Explorative studies of novel silicon surface passivation materials : Considerations and lessons learned

Despite the existence of several highly effective and well-characterized passivating materials for crystalline silicon surfaces, the topic of surface passivation and the investigation of new passivating materials remain of considerable interest for silicon photovoltaics research. However, the question of whether and under what circumstances a particular material will provide effective surface passivation remains difficult to answer. In this work, we provide an overview of recent insights relating to this question, drawing from our own work on novel passivation materials including MoOx, Nb2O5, TiOx, ZnO, and POx, and illustrated with experimental results. Factors that strongly influence the passivation performance include the use of pre-grown interfacial oxides, the film thickness, the annealing conditions, and the presence of capping layers. The impact of these factors on the surface passivation can vary widely from material to material. Therefore, all of these factors should be taken into account when investigating potential new surface passivation materials.


Introduction
Effective surface passivation is critical to realizing high-efficiency crystalline silicon (c-Si) solar cells and is becoming ever more important as wafer thicknesses are reduced and bulk Si quality improves. At the same time, new demands on the functionality of passivation layers have emerged, such as the ability to form passivating contact structures that in addition to passivation, enable the selective extraction of either positive or negative charges from the Si bulk [1,2]. These demands have stimulated significant research efforts aimed at evaluating the passivation properties of novel materials [1][2][3][4][5][6][7][8][9]. As a result, in the last decade, the list of materials known to be capable of effectively passivating c-Si surfaces has expanded significantly. The well-established Sibased passivation layers (SiO 2 [10], SiN x [11], a-Si:H [12], and more recently proven Al 2 O 3 [13][14][15][16]) have been joined by a range of new materials including HfO 2 [15,17,18], TiO x [19][20][21][22][23], Ta 2 O 5 [24] and Ga 2 O 3 [25][26][27], and new passivation materials continue to be reported, such as ZnO [28], Nb 2 O 5 [29,84], PO x [30,85], and ZrO x [31]. Table 1 gives a summary of state-of-the-art passivation results for these materials and a few other notable examples. Note that some of these passivation materials can also serve as passivating contacts, as apart from preventing minority carrier recombination at the c-Si surface, they also yield a sufficiently low contact resistance to c-Si for majority carriers. In Fig. 1, transmission electron microscopy (TEM) images for a selection of these passivating materials are shown.
The mechanism of surface passivation is generally understood to involve a reduction in the number of interface states ("chemical" passivation) and/or suppression of the concentration of either electrons or holes at the semiconductor surface (historically referred to as "fieldeffect" passivation) (in general, effective passivation almost always involves a combination of these two mechanisms). However, the reasons for when and why a particular material will provide passivation are still not entirely clear. Some materials that were long believed not to provide effective surface passivation have now been shown to do so under the right conditions. For instance, TiO x was replaced by SiN x as the preferred anti-reflection coating (ARC) for c-Si solar cells in the late 1990s due to the more effective passivation provided by SiN x , but it has recently been shown that TiO x can in fact also provide excellent passivation [19][20][21][22][23]. On the other hand, other materials that are potentially relevant to c-Si photovoltaics thus far do not seem to provide passivation, regardless of their preparation method and post-deposition treatment. However, the growing number of studies of different In our group, we have investigated a large number of materials prepared by atomic layer deposition (ALD) for surface passivation of c-Si. ALD provides distinct advantages for surface passivation studies, such as large-area uniformity, precise thickness control, and damagefree ('soft') deposition [7]. Materials that have recently been examined include Nb 2 O 5 , PO x , ZnO, TiO x , and MoO x . In this work, we will highlight several trends and exemplary cases taken from these and other datasets which provide insight into the factors that are important to surface passivation, with a particular emphasis on results that deviate from or go beyond conventional wisdom. While such examples do not yet permit us to outline a step-by-step procedure for reliably achieving surface passivation with new materials, they do highlight some important factors that ought to be taken into account when investigating novel materials for surface passivation properties. In particular we will highlight the role of pre-grown interfacial oxides (prepared by e.g. a standard Radio Corporation of America (RCA) clean or a low-temperature oxidation), the existence of different optima for film thickness, variation in optimum annealing temperature between different materials, and the potential role of capping layers in providing hydrogen for interface passivation. Since there are already a number of recent works which review passivating materials and/or passivating contacts [1][2][3][4][5][6][7][8][9], this work is not intended as a fully exhaustive review paper, but is instead primarily focused on providing the latest insights taken from a variety of relatively new passivation materials.

