Influence of firing temperature and silver–aluminum paste intermixing on front contact quality and performance of TOPCon silicon solar cells

Contents lists available at ScienceDirect Solar Energy Materials and Solar Cells journal homepage: www.elsevier.com/locate/solmat Influence of firing temperature and silver–aluminum paste intermixing on front contact quality and performance of TOPCon silicon solar cells Martin Krejcia, Julien Hurnib, Ezgi Gençb, Jaroslav Čechc, Jaroslav Kuličekd, Egor Ukraintsevd, Petr Haušildc, Bohuslav Rezekd, Franz-Josef Haugb,∗ aUniversity of Applied Sciences and Arts Northwestern Switzerland (FHNW), Institute of Mathematics and Natural Sciences, Klosterzelgstrasse 2, 5210 Windisch, Switzerland bEcole Polytechnique Fédérale de Lausanne (EPFL), PV-Lab, Maladière 71b, 2000 Neuchâtel, Switzerland cFaculty of Nuclear Sciences and Physical Engineering, Czech Technical University in Prague (CTU), Trojanova 339/13, 120 00 Prague 2, Czech Republic dFaculty of Electrical Engineering, Czech Technical University in Prague (CTU), Technická 2, 166 27 Prague 6, Czech Republic A R T I C L E I N F O Dataset link:10.5281/zenodo.14967307, 10.52 81/zenodo.17150543 Keywords: Silicon solar cell Passivating contacts Contact formation A B S T R A C T We present a microstructural analysis of the front metallization in TOPCon solar cells by using scanning electron microscopy, Raman spectroscopy, Kelvin probe microscopy and atomic force microscopy. Correlative imaging at the cross section of front contacts interface with secondary electrons and with back-scattered electrons is a suitable method to identify contact formation in small localized areas where silicon intermixes with elements from the metallization paste. Three different types of intermixing have been identified; two shallow types are located either in the valley between the pyramids of the surface texture or on their facets near the tips. For contacts formed at higher temperature we detected sporadically a third type whose depth typically exceeds one micrometer. Likely this type of contact pierces through the emitter region, leading to losses in the open circuit voltage by creating shunts across the 𝑝-𝑛 junction of the cell. By statistically evaluating the dimensions of the intermixed areas we estimate electrical contact area fractions of 0.3% and 1 % for a firing temperatures of 780 ◦C and 840 ◦C, respectively. These area fractions are consistent with discrepancies between our measured values for contact resistivity and reported data for ideal full area contacts. 1. Introduction The global photovoltaic market currently undergoes a change of its leading technology away from the 𝑝-type passivated emitter and rear cell (PERC) towards 𝑛-type cells with tunnel oxide passivating contacts (TOPCon) at the rear [1]. In the latter design, a 𝑝-𝑛 junction is created at the front by boron diffusion. After the deposition of a dielectric SiNx/AlOx anti-reflection coating, a contact grid is applied by screen-printing of an Ag/Al paste, and contact to the emitter is established by applying a rapid thermal annealing process called firing. Despite its ubiquitous application in research as well as in industrial manufacturing, the mechanisms underlying the actual contact formation are little understood. Whereas Ag is a preferred material for the metallization because of its high intrinsic conductivity and its excellent solderability, it has a comparatively high effective Schottky barrier of 0.6 eV to 𝑝-type silicon [2]. The addition of Al was reported to reduce the specific contact resistivity [3–6], but Al can significantly increase the bulk resistivity of the printed metallization [6]. Moreover, Al melts during the firing process and readily dissolves Si into the melt, ∗Corresponding author. E-mail address: [email protected] (F.-J. Haug). resulting in deep pyramidal pits that are filled with metal during the cool-down [7–9]. Since such spikes can easily shunt the junction, Si is often added in the printing paste to saturate the melt [8]. A different mitigation strategy is to reduce the Al content in the printing paste in combination with laser enhanced contact optimization (LECO) [10]. Glass-frit is added to the printing pastes to etch through the dielectric passivation layers on top of the emitter. After melting and solidification, this can leave behind a thin insulating layer between the sintered metallization paste and the highly doped emitter [11]. In this contribution we combine scanning electron microscopy (SEM) with energy dispersive X-ray spectroscopy (EDS), Raman mapping, atomic force microscopy (AFM) and Kelvin probe force microscopy (KPFM) to investigate the formation of the contact between diffused boron emitters of varying depth and an Ag–Al screen printing paste. A variation of the contact firing temperature between 780 ◦Cto 840 ◦C allows us to study a transition from underfired contacts with high contact resistivity to