Classification of different post-hyperdoping treatments for enhanced crystallinity of IR-sensitive femtosecond-laser processed silicon (2024)

Hyperdoping refers to the incorporation of dopants or impurities into a semiconductor substrate, such as silicon, above the solid solubility limit [1, 2]. It is commonly achieved by one of the following two procedures [3]. First, by ion implantation combined with subsequent rapid thermal process [4], or second, by ultrashort laser pulses that locally melt the substrate to a depth of tens to a few hundreds of nanometers below the surface [5]. The laser hyperdoping process often takes place in an atmosphere that contains the dopant in a gaseous form.

Hyperdoping can be used to incorporate deep level impurities in a sufficiently high density to form an intermediate band [6]. This enables the material to absorb photons with an energy below the original bandgap energy. The extended absorptance promotes this material system as a candidate for near-infrared photodetection or solar cells [5, 7–10]. Here, we use Si as a substrate and hyperdope it with sulfur via femtosecond laser pulses.

On the one hand the laser hyperdoping process works well in introducing impurities and thus elevating sub-bandgap absorptance, while on the other hand it induces a variety of detrimental crystallographic defects and amorphous phases [11]. These defects cause additional carrier recombination and lower the carrier mobility and thus, ultimately reduce the quantum efficiency [12]. The suitability of a hyperdoped material system for optoelectronic applications can be evaluated by a figure of merit, which is the ratio of the minority carrier lifetime in the hyperdoped region and the time it takes for the charge carrier to traverse this region [13–15]. The figure of merit, which ideally is as large as possible, thus depends on the thickness W of the hyperdoped layer, the carrier mobility µ, the carrier lifetime τ and the (sub-bandgap) absorption coefficient α [14, 15]. The impurity concentration N_t typically affects some of these parameters in an inverse manner. For example, an increase in N_t results in a larger α, i.e. a higher figure of merit, but usually in a lower µ and τ, and possibly also in a larger W. Sher et al have studied the impact of the sulfur concentration in Si on the figure of merit and found an optimum impurity concentration that results in a trade-off between these counteracting effects [15]. However, they have used a combination of ion implantation and subsequent pulsed laser melting as the two-step hyperdoping method for their study, which, in the form of the implantation dose, holds an additional degree of freedom to control the impurity concentration. Options to control the hyperdopant profile when using ultrashort laser pulses in a dopant containing atmosphere as a one-step method are more limited and require careful process control [16].

Post-hyperdoping processes can be used to manipulate the hyperdoped layer properties and/or to restore the crystallinity to enhance the figure of merit, as higher crystallinity indicates less amorphous phases, i.e. a higher carrier mobility and a higher carrier lifetime. We distinguish between (a) etching techniques, where the topmost surface layers are subsequently removed, and (b) thermal treatments, where defects get annealed. The etching approach benefits from the fact that segregated and excess dopants or other impurities concentrate at the very surface and get irreversibly removed. The effect of wet-chemical etching on the lifetime [17] and absorptance [17, 18] of fs-laser processed Si has been explored before. In addition, reactive ion etching (RIE) [19] and ion beam etching (IBE) [20] have been applied to fs-hSi and their influence on the absorptance has been studied. However, these reports either not examine hyperdoped material and/or use larger etching depths and/or do not take the impact on crystallinity into account.

As mentioned, thermal post treatment methods can also be used to manipulate the material properties. These treatments are characterized by the peak temperature, the heating and cooling rate and whether they heat the whole substrate or only the surface region [21]. Thermal annealing of fs-hSi for a duration of minutes at temperatures above ∼450 °C and below the melting temperature of Si proves to effectively restore crystallinity as it reduces amorphous and poly-crystalline phases [22]. However, in the case of hyperdoping Si with chalcogens, the sub-bandgap absorptance decreases significantly after such long-term thermal annealing [23, 24]. Note that the temperature at which deactivation occurs and the absolute degradation for a fixed temperature varies depending on the type of chalcogen, with Se and Te proving to be more stable compared to S [11, 25, 26]. Annealing with nanosecond laser pulses represents an extreme form of thermal annealing. Here, the surface region is locally heated and the substrate serves as a heat sink, providing high cooling rates that maintain [27] or even re-activate [28] sub-bandgap absorptance. In contrast to etching, thermal treatments do not remove material from the original sample, provided that the annealing temperatures do not exceed the evaporation temperature, or, in case of laser annealing, the ablation threshold.

