Temporal and spatial atomic layer deposition of Al-doped zinc oxide as a passivating conductive contact for silicon solar cells

deposition


Introduction
So-called passivating contacts have been an immensely popular topic in research, especially over the last decade [1][2][3][4][5][6].Many material systems have been studied and passivating contacts can nowadays be found in many commercial solar cells [7].In conventional homojunction contacts, carrier-selectivity arises from the asymmetric conductivity induced by the highly doped n + and p + regions within the c-Si wafer.This facilitates majority carrier extraction at the metal contact and shields minority carriers from the recombination-active Si-metal contact.Yet, high doping of the c-Si wafer comes at the expense of Auger-recombination losses, limiting the open-circuit voltage V oc .Passivating contacts improve upon conventional homojunction contacts by inducing carrier-selectivity from passivating films deposited on the wafer, i.e. they are heterojunctions.This renders high n + and p + doping of the wafer obsolete and moreover makes sure that there is no recombination-active direct Si-metal contact.Because of this, passivating contact cells typically boast much higher V oc values [1][2][3].
Many material classes have been explored for passivating contacts, including doped hydrogenated amorphous (a-Si:H) and polycrystalline silicon (poly-Si) as well as metal oxides, fluorides and nitrides (referred to here as metal compounds) [1][2][3].The main potential benefit of metal compounds over doped Si is their generally higher transparency because of their wide bandgap, enabling less parasitic absorption of light by the contact layers [1,2].Yet, to date the most efficient cells are still made using doped Si contacts, as doped Si makes up for its limited transparency (so relatively lower short-circuit current density J sc ) by having superior surface passivation and majority carrier extraction (hence high V oc and fill factor FF) [1].While likely not a fundamental issue, in practice all metal compound passivating contacts have one or more drawbacks in this respect, including non-perfect passivation, poor selective carrier extraction, and/or being susceptible to external influences that can impair the contact performance [1].The latter is caused by a Fermi level that is easily shifted in these typically wide bandgap, intrinsic materials.Because of this, doping of metal compounds is a promising approach to improve selectivity and resilience of the Fermi level to external influences, as also recently proposed by Michel et al. [1].
Within this broader context, it is interesting to consider doped ZnO as contact material.As outlined below and in previous work, doped ZnO can enable a hybrid approach between homo-and heterojunction contacts where a n + -doped Si surface is contacted by a passivating transparent conductive oxide (TCO) [8].Moreover, being a TCO, it can simultaneously aid in lateral conductivity and antireflection.For the purpose of clarity, the main findings so far are summarized here for this contact material.Specifically, in 2019 our group showed that ALD ZnO/Al 2 O 3 stacks can yield excellent passivation on n-type c-Si, as witnessed by implied V oc levels of 728 mV [9].While ZnO was not well-known for its surface passivating properties, in that work it was demonstrated that good surface passivation can be achieved provided a few key aspects are taken into account [9].Firstly, it is prerequisite to form a high-quality SiO 2 interlayer on the c-Si wafer prior to ALD of ZnO, as it appears that ALD ZnO does not naturally form a high-quality interfacial SiO 2 interface on c-Si.Secondly, the presence of an ALD Al 2 O 3 capping layer is essential to prevent effusion of hydrogen from the stack during an anneal at 400-450 • C that is necessary to activate the passivation.Finally, doping of the ZnO can enhance the field-effect passivation, since doping reduces the work function of the ZnO and thereby results in more downward band bending at the c-Si surface [9].
In subsequent work, we demonstrated that this ZnO-based stack can not only provide surface passivation, but can also yield a low contact resistivity to n + -diffused surfaces and poly-Si(n) contacts, as well as providing a low sheet resistance and good optical incoupling into c-Si [8].In addition to demonstrating the contact properties, many insights were gained in the working principle of this contact stack.Doping of both the ZnO:Al and n-type c-Si was shown to be required for achieving low contact resistivities down to 15 mΩcm 2 .The anneal step was found to lead to optimal passivation at temperatures around 450-500 • C, where higher annealing temperatures lead to rapid loss of passivation.The latter was shown by transmission electron microscopy (TEM) imaging to originate from break-up of the interfacial SiO 2 at high annealing temperatures.The anneal step also enables significant improvements in the carrier mobility (by approximately 50%) of the ZnO:Al (up to 38 cm 2 /Vs), such that resistivity and transparency levels typical to those of indium-based TCOs are reached.Specifically, depending on the doping level, resistivity values well below 1 mΩcm and down to 0.5 mΩcm were achieved, while optical modelling show a J sc potential in excess of 42 mA/cm 2 (ignoring front metallization) [8].Importantly, this improvement is only observed when an Al 2 O 3 capping layer is present, whereas ZnO:Al films without an Al 2 O 3 capping layer degrade strongly upon annealing.Similar to surface passivation, it is thought that passivation of defects in the ZnO by hydrogen play an important role in this mobility improvement.Moreover, TEM imaging elucidated two more effects of annealing that can contribute to an improved carrier mobility.Firstly, upon annealing the grain size increases due to coarsening.Secondly, the Al dopants that originally are present in planes due to the ALD supercycle deposition approach become more isotropically spread over the film [10].
In this work, we present the latest results and insights obtained for this contact stack.Firstly, the upscaling potential is highlighted by demonstrating that the ZnO:Al/Al 2 O 3 stack can also be prepared by spatial ALD, yielding passivation levels on both undiffused n-type and n + -diffused c-Si surfaces that are similar to those obtained by temporal ALD.Furthermore, we demonstrate that both the ZnO and Al 2 O 3 layers can be made significantly thinner than in previous studies without significant loss of passivation.Also, we provide details on a pH-controlled wet-etch that was developed to remove the Al 2 O 3 selectively from the ZnO:Al after the post-deposition anneal, such that the ZnO:Al can be contacted by a metal.Finally, we demonstrate the capability of ALD to deposit ZnO:Al layers selectively on oxidized regions of an otherwise HFlast treated c-Si surface.Such area-selective deposition opens up potential pathways for local, self-aligned contact formation.