Surface preparation and the role of pre-grown interfacial oxides
The conventional silicon surface preparation procedure prior to surface passivation involves one or several wet-chemical cleaning steps (e.g. a Radio Corporation of America (RCA) standard clean [49]) which typically results in the growth of a thin (1-2 nm) surface oxide. Immediately before the deposition of a passivating film, this oxide is removed by etching in dilute hydrofluoric acid (HF) to leave a hydrophobic, hydrogen-terminated surface. The latter is generally considered the ideal starting surface for passivation, as such low-temperature wet chemical oxides are generally regarded as being of low-qualityusually they provide negligible passivation on their ownand their removal enables the formation of a new high-quality interface. This is well-established for example for thermal SiO 2 [50], PECVD SiN x [51] and PECVD a-Si:H [34]. Note that, at least in the case of oxide materials, the formation of this new interface generally involves the growth of a new interfacial SiO x layer either during deposition or upon subsequent annealing, (as visible in the TEM images shown in Fig. 1 of Al 2 O 3 , TiO x , and Nb 2 O 5 deposited on HF-treated Si), where the thickness and other properties of this layer will depend on the specific processing conditions and interfacial chemistry.
Contrary to this picture, we have found that in some cases retaining the oxide grown during surface cleaning, or intentionally growing a thin chemical oxide, can be beneficial for surface passivation by metal Table 1 Selection of state-of-the-art passivation results reported for various passivation materials on n-type substrates with a bulk resistivity in the range of 1-7 Ω cm (unless otherwise specified). S eff,max values are calculated from the original lifetime data assuming an infinite bulk lifetime. In general we have selected the references with the lowest reported S eff,max in this resistivity range, giving preference to reports on more highly doped substrates in case of similar values. Representative values for the bandgap are given (additional references for these bandgap values have been included in cases where these are not mentioned in the reference for S eff,max ). The typical polarity (n or p) of the induced silicon space-charge region is specified. n + or p + indicate particularly strong band bending in the silicon, while p/n indicates that band bending is tunable through doping. We also specify the passivation layer thickness and whether a pre-grown SiO x interlayer (either thermally or chemically grown) was present. Materials that are capable of forming passivating contacts for electron or holes are indicated. oxides. In fact, using a starting surface with an existing thin oxide layer can sometimes even result in passivation in cases where the same procedure applied to a HF-treated surface would not. For instance, MoO x , which is being explored as hole-selective contact [52,56,57], is a material which typically yields very poor surface passivation when directly applied to the Si surface [54][55][56][57]. For this reason, it is either used as partial rear contact [56] or combined with a dedicated amorphous silicon passivation layer [52,53] as shown in Fig. 1(c). However, we recently found that MoO x prepared by ALD [42,45] can yield surface passivation when it is deposited on an RCA oxide and subsequently subjected to a forming gas anneal (FGA) treatment. This is shown in Fig. 2. These results suggest that the interfacial oxide between MoO x and HF-treated siliconthat naturally forms during processingis of poor quality and that this interfacial oxide cannot be made to passivate by an FGA treatment. Conversely, the well-defined RCA oxide can be used to achieve passivation by MoO x , with the best results being obtained after a one hour FGA treatment at 300°C. The beneficial effects of an RCA oxide are also observed for ALD ZnO:B/Al 2 O 3 stacks. Such stacks yield only moderate levels of passivation on HF-treated Si surfaces, whereas excellent passivation is achieved on an RCA oxide (see Fig. 1(f)). Therefore, also in the case of ALD ZnO it seems that the oxide that forms during deposition on HFtreated Si is not of high quality. Conversely, for ALD Nb 2 O 5 good passivation is achieved on both HF-treated Si and Si with an RCA SiO x layer. This shows that for this process a SiO x layer is formed during deposition and post-deposition annealing that enables good surface passivation [29]. This interfacial SiO x layer is also visible in Fig. 1(e).  [13]. (b) p-type poly-Si on SiO x prepared by low-temperature oxidation [39]. (c) ALD MoO x on a-Si:H prepared by chemical vapor deposition [40] and capped with ALD In 2 O 3 :H [41], after annealing at 200°C in N 2 [42]. (d) ALD TiO x prepared at 200°C on HF-treated Si (e) ALD Nb 2 O 5 prepared at 150°C on HF-treated Si and subjected to a 300°C FGA [29] (f) ALD boron-doped ZnO prepared at 200°C on c-Si with an RCA oxide and subjected to a 450°C FGA. The use of a chemical oxide can also strongly alter the passivation behavior of materials that do provide passivation to HF-treated Si. This is clearly illustrated by the case of ALD Nb 2 O 5 [29]. On HF-treated Si, the surface passivation quality improves when using thicker Nb 2 O 5 films and lower deposition temperatures, as shown in Fig. 3. However, the use of an RCA oxide not only yields a better passivation quality, but in addition it renders the passivation level much less dependent on the experimental conditions. Strikingly, this even enables excellent passivation of the surface using only 1 nm of ALD Nb 2 O 5 . The sensitivity of the passivation quality on HF-treated Si surfaces to the experimental conditions is thought to originate from differences in the formation of the interfacial oxide during deposition [29]. It is known that SiO x interfaces formed by different chemical processes [58] and by the same process at different temperatures (e.g. thermal oxidation [59]) can exhibit significant differences in interface state and fixed charge densities, which determine passivation quality. Conversely, the fact that the interfacial oxide is formed prior to Nb 2 O 5 deposition in the case of RCAtreated surfaces renders such surfaces much less sensitive to the exact process conditions.
