Kinking of GaP Nanowires Grown in an In Situ (S)TEM Gas Cell Holder

Nanowires are a promising structure to create new defect‐free heterostructures and optoelectronic devices. GaP nanowires grown via the VLS mechanism using tertiary‐butyl phosphine (TBP) and trimethylgallium (TMGa) as precursors in an in situ closed gas cell heating holder are shown. This holder is a model system to investigate the processes in metal‐organic vapour phase epitaxy (MOVPE). GaP nanowires change their growth direction after random distances by producing kinks. Statistics of these kink angles show dominant values of around 70.5°, 109.5°, and 123.7°. A custom holder tip capable of holding a single heating chip is used to perform scanning precession electron diffraction (SPED) measurements on the nanowire kinks. The results show that the predominant kink angles result from micro twins of first and second order. Understanding the defect formation and resulting geometry changes in GaP nanowires can lead to increased control over their shape during growth and mark a huge step toward applicable nanowire devices.


Introduction
Nanowires, especially those consisting of III/V semiconductor materials, are used for many technical applications, such as LEDs, lasers, photodetectors, and solar cells. [1][2][3][4][5] A widely used fabrication process for these materials is metal-organic vapor phase epitaxy (MOVPE). [6][7][8] A detailed understanding of their growth behavior, such as defect structure, growth rate, growth direction, or growth geometry, is required to develop such devices. However, investigation of the growth processes on the nanometer scale in a conventional MOVPE reactor is functionally impossible. To this end, a commercially available Protochips Inc. in situ system has been modified. [9][10][11] With this The kink distribution in GaP nanowires differs from that of Si and InP nanowires, as the kinks occurring most frequently are those of 70°. Earlier investigations on the transport properties of III/V semiconducting nanowires suggest that kinks can negatively influence electron mobility which is of high importance for technological application. [21,22] For this reason, a fundamental understanding of the origin of these kinks is crucial in order to change growth conditions in a way to control or completely avoid kinking.
In this work, scanning precession electron diffraction (SPED) is used to determine the origin of kinks in GaP nanowires. SPED has been previously applied to several nanowire systems such as to measure strain in GaP and GaAs nanowires, [23] to obtain orientation maps of Ge and P:Co (99% Co, 1% P), [24] and silver nanowires, [25] and to perform phase mapping of GaAs-GaAsSb nanowires. [26] This highlights the applicability of this technique to study the orientation relationships at the kinks.

Experimental Section
GaP nanowires were grown via the VLS Mechanism, [27] where heated metallic nanoparticles were used to catalyze the decomposition of precursor molecules. The growth atom species delivered by the precursors diffuse into the particles forming a liquid alloy, and, after supersaturation, crystal nucleation takes place at the liquid-solid interface. The growing crystal pushed the catalyst droplet forward and thereby created a thin wire. The GaP nanowires studied in this work were grown in a Protochips Inc. Atmosphere gas cell holder. The system has been modified to allow the usage of toxic and pyrophoric precursor gases. To this end, gas mixing, appropriate gas monitoring, and gas scrubbing systems have been added. [10] The precursor gases were stored in a DIN 12925 norm gas storage locker and their amount was so low that the TLV (threshold limit value) cannot be exceeded. Growth took place on a Protochips Inc. MEMS chip with electron transparent silicon nitride (SiN) windows. The SiN windows of the heating chip have a thickness of around 30 nm and were surrounded by a 120 nm thick SiC heating membrane. [28] A suspension of colloidal gold nanoparticles in isopropanol was drop-cast onto the heating chip. After evaporation, only the nanoparticles remained on the chip's SiN windows to act as catalysts. The size of the nanoparticles used in this study, obtained via JEOL JIB-4601F SEM, was around 20 nm ( Figure SA, Supporting Information). The size of nanoparticles determined nanowires' diameter. [29] With the given partial pressures GaP nanowires were expected to grow in zincblende structure. [30] Growth took place at 450 °C and partial pressures of tertiary-butyl phosphine (TBP) and trimethylgallium (TMGa) of 1 and 0.2 Pa, respectively. Additionally, 200 hPa of N 2 was used as a carrier gas. Images during growth were acquired in a double C s -corrected JEOL JEM 2200FS (S)TEM operating at 200 kV. Images were recorded under high-angle annular darkfield (HAADF) conditions which leads to Z-contrast in the images due to Rutherford-like scattering. The frame rate was 0.51 s −1 .
