Simulation-based roadmap for the integration of poly-silicon on oxide contacts into screen-printed crystalline silicon solar cells

We present a simulation-based study for identifying promising cell structures, which integrate poly-Si on oxide junctions into industrial crystalline silicon solar cells. The simulations use best-case measured input parameters to determine efficiency potentials. We also discuss the main challenges of industrially processing these structures. We find that structures based on p-type wafers in which the phosphorus diffusion is replaced by an n-type poly-Si on oxide junction (POLO) in combination with the conventional screen-printed and fired Al contacts show a high efficiency potential. The efficiency gains in comparsion to the 23.7% efficiency simulated for the PERC reference case are 1.0% for the POLO BJ (back junction) structure and 1.8% for the POLO IBC (interdigitated back contact) structure. The POLO BJ and the POLO IBC cells can be processed with lean process flows, which are built on major steps of the PERC process such as the screen-printed Al contacts and the Al2O3/SiN\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\text{Al}_\text{2 }\text{O}_\text{3 }/\text{SiN }$$\end{document} passivation. Cell concepts with contacts using poly-Si for both polarities (POLO2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\text{POLO}^2$$\end{document}-concepts) show an even higher efficiency gain potential of 1.3% for a POLO2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\text{POLO}^2$$\end{document} BJ cell and 2.2% for a POLO2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\text{POLO}^2$$\end{document} IBC cell in comparison to PERC. For these structures further research on poly-Si structuring and screen-printing on p-type poly-Si is necessary.


Simulation method
Two important aspects when evaluating new cell structures for their potential economic benefit in large-scale production are the process complexity and the efficiency gain compared to proven concepts. To estimate the potential efficiency gain, we perform numerical device simulations using the Quokka 13 implementation of the conductive boundary model 14 . These simulations require the knowledge of the resistive and recombination properties for the bulk, the passivated surfaces and the contacts as well as a charge-carrier generation profile for each structure. For a good comparability between all analyzed cell structures we choose equal input parameters for equal or comparable properties of the respective structures. For example, the recombination at a 200 nm thick n-type poly-Si junction is always assumed to be J 0 = 3fA /cm 2 although in reality this J 0 can vary for different cell structures due to the influence of other process steps on the passivation quality. In general, we choose best-case measured parameters achieved in lab-type production with industrial-type processes. We confine our analysis to screen-printed contacts, because most industrial solar cells are produced using screen-printing. Combining all these best-measured parameters into a hypothetical simulated cell gives an estimate for cell efficiencies to be expected in the next years for cells with POLO junctions and for PERC+ cells. All input parameters are detailed in Table 1.
For the optical performance of each structure we employ ray-tracing simulations using the SUNRAYS raytracer 15 with optical parameters chosen according to Refs. [16][17][18] . This program does, however, not allow simulating the complex unit cell shown at the top of Fig. 1. Therefore, we resolve the complex optical structure into simple unit-cells shown in the second row of Fig. 1 and simulate them separately. Each of these simulations yields a charge-carrier generation rate. Underneath the metal no optical generation is assumed. The generation profiles are averaged according to their area fractions. It should be noted that we assume an external mirror with ideal specular reflectance on the rear side in the simulations to represent a measuring chuck or white module backsheet. The effect of this mirror on the simulated cell efficiencies is comparable for all structures because they all have a textured front and a planar rear side. Note that while the area-averaging approach accounts for differences in front surface metallization and parasitic absorption in the poly-Si, it lacks the impact of multiple internal reflections at, e.g. metal contacts. However, we estimate the additional parasitic absorption at the contacts to be www.nature.com/scientificreports/ below 10% of the total parasitic absorption within the cell. In addition, all structures have similar light trapping schemes and, thus, no structure is optically favored.

