Synthesis and characterization of type II silicon clathrate films with low Na concentration

Si plays a central role in nearly all aspects of semiconductor technology. However, the most stable Si form, diamond-structured Si (d-Si), has an indirect bandgap and a low absorption coefficient which limit its application in light emission and absorption devices. Many other crystalline forms of Si exist^1–4 and in particular, type II Si clathrate is interesting since it has been predicted to have a direct or nearly direct bandgap of about 1.9 eV,^5–9 which would be an ideal range for Si-based LEDs or thin film solar cell applications.

Si clathrates tend to form as crystalline inclusion compounds, where the cage-like Si sublattice surrounds guest atoms such as Na, K, or Rb forming an open Si framework with a lower density than d-Si.^10–14 Na is the most commonly studied guest, and Na–Si clathrates can be formed in either type I (Na₈Si₄₆), or type II (Na_xSi₁₃₆, 0 < x < 24) phases.^10,11 Typically, Na in Si clathrates is viewed as an interstitial guest. However, if the Na content is low enough, which can only be achieved in type II Si clathrate Na_xSi₁₃₆ (x ≪ 1), the clathrate can become a semiconductor with Na acting as a shallow n-type dopant.^15,16 Most studies of Si clathrates have focused on powder samples.^10,11 Powders, however, are not the preferred form factor for optoelectronic devices. Realizing the potential of clathrates as optically efficient Si-based semiconductors requires the synthesis of high-quality Si clathrate films. This, in turn, relies on a fundamental understanding of the synthesis process and of defect structures and inhomogeneities that can form during the synthesis.

There have been several attempts by our group and others to grow type II Na_xSi₁₃₆ films on Si wafers^6,17–20 and other substrates.⁶ In almost all cases, Si clathrate films were formed using a two-step process adapted from powder synthesis.^10–12,21 It begins with formation of a Zintl precursor NaSi film by reacting Na metal with Si followed by thermal decomposition of the precursor into the clathrate film.¹⁶ However, this thermal decomposition technique is very sensitive to the local environment including the Na local vapor pressure,¹⁶ reactor geometry,⁶ Na purity and form,⁶ and annealing time.¹⁹ Many films on Si wafers adhered weakly to the substrate,⁶ which limited post-growth treatments to reduce the Na concentration and complicated measurements of film properties. Moreover, optical properties of Si clathrates have been difficult to verify. The previously reported optical data have been mostly from powder samples.⁸ Only a few preliminary optical results are published,^6,18,19 but the main features and interpretations of the results are complicated by the unclarified defect impurities.

In this paper, we explore the two-step film synthesis process in more detail. Using x-ray diffraction (XRD), Raman scattering spectroscopy, and Electron Paramagnetic Resonance (EPR), we find evidence of a disordered or amorphous Si-like phase in the film that dominates the film surface and grain boundaries and leads to inhomogeneous properties across the film. Using Time-of-Flight Secondary Ion Mass Spectrometry (TOF-SIMS) in conjunction with Electron Paramagnetic Resonance (EPR), we demonstrate inhomogeneity in the Na distribution which is likely associated with the grain structure and presence of impurity phases. We demonstrate approaches to remove the defective phases or minimize their impact allowing more reliable characterization of optical properties. We also identify key factors leading to lower Na concentration and thinner films. This leads to films with phase purity and reduced Na concentrations that are comparable to or better than the best powder samples that have been reported. It also provides a pathway to improved thin films for subsequent Si-based applications in photovoltaic PV or optoelectronic industries.

Our synthesis of type II Na_xSi₁₃₆ (x ≪1) clathrate films follows the approach described in Ref. 6 with some modifications aimed at producing thin, uniform, and well-adhered clathrate films with low Na concentration. The approach is adapted from powder clathrate synthesis processes.^12,15 Instead, however, of using d-Si powder as the silicon precursor, films are formed directly on Si wafers or alternative substrates, such as Si on sapphire (SOS). The latter simplifies optical characterization. For Si wafers, 100 mm, n-type Phosphorus doped Czochralski ⟨100⟩ wafers of thickness 380 μm, and a resistivity of 5–10 Ω.cm were used. For SOS (from Kyocera), a 100 mm wafer with a Si layer thickness of about 500 nm was used. It was nominally undoped with a high resistivity.