Previous studies on post-hyperdoping treatments mostly focus on a single or few techniques and/or treat different material systems, which impedes a comparison and hence classification of post-hyperdoping techniques. In our contribution, we evaluate three different etching techniques (IBE, RIE, HNA) as well as ns-annealing and minute-long thermal annealing on a hot-plate as a thermal treatment. The impact of each method on the fs-hSi properties is characterized by the change in respective sub-bandgap absorptance and in the Raman spectra. While the absorptance is proportional to the remaining optically active sulfur states in the material, we take the Raman spectra to assess the crystallinity of the material by comparing the signal intensity of amorphous and crystalline signatures. Here, we use a larger crystallinity as an indirect measure and a sufficient condition for a larger figure of merit. We furthermore include fs-hSi with different surface roughness to check for a potential dependence on specific post-treatments. In summary, this work gives a detailed overview on post-hyperdoping techniques that have been specifically optimized and designed for fs-hSi, which allows for a comparison and classification of these methods.

We use monocrystalline double-side polished, Czochralski-grown (100) boron doped (1–10 Ω cm) silicon wafers with a thickness of 280 ± 20 μm and a diameter of 2" and place them in a stainless steel vacuum chamber. The chamber has a glass window at the top end for laser beam entrance and is mounted on a motorized x–y-stage. We first evacuate the chamber to a pressure of <100 mbar and subsequently fill it with SF₆ to a partial pressure of 675 mbar. This pressure is then stabilized under a constant gas flow rate of 100 sccm. The constant gas flow removes residual air molecules and assures that no depletion of SF₆ occurs during processing.

The laser pulses are emitted from a Ti:sapphire laser amplifier (Spitfire Ace from Spectra Physics) and exhibit a central wavelength of 800 nm, a pulse duration of 100 fs. They are focused through the glass onto the Si substrate by a lens with a focal length of 1000 mm. The 1/e²-Gaussian beam diameter is ω₀ = (216 ± 5) μm.

We irradiate silicon wafers by scanning the surface with two (N, ϕ₀)-combinations, where N is the pulse density, given in pulses per spot (pps), and ϕ₀ the peak fluence in J cm⁻². The pulse density is defined by N = π ω₀ ² /d_xd_y with d_x and d_y being the distance between spots in x-and y-direction. We process six square-shaped (7 × 7) mm² fs-hSi areas on each wafer: three with a parameter combination of (20 pps, 0.6 J cm⁻²) and three with (100 pps, 1.2 J cm⁻²). The laser process parameters result in a different surface roughness, which we refer to as moderately (20 pps) and heavily structured (100 pps). All wafers are cleaned in a SC-1 and SC-2 bath after the laser process.

For ns-annealing, we use a ns-laser that is frequency doubled to 532 nm, and exhibits a pulse length of 8 nanoseconds, a Gaussian intensity profile and a 1/e² spot diameter of (250 ± 10) µm at a repetition rate of 100 kHz. We set the pulse density to 20 pps and apply 6 different peak fluences of 0.3 J cm⁻², 0.4 J cm⁻², 0.6 J cm⁻², 0.7 J cm⁻², 0.8 J cm⁻² and 1.0 J cm⁻².

Thermal annealing takes place on a hot plate in ambient air. Four wafers are annealed for 30 min at 360 °C, 440 °C, 520 °C and 600 °C, respectively.

