Continuous-flow electroreduction of carbon dioxide

Abbreviations: BID, Barrier ionization discharge detecto DEMS, Differential electrochemical mass spectrometry; daic efficiency; FT-IR, Fourier transform infrared spectro MEAs, Membrane electrode assemblies; NMR, Nuclear PVA, Poly(vinyl alcohol); PEM, Polymer electrolyte mem Rs, Solution resistance; SCCM, Standard cubic centimet copy; UV, Ultraviolet; Ru, Uncompensated resistance; diffraction * Corresponding author. E-mail address: janaky@chem.u-szeged.hu (C. Jan aky


Introduction
T a g g e d P Finding adequate solutions for a diversified and sustainable energy supply is undoubtedly one of the grand challenges of our society today [1]. It is imperative that renewable energy sources and solar/wind energy in particular, are increasingly used to improve the security of energy supplies and also ameliorate the environmental impact from carbon-based energy production and consumption. While solar and wind electricity generation already enjoy an important, and impressively increasing role in the global (and especially European) energy mix, storage is still an issue because of the intermittency of most renewable energy sources [2]. At the same time, the steeply rising level of carbon dioxide (CO 2 ) in the atmosphere calls for conceptually new approaches to capture and utilize this greenhouse gas. A solar fuels-based economy tackles the above parallel challenges admirably well, although many challenges remain before widespread use of such energy carriers (e.g., H 2 , methanol, ethanol, and methane) sees the light of day.
T a g g e d P CO 2 is a greenhouse gas; therefore using renewable energy to convert CO 2 to transportation fuels and commodity chemicals is a value-added approach to simultaneous generation of products and environmental remediation of carbon emissions. The large amounts of chemicals produced worldwide (Fig. 1) that can be potentially derived from the hydrogenation of CO 2, highlights further the importance of this strategy. Several industrial entities are interested in such technologies, ranging from energy/utilities companies through cement producing and processing firms to oil and gas companies.
T a g g e d P There are numerous routes for converting CO 2 to transportation fuels and other chemicals. The following three major pathways delineate how sunlight can be used to generate such products (e.g., CH 4 or CH 3 OH) from CO 2 (Fig. 1) [2À7].
T a g g e d P Photochemical (PC) or photosynthetic methods: Directly use sunlight to photochemically convert CO 2 to fuels using molecular-or suspended semiconductor (SC) photocatalysts [8À10].
T a g g e d P Electrochemical (EC) approaches: Here sunlight is first converted to electricity by a photovoltaic solar cell (PV) and CO 2 is then reduced electrochemically [11,12].
T a g g e d P Photoelectrochemical (PEC) route: Photogenerated electrons are utilized to reduce CO 2 either directly at a SC/electrolyte interface or indirectly employing a redox mediator [13].
T a g g e d P With the recent rapid drop in the cost of Si solar cells, the price of solar electricity has decreased to a level that in over 20 countries translates to grid parity. A recent study concluded that on a 20-25 year term it is not likely that any solar energy utilization pathways other than Si solar photovoltaic panels will have an industriallyrelevant role [14]. Another techno-economic analysis suggested that PV + EC conversion setups may attain »14% solar to H 2 efficiency (20% PV, 70% EC) in an economically feasible manner as the electricity price drops (which is clearly the case for both solar and wind power) [4]. These factors suggest that CO 2 conversion, at least on a short to intermediate term, will be driven in an EC configuration (note also the availability of other renewable electricity sources, such as wind). On a longer term basis, the PEC strategy also cannot be ruled out and in fact, further extensive research work is highly encouraged [4].
T a g g e d P Electrochemical (EC) and photoelectrochemical (PEC) conversion of CO 2 are multi-electron in nature (up to 8 efor conversion to methane) with considerable kinetic barriers to electron transfer. It therefore requires the use of carefully designed electrode surfaces to accelerate electron transfer rates to levels that make practical sense. In this vein, much has been written about the electrochemical, solidstate physics, theoretical, catalytic, and general materials science aspects of EC/PEC CO 2 reduction [15À17]. During the past 5 years, however, an accelerated progress has occurred, reflected in the number of published research articles and the citations they attracted (Fig. 2). Most of this work has focused on the development of new catalysts [18] and the enhancement of product selectivity [11,19]. Excitingly, we appear to be at the very cusp of a new era of electrochemical CO 2 conversion studies, which hopefully will lead to efficient CO 2 electrolyzers on a practical scale.
T a g g e d P At this juncture, however, it has to be noted that CO 2 reduction is a lot more complex than water splitting, simply because many different products can be formed via proton-coupled multi-electron transfer [20]. To drive this process in an economically attractive way, it is important to produce (i) any product as selectively as possible; (ii) products of economic value; and (iii) products that are easy to separate. A recently-performed techno-economic analysis on the process suggested that the picture is even murkier, because coproducing a low value product (such as methanol or ethylene) together with a high value product (such as formic acid or carbonmonoxide) can be a better strategy than producing them alone. In fact, the optimal scenario would be to co-produce two products that are in different phases (i.e., one in the liquid phase and the other in the gas phase) as the separation process becomes straightforward in this case [21].
T a g g e d P In addition, since the redox potential for proton to H 2 transformation is very close to the redox potential of the desired CO 2 reduction processes, there is always a competition between these two processes. Furthermore, although thermodynamic considerations would allow reduction of CO 2 at moderately negative potentials, electrochemical reduction of CO 2 is kinetically daunting with high overpotentials needed for its conversion to hydrocarbons and alcohols. Finally, in a simple batch reactor, the maximum achievable rate T a g g e d P for the reaction is often limited by the low solubility (»30 mM) of CO 2 in water. Using a simple semi-infinite diffusion model, the limiting current density attainable at this concentration is approximately I L = 60 mA cm -2 under vigorous stirring [22]. Importantly, the solubility of CO 2 strongly depends on the temperature, the solution pH, and the ionic strength as well, which should be taken into account when comparing data measured in different solutions [23]. T a g g e d P To increase the CO 2 conversion rate to a level of practical significance, EC CO 2 reduction must be performed in a continuous-flow setup to overcome mass-transport limitations. We note that there is a striking difference here compared to water splitting, where ample amounts of water molecules (55.5 M!) are available for the reaction. Interestingly, despite some successful pioneering studies [24], very little attention has been devoted to EC CO 2 reduction in continuous flow mode. Only a minor fraction (»5%) of the articles (see Fig. 2C) report on studies performed under such flow conditions. In fact, this trend can also be very problematic in the sense that conclusions drawn from batch experiments cannot be directly translated to flow situations (unlike for water splitting).
T a g g e d P There are a few reviews on electrochemical cell designs for CO 2 reduction: namely microfluidic reactors [25,26] and polymer electrolyte membrane (PEM) electrolyzers [27]. Solid-oxide electrolyzers, while an important category, fall outside the purview of the present study, because of their vastly different operational principles and conditions (high temperature, single CO product, etc.) [28]. In a continuous flow CO 2 electrolyzer, multiple parameters have to be simultaneously tailored to possibly achieve an economically viable process [21]. In two recent articles careful modeling on the effect of the various operational parameters (pH, concentration, temperature) was carried out, mostly through studying the CO 2 /carbonate family equilibrium [29,30]. In this review, we focus on real operational cells: materials aspects, device-related features, together with operational parameters are the three main categories forming the crux of this article. We compare various reports in the literature, to identify the role of the individual parameters, and to set guidelines for future studies. We also incorporate the lessons learned from the fuel cell and water electrolyzer communities, while highlighting also the most important differences.
T a g g e d P The prematurity of the field is also reflected in the lack of precise definition of electrochemical flow cells. A simple literature search for "flow-electrochemical cell" or "continuous-flow electrolyzer" can easily mislead the reader. For example, some studies in which the liquid phase is continuously purged with systematically varied flux of CO 2 during the measurement are claimed to be on "flow" systems. Although very important conclusions can be drawn from these studies on the effect of the different reaction parameters (e.g., gas flowrate, electrode composition, pH), these setups significantly differ from those À constituting the core of this study À in which a fresh solution/gas phase is fed to the electrodes continuously as detailed later [31]. Similarly, while in some cases both the catholyte and the T a g g e d P anolyte is continuously pumped (and thus refreshed), in other instances only one of them. It is also common that the circumstances of the measurements are not precisely defined, and therefore it is very difficult to determine whether the experiments were performed in a real flow electrolyzer or not.
T a g g e d P Finally, in our opinion, there is a strong need for setting the proper measurement protocol for assessing the properties of continuous flow EC cells for CO 2 reduction and for specifying a set of parameters for benchmarking purposes. Such an exercise is particularly relevant because researchers from diverse specialties (e.g., heterogeneous/homogeneous catalysis, fuel cells, water splitting, etc.) and even disciplines (chemistry, physics, and engineering) have congregated in the EC CO 2 reduction arena. These disparate communities often use very different terminologies. Hark back to the history of the fuel cells and water electrolyzer communities where it took decades to establish generally-accepted testing protocols and benchmarking parameters. In the interim period, literally hundreds/ thousands of reports accumulated in the literature describing results that could not be reliably intercompared with one another.

