1. Introduction
Recently, the chemical vapor deposition (CVD) method has been widely used for the manufacture of epitaxial silicon layers and epitaxial wafers. In the frame of the CVD approach, epitaxial layers can be deposited on relevant supporting substrates from a trichlorosilane source gas or a tetrachloride source gas at 1000 °C and above. The dopant gases are diboran or phosphin. The gas flows and process times must be carefully chosen to achieve the specified layer thickness, layer resistivity, and relevant requirements for defect density.
Deposition of epi-Si layers at high deposition rates followed by the lift-off step is extremely important for photovoltaics, which looks for kerfless/kerffree approaches to fabricate Si wafers in a cost-effective way to avoid losses of silicon during the wire sawing wafering step.
Today, high-quality monocrystalline silicon (Si) ingots are typically grown using the Czochralski (CZ) method and cut using wire sawing to thin wafers, creating the basis for different devices. The fabrication of thin wafers (around ~100 µm) is of increasing importance to reduce material consumption and to enable the processing of novel devices with increased functionality, such as flexible solar cells. However, due to the physical limitations of the sawing technology used today, especially the ratio between the saw thickness and the wafer thickness, an increasing amount of Si-material is lost in the cutting process as the wafer thickness continues to decrease. In fact, the thickness of wafers produced today is comparable to the thickness of the saw, giving about 50% Si material loss (kerf) per wafer produced. Although some material can be regained through kerf remelting [
1], the achieved yield is strongly limited by several issues, such as material contamination, increased energy usage, handling and breakage of the cut wafers, and an increased number of processing steps. For these reasons, new approaches and methods must be developed to reduce the associated energy and material costs (and the environmental footprint) and enable continued technological development.
There are several different methods for producing “kerfless/kerffree” Si wafers without casting and sawing: the Direct Wafer
® production process to fabricate high-performance “kerfless” silicon wafers directly from molten silicon [
2], the “Smart-cut” process [
3], epitaxial growth using CVD on a porous Si substrate followed by exfoliation of the epi-Si wafer [
4,
5], and some others, like stress-induced lift-off processes, which for several reasons have not achieved practical application in PV so far. Moreover, the edge-defined, film-fed growth (EFG) technique [
6] has been commercialised but has not been able to stay competitive with the Cz-Si approach due to low throughput, non-standard geometry, and high dislocation density resulting in lower cell efficiencies and increased brittleness.
The most technologically advanced ion cut process is the Smart-Cut process [
3], which involves hydrogen implantation followed by wafer bonding and annealing. This method allows the transfer of Si layers with required thickness. To obtain thicker layers more adapted to photovoltaic applications, hydrogen implantation at high energies (with few c) has been successfully implemented, providing freestanding Si layers with thicknesses in the range of 9 µm to 100 µm [
7,
8]. However, it has turned out that Smart-Cut technology is too expensive to be implemented in the mass production of Si wafers.
The alternative low-cost stress-induced spalling approaches that have been developed are: (i) glue-cleave [
9], (ii) SLIM-cut (stress-induced lift-off method) [
10], (iii) “cold split” [
11], and (iv) controlled spalling [
12]. These approaches do not require the costly formation of a weak layer (except “cold split”), as they are based on the formation of thermal stress inside the silicon bulk. The main drawback linked to stress-induced spalling methods is the lack of control over the thickness of the exfoliated Si layers.
Nevertheless, NexWafe, a spin-off of ISE Freiburg, has implemented a CVD deposition method to process a thick crystalline epi-Si layer (n-type, 50–180 µm thick) on a reusable template using trichlorosilane as a gaseous precursor at a deposition rate of about 2 µm/min. The fully grown wafer is then detached from the seed wafer, aiming at replacement of traditional CZ wafers, and ~20% efficiency for solar cells processed from such wafers is reached [
13]. According to NexWafe, their technology could save up to 60% of Si-loss during sawing, reduce energy consumption during manufacturing by up to 80%, and require 70% less investment cost for its scrap-free wafer production. IMEC has demonstrated conversion efficiency of about 22.5% [
14] for n-PERT solar cells based on CVD wafers grown by Crystal Solar.
Although CVD-based epitaxial wafers have excellent material quality, as evidenced by the minority-carrier lifetime measurements and cell efficiencies, usage of poisonous gases and the need for complex and expensive infrastructure and safety-related considerations create some barriers for penetration of this technology at the industrial level.
Therefore, the search for cost-effective alternatives to CVD approaches is an important task for Si-based PV and potentially for Si-based microelectronics technologies.