Substrates and pre-treatments
In this work, two types of c-Si substrates were used for the preparation of symmetrical lifetime samples.The first type are double-side mirror-polished floatzone (FZ) c-Si wafers, which are 285 μm-thick and are n-type doped with a resistivity of 3 Ωcm.Secondly, to evaluate the passivation performance on n + -diffused surfaces, Czochralski (CZ) c-Si wafers (n-type, 3 Ωcm, 185 μm-thick) with a 137 Ω/sq n + -emitter prepared by POCl 3 diffusion in a tube furnace were used.These wafers featured an interfacial oxide either grown by Radio Corporation of America (RCA) cleans or by UV/O 3 oxidation.UV/O 3 oxidation was performed for 30 min at room temperature in a Novascan PSD Series UV Ozone Cleaner, directly after a 1% HF dip for 1 min and rinse in DI water.

Film preparation
Both temporal and spatial ALD were used to prepare the ZnO:Al and Al 2 O 3 layers in this work.For both types of ALD, diethylzinc (DEZ) and dimethylaluminumisopropoxide (DMAI) were used as the zinc and aluminum precursors, respectively, and H 2 O as co-reactant.For Aldoping of the ZnO, DMAI was used as it is a more efficient doping precursor compared to trimethylaluminum (TMA) [11].On the temporal ALD tool, the DMAI and DEZ bubblers were kept at 50 and 25 • C, while on the spatial ALD tool they were kept at 25 and 35 • C, respectively.
For Al-doping of the ZnO:Al both the so-called supercycle and coinjection approaches were used.The supercycle method was used for temporal ALD, where a doped film is deposited by repeating a pattern of an integer n cycles of ZnO and one Al 2 O 3 cycle.In this approach, the doping level can be controlled accurately by varying the integer n.The co-injection method was used for spatial ALD, where the Zn and Al precursors are injected simultaneously every cycle and the doping level can be controlled by the mixing ratio of the two (or more) precursors.
Temporal ALD was performed in an Oxford Instruments OpAL™ reactor at a table temperature of 200 • C, similarly as described in our previous work [8].The dose times were 60 ms for both DEZ and DMAI and 200 ms for H 2 O.A 6 s purge was used after the DEZ and DMAI steps, while a 10 s purge was used after the H 2 O dose.Spatial ALD was performed at atmospheric pressure in an R&D tool of the company SALD (former SoLayTec) at 230 • C.This tool allows for single-sided deposition and in-line atmospheric pressure processing.Moreover, it is configurable for both co-injection and supercycles, where in this work we focused on co-injection.The system consists of a substrate table that is moved back and forth underneath gas nozzles with a velocity of mm/s.The gas nozzles are supplied with precursors from heated canisters through bubbling with N 2 , while the H 2 O co-reactant is supplied from an evaporator.Nitrogen gas curtains were used to prevent intermixing of the H 2 O and precursor gases present in the distinct precursor zones.For Al-doping of the ZnO, DMAI and DEZ were co-dosed using N flows of 0.5 slm and 2.0 slm, respectively.XPS showed successful incorporation of Al (1.5 at.%) and the carrier density was raised from 0.3 × 10 20 to 1.6 × 10 20 cm − 3 with respect to intrinsic ZnO (iZnO).
Silicon nitride (SiN x ) was tested as alternative capping layer and was deposited by plasma-enhanced chemical vapor deposition (PE-CVD) using an inductively coupled plasma tool (PlasmaLab 100 ICP) from Oxford Instruments, similarly as described in our earlier works [12,13].The substrate temperature was kept at 200 • C, where helium backflow was applied to the sample to ensure a good thermal contact between the sample holder and the substrate.Nitrogen and silane were injected with a flow of 24 and 16 sccm, respectively, and a plasma power of 500W was applied.The pressure was actively controlled by an automated pressure controller (APC) at 12 mTorr.From XPS, the Si:N ratio of these films was determined to be 1.3.
Post-deposition forming gas anneals (FGA, 10/90H 2 /N 2 ) were performed in a Jipelec rapid thermal anneal furnace.All anneals were done consecutively for 5 min at temperatures ranging from 300 to 650 • C at 50 • C increments.Selective wet-etching of Al 2 O 3 from the ZnO(:Al) stack was done using a 0.1 M solution of Na 2 CO 3 at 60 • C of which the pH was controlled to 11.8 by KOH dripping, similarly as done in our earlier work and in the work of Sun et al. [8,14].

Analysis
The passivation quality and the implied open-circuit voltage (iV oc ) were evaluated using a Sinton WCT-120TS lifetime tester.Spectroscopic ellipsometry (SE) was used for determination of the optical constants and thicknesses of ALD layers.The optical constants of ZnO:Al were modeled using a combination of a Tauc-Lorentz and a Drude oscillator, while a PSEMI-M0 model was used for undoped ZnO samples.The latter was done since the PSEMI-M0 model is better able to account for the sharper transition in absorption around the bandgap energy that is typical of undoped films.The Drude oscillator was used to evaluate the carrier density in the film [15].The Al 2 O 3 and SiN x dielectrics were modeled in their transparent regions (i.e.below the band gap) using a Cauchy function.The film composition was evaluated by X-ray photoelectron spectroscopy (XPS) using a Thermo Scientific KA1066 spectrometer.

