Bipolar-interface fuel cells – an underestimated membrane electrode assembly concept for PGM-free ORR catalysts

We present the ﬁ rst combination of a bipolar interface fuel cell with a commercial Fe – N/C catalyst as an alkaline cathode and a PGM-based, acidic anode, both separated by a proton exchange membrane (PEM). This membrane electrode assembly (MEA) concept enables the employment of Fe – N/C catalysts in a less corrosive alkaline environment, while simultaneously keeping the profound advantages of the hydrogen oxidation reaction in acidic media with extremely low PGM-material requirement. We compare two di ﬀ erent cases for the anion exchange polymer – proton exchange polymer (AEM|PEM) interface at the alkaline cathode and the acidic membrane. In one case the PEM is simply pressed against the alkaline electrode and in the other case a part of the PEM is deposited onto the alkaline electrode. We achieved power densities of about 38 mW cm (cid:1) 2 and 210 mW cm (cid:1) 2 respectively. This is corresponding to 2.1 W mg Pt (cid:1) 1 cm (cid:1) 2 . Our results show, that the bipolar interface design is one of the most important factors for performance optimization in BPM fuel cells. In addition, we compared a conventional PEM fuel cell with identical Fe – N/C cathode loading to the bipolar deposition case. After a 15 hour test run the PEMFC cell showed a strongly increased overpotential at lower current densities, whereas the overpotential increase for the bipolar cell was only marginally in the same current density region. With this work we show a facile manufacturing approach that enables bipolar interface fuel cells with Fe – N/C catalysts, showing promising power densities at low total PGM-loadings.


Introduction
In present-day research PEMFCs are the prevailing fuel cell type for PGM-free oxygen reduction reaction (ORR) catalysts, like Fe-N/C. This fact is mainly attributed to the high power densities achieved at low to medium temperatures and the sophisticated, commercially available fuel cell components like catalysts and highly conductive proton exchange membranes. But stability studies on Fe-N/C ORR-catalysts in acidic media revealed that catalyst deactivation occurs via carbon corrosion at potentials above 0.9 V and via Fe demetallation at potentials below 0.7 V. 1 This is one of the reasons anion exchange membrane fuel cells (AEMFCs) became a focus of interest. The alkaline environment facilitates the use of a broader range of material classes, like hydrocarbon based ionomers and provide a more stable operating environment for non-PGM ORR-catalysts. 2 However, the exchange current densities for the HOR in alkaline media are at least one order of magnitude slower than in acidic media. 3 This consequently increases the anode overpotential losses (which are nearly negligible in PEMFCs) in an AEMFC to an extent, where high Pt-loadings or extensive research on novel PGM-free HOR catalysts are necessary to compensate for those losses. In our opinion, when considering the AEMFC-technology as an option to stabilize PGM-free catalysts, to reduce overall device costs, it is counterproductive to increase the amount of PGMs at the anode to a level similar or even higher than in an entire state-of-the-art PEMFC. [2][3][4] An alternative, desirable implementation of PGM-free ORRcatalysts would combine the best of both worlds. By providing a non-corrosive, alkaline environment at the PGM-free cathode side the catalyst would be stabilized. Additionally, an acidic anode environment would reduce the anode overpotential and consequently the required PGM-loading to a minimum. In 2009 Unlü et al. introduced a fuel cell conguration concept, with one electrode operating at high and the other electrode operating at low pH. 5 The schematic illustration of such a hybrid fuel cell setup with a PGM-free cathode and its fundamental working principle is shown in Fig. 1.