All procedures in the synthesis were performed in a nitrogen purged glovebox (< 1 ppm O₂ and $∼0$ ppm H₂O) or in a furnace under flowing high-purity argon or vacuum (⁠$∼10$ mTorr) to minimize exposure of the Si and Na precursors and subsequent films to O₂ or H₂O. The first step involves the formation of the precursor NaSi film. The same process was followed for both Si wafer and SOS substrates. A Si substrate was diced into 70 × 17 mm² (Si wafer) or 20 × 20 mm² (SOS) pieces. A layer of metallic Na (99%, sodium cubes, Sigma-Aldrich) was spread on the bottom of a ceramic crucible with a slightly smaller opening than the size of the Si pieces. Sodium metal converted from sodium hydride (90%, powder, Sigma-Aldrich) by a preliminary annealing step at 550 °C under flowing Ar (ultra high purity) was also used. The Si substrate was placed on top of the crucible with the polished surface down. The distance between the Si substrate and Na source was about 15 mm. The crucible was then wrapped using 99% pure Al foil and inserted into a quartz furnace tube with sealed end caps that contained quick connects for subsequent argon purge. This was all done inside the glovebox. The sealed furnace tube was then removed from the glovebox and inserted into the furnace body. The crucible was annealed at 550 °C under flowing Ar to form the precursor NaSi film on the substrate. The temperature ramp during this step can affect synthesis, with clathrate films sometimes not forming if the ramp rate was too fast. A temperature ramp of 5 °C/min to 550 °C was used in this step.

The second step involves decomposition of the NaSi to form the desired clathrate (Na_xSi₁₃₆) phase. For the decomposition step, the sealed furnace tube was returned to the glovebox, and the sample was removed from the crucible and placed in a furnace tube with the film surface exposed. The furnace tube was sealed and returned to the furnace body where the film was annealed at 400 °C under vacuum (⁠$∼10$ mTorr) for 24–72 h. After the annealing stopped, the tube was left in the furnace body until arriving at room temperature to avoid abrupt change of temperature which can lead to less adhered films. The temperature ramp rate and annealing time during this step were adjusted to produce thinner films dominated by the type II Si clathrate phase. Our previous studies of powder synthesis have shown that there is a trade-off between the formation of type I and type II silicon clathrate phases, which depends on the Na overpressure during decomposition.¹² Given that NaSi is formed at a temperature where the Na partial pressure is close to 1 Torr, it is possible that the NaSi contains excess Na either as metallic Na or as excess Na within the crystal lattice although XRD does not show evidence of a large amount of metallic Na phase. Previous work on powders has shown that an excess of Na during the decomposition to clathrate can result in a large fraction of type I silicon clathrate. For this reason, and to try to lower the overall Na content in the final films, before the decomposition step, some of the films received an additional annealing step at 250 °C for 3–24 h under vacuum (⁠$∼10$ mTorr). This temperature is too low to decompose the NaSi but allows removal of excess Na from the sample.

After synthesis, all films were rinsed in ethanol followed by DI water to remove residual Na. During the water rinse, trace Na caused formation of H₂ bubbles and NaOH on the sample surface, but NaOH is diluted and washed away quite quickly with no etching of the samples observed. Samples were left in the water until this process was complete. This step-by-step approach to eliminating trace Na was particularly important for the wet, HF/HNO₃ etch treatment discussed below. Without these steps, rapid reaction between the etch solution and sample when residual Na was present often removed the film from the substrate surface as reported for previous films.⁶ Many films were prepared as part of this study. Here, we discuss in detail a representative series of samples which illustrate the effects of various growth parameters and treatments on film properties. Preparation conditions for these samples are listed in Table I.

Raman scattering measurements were performed in two setups. The first used a 514 nm laser source and the light was passed into a Spex 1877 Triplemate 0.7 m monochromator with a 1200 g/mm final grating. The bandwidth was set by the width of the final slit, which was 100 micrometers. The light was detected with an Andor DU970P electron multiplying CCD. The second setup is a WiTec Alpha 300 Confocal Raman Microscope equipped with a 532 nm laser, which has a 1 μm probing spot. Since the first Raman setup allowed more flexibility over power and filtering of scattered light, it was generally used in long time acquisitions and to probe lower wavenumbers. The confocal setup allowed more rapid measurements for screening and higher spatial resolution. The two systems, however, gave essentially the same results.

Micro-transmission, reflection, and absorption spectra were taken using an Ocean Optics USB 4000-VIS-NIR spectrometer with an Ocean Optics LS-1 Halogen Tungsten light source focused through a Nikon Metaphot V microscope. Newport Research blue-green and long pass colored glass filters and neutral density filters were used in the beam path to avoid detecting second order light or saturating the detector. These measurements required thin layers so were made on SOS prepared films. A sapphire substrate with the Si layer etched off was used as the transmission reference.

Photoluminescence (PL) spectra were acquired using an Acton 300i spectrometer with interchangeable 150 and 600 g/mm gratings and a Princeton Instruments Spec-10:100BR liquid nitrogen cooled silicon CCD detector array. Excitation sources were a 460 nm diode laser or the 514 nm line of an argon ion laser using 3 mW of power focused to an $∼1$ mm spot. Both sources passed through narrow bandpass filters before exciting the sample. In addition, long pass filters were inserted in the collection path to minimize scattering of laser light into the spectrometer. For 460 nm excitation, this was a 515 nm Newport long pass glass filter. For 514 nm excitation, a Semrock stopline notch filter was used. The filter began transmitting within $∼300$ cm⁻¹ from the laser line, allowing Raman scattering spectra to be collected from the same sample spot as the PL. Samples were mounted on the cold finger of a closed cycle helium refrigerator allowing measurements from 15 to 300 K.