Investigated solar cell structures
Most of the recently installed industrial solar cell production lines produce PERC and PERC+ solar cells, which have a high conversion efficiency and allow low production costs. However, the diffused contacts and front contact shading suggest an efficiency limitation at around 24% 19 . Figure 2a shows a synergistic efficiency gain analysis (SEGA) 20 of an industrial-type PERC+ cell with best-measured input parameters given below. It should be noted that we only consider busbar-less cells in this work. The SEGA yields potential efficiency gains in solar cells by switching off the respective loss channel in the cell simulation and calculating the difference in conversion efficiencies. Synergistic effects between multiple loss channels are analyzed by switching off multiple loss channels simultaneously. The simulated efficiency is 23.7%, which agrees with the efficiency to be expected from PERC+ production lines within the next years given today's efficiencies of around 22.5% and a 0.5% per year learning curve according to Ref. 21 . The efficiency gains due to a reduction of the front finger shading and a reduction of recombination in the emitter and at the front contact are the largest individual gains possible besides the synergistic gain between the extrinsic recombination channels. Therefore, the Ag contacts and the phosphorus diffusion are the bottlenecks on the road towards higher efficiencies. Here n-type POLO junctions offer better electrical performance. The so far most common approach for their implementation is the industrialization of the TOPCon 8,12 cell structure. The corresponding SEGA is shown in Fig. 2b. For the input parameters assumed in this study, the resulting simulated efficiency is 24.4%, i.e., 0.7% higher than that for PERC+. This difference is mainly a consequence of the mitigation of the contact recombination at the electron-collecting contact, the lower J 0 of the B-doped front emitter, and the synergy between both loss channels. We acknowledge that our simulated efficiency for industrial TOPCon has already been surpassed experimentally: To our knowledge, the highest ISO 17025/IEC 6094 calibrated efficiency reported so far is 24.8% 9 . Obviously, it is possible to optimize single components further than assumed in Table 1. However, as discussed below, it still has to be shown that all possible measures for the optimization of a record cell (e.g. extremely deep emitters) are compatible with mass production. The median efficiency in production is currently around 23% 8 . It is still not yet clear whether TOPCon is the best suited cell structure for an integration of POLO junctions into industrial production. Therefore, we investigate six alternative structures featuring POLO junctions and compare their performance potential and processing complexity with that of PERC+ cells and with a TOPCon 8,12 cell structure. It should be noted that these are cell structures we regard as interesting candidates for industrial integration after preliminary simulations and discussion, which are not shown here. There are, of course, many other possible cell structures and our list does not aim at being all-encompassing. In particular we focus our discussion on p-type doped Si solar cells, because it allows the re-utilization of established production steps. The six cell structures investigated are shown in Fig. 3 along with two reference structures. (a) the reference PERC+ cell, (b) the reference TOPCon cell based on n-type Si with a diffused boron emitter and a full-area n-type POLO junction on the rear, (c) the PERC+ POLO structure, which is a PERC+ cell with n-type POLO under the front contacts, (d) the POLO BJ  www.nature.com/scientificreports/ structure, which features Al-alloyed (Al-p+) contacts on the front and a full-area n-type POLO on the rear side 10 , (e) the POLO IBC structure, which is an IBC concept featuring Al-p+ and n-type POLO junctions 10,22 , (f) the POLO 2 BJ structure, which features local p-type POLO junctions on the front side and a full-area n-type POLO emitter on the rear 11 , (g) the POLO 2 IBC structure, which is an IBC concept with POLO junctions for both polarities 4 and (h) the n-Si POLO 2 BJ structure, which is the same structure as (f) but with reversed polarities. The six cell structures (c) through (f) feature either only one n-type poly-Si contact and an Al-p+ contact (POLO cell) or two POLO junctions ( POLO 2 cells) in either front junction (FJ), back junction (BJ) or interdigitated back contact (IBC) configuration. We acknowledge the pioneering work of other groups (including companies) on many of these or related cell structures. For example, although not all details have been disclosed, the concept of TetraSun 23 was reminiscent of the n-Si POLO 2 BJ structure, and SunPower′s Maxeon technology 7,24 seems to correspond to a n-Si POLO 2 IBC structure. It should be noted that for all structures either the structured layers underneath the contacts (selective emitter or POLO) are wider than the metal contact or the metal is wider than the LCO in case of Al contacts. This is due to alignment tolerances required for screen-printing, which we consider in our simulation study. The Ag contacts for which we assume a printing width of 30 µm are printed on 75 µm-wide stripes of either POLO layers or the laser-doped selective emitter. For the Al contacts we assume an LCO width of 13 µm and 50 µm-wide Al fingers. For all structures we individually optimize the front contact pitches. For the IBC cells we assume an emitter coverage of 80%. For the rear side, smaller contact pitches are in most cases preferable over larger pitches due to reduced resistance contributions. We, therefore, choose a rear contact pitch of 0.5 mm for all cells.

