Efficient surface passivation of germanium nanostructures with 1% reflectance

Germanium (Ge) is a vital element for applications that operate in near-infrared wavelengths. Recent progress in developing nanostructured Ge surfaces has resulted in >99% absorption in a wide wavelength range (300–1700 nm), promising unprecedented performance for optoelectronic devices. However, excellent optics alone is not enough for most of the devices (e.g. PIN photodiodes and solar cells) but efficient surface passivation is also essential. In this work, we tackle this challenge by applying extensive surface and interface characterization including transmission electron microscopy and x-ray photoelectron spectroscopy, which reveals the limiting factors for surface recombination velocity (SRV) of the nanostructures. With the help of the obtained results, we develop a surface passivation scheme consisting of atomic-layer-deposited aluminum oxide and sequential chemical treatment. We achieve SRV as low as 30 cm s−1 combined with ∼1% reflectance all the way from ultraviolet to NIR. Finally, we discuss the impact of the achieved results on the performance of Ge-based optoelectronic applications, such as photodetectors and thermophotovoltaic cells.


Introduction
Germanium (Ge) is a promising candidate in several technological areas, due to the attractiveness governed by its intrinsic properties [1][2][3][4][5][6][7]. In particular, Ge has a relatively small bandgap, making it an ideal candidate for optoelectronic devices operating in the near-infrared (NIR) region, such as photodiodes and thermophotovoltaic (TPV) devices [5][6][7][8][9]. Nevertheless, planar Ge surfaces suffer from high reflectance (∼40% at 1550 nm) resulting in significant optical loss limiting the performance of Ge-based optoelectronic devices. To address the optical losses, anti-reflective coatings have been the mainstream technology in the state-of-the-art Ge devices, although the coating can only be optimized for a limited wavelength range. In recent years, significant research efforts have been devoted to further reduce the surface reflectance of Ge. One approach is to use dielectric metasurfaces that aim to control the local polarizations, phases, and field amplitudes of light, resulting in enhanced absorption especially in thin layers [10,11]. Another alternative is to form a graded-index medium between air and the Ge substrate by nanostructures etched on Ge surface (i.e. black Ge) either via metal assisted chemical etching [7,8,12,13] or plasma etching (i.e. reactive-ion etching (RIE) or inductively-coupled plasma etching (ICP-RIE). Out of all these methods, ICP-RIE has been the most promising achieving as low as sub-1% surface reflectance all the way from ultraviolet up to NIR (∼1700 nm) [9,14,15]. The excellent optics provided by the nanostructures has already been utilized in some devices such as metal-semiconductor-metal and Schottky photodiodes [7,9]. However, most optoelectronic devices (such as PIN-photodiodes and solar cells) are minority carrier based, thus, besides optical properties, excellent electrical properties such as minority carrier lifetime are also of vital importance. In this regard, surface passivation-a process that aims to minimize surface recombination-plays a critical role for nanostructured surfaces. Quite surprisingly, while successful surface passivation of nanostructured silicon has been reported in literature [16][17][18][19], similar studies have not been reported for Ge. In other words, there are only some interface defect density results for microscale structures [8] but no direct data on surface passivation of nanostructured Ge. However, for planar Ge surfaces, there are extensive studies. A natural material choice for the surface passivation would be germanium oxides, but they are known to be highly unstable and watersoluble preventing their use in practical applications [20]. Recently, aluminum oxide (Al 2 O 3 ) deposited via atomic layer deposition (ALD) has become the forefront in surface passivation [21][22][23][24]. It has been demonstrated that ALD Al 2 O 3 forms negative fixed charge at the Al 2 O 3 -Ge interface resulting in excellent surface passivation through field-effect [21,24]. Prior to ALD either HCl or HF alone is often used instead of RCA cleaning, which has been shown to be too vigorous for Ge [25][26][27][28]. In addition to cleaning the surface, another purpose of this treatment is to remove unstable native germanium oxides that easily form on the surface when exposed to air [29][30][31]. These results serve as a good starting point for nanoscale structures as well due to the conformality of ALD. However, this is not enough as the nanostructures bring several additional challenges. Firstly, the surface area is significantly enhanced as a consequence of needle-like nanostructures inherently increasing the number of dangling bonds at the surface. Secondly, ion-bombardment as a part of dry etching raises the concern of creating crystallographic damage on the surface, which could result in a so-called 'dead-layer' with extensive amount of recombination sites near the surface [16,[32][33][34]. Thirdly, the use of processing gases during dry etching results in chemical residuals on the nanotextured surface [35]. To fully utilize the potential of black Ge in devices without tradeoffs in electrical and optical properties, these challenges need to be addressed.
In this work, we develop an efficient surface passivation scheme for black Ge based on ALD Al 2 O 3 . More specifically, we investigate and address the three aspects that are concerned when integrating black Ge into optoelectronic devices: (1) surface area enhancement, (2) crystallographic damage caused by ion-bombardment and (3) chemical residuals remaining on nanotextured surfaces. The corresponding electrical properties are evaluated using quasi-steady state photoconductance decay. The crystallinity of the Ge after subjecting to ICP-RIE is investigated by transmission electron microscopy (TEM) and the chemical residuals on the Ge surface at different stages are monitored by x-ray photoelectron spectroscopy (XPS). Finally, the impact of the obtained results on optoelectronic devices is discussed.

