Independent Control of Nucleation and Layer Growth in Nanowires.

Control of the crystallization process is central to developing nanomaterials with atomic precision to meet the demands of electronic and quantum technology applications. Semiconductor nanowires grown by the vapor-liquid-solid process are a promising material system in which the ability to form components with structure and composition not achievable in bulk is well-established. Here, we use in situ TEM imaging of Au-catalyzed GaAs nanowire growth to understand the processes by which the growth dynamics are connected to the experimental parameters. We find that two sequential steps in the crystallization process-nucleation and layer growth-can occur on similar time scales and can be controlled independently using different growth parameters. Importantly, the layer growth process contributes significantly to the growth time for all conditions and will play a major role in determining material properties such as compositional uniformity, dopant density, and impurity incorporation. The results are understood through theoretical simulations correlating the growth dynamics, liquid droplet, and experimental parameters. The key insights discussed here are not restricted to Au-catalyzed GaAs nanowire growth but can be extended to most compound nanowire growths in which the different growth species has very different solubility in the catalyst particle.

A central challenge in crystal growth is to understand the dynamic and transient processes underlying the nucleation and growth steps. The ability to independently control these two steps would greatly expand the potential to design the structure, morphology, and properties of the resulting material. Understanding the steps in crystallization is particularly important in the emerging fields of nanoscale and confined crystal growth, where a large interface-to-volume ratio enables new degrees of freedom in designing crystal properties.[1,2] Vapor–liquid–solid (VLS) growth of nanowires, which uses a nanoscale liquid melt to control nucleation and crystal growth, is a promising and versatile growth technique.[3−5] For semiconductor materials, the VLS growth process enables synthesis of metastable crystal phases,[4−9] metastable semiconductor alloys,[5,10,11] extremely high dopant and impurity incorporation,[12−14] and atomically precise lattice-mismatched[15−17] and crystal-phase heterostructures.[18−20] Understanding the nucleation and crystallization processes at the interface in VLS-growth, and how these are influenced by the finite size of the liquid droplet, is centrally important for controlled design of these nanomaterials. In particular, it is critical to understand on an atomic scale the mechanisms by which the growth process is correlated to accessible experimental parameters.

The VLS process is understood to occur by dissolution of semiconductor precursor species (or their derivatives) in a liquid metal (often Au-based), followed by precipitation of the solid semiconductor after the liquid becomes supersaturated.[21,22] This process occurs in two steps: formation of a critical nucleus at the liquid–solid interface followed by layer growth across this interface.[23,24] Since the metal droplet is relatively small, the number of atoms causing supersaturation of the liquid is finite and could be consumed during the formation of a layer.[25] After the growth of a complete layer following the nucleation, there typically is an “incubation” (waiting) period before the nucleation of the next layer.[26−28] Most theoretical models on nanowire growth have assumed the layer completion time to be negligible compared to incubation time.[26,29−31] In contrast with theoretical models, in situ investigations have in some cases demonstrated slow layer growth (and relatively insignificant incubation time);[27,28,32,33] however, these experiments were conducted at low precursor pressures yielding very low growth rates, and hence, were generally not considered to be representative of typical ex situ growths. In situ experiments at conditions similar to typical ex situ growths are thus necessary to understand the growth kinetics in the widely used conventional (ex situ) systems. This understanding combined with the serial nature of the nucleation and incubation steps suggest the potential to separately access the parameters controlling them and to use this as a means to design material properties.

In this work, we investigate GaAs nanowire growth using in situ transmission electron microscopy and demonstrate that nucleation and layer growth can be controlled independently. By correlating the incubation time and layer growth time to the partial pressures of Ga and As precursors, we observe that the time scales for the two steps are of similar magnitude but follow entirely different trends. For most of the studied parameter space, incubation can be tuned by varying the Ga precursor partial pressure, while the layer growth can be independently controlled using the flow of the As precursor. Toward the edges of the parameter space, interactions between these species lead to more complicated trends. Simulations of the growth show that the observed effects can be understood by considering the very different steady-state compositions of the two species (Ga and As) in the catalyst droplet during growth. Since this effect is fundamental in nature, it can be directly extrapolated to other growth conditions, growth methods and other binary semiconductor materials.