Results and discussion
The passivation quality of spatial ALD ZnO:Al/Al 2 O 3 stacks was evaluated on symmetrical lifetime samples on both undiffused FZ and n + -diffused (137 Ω/sq) CZ c-Si(n) wafers.As can be seen in Fig. 1, on both substrate types the passivation in the as-deposited state is poor, irrespective of the ZnO:Al doping level.Note that for the n + -diffused wafers the iV oc in the as-deposited state is significantly higher than for the undiffused wafers, since the n + -diffusion helps shield minority carriers (i.e.holes) from the poorly passivated surface [3].Upon annealing, the passivation level improves significantly for both substrate types, with an optimum around 400-450 • C of annealing, resulting in iV oc values up to 718 and 701 mV for undiffused and n + -diffused wafers, respectively.For annealing temperatures of 500 • C and higher, the passivation quality degrades rapidly.The same trend was observed in our previous work on temporal ALD ZnO:Al contacts, where a TEM study suggested that roughening and disordering of the interfacial SiO 2 at high annealing temperatures contributes to this loss of passivation [8].
On undiffused wafers, the doped ZnO:Al sample reaches an iV oc value of 718 mV, which is slightly better than the 711 mV obtained for iZnO.Also in our previous work on passivation by temporal ALD ZnO, it was found that the passivation level consistently improves slightly for increasing doping levels of the ZnO contacts.This was attributed to a change in interfacial band bending induced by the quickly diminishing work function (WF) of the ZnO contact upon stronger degenerate doping, which was supported by band structure simulations [9].Presumably, the well-passivated interfacial oxide helps reduce Fermi-level pinning such that the lower WF for doped films also effectively translates to more downward band bending at the c-Si interface.On the n + -diffused surfaces, no significant difference in the passivation level is observed between the iZnO and ZnO:Al samples, both of which reach an iV oc of (700 ± 2) mV.Although hard to pinpoint exactly from this dataset, plausible explanations include (1) that the carrier population at the c-Si interface is more strongly dictated by the n + doping of the c-Si rather than the ZnO:Al doping level (and hence WF) as compared to undiffused c-Si(n) wafers and (2) that for these samples any relatively small differences in surface recombination have a less pronounced effect on the iV oc , as Auger recombination in the highly-doped n + -region plays a role as well.Specifically, using EDNA 2 calculator the J 0 due to Auger recombination in the highly-doped has been estimated at 15 fA/cm 2 , which is on the same order as the surface recombination (see next paragraph).Also, this additional (Auger) recombination induced by n + doping of the c-Si subsurface is the most likely reason why lower iV oc values are obtained as compared to undiffused surfaces.
B. Macco et al. this work as well as temporal ALD results from previous work, where some datapoints were acquired on PERC half-fabricate cells [8].As can be seen, the ZnO-based contacts generally show a high level of passivation.The spatial ALD films specifically also give low J 0 values down to 32 fA/cm 2 on the 137 Ω/sq surface, yet are not completely on par with the result obtained for temporal ALD on the PERC half-fabricate cells.It should however be stressed that this might not be due to a fundamental difference between temporal and spatial ALD, but could also arise from the fact that in this specific case the temporal ALD ZnO:Al was more highly doped.Specifically, the temporal ALD sample prepared with a supercycle ratio n = 24 has a higher carrier density N e of 3.3 × 10 20 cm − 3 as compared to 1.6 × 10 20 cm − 3 for spatial ALD.
As mentioned in the introduction, the Al 2 O 3 capping layer is crucial for achieving surface passivation as it prevents the effusion of hydrogen from the passivation stack upon thermal annealing [9].However, thus far all passivation results were obtained for relatively thick Al 2 O 3 capping layers, i.e. 30 nm.In this work, the thickness required for achieving good passivation was investigated in detail.Such insight is relevant for implementation of the passivating stack, since the Al 2 O 3 layer is sacrificial, i.e. every nanometer of Al 2 O 3 that has been deposited also has to be etched of after the post-deposition anneal.Fig. 3 shows the passivation performance of spatial ALD 30 nm iZnO/x nm Al 2 O 3 passivation stacks as a function of annealing temperature.For 0 and 2 nm of Al 2 O 3 no appreciable passivation is observed.Importantly, for both these samples approximately millimeter-sized specks started to appear after forming gas annealing at 500 • C, while the films were completely removed after annealing at 550 • C. For annealing below 500 • C, such removal of the film and/or degradation was not observed.This suggests that the Al 2 O 3 capping layer not only plays a key role for hydrogenation but also helps prevent etching of the ZnO during forming gas annealing, provided the Al 2 O 3 is sufficiently thick and closed.For 4 nm of Al 2 O 3 capping layer, some activation of the surface passivation is observed, reaching modest iV oc values of around 600 mV.For Al 2 O 3 capping layers of 8 nm and thicker, equally good passivation is observed for annealing temperatures up to 450 • C.This demonstrates that for reaching a high level of passivation, the Al 2 O 3 capping layer can be significantly thinner than the previously-employed thickness of 30 nm.