Reactor designs
T a g g e d P Whether we consider technology scale-up or reliable rapid screening of materials in the laboratory, continuous-flow reactors have multiple benefits compared to their batch counterparts. Among others, these include increased mass transfer and improved mixing of different phases, better temperature and heat transfer control, and more precise influence on reaction mixture residence time in the reactor [32]. When moving from batch-type experiments to continuous-flow cells, the architecture and design of the reactor (electrochemical cell) must be first clearly defined. Note that this includes: the type of electrolyte used (liquid/gas), reactor materialreaction mixture compatibility, whether it can be pressurized or not, the applicable temperature and flow rate, and the possibility of using a reference electrode.
T a g g e d P Schematic drawings of the most frequently studied cell configurations are presented in Fig. 3. Before discussing each of them individually, we would like to emphasize the importance of the number of electrodes employed in the setup. Many designs only involve two electrodes (a working and a counter electrode); thus only current or voltage (not potential) control is possible in these cases. There is a considerable number of electrolyzer setups where a reference electrode is also integrated into the cell, close to the working electrode surface, to ensure the possibility of potential control. A fourelectrode setup is also possible (with two reference electrodes), if the water oxidation step is complicated, and thus the monitoring of both half-cells becomes important. Potential control is especially important in electrolyzers in which the catalysts (e.g., copper oxides [20]) change their chemical composition (or surface properties) Fig. 2. Results of a literature survey on electrochemical and photoelectrochemical CO 2 reduction. A: number of papers published, B: the citations gathered by these papers, C: papers claimed to use a flow-setup (this is a subset of Fig. 2A).
T a g g e d P during operation (i.e., aging). This often results in a change in the product distribution with time, as discussed in what follows.
T a g g e d P Fig. 3A shows the most universal architecture, from which all other setups can be derived. This parent configuration consists of two flow channels, one for the anolyte and another for the catholyte, separated by an ion-exchange membrane. The cathode electrocatalyst is immobilized on a gas diffusion layer (GDL), which is in contact with the flowing catholyte from one side, while CO 2 gas is directly fed on the other side [33]. This arrangement overcomes most of the problems associated with the other setups, namely: (i) current limitation due to the low concentration of CO 2 at the electrode; (ii) H + crossover from the anode and the consequent acidification of the catholyte; (iii) difficulty of inserting a reference electrode; (iv) diffusion of liquid-phase products to the anode, where they are oxidized (product crossover). Although no such instrument is commercially available on the industrial scale at the moment, most components of this setup (i.e., GDLs and catalysts) are available and ready for scaleup already. We note here that it seems to be very challenging to build large stacks based on this concept mostly because of the pressure sensitivity of this structure [34]. We expect therefore that the parallel operation of these instruments (the scale-out concept) will constitute the main track of industrialization in this case À at least in the near future. The other experimental devices (Fig. 3BÀE) can be derived by "removing" different elements of this general setup, thus T a g g e d P simplifying the overall architecture. By applying a single, undivided channel for flow of the common electrolyte, we get a typical microfluidic reactor (Fig. 3E). In this setup, the electrodes are separated by a thin spacer (usually well below 1 mm in thickness) and no membrane is included [35À39]. The reference electrode can be inserted in the electrolyte flow stream and the excess protons formed on the anode are drained from the cathode vicinity by electrolyte flow (although not as effectively as in the case of separated electrolyte channels).
T a g g e d P A popular class of devices feature separation of the two electrodes by an ion-exchange membrane (typically Nafion TM , but see also Section 3.5. for others) (Fig. 3BÀD). Usually the electrodes are pressed together and no flow channels are formed between them (zero gap cells). The electrolyte/gas is fed to the electrodes (mostly gas diffusion electrodes (GDEs), formed by immobilizing the catalyst on a GDL), and remains in the cell until reaching the exit point. The greatest advantage of these setups, compared to the microfluidic arrangements, is that it is relatively easy to pressurize the reactant and product flows. Furthermore, based on the similarity of these devices to PEM water electrolyzers, the scale-up of these setups to the industrial scale and construction of large sized stacks seems to be more straightforward.
T a g g e d P Variants of these setups differ in the type of electrolytes used. The reactants fed to the cathode and anode compartment can be (i) both T a g g e d P in liquid phase (Fig. 3B) [40]; (ii) CO 2 gas on the cathode and liquid phase anolyte (Fig. 3C) [41,42]; (iii) humidified gases on both electrodes (Fig. 3D) [43,44]. Although this may look like a minor difference at first sight, the reactant type has an important and complex effect on cell performance. When feeding pure CO 2 to the cathode, the reactant concentration remains very high on the catalyst, and therefore high reaction rate (current) can be achieved. In this case however, trustworthy measurement of the individual electrode potentials is far from trivial, although several promising attempts were made with all-gas fuel cells [45À48]. Previous studies have shown that the product distribution changes parallel to continuous ageing of certain catalysts during long-term operation [49,50]. The lack of a reference electrode complicates this situation even more, since in this case the change in the cell voltage/current stems from either anode or cathode degradation (or both), whose effects cannot be separated.
T a g g e d P The main drawback of these membrane-separated zero-gap devices paradoxically is the same as their main advantage, namely the proximity of the two electrodes to one another. This decreases the cell resistance and consequently, the IR-drop to a minimum level, but on the other hand, ion exchange leads to acidification of the catholyte and therefore to increased production of H 2 (instead of reduction of CO 2 ). Including a buffer layer between the electrodes can circumvent this effect, but at the same time, leads to increased cell resistance [51]. Although most studies on PEM CO 2 electrolyzers focus on the use of cation exchange membranes, anion exchange membranes may also bear considerable scope in CO 2 conversion. In such arrangements, OH ¡ ions are transported through the membrane, which results in a different product distribution compared to cation exchange membrane containing systems [52]. A new anionexchange membrane based electrolyzer was recently developed for both CO 2 conversion and water splitting [53]. The reactors, using the new anion exchange membrane based on polymers containing imidazolium and pyridinium groups, exhibited high durability at industrially relevant current densities (100 mA cm ¡2 for CO formation) [53]. Detailed discussion on the role of the ion-exchange membrane is given in Section 3.5.
T a g g e d P It is very important to underline that for industrial applications, large-sized, multiple-stack electrolyzers are required. The current status of this field is very far from this level, since aside from a few examples, all studies were carried out on 1À10 cm 2 sized electrodes [54,55]. Consequently, a grand challenge for future research and development is to construct experimental setups that can be readily translated to real industrial technologies.
T a g g e d P The operational principle of solid oxide electrolyzers is shown in Fig. 3F. Detailed discussion however, is omitted because of their completely different properties compared to both their microfluidic and PEM counterparts [28].

Materials
T a g g e d P As outlined in the previous section, various device architectures can be used for the continuous-flow electroreduction of CO 2 . There are certain components, however, which are common to all electrolyzer designs, and they are discussed in what follows (also see Table 1 for an overview).