The main difference between the hybrid conguration and the PEMFC/AEMFC congurations on their own are the sites of the water formation and consumption reaction: in the PEMFC water is produced at the cathode, in the AEMFC at the anode. In contrast to the conventional fuel cell congurations the water formation reaction is not located within the electrodes, but at the interface of the PEM and the AEM within the membrane. This combination of a PEM and an AEM is oen referred to as a 'bipolar membrane', since the interface forms a junction similar to the p-n junction in a semiconductor. For a more detailed description of the theoretical background of the bipolar "PEM|AEM" junction, seeÜnlü et al. 5 The overall electrochemical reaction for the bipolar membrane fuel cell (BPMFC) can be split into the half-cell and the interface reactions as following: Anode: Cathode: Interface: Full cell: Although the ORR in alkaline media is shied to lower potentials, compared to acidic media, the thermodynamically cell potential is 1.23 V, as the overall reaction remains identical to the reaction in classical fuel cell designs. The lower ORR potential is compensated by a potential at the bipolar membrane interface. This interfacial potential arises from xed charges on each side of the interface, creating an electric eld and consequently a potential difference between the AEM and PEM. 5 Since the development of the rst BPMFC 10 years ago nearly all studies were performed on Pt-based catalysts and mostly concentrated on the functionality and the development of different MEA architectures. 5,6 Although the employment of PGM-free catalysts was suggested since the rst BPMFC in 2009, only one detailed study enabling silver in the CCL was published since then by Xu et al. Due to an optimization of the CCL ionomer content they reached an optimized fuel cell performance of 19.3 mW cm À2 under dry gas feed conditions. 7 Additionally, Ramani and coworkers demonstrated another bipolar MEA conguration with a Ag based cathode under fully humidied conditions and a resulting power density of $50 mW cm À2 . 8 As demonstrated in previous work, a crucial element that needs to be taken into account, when manufacturing MEAs, is the quality of the individual layer interfaces. 9 In contrast to conventional FCs (PEMFC and AEMFC) in BPMFCs an additional interface is introduced by the "PEM|AEM" junction, creating a localized reaction zone for the water formation reaction in the membrane. The local connement of an electrochemical reaction is contrary to the conventional FCs, where the water formation reaction is distributed over the corresponding electrode and is not fully localized at the respective interface. Therefore, the main development focus for BPMFCs should be extended to the "PEM|AEM" interface and not only be restricted to an electrode optimization process.
Previous publications focused on the development of the CL|PEM interface in conventional PEMFCs, by employing a manufacturing approach called "direct membrane deposition" (DMD). [10][11][12] The DMD approach substitutes the catalyst coated membrane (CCM) with two gas diffusion electrodes (GDE) being coated with ionomer solution and forming the PEM aer solvent evaporation and additionally eliminating the need for a hot-pressing step. The application of the ionomer solution can be executed with a variety of manufacturing approaches like ink-jet printing, 10 ultrasonic spray coating 11 or as realized in this work via doctor-blading. DMD opens up the possibility to manipulate essential fuel cell membrane parameters like mechanical and chemical stability by implementing electrospun polymer nanobers 13 or radical scavengers like cerium-oxide. 14 Additionally DMD is capable of fabricating membranes as thin as 12 mm. MEAs fabricated with DMD have demonstrated a more than 100% increased power density of 4 W cm À2 (300 kPa O 2 ) compared to a CCM with identical membrane thickness and identical catalyst loading. 15 The main reason for the high power densities was attributed to a dramatically reduced charge transfer and mass transport Fig. 1 Working principle of a hybrid fuel cell, composed of a high pH cathode catalyst layer (CCL) enabling Fe-N/C and a low pH anode catalyst layer (ACL) enabling Pt/C. The pH of the individual electrodes is adjusted by the selection of a proton conducting ionomer (PEM) as ACL binder and an OH À conducting ionomer (AEM) as CCL binder. resistance, originating from the enhanced contact of the PEM and the ionomer in the electrode at the CL|PEM interface. 12 The manufacturing of state-of-the-art BPMFCs is typically performed by hot-pressing a PEM and an AEM to create the bipolar membrane. Also state-of-the-art BPMFCs rely on the usage of GDEs, attributable to material inherent manufacturing processes like the common KOH-treatment of the AEMmaterials. Our hypothesis was that DMD, in contrast to the commonly used PEM/AEM or GDE/BPM hot-pressing steps, might not only lower the internal resistance of the BPMFC, but also might have the potential to enhance the interfacial contact at the PEM|AEM junction, resulting in drastically increased power densities.
This work is the rst one that enables DMD as a facile tool to create and manipulate the additional PEM|AEM interface in BPMFCs. Moreover, this is the rst reported usage of an M-N/C ORR catalyst in a BPMFC. In addition, this work demonstrates the rst exploration of doctor-blading as a tool to manufacture half-MEAs via DMD. To emphasize the importance of this novel and easy manufacturing approach, it was desisted to employ self-made materials and rely on commercially available materials only.
During this study, we focus on the variation of the PEM|AEM interface to better understand its importance for BPMFCs. In the second part, we investigated the potential benets of BPMFCs for the mitigation of the Fe-N/C degradation. This degradation is commonly observed in classical PEMFCs. For this reason, we manufactured three different MEA architectures in this work: compressed BPMFC (c-BPMFC), deposited BPMFC (d-BPMFC), and reference PEMFCs. They are schematically depicted in Fig. 2.