For the IBE process, the process chamber is evacuated to a pressure of 3⋅10⁻⁶ mbar and filled with argon to a pressure of 2.4⋅10⁻⁴ mbar afterwards. A Kaufman ion source (NTG Kaufmanquelle) delivers an argon ion beam diameter of about 8 cm. The argon ions are extracted from the discharge plasma of the Kaufman source with an acceleration voltage of 100 V. Afterwards they are neutralized by electron emission from a tungsten cathode. We use one wafer and process it successively in steps of 4 min. These conditions result in an etching depth of 50 nm on a planar c-Si wafer, which we determine by profilometer measurements (DEKTAK3) on a partially photoresist-protected reference wafer. Each sample is characterized after each 4 min IBE-process. We repeat this cycle three times on the same wafer, so we end up with a cumulated process time of 12 min. In contrast to the following etching methods, the IBE process is a purely physical etching technique.

For the capacitively coupled RIE performed in a Vacutec Plasma System AB 1500, SF₆ at a pressure of 80 mbar with a gas flow rate of 20 sccm is ignited by an alternating frequency of 13.56 MHz at a power of 10 W, resulting in a bias voltage of −15 V to −20 V. Four wafers are etched for 80 s, 90 s, 110 s and 125 s respectively. For the RIE process, we find an average etching rate of 92 nm min⁻¹ on a planar c-Si reference. As we target a low etch rate to resolve the removal of the thin hyperdoped layer, we apply a low plasma power. This, however, tends to make the process less reproducible. In addition, the etch depth is non-linear in the process time, so the latter must be checked repeatedly on c-Si references. The RIE etching process we apply here is mainly of chemical nature with a smaller physical component, which is quantified by the bias voltage.

The anisotropic wet chemical etching (HNA) is done by using a diluted solution of 25 ml CH₃COOH (100 %), 11 ml HF (10 %), 66 ml HNO₃ (65 %) and 60 ml deionized H₂O. We process five wafers in a polytetrafluoroethylene container for 10 s, 20 s, 30 s, 40 s and 125 s respectively. The etch rate of this purely chemical process decreases over the etching processes due to a consumption of the dilute HF, so, similar to RIE, the etch rate is non-linear in the process time. We determine an initial etch rate of 166 nm min⁻¹ on a c-Si reference.

We use a photo spectrometer (Lambda750 by Perkin Elmer) with an integrating sphere with a diameter of 15 cm to measure the reflectance R and transmittance T. The absorptance A is determined by A = 1 − R − T. We utilise a Raman spectrometer (DXR3 by Thermo Fischer Scientific) to determine the crystallinity of the annealed areas. We normalize the Raman spectra by setting the peak value of the c-Si peak at 520 cm⁻¹ to 1. We characterize each of the six areas per wafer before and after each post-processing step and average the results for each laser parameter over the three sub-areas, if not mentioned otherwise. The surface morphology is investigated via a scanning electron microscope (Jeol JSM-6380LV). We acquire top-view micrographs under an angle of 20° of each sample after the hyperdoping process and after each respective post-treatment. The cross-section SEM micrographs were taken at an angle of 3°.

Figure 1 shows top-view SEM micrographs of the 20 and 100 pps samples directly after the hyperdoping process and after the different post-hyperdoping methods. The bright areas in the SEM micrographs show areas protruding from the surface. It has been shown that these protrusions contain most of the pressure-induced crystal defects, whereas resolidified amorphous silicon forms a thin conformal film on the surface [29, 30]. After hyperdoping, the two laser parameter combinations result in different levels of µm-sized surface roughness as shown in figures 1(a) and (b). We measure the height of the protrusions to be below 1 µm for the 20 pps samples and around 2 µm for the 100 pps samples, using cross-section micrographs as shown in figure 2(a). We therefore refer to the 20 pps samples as moderately and to the 100 pps samples as heavily structured fs-hSi, respectively. Cross-section micrographs (figure 2(a)) further show a nm-sized roughness on the surface after the hyperdoping process that we associate with chemical etching by an SF₆ plasma [31].