3D ion images and ion intensities were acquired using a IonTOF Time-of-flight secondary ion mass spectrometer (TOF.SIMS 5) with a primary ion beam of Bi⁺ and secondary ion beam of $O2+$⁠. EPR spectra were acquired using an X-band Bruker EMX EPR Spectrometer equipped with a continuous flow cryostat following the approaches outlined in Ref. 15. For the work presented here, measurements were made from 77 to 300 K using liquid nitrogen as a cryogen. Crystalline phases present were identified using a D500 Siemens x-ray diffractometer with a Cu Kα radiation tube (voltage 30 kV, current 25 mA). Cross-sectional and plan view SEM images were acquired with a Phenom-pro benchtop scanning electron microscope. Reactive Ion Etching (RIE) using SF₆ was performed on an AutoGlow 200 Plasma System. The film thickness of the SF₆ etched samples was measured using a KLA Tencor D-600 optical profilometer.

The properties of the synthesized Si clathrate films are very sensitive to growth parameters such as temperature ramp, annealing time, and treatments after film growth. These effects were therefore investigated in order to achieve a high-quality film. Before the thermal decomposition process, the precursor film formed in the first step was characterized using XRD to confirm that it was dominated by the NaSi Zintl phase. Since NaSi is flammable in contact with air, an airtight sample holder was used. The XRD peaks observed were consistent with those calculated based on the crystal data of NaSi (see Fig. S1 in the supplementary material).

In this two-step synthesis process, we expect that the thickness of the resulting clathrate film is primarily determined by the thickness of the NaSi layer created in the first step, and this is consistent with the results shown in Fig. 1. For these films, the annealing time of the first step was varied while the second step annealing was performed for 24 h for all the samples. Reducing the first step annealing time leads to thinner films. To fit the dependence of film thickness on first step annealing time, we considered a model of NaSi formation where Na vapor diffuses through the NaSi layer and new NaSi growth occurs at the buried NaSi/Si interface (inset in Fig. 1). This is analogous to the Deal–Grove model used to interpret the growth of a thermal oxide on the surface of Si wafer.²² More details of the fitting are given in Fig. S2 of the supplementary material. The fact that the fitted curve in this work is close to linear suggests that growth of the precursor film is more likely limited by reaction than diffusion, which is not surprising since Na is a fast diffusor in diamond Si.²³ 

Being able to produce relatively thin films which are less than 1 μm to 10's of micrometers is important both for characterization and applications of Si clathrate. Understanding the impact of the annealing time and the rate-determining step in the NaSi formation is key to this, and Fig. 1 shows for the NaSi formation temperature used, sub-1 h annealing times are required to produce relatively thin films. Most of the films reported here were produced using first step annealing times in this range (see Table I). In attempting to produce thin films, however, an issue with the presence of a disordered phase of silicon arose with the fraction of disordered phase increasing for thinner films. This is discussed next.

Characterization of the clathrate films provides direct evidence for the presence of a disordered or “amorphous-like” phase of silicon. Figure 2(a) shows the XRD pattern for sample 1 (prior to the acid etching discussed below) along with an XRD pattern for an amorphous silicon sample produced by conventional plasma enhanced chemical vapor deposition (PECVD) growth. This sample is one of those that received an additional anneal to reduce Na content between the NaSi and clathrate formation steps as discussed above. In general, the additional anneal did reduce Na content in the final film as determined by XRD. In addition to the sharp peaks associated with the presence of crystalline type II Si clathrate, sample 1's XRD pattern exhibits two broad background peaks at around 27° and 50° 2θ that match the amorphous Si peaks, suggesting that an amorphous Si-like phase exists in the film.

In general, decreasing the annealing time in the first step to produce thinner films increased the fraction of a disordered phase observed in XRD. Specifically, for the films thicker than 100 μm, the amorphous phase observed in the XRD pattern (see supplementary material Fig. S3) is extremely low relative to the crystalline clathrate phase, while for films grown for less than 40 minutes (< 10 μm) we did not observe a crystalline phase in the diffraction pattern.

The presence of a disordered silicon phase is supported by a comparison of Raman scattering measurements from the top surface of the film to amorphous silicon as shown in Fig. 2(b). The Raman spectrum of amorphous silicon has been studied extensively.^24,25 It exhibits a broad peak at $∼480$ cm⁻¹ as seen in the spectrum in Fig. 2(b). The remaining spectra which were taken at three different locations on the sample surface exhibit an amorphous-like characteristic with no evidence of the presence of clathrate. This was true across the surface. In obtaining Raman spectra careful attention was paid to keeping the laser power below the threshold for laser heating induced conversion to diamond silicon as illustrated in Fig. S4 of the supplementary material. The presence of a disordered phase on the film surface is further supported by the SEM images shown in Figs. 2(d) and 2(e) for sample 5 before acid etching (discussed below). The film is made up of grains from 10 to 100 μm in lateral size. A rough, disordered layer is visible at the surface of the film.