We designed the compressed BPMFC (c-BPMFC) conguration to create the BPM interface by physical compression of the two half-MEAs only. Contrary to that, the BPM interface in the deposited BPMFC (d-BPMFC) conguration emerges from directly casting half of the PEM on the alkaline part of the MEA. To prevent direct contact between PEM and catalytic active centres of the catalyst a thin layer of AEM was casted onto the alkaline CCL for both BPMFC-congurations beforehand. The total thickness of the PEM was kept constant for both congurations. The d-BPMFC was physically analysed with SEM-crosssection images and the evolution of the BPM interface was studied via cross-section and surface images to evaluate possible morphological changes of the electrode and the quality of the interfacial contact. The BPMFCs were characterized with a fuel cell characterisation setup, such as electrochemical impedance spectroscopy (EIS) and current-interrupt iR measurements, also in regards to pressure dependency. In the second part the d-BPMFC conguration was compared to the PEMFC conguration, in regards to fuel cell data collected before, during and aer continuous operation at constant current and potentials below 0.7 V.

Physical characterization
An electron image of the d-BPMFC cross-section is provided in Fig. 3. The image shows the layer structure of this design. Clearly visible is the junction between catalyst layer and membrane. Here the membrane conforms to the structure of the catalyst layer.
We documented the manufacturing process by imaging cross sections and surfaces following each major step. The steps include the deposition of the CCL, the inltration of the CCL with the AEM and the deposition of the PEM in the last step. Fig. 4 shows the surface images and the cross section images for the individual manufacturing steps. Both for the plain and the  AEM coated sample ( Fig. 4A and B) a rough surface is observed. Following the coating with Naon a smooth surface is observed (Fig. 4C). Thus a large interface is expected, if the Naon conforms to the surface of the catalyst layer. Fig. 4a and b also reveal that the AEM does not signicantly decrease the surface roughness at the scale accessible by a mechanical cross section. In general, no major change can be observed between the AEM coated and untreated catalyst layer. Fig. 4c indicates that the Naon conforms to the surface of the AEM treated catalyst layer.
Additionally, it appears to inltrate the upper layers of the catalyst layer. This could lead to an even larger bipolar interface. While we cannot assert this, without verifying that the AEM interlayer inltrates the catalyst layer to the same degree. The bipolar interface would be substantially large than both the geometric area and the surface roughness indicates.
The degree of inltration can be estimated by considering EDX maps of the sections. Especially the uorine signal gives an indication as to the degree of inltration. We have reproduced the uorine EDX map of the cross section in Fig. 4c. It shows the interface between PEM and CCL. The uorine signal matches the light areas in the backscattered electron image. Furthermore, even in pores not fully inltrated in the image a strong uorine signal is detected. This indicates at least partial inltration.

Interface studyelectrochemical characterization
To investigate the impact of the bipolar reaction-interface, the two BPMFC congurations d-BPMFC and c-BPMFC were analysed with respect to their overall device performance in a fuel cell setup (Fig. 6). The individual congurations were varied in the AEM|PEM-interface only. The catalyst loading, ionomer content in the CLs and the thickness of the PEM were constant for both cases. One bipolar interface was created via the direct deposition of a concentrated Naon-dispersion on the alkaline half-MEA (d-BPMFC -conguration). The second half of the PEM was manufactured by depositing a membrane with identical thickness on the acidic electrode. The second BPMFC was fabricated by depositing a Naon-membrane, with twice the thickness as the membrane on the individual half-MEAs in the PEM|PEM case, on the low pH electrode only. In this conguration (c-BPMFC-conguration) the BPM-interface was generated by physical compression of the two half-MEAs in the fuel cell setup.