Determining limiting factors for surface passivation
Increased surface area Figure 1 shows the surface morphology for the two sets of nanotexture used in this study, fabricated by using ICP-RIE with an ICP power of 1000 W. First, a 700 nm deep nanotexture ( figure 1(a)) was formed by using capacitively-coupled plasma (CCP) power of 4 W, which results in black Ge with a broadband surface reflectance of less than 1% [15]. The inset shows an optical image of the full 4 inch wafer to demonstrate the black appearance by naked eye. Second, a shallow nanotexture of ∼200 nm height ( figure 1(b)) was formed by using CCP power of 2 W, which results in completely different morphology i.e. only moderately increases the surface area compared to a planar surface. These two morphologies allow us to study the impact of the increased surface area on the passivation.
To determine the starting level of surface passivation of the Ge nanostructures, for the first experiment we chose the passivation scheme that has provided the best performance on planar Ge [36]. After subjecting to a HCl cleaning for 5 min, we deposited a 20 nm thick Al 2 O 3 conformally on the nanotexture and activated the surface passivation by performing a 400°C post-deposition anneal. As expected, for a planar germanium reference sample, ALD Al 2 O 3 provided excellent surface passivation resulting in a lifetime above 1 ms, as shown in figure 2 [21,24]. For the deep nanostructures, the same passivation scheme leads to substantially lower lifetime around 200 μs (blue markers). Surprisingly, for shallow nanostructures (green markers) there is no significant improvement despite the drastic difference between the surface morphology shown by the SEM images. Therefore, it is clear that surface area enhancement, due to the formation of nanotextures, is not the primary factor limiting the lifetime for the nanotextured samples.

Crystallographic damage
There has been a pervasive concern on surface damage induced by RIE [7,8], hence, to elucidate the role of ion-bombardment on the poor effective lifetime measured on nanotextured samples, the crystallinity of the nanotextured Ge was examined in detail at the Al 2 O 3 -Ge interface by using TEM. As shown in figure 3, Ge lattice at the proximity of the Al 2 O 3 -Ge interface remains crystalline as no amorphous-like structure is observed, akin to what has been demonstrated on black silicon fabricated by the same method [17]. Consequently, the poor effective lifetime shown above is not caused by ion-bombardment during the etching process. Since no crystallographic damage was observed for the deep nanotexture, the same conclusion can be drawn from the shallow structures, as the degree of ion-bombardment correlates to the magnitude of CCP power, which was reduced from 4 to 2 W [16]. In summary, this result clearly demonstrates that it is possible to fabricate Ge nanotexture by ICP-RIE that is free of crystallographic damage.