Electronic input parameters
The electronic input parameters chosen in this work are values measured mostly on reference samples manufactured with lab-type processes that can be transferred to industrial-scale production. Where possible we choose values measured at ISFH because this assures the best consistency between the different parameters. All input parameters are shown in Table 1. Values without a reference are unpublished ISFH internal measurements. At ISFH we measure the lifetime of non-metallized samples using a WCT-120 lifetime tester by Sinton instruments 32 and determine the surface recombination using the in-built Kane-Swanson 33 analysis tool. The recombination at metallized surfaces is determined by comparing samples with different metallization fractions and fitting the respective lifetime, measured by photoluminescence emission, using Quokka. Bulk and sheet resistances are determined using a four-point probe setup whereas contact resistivities are determined using the transfer-lengthmethod (TLM) 34,35 . It should be noted that some values differ from the ones used in our recently published paper in Ref. 11 , because they were either just recently measured or a publication exists for a similar value, which makes it easier for the reader to check the experimental details of the input parameters.
The boron-doped Czochralski-grown material that we use in our analysis was measured at ISFH with ρ b = 0.9 cm and mid-gap Shockley-Read-Hall lifetime parameters τ n0 = 2000 µs and τ p0 = 20,000 µs . The phosphorusdoped n-type material was also measured at ISFH with ρ b = 3 cm and an injection-independent SRH lifetime of τ b = 22,000 µs . For the surfaces passivated with an Al 2 O 3 /SiN stack we determine a J 0 of 1 fA /cm 2 on planar www.nature.com/scientificreports/ and 3 fA /cm 2 for the passivation on textured p-type material. For n-type material we assume a passivation quality equal to that on p-type. For the phosphorus-diffused surfaces we determine J 0 = 22 fA /cm 2 and R s = 133 �/ for the inter-finger region 25 . For the selective doping underneath the front contacts we determine J 0 = 100 fA /cm 2 and R s = 95 �/ . For the boron emitter in the TOPCon concept we take literature values for the shallow doped emitter 26 with J 0 = 14 fA /cm 2 and R s = 135 �/ and the selective emitter 27 with J 0 = 110 fA /cm 2 and R s = 95 �/ . For the planar n-and p-type POLO junctions we assume 200 nm thickness for the calculation of the optical performance. We assume this thickness to be required for a screen-printed POLO contact to retain its recombination properties. Lower thickness values are possible in combination with other metallization schemes that are not considered here. For planar n-type POLO layers we measure a J 0 of 3 fA /cm 2 and a sheet resistance of 50 �/ 22 . For p-type POLO layers we measure a J 0 of 5 fA /cm 2 and a sheet resistance of 200 �/ . We determine the contact recombination of Ag contacts on a P-diffusion to J 0 = 1400 fA /cm 2 in low injection. For the Ag on P contact we measure a contact resistivity of ρ c = 1.5 m cm 2 . For the Al-p+ contacts we determine J 0 = 400 fA /cm 2 and ρ c = 1.3 m cm 2 . For the Ag contact on boron diffusion we again refer to the literature with J 0 = 740 fA /cm 229 and a ρ c = 2 m cm 230 . For contacts on n-type POLO formed with firing-through pastes we observe that the contacts do not contribute to the recombination. These contacts were measured with a resistivity of ρ c = 0.9 m cm 2 . For p-type POLO the firing-through pastes yield high J 0 -values of around 250 fA /cm 2 as published by Mack et al. 36 . However, non-firing-through pastes on p-type POLO lead to insignificant recombination similar to n-type POLO contacts with a resistivity of ρ c = 5 m cm 2 as measured at ISFH 4,31 . Table 2 shows the simulated I-V parameters for each of the cell structures shown in Fig. 3. The results show clear trends in terms of increasing V OC and J SC : V OC increases with the reduction of recombination losses at conventional contacts and diffusions. Consequently, the POLO 2 concepts show the highest V OC followed by the POLO www.nature.com/scientificreports/ concepts. However, POLO layers always lead to parasitic optical absorption. Therefore, cell structures featuring structured POLO at the front show lower J SC than those without POLO junctions. In addition, POLO layers on the rear also contribute to the parasitic absorption, leading to further J SC reduction for concepts featuring fullarea rear side POLO layers. In the following discussion, we also comment on the bifaciality. For utility applications, it is mandatory to take this aspect into account when comparing different cell concepts on "equal footing". Please note that the projection of the achievable bifaciality values for all cell structures is difficult and depends strongly on e.g. future paste development (Al lines widths, mitigation of spiking through thin poly-Si layers, ...). Thus, all assessments of this aspect here are based on today's published values and our practical experiencedbased projection of the potential for further improvement. For the discussion whether a cell concept is industrially viable, technological