Structural properties of inkjet-printed and ultrasound-spray-coated PEM fuel cell catalyst layers and their impact on fuel cell performance

Structural properties of inkjet-printed and ultrasound-spray-coated PEM fuel cell catalyst layers and their impact on fuel cell performance M. Hala a , M. Prokop a , P. Capek b , M. Vesely b , T. Jedlicka a , T. Zubkova c , K. Heinrich c , A. Willert c , R. Zichner c , K. Bouzek a,* a Department of Inorganic Technology, University of Chemistry and Technology Prague, Technicka 5, Prague 6, 166 28, Czech Republic b Department of Organic Technology, University of Chemistry and Technology Prague, Technicka 5, Prague 6, 166 28, Czech Republic c Fraunhofer Institute for Electronic Nano Systems, Technologie-Campus 3, 09126, Chemnitz, Germany ARTICLE INFO Keywords: PEM Fuel cell Inkjet printing Catalyst layer Electrical conductivity Permeability Porosity Tomography ABSTRACT Catalyst layers (CLs) are critical to the performance and durability of proton exchange membrane fuel cells (PEMFCs). As shown previously, inkjet printing (IJP) represents an attractive CL production technology ensuring efficient catalyst utilization. This study provides a comparative structural analysis of CLs fabricated via IJP and ultrasonic spray coating (USC), aiming to explain the performance differences observed in operating fuel cells utilizing these two types of CL. This allows further optimisation of the CL deposition by IJP. A combination of advanced experimental and modeling techniques was employed to accomplish this task, including in-plane electron conductivity measurements, optical profilometry, FIB-SEM tomography, and 3D structure-based transport simulations. IJP CLs were consistently thinner (4–10 μ m vs. 6–18 μ m), smoother (R a ~0.38–0.44 μ m vs. ~0.55–0.63 μ m), and exhibited significantly fewer surface cracks (0.18–0.61 % vs. 1.34–4.05 %) compared to USC layers. Despite similar porosities at the microscale (40.0 % for IJP vs. 38.3 % for USC), IJP layers showed higher electrical conductivity (448 ±135 S m −1 vs. 387 ±96 S m −1 ) and more homogeneous Pt distribution. FIB-SEM reconstructions confirmed isotropic and statistically homogeneous structures of CLs produced by both methods, with negligible isolated porosity and comparable transport properties. However, macroscale features such as crack formation, layer thickness, and surface roughness strongly impacted overall performance and Pt utilization. These results highlight the critical role of deposition method in determining catalyst layer architecture and reveal inkjet printing as a highly promising approach for producing low-loading, high-performance CLs with potential for scalable, additive manufacturing. List of symbols Symbol Units Description am Distance a between two adjacent lattice nodes Am 2 Electrode surface area ECSA m 2 g −1 Electrochemical surface area g ii 1 Effective electrical conductivity h μ m CL thickness IA Electrical current I (s) (x)1 Indicator function for the solid phase (continued on next column) (continued) I (v) (x)1 Indicator function for the void phase l 1 , l 2 , and l 3 1 Integers m A,Pt mg cm −2 Pt loading N1 Sampling area (number of pixels) P (v) (δ) μ m −1 Pore-size probability density function q H,ML C m −2 Charge of adsorbed hydrogen monolayer R a μ m Arithmetic mean roughness R q μ m Root mean 2 roughness (continued on next page) Abbreviations: PEMFC, Proton exchange membrane fuel cell; CCM, Catalyst coated membrane; FIB-SEM, Focused ion beam scanning electron microscopy; FIBTEM, Focused ion beam transmission electron microscopy; IJP, Inkjet printing; IQR, Interquartile range; USC, Ultrasonic spray coating; CNC, Computer numerical control; FTO, Fluorine-doped tin oxide; GDL, Gas diffusion layer; MPL, Microporous layer; MEA, Membrane electrode assembly; CL, catalyst layer; R2R, roll-to-roll; Xray CT, X-ray computer tomography; EDS, energy-dispersive spectroscopy; CF, crack fraction; NaN, not a number; R a , Arithmetic Mean Square Roughness; R q , Root Mean Square Roughness. * Corresponding author. E-mail address: [email protected] (K. Bouzek). Contents lists available at ScienceDirect Chemical Engineering Journal journal homepage: www.elsevier.com/locate/cej https://doi.org/10.1016/j.cej.2025.171283 Received 5 September 2025; Received in revised form 3 November 2025; Accepted 25 November 2025 Chemical Engineering Journal 527 (2026) 171283 Available online 26 November 2025 1385-8947/© 2025 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license ( http://creativecommons.org/licenses/by/4.0/ ). (continued) sm 2 m −3 Specific surface S 1 (v) (x)1 One point probability function for the void phase S bulk S m −1 Average bulk electrical conductivity S CL S m −1 Catalyst layer electrical conductivity U lb V Lower boundary voltage U ub V Upper boundary voltage V–Convex region