overfired ones where even deep junctions become shunted. We find that contact between the metallization and https://doi.org/10.1016/j.solmat.2025.114085 Received 21 September 2025; Received in revised form 24 November 2025; Accepted 24 November 2025 Solar Energy Materials & Solar Cells 296 (2026) 114085 Available online 25 November 2025 0927-0248/© 2025 Published by Elsevier B.V. M. Krejci et al. Fig. 1. Process flow of co-annealed TOPCon solar cells, combining a variation of the boron content in the diffusion source layer and the peak temperature during the firing step of the metallization [12]. the underlying boron-diffused wafer surface is formed locally in small regions where elements from the paste intermix with the wafer. The intermixed regions can be classified into three types; two of them are comparatively narrow and located either on the sides near the peaks of the pyramids, or in the valleys between the pyramids. The third type is found in the valleys, but extends over a significantly larger region, and it is formed only at high firing temperature. For the first two types, Si and Ag are separated by a concentration peak of Al and N, whereas the third type shows a direct transition between Si and the Ag–Al metallization. 2. Experimental 2.1. Sample preparation 2.1.1. Fabrication of solar cells Solar cells were fabricated on 100-oriented float zone (FZ) 𝑛-type crystalline silicon (c-Si) with a resistivity of 2 Ω cm as discussed elsewhere [12]. We used single-side textured wafers with diameter of 4" and thickness of 200 μm. After standard RCA cleaning, both sides of the wafer were exposed to 1 min of UV-O3 to grow the interfacial tunnel oxide. The flat rear side was additionally treated for 1 min with N2O plasma in a parallel plate reactor for plasma enhanced chemical vapor deposition (PECVD) at 200 ◦C with an excitation frequency of 40 MHz (KAI-M, Unaxis). Next, the phosphorus-doped 𝑛-type layer was deposited by PECVD in the same reactor, using a precursor gas mix of SiH4, H2, and PH3. A small flow of CH4 was added to enhance adhesion and chemical resistance, resulting in a carbon content of approximately 2–3% [13,14]. On the textured front side, 𝑝-type SiOx layers were deposited by PECVD, using a precursor gas mix of SiH4, CO2, H2 and varying flows of Trimethylborane (TMB). After annealing at 900 ◦C for 1 h in a tube furnace (PEO 603, ATV), the boron-doped SiOx was etched from the front side and the surface was prepared for the growth of AlOx by dipping in ozonated DI-water and HF. Subsequently, AlOx was grown on the front by ALD, and SiNx layers were deposited by PECVD on both sides (SiNA, Meyer-Burger). Next, a grid of Ag/Al metal fingers was screen printed at the front to define five cell areas of 2 cm 𝑥 2 cm, and samples were fired at varying temperatures between 780 ◦Cto 840 ◦C in a firing furnace with ceramic rollers (Camini, Meyer-Burger). For the current study we focused on the extreme conditions as shown in Table 1. The rear side metallization was completed by removing the SiNx and sputtering of indium tin oxide (ITO) and Ag as discussed in Ref. [12]. Finally, a low-temperature Ag paste was used to print busbars on the front, followed by curing at 210 ◦C for 30 min. Fig. 1 illustrates the process flow of co-annealed TOPCon solar cells. Table 1 Samples which were characterized in detail. Sample TMB flow Firing temperature #1 30 sccm 780 ◦C #2 90 sccm 780 ◦C #3 30 sccm 840 ◦C #4 50 sccm 840 ◦C #5 90 sccm 840 ◦C 2.1.2. Preparation of cross section samples Cross-sectional samples were prepared to examine the interface between the Ag/Al metal fingers and the emitter region. For this purpose, the solar cells were scribed and cleaved in a perpendicular direction to the fingers (c.f. SI). Because soft silver paste is plastically deformed during cleaving, there was a significant risk that overhanging Ag/Al paste would cover parts of the area to be examined. For this reason, substantial effort was invested in the development of the scribe and cleave process with minimized paste overhang. In the first step of the optimized process, an approximately 3 mm long scribe was created on the top side of the cell using a 1064 nm marking laser, taking care that the scribe ran precisely along a ⟨100⟩ direction of the silicon crystal. In the second step, the sample was placed on a soft foam mat with the top side facing down and cleaved. To do this, the round end of tweezers was gently pressed above the cleave against the back side of the cell. This procedure resulted in uniform bending of the sample and smooth cleave propagation along a crystal orientation. Because only a small force had to be applied to form the cleave, the plastic deformation of the silver paste and the resulting overhang were less than if the sample was broken with a large force, for example, over the edge of a glass substrate. The scribe and cleave process is illustrated in Fig. 2. In order to reduce the plasticity of the silver paste, some samples were shock cooled in liquid nitrogen prior to cleaving. This step appeared to further reduce the overhang of the paste, but the effect was not investigated in detail. The systematic execution of the optimized scribe and cleave process resulted in only 5% of the examined area being affected by an overhang. 