In particular, recent technological advances based on developments of physical vapor deposition (PVD) processes, such as electron beam (e-beam) deposition, have enabled the possibility to process reasonable quality Si films using laser or solid-phase crystallization processes applied to e-beam deposited Si films [
15,
16] or by the direct growth of Si layers using e-beam deposition at elevated temperatures (below 650 °C [
16]). It should be noted that the processing of Si-based layers through the e-beam evaporation of Si offers several advantages over CVD-based methods due to the elimination of toxic gases, feedstock flexibility, and the possibility for direct growth of p/n junctions using p/n Si feedstock for e-beam evaporation. Moreover, a high deposition rate (up to 30 µm/minute) at elevated temperatures (~800 °C) has been demonstrated [
17,
18]. Since the deposition temperature was not high enough, high-quality epi-Si growth was not achieved. Epi-Si layers with high quality have been grown using CVD at high temperatures (1000 °C and above). However, the growth of epi-Si layers at such high temperatures and their analysis have not, to the best of our knowledge, been reported for the e-beam deposition-based process so far.
The main goal of this article is to present and compare the initial results of e-beam Si layers deposited at different elevated temperatures, 600 °C, 800 °C, and 1000 °C, all with a high deposition rate (about 1.5 µm/min), realised by the implementation of an advanced dedicated e-beam system.
3. Results and Discussion
The silicon layers deposited on polished six-inch Si substrates with a high deposition rate of ~1.5 µm/minute at 600 and 800 °C visually demonstrated areas of different appearance; some areas appeared shiny, whereas others appeared grey and matte. The sample deposited at 800 °C had a higher ratio of shiny areas than the one deposited at 600 °C. On the other hand, the layer deposited at 1000 °C with the same deposition process conditions did not have any grey areas and was fully shiny. As shown in the SEM images in
Figure 1, the grey parts had a rough microstructure, while the shiny parts were flat. The SEM images confirmed an overall flat microstructure for the layer deposited at 1000 °C, and no areas with a rough microstructure could be observed for this sample. The areas with the rough microstructure in the layers deposited at 600 and 800 °C probably appeared because of lower surface diffusion energy for Si atoms due to the lower temperature of the substrates during deposition.
A cross section EBSD image of a film deposited at 1000 °C is shown in
Figure 2. The EBSD orientation map shows that the deposited layer had the same orientation as the substrate and that the orientation perpendicular to the sample surface was (100), clearly demonstrating the epitaxial nature of the film.
Figure 3 shows the Raman spectra for the shiny areas of the silicon layers deposited at 600, 800, and 1000 °C.
Figure 3 shows that at all three selected deposition temperatures, the Raman peak was located at 520 cm
−1, which is the representative wavenumber for crystalline and stress-free silicon material. The same peak was present in the reference Si (100) wafer.
Figure 4 depicts the Raman spectra from the grey parts of the 600 and 800 °C films, showing the crystalline phase (c-Si) around 520 cm
−1 and the so-called defective/mixed crystalline phase around 510 cm
−1 obtained through deconvolution of the Raman spectra. The Raman peak around 510 cm
−1 is also sometimes referred to as a mixed crystallinity peak (mc-Si) [
19].
The defective/mixed crystalline phase is usually attributed to microcrystalline Silicon [
19,
20], which was formed in the grey parts of the 600 °C and 800 °C films as a transition phase between amorphous and crystalline silicon during the crystallization process. More investigations are required to clarify the nature and reason for the formation of such areas with a defective/mixed crystallinity phase. Such studies are outside the scope of this article. However, it can be concluded that a higher deposition temperature fully eliminates the formation of such a defective/mixed crystallinity phase.
The fitting of the Raman spectra for the grey areas was conducted using Lorentzian shapes for the deconvoluted lines. Three peaks around 520, 510, and 480 cm
−1, representing the crystalline, defective, or mixed crystalline peak and the amorphous peak, respectively, were chosen for the fittings. It is important to note that the positions of these peaks were not fixed during the fitting procedure, and, as a result of fitting, they could therefore deviate from the initial values. Prior to the fitting process, the raw data underwent normalization and smoothening. Firstly, the data were normalised such that the maximum peak was assigned the value 1. A moving average smoothing approach was then applied to data with a pixel size of 10. The resulting refined positions for the different constituents are given in
Table 1. From
Table 1, the defective peak appeared at a lower wavenumber for the Si film deposited at 600 °C compared to the one deposited at 800 °C. This could indicate the presence of phases with less crystallinity and therefore the lower quality of the film deposited at 600 °C compared to the film deposited at 800 °C. Defective peaks did not appear for Si layers deposited at 1000 °C, showing that such layers obtained higher quality than those deposited at lower temperatures. Therefore, Si films deposited at 600 °C were not analysed in further detail.
It should be noted that since the surface morphology of the Si films deposited at 600 °C and 800 °C was different (
Figure 1), direct comparison of the crystallinity levels for these films could not be performed correctly.
Hall effect measurements were performed on the deposited films in the van der Pauw configuration, and the results are given in
Table 2.