of space filled by the medium vV s −1 Scan rate w i 1 Mass fraction of component/phase i x μ m Position vector x =(x 1 , x 2 , x 3 ) =(al 1 , al 2 , al 3 ) z i μ m Local height of CL z μ m Mean height of CL (averaged over all pixels in the map) βm 2 Transport parameter δmDistance between the nearest point on the pore-solid interface ϕ v 1 Total porosity ϕ vi 1 Isolated porosity κm Transport parameter σ β m 2 Joint confidence limit for β σ κ m Joint confidence limit for κ ψ ii 1 Calculated geometric factor 1. Introduction Proton exchange membrane fuel cells (PEMFCs) represent a key technology for decarbonisation across transportation and stationary power sectors due to their high energy conversion efficiency and compact design [1–4]. However, large-scale adoption is limited by high catalyst costs and performance variability, both of which are closely linked to the structure and fabrication method of the catalyst layer (CL). As the functional site for electrochemical reactions, the CL architecture influences catalyst utilization, gas transport, ionic and electron conductivity, and durability [5–8]. To maximize performance, CLs are often deposited directly onto the membrane, forming a catalyst coated membrane (CCM). This approach enhances the connection between ion conductive phases and significantly improves PEMFC performance [9]. In addition to composition, the deposition technique chosen for production of CCMs plays a crucial role in determining the structural and functional properties of CLs. Advanced methods allowing precise control over the structure and composition of CCMs are essential, particularly in considering economic factors such as process scalability and waste reduction. Among the largescale manufacturing methods, roll-to-roll (R2R) techniques, such as slotdie coating, are commonly reported for high-throughput production of membrane electrode assemblies (MEAs) with identical geometry [10,11]. Additive manufacturing technologies, such as inkjet printing (IJP) offer an attractive alternative. IJP deposits the catalyst ink onto the membrane surface in a highly controlled manner using piezoelectric actuators. This technique is particularly advantageous for producing low platinum loaded layers with high precision and minimal material waste [12–23]. Additionally, IJP is characterized by high flexibility and variability, allowing for rapid on-demand deposition of customized structures and graded catalyst layer geometries enhancing utilization of Pt and/or efficiency of the electrochemical processes [24–26]. This makes IJP highly attractive and suitable for a wide range of applications at small to medium production rates. Ultrasonic spray coating (USC) is a widely used scalable technique for CCM fabrication [46–48]. It can use catalyst inks of identical, or similar, composition as IJP, while it relies on a different deposition technique based on atomization of the ink with the following gravitational deposition of the formed droplets to substrate. In contrast to the other layer deposition methods, both the USC and IJP are scalable from laboratory application to industrial production. This makes USC both an industrially relevant benchmark and a counterpoint in the deposition mechanism to IJP allowing us to separate effects of the deposition method and the composition of the ink. Our previous study [27] demonstrated that CLs produced by industrial-scale IJP outperformed those deposited by USC, particularly at low platinum loadings. This superior performance was linked to differences in internal structure and transport properties resulting from the chosen deposition technique. To achieve a deeper and more accurate understanding of CL morphology, advanced imaging tools such as focused ion beam scanning electron microscopy (FIB-SEM), transmission electron microscopy (FIB-TEM), and X-ray computed tomography (X-ray CT) are increasingly utilized [8,28–36]. Structural investigations of IJP catalyst layers have been reported by others, including Shukla et al. [22,37] and M. Sabharwal et al. [38], who provided valuable insights using methods such as SEM cross-section analysis, mercury porosimetry, and FIB-SEM reconstruction. However, most studies typically focused on selected structural features and analysis methods, without comparison to other fabrication methods and actual performance in an experimental fuel cell. To achieve a deeper understanding of how deposition influences catalyst layer functionality, a comprehensive approach that builds directly on the performance data from our previous work [27] was adopted in this study. The aim was to compare side-by-side IJP and USC produced catalyst layer using identical inks and substrates. This approach allows to eliminate other parameters and to link observed differences in the performance exclusively to the deposition technique used. By combining high-resolution FIB-SEM tomography, optical profilometry, electron conductivity measurements, numerical simulation and experimental determination of transport parameters of CLs produced by IJP and USC, a critical set of information was collected. These techniques were used as a compromise between adequate resolution and sufficient size of the imaged/reconstructed region, which was necessary to draw relevant conclusions. It explains the performance differences observed in operating PEMFCs and assess the structural advantages of IJP for scalable, high-efficiency MEA production. This knowledge is crucial for further optimisation of the ink formulation and printing process parameters. 