For post-growth investigations higher precursor pressures were used to grow a larger amount of material. This growth took place at 450 °C and partial pressures of tertiary-butyl phosphine (TBP) and trimethylgallium (TMGa) of 1 and 0.1 hPa, respectively. Additionally, 400 hPa of N 2 was used as a carrier gas. After 1 h, the growth was stopped by reducing the temperature to ambient.
To investigate such grown nanowires post-growth, a custom TEM holder tip was devised capable of holding a MEMS chip and enabling sample observation in vacuum. This greatly improved image quality, since the enclosed gas volume in the heating cell would otherwise lead to beam-induced contamination. Additionally, by removing the necessity for a second window chip, this solution reduced the total amount of amorphous SiN in the image background. This holder tip with a mounted MEMS chip is shown in Figure 1a. Figure 1b shows this holder tip mounted onto a single-tilt TEM holder from JEOL Ltd. The MEMS chip was flipped to place the sample within the eucentric height of the goniometer and the focal plane of the lens.
This setup was used in a JEOL JEM 3010 operating at 300 kV in combination with a NanoMegas's ASTAR system [31] to acquire 4D SPED data sets (x, y, k x , k y ) and thereby investigate the crystallographic origin of the dominant kink angles.
A diffraction pattern of the sample was obtained by precessing a tilted electron beam around the optical axis with the help of two-stage deflection coils. Another set of two-stage deflection coils offsets the incident beam tilt and steadied the diffraction pattern. The resulting diffraction spots were the sum of a series of Laue circles, suppressing dynamical effects and giving quasi-kinematical conditions. By scanning the precessing nanobeam over the sample, diffraction patterns of every www.advmatinterfaces.de scan point can be obtained, resulting in 4D data sets. These data sets were used to generate spatially resolved crystal orientation maps using an ultrafast pattern-matching algorithm. The ASTAR software package finds the best matching crystal orientation at each scan point by comparing with a library of simulated diffraction patterns [31,32] (Figure SB, Supporting Information). In comparison to other scanning TEM techniques, the spatial resolution of SPED was reduced, since precession exacerbates the effects of lens aberrations, thereby increasing the probe size. [33] The probe sizes for precession angles between 0° and 0.6° were approximately 6 to 12 nm. The SPED 4D data sets were typically acquired using a precession angle of 0.249° to 0.502° with a step size of about 2 nm. With the current growth setup and procedure, the size of micro twin domains of the GaP nanowires was often at the resolution limit (≈7 nm) of SPED.
Therefore, the nanowires were additionally investigated using high-resolution TEM (HRTEM) imaging. This method was able to resolve the defects which lead to kinks at the atomic scale. However, to acquire HRTEM images a sample orientation in zone axis is mandatory. For this reason, HRTEM imaging of nanowires grown on a MEMS chip is often challenging due to MEMS chip holders being limited to a single tilt axis. The probability of finding nanowires that can be tilted in the required zone axis conditions is relatively low. In contrast, SPED orientation mapping is able to determine crystal orientations independently from the sample orientation. Furthermore, the presence of the amorphous SiN windows deteriorated the contrast in HRTEM imaging, especially in an uncorrected microscope. These factors make the approach of HRTEM imaging to characterize micro twins laborious. However, if a nanowire in zone axis was found, HRTEM images can serve as a useful addition, to aid in the interpretation of orientation maps.