Results and Discussions

Nanowire Growth

Gold is the most commonly used catalyst for growing III–V nanowires;[4,34] hence, GaAs nanowires were grown in this study using Au nanoparticles. GaAs nanowires were grown in a Hitachi HF-3300S environmental TEM on SiN-coated heating chips on which Au aerosol nanoparticles (∼30 nm diameter) had been deposited. Trimethylgallium (TMGa) and arsine (AsH3) were used as the precursors as in a typical ex situ metal–organic chemical vapor deposition (MOCVD) system. These precursors were supplied via capillary tubes fed through the sample holder such that the gases entered the microscope within 4 mm of the sample area. Gas flows were controlled by mass flow controllers and pressure valves, and monitored during growth with a residual gas analyzer in the exhaust gas which had been calibrated to give partial pressures at the sample. Gas flows were chosen to give AsH3 partial pressure in the same range as typical ex situ MOCVD nanowire growth, with slightly lower TMGa flows giving average nanowire growth rates in the range of 0.1–1 nm/s (comparable to typical ex situ GaAs nanowire growth rates of about 0.5–5 nm/s[35,36]). Details of the experimental setup are found in the Methods. TMGa and AsH3 were supplied to initiate nanowire growth, after which a suitable nanowire was selected based on its orientation relative to a hole in the supporting SiN film and to other nanowires, in order to follow the growth as parameters were varied. Nanowire growth was imaged continuously with 50 ms exposure time per image using conventional parallel beam TEM.

We observe that growth occurs layer-by-layer across the nanowire interface (Figure a–f), consistent with previous reports.[27,28,32] There is also a measurable “incubation time” between successive layer growths, which we interpret as a waiting period where the droplet accumulates enough material to overcome the nucleation barrier for a new layer. This interpretation implies that there is a “nucleation step” during which the critical nucleus (and potentially a small part of the layer) forms. This step is extremely fast and not visible in such experiments, and so will not be considered in this analysis. In the following we use the terms “layer completion time” for the time observed for completion of each GaAs layer and “incubation time” as the time between completion of one layer and the starting of the next layer. (In the literature, the “layer completion time” is also sometimes referred to as “step-flow time”.) Interestingly, we observe that the order of magnitude of the incubation and layer completion times is similar, with layer completion times ranging from 0.1–0.5 s, and incubation times ranging from about 0.2 to 5 s; an example is shown in Figure g,h. This is seemingly in contrast to the theoretical models on nanowire growth that assumed the layer completion time to be instantaeous.[26,29−31] Our observations suggest that, for conditions comparable to typical ex situ nanowire growth, both steps–incubation and layer completion–contribute in a significant way to the overall growth process.

Figure 1

Layer growth in GaAs nanowires. (a) TEM image of a growing GaAs nanowire NW along with the catalyst. A partially grown layer can be observed in this image (indicated by the arrow). (b–f) Sequence of frames (cropped) showing layer growth from a recorded video of another nanowire. The frame number (fr.) and time elapsed from the image in (b) is denoted on the top right of each section. The layer completion time, i.e., the time between the starting and ending of this layer corresponds to 4 frames (i.e., 0.22 s). After the ending of a layer in frame 5 (e) a new layer starts only in frame 11 (f). Thus, the “incubation time” or the waiting time in this example is 6 frames (i.e., 0.33 s). Scale bars indicate 5 nm. (See also Supplementary Video 1.) (g, h) Histograms of layer completion time and incubation time, respectively, of some layers grown at the same growth condition. (The frame rate for the video is 18.3 fps on an average.).