2.2. Sample characterization 2.2.1. Electrical characterization The boron concentration profiles of the emitters (c.f. Fig. SI3) were investigated by electrochemical capacitance voltage profiling (ECV), using 0.1 M ammonium fluoride (CVP21, WEP). The contact resistivity was investigated by the transfer length method (TLM) The effective Solar Energy Materials and Solar Cells 296 (2026) 114085 2 M. Krejci et al. Fig. 2. Illustration of the optimized scribe and cleave process: (a) optical microscopy image of laser scribe, (b) schematic drawing of cleave process, (c) SEM image of cleaved facet. minority carrier lifetime and the implied open circuit voltage (𝑖𝑉𝑜𝑐 ) was assessed at several steps of the fabrication sequence by measuring the photoconductance decay (WCT120, Sinton Technologies). Uniformity was assessed by photoluminescence imaging using laser illumination with a wavelength of 808 nm (Ostech, intensity equivalent to 1.2 suns) and recording with a silicon-based camera (PIXIS, Princeton Instruments) through a 850 nm long-pass filter. The performance of the solar cells was assessed by measuring their current voltage characteristics under standard test conditions. The cells were illuminated by a class AAA solar simulator (Wacom) and the current was measured at each step of a voltage sweep, using a source meter (Keithley-2601 A). From these measurements, we obtained the short-circuit current density (𝑗𝑠𝑐 ) by dividing with the cell area, the open-circuit voltage (𝑉𝑜𝑐 ), the fill factor (𝐹𝐹 ), and the power-conversion efficiency (PCE). 2.2.2. Scanning electron microscopy and atomic force microscopy SEM measurements, which served to statistically analyze locations in which intermixing of material from the metal finger into the semiconductor had occurred during the firing process, were carried out in secondary electron (SE) and backscattered electron (BSE) regimes in ZEISS EVO 10 microscope. The best results were obtained at 10 kV accelerating voltage and 50 pA probe current. The SEM was equipped with a Litescope Atomic Force Microscope (AFM; NenoVision) module, which is capable of performing Correlative Probe and Electron Microscopy (CPEM). In this study, the method was used to analyze a relationship between AFM surface topography of the intermixed regions at the front contact interface, their local electrical surface potential obtained in Kelvin Probe Force Microscopy (KPFM) regime, and their correlation with contrasts visible in SEM images [15]. The images of 30 ×30 μm2 were obtained at 1024 × 1024 px2 resolution in 12.5 h. SEM data were obtained in-situ but in a separate measurement from AFM and KPFM data because the employed conductive NenoProbe Conductive (NPC) cantilever is not a nose type cantilever and thus minimal offset between NPC cantilever probe apex and electron beam is 35 μm. By KPFM we analyzed the relative surface potential profiles across the junction. Thus, an absolute KPFM cantilever work function calibration on a reference metal sample was not needed. We tested the correct KPFM operation and reproducibility directly on the doped silicon substrate. Topography AFM image was used as a reference to modify the distortion, offset, rotation and scale of SEM data to make a correlative 3D CPEM image [16]. In the CPEM image the 3D topography data corresponds to height and geometrical shapes of structural features whereas colors correspond to the intensity of SE or BSE signal from SEM. EDS investigations in the scanning electron microscope of different areas of selected samples were performed on a Jeol JSM-IT500HR SEM with field emission gun equipped with an EDAX Octane Super Elite EDS system at 7 kV accelerating voltage. On the one hand, the investigations were used to confirm and substantiate the comparison between the SE and BSE signals and, on the other hand, to draw conclusions about the intermixing of chemical elements from the paste into the semiconductor. 2.2.3. Raman spectroscopy Raman Spectroscopy was applied to identify interface regions below the finger, for which the SiNx/AlOx layer was not etched by the glass frit during the firing process. The measurements were performed using a WITec alpha300 RAS Raman spectromicroscope equipped with a single-mode laser, emitting 2 mW power at a center wavelength of 532 nm. The Raman spectra were established by an objective with a magnification of 100× and a numerical aperture of 0.9 and a UHTS 300 VIS spectrometer. The integration time was 30 s and each scan was repeated three times resulting in a measurement time of 1.5 h per finger. The spectra were collected and processed by the WITec Control 6.1 and WITec Project 6.1 software. An X-Y piezo-flexure stage was used for sample lateral scanning to obtain Raman spectral maps of a 60 × 15μm2 large area. The scan speed was 30 s/line, resulting in an image resolution of 150 × 150 pixels. 