The nature of the pre-grown oxide is also important. Ultrathin oxides may be prepared by a wide variety of methods, including growth in air (native), wet-chemical (e.g. nitric acid oxidation (NAOS), RCA, boiling water), UV/ozone, O 2 plasma, and thermal oxidation, and these can yield significantly different levels of surface passivation [60][61][62][63][64][65][66][67]. An example of this can be seen from the comparison of lifetimes for TiO x -passivated samples with RCA and NAOS oxides in Fig. 4, where the RCA oxide results in significantly better passivation than the NAOS. Note that, based on our own data and that of others, it does not so far appear possible to establish a simple generally applicable correlation between the readily observable macroscopic properties (thickness, stoichiometry/oxidation state) of a given SiO x layer (whether pregrown or formed during processing) and its suitability for passivation, so that this needs to be tested directly in each case.
Note that beneficial effects from pre-grown interfacial oxides have been reported before. For example, Mihailetchi et al. reported that the use of a 1.5 nm oxide grown by NAOS enabled effective passivation of p + Si surfaces by PECVD SiN x [68]. Ultrathin oxides are also routinely employed as interlayers in polycrystalline silicon (poly-Si) passivating contacts [69][70][71][72].
While the use of a pre-grown oxide can enable passivation in some cases, in other cases an HF-treated starting surface results in better passivation, as seen for example for the TiO x films deposited at 50°C in Fig. 4. This is also generally the case for ALD Al 2 O 3 which provides a better surface passivation quality on HF-treated silicon in comparison to chemically pre-oxidized silicon [73]. Generally, it appears that some materials and some deposition processes are capable of forming their own high-quality interfaces, while others benefit from the presence of a pre-formed oxide interface. Therefore, the influence of the starting surface needs to be investigated for different materials and processes on an individual basis.

Influence of film thickness
The thickness of a passivation layer is another known parameter affecting passivation quality. It is conventionally expected that very thin films will provide worse passivation than thicker films. More specifically, the passivation quality is usually observed to improve with film thickness up to a certain point (typically~10 nm) and then to saturate. This is well-established for example for thermal SiO 2 [50], PECVD SiN x [51], and ALD Al 2 O 3 [74][75][76]. At least in case of Al 2 O 3 , which owes its excellent passivation partly to a strong field-effect passivation, it is the chemical passivation which deteriorates when moving to very thin films [77]; this may also apply to other passivating materials. Note that for a-Si:H, saturation in fact appears not to occur even for thicknesses up to several hundred nm, but the passivation quality is still observed to improve monotonically with thickness [34].
In contrast to this simple picture, we have found that for some materials passivation is not a monotonic function of film thickness. In certain cases there is a clear optimum at a relatively low thickness value, as is illustrated in Fig. 5 for TiO x after an FGA treatment. A similar optimum thickness behavior was also observed for TiO x by Yang et al. who also used an FGA treatment [22,23]. Consequently, if only relatively thick films had been investigated in these cases, as is often the case, passivation would have appeared considerably worse.
Our results and those of Yang et al. [22,23] are in contrast to those of Cui et al., who used an N 2 annealing treatment and observed saturation of passivation for TiO x thicknesses between~8 and 63 nm, which resembles the classical case where the passivation quality saturates above a certain layer thickness [21]. Therefore, the thickness dependence of passivation appears not to be an intrinsic property of TiO x , but depends on the processing history of the films. Since each of these studies employed different ALD precursors, deposition temperatures, and annealing treatments, these may account for the difference in passivation behavior.