The presence of a disordered Si phase in the films is not surprising. Soon after the first synthesis of type II silicon clathrate powder, Mott argued that an “amorphous region” existed between individual Si clathrate grains in these powders based on the observation of thermally activated hopping conductivity.²⁶ Recently, our group found evidence for this disordered phase in EPR studies of powder samples.¹⁵ This was achieved through detection of dangling bond defects which arise from unterminated bonds and have been studied extensively in amorphous silicon.^27,28 We also demonstrated that etching the powders reduced the relative size of the dangling bond signal allowing us to associate the disordered Si with grain surfaces. In a detailed EPR study of similar powder samples with varying Na content, Yamaga et al. also identified the dangling bond signature based on the g-value and width of the associated EPR peak.²⁹ They came to the same conclusion that its presence was evidence for an amorphous phase in their samples.

The same dangling bond signature is present in the films studied here. Following the approach used to analyze the room temperature EPR spectrum of powder samples,¹⁵ Fig. 2(d) presents the room temperature EPR spectrum of sample 6, an amorphous silicon EPR spectrum, and the difference between the two after scaling the amorphous spectrum to match the low field intensity of the clathrate spectrum. The g $∼2.005$ feature in the amorphous spectrum arises from Si dangling bonds and is also present in the clathrate. After subtraction, the EPR line at g $∼2.003$ that remains is attributed to free carriers donated by the Na.^15,29–31 The presence of dangling bonds in the clathrate films is further evidence for the presence of disordered silicon in the films.

The spatially resolved TOF-SIMS results in Fig. 3 provide evidence of an additional inhomogeneity in the films associated with the Na distribution. In the image in Fig. 3(d), we see high Na content at the film surface. Perpendicular to the surface, Na-enriched regions extend into the film. Between these regions are Si rich areas. The depth dependence is further explored in the traces of ion yield as a function of sputter time in Fig. 3. Figure 3(a) shows the normalized ion yield as a function of sputter time within the full 150 $×150$ μm² region of the image in Fig. 3(d). Smaller region of interest (ROI) profiles of high and low Na content regions were generated from the original 150 $×150$ μm² profiles indicated in the image of Fig. 3(d). Since we do not have a standard to calibrate ion yields in the clathrate structure, we cannot determine absolute elemental concentrations. Instead, the results are normalized point by point so the total relative yield is unity. This allows relative intensities of different elements to be tracked as a function of sputter time. Because the surface is rough, depth calibration is also difficult although SEM images of the etch pit after measurement indicated sputtering through most of the film.

For all the traces, the top surface shows an excess of Na. For the high Na content region [Fig. 3(b)] relative Na yield exceeds Si throughout the sputter depth. For the low Na content region [Fig. 3(c)] Na intensity is about two orders of magnitude lower than Si beyond the top surface, which is consistent with the Na relative concentration in the clathrate phase estimated from x-ray refinement (∼Na₁Si₁₃₆). We note, however, that the ion sensitivity factor of Na is generally much higher than Si in this type of measurement.³² Hence, while the trends would be the same, the absolute Na content is likely much lower relative to Si. The fact that Na intensity is stronger than Si over the entire sputtering time at the high Na content region leads to a Na intensity close to Si in the overall profile in Fig. 3(a). Further calibration based on the sensitivity factor will be performed to obtain more accurate results. K⁺ ions which roughly follow the profile of Na are also observed. This film was prepared using NaH powder (90% pure) and we believe that K is an impurity in the powder which diffuses into the silicon substrate along with the Na during NaSi formation in step 1. Whether or not that K resides in the clathrate cages will need further investigation.

We associate high Na content with the surface and grain boundaries. It's natural to suggest these are also areas where disordered Si is present with the high-quality clathrate phase located in low Na content regions below the surface.

Raman probes the optical absorption depth in the sample making it surface sensitive. XRD samples the entire film. The fact that the Raman spectra in Fig. 2(b) only exhibit a disordered characteristic while XRD [Fig. 2(a)] indicates that a large fraction of the film is clathrate supports the suggestion above that the clathrate phase lies below a disordered Si surface. To test this, a buried interface was exposed and studied by adhering tape to the film surface and pulling the film off the Si substrate.