The samples annealed with a ns-laser fluence of 0.7 J cm⁻² are shown in figures 1(c) and (d). Because this fluency is below the ablation threshold, at least for the heavily structured samples, the changes in surface morphology are only minor.

The micrographs of the etched samples are shown in figures 1(g)–(l) and correspond to an etch depth of 150 ± 20 nm respectively, as measured on a c-Si reference. The etch time is 240 s for IBE, 125 s for HNA and 150 s and 110 s for RIE for the 20 pps and 100 pps samples respectively. All etching processes lead to a partial planarization of the 20 pps samples, what can be seen in figures 1 (g), (i) and (k). The heavily structured 100 pps samples retain their structure through the IBE process, as shown in figure 1(h). However, the comparison of the cross-section micrographs in figures 2(a) and (b) show that the nm-scale roughness is smoothed out. Also, the heavily structured samples processed by RIE retain their µm structure (figures 1(j) and 2(c)). In contrast to IBE and RIE, the heavily structured samples etched with HNA show a planarization of the µm structure. The cross-section micrographs in figure 2(d) show that the µm structure becomes more planar and the height of the cones decreases to values below 1 µm. The top view micrographs in figure 1(l) show that especially the volumes that previously protruded from the surface are removed and depressions are formed there after etching.

We measure Raman and absorptance spectra after the different post-hyperdoping methods. In figure 3 all measured Raman and absorption spectra for the heavily structured 100 pps samples, before (black lines) and after the investigated post-treatment methods are shown. The moderately structured 20 pps samples are shown in the supplementary material. In each subfigure, we also include the respective spectra of the untreated c-Si reference (grey lines). The peak at a Raman shift of 300 cm⁻¹ also originates from the Si-I lattice. Analogous to Smith et al we assign the two peaks at a Raman shift of 354 cm⁻¹ and 395 cm⁻¹ to the Si-XII phase and the two peaks at 387 cm⁻¹ and 443 cm⁻¹ to the Si-III phase [29]. Both phases are pressure-induced poly-crystalline phases. For the 100 pps samples shown in figure 3, the two peaks of the Si-XII phase at 354 cm⁻¹ and 395 cm⁻¹ overlap with the Si-III phase at 387 cm⁻¹ and form one peak. The increased Raman intensities for the wavenumber ranges of 50 cm⁻¹–200 cm⁻¹ and 470 cm⁻¹–490 cm⁻¹ are due to resolidified amorphous silicon phases. These amorphous phases originate from the large resolidification velocities of the melt, which are associated with the ultrashort laser pulses. They are especially pronounced for fluences below the ablation threshold [29]. An ideal post-hyperdoping process reduces or removes the non-crystalline signatures in the Raman spectra while simultaneously maintaining the high sub-bandgap absorptance in the spectral region between 1.1 µm and 2 µm.

All post-hyperdoping processes we apply lead to a reduction of non-crystalline phases but also result in a reduction of the absorptance. To quantify the treatments and allow for an easier comparison between them, we introduce a measure for crystallinity analogous to Franta et al [27]. The crystallinity is determined by integrating the area under the Raman spectrum for the wavenumber from 470 cm⁻¹ to 490 cm⁻¹ and then dividing it by the c-Si integral from 515 cm⁻¹ to 525 cm⁻¹, see also equation (1).

The reference value for the crystallinity is determined with the same method, except that the spectrum of a monocrystalline and unstructured wafer is used. This leads to an offset value in the crystallinity measurements shown in figure 4.

To obtain a quantitative value for the IR absorption, the absorptance at 1450 nm is used as a representative value. Figure 4 shows the IR absorption plotted against the crystallinity as determined by equation (1). We account for variations in the initial state of the samples after hyperdoping by calculating relative changes for the different methods according to A/A₀ , and X_a-Si /X_{a-Si 0} respectively, where A and X_a-Si is the absorptance or the crystallinity value after the post-hyperdoping process and A₀ and X_{a-Si 0} the corresponding values in the initial state. Thus, the starting material shows a relative absorption and an amorphous silicon signal of one each. A good post-treatment method produces a low amorphous silicon signal and a high IR absorption, located in the top left corner of the graph.