Alternatively, for some materials saturation may occur, but much more rapidly than for SiO 2 or Al 2 O 3 . Indeed saturation of passivation  can occur after only a few monolayers for some materials. An example of this was shown in Fig. 3, where passivation by Nb 2 O 5 apparently saturates already after 1 nm on RCA oxide surfaces. Film thickness can therefore influence passivation in unexpected ways and the influence of film thickness on passivation should be carefully considered especially when investigating new materials.

Post-deposition annealing: Temperature optima and the role of capping layers
It is well-known that a post-deposition annealing treatment, either in N 2 or in a hydrogen-containing gas mixture (FGA), can improve the surface passivation quality. This is for instance the case for materials such as SiO 2 , a-Si:H, SiN x , and Al 2 O 3 [16]. The general aim of such an annealing treatment is to effectively hydrogenate and passivate the dangling bonds present at the silicon surface, i.e. to improve the chemical passivation. Generally, a sufficiently high temperature has to be applied to enable the diffusion of hydrogen from either the passivating film itself or from a hydrogen-containing ambient (e.g. forming gas) towards the silicon interface, while at the same time avoiding excessive hydrogen loss from the film via effusion. In the case of both SiO 2 and Al 2 O 3 , the optimal annealing temperature is known to be close to 400°C [16], while much lower temperatures around 180°C appear optimal for a-Si:H [78]. This already indicates that the optimal annealing temperature may vary from material to material.
This point is made even more obvious when looking at the wide range of novel passivating materials, as is shown in Fig. 6. It is striking to see that the optimal annealing temperatures for MoO x , Nb 2 O 5 , Al 2 O 3 , ZnO:Al, and PO x are all different, indicating different annealing kinetics for each of these materials. It should be noted that since the annealing step time was not equal for all of the materials, a one-to-one comparison cannot be made for all the materials in this graph. Nonetheless, the observed trends should still be dominated by the temperature rather than by the anneal step time, since the rate of thermally-activated processes depends much more strongly on temperature. This is for example seen in the case of SiO 2 /Al 2 O 3 stacks, where the apparent saturated level of passivation differs for different annealing temperatures [79].
When comparing the three samples on the RCA oxide, i.e. MoO x , Nb 2 O 5 , and ZnO:Al/Al 2 O 3 , it can be seen that the ZnO:Al/Al 2 O 3 sample has a much higher onset and optimum temperature. This is striking if one assumes that the annealing kinetics depend on the hydrogenation reaction at the interface, as it is the same RCA SiO x interface that needs to be passivated during the anneal treatment. These differences are therefore likely linked to differences in the kinetics of hydrogen diffusion through the material and toward the interface and effusion of hydrogen from the material. Conceivably, the transport of hydrogen takes place more readily and at lower temperatures in the amorphous MoO x and Nb 2 O 5 films as compared to the polycrystalline ZnO/Al 2 O 3 stack. Since diffusion plays a role, film thickness is also likely to play a role in the exact annealing trend. Also material stability and possible crystallization during annealing can play an important role in degradation at higher temperatures and thus the location of the optimum. For the case of MoO x , macroscopically visible formation of etch pits as well as a reduction of the material from near-stoichiometric MoO 3 to MoO 2 was observed by XPS when performing FGA treatments exceeding 300°C, explaining the strong reduction in lifetime observed.
Capping of passivation layers prior to annealing by other materials, most notably Al 2 O 3 and SiN x , has proven to be a highly successful approach to improve or enable passivation. As illustrated in Fig. 7, it has for example been demonstrated that applying an Al 2 O 3 capping layer can greatly enhance the passivation provided by ALD SiO 2 , SiO x /poly-Si, and SiO x /ALD ZnO. For SiO 2 and poly-Si, it has been shown that the Al 2 O 3 layer serves both as a hydrogen source and a hydrogen effusion barrier [78,80,81]. Note that hydrogen effusion barriers can also be formed by other materials, such as SiN x [82]. Conversely, in the case of ALD ZnO, ample hydrogen for passivation is already present at the SiO x interface and in the ZnO layer after the ALD process. In this case, the Al 2 O 3 capping layer merely serves as a hydrogen effusion barrier that prevents a significant effusion of hydrogen at 400°C [28].