The kink angles of the nanowires were obtained by performing tilt series on a JEOL JIB-4601F SEM. In a tilt series, several images were recorded at different stage tilt angles in the SEM ranging from −8° to 55°. Out of these image series, the kink angles were calculated via 3D reconstruction.

Results and Discussion
In the results depicted in the following, first an in situ observation of a kinking nanowire will be shown. To investigate the origin of such nanowire kinks, post-growth studies will be depicted and discussed afterward. Figure 2 shows the evolution of a nanowire kink. A complete video of these kink can be seen in Movie SC, Supporting Information. The bright gold particle can be seen on top of a nanowire of 25 nm radius growing from bottom to top. The dark background consists of the SiN window. The growth interfaces are highlighted in yellow for each frame and maintained to the following. The coordinate system is chosen according to an initial (111) growth front viewed along [110]. Initially, the catalyst droplet is symmetrically shaped on the tip of the straightgrowing nanowire. From frame (b) -(d), truncating tilted growth planes occur, resulting in a deformation of the catalyst particle. The tilt angle increases up to an angle of 70° forming the new growth front. In the successive frames (e) -(g), the former growth plane transitions to the 70° tilted one, completed in frame (h). Frame (i) shows the resulting 109° kink with former (111) growth planes in blue, new (111) growth planes in red and intermediate planes in yellow.
Since these in situ observation of these growing nanowire kinks is a singular event, no statistical statements of the occurrence of such kinks can be made. Therefore, in the upcoming results, ex situ investigations will be shown in order to study the origin of such kinks.
In the following the angle distribution of GaP nanowire kinks is discussed and dominant angles are further investigated via SPED to reveal their crystallographic origin. Figure 3 shows post-growth images of GaP nanowires grown on an in situ MEMS chip at a growth temperature of 450 °C, precursor partial pressures of 1 hPa TBP and 0.1 hPa TMGa, and 400 hPa N2 carrier gas pressure. Figure 3a is an overview image taken in an SEM. The bright parts are the grown GaP nanowires. The six disks with dark contrast in the background are the SiN windows of the MEMS chip. The grey area around the SiN windows is the SiC heating membrane. Due to the amorphous nature of the SiN windows the nanowires grow in random directions. This may be different from the case where a defined substrate is used to grow nanowires epitaxial. However, the influence of the substrate on the kinking mechanism will diminish for longer nanowires and the kinking mechanism is only affected by the growth conditions and thermodynamic stability of the nanowire system. Accordingly, our results can be transferred to the epitaxial situation as well. Nanowires in the vicinity of the windows often grow comparatively straight and long. In the outer areas, the nanowires are smaller with a much higher number of kinks. This is due to a temperature gradient from the inner to the outer regions of the MEMS chip. The observation that lower temperatures lead to more kinks already hints at crystallographic defects being the origin of these kinks. [34] Figure 3b shows a conventional TEM image of nanowires grown on a SiN window. Due to the imaging conditions, the SiN window appears bright (because of the weak interaction of the electrons with the SiN), and the nanowires and the SiC heating membrane appear dark because of the increased thickness and scattering cross-section. Although the used gold nanoparticles have diameters of around 20 ± 9 nm, the observed nanowire diameters vary more. This is due to the formation of gold nanoparticle clusters, which coalesce into larger gold droplets during heating. Additionally, the heated droplets are subject to Oswald-ripening. [35] Since the thickness of nanowires grown via the VLS mechanism correlates to the size of the catalyst droplet. [29] While some of the observed nanowires grow in random directions out of the substrate plane, others grow directly on the surface of the window. While the latter largely follow the rough morphology of the SiN window, both growth modes show straight nanowires, which change their direction abruptly by kinking at random distances. Some examples of such kinks are marked with red arrows.
