Sabatini, B. Santra, M. Schlipf, A. P. Seitsonen, A. Smogunov, [4] F. Glas, J. C. Harmand, G. Patriarche, Phys. Rev. Lett. 2007, 99, I. Timrov, T. Thonhauser, P. Umari, N. Vast, X. Wu, S. Baroni, 146101. J. Phys. Condens. Matter 2017, 29, 465901. [5] X. Yuan, P. Caroff, J. Wong-Leung, L. Fu, H. H. Tan, C. Jagadish, Adv. [33] D. Vanderbilt, Phys. Rev. B 1990, 41, 7892. Mater. 2015, 27, 6096. [34] S. Plimpton, J. Comput. Phys. 1995, 117, 1. [6] R. R. Zamani, S. Gorji Ghalamestani, J. Niu, N. Sköld, K. A. Dick, [35] M. Tuckerman, B. J. Berne, G. J. Martyna, J. Chem. Phys. 1992, 97, Nanoscale 2017, 9, 3159. 1990. Adv. Mater. 2020, 32, 2001030 2001030 (7 of 8) © 2020 The Authors. Published by Wiley-VCH GmbH www.advancedsciencenews.com www.advmat.de [36] V. Kapil, M. Rossi, O. Marsalek, R. Petraglia, Y. Litman, [39] L. Jones, H. Yang, T. J. Pennycook, M. S. J. Marshall, S. Van Aert, T. Spura, B. Cheng, A. Cuzzocrea, R. H. Meißner, D. M. Wilkins, N. D. Browning, M. R. Castell, P. D. Nellist, Adv. Struct. Chem. B. A. Helfrecht, P. Juda, S. P. Bienvenue, W. Fang, J. Kessler, Imaging 2015, 1, 8. I. Poltavsky, S. Vandenbrande, J. Wieme, C. Corminboeuf, [40] M. Ceriotti, G. Bussi, M. Parrinello, J. Chem. Theory Comput. 2010, T. D. Kühne, D. E. Manolopoulos, T. E. Markland, J. O. Richardson, 6, 1170. A. Tkatchenko, G. A. Tribello, V. Van Speybroeck, M. Ceriotti, [41] G. Bussi, D. Donadio, M. Parrinello, J. Chem. Phys. 2007, 126, Comput. Phys. Commun. 2019, 236, 214. 014101. [37] W. J. Hulsman, J. F. Peters, M. J. Zwanenburg, S. A. De Vries, [42] P. Raiteri, J. D. Gale, G. Bussi, J. Phys. Condens. Matter 2011, 23, T. E. Derry, D. Abernathy, J. F. Van der Veen, Nature 1997, 390, 334213. 379. [43] J. P. Perdew, K. Burke, M. Ernzerhof, Phys. Rev. Lett. 1996, 77, 3865. [38] J. Arbiol, A. Fontcuberta I Morral, S. Estradé, F. Peiró, B. Kalache, [44] G. Prandini, A. Marrazzo, I. E. Castelli, N. Mounet, N. Marzari, npj P. Roca I Cabarrocas, J. R. Morante, J. Appl. Phys. 2008, 104, Comput. Mater. 2018, 4, 72. 064312. [45] H. J. Monkhorst, J. D. Pack, Phys. Rev. B 1976, 13, 5188. Adv. Mater. 2020, 32, 2001030 2001030 (8 of 8) © 2020 The Authors. Published by Wiley-VCH GmbH