Kinetics as a Function of As-Precursor Supply

In order to understand the mechanisms controlling the incubation and layer growth, we first discuss the dependence of growth kinetics on the AsH3 flow at 420 °C and TMGa pressure of 13 × 10–5 Pa (Figure ). AsH3 partial pressure was initially set to 1 Pa (see also Supplementary Video 2) and subsequently increased monotonically. After conducting growth at the highest AsH3 pressure (5.6 Pa), the AsH3 flow was reduced backto 1 Pa to verify reproducibility of the observations. The time for completing each layer (purple squares) decreases with increasing AsH3 supply, indicating that layer completion is restricted by the amount of arsenic present in the catalyst. This suggests that there are not sufficient (excess) As atoms present in the droplet at any instant to form a complete layer. This is not surprising since the solubility of As in the Au–Ga alloy is very low, and experimental reports[37,38] and theoretical predictions[39] show very low concentration of As species in the catalyst. Incubation time is plotted as a function of the AsH3 partial pressure as cyan circles in Figure ; we see that incubation, unlike layer completion, increases slightly as we increase the As precursor pressure. There was no visible change in the catalyst volume or nanowire geometry over this series.

Figure 2

Incubation and layer completion as a function of AsH3. The average layer completion time (purple squares, axis on left side) and incubation time (cyan circles) for growing a layer is plotted as a function of the As-precursor pressure. The layer growth of each individual layer is faster at higher AsH3 flow. Error bars denote standard deviation across the measured 10 or more events.

Kinetics as a Function of Ga-Precursor Supply

Next, we discuss the effect of TMGa pressure while keeping the AsH3 constant (1 Pa) at the same temperature of 420 °C (Figure ). TEM images of the nanowire catalyst during growth are shown in Figure a,b. At the low Ga fluxes investigated, the volume (and the Ga content) of the catalyst remains almost the same, while at the highest Ga flux shown here there is a noticeable volume increase indicating more Ga content in the catalyst.[37] As shown in Figure c, we observe that incubation time decreases with increasing TMGa flow (i.e., nucleation occurs after a shorter average waiting time). This trend is strongest at low TMGa flow, while for high TMGa flow, the incubation time is very brief, and there is very little change with increasing flow. There is a large variability in incubation times measured for different layers growing at low TMGa flow, resulting in large error bars; for higher flows, however, the spread is very small (not visible on the scale of the figure). On the other hand, the layer completion time does not depend in an obvious way on TMGa flow for most of the studied TMGa flows (9–60 × 10–5 Pa). For the very lowest TMGa flow investigated (9 × 10–5 Pa), there is a slight increase in layer completion time, suggesting that for this extremely low Ga supply there might not be sufficient excess Ga atoms to form one complete layer. Finally, we observe that for a significant part of the parameter space covered in the measurement (TMGa > 25 × 10–5 Pa), the layer completion time is actually longer than the incubation time. We note that this experiment was conducted on a different nanowire and day, and so the absolute values of the incubation time and layer completion time for the same “set” experimental parameters differ slightly between experiments. This may be related to differences in local environment at the nanowire or the pumping efficiency of the instrument on different days; however, the observed qualitative trends are reproducible.

Figure 3

Incubation and layer completion as a function of TMGa. TEM images of the nanowire catalyst at a low TMGa partial pressure of 9 × 10–5 Pa is shown in (a) and that at 56 × 10–5 Pa is shown in (b). (c) Layer completion time (purple squares) and incubation time (cyan circles) are plotted as a function of the Ga-precursor flow at fixed AsH3 partial pressure of 1 Pa. With increasing Ga-precursor flux, the incubation time decreases, indicating that the nucleation of each new layer is controlled by the Ga supply to the catalyst particle. For TMGa pressure above 10 × 10–5 Pa, the layer completion time stabilizes at a nonzero value indicating that layer growth is limited by the As availability. Error bars denote standard deviation across the measured 10 or more events.