2. Experimental For the comparison purposes, an ink with standard composition, suitable for both, the inkjet printing and ultrasonic spray coating, was developed and optimised. This was necessary in order to compare properties induced by the CL deposition method, while eliminating potential impact of the ink composition. 2.1. Ink preparation The catalyst ink was prepared by mixing 4.7 g of ionomer dispersion Nafion™ D521, 5 wt%, (Chemours), 0.557 g Pt/C catalyst TEC10V30E (Tanaka Kikinzoku) in 27.8 g of water:alcohol mixture (30:70). The mixture was sonicated in an ultrasonic bath for 20 min and subsequently treated for 10 min (pulse 0.5 s) with ultrasonic probe to homogenize the ink and break up agglomerates. 2.2. Catalyst layer deposition Two experimental methods of catalyst layer deposition were used, IJP and USC. In order to compare reliably properties of CLs deposited by both techniques, deposition parameters were kept constant, independently of the substrate type. 2.2.1. Inkjet-printing For the inkjet printing Q-Class Sapphire QS-256/80 industrial printhead (Fujifilm Dimatix, USA) was used. The native drop volume of each nozzle is 80⋅10 3 μ m 3 . The native resolution of the printhead is 100 dpi, corresponding to the distance between the nozzles of 254 μ m. The QS-printhead was driven by the PixDRO LP50 (SÜSS MicroTec SE, Germany). Further details and printing process optimization are described in our previous study [27]. M. Hala et al. Chemical Engineering Journal 527 (2026) 171283 2 2.2.2. Ultrasonic spray coating Catalyst layers were deposited using an in-house made system, consisting of an ultrasonication nozzle with integrated focusing N 2 gas delivery (50 cm 3 min −1 ) with a controller (Cheersonic), integrated into an x-y-z CNC platform (CZ Robotics). Details of the deposition are described in our previous study [27]. 2.3. Catalyst layer electron conductivity Electron conductivity of prepared layers was determined using procedure described in previous work [30]. Catalyst layers were deposited using procedures described in Sections 2.2.1 and 2.2.2 on a fluorinedoped tin oxide (FTO) glass TCO22–7/LI (Solaronics) with conductivity of 286,500 S m −1 and thickness of 500 nm. A series of CL samples was prepared with Pt loading varying between 0.1 and 0.3 mg cm −2 . The list of the samples tested is summarized in Table 1. Due to the nature of the layer surface, the thickness of the catalyst layers was determined using profilometry data analysed by four distinct methods: 1) Smooth 95 Percentile Method: This method calculates the layer thickness as the 95 th percentile of the smoothed height profile. The uncertainty is determined from the standard deviation of the top 10 % of the height values. 2) Average Regions Max Method: The height profile is divided into five equal regions, and the average height of each region is calculated. The thickness is determined as the height of the highest region, and the uncertainty is derived from the standard deviation of the regional averages. 3) Average Regions Min Method: The height profile is divided into five equal regions, and the average height of each region is calculated. The thickness is determined as the height of the region with the lowest thickness, and the uncertainty is derived from the standard deviation of the regional averages. 4) Compact Layer Method: This method applies strong smoothing to the height profile to eliminate surface roughness. The compact layer is identified using the interquartile range (IQR) to exclude outliers, and the thickness is calculated as the mean height of the compact layer. The uncertainty is determined from the standard deviation of the heights within the compact layer. Each method provides complementary insights into the layer thickness, accounting for different structural features and measurement uncertainties. The results were compared, and the 4) Compact Layer Method was selected as the most suitable one. The profilometry measurements were performed using a DektakXT profilometer (Bruker) in line mode across the entire catalyst layer, with normalization against uncoated, flat sections of FTO glass. The results