The crystallites making up the film were effectively inverted by the exfoliation allowing a Raman spectrum from the buried interface to be recorded (Fig. 4). The spectrum is consistent with a high-quality Si clathrate crystal structure. The dominant Raman lines match well with published spectra^33–36 of type II Si clathrate. We also observed second order peaks involving two phonons for the first time which are between 600 and 1000 cm^–1. We were able to see these peaks on multiple Si clathrate films using different Raman spectroscopy systems. The much more dramatic drop in scattering intensity beyond 1000 cm^–1 relative to diamond Si, also shown in the figure, further supports the identification of second order scattering. Comparing the spectra from the buried layer and from the top surface in Fig. 4, it clearly shows that the top of the film is highly disordered while the high crystalline quality clathrate material lies deeper in the film.

As part of this study, Raman spectra were acquired from multiple points on the buried interface and also from the surface of the silicon substrate after the tape pull. It was still possible to find regions on both surfaces that showed evidence of disordered silicon although the clathrate spectrum was dominant. An example Raman spectrum from the substrate after the tape pull is shown in Fig. S5 of the supplementary material. It is consistent with the Si clathrate structure indicating that after exfoliation some of the film still remains on the substrate. The Raman lines, however, are broadened, and there is excess background in the 480 cm^–1 region which could suggest the presence of disorder.

From the results above, the picture that emerges of the as-prepared clathrate films involves a highly defective or disordered top surface of the film with this disorder extending along the cracks or grain boundaries in the film. Na concentration is also high in the vicinity of these disordered regions. High-quality clathrate with low Na content is then found deeper within the film and away from grain surfaces. It's interesting to speculate that the disorder arises in regions with a high Na concentration and/or high rates of Na diffusion through the region, particularly during the second, decomposition step of clathrate formation. During the NaSi decomposition, a large amount of Na vapor escapes from the film through the surface and boundaries. If this resulted in collapsed cage structures or hindered formation of Si-Si bonds, unterminated Si bonds might result. We note that during intercalation of high concentrations of lithium into Si anodes for battery applications, a similar conversion of the silicon into an amorphous phase is reported with repeated cycling.^37–39 While the phase diagram of Li silicides is more complicated than that of NaSi, it is interesting to speculate that something similar may be occurring in our films. Whatever the origin of the disordered Si is, it complicates measurement of intrinsic properties and is an issue for applications. This is particularly true for thinner films where the disordered layers can be a larger fraction of the film composition. We next discuss some growth techniques we have found that reduce the amount of disorder and post growth treatments, in addition to exfoliation, that help isolate the type II clathrate phase.

For films thicker than 20 μm, a heating rate of 5 °C/min (80 min to reach the 400 °C clathrate formation temperature) results in an XRD pattern with both crystalline clathrate and disordered Si phases as discussed above. However, for thinner films of 5–10 μm (⁠$∼40$ min annealing for the first step) using the same temperature ramp, the NaSi film converts into a highly disordered and porous structure (see the cross-sectional SEM image of sample 2 in Fig. S6 of the supplementary material). The XRD pattern of sample 2 is also characteristic of a disordered, amorphous-like phase (Fig. 5). When we increased the ramp rate to 40 °C/min (10 min to reach the desired temperature), the type II phase was dominant in the XRD pattern (sample3 in Fig. 5), and cross-sectional SEM indicated a compact film (Fig. S6 in the supplementary material). However, the fact that we can still see evidence of disordered Si in the XRD pattern of sample 3 indicates that the rapid annealing doesn't eliminate this phase. 40 °C/min was at the limit of our furnace. Future work will be performed to see if an even faster ramp further improves crystallinity. We do not completely understand why increasing the ramp temperature improves crystallinity, but as noted above, the changes in Na concentration, rate of diffusion, and associated activity all have a significant effect on decomposition of NaSi into clathrate especially in the surface layer. The rate at which the sample reaches the decomposition temperature could certainly have an effect on Na activity during decomposition. We find other factors which might affect Na activity also effect crystallinity. For example, minor oxygen contamination of the Na metal during silicide and clathrate formation often results in films with low or no clathrate phase.

We have previously shown that HF/HNO₃ wet etching effectively decreases the Na concentration and amount of type I phase in powder samples of type II Si clathrate.¹² However, this technique has not been applied to clathrate films in part because the structural integrity of previous films was not sufficient to allow the film to withstand the acid etching.⁶ Working with thinner and better adhered films and using the step-by-step washing techniques described in the experimental methods section to remove residual Na, we can routinely use an HF/HNO₃ etch (3 vol. % HF and 10 vol. % HNO₃ in DI water) with films. Similar to powders,¹² the type I phase is undetectable after etching and Na occupancy is largely reduced (Fig. S7 in the supplementary material). Rietveld refinement of the XRD patterns of the Na_xSi₁₃₆ (0 < x < 24) film in Fig. S7 of the supplementary material gives Na contents before and after etching of x = 7.2 and 2.9, respectively. The latter is near the detection limit of XRD. Following the approach applied to powders,¹⁵ we used the integrated intensity of Na spin related signals in EPR to obtain a better estimate of Na content for one of the lowest Na concentration films (sample 6, etched). The Na level is comparable to that of etched powder samples where we found a total Na concentration of 5 × 10¹⁸/cm³ and isolated Na concentration of 3 × 10¹⁷/cm³ taking the clathrate film into the range of a heavily doped semiconductor. A comparison of the EPR spectra and spin densities of the film and powder is given in Fig. S8 of the supplementary material.

