Direct observation of hydrogen at defects in multicrystalline silicon

Hydrogen passivation is a key industrial technique used to reduce the recombination activity of defects in multicrystalline silicon (mc‐Si). However, not all dislocations and grain boundaries respond well to traditional hydrogen passivation techniques. In order to understand the reasons for these different behaviours, and how superior passivation might be achieved, a method is required for the direct observation of hydrogen at these defects. Here, we present a novel characterisation technique based on a combination of transmission Kikuchi diffraction (TKD), atom probe tomography (APT), and isotopic substitution that enables unambiguous detection and quantification of hydrogen atoms present at crystallographic defects in mc‐Si.


| INTRODUCTION
Currently, over 60% of silicon solar cells are manufactured using multicrystalline silicon (mc-Si). 1 The performance of these cells is limited by carrier recombination at crystallographic defects, such as grain boundaries and dislocations, and the impurities that decorate them. 2 The reduction of these sources of recombination has been identified as one of the key challenges for the future of solar cell development. 3,4 The introduction of atomic hydrogen from dielectric layers during high temperature steps in the processing of silicon has been shown to result in increases in bulk carrier lifetime by almost an order of magnitude, 5 and the complete electrical deactivation of certain grain boundaries. [6][7][8][9] Although hydrogen passivation is the most effective current commercial process in reducing this recombination, there remain a number of significant unanswered questions about the process. 8,10 In particular, it is not understood why hydrogen passivation is effective on some defects but not others. 3 For example, Bertoni et al, 8 recently showed that while a random-angle grain boundary clearly responds to hydrogen passivation, a Σ27 boundary subject to the same process remains electrically active. Also, unresolved are the mechanisms by which charge carrier injection (by light or biased junctions), hydrogen and defect charge states, and/or prior gettering steps act to change the effectiveness of hydrogen passivation. [11][12][13] The primary reason so little is known about the hydrogen passivation process is that the bulk hydrogen concentrations (10 14 -10 16 cm −3 ) involved in these processes are well below the detection limits of most available techniques. [14][15][16] Indeed, the characterisation of hydrogen in material microstructure remains a formidable challenge even for modern high-resolution microscopes. This inability to directly image hydrogen, particularly at the atomic scale, has limited the understanding of its role in affecting the performance of a wide range of materials.
APT is a 3D nanoscale microscopy technique that provides both chemical identity and precise spatial location of individual atoms within a material microstructure. More specifically, recent studies have demonstrated the unique insights provided by APT in the characterisation of grain boundaries within mc-Si, using a multimicroscopical approach of electron backscatter diffraction (EBSD), electron beaminduced current (EBIC), and APT. [17][18][19] Utilising these methods, recombination active random-angled grain boundaries enriched in C, O, and Cu have been observed. 20,21 Since APT provides both spatial and compositional information at the atomic scale, it has obvious potential to bring new insights to the mechanisms behind the hydrogen passivation of defects in silicon. However, a major drawback in the design of current atom probes is the significant amount of ambient hydrogen that is present in the analysis chamber during operation. This hydrogen adsorbs onto the specimen and is then evaporated from its surface just like the constituent atoms. 22  In this study by combining hydrogen passivation, underpinned by isotopic substitution from an atomic source, with the 3D atomic-scale characterisation technique atom probe tomography (APT) and crystallographic information by transmission Kikuchi diffraction (TKD), we present a novel experimental protocol for the direct and unambiguous imaging of individual hydrogen atoms at defects within mc-Si.