Simulation of the Growth of Layers in GaAs Nanowires

The observations show that under typical growth conditions nucleation of a new layer is controlled by the Ga species, while the arrival of As species controls the layer formation. To understand this better, we conducted Monte Carlo simulations of the nanowire growth based on mass transport and nucleation theory, which uses the effective impingement rates of As and Ga as the main input parameters.[39] A complete description of the model is found in Supporting Information section S1; here we summarize the simulation process and how it was used in this study. The composition of the seed particle is calculated iteratively for a fixed time interval (dt = 1 ms) based on the effective impingements, the evaporation of As and Ga and potentially on Ga–As pairs incorporated in the layer growth. The nucleation rates, which depend on the properties of the seed (such as supersaturation), are calculated using classical nucleation theory, modified to take droplet depletion[25] into consideration. Whether or not a nucleation event occurred in the time step is decided by random numbers and the nucleation probabilities. Once a nucleation event has occurred (and layer growth is not yet completed), Ga–As pairs are moved from the liquid seed to the solid layer in each time step, as long as the supersaturation is high enough for it to be energetically favorable to grow the nucleus further. The layer completion time and incubation time are thus obtained for each cycle.

An illustrative example of the simulated growth is shown in Figure . The Ga and As concentrations in the catalyst droplet (and several other related parameters) change in a cyclic way: increase until a critical supersaturation is reached, and then decrease when nucleation occurs (i) and a layer forms (ii). Following completion of a layer (iii), the concentrations again increase. Arsenic has a high vapor pressure and very low equilibrium solubility in Au;[39] droplet supersaturation is thus very sensitive to the addition of As atoms, and As concentration rapidly equilibrates with the ambient AsH3 pressure. Gallium, on the other hand, easily forms metallic liquid alloys with Au,[40] and the Ga concentration in the droplet is high during nanowire growth (measured experimentally to be on the order of 25–55 atomic %),[37] with additional Ga atoms having a relatively much smaller effect on the droplet supersaturation. In order for nucleation to occur, the species in the droplet must exceed the nucleation barrier as determined by the supersaturation of Ga and As in the droplet relative to the GaAs crystal. When the As concentration has equilibrated with the vapor, the probability of overcoming this barrier is typically low enough to prevent nucleation. While the As concentration remains flat, the Ga concentration continues to increase (Figure a) resulting in a higher nucleation probability. Thus, the nucleation of a new layer is controlled by Ga for most of the parameter space. When a nucleation event occurs the number of available As atoms drops rapidly and is insufficient to form a complete monolayer, and so layer growth will primarily be controlled by the rate at which As atoms arrive.

Figure 4

Simulation of the layer growth process in GaAs nanowires. (a) Simulation of the Ga and As concentration in the Au–Ga–As catalyst droplet as a function of time. The collection and depletion of (excess) Ga is relatively slower than As. The As concentration is several orders of magnitude lower than Ga concentration. Different stages of layer growth (nucleation, progression of a layer and completion of a layer) are denoted in the figure as i, ii, and iii, respectively. (b) Schematic progression (not simulated) of these steps is shown for illustrative purposes; this schematic is not intended to display the real geometry of the growing layer, which evolves in a complex way as in ref (28).

Monte Carlo simulations were performed for about 35 cycles of growth for each precursor flow to find the average layer completion and incubation times (Figure ). The error bars represent the standard deviation among the multiple “grown” layers in the simulation. We can see that the simulation reproduces the trends observed experimentally (Figure , 3): Layer completion time decreases strongly with increasing As partial pressure while showing a very weak, almost flat, dependence on Ga partial pressure. Incubation, on the other hand, decreases strongly with increasing Ga partial pressure while increasing slightly with increasing As partial pressure. The observation of a small increase in incubation with As pressure in the simulation allows us to understand its origin: the Ga atoms required for supersaturating and growing the layers are collected in the catalyst during both the incubation and layer growth steps; so when the layer completion time becomes smaller at high AsH3 pressures, the incubation time becomes longer correspondingly so as to accumulate the required Ga atoms. Finally, the simulation also shows that the spread in incubation times between individual layers is much larger for low Ga partial pressure, and that layer-to-layer variability in completion time is very small for all conditions.