challenges, process complexity and cost of ownership aspects (e.g. tool and material costs, waste management aspects, ...) are also relevant. It is hardly possible to compare all of these aspects for all structures on equal footing. There is a large variety of processing techniques, e.g. for poly-Si deposition and structuring, with specific advantages and disadvantages. These process techniques can, in principle, be combined to numerous process sequences for each cell concept. Some of them are less mature than others, and some of them have not even been demonstrated experimentally. We therefore refrain here from proposing hypothetical (and partially speculative) sequences, which could be misleading for the assessment of the above-mentioned practical issues. Rather, we remark for all cell structures the major possible or proven technological challenges that we are aware of after 10 years of PERC and after 7 years of industrial POLO development at ISFH. For all structures with POLO junctions (regardless whether for one or two polarities), growth of the interfacial oxide, poly-Si deposition and a certain post-deposition high-temperature anneal are mandatory. For interfacial oxide growth, several methods like ozone oxidation 37,38 , wet chemical oxidation 12,38,39 , short thermal oxidation or the Plasma Enhanced Chemical Vapor Deposition (PE-CVD) of SiO x 40 have been successfully demonstrated. The most viable option for industrial processing, e.g. with respect to throughput, reproducibility, homogeneity and robustness against contaminations, will have to be identified in pilot production. For the deposition of amorphous or polycrystalline Si, different methods like Low Pressure (LP-) CVD 38,39,41-43 , PE-CVD 12,44-48 , Atmospheric Pressure (AP-) CVD 49 , sputtering 50 and e-beam evaporation 51 have demonstrated excellent results. The process versatility of both, interfacial oxide growth and (poly-) Si deposition, hints to the robustness of the POLO concept. Apparently, the high temperature annealing reliably allows the system of crystalline Si, interfacial oxide and polycrystalline Si to re-configure into a low-energetic state. Regarding industrialization of the poly-Si deposition methods, the most recent ITRPV 52 gives some indications, but we think it should be considered with some caution because new deposition techniques like sputtering are emerging and different cell concepts may favor different deposition techniques. Technological challenges are, besides aspects like homogeneity of Si layer thickness and (if present) in-situ dopant distribution, the absence of blistering and in particular the mitigation of wrap-around-also for nominal single-side deposition techniques like PE-CVD and AP-CVD. Alternatively, the process sequence can be designed in a way that it tolerates or even utilizes double-sided Si deposition 11 . For the post deposition hightemperature treatment, quartz tube furnace anneals (either in inert or oxidizing ambient or within a POCl 3 44 or BBr 3 diffusion 45 ) are currently standard. For some cell concepts, they might be replaced by so called "firing-only contacts" 49,53 in the future, but the latter still have to demonstrate high selectivities (in particular low junction resistances), sufficiently low sheet resistances, compatibility with high-temperature screen-printing and so on.

Results and discussion
In the following, we discuss the different cell structures in detail: a) Our reference PERC+ cell shows a frontside efficiency of 23.7%. The SEGA is shown in Fig. 2. One should note that the bifaciality of PERC+ in the range of 80% 54 allows for a significantly increased energy yield (10-20% 55 ) for utility applications. This has to be taken into account when comparing the performance with other cell structures with higher or lower bifaciality.
From a technological point of view, a certain process sequence "standard" or "main route" has been established for PERC or PERC+ 21 . It comprises at least one high temperature step ( POCl 3 diffusion for introduction of n+type doping), while an ex-situ oxidation of the emitter is optional. Structuring for the selectively doped emitter is mainly performed via laser-doping from the phosphor silicate glass, which is an inexpensive process without the necessity of any waste management. A further technological advantage of the PERC+ structure is that Al alloying during firing forms the contact for the second polarity (holes) inexpensively. The rear side metallization furthermore only contains a limited amount of silver (for soldering pads, not for fingers). Thus, the PERC+ structure poses a tough benchmark (with respect to cost per watt, but also to Levelized Costs of Electricity) for any other cell concept. www.nature.com/scientificreports/ b) A widely evaluated concept is the TOPCon concept based on an n-type Si wafer with n-type POLO on the rear and a diffused boron emitter on the front. We include this concept as a second reference. In our simulation study, it yields a front-side efficiency potential of 24.4%, i.e., 0.7% higher than that of the PERC+ cell. The bifaciality of industrial TOPCon cells is reported to be in the range of 80 % 56 , i.e., comparable to that of PERC+.