The two thermal treatment methods, ns- and thermal annealing, represent the best and the worse post-treatment in point of view of achieving simultaneously high crystallinity and high IR absorptance. This holds for both surface morphologies. Thermal annealing provides the largest relative decrease in IR absorption for a given crystallinity. In contrast, ns-annealing leads to the highest absorptance with simultaneous improvement of crystallinity. Note that in both graphs the ns-anneal fluence increases going from right to left. Only when the ns-laser fluences exceed the ablation threshold, a reduction of the IR absorption takes place. The onset of diminishing IR absorptance sets in at 0.6 J cm⁻² for the 20 pps samples and at 0.8 J cm⁻² for the 100 pps samples. This difference is due to the varying degree of structuring and the resulting larger effective area of the 100 pps samples.

The etching processes are located between these two thermal processes. For the 20 pps samples, the etching processes provide higher absorptance for the same crystallinity than thermal annealing and lower ones compared to ns-annealing. However, no difference between the etching processes can be observed here. For the 100 pps samples, the etching processes also provide higher absorption compared to thermal annealing. Here, however, an influence of the etching process can be seen. RIE and wet chemical etching show a similar behavior, whereas the purely physical removal of the surface by IBE leads to higher absorption for a fixed crystallinity value. In the evaluation of the SEM micrographs, we have found that the µm-scale structure of the 100 pps samples is preserved by the IBE process, see figure 1(h) and the cross-section micrographs in figure 2(b). In the following, we discuss what leads to the different performances of the etching processes.

Note, that the sub-bandgap absorptance does not approach zero, as would be the case for a complete deactivation of optically active sulfur or removal of the hyperdoped layer. The first reason for this is simply that the post-hyperdoping processes applied here undershoot this transition. The other reason is due to a systematic measurement error, as has been described for this sample type in [32], when calculating the absorptance from reflectance and transmittance. Briefly summarized, the scattering properties of the sample unavoidably result in non-detected photons, which are then counted as being absorbed. This error depends on the sample properties and may overestimate the sub-bandgap absorptance by up to 20%_abs.

Figure 4(a) shows that there is no etching process to be preferred for the 20 pps samples. However, for the 100 pps samples, the IBE process shows a weaker decrease in IR absorptance for a given crystallinity. We investigate the planarization effects as it yields insights on how the morphology changes due to the different post-treatments. We use the absorptance at 750 nm as a measure of the roughness of the surface. Since silicon shows no transmission at 750 nm, the value depends only on the reflectance of the silicon surface and thus is independent of the sulfur concentration. We further determine the optical depth of the hyperdoped layer, i.e. the product of the absorption coefficient α and its effective thickness (see equation (2)), by applying the model of Schaefer et al on the measured absorptances [32]. The optical depth is a measure for the sub-bandgap absorptance of the hyperdoped layer and allows for a better comparison of the samples as it corrects for the different surface reflectance. The optical depth a is described in equation (2).

Here α is the absorption coefficient and W the layer thickness, in this case the thickness of the sulfur-doped layer. The absorption coefficient depends on the concentration of optically active sulfur atoms. Removal of the surface by etching processes reduces both the film thickness W and the absorption coefficient α, since the sulfur concentration decreases with increasing distance from the surface [33].

In figure 5 we plot the absorptance at 750 nm versus the optical depth for the used etching techniques. We also compare them to the thermal post-treatment methods. The dashed lines show a linear fit for each method. To show the different degrees of planarization for each etching technique we use the optical thickness as a measure for the etching depth of the respective etching processes and the absorptance at 750 nm as an indicator of the planarity of the surface.