from all methods were compared to ensure robustness and consistency in the thickness evaluation. Determination of in-plane electron conductivity was performed using the four-electrode setup described in [30] using chronopotentiometric DC current polarisation with potentiostat PARSTAT MC (Ametek) ranging from 1 to 10 mA. Resistance of layer was evaluated using 3D mathematical model of potential distribution in sample defined by Laplace equation. The bulk conductivity of a fully compressed CL was measured using the method developed in [30]. A catalyst layer removed from the FTO glass and homogenized was used for this experiment. From the graph of the dependence of the layer's electrical conductivity on reciprocal compression level, the value of conductivity at maximum compression level was extrapolated. 2.4. Catalyst layer permeability The method used to obtain effective transport properties of the catalyst layer involved quasi-stationary permeation experiments using pure gases at ambient temperature. Due to the mechanical instability of the CL, it was deposited on a SIGRACET 38BC gas diffusion layer (GDL), allowing for the estimation of transport parameters by first measuring the GDL alone and then the GDL with the CL. The permeation cell consisted of two chambers separated by impermeable discs with a cylindrical opening, where the porous sample is fastened. The mathematical model assumed quasi-stationary gas transport in multilayer porous media and allowed the estimation of effective transport parameters appearing in the Darcy equation modified for gas transport. Further details on this method can be found in previous work [30]. 2.5. Optical surface characterization Optical observation and surface roughness of prepared samples on Sigracet 38BC GDL were carried out by optical profilometry Sensofar S Neox (Sensofar) working in confocal mode. The confocal Z-scanning were performed using 20×objective (TU Plan Fluor EPI, Nikon) with vertical resolution of 5 nm and pixel size of 0.69 μ m. To cover statistically significant area of the sample surface comparable to the sample area exposed to the gas permeation in the permeation cell, 4 ×3 field-ofviews for each sample was acquired. After installation, the total inspected area of the sample was 3.15 mm ×2.37 mm. The final sample map recorded contained two types of pixels: I) pixels with information about the height in case that the light from microscope illumination went through the focus, and II) pixels with “Not-a-number” value in case of the light did not find a focus within the total scanning depth of 100 μ m. In the case of these materials, the “Not-a-number” values typically indicate cracks in the sample surface. Due to a natural tilt of the sample on the large inspected area, the resulting map was corrected for sample tilt by identifying the background. The background correction was based on morphological opening, which was applied only on homogeneous part of the map; i.e., all the pixels containing height values. As a structuring element, the disc with a diameter of 20 pixels was used. The background subtracted map was consequently processed to calculate characteristics to quantify surface roughness [39], the Arithmetic Mean Roughness (R a , Eq. (1)) and the Root Mean Square Roughness (R q , Eq. (2)) Ra=1 N∑N i=1|zi−z|(1) Rq= 1 N∑N i=1(zi−z)2 √(2) where N is the number of pixels in the sampling area, zi is the local height (layer thickness) at pixel i and z is the mean height over all pixels Table 1 Pt loading and thickness of catalyst layers on FTO glass based on the Compact Layer method used for the evaluation of electrical conductivity. Sample ID Pt loading / mg cm −2 Thickness / μ m FTO_USC-1 0.097 6.89 ±0.83 FTO_USC-2 0.104 7.74 ±1.06 FTO_USC-3 0.167 9.79 ±0.85 FTO_USC-4 0.164 9.29 ±0.95 FTO_USC-5 0.217 12.67 ±1.05 FTO_USC-6 0.217 13.28 ±0.91 FTO_USC-7 0.294 16.17 ±1.09 FTO_USC-8 0.311 17.52 ±1.11 FTO_IJP-1 0.089 4.21 ±0.62 FTO_IJP-2 0.091 4.38 ±0.72 FTO_IJP-3 0.132 7.01 ±0.48 FTO_IJP-4 0.150 5.20 ±0.51 FTO_IJP-5 0.180 8.13 ±0.46 FTO_IJP-6 0.175 9.00 ±0.62 FTO_IJP-7 0.236 9.22 ±0.64 FTO_IJP-8 0.239 8.99 ±0.89 M. Hala et al. Chemical Engineering Journal 527 (2026) 171283 3 in the map. To quantify the extent of cracks, we introduced Crack fraction (CF), which is the ratio of “Not-a-number” pixels divided by total amount of pixels in the map. 2.6. 