3. Results and discussion 3.1. Electrical characterization Fig. 3 summarizes the main solar cell performance parameters with respect to the TMB flow used for the deposition of SiOx dopant source layer and the firing temperature used during contact sintering. Overall, the power conversion efficiency (PCE) follows most closely the behavior of the 𝐹𝐹 which increases with increasing TMB flow as well as with increasing firing temperature. However, there is a trade-off with losses in both 𝑉𝑜𝑐 and 𝑗𝑠𝑐 for the highest TMB flow, such that the highest device efficiency of 21% is observed for the intermediate TMB flow of 50 sccm. A more detailed discussion of the solar cell results can be found elsewhere [12], but we note that the behavior of the 𝐹𝐹 is dominated by the contact resistivity between the boron-diffused emitter and the Ag–Al metallization. Fig. 4 shows the contact resistivity with respect to the peak of the active doping concentration of our boron-diffused emitter. We include reported data for similar contact structures [17] as well as data for pure Al-Si(p) Schottky junctions [18–21]. For the shown range of doping concentrations we are aware of only one report on Ag-Si(p) Solar Energy Materials and Solar Cells 296 (2026) 114085 3 M. Krejci et al. Fig. 3. Power conversion efficiency (PCE), fill factor (𝐹𝐹 ), open circuit voltage (𝑉𝑜𝑐 ) and short circuit current density (𝑗𝑠𝑐 ) of the solar cells using the front contacts analyzed in this contribution. junctions [22], but judging from their higher effective Schottky barrier, we can expect that the contact resistivities would fall in a similarly broad band about one order of magnitude higher than the Al-Si(p) contacts. The contact resistivities of the screen-printed Ag–Al pastes show a decrease with peak doping concentration (c.f. Fig. SI3), and in each case there is an additional decrease with increasing firing temperature, but in all cases they are much two to three orders of magnitude higher than the pure Schottky junctions. The discrepancy was explained by two geometrical effects [9]; first, the area under the screen-printed metallization is largely covered by insulating glass-frit whereas the actual contact between the metal and the semiconductor forms in a small fraction of the metallized area. In these regions the contact areas assume the shape of inverted pyramids that are filled by the metal. Second, as the metal-filled pyramids extend into the borondiffusion profile, only a small region close to the base is in contact with high doping concentrations whereas most of the facet areas and especially the tip region is in contact with the lowly doped tail of the boron-diffusion profile. Higher firing temperatures can yield deeper etch-pits and thus larger contact areas between the metallization and the boron emitter, resulting in the lower contact resistivities shown in Fig. 4. At the same time, there is a risk that the tip of pit pierces through the emitter and yields a short-circuit of the junction. Likely this explains the degradation between the 𝑖𝑉𝑜𝑐 before metallization (c.f. Fig. SI4) and the actual solar cell 𝑉𝑜𝑐 . In case of low TMB flow this degradation is most severe since this condition yields the shallowest junctions. For TMB flows of 50 and 90 sccm the junctions are increasingly deeper and 𝑉𝑜𝑐 loss with increasing firing temperature is no longer observed. However, excessively high dopant concentrations in the case of 90 sccm reduce the 𝑉𝑜𝑐 due to increased surface recombination at the interface with the AlOx passivation layer and also due to increased Auger recombination in the emitter itself. Fig. 4. Contact resistivities of our samples (box plots); for clarity each group was split laterally around its respective peak doping concentration. Open circles denote reported data for firing at set temperatures of 750 and 770 ◦C [17]. Filled symbols represent literature data for pure Schottky junctions of Al-Si(p) (squares [18], circles [19], up triangles [20], down triangles [21]), and of Ag-Si(p) (diamonds, [22]). 