The angles of these kinks were determined by performing tilt series. An example of such a tilt series is shown in Figure 4a-c. The corresponding stage tilting angles are shown in white. Some exemplary kinks measured this way are indicated with red arrows and their estimated angles are 1) 123.2° ± 2.3°, 2) 111.4° ± 1.9°, 3) 124.0° ± 3.9, 4) 124.1° ± 2.4°, and 5) 70.5° ± 3.7°. The actual magnitude of the errors varies considerably, because it depends on the degree of 3D projection of the kink during the tilt series. Figure 4d shows a histogram of all measured kink angles in orange. The blue line is a representation of the estimated angular distribution, obtained by adaptive kernel  www.advmatinterfaces.de density estimation [36] in which also the errors are considered. The kink angles are not randomly distributed. There are three dominant peaks in the distribution, positioned at around 72°, 111°, and 124°. Both, the 72° and the 111° angles match with the characteristic crystallographic angles of 70.5° and 109.5° which are the angles between <111> directions in the zincblende crystal structure. The third angle of 123.7° can be a result of second-order twinning which will be discussed later. For this reason, the dominant angles will be referred to by their crystallographic values in the following discussion. The distribution shows that these dominant angles don't appear equally often. 70.5° and 123.7° kinks occur most frequently, while the 109.5° kinks are rare in comparison.
In the following results, we will first describe twins in straight nanowire segments. Following this, the origin of the 109.5° and 70.5° kinks will be shown. The third part deals with the origin of the 123.7° angles. The relative occurrence of all these angles will be discussed at the end. Figure 5a shows a high-resolution TEM image of a straight nanowire segment. The alternating regions of darker and brighter contrast clearly reveal that the wire consists of segments of differing crystal orientations. From one segment to another, the stacking of {111} planes shifts slightly to the right (dark segments) or to the left (bright segments), resulting in {111}-facets at the edges of the nanowire, indicated by the dashed white lines in Figure 5a. These facets have an angle of 141° with respect to each other, which can be described as the angle between stacked octahedrons. [17,18] This angle is a result of a crystal rotation of 60° around the [111] growth direction.  The angle between these planes is given. b) Simulated crystal model with a twin of first order in the same orientation as the nanowire in (a). The twin boundary as well as the 141° facet is shown by the black dashed line. c) SPED data combined with the index and reliability map. The color code for (c) is shown in the inset. d) The misorientation from the origin is plotted against the distance. The arrow in (c) shows the trace of these line scan.

www.advmatinterfaces.de
The two stacked octahedral crystal segments are twins with a common (111) plane, referred to as a twin boundary. [37] To illustrate this, a crystal model of such a twin boundary is shown in Figure 5b. If we define the growth direction from bottom to top, the growth planes are P-terminated which is in other words the B-polar growth which is in agreement with the literature. [18] The model also shows that the termination of the growth planes stays the same when the growth direction does not change, since the surface of the (111) growth plane is occupied exclusively by gallium or phosphorus before as well as after the twin boundary. Figure 5c shows a combination of the orientation, index, and reliability map of a SPED measurement. The orientation map indicates different crystals' orientation with different colors. The index map displays the correlation index value in greyscale at every pixel. The higher the index value, the better the match between the recorded diffraction pattern and the selected template. The inclusion of the index map de-emphasizes amorphous regions. Orientation reliability is a measure of the difference between the index values for the two best-matching templates for the selected point displayed by the reliability map. This de-emphasizes regions with overlapping crystals, where the orientation is ambiguous. The index and reliability maps, together with the raw orientation map, can be found in Figure SD, Supporting Information. The area of Figure 5a is marked with the dashed white line in Figure 5c. Since the segments with the same crystal orientation have the same color, it can be seen that the orientation changes back and forth over the whole measurement region. The small bright segment from Figure 5a cannot be seen in Figure 5c because of the limited spatial resolution of SPED. Figure 5d shows the misorientation along the overlaid arrow with respect to the first pixel of the arrow (referred to as misorientation line scan hereafter), shown in Figure 5c. Each point in the plot represents one scanning pixel of the SPED. Between every neighboring segment, the orientation is rotated by 60° around the [111] axis. The rotation axis can be read out via the inverse pole Figure in the ASTAR software. Together, this is a proof of twin boundaries between the segments. All-in-all, Figure 5 shows a straight-grown nanowire that has several twin boundaries along the growth direction. After some distance, the crystal rotates back and forth, which results in faceting at the edges while keeping the growth direction straight. An overview of the measured nanowire as stitched TEM images and an overall SPED measurement can be found in Figure SE, Supporting Information.