Electrocatalysts
T a g g e d P As it dictates both the kinetics and thermodynamics of the electrolysis process, the electrocatalyst is the heart of such devices. The most extensively scrutinized cathode catalysts and their most important features are summarized in Table 1. When comparing and contrasting these to the catalysts employed in batch experiments we can conclude that (i) the catalyst candidates that proved to be promising in batch setups show similar, or even, higher T a g g e d P electrochemical activity in the corresponding flow setup; (ii) the formed products are very similar for both batch and flow experiments; (iii) the reported potential values differ significantly in many cases. The most intensively studied electrocatalysts are Sn, on which formate is produced almost exclusively; Ag with preferred CO formation; and Cu, on which a wide variety of products is formed, depending on the experimental circumstances. Several catalysts (e. g., metal alloys) that were promising in batch experiments however, are still waiting to be tested in flow cells. An example of new generation catalysts is metal organic frameworks (MOFs), which were very recently studied in flow-reactors [56]. Much less effort has been devoted to the anode catalyst, where mostly IrO x and Pt was employed to facilitate water oxidation (O 2 evolution) [57]. In the outlook section, we present some possible future R&D avenues, where the anode electrocatalyst gets higher importance.
T a g g e d P The amount of the electrocatalyst in these devices varies in a relatively broad range, from 0.2 to 10 mg cm -2 . The most commonly applied catalyst loading however, is around 1 mg cm ¡2 , independently of the catalyst used. This latter fact is indeed very surprising, since the molar mass, density, and specific surface area of the different catalysts can differ severely. This may cause several orders of magnitude difference in the number of electrocatalyst atoms, and more specifically in that of the surface atoms, which can interact with the CO 2 molecules. To get a meaningful comparison on the electrocatalytic properties of different catalysts, one must therefore always normalize the measured current values in terms of either the electrochemically active surface area, or with the number of surface atoms, but not with respect to the geometric surface area or electrode mass. This is even more important in the case of thick porous electrodes, where the current density can be influenced by electrode thickness, without affecting the electrode kinetics.
T a g g e d P At this juncture, however we note that researchers need to make sure that all measurements were taken in the kinetically-controlled regime and not the mass transfer-limited regime, before they compare reaction rates across different catalytic systems. Different strategies can be employed to determine mass transfer limitations, for example obtaining breakthrough curves (current density vs. catalyst loading) [78]. An example is shown in Fig. 4, where LSV curves are shown for Sn-based GDEs with different Sn-loading. As seen in Fig. 4A, after a certain Sn-content, the mass-transport limited regime is reached. This trend is directly visualized in Fig. 4B, where the partial current density related to CO 2 reduction is plotted vs. the catalyst loading (note the constant FE values, confirming that the chemical process is identical).
T a g g e d P It should also be noted that although attempts were made to directly compare the results of batch experiments with those measured in flow setups, such a comparison is not straightforward because of several reasons [33]. The continuously refreshed solution, reaching the electrode surface leads to striking differences as it has a massive influence on the (i) mass-transport (diffusion layer thickness); (ii) local pH effects; (iii) product accumulation in the close vicinity of the cathode; (iv) residence time of CO 2 molecules at the electrode surface. When applying (humidified) CO 2 gas as "catholyte," its effective concentration on the surface is obviously higher than in the case of aqueous solutions, which leads to higher current densities at the same "potential". Here we refer to our earlier point, namely that defining and measuring electrode potential in such arrangements is also problematic.

Effects of catalyst size and morphology
T a g g e d P Particle size effects [79] have not been extensively studied in flow cells; however, there are some nice examples on copper nanoparticles and thin films (3-21 nm thick) [80]. First it was shown that nanoparticulate Cu (n-Cu) behaves differently than Cu foils (Fig. 5A). In addition, from the film thickness dependent methanation Faradaic T a g g e d P efficiency and gravimetric methanation current experiments, it was demonstrated that thin evaporated films behaved like the n-Cu/C electrodes while thick evaporated films behaved like copper foils (Fig. 5B). In this study, it was also suggested that n-Cu was ideal for preparing GDLs with lowered polarization losses, thus maximizing the energy efficiency of the electrolyzer [80].
T a g g e d P In another study, Cu electrodes with different morphologies were prepared [81]. The first important observation was that higher active surface area resulted in an improved total FE compared to the smooth Cu plate. As for the product selectivity: (i) electroplated Cu T a g g e d P (on Cu foil) favored the production of formate; (ii) electrofaceting of the Cu foil moved the selectivity towards CH 4 formation; (iii) deposition of Cu onto carbon cloth resulted in the formation of C 2 H 4 [81]. The effect of tin loading and particle size were studied in a filterpress cell for CO 2 reduction [82]. Tin particles of different sizes (150 nm, 10 mm, and 150 mm) were studied, and certain size effects were reported with an optimal behavior for the smallest particles. Again, meaningful normalization is a mandatory exercise in such studies, to deconvolute the simple surface area effect from other, chemical underpinnings.

Role of the catalyst support
T a g g e d P To achieve current densities of practical relevance, it is vital to increase the electroactive surface area of the electrocatalyst. Moreover, in the case of flow electrolyzers, one must also consider that the electrolyte flow continuously removes the products (or intermediates) from the electrode surface, leading to relatively short remanence time. This severely influences both the Faradaic efficiency and product distribution. Therefore, the CO 2 containing electrolyte must be forced to travel through a long path, while it remains in contact with the electrocatalyst for a sufficiently long time period. This can be achieved by immobilizing the catalyst on a GDL, which is a porous substrate with large surface area. The GDL has a dual function in the cell by allowing transport of materials between the catalyst and the flow channel while also maintaining proper electronic communication between the current collector and the electrocatalyst. The most frequently (almost exclusively) applied GDLs are porous carbon supports, formed from carbon fibers or pressed carbon particles. These carbon cloths and carbon papers are T a g g e d P frequently used supports in fuel cells and water electrolyzers on the cathode side. Interestingly, although several attempts were made to employ different metal foils or meshes as catalyst supports in CO 2 flow electrolysis cells, the use of different carbonaceous substrates is reported in almost all the recent studies. As a specific example, we mention the comparison of two continuous-flow electrolyzers, where an In metal foil electrode was compared with In nanoparticles (100À300 nm) immobilized on a carbon GDL. Seven-fold higher partial current densities towards HCOOH formation were detected in the latter case, compared to the simple indium foil [83].
T a g g e d P Fig. 6 summarizes the typical components of a GDE, which includes the GDL, the microporous layer, and the catalyst (possibly together with an ionomer) [78]. The importance of the tri-phasic solid/liquid/gas interface is also highlighted. The effect of GDE structure (e.g., thickness, porosity, and density) on the electrocatalytic properties was extensively studied in fuel cells and water electrolyzers; much less attention has been dedicated to these parameters in the case of flow CO 2 electrolyzers so far. As an exception, optimization of the gas diffusion electrodes consisting of a carbon fiber  T a g g e d P substrate, a microporous layer, and a catalyst layer was performed very recently [84]. The optimized electrode exhibited a higher partial current density for CO production than Sigracet 35BC, a commercially available GDE. Overall, we are convinced that there is a significant opportunity for the rational design of GDEs for further improvement of such devices.
T a g g e d P Another very important aspect in the case of flow electrolyzers working with liquid electrolytes is the hydrophobic/hydrophilic properties of the GDL. In the case of fuel cells, an important role of the hydrophobic GDL is the removal of excess water. The case is however different for flow CO 2 electrolyzers: the solution must be kept in contact with the catalyst surface for a sufficiently long time. Proper wetting of the top of the carbon support therefore is a prerequisite, but too much hydrophilicity should also be avoided, because H 2 evolution would be favored in that case. Different ionomers are employed to circumvent the above contradiction, which in turn also contributes to the fixation of the catalyst on the surface of the gas diffusion electrodes (Fig. 6). We also note that controlling the interfacial chemistry between the components of the electrode assembly is of prime importance to ensure high performance and stability at the same time.