As the etch depth increases, the optical depth decreases. In addition, the absorptance at 750 nm decreases because the etching processes levels the surface of the silicon, causing it to reflect more light. The slope contains information on how the surface roughness alters during the post-hyperdoping treatment: a small slope in the range of zero indicates a conservation of surface roughness under the removal or optical deactivation of hyperdoped material, while a larger positive slope points to a stronger planarization. A negative slope, as we observe for the ns-annealing, yields that while the surface becomes smoother, the optical depth increases and, thus, part of the previously optically inactive impurity states are transferred into optically active configurations. This could be explained by the dissociation of optically inactive S-complexes by the ns laser pulses, resulting in surplus optically active sulfur atoms. Alternatively, the increase in crystallinity results in significantly fewer grain boundaries and also less amorphous silicon in the crystal structure. Sulfur that was trapped in these grain boundaries, or present as optically inactive sulfur within the amorphous volumes, is now dissolved in the silicon lattice and becomes optically active. This corresponds to findings of Franta et al who report on an increase of sub-bandgap absorption upon ns-annealing [27].

We find that the optical depth of the thermally annealed samples decreases with increasing temperature. This is due to the reduction of the concentration of the optically active sulfur [23]. However, the temperature is not high enough to cause a smoothing of the surface and thus affect the absorptance at 750 nm within the scope of our measurement accuracy. This is the reason why the slope for thermal annealing is about zero, both for the 20 and 100 pps samples.

Now we discuss the different results for the etching processes. IBE, RIE and wet chemical etching exhibit a comparable slope for moderately structured samples shown in figure 5(a). This means that for all three processes an increasing etch time, i.e. decreasing optical depth, leads to a similar surface planarization behavior and thus slope in figure 5(a). This finding explains that, regardless of the etching mechanism used, we do not measure a significant difference in performance for the medium-structured 20 pps samples, see figure 4(a). The smoothening of the surface is also clearly visible in the SEM micrographs shown in figures 1 (g), (i) and (k).

For the 100 pps samples shown in figure 5(b), the HNA and RIE process have a similar slope. The IBE process in relation, however, has a smaller slope. So, an increased etching depth with the IBE does not result in an as severe leveling of the surface as with the other etching processes. A possible explanation for the different etching effects is the different selectivity of the etching processes. That is, the ratio of the etch rate of amorphous to crystalline silicon. Wet chemical etching has a higher selectivity compared to dry etching methods [34]. It is known that after hyperdoping mainly the cones and the structures protruding from the surface are amorphous, whereas the valleys of the structure have a smaller amorphous fraction [29, 35]. The higher etch rate of the a-Si by the HNA solution leads to a preferred removal of the structures protruding from the surface and a strong planarization occurs. This effect can be observed on the SEM micrographs in figures 1(k) and (l) as well, both for the 20 pps samples and the 100 pps samples. The structures protruding from the surface, with a high crystal defect density are completely removed by the wet chemical etching, and even etch pits and troughs are formed in these places. As a result, the surface is more planar than before.

The difference between the RIE and IBE process, however, can be explained by the anisotropy and angular dependence etch rate of the IBE process, compared to the isotropic removal by RIE. The sputtering yield, i.e. etch rate of the IBE process is dependent on the angle between the ion beam and the surface normal [36–38]. We use the software TRIM [39] to simulate the angle dependent etch rate of silicon under bombardment with argon ions (see figure S2 in supplementary). In order to simulate the ablation by RIE we assume a completely isotropic ablation. Figure 6 shows the simulated evolution of the surface contour under (a) IBE and (b) RIE treatment. We extract the original contour from cross-section SEM micrographs (see figure 2(a)). The details of the simulation, used to create figure 6, are shown in the supplementary material. We find that the IBE process preserves protrusions with a larger texture angle and material gets removed at the top and in between the cone structures. For the RIE process, the protrusions first become thinner, as can be seen in the SEM cross-section micrograph in figure 2(c) as well as the simulation shown in figure 6(b). When simulating higher etch depths the needles are eventually almost removed by RIE as shown in figure 6(b).