3D imaging of catalyst layers For the 3D imaging, a dual beam Lyra 3 GMU instrument (TescanOrsay) equipped with a high-power electron beam and a gallium ion column was used. The details of the data acquisition procedure are described in a recent paper [29] with one exception, which is the voxel size. By fine-tuning the whole procedure, the voxel size was reduced to 5 nm, which allowed following microstructural details at a higher resolution. Raw 2D images had to be re-aligned, cropped and packaged into one volumetric (3D) image for each material. After correcting for uneven illumination in the 3D images, these images were further processed by filtering in the spatial domain. A proven combination of an adaptive median filter and an anisotropic 3D diffusion filter was used to adjust the images prior to segmentation. Subsequent segmentation was performed using a graph-based algorithm called power watershed. The details of the entire procedure are again described in paper [29]. 2.7. Mathematical description of microstructure The structure of the digitised catalyst layer consisting of the void and solid phases can be defined in terms of the indicator function for the void phase (Eq. (3)) [40]. I(v)(x)={1,if xbelongs to the void phase 0,otherwise (3) where x is the position vector, x ∈V, and V is the convex region of space filled by the medium. Discrete values of x =(x 1 , x 2 , x 3 ) are determined by a simple cubic lattice of size al 1 ×al 2 ×al 3 in the 3D medium. The distance a between two adjacent lattice nodes corresponds to the cubic voxel size, and l 1 , l 2 , and l 3 are integers. Note that there is the simple functional relationship between the indicator functions for the void and solid phases: I (v) (x) +I (s) (x) =1. The mathematical model of the microstructure represented by the indicator function is subsequently referred to as the replica. It is also worth noting that the third coordinate (x 3 ) is related to the thickness of the catalyst layers. In order to show basic statistical characteristics of region V in a condensed form, the one-point probability function, S(v) 1(x), was computed for the void phase in three principal directions. The function S(v) 1(x)gives the probability of finding the void phase at the position x and is interpreted as a position-dependent volume fraction of the void phase [40]. If the values of this function are independent of position, the microstructure is said to be statistically homogeneous. Consequently, an ergodic hypothesis holds, i.e., one sufficiently voluminous region V provides complete probabilistic information. It is also meaningful to define volume averages for V. Since the region V had a relatively small volume (due to the limits of FIB-SEM), the function S(v) 1(x)allowed us to test whether the ergodic hypothesis was valid and whether a volume of the region V was large enough to evaluate volume averages (effective transport properties here). Averaging S(v) 1(x)over the entire region V gives the mean void fraction (Eq. (4)). ϕv=∫V S(v) 1dx(4) To characterize pore sizes of V, a microstructural descriptor called the pore-size probability density function P (v) (δ) was chosen. The product P (v) (δ) dδ represents the probability that a randomly chosen point in the pore space lies at a distance between δ and δ +dδ from the nearest point on the pore-solid interface [40]. Note that the definition assumes the statistically homogeneous microstructure in V. Since only one realization of V for each catalyst layer was available, the ergodic hypothesis in the calculation of the P (v) (δ) function had to be used. The function P (v) has the two extreme values (Eqs. (5) and (6)) P(v)(0) = s/ϕv(5) P(v)(∞) = 0 (6) where s stands for the specific surface defined to be the interface area per unit volume of V. The complementary cumulative probability function F (v) (δ) complemented the function P (v) (δ) (Eq. (7)). F(v)(δ) = ∫∞ δ P(v)(r)dr(7) Put another way, F (v) (δ) is the fraction of pore space that has a pore radius larger than δ. By analogy, the functions P (s) (δ) and F (v) (δ) for the solid phase were also calculated. A mean radius of spherical regions of phase i was also used (Eq. (8)). 〈δi〉 = ∫∞ 0 δP(i)(δ)dδ=∫∞ 0 F(i)(δ)dδ(8) Connectivity of the void phase was characterized using the two-point cluster function C(v) 2(x1,x2)[40]. For a statistically homogeneous medium, the function is defined as the probability that both ends of a line segment of length u= ‖x2−x1‖tossed at random into the medium are located in the same cluster of the void phase. Note that the 26-neighbour connectivity rule was used for clustering of the void-phase voxels. In order to verify the pore structure information obtained by the 3D imaging and subsequent mathematical treatment, pore size estimation with mercury porosimetry was determined using AutoPore IV (Micromeritics) porosimeter and BET surface area was obtained using ASAP 2050 (Micromeritics) extended pressure adsorption analyzer. 