| METHODS
The material used in this study is a 2 × 2 cm wafer section taken from a p-type high-performance multicrystalline silicon (HPMC) ingot. 2,25 The surface saw damage was removed using a chemical etch of  29 Si H at a grain boundary, with a small fraction of Si atoms (grey) shown for reference hydrofluoric acid (50 mL), nitric acid (220 mL), and acetic acid (30 mL).
The samples were then etched using a Secco etch, as described in The samples were then exposed to a remote 2 H plasma at 200°C for 60 minutes with a plasma power of 30 W. The use of atomic 2 H is important as it is much more effective at introducing hydrogen into silicon than molecular sources and more closely mimics hydrogen sources used in solar cell production. 3,27,28 Atom probe samples were produced by selecting individual defects using focussed ion beam (FIB) methods via a Zeiss NVision Dual Beam SEM/FIB, after the method described in Lotharukpong et al. 29 This method used FIB techniques to create a lift-out bar containing the grain boundary of interest, which was then attached to the end of a transmission electron microscopy (TEM) half grid. Doughnut milling was then used for the production of needles with the grain boundary running along the length of the needle, enabling a comparatively large section of the boundary to be analysed. TKD [30][31][32]   increase the signal-to-noise ratio for the detection of low levels of impurities. 33 3 | RESULTS Figure 1 shows the analysis applied to a random-angle grain boundary isolated within the hydrogenated silicon. This type of grain boundary is known to be electrically active but also susceptible to hydrogen passivation. 7 Figure 1A presents an SEM image of the silicon wafer surface, with the etch pits formed during Secco etching marking the grain boundaries clearly visible. The grain boundary was then lifted out using FIB methods described in Lotharukpong et al, 29 to produce the sharp needle-shaped geometry required for APT ( Figure 1B). Figure 1C shows a TKD image of the needle, used to ensure the alignment of the grain boundary along the axis of the needle and to characterise the misorientation and rotation axis of the boundary. In Figure 1D, Within the same 2 H-charged silicon sample, hydrogen was also found at three dislocations close to a twin boundary, as presented in  of dislocations observed after hydrogen passivation. 12,28 In contrast to dislocations, perfectly coherent Σ3 twin boundaries are known to be electrically inactive in mc-Si. 34 However, it is not known whether hydrogen segregates to these boundaries during passivation. Figure 3 30 Si + at 30 Da; therefore, a deconvolution analysis must be applied to estimate the contribution of the 2 H containing ions to each of these peaks.
By assuming that the isotopes of silicon are observed in their natural relative abundances, 36 and the fact that the 28 Si + peak has no significant overlaps in this system, it is possible to determine the relative contributions to these peaks. This deconvolution was performed by extracting a region of interest from the dataset containing an area of the grain boundary or along the length of a dislocation.
Then using the programme AtomProbeLab (https://sourceforge.net/ projects/atomprobelab/), which uses the maximum likelihood method, the quantity of 2 H present was determined within confidence intervals (95% in this study), as described in London et al. 37 Since the detector has an efficiency of only 52%, a calculation of dividing the number of counts by 0.52 was required to determine the total number of 2 H + ions present. This deconvolution of the peaks maximised the likelihood of composition, based on the atomic abundance of silicon and the counts in the ranges (28)(29)(30)(31)(32) Figure 2.
Although this study does not allow for the determination of the cause of the variation between dislocations, it clearly shows the ability of the deconvolution method to allow for quantitative comparisons regarding the magnitude of the hydrogen present to be made. To give a rough estimation of number of 2 H per atomic site along the dislocations,

FIGURE 4
A, Mass spectra of a random-angle grain boundary from a sample, which has not received any hydrogen treatment. B, Mass spectra of the random-angle grain boundary shown in Figure 1, from a sample that has been passivated by H

| CONCLUSION
The ability to quantify the amount of hydrogen present at a grain boundary or dislocation is an important breakthrough for the characterisation of hydrogenation in mc-Si for solar cell applications. With regard to electrically active grain boundaries that are less susceptible to hydrogen passivation, it will allow determination of whether this is due to a lesser affinity of hydrogen to segregate at these features or if there are other causes such as the presence of specific impurities that are not passivated. The technique is also applicable to the crystallographic defects that limit the performance of cast monocrystalline solar cells.
This information is invaluable for informing the further development of casting methods for crystalline silicon growth in order to avoid the formation of defects than cannot be passivated. 2,25 It can also be observed how specific processing conditions, such as temperature and carrier injection, 13,40 can either increase or decrease the amount of hydrogen present at crystallographic defects. A better understanding of these effects is critical for optimising such processes and continuing to improve the performance of solar cells based on these materials.

ACKNOWLEDGEMENT
This work has been supported by the UK Government through the EPSRC (Supersilicon grant, EP/M024911/1). The research materials supporting this paper can be accessed at ora.ox.ac.uk under the paper title.  Quantity of 2 H calculated as a Gibbsian interfacial excess for the random angle-grain boundary shown in Figure 1, for the three dislocations shown in Figure 2 and for the Σ3 grain boundary shown in Figure 3 RA Note. Upper and lower bounds are determined, within 95% confidence intervals. RA, random angle.
[Correction added on 19 August 2019, after first online publication: the columns in Table 1 were incorrectly aligned and this has now been corrected.]