Yet, upon annealing at 500 • C it is observed that a thicker capping layer does help to mitigate the loss of passivation, presumably by preventing the effusion of hydrogen more effectively.After annealing at 550 • C, the passivation is effectively lost for all Al 2 O 3 capping layer thicknesses, which is thought to originate from aforementioned disintegration of the interfacial oxide as observed in TEM [8].
Also 40 nm of PE-CVD SiN x was tested as an alternative capping layer, either directly on iZnO or in a 8 nm Al 2 O 3 /SiN x stack [24] Our experiments showed very low lifetimes around 1 μs for the iZnO/SiN x stack, which could not be recovered by post-deposition annealing (not shown).For the 8 nm Al 2 O 3 /SiN x stack, annealing leads to only slight surface passivation, where the achieved iV oc values are well below 600 mV.The low passivation could potentially be due to plasma-induced damage by the SiN x deposition as has been observed for various systems, including a-Si:H and SiO 2 /poly-Si passivation layers, and has often been attributed to photons and ions [25][26][27][28][29]. Though still a topic of study, this does show that the Al 2 O 3 capping layer is not easily replaced by other common capping layers.
Furthermore, it was found that the ZnO layer in the ZnO/Al 2 O 3 stack can be made relatively thin, without significant loss of passivation.As can be seen in Fig. 4, the passivation quality only decreases slightly when going from 75 nm thick ZnO (the optimal for antireflection) down to 5 nm.For such thin layers, one of the potential reasons for reduced passivation could be the fact that the film is not closed.Indeed, in earlier work it was shown that ALD ZnO grows in a so-called island mode, where it takes close to 10 nm for coalescence of these islands (i.e.film closure) to complete [26].For the 0 nm ZnO data point (i.e.only SiO 2 /Al 2 O 3 ) the passivation level is relatively low.While such a stack should yield a high level of chemical passivation, the lower passivation level can partly originate from the negative fixed charge in such a stack which is better suited for passivating p-type c-Si surfaces [30].Yet, the main difference is thought to originate from the employed anneal temperature of 450 • C which is optimized for the ZnO contacts, whereas the best passivation for Al 2 O 3 layers themselves is typically found using a 400 • C annealing temperature.
For selective etching of the Al 2 O 3 from the ZnO:Al layers, a procedure based on the work of Sun et al. was developed [14].This pH-controlled selective wet-etch is based on a 0.1 M Na 2 CO 3 solution at 60 • C of which the pH is raised to 11.8 by KOH dripping.As shown in Fig. 3. Influence of the thickness of the Al 2 O 3 capping layer thickness on the passivation quality for 30 nm iZnO/x nm Al 2 O 3 stacks prepared by spatial ALD on undiffused FZ c-Si(n) wafers with an RCA SiO 2 layer.To evaluate the possibility of using SiN x as an alternative capping layer, 30 nm iZnO/40 nm SiN x stacks and 30 nm iZnO/8 nm Al 2 O 3 /40 nm SiN x stacks were prepared.For the iZnO/SiN x sample the iV oc could not be determined reliably due to extremely poor passivation.Consecutive annealing steps were performed for 5 min in forming gas ambient.This reduction in wet-etch rate upon annealing is reminiscent of SiN x which becomes much more resistant to etching in hydrofluoric acid after high temperature annealing [31].This was attributed to densification and release of H from the layer, resulting in more strong Si-N bonds.Also the Al 2 O 3 layers studied in this work densify significantly upon annealing, as shown in Fig. 6.The thickness of the Al 2 O 3 layers decreases upon annealing, which is accompanied by an increase in refractive index n.The Al 2 O 3 films prepared at 200 • C have a slightly lower refractive index as compared to films prepared at 230 • C. Also, the films prepared at 200 • C lose relatively more thickness, specifically a little over 9% after annealing at 600 • C as compared to close to 6% for films prepared at 230 • C. Importantly, even though the reduction in wet-etch rate upon annealing is concomitant with densification, the wet-etch rate of the films prepared at 200 and 230 • C is approximately equal after annealing temperatures of 350 • C and higher, while the films prepared a 230 • C still have a higher refractive index.