3. Results and discussion In the following text, individual characteristics of the CLs produced by the IJP and by the USC are determined and compared. The aim is to link these properties to the CL performance in the fuel cells reported previously [27]. This allows gaining deeper understanding of the deposition method impact on the catalyst layer functionality and, consequently, fuel cell performance. As can be found in the Fig. S1 in Supplementary information, the IJP layers exhibit high activity and efficient catalyst utilization at low cathode loadings (0.1 mg Pt cm −2 ). As catalyst loading increases, however, the load curves show transition toward transport-limited behaviour. In contrast, USC layers show kinetically hindered performance at low loading. With increasing loading both mass and charge transport are improved resulting in improved fuel cell performance. This leads to the similar peak power performance of IJP layers with 0.1 mg Pt cm −2 catalyst loading and USC layers with 0.3 mg Pt cm −2 . 3.1. SEM microscopy The first insights into the morphological characteristics of the catalyst layers were obtained using scanning electron microscopy (SEM). This method enabled the visual assessment of surface texture, crack patterns, and deposition-related artefacts, forming the basis for subsequent quantitative and structural analyses. It is important to keep on our mind, that these CL defects have potential influence on the fuel cell performance, e.g. due to the impact on the CL longitudinal conductivity, or water management. Whereas Fig. 1 summarises typical morphologies of CLs produced by USC and by IJP, Fig. 2 provides results of the surface EDS analysis. Samples USC 10L (i.e. 10 layers) and IJP 6L are displayed here because they were also used for the structure reconstruction, and sample USC 20L is displayed to show the contrast between samples with low and high Pt loading. Results obtained for the remaining samples are M. Hala et al. Chemical Engineering Journal 527 (2026) 171283 4 provided in Figs. S2 to S5 in the Supplementary information. Following conclusions may be derived from these observations. Firstly, the USC layers exhibit more visible cracks with sharper edges compared to the IJP layers. This can be attributed to differences in the thickness of a single deposited layer and the drying rates between the two methods. Individual layers deposited by USC contain more material and are drying more rapidly. This leads to an increased stress and more pronounced crack formation. Secondly, in the IJP sample, repeating parallel lines are evident. This is clearly result of the printing process. Despite this, surface morphology indicates more compact and ordered CL structure in the case of CL produced by IJP. This coincides with the profilometry results (see Section 3.2, Table 2). Surprisingly, on the surface of the USC layers, spherical particles of 5 μ m in radius may be observed. EDS analysis confirmed these structures to be of the same composition as the CL Fig. 1. SEM images of USC 10L (A, B, C,) and IJP 6L (E, F, G) at ×50 (A, E), ×250 (B, F) and ×1000 (C, G), detail of the spheres (I), and cross-sections of USC (D) and IJP (H) CCMs. M. Hala et al. Chemical Engineering Journal 527 (2026) 171283 5 surface. Accordingly, the spherical particles were identified as drops of the ink dried in-flight before impacting the surface of the CCM [41]. Some of these particles disperse when next layers are deposited onto them, as seen in cross-section images of catalyst layers (Fig. 1D)). Thirdly, as Fig. 2 shows, Pt was located in unconnected islands in the case of USC 10L. These islands were not present anymore in the USC 20L sample, leading to a more homogeneous catalyst and ionomer distribution. This demonstrates the USC method is not suitable for deposition of CLs with low Pt loadings. In contrast to this, even IJP 6L shows homogeneous Pt distribution. 3.2. Optical profilometry Whereas SEM provides qualitative images of surface features, optical profilometry allows to quantify surface roughness and crack distribution across representative areas. A combination of these two methods gives a comprehensive picture of surface morphology and layer integrity. This information is important for the following structural evaluation. An overview optical image was acquired (see Fig. 3A and B) to navigate over the sample and to identify the field-of-views for confocal scanning. Overview images of all samples can be found in Figs. S6 and S7. The Not a Number (NaN) pixels corresponding to the cracks are used to determine Crack Fraction of the sample surface. The Crack Fraction value is calculated based on binarization of the confocal image into the region of Fig. 2. EDS analysis of Pt (A, E, I), F (B, F, J), C (C, G, K) of samples IJP 6L (A, B, C, D), USC 10L (E, F, G, H) and USC 20L (I, J, K, L) and corresponding SEM images (D, H, L). Table 2 Crack fraction, R a and R q numbers for USC and IJP samples. SIGRACET 38BC is used as GDL. Sample Crack fraction / % R a / μ mR q / μ m USC 10L 4.0478 0.550 0.640 USC 20L 4.1603 0.625 0.745 USC 30L 2.5032 0.570 0.635 USC 40L 1.3394 0.625 0.730 IJP 6L 0.6187 0.375 0.195 IJP 9L 0.5357 0.390 0.210 IJP 12L 0.3276 0.400 0.200 