The HF/HNO₃ etching process also removes the rough layer at the film surface. This is apparent in both the top view and cross-sectional SEM images in Fig. 6 for the same sample shown in Fig. 2(d) but after etching. EPR spectra at 77 K of etched films [Fig. 6(c)] indicate a decrease in the relative concentration of Si dangling bonds. Four Na hyperfine lines are present at 314, 326, 339, and 352 mT. The line at 339 mT is obscured by the dangling bond and free carrier lines which were discussed above in connection with Fig. 2. These two lines are indicated by arrows A and B, respectively, in the inset. The fact that the ratio of the dangling bond line intensity to the hyperfine line intensity decreased after removal of the rough surface layer by etching (15 min using 3 vol. % HF and 10 vol. % HNO₃ solution) suggests that this disordered layer is highly defective with a high concentration of Si dangling bonds.

Several films were sequentially etched in HF/HNO₃ with Raman spectra acquired after each etch step. At first, all measured spots on the surface showed a disordered Raman spectrum. For etch times approaching 10 min, a transition to a mixture of disordered and clathrate spectra was observed (representative spectra are shown in Fig. 7), presumably because the etch has removed enough disordered material to expose buried clathrate regions. In order to further reduce the disordered phase, we applied higher acid concentrations (5 vol.% HF and 15 vol. % HNO₃ solution) and etched the film longer (50 min). Type II Si clathrate Raman spectra were observed with little evidence for disordered Si (Fig. S9 in the supplementary material). However, the fact that a strong Raman line associated with the Si substrate is also seen suggests that most of the film was etched away. The observation from Raman spectroscopy that disordered Si remains in film for etch times that leave a reasonable amount of the film on the surface is consistent with the XRD results in Fig. S7 of the supplementary material where a clear signature of the broad peaks due to disordered Si is visible in the baseline of the patterns both before and after etching.

Although HF/HNO₃ wet etching reduces the disordered layer at the surface, disordered Si is still present after short etches. Yet long etches remove most of the film which is not ideal for subsequent measurements. We therefore explored dry etching using SF₆.

Reactive ion etching was performed in an SF₆ plasma. The etch rate of the clathrate film was calibrated to be about 3.4 μm/min [Fig. S10(b) in the supplementary material], and the film after 2 min etching has a thickness of about 8 μm. Figure 8(a) shows the XRD patterns of film 8 before and after 2 min dry etching. The gray and orange curves are amorphous Si patterns which have been scaled to fit the backgrounds due to disordered Si in the film patterns. They are used to estimate the intensity of the disordered phase in the film [see Fig. S10(a) in the supplementary material]. The film XRD pattern after etching shows a significant reduction in the amorphous phase relative to type II clathrate indicating that most of the disordered material was removed by the SF₆ plasma. Rietveld refinement also shows that the Na concentration decreases from $x∼11$ to $x∼3$ in Na_xSi₁₃₆. The type I phase was also eliminated by this treatment. These conclusions are confirmed by the Raman spectra in Fig. 8 which show disorder before etching but the desired clathrate spectrum after the etch. It was still possible to find spots that gave a disordered signature, but the majority of the etched surface exhibited a clathrate signature. Gas phase etching appears to be a much more effective way to isolate low Na content, high crystalline quality type II clathrate compared with wet etching.

Absorption spectra for a Si clathrate film on sapphire with a film thickness of about 700 nm are shown in Fig. 9(a). This sample received a HF/HNO₃ etch. The anneal for the NaSi formation step was much longer than would be required to form a 700 nm film (Fig. 1) but a rapid temperature ramp for the clathrate formation step was not used. We assume that the film has sufficient absorption, so no multi-bounce transmission or reflection occurs. To determine absorption, reflection at the top surface and film-sapphire interface are considered. The absorption coefficient, α, is calculated using α = $1dln((1−Rtop)(1−Rsapphire)T)$⁠, where d is the film thickness, R_top is the reflectance from the film surface, R_sapphire is reflectance from a sapphire substrate with the Si layer removed, and T is the measured transmission also referenced to the sapphire substrate.

Unlike the reported absorption data of clathrates in Refs. 6, 16, 19, and 41, no increase in absorption at long wavelength (near 950 nm) is observed. This has been attributed to free carrier absorption resulted from carriers donated by Na. The absence of this absorption tail is consistent with a lower Na content in the present work.