Catalyst immobilization
T a g g e d P The technique employed for the catalyst immobilization has a decisive influence on the performance of flow electrolysis setups. The most often used techniques include two steps: first the T a g g e d P electrocatalyst is prepared and subsequently, it is transferred to the GDL (or other substrate) or more frequently to the membrane to form the so-called membrane electrode assembly (MEA), via a physical method. We note here that no "standard" catalyst immobilization method exists, but paint-and air-brush techniques are the most common ones. As demonstrated in a recent study, the immobilization technique influences both the measured current values and the product distribution (Fig. 7) [58]. In addition, according to our own experience, the catalyst deposition method affects the stability of the cell as well: channel formation and degradation can be observed if the catalyst is not evenly dispersed, rooted in the uneven distribution of the current flow (see also Fig. 7 for catalyst distribution details).
T a g g e d P In situ deposition methods, in which the catalyst is directly formed on the substrate, constitute the other large, but less frequently applied class of immobilization techniques. Electrochemical deposition of different catalysts, exploiting the conductive nature of the carbonaceous GDLs is a particularly promising avenue to form such architectures. The catalyst/GDL structure is thus formed in a single step, and intimate electrical and physical contact between the electrocatalyst and the carbon substrate is ensured. Further, after careful pretreatment of the GDL layer (to tune its hydrophobic/ hydrophilic character) not only the top of the GDL is decorated with catalyst (nano)particles, but the inner regions as well. This can lead to increased current values because of the enlarged electrochemically active surface area. A recent study presented the electrodeposition of Sn on carbon fibers thus forming a GDE as a promising T a g g e d P example [85]. Note however that this approach requires careful optimization of the synthesis circumstances for each catalyst/GDL pair, and is therefore not compatible with rapid screening needs.
T a g g e d P One further aspect to consider when immobilizing a catalyst is the use of any binder material (e.g., Nafion TM dispersion). Adding an ionomer to the dispersion of the catalyst before transferring it to the membrane/GDE, influences the subsequently formed catalyst layer. It has a substantial effect on (i) the electroactive surface area of the immobilized catalyst (three dimensional reaction zone) (ii) the ionic (proton) conductivity of the catalyst layer (iii) the hydrophilic/ hydrophobic property of the catalyst layer (iv) the interaction (electrical) between the catalyst layer and the membrane. The effect of the relative ionomer content was extensively studied for PEM fuelcells [86À89]. These works revealed that the cell performance can be enhanced by adding small amount (typically 10À20 wt.%) of the ionomer into the catalyst layer. This results in a large decrease in the ohmic and transport resistance of the layer and extends the active surface area where the electrochemical reaction occurs.
T a g g e d P As for the electroreduction of CO 2 , the same effect was demonstrated by forming catalyst layers of Sn nanoparticles (100 nm in size) with the same catalyst loading, but by systematically varying the ionomer content [78]. Similarly to the case of the PEM fuel-cells, an initial increase in both the current and the FE was witnessed with the increasing ionomer content, followed by a sharp decrease in the cell performance when the Nafion content was further increased.
3.5. The role of ion-exchange membranes T a g g e d P In those setups that contain an ion-exchange membrane (see Fig. 3), its properties also affect the performance of the cell. The chemical structure, thickness, and the wetting properties all have substantial influence, although their precise contribution have not been fully explored yet. While the clear majority of the studies employ cation exchange membranes, and specifically Nafion TM , as the ion-exchange membrane (see some other examples in Fig. 8), there is a considerable scope in tailoring physical-chemical T a g g e d P properties of the membrane towards CO 2 electroreduction (or even towards specific product formation) [66,90]. Chemical modification of well-known polymers with new organic mediators may open up a new avenue in the future [91,92]. Finally, in addition to chemical factors, durability and pressure handling are two equally important physical attributes to consider.
T a g g e d P Four different membranes (Nafion TM , SPEEK, alkali-doped PVA, and Amberlyst/SPEEK) were compared in a continuous-flow cell using Pt/C anode and Cu/C cathode for CO 2 reduction [52]. Both the measured current density as well as the relative amounts of the products showed a notable variation as a function of the chemical properties of the membrane. This distinction was rationalized predominantly by the vastly different ionic conductivity of the membranes [52]. In another recent study several chemically different membranes were compared, and an ionic-liquid inspired polymer (anion exchanging Sustainion TM ) was introduced, providing both high efficiency and stability [93]. For those applications where H 2 formation is to be avoided (e.g., space utilization, see also the Outlook section) the use of anion exchange membranes may help by suppressing proton reduction. This trend was demonstrated when an anion-and cation exchange membranes were compared using a Ag-cathode [94]. Most recently, bipolar membranes have offered enhanced water splitting via steady-state pH gradients and product separation, and these novel materials may be worth exploiting for CO 2 reduction studies as well [95,96]. Finally, we note that product crossover is an important factor to be considered for these membranes from the early development stage.
3.6. The role of the current collectors, bipolar plates, and cell body T a g g e d P The mechanical stability of an electrolyzer is ensured by the cell body, the current collectors (end-plates), and the bipolar plates. As the cell body is a passive element in the setup, the most important requirements to fulfill are chemical inertness and mechanically stability, hence it must not change its shape or dimensions during the electrolysis. This is particularly relevant when the cell operates at high pressure and/or high temperature conditions. The cell body is therefore most frequently made of stainless steel, and its two parts are pressed and held together by several steel screws.
T a g g e d P When considering the industrial application of continuous-flow CO 2 electrolyzers, we must distinguish between the microfluidic and the PEM arrangements (see again Fig. 3). While for the first group the scale-out strategy (and therefore the parallel operation of large surface area setups) seems to be favored, the scale-up strategy seems to be the best way of industrial application of PEM CO 2 electrolyzers À very similarly to PEM water electrolyzer and PEM fuelcells. In this case, several electrocatalyst layers and membranes (MEAs) are coupled in series, separated by bipolar plates, functioning T a g g e d P as anode on one side and cathode on the other side (see a typical PEM fuel cell setup in Fig. 9 as an example). The function of these bipolar plates and end-plates is complex: (i) they form the electrodes which are in contact with the catalyst layers, (ii) as the reactants are fed to the catalyst layer through the channels formed in these plates, they are responsible for the reactants supply to the cell active area, and for the proper outlet of the products. Furthermore, they play a significant role in the water and heat management of the cell (most importantly in the case of PEM setups, fed with humidified gases) [98]. To serve this purpose, flow-channels are formed on these plates to increase the surface area, and to help the transport processes [97]. As it was shown for PEM fuel-cells, the different flowfield designs (e.g., straight flow channels, single-or multiple serpentine channels, etc.) have both pros and cons, and therefore this pattern must be always optimized towards the targeted application in the employed setup. The use of current collectors with flow-patterns in continuous-flow CO 2 electrolyzers needs extensive investigations and use of different flow-patterns might contribute to the scale-up and industrialization of this process.

Operation
T a g g e d P All factors governing the performance on a sub-reactor level were summarized in the previous sections. Now we turn the focus to the operation of the complete electrochemical cell, and review parameters that affect the EC CO 2 reduction reaction in the continuous-flow mode. Note that the effects of the discussed parameters are not independent of each other. In fact, their influences are rather complex and convoluted; therefore it is often difficult to carry out studies where only one parameter is varied. Table 2 lists selected examples of continuous-flow CO 2 electroreduction studies from the literature, where the effect of the most important factors is highlighted.