IJP 15L 0.1760 0.440 0.250 GDL 1.3542 0.398 0.235 M. Hala et al. Chemical Engineering Journal 527 (2026) 171283 6 the homogeneous material and regions with cracks (Fig. 3E) and F)). In addition to crack fraction quantification, confocal profilometry image was also used to calculate Arithmetic Mean Roughness (R a ), and Root Mean Square Roughness (R q ). The R a value shall be interpreted as average height of the peaks and valleys on the surface. R q value is more sensitive to higher peaks and deeper valleys than R a value. Lower R q values correspond to more uniform/flat surface. The resulting values are summarized in Table 2 for USC and IJP samples and substrate GDLs. Following conclusions may be derived from the analysis performed: •GDLs used show homogeneous properties and do not have any effect on the properties of CLs deposited. •USC prepared CLs show rougher surface with higher number of the well-pronounced cracks. These cracks are formed during the deposition process and their number decreases with the number of layers deposited. •IJP produced CLs are generally smoother and more homogeneous. During printing, catalyst ink fills cracks occurring in the previous layer, including the GDL substrate. This leads to a significantly less disturbed CL. •Smoothness of the CLs produced depends solely on the deposition method and does not change with the number of the layers deposited. These results are in contrast to crack formation behaviour of the CLs printed on membranes from our previous study [27]. There, the thicker CLs (more individual layers) exhibited higher crack fraction. These differences stem from the mechanism of the ink-substrate Fig. 3. (A, B) Optical overview image of sample USC 20L (A) and IJP 12L (B). Cracks (black whiskers) are statistically significant feature of the sample surface for all investigated samples. (C, D) Confocal profilometry of sample USC 20L (C) and IJP 12L (D). Color bar denotes to the normalized, background subtracted, height of the sample surface. White areas in the images represent NaN values. (E, F) Binary image of the USC 20L (E) and IJP 12L (F) surface. Black pixels denote to regular homogeneous surface, white pixels denote to cracks in the surface. M. Hala et al. Chemical Engineering Journal 527 (2026) 171283 7 interaction – the membrane swells and wrinkles in contact with the ink and the ink dries slower, while the GDL is not affected by the wetting of its surface by the ink and has a higher heat transfer coefficient leading to faster ink drying rate. 3.3. Porosity estimated from apparent and skeletal densities of CL Following the characterization of surface morphology, the next step was to estimate the porosity of the layers based on the known composition and the thickness determined via profilometry. This approach provides a bulk-level assessment and serves as a first comparison of structural differences induced by the deposition method. Firstly, thickness of the layer deposited on top of the FTO glass needs to be determined. The results obtained using the Compact Layer method are summarized in Fig. 4 (comparison of all the methods can be found in Fig. S8 in Supplementary information). As can be seen, CL layer thickness increased linearly with the Pt loading for both methods. USC layers, however, were consistently thicker than IJP layers with the same Pt loading. This suggests that USC formed CLs characterized by a less dense/more open structure. The mass, composition and thickness of the CLs can be transformed into the total porosity ϵV using Eq. (9) [15,42]. ϵV=1−mPt L(1 ρ Pt +wC wPt ρ C +wN (1−wN)wPt ρ N)(9) where mPt is the platinum loading in g cm −2 , L is the catalyst layer thickness in cm, ρ C, ρ Pt and ρ N are densities of carbon, platinum and Nafion taken as 2.0 g cm −3 , 21.5 g cm −3 and 1.8 g cm −3 , respectively. The composition of the heterogeneous mixture of platinum and carbon, i.e., platinum deposited on the carbon support, is described by the respective mass fractions wPt and wC (wPt +wC=1), while the Nafion mass fraction wN refers to the entire mass of CL. The mass fractions of the components were calculated from the composition of the ink: wPt =0.290, wC=0.710 and wN=0.298. The results are collected in Table 3. 