3.2. Identification of intermixed regions The main method to identify intermixed locations at the emitter contact surface was the comparison of SE and BSE images. This procedure is illustrated in Fig. 5. In the BSE image of Fig. 5(a), the intermixed region, indicated in the image by an arrow, is visible as a result of the material contrast provided by the method. In comparison, the SE image in Fig. 5(b) exhibits little or no contrast in the intermixed region, because the formation of secondary electrons in the sample depends Solar Energy Materials and Solar Cells 296 (2026) 114085 4 M. Krejci et al. Fig. 5. (a) BSE and (b) SE image of an intermixed region. Fig. 6. CPEM investigation, (a) SEM image of the AFM cantilever at a region of interest, (b) SEM image with an overhanging feature marked by a circle, (c) KPFM and (d) AFM 2D scan of the region of interest, (e) combination of SE image and AFM topography information. much less on atomic mass. The difference in contrast between the SE and BSE imaging methods was a fast method for identifying areas where the ohmic contact between the contact finger and the semiconductor had been established. CPEM studies supplied a correlation between topography and SEM contrast, which contributed to the understanding of the SEM images, as illustrated in Fig. 6. The SE image displayed in Fig. 6(a) shows the AFM cantilever positioned above a cleaved examination area of a sample with 840 ◦C firing temperature, illustrating the CPEM method. Fig. 6(b) shows the SE image of the cleaved surface, Figs. 6(c) and 6(d) show the KPFM potential and the topography, respectively. Based on the SE contrast alone, the material properties of the surface cannot be distinguished from topographic structures. In particular, the nature of the bright contrast circled in red cannot be conclusively determined from the SE image alone. Only the additional AFM examination of the same region provides the missing information on the correlation between SE contrast and topography. For example, the combination of the SE contrast and the AFM topography shown in Fig. 6(e) confirms that the contrast circled in red is a surface structure which was probably created as an artifact during cleaving. As expected from the nature of the cleaving process, a flat semiconductor surface and plastically deformed silver paste are observed. This topography contrast was used to identify the location of the interface between emitter and contact finger. The color range of the KPFM map was set from 0.2 V to 1 V with the intention of illustrating the potential differences in the interfacial region. The results of the KPFM investigation is discussed in Section 3.2.2 based on linescans evaluated at three typical regions. The dissolution of the SiNx/AlOx anti-reflection coating and the intermixing of the metallization paste and Si was further examined by Energy Dispersive X-ray Spectroscopy and Raman Spectroscopy on selected samples. The motivation for the investigations was to confirm that the contrast difference in the SE and BSE images can be explained by intermixing. Raman Spectroscopy imaging potentially provides a simple way to detect areas of a sample in which the antireflective coating has been dissolved during the firing process. 3.2.1. Energy dispersive X-ray spectroscopy Reference areas were examined for the evaluation of the EDS investigations on chemical intermixing at the interface between the silicon absorber and the metal contact finger. As shown in the SEM images of Fig. 7(a), the regions were selected according to the criterion that a distinct and sharp boundary layer was observed between the emitter and the metal finger. The yellow line in the figure illustrates the EDS line scan that was performed across the interface, and the rectangle around the line indicates the spatial resolution of the investigation, which was estimated to be approximately 0.4 μm to 0.6 μm based on the spot size of the electron beam, the applied accelerating voltage and the sample chemistry. The EDS line scan is shown in Fig. 7(b). Because the EDS examination of an inhomogeneous sample involves a high degree of uncertainty when making a statement about the absolute atomic concentration, the raw data (Counts) are discussed in the following. Striking are the three regions 1, 2 and 3. Adjacent to the silicon emitter is the thin anti-reflection coating consisting of an aluminum oxide (1a) and a silicon nitride (1b) layer. Then comes a region in which the signals of the elements boron, silicon, lead, oxygen, and nitrogen are approximately constant (2); presumably this is a region of borosilicate material used in the glass-frit. The adjacent region (3) consists mainly of silver, which partially penetrates the borosilicate region. The behavior of aluminum is as follows: At the interface between silver and the borosilicate, the concentration is maximal and decreases toward the Solar Energy Materials and Solar Cells 296 (2026) 114085 5 M. Krejci et al. Fig. 7. (a) BSE and SE image of a region used as reference for the EDS measurement, (b) EDS line scan of the reference region, (c) schematic drawing of the layer stack, (d) 𝑅=0.5 μm wide blurring caused by the limited spatial EDS resolution. anti-reflective layer. A schematic drawing of the layer stack is shown in Fig. 7(c). As a consequence of the limited spatial resolution of the EDS measurement, transitions at sharp interfaces are smeared, falsely suggesting the intermixing at the interface. Figure Fig. 7(d) shows a magnification of the line scan of the reference sample at the interface between the emitter and the metal finger. The