The optical bandgap energy was estimated based on the direct band absorption formula α = $A(hυ−Eg)1/2/hυ$⁠, where α, A, h $υ$⁠, and E_g represent the absorption coefficient, a fitting constant, incident photon energy, and the direct bandgap, respectively. The plot of $(αhυ)2$ vs incident photon energy is shown in Fig. 9(b), where the bandgap was estimated to be about 1.65 eV. From an application standpoint a bandgap much larger than d-Si is interesting. In addition, compared with d-Si, the absorption coefficient of the clathrate film [Fig. 9(a)] is about two orders of magnitude higher suggesting that a film thickness of 2 μm should result in > 98% above bandgap light absorption. While these results are consistent with previous powder^5,8 and film studies¹⁹ of type II Si clathrate reported in the literature, we are concerned that the presence of disordered Si could be affecting our measurements. Amorphous silicon, for example, exhibits an optical gap in the 1.5 eV range and a much larger absorption coefficient than diamond silicon. Raman spectra of the etched surface of the film studied here showed evidence that disordered material was present. We therefore measured SF₆ etched SOS samples where Raman showed most of the disordered material was removed (see the supplementary material Fig. S11). The measurements focused on regions exhibiting clathrate Raman signatures. While the regions were small, pin holes in the film affected results, and we could not easily calibrate the thicknesses of regions studied, the absorption coefficient and bandgap were consistent with Fig. 9. This suggests that the absorption data here is already very close to the clathrate properties. We will continue working on this topic to further reduce the effect of disordered materials.

Sample inhomogeneity also has a significant impact on PL measurements. Spectra obtained from the top surface of films, even after a 15 min HF/HNO₃ etch, varied markedly from one location to the next (see the supplementary material Fig. S12). Generally, an emission band is observed at energies well above the predicted 2.0 eV bandgap of type II Si clathrate with relative intensity that varies significantly across the sample. There is often an emission band near 700 nm and additional emission extending further into the infrared. The above bandgap emission has previously been reported for type II clathrate by our group¹⁶ and others,¹⁹ with speculation that it is associated with defects. Band filling along with formation of an impurity band in heavily doped material is another way of obtaining above bandgap PL. We now believe, however, that other phases present in the sample are responsible for some of the PL we observe. For example, sodium silicate glass which could form from trace oxygen contamination might explain the higher energy emission. The presence of amorphous material could contribute to infrared emission. Hydrogenated amorphous silicon exhibits a broad emission peak near 900 nm. The preparation steps reported here, however, are likely to cause the disordered/amorphous silicon phase in the films to be H free, which generally is associated with low emission efficiency.^42,43

To isolate emission arising from the clathrate phase, a film was prepared, exfoliated, and mounted on the cold finger of the refrigerator with the buried interface exposed allowing the buried surface to be studied. Raman spectra were measured using the 600 g/mm grating of the PL setup to identify regions with a strong type II clathrate spectrum. PL measurements were then made using the 150 g/mm grating to allow a broad spectral region to be recorded. While the spectral resolution was low with this grating, the dominant clathrate Raman lines near 460 cm^–1 could clearly be detected in the high energy tail of the PL measurements as shown in the inset to Fig. 10(a). A diamond silicon Raman spectrum acquired with the same parameters is shown for comparison. This allowed us to conclude the volume under study was composed of type II Si clathrate. The resulting temperature dependent PL spectra [Fig. 10(a)] exhibit a broad (∼230 nm FWHM) peak near 700 nm. The same spectral shape was observed at all spots that showed a clathrate Raman signal, although the intensity varied from region to region. While spectra are only presented to T = 250 K, emission at room temperature was easily observed. In addition, the energy of the peak, 1.7 eV, is consistent with the predicted (2.0 eV) and measured (1.7 eV) bandgaps. The high energy tail of the emission does extend above 2.0 eV. This could be tied to the width of the emission band, which is not well understood. PL emission can be broadened by intrinsic effects, such as strong phonon side bands or the spatial separation of donors and acceptors in donor-acceptor pair luminescence. We suspect, however, that additional inhomogeneity in the material is responsible. For instance, we have previously reported that the Na distribution in powder clathrate samples appears to be inhomogeneous.¹⁵ Localized regions with higher and lower Na content could lead to different band filling effects and shifts in emission energy.