While in most cases the formation of a twin does not change the growth direction, in some cases a twin initiates a change from one [111] growth direction to another. This can result in kink angles of either 109.5° or 70.5°. One example of a 109.5° kink is shown in Figure 6, with a growth direction from right to left. The 3D reconstruction by tilt series gives an angle of 110.2° ± 6.1° for this kink and the high-resolution image in Figure 6a clearly confirms an angle of 109.5°. The kink is divided into two parts, A and B, separated by a staggered twin boundary, indicated by dashed white lines. An additional twin boundary that doesn't result in a kink can be seen on the right of the image. The white arrows at the bottom define the crystal orientation of the two parts. In Figure 6b fast Fourier transforms (FFTs) of segments A and B are shown. These also show the different orientations of the segments as indicated by the labeled spots and colors. Each zone axis is written in the upper right and is [011] for part A and [110] for part B. Figure 6c shows a SPED measurement of the same kink. The individual index and reliability maps can be found in Figure SF, Supporting Information. The dashed white box shows the area of the high resolution from Figure 6a. Area B corresponds to the red orientation and area A to the cyan one. Figure 6d shows the misorientation line scan along the arrow in Figure 6c. A misorientation of 60° with rotation around the [111] axes between the cyan and the red areas is observed, which confirms that the two areas are twins with a (111) twin boundary. It is unclear whether the fact that this twin boundary is staggered plays a role in the kink formation. Figure 7a shows a TEM image of a 70.5° nanowire kink. The kink angle was determined via tilt series which resulted in an angle of 70.3° ± 2°. The growth direction of the nanowire is from right to left. The black box shows the area where a SPED measurement was performed. The combined orientation, index, and reliability maps, are shown in Figure 7b. The individual maps can be found in Figure SG, Supporting Information. The diameter of the investigated nanowire in a) is around 50 nm. However, the apparent diameter of the nanowire in b) is reduced to about 38 nm. This can be explained by an amorphous shell, consisting most probably of an oxide, surrounding the nanowire, which gives no diffraction spots and therefore is not matched in SPED. The region marked with the blue arrow in Figure 7a is contamination deposited on the SiN window during the alignment of the PED system. This is also amorphous, and thus it is also not visible in Figure 7b. Figure 7c shows a misorientation line scan along the line indicated by the white arrow in Figure 7b. It can be seen that there is a misorientation of 60° between the purple and the orange areas that correspond to the expected misorientation between twins. After approximately 15 nm, the orientation turns back to purple and then again to orange. Each change in orientation results in a relative misorientation of 60° with rotation around the [111] axis, which is again a proof for (111) twin boundaries between neighboring segments.
From the results presented in Figures 5-7, we can conclude the following: First, that GaP nanowires regularly form twins with (111) twin boundaries, that most often leave the growth direction unchanged. Second, when these twin boundaries cause growth direction changes, kinks of 109.5° and 70.5° are produced. Whether or not a twin produces a kink is most probably a process based on the thermodynamic stability of the catalyst droplet.