The different effects of the etching processes thus lead to different surface geometries, which explain the different absorptance values in the visible wavelength regime and hence the smaller slope of the linear fit for the heavily structured sample processed by IBE compared to the other etching techniques in figure 5(b). When we compare the different etching methods for the heavily structured samples IBE leads to the highest absorptance values at a given crystallinity (see figure 4(b)).

The ns-annealing process achieves the best results in terms of crystallinity and absorption. Our results can be compared with Franta et al. Analogous to Franta et al we observe an increase in crystallinity while maintaining IR absorptance. Only when the ns-laser fluence exceeds the ablation threshold, the absorptance in the visible and IR range reduces. The ablation threshold depends on the surface morphology in that the threshold fluence for ablation increases with the height texture angle of the protrusions. The higher texture angle for the 100 pps samples results in a larger effective surface area and thus a reduced local fluence.

We study three different etching techniques and two thermal treatments with different temperature gradients in their ability to increase the crystallinity while maintaining the sub-bandgap photosensitivity of femtosecond-laser pulse, sulfur hyperdoped silicon (fs-hSi). The respective impacts are compared for two different surface morphologies that are laser processed with different pulse densities. In our classification, we take the crystallinity, i.e. the absence of amorphous or poly-crystalline phases that are induced by the laser hyperdoping process, as an indirect measure for a larger figure of merit. We find that post-processing via thermal annealing largely depends on the temperature gradients involved. The two extremes we investigate, high-gradient ns-laser pulse annealing and small-gradient long-term thermal annealing on a hot plate, both effectively reduce amorphous phases, but differ in their ability to keep the material photosensitive for sub-bandgap photons. In this respect, ns-annealing outperforms all other investigated post-processes, provided the fluence of the ns-pulses is low enough to not ablate hyperdoped material. The onset of ablation depends on the surface roughness of the material. For moderately and heavily structured fs-hSi, we find fluences of 0.6 J cm⁻² and 0.8 J cm⁻², respectively, to conserve at least 90 % of the sub-bandgap absorptance and reduce the amorphous to crystalline Raman peak intensity by at least 80%. Thermal annealing on a hot plate, however, proves to be inferior to the other post-processes in our classification criteria, as it decreases photosensitivity the most.

The etching techniques range in between these extremes for both morphologies and do not show any preferences, when it comes to moderately structured fs-hSi. Here, we find a monotonous lowering of absorptance and crystallinity that is proportional to the exposure time and hence points to an isotropic removal of the hyperdoped layer and a successive planarization of the surface. For heavily structured samples, however, the IBE outperforms RIE and HNA processes. The directed ion beam results in a non-isotropic, physical removal of hyperdoped material and conserves the cone-like structures. This is beneficial for a post-hyperdoping process, as it becomes more controllable and prevents overetching. The other etch techniques, i.e. RIE and HNA are more isotropic also for the heavily structured fs-hSi. The HNA technique shows a pronounced etching of the defect-rich cone structures. This specific selectivity must be accounted for, when applying the HNA technique.

We have shown that a manipulation of hyperdoped Si is feasible for a one-step fabrication method of hyperdoped Si with fs-laser pulses. Our classification criteria of high absorptance and crystallinity yield that ns-annealing is most suited as post-hyperdoping process for fs-hSi. In addition, IBE should be preferred over other etch processes when it is desired to reduce sulfur concentration or the hyperdoped film thickness. We can also think of a combination of different post-processes to further tailor the sample characteristics for an optimum performance in a, opto-electronic device.

The authors acknowledge financial support from the Deutsche Forschungsgemeinschaft DFG under Grant No. 429413061 and the Federal Ministry of Education and Research of Germany under Grant No. 03INT701AA.

The authors would like to thank Professor Dr Jutta Kerpen and her team for providing us the opportunity of measuring with the Raman spectrometer and thank Ingo Lebershausen for technical support.