Chemical residues
Since neither the enhanced surface area nor the crystal damage explains the poor lifetime of the black Ge, next we studied whether some chemical residuals formed during dry etching could be the root cause. Figures 4(a) and (b) shows the XPS spectra of the S 2p and F 1s bonding states measured at the surface of the nanotexture directly after the ICP-RIE process. A significant amount of sulfur and fluoride elements were detected on both deep and shallow nanotextures. (Note that there is a much higher amount of residuals detected on the shallow nanotexture, which could be due to the use of higher process gas flows and a lower CCP power during the ICP-RIE etching process.) As shown in figure 4(c), after 5 min of HCl cleaning, the sulfur elements with a higher binding energy were removed, while the ones with a lower binding energy remain on the surface. On the other hand, all the fluoride elements were removed for both sets of nanotexture ( figure 4(d)). The removed elements are likely associated with oxide compounds, as HCl has shown to be efficient in removing oxide from germanium surface [31]. To conclude, HCl is not able to remove all the chemical residues and most likely the remaining sulfur-related residuals cause the poor surface passivation. We speculate the following mechanisms for this: (i) they could act as recombination-sites at the Al 2 O 3 -Ge interface, which increases the interface defect density, (ii) they could also hinder the subsequent activation of the Al 2 O 3 layer during the postdeposition annealing, which reduces the fixed charge density at the interface. Nevertheless, the exact mechanism is difficult to identify, due to the ambiguous analysis of C-V measurements on nanotextured surface.

Removal of chemical residues
As the XPS results demonstrated above, HCl alone leaves some sulfur-related residuals on the surface of Ge nanotexture. In order to remove them prior to thin film deposition, we applied a chemical cleaning process composed of H 2 O 2 and HCl sequentially. Since germanium dioxide is water soluble, the Ge atoms at the surface, when subjecting to H 2 O 2 solution, are readily oxidized and then dissolved in the solution [37]. Therefore, we speculate that the sulfur-related residuals could be removed along with the Ge atoms by performing an additional H 2 O 2 dip. The subsequent HCl dip, on the other hand, efficiently removes any sub-oxide remaining on the Ge nanotexture and results in Cl-termination of the surface, which protects the Ge nanotexture from being oxidized in the air prior to ALD passivation [31,37]. As shown in figure 5, according to our hypothesis, the sulfurrelated residuals detected on the nanotexture after HCl cleaning alone were found to be removed completely after an addition of a 10 s H 2 O 2 dip. As a result, the level of surface contamination for the Ge nanotexture after 10 s of H 2 O 2 cleaning should be similar to the planar Ge surface without any ICP-RIE etching.      6(b)), as the features are largely intact, with only some rounding observed at the tip of the nanotexture. As another significant observation, it was found that the deep nanotexture with an addition of 10 s H 2 O 2 treatment (orange star) resulted in a noticeably higher lifetime (400 μs versus 240 μs) when compared to the shallow nanotexture treated with HCl alone (green triangle) despite having a substantially higher surface area and a lower surface reflectance. Since the attribute of crystal damage has been excluded by the TEM image shown above, it is likely that the difference in surface contamination prior to ALD deposition, due to the addition of H 2 O 2 cleaning, contributes to the difference in lifetime measured for both sets of nanotextures. Removing all the chemical residuals is, therefore, demonstrated to be a crucial factor in achieving good surface passivation for Ge nanotexture created by ICP-RIE process.
In the case of shallow nanostructures the addition of H 2 O 2 cleaning was found to affect significantly the morphology by removing most of the features resulting in a planarlike surface. Consequently, the lifetime was improved from sub-300 μs to over 900 μs. The obtained value is clearly higher than in the deep nanostructure counterpart. This difference is likely due to the difference in the surface area. Consequently, it is suggested that the lifetime of the deep nanotexture after 10 s of H 2 O 2 cleaning is limited by the significantly larger surface area when compared to the planarlike surface.