In terms of technology, the process sequence for the fabrication of a TOPCon cell strongly depends on the availability of real single-side Si deposition techniques (without wrap-around) and on the availability of firingonly contacts. Without the latter, two high temperature processes-one for BBr 3 diffusion and one for POLO junction formation-are required. The reason is that the thermal budgets of both processes are typically quite different (unless "thick" interfacial oxide is used), and that the boron silicate glass and the boron doping from the rear has to be removed prior to the poly-Si deposition. Replacing the BBr 3 diffusion by single-side boron doping techniques like AP-CVD deposition of boron-doped glasses could eliminate the latter necessity. The boron doping profile can be driven-in several micrometers in order to mitigate contact recombination on the front -in expense of process time. Selectively doped boron emitters are less mature than their phosphorus-doped counterparts, and most work in this direction is so far based on etch-back approaches 57 rather than on laser doping. If the poly-Si deposition yields an intolerable wrap-around, the poly-Si has to be removed selectively from the already present boron front-side emitter. To enable this step, an alkaline-resistant etch-stop layer on the front side (deposited prior to poly-Si deposition) could be required. This layer could also serve as a diffusion barrier when doping the poly-Si on the rear in-situ via POCl 3 diffusion 8 . A cost disadvantage of the TOPCon structure-as compared to the PERC+ cell-is that the Ag consumption on the rear is increased due to the necessity to print Ag fingers for contacting the poly-Si without alloying through it. Besides cost aspects, it is worth to mention that contacting of n+-type poly-Si with firing-through Ag pastes without degrading the passivation quality works fairly well. Since there are controversial statements on cost perspectives for n-type wafers from the industry, it is difficult from our perspective to assess this issue. Also the relevance of the often cited higher robustness of n-type material with respect to some metal contaminations 58 is difficult to assess. Under "clean" process conditions, the bulk lifetime in p-type material (and, more important, the diffusion length) can be comparable to that of excellent n-type material (see our results above), but the conditions in industrial environment might be different 59 . Last but not least, the absence of Light Induced Degradation is currently an advantage of n-type material, which might vanish in the near future due to the availability of low [O i ] B-doped or of Ga-doped p-type material, as well as due to the readiness of industrial feasible permanent BO deactivation schemes 60 .
c) The PERC+ POLO cell replaces the selective emitter of the PERC+ cell with a local n-type POLO layer. This brings an efficiency gain of 0.3% due to a higher open-circuit voltage (+ 15 mV) and fill factor (+ 0.3%) because of the reduced recombination and contact resistance. This positive effect overcompensates the loss in the short-circuit current density ( − 0.4 mA /cm 2 ) due to absorption and reflection by the planar POLO layer in the finger regions.
In terms of technology, the process sequence for the fabrication of a PERC+ POLO cell is identical to that of a PERC+ cell for passivation and all steps afterwards. However, the necessity to structure the poly-Si on the front can imply a significant increase in process complexity. The straightforward approach adds the steps interfacial oxide growth, full-area deposition of poly-Si, print of an etch mask, poly-Si removal in the inter-finger regions, removal of the etch mask + cleaning and, for organic barriers, implies the necessity of waste treatment. Laser processing, e.g. laser oxidation 11 , might avoid the latter aspect and even the necessity to deposit or to print an etch barrier. The most attractive option would be to deposit the poly-Si already structured, e.g. by using shadow masks 61 , laser transfer techniques or by utilizing local print of Si inks or liquid Si 62 . All of these approaches are non-standard processes, which need development before becoming an option for industrial production. The rather small simulated efficiency gain of 0.3% questions the overall advantage of this cell structure.
d) The POLO BJ cell yields an efficiency potential that is 1.0% above the reference PERC+ process. The SEGA of this structure is shown in Fig. 4. The larger shading losses of 0.7% (J sc : −0.7 mA /cm 2 ) due to the 50 µm Al fingers are overcompensated by the gain in the open-circuit voltage (+ 39 mV) due to avoiding the recombination at the conventional P-diffusion and Ag contacts (0.45% and 0.35%, respectively). The disadvantage of larger shading will be reduced in the module, because a significant fraction of the light reflected from the finger will be redirected to the cell. Besides the shading loss, the resistance due to the large optimal finger spacing, and the synergies between the remaining recombination channels are the main bottlenecks of this structure. Regarding bifaciality, similar values as for TOPCon (i.e., 80%) are expected and we measured values in this range, which will be published separately.