However, the success of such Al 2 O 3 capping layers does not necessarily translate to other passivation materials, as Fig. 7 shows for the cases of Nb 2 O 5 , MoO x and TiO x . For these three materials, the addition of the Al 2 O 3 capping layer leads to a strong reduction in passivation. In the case of MoO x capped with Al 2 O 3 , the deterioration is likely related to the visible formation of blisters when annealing at 300°C. Such blistering was not observed for Nb 2 O 5 and TiO x capped by Al 2 O 3 , and the precise reasons for the detrimental effect of capping by Al 2 O 3 are so far not known. Conceivably, the photocatalytic nature of both TiO x and Nb 2 O 5 plays a role. Both TiO x and Nb 2 O 5 yield improved passivation upon light soaking, which is thought to originate from an enhanced  field-effect passivation arising from more net negative fixed charge in the dielectric [20,29,83]. In the case of TiO x , it has been argued that this enhancement of the negative charge upon light soaking originates from a photocatalytic reaction between the TiO x surface and molecules from the ambient [83]. When an Al 2 O 3 capping layer is applied, the photocatalytic surface of TiO x or Nb 2 O 5 is no longer exposed to the ambient. Therefore, such light-induced charging would be hindered, resulting in a lower field-effect passivation.

Discussion
We have shown how the choice of starting surface, the film thickness, and the use of capping layers can all be critical for achieving surface passivation by deposited layers. While we have considered these factors separately, we should stress that in reality different factors can be intertwined. The use of a pre-grown tunnel oxide can be very beneficial for TiO x passivation at one deposition temperature for example, but for another deposition temperature it can give poor results. The point is that all of these factors need to be considered when investigating new materials for surface passivating properties.
The corollary to this point is that it is much harder to say with confidence that a material does not passivatethe negative casethan to confirm the positive case that it does. It may simply be that one has not yet found the necessary combination of surface pretreatments, process settings, and post-deposition treatments to achieve passivation. There is more than one example of a material known now to be effective for passivation that was written off in initial investigations.
Of course additional factors not discussed in this work, but wellknown to affect passivation, including the substrate dopant density and orientation, the details of the deposition process (temperature, pressure, chemistry, etc.) and the presence of extrinsic doping in the passivating film, will also be highly important. In this work we have confined ourselves to aspects that appeared to us to be not so widely appreciated or documented in the literature, and which we could illustrate with our own data.
It should also be noted that, while we have talked broadly in terms of achieving or not achieving passivation, optimized passivation quality varies significantly between different materials and processes, as can be seen from an examination of the lifetime values reported throughout this work. Passivation quality is of course also not the only criterion that needs to be considered for device applications; optical properties, process compatibility, and environmental stability, among other factors, are all important. A low contact resistivity is an additional criterion important for materials intended for passivating contact applications [1]. However, the relevance of these latter aspects depends in the first place on effective surface passivation, underlining this as a key design criterion for all layers in direct contact with the silicon surface in c-Si solar cells.

Conclusions
Silicon surface passivation remains a highly important research topic, which has lately also been illustrated by the surging interest in passivating contact materials. This emerging field has prompted the investigation of numerous novel materials for silicon surface passivation. For many of these materials, it seems that considerations relevant to passivation such as starting surface, film thickness, annealing treatment, and the role of capping layers come into play in ways that sometimes run counter to conventional assumptions. In this work, we have explored these considerations both for more conventional and for recently-discovered passivation materials including MoO x , ZnO, TiO x , PO x , and Nb 2 O 5 .
The main conclusion that can be drawn from this exploration is the fact that all these considerations strongly influence the passivation performance of most passivation materials. In addition, the optimal choice of one experimental parameter is also often found to depend on the choice of another. Therefore, the interdependence of all these factors needs to be carefully considered when investigating new materials for surface passivating properties. These lessons can prove useful in the design and optimization of future passivation schemes and passivating contacts for c-Si solar cells. Fig. 7. Effective excess carrier lifetime at an excess carrier concentration of 10 15 cm −3 for a variety of materials with and without capping by 30 nm thermal ALD Al 2 O 3 after optimized annealing treatments (10 min at 400°C in N 2 for SiO 2 , 1 h at 400°C in forming gas for poly-Si, 30 min at 450°C in forming gas for ZnO:B, and 2 h, 1 h, and 30 min at 300°C in forming gas for Nb 2 O 5 , MoO x , and TiO x , respectively). Data on ALD SiO 2 have been taken from [15]. poly-Si, ZnO [28], Nb 2 O 5 and MoO x were prepared on RCA SiO x layers, while TiO x layers were prepared on HF-treated Si.