Area-selective ALD of ZnO:Al
Another important aspect when implementing these ZnO-based contacts is the fact that film growth, and specifically the nucleation, depends strongly on the surface preparation of the c-Si wafer.As can be seen in the top panel of Fig. 7, on HF-last c-Si surfaces it takes approximately 35 ALD cycles for iZnO to nucleate, while iZnO readily nucleates on an RCA-prepared SiO 2 surface.With a growth-per-cycle (GPC) in the steady-growth regime of 0.18 nm for ZnO, this means that for an identical number of cycles, the resulting film will be about 7 nm thinner on HF-last surfaces than on SiO 2 surfaces.This nucleation delay is thought to originate from a lower reactivity of the DEZ precursor towards the relatively inert Si-H terminated surface, as compared to the hydroxyl groups found on an oxide surface.This statement is motivated by earlier work of Mameli et al., where we already showed that ALD ZnO from the reactants DEZ and H 2 O has a strong nucleation delay on Si-H terminated a-Si:H layers, while ZnO nucleates readily on intentionally oxidized a-Si: H [32]. Density functional theory calculations indicated that a lack of reactivity of DEZ to the relatively inert Si-H surface groups stands at the basis of this nucleation delay.Specifically, while the chemisorption of DEZ on surface Si-H and Si-OH groups is exothermic in both cases with an energy gain of 0.98 and 1.25 eV, respectively, there is a much stronger activation energy on Si-H surfaces (1.16 eV) as compared to Si-OH surfaces (0.81 eV).Importantly, it was excluded that oxidation of the Si-H surface during H 2 O exposure in the ALD cycle is the reason for loss of selectivity as this process is much slower.This is in line with the common notion that HF-dipped c-Si surfaces are relatively resilient to oxidation in ambient conditions.
Mameli et al. leveraged this surface-dependent nucleation effect to grow ZnO selectively on a-Si:H surfaces that were oxidized locally [32].Such an approach is often coined area-selective ALD (AS-ALD), a field that is considered an upcoming enabling technology in the semiconductor industry for bottom-up fabrication of nanostructures, yet   likely less well-known in the PV community [33].Since this similar effect is observed on c-Si, in principle it can be used to grow ALD ZnO area-selectively on an HF-wafer that is oxidized locally, or conversely an oxidized wafer which is HF-treated locally.Here, we validate the former approach by exposing an HF-treated c-Si wafer to an ICP O 2 plasma for 4 s through a mask in the temporal ALD reactor.Subsequently, the mask was removed and 60 ALD iZnO cycles (nominal thickness of 10 nm) and 100 cycles ALD Al 2 O 3 (nominal thickness of 8 nm) were performed.After ALD, the masked pattern was clearly visible to the naked eye.On the O 2 plasma-treated surface, ellipsometry measurements confirmed that this results in a near nominal stack thickness of 9.2/7.9nm iZnO/Al 2 O 3 .The iZnO thickness was markedly lower on the non-treated area, specifically the iZnO/Al 2 O 3 stack was 3.3/8.6nm thick.This is in line with the nucleation delay observed on HF-last surfaces and confirms that approximately 6 nm of iZnO can be grown selectively on SiO 2 -terminated surfaces before film growth starts on HF-last surfaces.As was shown in Fig. 4, such a thickness is sufficient to have a well-passivating ZnO layer on the SiO 2 -terminated area.Although a specific cell implementation or process flow of such area-selective ALD of ZnO is not directly evident, it does open up a potential innovative integration pathway for these ALD contacts.Especially since ALD ZnO only yields passivation on thin interfacial silicon oxides but not on HF-last surfaces, it can be advantageous to grow ALD ZnO only on predefined SiO 2 areas in a self-aligned manner.Yet, since doping of the ZnO is a strong prerequisite to form low-Ohmic contacts to c-Si, such approach would fail if the doping step in the supercycle would lead to nucleation on the HF-last surfaces.For example, ALD Al 2 O 3 grown from TMA and H 2 O -probably the most widely adopted ALD process in the PV industry -is known to nucleate readily on HF-last surfaces.Hence, using TMA to dope ZnO:Al in a supercycle manner should lead to rapid loss of selective deposition.Here we show that the less reactive DMAI precursor used in this work also has an appreciable nucleation delay on HF-last surfaces and can thus be used to dope the ZnO:Al without significant loss of area-selectivity.As can be seen in the bottom panel of Fig. 7, Al 2 O 3 grown by DMAI and H 2 O has no appreciable nucleation delay on an SiO 2 surface, while a delay of almost 20 cycles is present on the HF-last surface.The top panel of Fig. 7 shows the nucleation delay on HF-last c-Si of ZnO:Al prepared using supercycle ratios 12 and 48.For a supercycle ratio of 48, the nucleation behavior is very similar as for iZnO.This would point to negligible if any reduction of the selectivity by the Al-doping step, albeit that the first Al-doping cycle is performed only after 24 ALD ZnO cycles, which is close to the nucleation delay itself of approximately 35 cycles.For the n = 12 sample the nucleation delay is slightly reduced to approximately 27 cycles, at which point the surface has already experienced two Al-doping cycles (after ZnO cycles 6 and 18).This demonstrates that the DMAI step reduces the selectivity only slightly, even for extremely highly doped films prepared with n = 12.