observed interfacial blurring, 𝑅≈ 0.5 μm wide, corresponds to the estimated resolution limit of the EDS investigation method obtained by Castaing’s formula. In the following, this value will serve as a reference for the evaluation of the intermixing of elements from the metal finger at the silicon surface. To determine whether the EDS signals, which are framed in black in the border area of the silicon emitter in Fig. 7(b), are related to the presence of paste material that intermixed with the semiconductor, or whether the signals are background noise, EDS point measurements were carried out at several locations of the emitter. The spectrum shown in Fig. 8(a) was measured at a location 0.7 μm away from the boundary layer. For all elements except silicon, oxygen, and carbon, only a background signal was detected at this location. Because the cleaved surface is oxidized and carbon is deposited on the sample surface during SEM examination, no statements can be made about the significance of the oxygen and carbon peak. For the other elements in the paste, namely lead, nitrogen, aluminum, and silver, the concentration in the emitter is below the detection limit of the EDS measuring system. The EDS spectrum of an intermixed regions, recorded about 400 nm below the interface between metal finger and p type emitter is shown in 8(b). The comparison between the spectra (a) and (b) suggests the presence of the chemical elements Ag, Al, N, O and Pb in the investigated region. Because the spatial resolution of the EDS measurements was limited to about 0.5 μm a part of the EDS signal in 8(b) was likely generated in a region outside of the emitter. The sample locations at which intermixing was suspected based on the comparison between SE and BSE measurements can be categorized into three types, which differ in size, shape, and location at the interface between the metal finger and the emitter. For the first type, intermixing takes place in the valley of the pyramid-shaped emitter surface, as illustrated by the SEM images in Fig. 9(a). The circled and bright BSE contrast suggests that the intermixed region is also pyramidal. The red arrow indicates the length 𝓁= 0.8 μm of the intermixed section at the interface along the cleave. Again, the yellow line and the dashed Fig. 8. EDS spectrum of (a) a reference region recorded 0.7 μm inside of silicon and (b) and intermixed region. box in the SE image indicate the location of the EDS line scan and the approximate local blurring of the scan, respectively. The measured values of the scan are shown in the diagram in Fig. 9(b). The two dashed lines with distance 𝑅= 0.5 μm indicate the uncertainty of the boundary region of the reference location and serve to assess whether the elements of the metal finger intermixed with the underlying silicon. Compared with the scan in Fig. 7(d) of the reference location, the two peaks 1a and 1b of the anti-reflection coating have disappeared. Instead, we find that the signals of aluminum and nitrogen peak at the same position just outside the blurred zone. Earlier work found that during firing liquid Al reacts readily with SiNx. The glass frit Solar Energy Materials and Solar Cells 296 (2026) 114085 6 M. Krejci et al. Fig. 9. (a) BSE and SE image of a pyramid shaped intermixed region, (b) EDS line scan across the intermixed region. Fig. 10. (a) BSE and SE image of a hill shaped intermixed region, (b) EDS line scan across the intermixed region. and an underlying film of native SiOx are sufficient to confine the reaction to the thickness of the SiNx layer, but it can proceed for several micrometers within the film, forming an Al-N compound largely free of Si. [23,24] . The second type of intermixing occurs on protrusions of the semiconductor surface, which were created during the firing process as a result of etched pyramid tips. An example of this intermixing type is shown in the SEM images in Fig. 10(a). Unlike the intermixed regios at the bottom of a valley, in this case the intermixed region appears wide and shallow, as illustrated by the circled area of the BSE image. Consequently, the length 𝓁= 1.4 μm of the section along the cleave is larger than for the section in Fig. 9(a). The intermixing behavior can be assessed on the basis of the EDS line scan in Fig. 10(b). As in Fig. 9(b) the peaks associated to the anti-reflection coating are missing and also in this case the signals of Al and 𝑁 peak simultaneously at the outer edge of the uncertainty region 𝑅. An example of the third type of intermixing is illustrated by the circled bright area of the BSE image in Fig. 11(a) As with type 1, the intermixing occurs near the bottom of a pyramidal valley, but the length 𝓁≈ 1.7 μm and depth of the intermixed region is significantly greater. The intermixing behavior illustrated by the EDS line scan of Fig. 11(b) appears to be different from that of the other two types. The presence of silver is most pronounced, followed by that