Feedstock
T a g g e d P The first question related to the input flow turns back to the reactor design (see also Fig. 3). It seems that in most studies there is a parallel feed of liquid electrolyte and CO 2 gas; however, there are reports with liquid or gas feed only. Here the reader must be very careful, because certain articles claim solution + gas flow, although when examined closely, only liquid flow (which was previously saturated with CO 2 ) was deployed. It is also worth noting that all those studies where notably high currents were reported employed both gas and solution feed. While the gas feed was almost always pure CO 2 (see discussion later on pressure effects and an exception where Ar/CO 2 mixtures were studied [85]), the composition of the liquid electrolyte varied massively throughout the studies (Table 2). This T a g g e d P variation is quite surprising, considering the prominent role the electrolyte may play in the electrochemical process [39].
T a g g e d P The role of the electrolyte, by varying both concentration and chemical makeup, was investigated using a Ag-based gas diffusion electrode as cathode [39]. It was found that anions play a significant role in the process, as reflected in the onset potential for CO formation shifting in the order OH -< HCO 3 -< Cl - (Fig. 10A and B).
This trend was explained by the interplay of several factors such as pH, conductivity, and, more importantly, specific adsorption of certain anions on the electrode surface. Similar conclusions were drawn for a Sn-based electrolyzer, where OH -, HCO 3 -, Cl -, and HSO 4 anions were studied [99]. The effect of electrolyte anions and cations was further elucidated, by investigating four cations (Na + , K + , Rb + , Cs + ) and four anions (i.e., Cl ¡ , Br ¡ , I ¡ , OH ¡ ) [100]. A major size dependence was observed for the cations, namely an increased CO 2 reduction (and decreased H 2 evolution) activity was detected when larger cations were present in the solution. This trend was rationalized by the better hydration of smaller cations which thus are less likely to adsorb on an electrode surface. The larger cations were thought to adsorb on the cathode repelling H + ions from the cathode and stabilizing the "CO 2 ¡ " intermediate on the electrode surface [100].
T a g g e d P As for the effect of electrolyte concentration, a monotonic increase in the current density (as well as the energy efficiency) with increasing electrolyte concentration was noted (Fig. 10C and D). Electrochemical impedance spectroscopy (EIS) showed that both the charge transfer resistance (R ct ) and the cell resistance (R cell ) decreased with increasing KOH concentration [39]. The effect of catholyte concentration (varied in the range: from 0.1 mol dm -3 to 2.0 mol dm -3 ) on product (formate) selectivity was also studied using a tinned copper mesh electrode [54]. An optimal intermediate KHCO 3 concentration (0.45 mol dm -3 ) was found, and rationalized in terms of the competing effects of surface speciation, ionic conductivity, and CO 2 solubility. We note here that CO 2 -saturated T a g g e d P KOH and KHCO 3 solutions are very similar in nature, only the species distribution and the pH being different. As seen in Table 2. there are examples for the addition of an inert electrolyte (e.g., KCl or NaClO 4 ) to such solutions, further improving the electrical (ionic) conductivity without severely affecting the other parameters mentioned above [54,55,82,100]. As a general conclusion, we can state that higher electrolyte concentration leads to higher current densities, unless there is a specific adsorptive interaction with the electrode surface.
T a g g e d P Ionic liquids are an emerging class of solvents in CO 2 electroreduction [108]. These materials are special in many aspects, for example they can even stabilize the formed reaction intermediates in CO 2 reduction. Synthetic chemistry allows considerable latitude for tailoring the molecular structure of these liquids, and thereby enhance CO 2 solubility via specific chemical interactions [53,108,109]. The chemistry learned for ionic liquids might be exploited as either surface modifiers or ionomers in membrane electrode assemblies (MEAs) discussed below.
T a g g e d P The effect of solution pH is also rather complex and the conclusions cannot be generalized for different electrocatalysts and targeted reduction products. In general, at lower pH values, the formation of H 2 is more favored. However, to form hydrogenated CO 2 reduction products (e.g., formate and methanol), a lower pH is beneficial, while higher pH is suitable for CO formation. The pH effect was studied in a microfluidic reactor, employing a Sn cathode and a Pt black anode [35]. Experiments were carried out at three different pH values, and the cell voltage, the electrode potentials, and the efficiencies (energy and Faradaic) were monitored (Fig. 11). The pH is seen to exert a more prominent effect on the cathode process. At lower pH, the formation of formic acid was favored, both in terms of higher current density and selectivity [35]. This study also calls attention to the importance of having two reference electrodes in the cell, enabling the monitoring of both half-cell processes separately. T a g g e d P More recently, a dual electrolyte microfluidic reactor was designed, and the effect of using anolyte and catholyte of different pH was systematically studied. A set of electrocatalysts was investigated, and it was concluded that catholyte pH = 2 and anolyte pH = 14 resulted in an optimal cell performance. Furthermore, a three-fold increase was witnessed in the overall performance, compared to the single neutral (pH = 7) electrolyte arrangement [37].

Liquid/gas flow rate
T a g g e d P With the aim of converting large amounts of CO 2 , the question occurs instantly: how does the flow rate of the liquid/gas input influence the current density and the Faradaic efficiency (and thus the overall CO 2 conversion process)? Considering the importance T a g g e d P of this parameter, it is very surprising that there is no precise definition to meaningfully present the CO 2 flux in the electrolyzer. In almost all the cases, the unit of standard cubic centimeters per minute (sccm) was employed to characterize the gas flow, which equals cm 3 /min at standard temperature and pressure. While this unit sufficiently describes the overall gas flow, it gives only very limited information on the actual flux of CO 2 reaching the electrocatalyst surface in the electrolyzer. Consequently, it is very difficult to compare studies performed in different laboratories on electrolyzers of different size. It seems to be a useful exercise to normalize the flow rate with the electrochemically-active surface area of the electrode, and/or to the free volume of the cathode compartment of the electrolyzer, thus obtaining the actual flux of CO 2 (see below). In addition, the fact that more often than not,  T a g g e d P pure CO 2 (or solutions with different CO 2 concentrations) is fed to the cathode, implies that the flow rate should be also normalized with respect to the CO 2 content. This exercise was nicely performed in a recent study on Sn-GDE electrodes, where the Ar/CO 2 ratio was systematically varied, while the flow rate was kept constant [85].
T a g g e d P In a very recent study, copper hollow fiber electrodes were employed as both the gas diffusion layer and cathode electrocatalyst; and the effect of CO 2 flow rate on the cell performance was studied [31]. As presented in Fig. 12, there is a considerable effect on both the overall current density and the FE for CO generation. The enhancement in FE towards CO as a function of CO 2 flow rate is consistent with a concurrent increase in current density, suggesting that the FE of CO strongly depends on the efficiency of mass transfer of CO 2 to the electrode surface. After reaching a certain flow rate, the achieved steady behavior suggests that most active catalyst sites are involved in the reaction and the kinetically controlled regime was reached [31]. The importance of CO 2 transport in nanostructured electrodes was also demonstrated for electrodeposited Cu nanofoams, where a thickness-dependent selectivity was observed for the formation of formic acid and CO (vis-a-vis hydrogen evolution) [110].
T a g g e d P A similar trend was reported for a dual microfluidic reactor, where after reaching a certain threshold, further increase of the flow rate had no effect on cell performance [37]. Over this flow rate, mass transfer rate of CO 2 through the GDE becomes the bottleneck. A higher porosity and smaller thickness of the GDL may help to improve the performance, however, its coherent structure has to be maintained to sustain its mechanical stability and its electronic communication with the current collector.
T a g g e d P Besides the achievable current density, product distribution can also be tailored by the flow rate. For example, the effect of flow rate on the H 2 /CO ratio was studied in a hybrid PEM/microfluidic reactor, having a Ag cathode and Ru anode [59]. It was demonstrated that by controlling the current density and the CO 2 flow, it was possible to tailor the H 2 :CO product ratio between 1:4 and 9:1. Not surprisingly, by limiting the amount of CO 2 reaching the electrode surface, an increase in the H 2 :CO ratio was seen. At the same time, this increase in the ratio was not accompanied by any change in the cell voltage or cathode potential [59]. Importantly, after a short induction period, the product ratio was maintained for a relatively long time (5 h). These data suggest that there is considerable scope for tailoring the product ratio, for syngas formation, simply by optimizing the flow rate. Notably, such output can be directly utilized in different chemical processes, for example in the FischerÀTropsch synthesis.
T a g g e d P Finally, we highlight a study where considerable attempt was made to normalize the flow rate, although with the geometrical surface area. Experiments were carried out at three different electrolyte flow (Q) / electrode area (A) ratios and both the amount of the T a g g e d P formed products, as well as their distribution varied [50]. Even more interestingly, a Cu electrocatalyst with two different morphologies (Cu plate and Cu 2 O derived Cu on carbon support) were compared, and different behavior was seen (Fig. 13). This may be rooted, at least partially, in the different electrochemically-active surface area in the two cases, further highlighting the importance of the suggested normalization protocol (see Section 5).