The MEAs were then assembled as described in previous work. [10][11][12] When comparing the polarization and power density data for both BPMFCs: the increased maximum power density of the d-BPMFC (209 mW cm À2 ), compared to the c-BPMFC (38 mW cm À2 ) was visible at rst sight. Although it seemed natural to claim the superiority of the d-BPMFC-conguration, it was necessary to understand the impact of the interface in a more detailed manner. Therefore, we investigated the reasons for the differences between the two congurations. The over-potentials at lower current densities (up to 100 mA cm À2 ) seemed to be signicantly higher for the c-BPMFC. In this low current density regime, the potential losses of conventional PEMFCs are usually dominated by the ORR-kinetics of the catalyst. 16 However, when comparing the d-BPMFC conguration to the c-BPMFC conguration it is obvious that the kinetic regions, in spite of the same electrode congurations, were strongly different. Moreover, in PEMFCs the difference between the theoretical, thermodynamically potential and the OCV is usually an indicator for the magnitude of parasitic current arising from electrical shorts or fuel cross-over. 17 When considering the lower OCV for the c-BPMFC conguration (Fig. 6) compared to the d-BPMFC-conguration, despite identical membrane properties, it is reasonable to assume an additional inuence of the interface quality on the achievable OCV. Therefore, we propose the opencircuit-voltage (OCV) as a good indicator for the state of the  BPM-interface, as the interfacial reaction, according to eqn (1)-(4), accounts for $66% of the achievable potential. Aer identifying the major differences in the current-voltage characteristics it was necessary to probe for intrinsic characteristics of the MEAs, that would allow for a qualitative visualization of the respective bipolar reaction interface quality and the reasons behind the drastically different performances. One intrinsic parameter to look at is the ohmic cell resistance (R U ). The total ohmic resistance of the fuel cell is the sum of contact resistances and ohmic resistances of all individual components and interfaces. Those include the ohmic resistance of the membrane, the catalyst layer, the gas diffusion media, the bipolar-plates and all resulting contact resistances at the respective interfaces. 18 When excluding variations of the individual cell components, a possible difference of R U for the two BPMFC congurations, may be traced back directly to the interfacial contact quality and nature of the AEM|PEM interface. R U can be determined with a variety of electrical on-line testmethods, like AC resistance, current interrupt (iR), electrochemical impedance spectroscopy (EIS) and high frequency resistance (HFR). 19 For the determination of R U during this study, the system integrated current interrupt method was used. 19 As can be extracted form Fig. 7, R U was reduced by more than 50% when switching from the c-BPMFC to the d-BPMFC-conguration at 100 mA cm À2 and a gas backpressure of 200 kPa gauge . This enhancement should arise from a reduced contact resistance at the PEM|AEM interface, since the other ohmic resistance contributions remain identical for both congurations. During the conventional DMD process for PEMFCs, Naon inltrates into the electrodes to a certain degree, which was proposed as one of the main reasons for an enhanced PEM|CL interface contact and the increased power output. [10][11][12] Since the AEM was also fabricated via DMD in this work, the AEM should be viewed as a very thin layer covering the catalyst agglomerates in the upper electrode region aer inltration and not as a planar membrane on top of the alkaline electrode. As visible from the SEM images, the electrode pore-space was still accessible from the alkaline electrode side aer the AEM deposition. Therefore, as shown in Fig. 5 the PEM is partly incorporated into the upper pore space region of the alkaline half MEA. The inltration with the Naon dispersion consequently leads to a larger AEM|PEM interfacial area, compared to the physical compression of a nearly planar PEM-surface on a porous substrate. This inltration consequently leads to a reduction of the PEM|AEM contact resistance. Interestingly, for the c-BPMFC-conguration a strong dependence of R U from the applied gas backpressure was observed. As shown in Fig. 7 the ohmic resistance of the c-BPMFC increased drastically for backpressure values below 75 kPa gauge , whereas R U of the d-BPMFC varied little with the applied gas pressure at  a constant current. Furthermore, the d-BPMFC's resistances at 200 kPa gauge and 0 kPa gauge showed good agreement over a broad current range. We attributed the reduction of R U with increasing backpressure in the c-BPMFC-conguration mainly to an improved physical lamination of the AEM|PEM interface and an enhancement of the reaction area, as the so membrane may be pressed further into the porous electrode structure. Whereas in the d-BPMFC-conguration the BPM-interface may not be altered via additional physical compression, due to the unique possibility of PEM-electrode-inltration during the DMD-process. When reversing the adjustment of applied backpressure from 200 kPa gauge to 0 kPa gauge , the increase of R U in the c-BPMFC-conguration implied a beginning delamination of the AEM|PEM-interface as a consequence of insufficient adhesion between both interfaces. The DMD-manufacturing approach for the d-BPMFC-conguration enhanced the performance in two ways. First direct deposition of the PEM decreased R U via the enhancement of the AEM|PEM interfacial area. Second the polymer inltration also appears to promote the adhesion between the alkaline half MEA and the PEM. This adhesion was concluded from an unaltered R U at atmospheric gas pressure and its endurance over a broad range of current densities.