of aluminum. In contrast to the two other cases discussed, no significant nitrogen content was observed. This type of intermixing was observed only for samples fired at the highest temperature of 840 ◦C. 3.2.2. KPFM We used KPFM to investigate the variation of the electrostatic potential across the junction and to assess the impact of local intermixing. Fig. 12(a) shows a KPFM map of a region in which different intermixing behaviors at the interface were observed using SE and BSE investigations. The linescan profiles marked (1), (2), and (3) each run through a characteristic boundary region between the emitter and the contact finger. The contrast in the BSE images shown in Fig. 12(b) allowed us to identify (1) a region on the facet without intermixing similar to Fig. 7, (2) a region with intermixing at the tip of a pyramid similar to Fig. 10, and (3) a region with deep intermixing in the valley between two pyramids similar to Fig. 11. The topography information of the profile shown in Fig. 12(c) was used to define the interface between emitter and finger as position 0. This position is marked in Fig. 12(a) by the yellow dashed line. For the linescans along (1) and (2), Fig. 12(d) suggests an increase of the KPFM potential across the junction from the 𝑛-type wafer towards the 𝑝-type emitter and a constant level across the interface with the Ag–Al metallization, whereas there is no appreciable variation along the linescan (3). Based on the contrast of the BSE image, we can assume that region (1) is a pristine 𝑝𝑛-junction without intermixing. From the dopant concentrations we would expect a built-in voltage in excess of 800 mV, but the measured KPFM potential varies by less than 100 mV. Recent KPFM investigations of the contact region in TOPCon solar cells report similar differences between the built-in voltage and measured values [25,26]. The discrepancy is often attributed to surface damage during preparation and polishing of cross section samples, but also cleaved surfaces like the ones investigated here form interfacial defects which tend to pin the Fermi level close to mid-gap [27]. Moreover, storage in ambient air grows a native oxide with positive fixed charge [28,29] which accumulates electrons to the surface and repels holes from away from it, in both cases lowering the work function by a downwards band bending. Invoking charge neutrality, we estimate that the work function is reduced by about 170 mV to 200 mV for the lowly doped 𝑛-type wafer, but only by about 15 mV to 20 mV for the highly doped 𝑝-type emitter. Nevertheless, surface band bending alone cannot fully explain the small extent of the KPFM potential variation across the junction. The BSE image suggests intermixing for both regions (2) and (3), but the variation of the KPFM potential in Fig. 12(c) shows that region (2) at the tip of the pyramid still forms the same built-in voltage as the reference region (1). Region (3) behaves very differently, showing virtually no potential variation. We can tentatively explain this Solar Energy Materials and Solar Cells 296 (2026) 114085 7 M. Krejci et al. Fig. 11. (a) BSE and SE image of a deep intermixed region, (b) EDS line scan across the intermixed region. Fig. 12. (a) KPFM map showing the locations (1), (2) and (3) of selected linescan profiles (b) BSE images of the corresponding linescan locations, (c) topography of the selected profiles, (d) KPFM potential of the selected profiles. observation by the properties of boron diffusion in the presence of surface texture [30,31]. Since the dimensions of the pyramids are larger than the depth of the diffusion profiles, diffusion is much deeper at the tip of the pyramids compared to the facets and especially the valleys between them. Thus we conclude that some intermixing can be tolerated around the tips of the pyramids without shunting of the junction, but the same depth of intermixing can easily create shunt paths through the shallower diffusion profiles at the bottom of the valleys between the pyramids. The analysis of the BSE images above showed that this type of deep intermixing was found only for the highest firing temperatures, corroborating the observation that in this case the losses in 𝑉𝑜𝑐 were the most pronounced. At this stage it remains unclear why the KPFM potential of line scan (3) is 0.06 V lower in the metalized region compared to line scans (1) and (2). A possible reason is an inhomogeneity in the Al composition in this case which could yield a different work function or a different oxide coverage. 