Finding balance between reflectance and surface recombination
From the perspective of a device, it is crucial to consider both electrical and optical properties of black Ge. The successful removal of chemical residuals on the Ge nanotexture resulted in significant improvement in lifetime. However, the lifetime was still found out to be much lower than the corresponding planar sample. This means that after chemical residue removal the lifetime of black Ge is limited by the surface area -a factor which also contri-ance of black Ge. Hence, there is a clear need for a tradeoff between the reflectance and the lifetime. We aimed to minimize this tradeoff by modulating the surface morphology with the application of an extended H 2 O 2 treatment, i.e. by reducing the surface area as much as possible without meaningfully increasing the surface reflectance. Figure 7 shows the impact of an extended H 2 O 2 treatment on the morphology of a deep nanotexture. As mentioned earlier, a 700 nm deep nanotexture is formed on Ge surface by ICP-RIE, whereas some rounding at the tip (due to the combination of oxidation and dissolution of Ge atoms) is observed after subjecting to 10 s H 2 O 2 treatment. As the duration of H 2 O 2 treatment increases, the modulation of the surface morphology becomes increasingly significant, leading to pyramidal structures with a ∼54°facet angle after 120 s. Interestingly, the facet angle for the nanotexture observed in figure 7(d) coincides with the angle difference between the {100} and {111} for a diamond cubic crystal structure, which is commonly observed in silicon with alkaline chemical texturing [38]. It has also been observed in germanium with inverted pyramidal surface fabricated by the combination of lithography and metal-assisted chemical etching [7,39].
In order to analyze the tradeoff between electrical and optical properties after different H 2 O 2 cleaning durations, figure 8 summarizes both the surface recombination velocity (SRV) and the front surface reflectance at 1550 nm for all the samples. The deep nanotexture reduces the surface reflectance compared to the planar surface from over 35% to sub-1%. The surface reflectance of the deep nanotexture does not meaningfully increase by the H 2 O 2 treatment until 30 s is reached, beyond which it increases significantly to over 20% after 120 s. This significant increase in surface reflectance is expected, as the surface morphology is drastically modified after 120 s of H 2 O 2 treatment. On the other hand, the formation of the deep nanotexture substantially increases the SRV as compared to the planar surface from 6 cm s −1 to over 100 cm s −1 . After 10 s H 2 O 2 treatment, the SRV reduces by half, reaching 40 cm s −1 . Further increase in duration leads to further reduction in SRV, with 30 cm s −1 and 10 cm s −1 measured after 30 s and 120 s, respectively. Consequently, a very good combination of both optical and electrical quality is reached with 30 s H 2 O 2 cleaning resulting in ∼1% reflectance and SRV of ∼30 cm s −1 . The exact optimal etching time depends on the desired properties of the device.

Discussion
The above results demonstrate successful electrical passivation of black Ge nanostructures without increasing the surface reflectance. This was achieved by combining H 2 O 2 cleaning to remove chemical residuals originating from ICP-RIE with ALD Al 2 O 3 surface passivation, which is industrially feasible and should be relatively straightforward to integrate into device fabrication. The achieved excellent opto-electrical properties have tremendous potential in opening new opportunities for further advancing the performance of Ge-based  optoelectronic devices and/or fully utilizing the already existing advantages (e.g. quantum confinement, photo gating, negative differential resistance) demonstrated by nanostructured Ge devices [40][41][42][43]. For instance, in NIR photodiodes the method should enable much higher photoresponsivity around the two main wavelengths used in fiber optics (1300 and 1550 nm). In addition to NIR region, the results show a high promise also at lower wavelengths all the way to UV (300 nm). Recently, in silicon photodiodes, carrier multiplication at short wavelengths has been successfully harnessed for extremely high EQE [36]. Since Ge has much lower bandgap (0.66 eV at 300 K), harnessing carrier multiplication should be possible in much wider wavelength range possibly leading to above unity EQE up to ∼600 nm.
Another potential application where black Ge with low surface recombination should bring significant benefit is a TPV cell. It could be directly incorporated into the interdigitated-back-contact (IBC) cell architecture-a TPV cell design with higher theoretical conversion efficiency (up to 25%) [44] than conventional state-of-the-art Ge-based TPV devices (16.5%) [45,46]-creating an avenue for achieving near-unity EQE, akin to what has been demonstrated in black Si IBC cell for different wavelength spectrum, whereas the EQE of current standalone Ge TPV devices fall between 60% and 80% [6,47,48].