In terms of technology, this cell can possibly be manufactured with a lean process flow since no structuring of the poly-Si is required and the hole contact is formed by Al alloying during firing (as for the PERC+ cell). The latter aspect also implies comparable metallization costs for the POLO BJ cell as for the PERC+ benchmark. A small wrap-around of nominal single-sided poly-Si deposition techniques might be less detrimental than for e.g. the TOPCon structure since the poly-Si will not be in contact with highly doped regions of the opposite polarity. If in-situ doped firing-only contacts are available and the bulk lifetime is already sufficiently high even without any gettering step, a process sequence without any high temperature step in a quartz tube furnace might be possible. Achieving or maintaining a sufficiently high bulk lifetime, which is essential for all cell concepts and in particular for all back-junction concepts, is one of the technological challenges. To estimate the impact of different final lifetimes we run a variation of the bulk lifetime for all cell concepts based on p-type Si, which we show below. Printing narrow Al contacts (here 50 µm width) with good electrical properties and aligning them with respect to the laser contact openings in industrial production is another challenge, and requires further development. It is interesting to note that Al fingers as narrow as 50 µm were recently reported 63,64 . A further challenge is the reliable and, under industrial environment (automated handling, no cleanroom atmosphere), robust passivation of a lightly doped textured surface. Finally yet importantly, the inter-connection of the cells for module integration has to be developed. Besides geometrical constraints for possible soldering pads on the www.nature.com/scientificreports/ front due to shading, there are constraints on the SiN x thickness in order to still fulfill antireflection purposes. Thus, metal spiking (i.e., by the Ag soldering pads) through the passivation layer cannot be mitigated by simply increasing the SiN x thickness (as on the rear of a PERC+ cell). The comparably high efficiency potential of the POLO BJ structure comes with a potentially lean process flow and thus makes managing the above mentioned technological challenges a worthwhile research target. e) Conceptually, the POLO IBC has the same advantages as the BJ version discussed above, yielding a high gain in open-circuit voltage of + 36 mV compared to PERC+. The SEGA for this structure is shown in Fig. 5. When comparing the SEGAs of the POLO BJ and IBC structure, we see that the loss due to shading (0.71%) is completely avoided in the IBC design. In contrast the loss due to recombination at the Al-p+ contacts is increased  www.nature.com/scientificreports/ because of the reduced contact spacing (+ 0.2%). The reduced contact spacing, however, also yields a reduction of the loss due to the lateral resistance between the contacts ( − 0.24% ). All effects combined yield an efficiency gain of 0.8% compared to the POLO BJ structure, which is 1.8% higher than for the reference PERC+ cell. The bifaciality of this cell structure can be expected to be slightly reduced from that of the abovementioned structures since both metal grids are on the rear. We expect a bifaciality factor of 70%.
In terms of technology, this cell structure also has the advantage of an inexpensive and elegant formation of the hole collecting contact by Al alloying. Unfortunately, it also combines challenges from the PERC+ POLO cell (necessity for structuring of the poly-Si) with challenges from the POLO BJ cell (requirement for high bulk lifetimes, passivation of lightly doped textured surfaces, avoidance of metal spiking through passivation layers, adapted cell interconnection scheme). Again, the increase of the SiN x thickness on the rear is not a perfect measure for the mitigation of metal spiking (e.g. from the soldering pads) since the Ag paste still has to fire through the stack in the emitter (n+ POLO) regions. Assuming that these technological challenges can be solved, the high efficiency potential of this structure can also be combined with a lean process flow 22 .
f) and g) The POLO IBC concept is limited to 25.5% mainly by the screen-printed Al-p+ contacts. The logical next step is thus to also replace the Al-p+ contacts by p-type POLO junctions, promising an efficiency gain of 0.38%. We study both, f) the POLO 2 BJ and g) the POLO 2 IBC cell as improvements over the POLO BJ and IBC cells, respectively. Both cells are identical to the POLO BJ and IBC concepts except for the base contacts, which are replaced by p-type POLO junctions. The POLO 2 BJ structure shows an efficiency gain of 1.3% in comparison to PERC+ and 0.3% in comparison with the POLO BJ concept. In comparison to PERC+ this efficiency gain originates mainly from the large voltage gain of +45 mV and in comparison to POLO BJ from the reduced shading and recombination at the Al-p+ contacts. The IBC version of the POLO 2 concept yields the highest efficiency in our comparison with 25.9%, which is another 0.4% higher than the POLO IBC concept by avoiding the recombination at the Al-p+ contacts and 2.2% higher than PERC+. The SEGA for the POLO 2 IBC cell is shown in Fig. 6. We see that the main losses are due to recombination within the bulk and at the textured front surface. A further efficiency increase requires, thus, extensive research on Si material with reduced SRH defects and better surface passivation. All efficiency gains yield a final efficiency of 27.7%. The remaining margin to the efficiency limit of 29.6% 65 is due to parasitic absorption, front surface reflection and imperfect light trapping, which do not show up in our analyses. Also the cell thickness and doping do not have the values required for the 29.6% limit. Bifaciality is expected to be lowest for POLO 2 IBC (60%), since the rear side of this structure features the metallization for both polarities as well as parasitically absorbing poly-Si on almost the entire area of the planar surface. Nevertheless, this cell structure will have the highest energy yield for any monofacial application such as rooftop, vehicle integrated PV and others.