Temperature and pressure
T a g g e d P As seen in Table 2, there are only few examples where the pressure and temperature were varied from ambient values. The number of systematic studies is even smaller [59,60,103,111], but it seems that higher operating temperatures correlate with higher currents (Fig. 14). This effect however is also complex, because higher temperature leads to lower CO 2 solubility (causing also a change in the pH), but higher diffusion coefficient and reactivity [29]. In these studies the temperature had little effect on the selectivity [103], but this is likely to vary a lot depending on the electrocatalyst. Over a certain temperature, current increase is associated with a decrease in FE of CO 2 conversion, and thus increased H 2 evolution [60]. Our own, currently ongoing work suggests that lower temperatures (e.g., 3°C) promote ethylene formation over methane for copper-based cathodes in a GDL-configuration. Considering the complex effects of temperature, many more studies are needed in this direction with in situ monitoring of the temperature in the electrolyzer, which is not trivial.  T a g g e d P The effect of pressure is even less studied although increased pressure: (i) increases the solubility of CO 2 according to Henry's law; (ii) enables work at higher temperatures; (iii) makes it easier to store, or to react the products (e.g., syngas) further to value-added products. The pressure was gradually increased in a combined PEM + microfluidic reactor setup (see Fig. 3A) from atmospheric to 20 atm. A five-fold increase in FE towards CO evolution was seen both at room temperature and at 60°C (Fig. 15) [60].

Effect of the applied potential/voltage
T a g g e d P The potential/current dependent product distribution has been extensively studied in batch reactors. Much less is known about continuous-flow setups, but a comprehensive study using a PEM electrolyzer (Cu cathode, Pt anode) was able to detect as many as 16 reaction products (Fig. 16). At low overpotentials, only hydrogen, CO, and formate were observed, with the dominance of hydrogen formation at very low overpotentials. As the overpotential increased, cumulative FE of CO 2 reduction products increased versus H 2 formation. First hydrocarbons (methane and ethane) were formed, while at even higher voltages, C2ÀC3 products were produced [19].
T a g g e d P The effect of the applied potential is less complex in those cases where only one product (e.g., formate, CO) is formed [38,112,113]. Note however that H 2 evolution is a competing reaction even in these instances. As the external driving force (potential/voltage) is increased, this latter becomes more and more prominent, leading to a volcano-type plot for the Faradaic efficiency of the product with the applied potential. One hand, this is disadvantageous, since the production rate of the intended product cannot be increased while maintaining a high energy-efficiency. On the other hand, this gives the opportunity to form specific product mixtures. For example, using a silver containing GDE cathode, the H 2 /CO ratio was tuned to form readily processable syngas simply by varying the applied current density (and hence the potential) [59].

Timescale of the experiments
T a g g e d P As seen in Table 2, the timeframe of the electrolysis experiments span a few minutes to a few hours. Interestingly, in many cases, no T a g g e d P stationary electrolysis was performed, and the current values were gathered from linear sweep voltammetry (LSV) experiments. There are only a few studies where the product distribution was reported as a function of time. This information would be crucially important in assessing the stability of the electrocatalyst and the entire electrolyzer assembly. For example, the H 2 /CO ratio was monitored during a galvanostatic experiment (Fig. 17), using a silver GDE. The monotonic increase in the H 2 evolution ratio (and the decrease of FE co ) was rationalized by alteration of the electrode, in particular increased wetting of the GDL [103]. A recent study, employing a solid catholyte and an anion-exchange membrane showed stable selectivity and cell voltage over 250 hours of operation at 100 mA cm ¡2 [53].   5. How to benchmark a CO 2 electrolyzer correctly?
T a g g e d P As data-mining turns into a major component of every research project, it is now more important than ever to report experimental data (and the drawn conclusions) in a manner where comparisons from laboratory to laboratory can be easily made. According to lessons learned from allied R&D fields (most importantly fuel cells and electrochemical water electrolysis), setting up a proper benchmarking protocol leads to a more coherent knowledge of these systems, and consequently to an accelerated development of these fields T a g g e d P [114À118]. Such a protocol should include characterization of both the electrochemical cell-forming components separately as well as the complete cell itself, via a set of systematic studies. We suggest the following benchmarking experiments to be carried out and a list of figures-of-merit to report, as a framework for testing a new electrolyzer setup. Table 3 briefly summarizes these items, and each of them is further elaborated in what follows.
T a g g e d P Reporting the exact details of the cell is of prime importance. This should cover the dimensions of the cell components, their physical and chemical properties (e.g., membrane and GDL type and thickness, chemical composition, pretreatment method), and the degree of compression. We reiterate our earlier point, that without this information, the flow rate of the electrolyte (or the gas)-which is usually provided in units such as sccm or ml/min-gives only very limited information on the retention time of CO 2 in the cell. The flow rate of the electrolyte therefore should be always normalized by either the free volume of the half-cell or with respect to the electrochemically-active surface area of the electrode (or both).

Pre-operational characterization
T a g g e d P The electrocatalyst is the core component of any electrochemical setup, therefore its comprehensive physical and chemical characterization must be the first step of every experimental work. These Table 3 Proposed benchmarking protocol for continuous-flow CO 2 electrolyzers.

Characterization method Requirements Data to report
Physical and chemical characterization methods (XRD, XPS, TEM, SEM, EDS, Contact angle, and surface area measurements, etc.) Multiple scale analysis has to be performed both for the electrocatalyst and the support.
Chemical composition, size distribution, shape, dominant crystal orientation, surface properties Cyclic voltammetry and linear sweep voltammetry of the individual electrodes.
Measurement of Tafel-plots No Faraday process shall occur during CV, reference electrode is needed. Both separately for the two electrodes, and before operation inside the cell. IR-compensation shall be carried out Electrochemically active surface area, onset potential value at a given current density.

Electrochemical impedance spectroscopy
Multiple voltage values shall be used. R s , R ct , C

Linear sweep voltammetry
Multiple scan rates. Several subsequent scans, till the stationary behaviour is reached. Comparison of measurements with and without CO 2 shall be carried out, with solutions of the same pH.
LSV curves, current density normalized to both geometric and electrochemically active surface area. Onset potential value at a given current density.

Stationary electrolysis (chronoamperometry/chronopotentiometry) coupled with product analysis
Online analyis of both gas and liquid phase products. Measure on multiple current/voltage values (at least 5).
Faradaic efficiencies for ALL products, including H 2 (for all current/voltage values).

Long term stationary electrolysis (chronoamperometry/chronopotentiometry) coupled with product analysis
Online analyis of both gas and liquid phase products after selected time periods.
Faradaic efficiencies for ALL products, including H 2 (for all current/voltage values) with respect to the elapsed time.
Electrochemical impedance spectroscopy Multiple voltage values shall be used. R s , R ct , C Cyclic voltammetry and linear sweep voltammetry of the individual electrodes.
Measurement of Tafel-plots The same measurements as before operation both inside and outside the cell Electrochemically active surface area Physical and chemical characterization methods (XRD, XPS, TEM, SEM, EDS, Contact angle, and surface area measurements, etc.) The same measurements as before operation Chemical composition, size distribution, shape, dominant crystal orientation, surface properties