The Si wafers, which were used as a feedstock to fill the crucible for the e-beam process, obtained a resistance of 5.5–5.7 Ω·cm, and a boron concentration of ~2.7 × 10
15 B-atoms/cm
3. The specific resistance and carrier concentration of the film deposited at 1000 °C were about 5.01 Ω·cm and 4.93 × 10
15 1/cm
3, respectively. Hence, the resistivity of the deposited Si film was a bit lower and the carrier concentration (therefore the concentration of boron) was a bit higher than those of the Si feedstock. This indicates that some additional doping of boron from the Si feedstock occurred, which means a faster transfer of B than Si upon the introduction of the e-beam evaporation process. It should be noted that the co-evaporation of silicon and boron in a vacuum chamber using e-beam showed some peculiarities. First of all, the temperature during the e-beam process could be as high as 3000 °C under the electron beam spot and, secondly, the vacuum pressures of silicon and boron were different, which could lead to different evaporation rates for these elements. Moreover, some segregation processes of boron initiated at the solid/liquid Si interface could occur in the crucible with the boron-doped Si feedstock during melting using e-beam deposition. At this stage of development, it can be only stated that more boron atoms compared to that of silicon were evaporated and transferred to Si films during the e-beam deposition process. This means that the remaining silicon material in the crucible contained less boron after e-beam treatment compared to initial Si feedstock. Similar results were reported in [
21], where the purification of upgraded metallurgical silicon by means of the extraction of boron and phosphorus was experimentally demonstrated using concentrated solar radiation in the temperature range of 1550–1700 °C. Further detailed investigations concerning the transferability of dopants (boron in our case) should be performed. The p-type hole mobility was about 253 cm
2/(V·s), which indicated that the epi-Si layer, deposited at 1000 °C, was still not as perfect as the Cz-Si wafer, which has been used as a feedstock for e-beam evaporation (the hole mobility is about 408.5 cm
2/(V·s). However, the measured numbers were quite close to each other even without significant process optimization, indicating the potential of the e-beam deposition approach.
The film deposited at 800 °C had about five times higher resistance, mainly due to a lower carrier concentration, and a comparable mobility.
In the flat and crystalline regions of the e-beam deposited layers, some defects, often squares, appeared (
Figure 5a,b). Many of these defects appear to have originated around silicon droplets/particles in the deposited layer, as can be seen in the SEM images in
Figure 5a,b. The defects in the 1000 °C film were accompanied by the formation of twin structures. The size of the defects/squares varied, which indicated that they originated from different depths in the film. Despite these defects, most of the surface area was flat, homogeneous, and smooth. It should be noted that similar defects have been observed after the CVD growth of Si layers [
22]. Investigations to reduce the number of defects through more careful control of the process parameters have shown that regular cleaning of the e-beam chamber is the most efficient means to reduce the number of defects in the films.
Figure 5c,d compares the surface structure of two layers deposited at 1000 °C before chamber cleaning 5c and directly after cleaning 5d.
Figure 5 shows that such simple measures can significantly reduce the particles and associated defects. Other attempts to reduce the number of defects, such as using a reduced deposition rate at the beginning of the process, did not give a reduced number of defects. Usually, the growth of highly crystalline films requires very slow deposition rates ranging from several to several tens μm/h [
23], and if the rate is much higher, it can affect the single-crystal film growth. The high deposition rates can potentially create specific defects [
22], but this appeared not to be the only reason for the defect formation.
Another related reason, which caused the appearance of defects upon high deposition rate e-beam evaporation, was the formation of metal spits/droplets from the melt reaching the surface of the substrate during the evaporation process, where they solidified as particles in the deposited film. “Spitting” can usually be resolved by proper rise and soak times during the homogenization of the melt. Additional initiatives aimed at enhancing the process conditions to effectively manage and decrease defect density need to be implemented. One of the options is to raise the deposition temperature above 1000 °C, up to 1100–1200 °C, as carried out in [
24] for epi-Si-based layers deposited using CVD-based processes.
Secco etching was performed for 5 min to reveal the extended defects of the films, and
Figure 6 shows etch pits on films deposited at 800 and 1000 °C. The dislocation densities were determined to be ∼2 × 10
7 cm
−2 and ∼5 × 10
6 cm
−2 at 800 and 1000 °C, respectively. It is obvious that the structural quality of the films depends significantly on the deposition temperature and that the film deposited at the highest deposition temperature (1000 °C) had the lowest density of defects.
Initial tests were performed to verify the feasibility of the exfoliation of e-beam epi-Si layers from a support substrate. Here, a ~140 µm thick layer was deposited onto a Si “mother” substrate with a porous layer, similar to that reported in [
25,
26]. An image of the as-deposited epi-Si/porous Si structure is shown in
Figure 7. A partial exfoliation of the central part of the epi-Si layer can be seen after fast cooling of the epi-Si/porous Si/Si supporting substrate structure. This result serves as an initial and indicative proof of concept, demonstrating the feasibility of executing a lift-off process for e-beam deposited Si epi-layers deposited on supporting Si substrates with a porous Si weak layer. It should be noted that the “onset” of the exfoliation of the epi-Si layer in our case was observed without the use of the laser cutting of edges, laser notching, or any implementation of any metallic stressor, which were used in [
26]. Improvements to the lift-off process for e-beam deposited layers are needed, but these initial findings indicate its feasibility. The observed initiation of exfoliation is akin to the blistering phenomenon [
27], previously studied as an indicator of the feasibility to implement the “Smart-cut” process.