Figure 5

Simulation of incubation and layer growth as a function of precursor pressures. Layer completion time (square) and incubation time (circle) obtained from Monte Carlo simulations for (a) the AsH3 series and (b) TMGa series. The error bars represent the standard deviation among “grown” layers.

Understanding the Nanowire Growth Parameter Space

This section discusses nanowire growth regimes in the context of terminology used in the nanowire community. Experimental reports typically refer to “As-limited” (or V-limited) and “Ga-limited” (or III-limited) growth, assuming that if one species is in excess, the growth is controlled by the other species. The observation that incubation/nucleation and layer growth can be controlled by different species means however that the overall picture is more complex. A complete picture of GaAs nanowire growth can be obtained by combining our experimental and simulation results to consider the entire parameter space. On the basis of this understanding we identify the following regimes (summarized in Table ):

Table 1

Growth Regimes: Different Growth Regimes Can Be Identified According to What Species Determines the Layer Growth and Nucleation Processesa

The different regimes are ordered here in a way that on the left column As flow is low (i.e., Ga flow is relatively high) and on the right column it reverses to high As flow (i.e., Ga flow is relatively low).

As-Limited Regime

At low AsH3 flow and low V/III ratio, As limits both incubation (which will be very short, approaching zero) and layer growth (which can become very long) and hence growth becomes As-limited. In this regime Ga will rapidly accumulate in the droplet, and so the parameter space for stable growth is very narrow.[35] (Moderately high Ga accumulation in the seed particle can increase the droplet contact angle resulting in truncated corners that could in turn modify the layer growth dynamics.[27] The experiments discussed here are performed intentionally at conditions where a truncation is not present. Hence, the effects of truncation on the layer growth kinetics are not considered in these simulations.)

Bi-Limited Regime

At slightly higher AsH3 flow than the As-limited regime, there would exist an intermediate regime where the steady-state Ga and As concentrations in the droplet are correlated and incubation time determined by both flows, while layer growth is still controlled by AsH3. In this regime, the droplet composition will be a strong function of TMGa pressure, but Ga does not accumulate significantly with time at any fixed TMGa pressure. This behavior was described in a previous in situ study for high TMGa pressures[37] and can be observed for the highest TMGa pressure studied here (Figure b).

Semi-Ga-Limited Regime

With further increase in AsH3 flow the V/III ratio is high enough that the As concentration in the droplet rapidly equilibrates with the vapor; consequently, TMGa flow will control nucleation, but AsH3 will still control layer completion. Almost the entire parameter space accessible in our in situ experiments falls within this regime.

Quasi-Ga-Limited Regime

At even higher V/III ratio (with moderate or low Ga supply, but high AsH3 flow) there exists another intermediate regime where both Ga and As availability at the growth-front determine the layer completion time. The slight increase of layer completion time observed at the lowest TMGa flux studied experimentally might be an indication of this regime.

True-Ga-Limited Regime

At very high absolute AsH3 flow the incubation time would be limited by Ga. The layer completion could either be instantaneous (if both Ga and As are sufficiently high) or be controlled by Ga (for very low Ga). For conditions where both the incident Ga and As precursor fluxes are high, the simulation predicts layer completion time tending to zero (due to the surplus supply of atoms stored in the liquid), but incubation time remains non-zero and limited by Ga. According to rough estimates from the simulations, arsenic pressures of over 40 Pa would be a minimum requirement to have more As atoms in the seed particle than in one complete bilayer, and thus allow a complete layer to form instantly (calculated for 420 °C and nanowire radius = 16 nm). AsH3 partial pressures close to 40 Pa are possible in ex situ MOCVD but rarely used for GaAs nanowire growth.[35] Most reports on ex situ growth use AsH3 partial pressures in the 0.5–6 Pa range, so we anticipate that layer growth would not be instantaneous under most common MOCVD conditions.