In the third part of the results, we will focus on examining the origin of the 123.7° kinks. Such a kink is shown in Figure 8. Its kink angle is determined via tilt series to be 122.1°± 1°. Figure 8a shows the combined orientation, index, and reliability map of the nanowire with growth direction from the left to the right. The individual maps can be found in Figure SH, Supporting Information. The white arrow shows the position of the misorientation line scan, which is shown in Figure 8b. We find a misorientation between the green and the orange areas of 60° with a rotation around the [111] axis, which again indicates a twin boundary. Between the orange and the purple area, there is also a misorientation of 60° with www.advmatinterfaces.de a rotation around the [111] axis. Additionally, there is a small region of low reliability in between (bright green pixels). This region is a result of the overlap of the crystal domains of the orange and the purple area, whose orientation therefore can not be definitively determined. Furthermore, a misorientation of around 39° is observed between the green and the purple  area. This misorientation is expected for a twin of second order, where two crystal segments share a common twin (in this case the orange segment), but no common twin boundary. A model of a second-order twin is shown in Figure SI, Supporting Information. Additionally, if we progress further into the cyan region we find an overall misorientation of around 36° which is expected for a twin of third order. An overview of the expected misorientations for higher-order twins is shown in Figure 8c. Overall, the 123.7° kink consists of two consecutive changes in growth direction, adding up to a second-order twin.
Since the origin of the dominant GaP nanowire kinks has been found, their relative occurrence can be discussed. From a geometrical point of view, the 109.5° kinks should be expected to occur more often, since the required direction change of the gold droplet is not as large as compared to 70.5° kinks. In practice, however, 70.5° kinks are observed more frequently. This apparent discrepancy can be related to changes in the termination of the growth plane. As seen in Figure 5b, a twin boundary without a change in growth direction leaves the termination of the growth plane unchanged. Similarly, a change in growth direction resulting in a kink of 70.5°, for example, from [111] to [111], also does not invert the termination. However, when the growth direction changes to form a 109.5° kink, for example, to [111], the termination switches. Due to their different surface energies, it is expected that during growth one termination is favored over the other, making a singular 109.5° kink more unlikely. This results in two 109.5° kinks often occurring in quick succession forming second-order twins, resulting in an overall angle of 123.7°. We speculate that this mechanism reduces the occurrence of single 109.5° kinks and explains the relatively high amount of 123.7° kinks. These findings, for the material system GaP, are different from the relative occurrence in other material systems reported in the literature, where in InP nanowires the most frequent occurrence is at 109.5°. [20] As stated above, the formation of crystal defects like micro twins is most probably a thermodynamic effect that occurs spontaneously. This leads to the suggestion that different growth conditions like a different growth temperature could suppress the formation of kinks. To confirm this assumption, further growth experiments with different growth parameters need to be done. Another way to suppress the formation of kinks could be the growth in wurzite instead of zincblende structure. Preliminary modeling suggests that this change in crystal could entirely avoid kink formation by twinning. Growth in wurzite structure can be achieved by using much larger V/III ratios, which leads to smaller catalyst droplets, thereby changing the nucleation site. [38]

Conclusion
In this study, we demonstrated the possibility of growing GaP nanowires in a gas cell (S)TEM in situ setup and their subsequent characterization with SPED, facilitated by a custom TEM holder tip. With this, the origin of the dominant kink angles, namely 70.5°, 109.5°, and 123.7°, could be identified. Orientation mapping of the zincblende crystal structure of the GaP nanowires showed that single twin boundaries can initiate 70.5° and 109.5° kinks. This is in agreement with literature, where these angles are reported for similar nanowire material systems. The frequent 123.7° kinks can be explained by second-order twins. The relative occurrence of the dominant kink angles is not yet finally clarified, but can presumably be explained by the change in elemental termination of the growth surface in 109.5° kinks. The twinning of the second-order by two subsequent 109.5° kinks could prevent this change in termination and thus explain the suppression of the 109.5° kink angles as well as the high occurrence of 123.7° kinks.
It is expected that the creation of such microtwins is a thermodynamic effect and could be suppressed by other growth parameters like another growth temperature. Additionally, a growth in wurzite crystal structure could avoid kink formation by twinning entirely. Besides this, GaP in wurzite structure would be a direct semiconductor which is of great interest in optoelectronic device applications.
These findings further deepen the understanding and control of GaP nanowire growth and geometry, which are necessary for moving toward applicable nanowire devices.

Supporting Information
Supporting Information is available from the Wiley Online Library or from the author.