Electrochemical impedance spectroscopy (EIS) was used as another diagnostic tool for the evaluation of the processes within the BPMFCs (Fig. 8).
Usually the EIS-spectrum for a PEMFC creates a semi-circle loop, which is oen referred to as the kinetic loop, as the diameter of the semi-circle is usually controlled by the kinetic processes of the ORR at low current densities. 18 Calculating the difference between the two x-axis intercepts of the semi-circle is usually a facile method for determining the charge transfer resistance (R CT ) of the ORR. This resistance depends mainly on inherent properties of the cathode such as the electrochemical surface area of the catalyst, the utilization of the catalytic active sites or the catalyst loading. 18 As visible from the EIS spectra in Fig. 8, the diameter of the c-BPMFC-semi-circle was nearly double the diameter of the d-BPMFC-semi-circle. The difference resulted mainly from an increase in low-frequency resistance (R LFR ) for the c-BPMFC-conguration, whereas the increase of the high-frequency resistance (HFR), which can be interpreted like R U , was comparatively small. Since both cells share the same cathode characteristics, a facile ORR-R CT interpretation of the EIS data may not be fully adequate for the BPMFCs used in this study. As already presumed from the polarization data in Fig. 6, the large polarization difference in the low current density regime could not be explained by different cathode characteristics. Now the EIS data provided a measurable value that may account for a majority of the fuel cell performance discrepancy. Although different interpretations are plausible, to explain the differences in the EIS data, it is most feasible to think of an additional resistance term in BPMFCs, which includes the H + /OH À charge recombination at the AEM|PEMinterface (R CR ).
Grew et al. have discussed the transport of the charge recombination at the BPM-interface. 20 They found that the charge recombination most likely occurs via a trap-assisted mechanism. They have underpinned their assumption, among others, with the high current densities achieved with BPMFCs over the last years, which would not be possible for a different type of recombination mechanism. Additionally, they bring up the fact that AC impedance spectra of BPMFCs display  a classical charge-transfer behaviour with a R CT in the same order of magnitude as the predicted trap-assisted charge recombination resistance of H + and OH À at the BPM interface. They suggest that the charge recombination at the BPMFC might even be a rate limiting step in the overall cell and that a development of an impedance model for BPMFCs is highly recommended. 20 Ramani and co-workers concluded similarly to explain the large charge transfer resistance difference (measured at OCV conditions) from their bipolar MEA to their purely alkaline reference fuel cell. 8 Due to the manufacturing related manipulation of the bipolar interface, our EIS data, collected at 100 mA cm À2 , for the rst time provided signicant experimental conrmation for the assumptions of Grew et al. and Ramani et al. As the impedance spectra in Fig. 8 showed, a change in the nature of the AEM|PEM interface not only inuenced the ohmic cell resistance (HFR), as it was observed for alterations of the CL|PEM interface in PEMFCs at similarly low current densities, but also increased the apparent charge transfer resistance. 12 The impedance data suggest, that the AEM|PEM interface differs signicantly from other interfaces in the MEA, as the total cell performance was strongly dependent on the quality of the bipolar interface. At this point it would be too speculative to discuss more detailed, manufacturing related dependencies of R CR , but as evaluated from the current-interrupt-measurements the AEM|PEM contact resistance and adhesion properties between the two components appeared to be crucial parameters of the AEM|PEM-interface quality. When designing BPMFCs with an alkaline electrode and an acidic membrane, an alteration of the electrode composition may also inuence the interfacial contact behaviour between the PEM and the alkaline CL and consequently R CR . When evaluating such BPMFC electrode optimizations with EIS, the data should be interpreted with care, since the facile approaches for PEMFCs may not be transferrable to BPMFCs.

Comparison between BPMFC and PEMFC
Comparing the fuel cell performance of the d-BPMFC to a classical PEMFC ( Fig. 9) with identical Fe-N/C cathode loading and PEM-thickness, it becomes obvious that the measured maximum power density of the BPMFC was still relatively lower than the power density achieved with the conventional PEMFC.