3.2.3. Raman spectroscopy The comparison between the SE/BSE images and Raman maps of the Si and the SiNx vibration signal, all recorded in the same region of a sample with 780 ◦C firing temperature is shown in Fig. 13. The Raman Si map shows the intensity distribution measured for the signal at 520 cm−1, which can be attributed to crystalline silicon [32], while the region 800 −1000 cm−1 being used to map the broad band SiNx Raman signal [33]. The scale on the left side of the images relates the colors of the Raman maps to the signal intensity recorded by the CCD camera. The dark orange line in the Raman spectra shows the interfacial region between crystalline silicon and the glass frit obtained based on the Raman measurement. The circles in red indicate positions, at which intermixed regions have been identified, while an overhang of the glass frit covering the cleaved facet is located in the region of the yellow square. The regions for which intermixing was identified on the basis of the SE/BSE investigations also show a reduced or missing signal intensity in the SiNx Raman map. Fig. 14 shows the same comparison as Fig. 13, but for a sample with 840 ◦C firing temperature. The Raman SiNx map shows far less regions with high count numbers for the SiNx vibration signal than the Raman SiNx map of the 780 ◦C sample. This is in agreement with the expectation that the dilution of the SiNx/AlOx layer is more pronounced for a higher firing temperature. Same as for the 780 ◦C sample, the regions, for which intermixing was identified based on the SE and BSE contrast, exhibit in the SiNx map low count numbers. There are regions in both Figs. 13 and 14 with missing or reduced intensity in the Raman SiNx map, which do not exhibit a corresponding intermixing contrast in the SE/BSE images. A possible explanation for this discrepancy is that the intermixing of metal elements with silicon Solar Energy Materials and Solar Cells 296 (2026) 114085 8 M. Krejci et al. Fig. 13. Comparison between SE/BSE images and Raman maps of Si and SiNx Raman peaks for a region of a saple fired at 780 ◦C. Fig. 14. Comparison between SE/BSE images and Raman maps of Si and SiNx Raman peaks for a region of a sample fired at 840 ◦C. does not occur at all locations with a diluted SiNx/AlOx layer or that these are due to intermixed regions that are buried beneath the imaging plane by the 3D geometry of the pyramidal surface texture. Collection from buried regions could also explain why the SiNx layer thickness appears to exceed 2 μm whereas it is only a few tens of nanometers thin. Nevertheless, we are confident that the signal originates from the SiNx layer since Raman spectroscopy can detect signals even from features that are well below its optical resolution. For example, silicon nanocrystals up to 160 nm were detected in an amorphous silicon matrix [34]. Thus, Raman mapping of cross-section samples is a powerful tool to characterize disruptions in the SiNx layer, but due to the convolution of the layer geometry and the optical resolution of the Raman micro-spectrometer, or more generally with the instrument transfer function, it cannot inform about the true layer thickness. 3.3. Statistical evaluation For the purpose of explaining possible correlations between the electrical characteristics of the solar cells and the firing temperature, the intermixed locations identified and geometrically measured by comparing SE and BSE images were statistically analyzed. The relevant statistical parameter was the length 𝓁 of the intermixed interface section along the cleave, which is described in Section 3.2. The histograms shown in Fig. 15 illustrate the section length distribution for (a) the 780 ◦C and (b) the 840 ◦C samples. 0.2 μm long intervals were combined in a bin, starting at 𝓁= 0. Both distributions are asymmetric with a tail a higher length values. Therefore, we decided to calculate the median length value and the standard deviation assuming a log-normal distribution. The length values of the 840 ◦C samples are distributed around the median value of 0.98 μm with a standard deviation of 0.42 μm. For Fig. 15. Histograms of the lengths 𝓁 of the intermixed interface section visible in the BSE images for the (a) 780 ◦C samples and (b) 840 ◦C samples. the 780 ◦C samples, the median length distribution is comparatively lower at 0.81 μm um, with a slightly smaller standard deviation of 0.38 μm. A statistical summary of the detected intermixing locations is listed in Table 2. As shown in Table 1 the samples #1 and #2 were fired at 780 ◦C while the firing temperature of the samples #3, #4 and #5 was 840 ◦C. Within the two firing temperatures, the samples differ by the trimethylboron (TMB) flow, which has compared to the firing temperature only a slight influence on the intermixing between semiconductor and metal fingers. For the two 780 ◦C samples, 74 intermixing locations were identified in 27 investigated fingers, which is on average 2.74 locations per finger. The investigation of 44 fingers belonging to the three 840 ◦C samples resulted in 197 intermixed locations which corresponds to 4.48 locations per finger. The 840 ◦C firing temperature not only resulted in a 1.64 times higher concentration of intermixed locations but also in longer intermixed interface sections. The mean Solar Energy Materials and Solar Cells 296 (2026) 114085 9