In terms of technology, these POLO 2 structures add further challenges to those already mentioned for POLO IBC: the necessity to dope the poly-Si locally or at least single-sidedly, the necessity to contact p+-type poly-Si with high temperature screen-printing without alloying through it, and, for POLO 2 IBC, the necessity to electrically separate the n+-type and p+-type poly-Si regions [66][67][68] . For the first aspect, we proposed in Ref. 11 the overcompensation concept, which can be applied for both POLO 2 BJ and POLO 2 IBC. Alternative approaches would be the single-sided or even structured deposition of doping sources (e.g. boron or phosphorus containing glasses 69 , doping inks 70 , masked ion implantation 68,71 ) on initially intrinsic poly-Si, the single-sided or even www.nature.com/scientificreports/ structured deposition of in-situ doped poly-Si (structuring e.g. via shadow masks or inkjet printing of liquid silicon), or the extensive use of (structured) diffusion barriers. Many of the latter approaches are still in a very early stage of their development and have not proven their viability yet. For the second additional challenge (contacting of p+-type poly-Si via screen-printing), it will be essential to what extent the paste manufacturers can improve their products. We have reported our current status ( J 0 = 500 fA /cm 2 , ρ c = 5 m cm 2 ) in Ref. 11 , and these results are consistent with those of other groups 36 . Since it could be a fundamental aspect that a certain amount of Al in the paste is required to form a low contact resistance to p+-type (poly-) Si 72 while the Al always yields a certain degradation of the passivation quality, alternative metallization concepts like plating or Physical Vapor Deposition (PVD) might become more attractive for POLO 2 structures than they used to be for PERC+ cells. For the third additional challenge (separation of n+-type and p+-type poly-Si for POLO 2 IBC), different approaches can be used: oxidation of the poly-Si between the fingers 73 , introduction of a trench between the fingers 67,74-77 , or introduction of an initially intrinsic poly-Si region 4,66-68 . To summarize, both POLO 2 concepts are likely to require more time for industrial integration than the POLO BJ and IBC concepts. (h) The cell structures (c) through (g) discussed above use p-type Si wafers as does today's mainstream PERC+ technology. Switching to n-type material brings a slight efficiency advantage due to the higher lifetime: The n-Si POLO 2 BJ concept shows an efficiency potential of 25.1%, 0.1% higher than in the p-Si concept, that is otherwise identical. The benefit of integrating n-type Si into cell concepts is small due to the slightly higher recombination at the p-type POLO junction (5 fA /cm 2 compared to 3 fA /cm 2 for n-type POLO). In terms of technology, the challenges for n-Si POLO 2 BJ are-except for the fact that another passivation layer might be required for the lightly doped inter-finger regions on the front-comparable to those for the p-Si POLO 2 BJ cell.

Sensitivity of the efficiency to bulk lifetime variations
The bulk lifetime is an important parameter for finding the best cell type with POLO junctions for industrial production. In general, the bulk lifetime depends on the cell process due to gettering effects and defect kinetics. This is especially true for p-type Si wafers. We, therefore, vary the bulk lifetime of the various cell types to analyze the sensitivity of the efficiency to bulk lifetime variations. The cell geometries are those optimized for high lifetimes. The optimized front finger pitch, however, depends mainly on the trade-off between resistance contributions and shading. Therefore, the variation of the optimum finger pitch with varying lifetime is small. Figure 7 shows the efficiencies of all cell types as a function of a fixed, i.e. injection-independent, SRH bulk lifetime. The lifetime has only a minor effect on the efficiencies of the PERC+ and PERC+ POLO concepts of less than 0.5% for lifetimes between 500 µs and 10,000 µs . The benefit of integrating POLO into BJ or IBC p-type Si cells, however, depends strongly on the bulk lifetime. For a low lifetime of 500 µs the benefit of the POLO BJ and IBC are only 0.22% and 1.09%, respectively, and for the POLO 2 BJ and IBC concepts 0.48% and 1.26%, respectively. Compared to the expected efficiency gain at τ b = 10,000 µs , this is ca. 1.2% lower for the POLO concepts and 1.4% for the POLO 2 concepts and probably hardly justifies the increased process complexity. Thus, the economic feasibility relies on good quality Si and efficient gettering. In this context, it is interesting, that POLO layers also serve as gettering layers in high temperature steps 78 .