PRE operation
Cell configuration and size, properties and dimensions of the individual components (e.g., membrane and GDL thickness), catalyst loading pH, temperature, pressure, electrolyte concentration, surface area/free volume normalized flow rate Data to report Fig. 17. Alteration of the product selectivity with time in a pressurized PEM electrolyzer with a silver cathode. Reproduced with permission from ref. [103].
T a g g e d P studies must involve physical, chemical, and surface characterization via different spectroscopic techniques (FT-IR, Raman Spectroscopy), electron microscopy (TEM, SEM), elemental composition analyses (EDX, XRF), surface composition and functional groups analysis (XPS) and crystal phase determination (XRD). Similar characterization of the supporting layer (GDL) must be also performed, supplemented with the characterization of its pore structure (e.g., N 2 adsorption/desorption). T a g g e d P Although the catalyst morphology is one of the most important factors after the chemical composition, often little information could be found in the different articles. We recommend multi-level analysis to be performed before and after immobilizing the catalyst on the substrate, revealing crystallite size, crystallinity of the material, degree of aggregation, and surface properties of the particles. The geometric surface area of the immobilized catalyst can be estimated from the recorded microscopic images, but it must be kept in mind that the electrochemically-active surface area can immensely differ from this value. Therefore this latter parameter has to be calculated from EIS measurements, or by simply recording cyclic voltammograms in a potential window where no Faradaic process interferes [119]. For certain electrocatalysts, the use of model redox active compounds can also be useful.
T a g g e d P The activity of an electrocatalyst is most usually characterized by recording linear sweep or cyclic voltammograms. As these are nonstationary measurements (even at low polarization rates), the values of derived kinetic parameters from these, must be treated with caution. Recording of polarization curves provides complementary data. In this case, the stationary potential is recorded at different current densities (spanning several orders of magnitude), and then plotted versus the logarithm of the current densities to afford Tafel plots (Fig. 18). The linear region of the curve is fitted to give the Tafelslope, which À together with the onset potential of the process À is the most important parameter describing the electrocatalyst.
T a g g e d P By performing LSV measurements for the single electrodes in a classical three-electrode electrochemical cell and in the assembled electrolyzer under otherwise identical circumstances (electrolyte concentration, pH, temperature etc.), the effect of the cell construction method (hot pressing, mechanical fixation, etc.) on the catalyst properties can be studied directly. EIS measurements are very useful at this early stage of investigating the cell, as they can reveal any additional resistances (R s , R ct ), arising from failure of the cell assembly and leading to severe losses due to IR-drops in the system. Note, that during electrolysis, as a current (I) is driven between the working and the counter electrode, a potential drop (IR) develops in the setup; originating from the resistance of the cell constituents, the electrolyte, and the membrane. In a three-electrode electrochemical setup, the potential of the working electrode is measured against a stable reference electrode. Although these electrodes are placed very close to each other, some portion of this potential drop is still present between them, which is caused by the so called T a g g e d P uncompensated resistance (R u ).The IR u -drop is dictated by the cell geometry and is dependent on the applied current and its presence leads to distortion of the electrochemical measurements [78]. To secure reliable data on the catalyst activity, the measured potentials must be always corrected for the value of this uncompensated potential drop. For this purpose, the uncompensated resistance has to be measured, which is typically done by either EIS or by the current-interruption technique [121,122]. Overall, the discrepancy between actual and applied voltage requires the use of IR compensation to get reproducible data that is comparable between different laboratories for the various cells, especially those operating at high current densities.

In operando characterization
T a g g e d P The most important part of these studies is the investigation of the assembled flow cell during operation. To study the kinetics of CO 2 reduction in the given setup, LSV measurements should be carried out at multiple scan rates. Note that measurements at the first sweep rate must be repeated until steady-state electrochemical behavior is reached. Information gathered from the first cycles can often be misleading, because the measured current can originate from multiple sources, including the reduction of the electrode support or the electrocatalyst itself-as it was observed for the case of Cu 2 O derived Cu catalysts [20]. The LSV curves must be always normalized to the geometric surface area and/or to the electrochemically active surface area of the cathode. Comparison of such data to results measured with chemically-equivalent, but different sized catalyst particles can help to understand different size effects beyond the trivial increased surface area/volume ratio. Here we refer to the discussion on the importance of having one or more reference electrodes in the cell (see Fig. 3).
T a g g e d P The estimation of the onset potential of a process is always a difficult question. Some researchers derive this value graphically from the LSV curves, while others give the potential value at which the current reaches a certain threshold value. Consequently, the comparison of these data is difficult and often misleading. To avoid confusion, we suggest 1 and 10 mA cm -2 as a benchmarking current density (note the similarities with solar powered water oxidation and H 2 evolution) [118]. The LSV measurements serve as a decent guide for choosing the proper potential/current region, where CO 2 reduction studies have practical relevance. Subsequently performed stationary measurements, either at constant current or constant potential/voltage give real insights into the performance of a given setup. These time-dependent experiments must be coupled with in situ product analysis, both for the liquid and the gas phases, since these are better tools for determining the onset voltage (potential) than just using LSV curves.
T a g g e d P Designing a proper product-detection setup, however is not straightforward, because there is no single analytical method which can characterize both quantitatively and qualitatively all reduction products generated from CO 2 . First, for the analysis of the gas and liquid phase products, we generally need different analysis tools. An online sampling system should be connected to a gas chromatograph (for gas phase analyses) and sealed to avoid any leakage. We highlight a recent development, the barrier ionization discharge (BID) detector, which is universal with greater sensitivity than both the thermal conductivity-(TCD), and flame ionization detectors (FID). Consequently, there is no need for two different detectors to measure H 2 gas and the CO 2 derived products [123].
T a g g e d P Analysis of the liquid phase is a little more complicated, because the electrolyte ions must be either removed for the most sensitive GCÀMS analysis, or higher sample concentrations are needed for NMR analysis. The performance of the NMR analysis, however, can be improved by special cell design with high surface area working electrode and very low electrolyte volume [124]. Differential T a g g e d P electrochemical mass spectrometry (DEMS) is an effective tool for the real-time detection of electrochemical reduction products, however, the interpretation of the results can be complicated for a reaction which gives a mixture of different products [125]. For example, DEMS cannot be used for detection of CO, because CO 2 gives the same ionization fragments. The quantitative detection of formic acid, because of its deprotonation, is challenging by GC and DEMS as well. The amount of formate can be reliably quantified by ion chromatography. By this technique, however, other solution compounds (e.g. different alcohols) cannot be analyzed [126]. It is therefore of utmost importance to follow the reaction with different instrumental techniques, and to account for all the charge passed during the electrolysis.
T a g g e d P It is recommended to clearly state in every manuscript: (i) detection limit for the employed analytical instrument and (ii) the partial current density for the observed product at the onset voltage. We emphasize that because of the frequent use of carbon-based catalyst supports, studies with 13 C labeled CO 2 should be performed to clarify the origin of the carbon-containing products. Finally, we note that the selected analytical methods have to be compatible with the electrolyzer setup, possibly limiting the application of certain (otherwise promising) techniques.
T a g g e d P The Faradaic efficiency (FE) for all the formed compounds as well as the overall FE has to be reported. Note that FE as defined is strictly the ratio of the charge consumed on a given process and the passed total charge. Some authors-without analyzing the liquid or the gas phase products-mistakenly and misleadingly refer to the ratio of the given product amount and the detected total amount of all the products as the Faradaic efficiency. Similarly, H 2 is also a product (although not CO 2 -derived), therefore must be taken into account for these calculations.
T a g g e d P The operation of the electrolyzer must be also characterized on a longer timescale. Therefore, the above listed measurements-or at least those which were promising in the sense of current density and product distribution -has to be conducted for several days, while the product distribution is traced after regular time periods. Keep in mind that water electrolyzer cells, working on very similar principles and built from almost identical materials have a reported lifetime in the range of 10,000 h [127], and similarly long lifetimes are required for CO 2 electrolyzers in practical applications.

Post-operational characterization
T a g g e d P After the long-term operation, degradation of the electrochemical cell must be investigated at every level, by repeating the same measurements as before operation, in the reversed order. Thus, an EIS measurement of the cell can reveal corrosion of the cell constituents, while change in the onset potential (derived from LSV measurements) can indicate chemical change of the catalyst. After disassembling the cell, the catalysts must be characterized separately with the formerly used experimental techniques, revealing changes in the catalyst particle size, aggregation, composition, and in the structure of the membrane or the GDL, etc. [128].