Conclusions
ZnO-based contacts present an interesting approach for fabricating c-Si solar cell contacts, as they can combine high levels of surface passivation and low-Ohmic contacting, as well as lateral conductivity and antireflection.In this work, we provide the latest insights gained into both the practical aspects and working mechanism of this contacting approach.Firstly, it has been shown that spatial ALD is a viable method to prepare the contact stack with passivation performance on a similar level as found in earlier studies using lab-scale temporal ALD, underlining the potential for industrialization.Additionally, it was found that the thickness of the ZnO and Al 2 O 3 layers can be reduced considerably from 75 to 30 nm in previous studies to at least 5 and 8 nm, respectively, without significant loss of passivation quality.Selective etching of the sacrificial Al 2 O 3 from the ZnO was shown to be fast and effective in a basic Na 2 CO 3 solution.Although the wet-etch rate of Al 2 O 3 drops significantly after post-deposition annealing as the layer densifies, still practically high etch rates over 20 nm/min were achieved.Yet, the influence of etching in a Na 2 CO 3 solution on the solar cell process flow has not been investigated yet and will be left for future work.Finally, since both ALD ZnO and Al 2 O 3 exhibit a strong nucleation delay on HFtreated surfaces, ZnO:Al layers can be grown area-selectively on a locally oxidized region of HF-treated wafers, opening up potential innovative integration pathways for this contact material.