In the experiments reported in this paper the growth was never observed to be either exclusively As-limited or True Ga-limited. Experimental conditions for obtaining As-limited growth are achievable experimentally, for instance, in the case of Ga-catalyzed GaAs nanowire growth,[41−43] but since Ga accumulates in the droplet with time, nanowire stability is limited and the parameter range for successful growth is typically very small. True Ga-limited growth of Au-catalyzed GaAs nanowires can occur only at much higher absolute AsH3 pressure, potentially achievable for certain ex situ MOCVD conditions. As noted earlier, parameters in this study were chosen to give nanowire growth rates comparable to the lower end of typical growth rates observed for ex situ MOCVD, achieved by using similar AsH3 pressures and lower TMGa pressures. Our analysis thus indicates that the parameter space for ex situ MOCVD nanowire growth will also be dominated by the semi-Ga-limited regime, in which incubation and layer growth are controlled by different species, while As-limited and True Ga-limited growth may be achieved at the extremes of the parameter space. For extremely high fluxes of both species, it can also be predicted that growth rates could be determined by diffusion of As or Ga in the liquid, or kinetic factors such as the actual crystallization time, which are not considered in the simulation. This regime is anticipated to fall far beyond the conditions accessible in MOCVD.

Discussion

The observation that incubation/nucleation and layer growth are primarily controlled by different experimental parameters (Ga and As precursor partial pressures, respectively) means that under appropriate conditions it is possible to independently control the two steps. The evolution of the droplet composition during the two steps will control such properties as composition, uniformity, and crystal structure; clearly understanding the time scales for the two events thus provides a greater freedom to design the material properties. The crystal structure, for instance, is generally controlled by nucleation, assuming the absence of structural defects during layer growth or postgrowth structural transformations. On the other hand, compositional uniformity and impurity/dopant incorporation can also strongly depend on the layer growth step. For example, incorporation of impurities by step trapping, which can result in very high concentrations of dopants or other trace elements, is directly proportional to the layer propagation rate rather than overall growth rate.[10,44] Similarly, the performance of heterostructure devices such as tunnel field-effect transistors depends critically on the composition of the heterojunction,[45] which will be sensitive to the relative rates of both the incubation and layer growth processes.

The key insights discussed here can be extended to most compound nanowires and are not restricted to Au-catalyzed GaAs nanowire growth by MOCVD. The prevailing mechanistic understanding of nanowire growth (which assumes instantaneous layer completion) was initially postulated for elemental nanowires, and was later adopted for compound nanowires as well. However, the different miscibility of the constituent growth species in the catalyst lead to more complex growth kinetics for compound nanowires, in turn offering more opportunities to control the process. The fact that Ga readily alloys with the metallic catalyst while As incorporation into the catalyst is minimal is the underlying reason why Ga and As control growth in very distinct ways in most of the accessible experimental conditions. Most other III–V and II–VI binary semiconductors are also composed of one metallic (group II or III, e.g., Ga, In, Zn, Mg) element and one nonmetallic (group V or VI, e.g., As, N, O, P, S) element, which in most cases exhibit similar alloying trends with gold or other typical transition metal catalysts.[40,46] The nonmetallic species generally has low solubility in the metallic catalyst and would thus tend to limit layer completion. It is thus reasonable to hypothesize that it will be possible to independently control layer nucleation (typically with the group II/III precursor) and layer growth (typically with the group V/VI precursor) for many of these material combinations. In addition, since the effects are not related to the types of precursors used, they should also apply equally well to other growth techniques (such as molecular beam epitaxy).