Further analyzing the polarization characteristics of the two FC congurations, the more than twofold higher maximum power density of $550 mW cm À2 , achieved with the PEMFC, seemed to originate from reduced overpotentials in the low current density regime and not from a lowered ohmic cell resistance. This assertion is underpinned by the current dependent measurements of R U for both cells (Fig. 10). The internal cell resistance was only marginally increased for the BPMFC compared to the PEMFC. A high increase would imply high interfacial contact resistance at the AEM|PEM interface as described previously on the one hand or inadequate OH À transport through the CCL towards the reaction interface on the other hand. The high overpotential observed for the BPMFC in the low current density regime may therefore either be attributed to the charge-transfer processes at the bipolar interface or to a non-optimized electrode design. At this point we cannot clearly assign the increased overpotential to one of the mentioned contributions alone. But it is reasonable to assume, that apart from the bipolar interface, there is a strong impact of the electrode ionomer content, the catalyst loading, the ionomer equivalent weight and the lack of additives on the overall cell performance. This was demonstrated in literature for AEMFCs as well as for BPMFCs and will therefore be an additional focus in our future work. 7,21-23 Moreover, the used Fe-N/C in this study was designed and optimized for the application in PEMFCs and not exclusively for the use in a high pH environment.
For further comparison between the two FC congurations a constant current experiment was performed. For both cells the current was adjusted to a correlating cell potential of $0.45 V (as visible in Fig. 9). This value was chosen to operate at potentials below the stable operation potential window of 0.9-0.7 V. Due to the high overpotential of the BPMFC the potential was adjusted to values lower than 0.5 V to operate in the ohmic regime of both fuel cells. Aer the current hold for 15 h (Fig. 11) additional polarization data was collected. As visible from Fig. 11 the voltage decay of the PEMFC and the BPMFC aer 15 h is $66 mV and $17 mV respectively.
Although the overall potential loss for the BPMFC is smaller aer 15 h, compared to the PEMFC, the potential of the BPMFC uctuates more than 25 mV in shorter periods of time. Interestingly the strong potential uctuations of the BPMFC visible during the 15 h test run create a repetitive pattern of alternating periods of nearly constant potential at smaller time scales. The potential alternates between 0.45 V and 0.475 V in time periods of 10-15 minutes. The origin of the potential uctuations seems strongly connected with the R U of the BPMFC. The conventional PEMFC also showed uctuations of the potential and the R U , but in a range of one order of magnitude smaller compared to the BPMFC. Since the location of the water formation reaction and the additional need for H 2 O as a reactant at the cathode in the BPMFC is signicantly changed compared to the PEMFC, we believe that the alternating change of the internal cell resistance may be a consequence of an unbalanced water transport within the MEA. Further investigations will be necessary to understand this phenomenon in more detail.
The polarization data collected aer the constant current experiment (Fig. 12) reveals different insights into the aging processes of the two cell congurations.
When looking at the lower current density regime for both FC congurations, the polarization characteristics until 0.6 V changed only marginally aer 15 h current hold at 250 mA cm À2 for the BPMFC, whereas a greatly increased overpotential can be observed for the PEMFC. At a potential of 0.45 V the resulting current density has decreased by 40 mA cm À2 and 375 mA cm À2 for the BPMFC and the PEMFC respectively. But despite the reduced degradation of the BPMFC observed for potentials above 0.45 V, a characteristic mass transport restricted voltage loss could be observed for potentials below 0.3 V aer the constant current hold. We assumed that the mass transport losses aer the current hold either arose from an excessive ooding of the CL micropores, which is followed by a strongly increased oxygen transport resistance, or from changes of the ion and water transport properties towards and from the AEM|PEM interface, which could limit the water recombination rate and as a consequence reduce the limiting current density. We exclude delamination of the AEM|PEM interface, since no signicant change of R U could be observed before and aer the constant current hold. For future work it is strongly necessary to further investigate such degradation phenomena occurring in BPMFC congurations. In summary there are different possible reasons for the shown degradation, but currently we cannot be certain as to the cause of the degradation.