Fig. 1 .
Fig. 1. iV oc values for SiO x /ZnO(:Al)/Al 2 O 3 stacks prepared by spatial ALD on undiffused FZ and n + -diffused CZ c-Si(n) wafers with RCA and UV/O 3 SiO 2 layers, respectively.The ZnO and Al 2 O 3 films had a thickness of 30 nm.Consecutive annealing steps were performed for 5 min in forming gas ambient.The corresponding J 0 values on n + -diffused surfaces can be found in the Supplementary information, Fig.S1.

Fig. 4 .
Fig. 4. Influence of the thickness of the iZnO layer thickness on the passivation quality for iZnO/30 nm Al 2 O 3 stacks prepared by temporal ALD on FZ c-Si(n) substrates that feature an SiO 2 interlayer prepared by UV/O 3 oxidation.The samples were subjected to a post-deposition anneal in forming gas ambient at 450 • C for 30 min.The dashed line is a guide to the eye.

Fig. 5 ,
Fig.5, very high etch rates for Al 2 O 3 can be achieved with this etching solution, while no appreciable etching of ZnO occurs.Importantly, the etch rate depends on the deposition temperature and decreases strongly after post-deposition annealing.Specifically, ALD Al 2 O 3 prepared at 200 • C is etched faster than Al 2 O 3 prepared at 230 • C. Upon annealing the etch rates decrease and converge for annealing temperatures of 350 • C and higher.This reduction in wet-etch rate upon annealing is reminiscent of SiN x which becomes much more resistant to etching in hydrofluoric acid after high temperature annealing[31].This was attributed to densification and release of H from the layer, resulting in more strong Si-N bonds.Also the Al 2 O 3 layers studied in this work densify significantly upon annealing, as shown in Fig.6.The thickness of the Al 2 O 3 layers decreases upon annealing, which is accompanied by an increase in

Fig. 5 .
Fig. 5. Influence of annealing on the wet etch rate of Al 2 O 3 in a 0.1 M solution of Na 2 CO 3 (60 • C, pH = 11.8 by KOH dripping).ALD Al 2 O 3 films were prepared by temporal ALD at 200 and 230 • C. Consecutive annealing steps were performed for 5 min in forming gas ambient.The etch rate was determined by spectroscopic ellipsometry.It was verified that both ZnO and SiN x are not etched by this solution.

Fig. 6 .
Fig. 6.Change in refractive index n and relative thickness of Al 2 O 3 films upon post-deposition annealing as determined from spectroscopic ellipsometry.ALD Al 2 O 3 films were prepared by temporal ALD at 200 and 230 • C. Consecutive annealing steps were performed for 5 min in forming gas ambient.

Fig. 7 .
Fig. 7. Nucleation curves of temporal ALD ZnO:Al (top panel) and Al 2 O 3 (bottom panel) on H-terminated Si surfaces (labeled HF-last) and Si surfaces with an SiO 2 layer prepared by RCA cleans.The film thicknesses were tracked in-situ by spectroscopic ellipsometry.