Conclusions

Understanding the mechanistic processes controlling nucleation and crystal growth is key to designing materials with desired properties. In this study we have investigated in situ the nucleation and layer growth processes in VLS growth of GaAs nanowires. We observe that the nucleation (controlled by an incubation process) and layer growth processes can be independent, occur on similar time scales and are controlled by different parameters for most of the growth parameter space. This is a consequence of the confined multiphase nature of the VLS process: Although material is supplied via the gas phase, the kinetics of the process are controlled by the catalyst nanodroplet. This means that the nucleation and initial layer growth can significantly change the composition of the droplet, depleting it of supersaturating atoms. For binary materials composed of two species that interact very differently with the catalyst droplet material, this opens up a broad parameter space in which nucleation and layer growth are independently controlled. Specifically, nucleation can be controlled by the species that easily alloys with the catalyst (such that each atom changes the chemical potential very little) while layer growth is controlled by the other species, for which very few atoms are required to supersaturate the droplet. The similar time scales of these two steps indicate that both must be considered in order to understand the nanowire growth process and to control the material properties. Controlling the parameters dominating the independent steps offers an opportunity to engineer nanoscale devices with controlled properties, such as crystal structure, composition, and impurity incorporation.

Methods

Data Acquisition and Presentation

Videos of atomic resolved images were recorded with an AMT XR401 sCMOS camera with an exposure time of 50 ms per image. Continuously recording results in videos with about 18.3 frames per second (fps) on average. The TEM images shown here were extracted from these videos and processed. The incubation time and layer completion time are measured from the videos. The error bars in the plots of incubation time and layer completion time represent the standard deviation among multiple layer growth events.

In Situ Growth

GaAs nanowires were grown in a Hitachi HF-3300S environmental transmission electron microscope (ETEM) with a cold field emission gun and a CEOS B-COR-aberration-corrector. The wires grew along the ⟨111⟩ direction (cubic notation or equivalent ⟨0001⟩ direction in hexagonal notation). Blaze software supplied by Hitachi was used to control the local sample temperature using Joule heating in a constant resistance mode. The heating chips used had thin SiN windows on which nanowire growth could be followed; some contained holes so that growth could be viewed without SiN in the background. The ETEM was connected to a gas handling system with the MOCVD gases. Further details of the experimental setup can be found in ref (37).

Precursor Supply

AsH3 and TMGa were used as the precursors. The AsH3 flow was controlled exclusively by the mass flow controller. The TMGa was supplied via a bubbler maintained at −10 °C with H2 bubbled through it. A portion of the TMGa/H2 flow passing the mass flow controller was bypassed to the vent line to restrict the TMGa pressure reaching the microscope. The precursor fluxes sent to the ETEM were monitored with a residual gas analyzer (SRS RGA 300) using mass spectrometry. The precursor flows were calibrated to find the partial pressure at the sample. More details on the calibration methodology is given in ref (37). For this calibration we assume that TMGa and H2 are being pumped out from the ETEM at the same efficiency, and thus, the TMGa partial pressure could be underestimated in this report.[37]

AsH3 Series Experiments

The AsH3 flow was varied at constant TMGa pressure (13.1 × 10–5 Pa) and temperature (420 °C). The AsH3 supply partial pressure was initially set to 1 Pa, then increased in steps to 5.6 Pa, and finally decreased back to 1 Pa.

Ga Series Experiments

The TMGa partial pressure was varied at a fixed AsH3 supply (∼1 Pa) and temperature (420 °C). This experiment was started at a TMGa pressure of 11 × 10–5 Pa and successively set to 9, 13, 27, and finally 55 × 10–5 Pa.

Regarding the Model

In this model, we assume that the Ga (or As) flux into the catalyst is directly proportional to the TMGa (or AsH3) partial pressure. Due to the slight difference between the absolute values measured for the two different experiments, the proportionality constant between AsH3 partial pressure and the As flux into the particle was adjusted. (As a consequence, the 1 Pa AsH3 pressure used for the Ga series experiment was set in the simulation to be equivalent to 1.5 Pa AsH3 of the AsH3 series.) The proportionality constants will be affected by, for instance, V/III or pressure -dependent precursor decomposition, TMGa-dependent Ga surface diffusion, effect of local environment near the nanowire, and the spatial variation of the gas flow (in terms of fluid dynamics); therefore, these effects are not explicitly accounted for. Further details about the model are found in Supporting Information section S1.