Conclusion and outlook
This work demonstrated a facile manufacturing approach for a MEA with a non-PGM cathode catalyst layer operating at high pH, a proton conducting membrane and a low-Pt loading anode catalyst layer operating at low pH. Polarization data and EIS analysis revealed the interface at the alkaline CCL and the PEM (AEM|PEM-interface) as a crucial element for designing MEAs operating in a bipolar conguration. According to the EIS analysis the AEM|PEM interface directly inuences the measured apparent charge-transfer resistance, even at low current densities, which leads to the conclusion of the necessity of an altered equivalent-circuit model for the interpretation of  BPMFCs compared to the interpretation of conventional PEMFCs. The ohmic cell resistance of the BPMFCs can be an adequate indicator for the quality of the respective AEM|PEM interface, as pressure dependent measurements could estimate the characteristics of the lamination and adhesion between the electrode and the PEM. In our opinion an increased surface area of the AEM|PEM-interface facilitates the lamination and adhesion of the two polymers on the one hand and decreases the overall resistance for the water-formation reaction on the other hand. A comparison between the best-performing BPMFC and a conventional PEMFC still pictured a superiority of the PEMFC regarding the measured maximum power density, but a constant current experiment with an identical starting potential of 0.45 V indicated a lower catalyst degradation rate. In summary we see two main advantages of a bipolar MEA architecture (alkaline CCL and acidic ACL) over the conventional PEMFC and AEMFC technology for the employment of PGM-free ORR catalysts. Firstly, existing technology can be used to provide a more stable cathode operating environment, compared to the conventional PEMFC. Additionally, and in contrast to AEMFC technology, it is still possible to prot from low-Pt loadings on the anode side. The still relatively low power densities might be drastically enhanced by the optimized selection and rened processing of already existing materials. Additional improvements may result from an enhanced focus on the manufacturing of the bipolar AEM|PEM interface for fuel cell applications. For future work, the role of the interfacial AEM-layer will be investigated in more detail, by the adjustment of the polymer properties and possible catalytic active additives to facilitate the water recombination at the respective AEM|PEM interface.

MEA-preparation
The ink for the Pt/C-anode-GDE fabrication comprised of a total 1 wt% solids in a solvent mixture of 20 wt% isopropylalcohol (IPA) in H 2 O. The solid fraction consisted of 70 wt% Pt/C (HiSPEC4000; 40 wt% Pt on carbon) and 30 wt% ionomer (Naon D520; DuPont). The ink was homogenised at 0 C with an ultrasonic horn (Hielscher) at 60 W for 20 min. The catalyst ink was applied onto a Freudenberg H23C8 gas diffusion media with an ultrasonic spraycoater (Biouidix) on a heated stage at 85 C. The ink ow-rate and the movement-speed of the sprayhead was controlled to a deposition rate of approximately 6 mg Pt cm À2 per deposition cycle, until a total Pt-loading of 0.1 mg cm À2 was reached. The Pt loading of the GDEs was measured by weighing (Sartorius Cubis®, 0.001 mg) the samples before and aer the catalyst ink spray deposition. The PGM-free high pH cathode was fabricated from an ink comprised of a total 10 wt% solids in 1-propanol. The solid fraction was made up by 70 wt% Fe-N/C (Pajarito Powder) and 30 wt% ionomer (Aemion HNN8-00-X, Ionomer) with IEC > 2.4. The ionomer was dissolved in the solvent and then added to the catalyst powder. The resulting ink was mechanically stirred for one hour, placed in an ultrasonication bath for one hour, stirred overnight and sonicated again for another hour on the next day. Aer that the ink was stirred until usage. The ink was applied onto a Freudenberg H23C8 gas diffusion media (4 Â 4 cm) with an automated lm applicator (ZAA 2300, Zehnter). The wet lm thickness was determined by adjusting the gap-height on the doctor blade. The gap-height was set to 350 mm, which resulted in an average loading of $1 mg cm À2 . Aer that the samples were dried at 40 C for 1 h and at 40 C under reduced pressure for an additional hour. The Fe-N/C loading of the GDEs was measured by weighing (Sartorius Cubis®, 0.001 mg) the samples before and aer the catalyst ink deposition and solvent evaporation. For the fabrication of the AEM layer on the high-pH GDE, a 10 wt% solution of ionomer (Aemion HNN8-00-X, Ionomer) in DMSO (for gas chromatography, Sigma Aldrich), was applied to the previously prepared GDEs with the automated lm applicator and a doctor blade gap height of 100 mm. The samples were dried at 40 C for four hours and at 40 C under reduced pressure for two hours. Aer that the samples were placed in 1 M KOH for 48 h for a complete ion exchange. The samples were rinsed with H 2 O multiple times, until a neutral pH was measured in the washing solution. The samples were dried at room temperature. For the direct deposition of the proton exchange membrane in the PEM|PEM BPMFC conguration, a 20 wt% Naon dispersion (D2021, DuPont) was applied to the high-pH cathode and the low-pH anode with the automated lm applicator. The gap-height at the doctor blade was adjusted to 150 mm, which resulted in a PEM thickness of $7.5 mm on each electrode. For the AEM|PEM BPMFC conguration the 20 wt% Naon dispersion was applied to the low-pH anode only. The doctor blade gap height was set to 300 mm, resulting in a PEM thickness of $15 mm. All samples were dried at room temperature for at least two hours. For the PEMFC-reference the low-pH cathode was fabricated from an ink comprised of a total 10 wt% solids in a H 2 O/2-propanol (2 : 1 v/v) mixture. The solid fraction was made up by 55 wt% Fe-N/C (Pajarito Powder) and 45 wt% ionomer (Naon D2021, DuPont). The ink was further processed in the identical manner as described prior. The ink was applied to the gas diffusion media with a doctor blade gap height of 300 mm. Aer that the electrodes were dried at 40 C for 1 h and at 40 C under reduced pressure for an additional hour. It was necessary to perform the coating and solvent evaporation two times to achieve a total Fe-N/C loading of 1 mg cm À2 .