Conclusions
This paper demonstrates, for the first time, efficient surface passivation of nanostructured Ge surfaces fabricated by ICP-RIE. First, we defined the limiting factors preventing effective surface passivation by conventional ALD Al 2 O 3 . While TEM analysis ruled out the presence of crystallographic damage, XPS confirmed that the root cause lies in sulfur-related chemical residues that were remaining on the surface after HCl cleaning. By adding a short H 2 O 2 dip, we were able to remove all the chemical residuals on the nanostructured surfaces, resulting in a two-fold increase in effective lifetime. After chemical residue removal, the next factor limiting the lifetime was found to be the increased surface area. To find a balance between electrical and optical properties, the surface area was reduced by extending the duration of the H 2 O 2 dip. Consequently, the SRV of nanostructured Ge further reduced from ∼100 to ∼30 cm s −1 without impact on the front surface reflectance. These results enable nanostructured Ge to be a key building-block for further advancing the performance of Ge-based optoelectronics devices, such as photodetectors and TPV cells.

Experimental section
Sample fabrication. A batch of 187 μm thick, double-sided polished 4 inch n-type {100} Czochralski (CZ) germanium wafers with a 18-25 Ω cm base resistivity were used in these experiments. The Ge nanotexture was fabricated by using cryogenic inductively-coupled plasma reactive-ion etching (ICP-RIE, Plasmalab System 100, Oxford Instruments) at −120°C for 15 min using SF 6 and O 2 as the processing gases. Whilst the power of the ICP source was kept at 1000 W for both processes, a CCP power of 4 W and 2 W was used to form a deep and a shallow nanotexture, respectively. During both deep and shallow nanotexture etching, the DC bias was measured to be 0 V. For the deep nanotexture sample, the gas flows for SF 6 and O 2 were 17 and 22.8 sccm, respectively. For the shallow nanotexture sample, due to the significant reduction in etch-rate, the gas flows were increased to 21.5 and 28.4 sccm for SF 6 and O 2 , respectively. The chamber pressure was kept at 10 mTorr for all samples.
Two chemical cleaning processes were investigated in this study: (1) a HCl dip and (2) a sequential chemical treatment. In the HCl cleaning process, the samples were cleaned by dipping them into HCl (31.6% v/v) for 5 min without a subsequent DIW rinsing (to avoid oxidation) prior to ALD deposition. For the sequential cleaning process, the samples were cleaned by dipping them into different chemical solutions in the following order: (1) 3% v/v of H 2 O 2 for 2 min, (2) 31.6% v/v of HCl for 5 min, (3) 3% v/v of H 2 O 2 for varied duration and (4) 31.6% v/v of HCl for 5 min By varying the duration of step (3), the amount of Ge atoms being removed by this process was modulated. DIW rinsing was applied between each step, but not after step (4) was completed. An Al 2 O 3 layer (n = 1.65 at 633 nm, thickness = 22.5 nm) was subsequently deposited by thermal ALD at 200°C. The number of cycles was 200 and H 2 O and TMA were used as precursors. Finally, the samples were annealed at 400°C for 30 min under nitrogen ambient to activate the passivation.
Sample characterization. The injection-dependent effective minority carrier lifetime was measured by using QSS-μPCD technique in a Semilab PV2000A characterization tool. The measured lifetime was then converted to SRV, a useful metric in examining the quality of surface passivation, by the where W (cm) and Δn (cm −3 ) represent the wafer thickness and excess carrier density, respectively. The above calculation assumes that the bulk recombination is negligible, and hence the calculated S eff,max represents the maximum value. For single-sided RIE-etched sample, the SRV of the textured surface was evaluated by the following equation where S eff,planar represents the SRV of the planar surface extracted from a planar reference sample using equation (1). The front surface reflectance between 300 nm and 1700 nm was measured by using an integrating sphere in Agilent Cary 5000 UV-vis-NIR spectrophotometer. The surface morphology of the nanotextures was inspected by cross-sectional scanning electron microscopy using Zeiss Supra 40. The crystallinity of the germanium at the Al 2 O 3 -Ge interface was inspected by TEM using JEOL JEM-2200FS. The TEM lamella was prepared by lift-out focused-ion beam (JEOL JIB-4700F) with sputtered platinum acting as a protective layer. The surface chemistry of the nanotexture prior to ALD was analyzed by XPS using Thermo Scientific (tm) Nexsa instrument with monochromated Al Kα source.