Roadmap for future cell development
The results presented above can serve as a roadmap for a step by step improvement of crystalline Si solar cells using POLO junctions. Figure 8 shows this roadmap. The POLO BJ and IBC concept show a high efficiency gain of 1.0% and 1.8%, respectively, provided the high lifetimes measured on test structures can be transferred to the finished cell. Both concepts have potentially lean process flows with no and with one structuring process for the  81,82 . This V OC is still below the simulated value (due to a non-optimized Al-p+ contact), but already the highest experimental V OC reported so far for p-type solar cells with Al-p+ contacts. We think that these first experimental achievements are promising and show that our simulation-based roadmap is reasonable. A promising next development target are industrial-type POLO 2 cells based on p-type Si both as IBC and BJ structure. We already fabricated a POLO 2 IBC cell with 26.1% efficiency 77 using lab-type processes and evaporated contacts. The POLO 2 concepts are ultimately limited by their optical performance and material quality and of course the intrinsic transmission and thermalization losses. To address the latter loss channel, a further valuable research goal is the development of perovskite tandem cells with the Si POLO 2 concepts as bottom cells for 2-and 3-terminal devices. The analysis of these tandem cells is not part of this study but we estimate an efficiency of above 33% by adding an approximate 21% efficiency for the top and half of the efficiency simulated here for the bottom cell. The roadmap shown in Fig. 8 is one possibility for approaching highest efficiencies. It is, however, also possible to directly develop POLO 2 structures or implement tandem cells in earlier development stages like the POLO concepts or even at the current PERC+ state.

Conclusion
In the simulation study presented here we analyzed various cell structures for their suitability for industrial integration. We simulated eight cell types: a PERC+ reference, a TOPCon reference, three cell concepts with n-type POLO and Al base contacts, and three cell concepts with POLO junctions for both polarities, two of which are based on p-and one based on n-type Si. The input parameters are chosen to represent measured best-case parameters achieved with processes relevant to industrial production. We simulated an efficiency gain compared to the reference PERC+ cell of 1.0% and 1.8% for the POLO BJ and IBC concepts, respectively, which feature n-type POLO emitters and Al base contacts. We further simulated an efficiency gain of 1.3% and 2.2% for the POLO 2 BJ and IBC concepts, respectively. We also analyzed the concepts based on p-type Si for their sensitivity to the bulk lifetime showing a 1.2% efficiency gain difference for the POLO concepts and 1.4% for the POLO 2 concepts for SRH bulk lifetimes ranging from 500 to 10,000 µs . For the integration of POLO junctions into industrial production in the near future the POLO BJ and IBC concepts appear attractive. Both utilize the base contacts and passivation surfaces from PERC+ and can, therefore, build on process steps from existing production lines. The next development step is to replace the Al base contacts with p-type POLO junctions also in either Figure 8. Roadmap for further cell development: Starting from our PERC+ cells we focus on the POLO IBC structure for industrial Si cells in the near future. We also exploit the large synergy between the POLO IBC and BJ concepts to also realize the BJ concept as it has the benefit of being compatible with conventional cell interconnection. As a further development step we work on the POLO 2 concepts, which require development in terms of POLO structuring and screen-printing on p-type poly-Si. The development step after the POLO 2 concepts is to combine the POLO 2 as bottom cells to perovskite tandem cells. www.nature.com/scientificreports/ BJ or IBC configuration. Even though the industrial integration of these concepts may be a few years down the road, the building blocks for their realization are currently under development 11,28 . Perovskite-Si tandem cells are still in a research state where their economic feasibility is not yet proven. Nevertheless, perovskite-Si tandem cells using the Si POLO 2 as bottom cells are an option for further improvement after the power losses in Si cells approach the intrinsic limit.