Fuel cell testing
The fuel cells were assembled using 150 mm glass-bre enforced PTFE gaskets for the anode and 230 mm for cathode side and a 50 mm PTFE foil as a sub-gasket between the GDEs. The subgasket had an opening of 2 Â 2 cm, which reduced the active area of the fuel cell to 4 cm 2 . 15 No additional hot-pressing steps were performed. The cell was mounted with a torque of 5 N m. The fuel cells were operated at 80 C with 200 kPa symmetrical backpressure on a Scribner 850e (Scribner Associates) under power-optimized conditions with H 2 (0.25 L min À1 ) and pure O 2 (0.5 L min À1 ) and a relative humidity (RH) of 100%. The polarization data was measured galvanostatically with a stepsize of 0.05 A per point and 1 min per point up to 1 A and at higher currents with a step-size of 0.2 A per point and 1 min per point. The polarization data collection was repeated, until no signicant changes could be observed to the prior polarization data. The integrated on-line current-interrupt measurement was used to evaluate the ohmic cell resistance (R U ) at every recorded polarization data point. 19

EIS-measurement
The EIS data were collected galvanostatically with the Scribner 885-HS Fuel Cell Potentiostat (Scribner Associates). Before the EIS measurement a constant current hold was performed at the desired current density for 15 min. The frequency sweep was performed from 10 kHz to 0.1 Hz (10 steps per decade), with an AC RMS amplitude of 5% of the DC current.

Preparation of SEM samples
The MEA and GDE cross sections were obtained by sanding and polishing an embedded and inltrated sample. For this purpose, we embedded the MEAs and respectively GDEs in an Epoxy Resin (Buehler Inc, EpoThin). Inltration was facilitated by placing the mould with liquid epoxy in a desiccator and applying a vacuum. Following overnight curing of the resin, we ground the sample using SiC sanding paper (Struers GmbH) up to 4000 grain. The sanding was followed by a two-step polishing procedure. For the rst step a 3 mm polishing agent (ATM GmbH) was used with a MDdac polishing plate (Struers GmbH). This was followed by polishing using a 250 nm Diamond solution (ATM GmbH) with MDmol plate (Struers GmbH). In a nal step the samples were coated with a conductive layer of carbon using a thread evaporation coater (Balzers Union, MED 010).
The samples for surface images where prepared on aluminum sample stubs. First, a small piece of each sample (ca. 5 Â 8 mm) where cut out from the GDEs using a razor blade. Second, the pieces where attached to aluminum sample stubs using carbon adhesive tabs (Plano GmbH).

SEM imaging
All SEM images where recorded using a Gemini II electron column of a Zeiss Crossbeam 540 microscope. The imaging parameters are given in the table below. The cross-section images where recorded using the microscopes four-quadrant backscatter detector in compositional mode (5 kV accelerating voltage, 2 nA beam current). The surface images were recorded using the secondary electron detector (3 kV accelerating voltage, 2 nA beam current). EDX maps were recorded using a silicon dri detector (X-Max 150, Oxford Instruments), using an accelerating voltage of 5 kV and a beam current of 2 nA. The data was processed using Aztec (Oxford Instruments).

XPS measurements
Measurements were carried out using a Quantera II (Physical Electronics Inc.). Cutouts where made from GDEs at 2 stages of the manufacturing process: following the deposition of the catalyst layer and following the deposition of the AEM interlayer. The samples were not treated in KOH and thus the ion exchange has not taken place yet. The samples were mounted on the XPS sample stage using double sided tape. The XPS spectra were analysed using CasaXPS (Casa Soware Ltd.).

Conflicts of interest
There are no conicts to declare.