The temperature dependence of the integrated intensity of the PL peak is shown in the Arrhenius plot in Fig. 10(b). Here, the decrease in the integrated intensity at temperature T compared with the integrated intensity at 15 K ((I_15K−I_T)/I_T) is plotted as a function of inverse temperature. In a simple model where the PL intensity arises from recombination of photoexcited electrons captured on Na donors, and the loss of PL intensity is due to thermal excitation out of the donor, we would expect the straight line that is observed in the Arrhenius plot. The slope gives an activation energy of 19 meV. This is close to the Na donor binding energy previously inferred from the temperature dependence of the Na related hyperfine lines in EPR measurements.³⁰ 

It is interesting to compare the present work with previous studies on clathrate films from our group and others. Ohashi et al. and Kume et al. reported the synthesis of Si clathrate films using the same two-step process.^17,18,33 Although not discussed in the publications, two broad XRD peaks around 27° and 50° 2θ indicating disordered Si were visible in the NaSi XRD pattern in Ref. 33 formed after the first step and may be present after clathrate formation in Ref. 17 although other contributions to the XRD baseline were also present. This group also reported the observation of an amorphous phase in TEM measurments.⁴⁴ These observations are consistent with our identification of a disordered phase present in the Si clathrate films. While the growth kinetics and optimal synthesis conditions for Si and Ge type II clathrate films are likely to differ, published studies of Ge clathrate thin film formation on sapphire substrates by Kumar et al. using a similar two-step process complement the results presented here.^41,45,46 XRD patterns and Raman spectra in Refs. 41 and 46 showed evidence of an amorphous Ge phase. Reference 45 showed that by modifying the two-step process using evaporated Na and IR lamp annealing the amount of the amorphous phase was minimized. Since the IR lamp should allow a faster temperature ramp rate than a conventional furnace, this could support our finding that a rapid temperature ramp during the decomposition from silicide to clathrate leads to a less disordered phase.

This work presents a systematic pathway to improved Si clathrate thin films by investigating key factors during two-step film synthesis that lead to inhomogeneities and by identifying post growth processing that can isolate the desired type II clathrate phase. As-grown clathrate films are found to contain a significant fraction of disordered silicon, as well as other minority phases including type I clathrate and diamond silicon. The disordered silicon is localized at the surface of the films, but can be observed penetrating further into the films, presumably along cracks and grain boundaries. Na content which is also found to be high at the surface and along grain edges may be connected to the presence of disordered Si. In particular, we speculate that NaSi does not decompose into the clathrate structure in regions with high Na concentrations and/or high rates of Na diffusion. These effects are particularly important in thinner films where disordered or amorphous-like Si can become the dominant phase.

Several growth and post growth treatments have been demonstrated for the first time to improve film quality and crystallinity including film exfoliation, rapid thermal annealing, and wet and dry etching. Exfoliation to expose buried high-quality clathrate and reactive ion etching, which seems to selectively remove the disordered phase while also decreasing the Na content in the films, were shown to be promising ways of isolating high-quality material for study.

Using these growth and post growth techniques, absorption and photoluminescence measurements were performed. While absorption is probably still influenced by residual disordered material, a bandgap near 1.65 eV consistent with prior experimental work was observed. An absorption coefficient about two orders of magnitude higher than d-Si was obtained showing a 2 μm film should be sufficient for full absorption of above bandgap light in photovoltaic devices. Room temperature photoluminescence from film regions proven to be composed of type II Si clathrate in an energy range consistent with the bandgap was observed for the first time opening up the possibility of efficient light emission in the near infrared from a silicon-based material. An analysis of the thermally activated PL emission confirms that Na in the type II Si clathrate films can be a shallow donor. These findings suggest that this alternative crystalline form of Si has the potential to address key limitations that arise from the indirect bandgap of d-Si limiting its use in optical applications.

See the supplementary material for (XRD pattern of NaSi film shown in Fig. S1. Fitting of the Si clathrate film thickness at different first step annealing times shown in Fig. S2. XRD pattern of Si clathrate film annealed for 20 h in the first step shown in Fig. S3. Raman spectra of film top surface with increasing laser power shown in Fig. S4. Raman spectrum of the remaining film on the Si substrate after pulling the surface off shown in Fig. S5. Cross-sectional SEM images of two synthesized films with different annealing rates in the second step shown in Fig. S6. Normalized XRD patterns of film before and after wet etch shown in Fig. S7. Comparison of the EPR spectra for the Si clathrate film and powder at 77 K shown in Fig. S8. Raman spectrum of the film top surface after 50 min wet etch in Fig. S9. Ratio of the integrated XRD intensity of amorphous Si to the integrated XRD intensity of 17° 2θ peak at different etching times, and amount of clathrate material etched off at different etch times of RIE shown in Fig. S10. Optical absorption coefficient of the Si clathrate film on sapphire after dry etch shown in Fig. S11. PL spectra of sample top surface at T = 15 K as shown in Fig. S12.]

This work was supported by the National Science Foundation, Grant No. 1810463. This material makes use of the TOF-SIMS system at the Colorado School of Mines, which was supported by the National Science Foundation under Grant No. 1726898. The authors thank N. Hadacek for his work on the SEM measurement and A. R. Meyer for her help on the EPR measurement. The authors thank D. L. Williamson for his guidance on XRD and N. Fennell for his help with RIE.

The authors have no conflicts to disclose.

The data that support the findings of this study are available from the corresponding author upon reasonable request.