3.4. Microstructure of catalyst layers To obtain more detailed information on the internal structure of the catalyst layers and to verify the trends suggested by surface analysis and fuel cell performance, FIB-SEM tomography was used to reconstruct and examine their 3D microstructures. The 3D reconstruction of the CLs microstructure allows a direct visual and mathematical assessment of porosity, phase connectivity, and isotropy. Two thinnest CLs labelled IJP 6L and USC 10L were selected for this study. Thin layers are less prone to a structure collapse during the FIB milling than the thicker layers. Additionally, thin layers with low Pt loading are more interesting from the application point of view. Resulting primary volume images were processed by combining the 3D anisotropic diffusion filter and segmentation based on the watershed algorithm [29,30]. As a result, the visualisation of the reconstructed void and solid phases distribution was obtained (see Fig. 5). The segmented data was stored as an indicator function uniquely assigning each point in the 3D discrete space to either a void phase (pore) or a solid phase. Before any use, the indicator functions were checked for the presence of isolated (discontinuous) clusters of solid phases in the inner regions. Because their existence is non-physical, all solid phase clusters were identified using the 6-neighbour connectivity rule. Those clusters not connected to the rest of the solid phase in the inner regions were excluded from further considerations [43]. This operation slightly increased the void phase fraction, say by about 2 ×10 −4 . There is no physical reason for a similar operation with the void phase because isolated pores may exist. However, to keep the original porosity, all isolated pore clusters consisting of less than a specified number of voxels were replaced with solid-phase voxels. We note that the clustering of the void phase was governed by the 26neighbour connectivity rule. The indicator functions serve then as key Fig. 4. Dependence of catalyst layer thickness on Pt loading. Table 3 Platinum loading mPt, thickness L (determined using the Compact Layer Method) and total porosity ϵV calculated using Eq. (9) of catalyst layers on GDL. Sample ID mPt / mg cm −2 L / μ mϵV USC 10L 0.047 4.41 0.78 USC 20L 0.074 5.72 0.73 USC 30L 0.115 7.71 0.69 USC 40L 0.146 9.22 0.67 IJP 6L 0.090 4.64 0.60 IJP 9L 0.141 6.34 0.54 IJP 12L 0.178 7.57 0.51 IJP 15L 0.238 9.56 0.48 M. Hala et al. Chemical Engineering Journal 527 (2026) 171283 8 inputs for all subsequent calculations, including statistical measures, random-walk simulations and permeability calculations. Note that the term replica is used in this study for the indicator function. Basic microstructural characteristics of the reconstructed samples of IJP 6L and USC 10L are collected in Table 4. The void fractions ϕ v in Table 4 represent the values averaged over the entire 3D regions, the sizes of which are given there. Since the volume fractions of both replicas differ little, their microstructure can be tentatively characterized as similar. The isolated porosity, which is negligible in both cases, can also be seen as a reconstruction artefact. Fluctuations of S(v) 1 around ϕ v are shown in the top row of Fig. 6. Although these fluctuations are quite large for both replicas, there are no significant differences between the courses measured in the principal directions. In addition, most of the region V shows a local volume fraction in the interval from 0.3 to 0.5. Therefore, the microstructures of R-IJP 6L and R-USC 10L (excluding cracks) appear to be statistically homogeneous. Thus, the replicas can be considered as representative elementary volumes of both CLs. However, it must be kept in mind that the replicas are small, which limits the significance of the conclusions. The sizes of the spherical regions that fit into the void or solid phase are shown in the middle row of Fig. 6. The density probability functions P (v) and P (s) decrease very rapidly near the origin, indicating the presence of a large number of small spherical regions at the edges of other, more compact regions of both phases. The solid phase of both replicas forms spherical regions with radii between 10 nm and 40 nm, which are approximately equally represented, see the flat regions of the blue curves. The functions P (v) do not show a similar behaviour for either replica, i.e., they do not involve flat regions. Considering that ϕ v = 0.4004 for R-IJP 6L, it is interesting that the largest spherical regions in the void phase are even larger than the largest spherical regions in the solid phase, see the blue line crossing the red one in Fig. 6. In contrast, Fig. 5. 3D replicas of catalyst layers from FIB-SEM tomography supplemented with pore space skeletons: A) IJP 6L and B) USC 10L. Dimensions of the insets: IJP CL 2945 nm ×1855 nm ×1405 nm, IJP MPL 3415 nm ×1955 nm ×1405 nm, USC CL 4090 nm ×3050 nm ×3690 nm, USC MPL 4185 nm ×2050 nm ×3690 nm. Table 4 Basic characteristics of FIB-SEM replicas: size in voxels, total porosity ϕ v , isolated porosity ϕ vi , specific surface s (interface area per unit sample volume) and interface area per unit pore volume s/ϕ v . The voxel size is 5 nm in both cases. Replica Size, voxel ϕ v ϕ vi s / μ m −1 s ϕ v −1 / μ m −1 R-IJP 6L 1685 ×451 ×282 0.4004 0.0009 14.8 36.9 R-USC 10L 2079 ×817 ×739 0.3826 0.0007 12.3 32.1 M. Hala et al. Chemical Engineering Journal 527 (2026) 171283 9