Most important metrics to report
T a g g e d P Talking about the efficiency of the cell, one must distinguish among current-, electrical-, and energy efficiency. The current efficiency (or Faradaic efficiency, e Faradaic ) of the process is a direct relation between the number of electrons consumed in a given electrode process and the total amount of the charge passed during the electrolysis. Energy efficiency of a given cell is somewhat more complex concept, as it is not only influenced by the thermodynamics and kinetics of the electrode processes, but electrical parameters of the cell as well. The measurement conditions and heat exchange efficacy during the process may also be important parameters. These T a g g e d P can be defined in different ways, and therefore are a common source of confusion [129].
T a g g e d P The most commonly applied, and most straightforward definition is the voltage efficiency of the cell, which gives a good approximation on its operation. This is calculated as the ratio of the thermodynamic voltage needed to drive the given electrochemical process (given by the redox potential of the half-reactions) and the total voltage applied to the electrolysis cell (V cell ). Beyond the thermodynamic requirements of the reaction, this latter value is dictated by different energy losses, related to the catalysts (anodic and cathodic overpotential, h i ), to mass-transport, and to the cell itself (IR-drop). Fig. 19 summarizes the most common mechanisms, which lead to an increased cell voltage. Note that the shape of the curves may also be useful for diagnostic purposes when an electrolyzer fails. In addition, these factors can be separated by the above described benchmarking protocol, thus uncovering future development avenues [130].
T a g g e d P The above-mentioned efficiency units alone, however, are of low industrial relevance. To give an exact measure on its operation, the energy efficiency of the cell must be calculated during longterm operation. The energy efficiency is most commonly calculated as the product of the voltage-and Faradaic-efficiencies [15]. This is only true, however, if only one single product is formed. If this is not the case (i.e., multiple products are formed in parallel), the individual redox potentials (e.g., E°C O2/CH4 ) have to be taken into consideration Eq. (2).
T a g g e d P The energy efficiency can be also defined as the energy content (E content ) of the given product over the electrical energy consumed during electrolysis, where t is the time of electrolysis to form unity amount of the product. The chemical energy of the produced molecules can be expressed by their Gibbs free energy per mole (DG 0 ). In some articles, the heat of combustion, or enthalpy, has been used to represent the energy content: 6. Photoelectrochemical reduction of CO 2 in continuous-flow  Fig. 19. Different loss-mechanisms potentially affecting the performance of an electrolyzer.
T a g g e d P photoelectrochemical (PEC) cells, demonstrated in CO 2 conversion. Historically, PEC solar fuel generation was predominantly focused on water splitting (H 2 evolution), and much less attention was paid to CO 2 reduction, mainly because of kinetic constraints [131]. In direct PEC CO 2 reduction, upon excitation of an electron from the valence band to the conduction band (by UV or visible light irradiation) of a p-type semiconductor (SC) the separated photoelectrons are driven to the surface of the SC and react with CO 2 (or a redox mediator/co-catalyst present either on the surface or in the solution). The CO 2 -reducing ability of the photo-electrocatalyst is dictated by the position of the conduction band (CB) edge: it has to be at a more negative potential than the targeted CO 2 reduction reaction. In such a procedure, it is possible to reduce CO 2 at less negative potential relative to the thermodynamic potential (vs. electrochemical reduction). The indirect PEC approach is also possible, where an n-type semiconductor photoanode is employed, and CO 2 reduction takes place at the cathode in the dark [132]. By employing either PEC approach, the necessary energy input (cell voltage) can be decreased by the photopotential, which is dictated by the CB edge of the SC photocathode and the redox potential of the CO 2 reduction process (or the VB edge of the photoanode and the water oxidation potential, for the indirect approach).
T a g g e d P To enhance the current densities to a level which makes practical significance, however, photoelectrodes with large specific surface area are needed. This necessitates the use of nanostructured photoelectrodes, which is a surprisingly unexplored area, in fact, the longstanding theory of PEC builds on thick electrode films (with film thickness of over several micrometers) [133]. To efficiently drive CO 2 reduction reactions, these carriers need to reach the photoelectrodes interfaces at the electrolyte and at the back contact. Therefore it will be very important to understand the effect of nanostructuring on carrier generation and collection, surface recombination, and on the size of space charge layers [134].
T a g g e d P There are only a few examples in the literature, which at least demonstrate the concept of continuous-flow PEC CO 2 conversion. In a recent proof-of-concept study, hybrid CuO/Cu 2 O nanorod arrays were incorporated as the photocathode in a new continuous-flow design [68]. The performance of the photocathode was compared to a conventional two-compartment batch type PEC cell. The primary products were alcohols, with a yield which was »6 times higher than the batch design, and significantly enough, also showed longerchain alcohol products up to C2-C3 (ethanol and isopropanol). The high surface area-to-volume ratio resulting from the narrow T a g g e d P reaction channels resulted in an enhance photocurrent density and Faradaic efficiency.
T a g g e d P Even more recently, NASA researchers reported a PEC conversion device, operated at room temperature and ambient pressure with only ultraviolet radiation [135]. They used a nanocomposite electrode which combines a photocatalyst and an electrocatalyst, capable of reducing gaseous CO 2 to methane without the need of external electricity input. Considering the typical solar flux, the achievable current densities are typically lower for these cells (»10 mA cm ¡2 ) compared to the electrolyzers presented earlier. In addition, long term stability is an addition hurdle to face, therefore stability tests are even more important in these instances. In other instances, the indirect approach was employed, where the oxidation half reaction took place on a photoexcited n-type semiconductor (specifically, TiO 2 ). As shown in Fig. 20, the overall cell design is very similar to the conventional PEM electrolyzers, except that the anode is irradiated [132,136]. The principal benefit of this setup is that all the knowledge gathered for the cathode reaction (CO 2 conversion) can be implemented, while the solar energy input is harnessed.
T a g g e d P In the case of PEC cells, the photon to product conversion efficiency is defined as the output power, namely, the product of voltage, partial current densities, and Faradaic efficiency for the formation of different products, divided by the denominator, namely, the solar energy input from sunlight: where j photo is the photocurrent density (mA cm ¡2 ) normalized to the illuminated electrode area, V redox is the thermodynamic potential of the given process, V bias is the applied voltage, e Faradaic,i are the partial Faradaic efficiencies for the different products, and P light is the power of the incident illumination [137,138]. T a g g e d P Alternatively, Eq. (5) uses the chemical energy of the products formed divided by the solar energy input from sunlight incident on the electrode. In this expression, the nominator is directly related to the formation rate of a given product, establishing a direct relation to product analysis during the measurements. Note that this definition is similar to that conventionally used for solar to hydrogen (STH) conversion [139], and is only applicable if no external bias is employed (i.e., at short circuit conditions). T a g g e d P where v i is the formation rate of a given product (mmol s ¡1 ), DG i is the Gibbs free energy per mole of a given product (J mol ¡1 ), P total is the incident illumination power density (mW cm ¡2 ), and A is the illuminated electrode area (cm 2 ).

Summary and outlook
T a g g e d P As demonstrated in this review article, multiple parameters have to be optimized simultaneously to efficiently perform continuousflow electroreduction of CO 2 . Some of them are well understood, while others still need to be carefully studied. The effects of high pressure and temperature are of particular interest to achieve reasonable current density and selectivity. These parameters will also affect the surface of the catalysts [140], which is another factor to be studied in continuous-flow cells. Computational modeling can contribute to the rational design of electrolyzer configuration. In this vein, the reactor performance can be numerically simulated to unravel the influence of flow rate and channel geometry on CO 2 conversion and consumption rate. Similarly, recent advances in 3D printing allows rapid prototyping of different cell geometries and thus will be a powerful tool in the hand of electrochemists [141]. Furthermore, we believe that successful studies in vapor phase will open up the opportunity to use industrial exhaust fume (rich in both CO 2 and H 2 O) directly as feedstock for solar fuel generation. Accordingly, different model gases containing typical impurities should be studied in the future.
T a g g e d P As for future development avenues, we would like to emphasize two directions. One is coming from the materials perspective: the need for intricate architectures where the elements of the GDE are simultaneously optimized. As shown in Sections 3.3À3.5, rationally designed interfaces are required for efficient CO 2 conversion. In this endeavor, the cooperation of chemists, materials scientists, and engineers is highly recommended. The second R&D path is rooted in the fact that the anode reaction was oxygen evolution (water oxidation, OER) in almost all the presented studies. In such cases the formed oxygen is considered as a non-harmful by-product, and is simply let to the atmosphere without using it for any purpose. We also note that OER as the anode process can be important in Space applications, namely as a root for the recovery of O 2 from CO 2 . With the interest of deep space exploration (i.e., possible prolonged missions to Mars), it is of high importance to improve such key enabling technologies. As for terrestrial applications, the formed oxygen can be compressed and sold, but driving a more beneficial electrochemical procedure on the anode could be a value-added approach. In this manner, CO 2 electrolyzers could be easily integrated in other industrial processes, in which the main product is formed on the anode.
T a g g e d P There are several candidates, for example, using the oxidation of organic pollutants on the anode, which is a kinetically-facile reaction. Thus the electrolyzer can be employed as both CO 2 converter and water purifier (i.e., wastewater treatment with a net zero CO 2 emission) adding value to the overall process [142]. This can be envisioned by either directly oxidizing the organic pollutants, or indirectly, by generating ozone on the anode. This concept is wellknown for water electrolyzers, in which hydrogen is produced on the cathode, while oxidation of water pollutant occurs on the anode [143]. Chlorine (Cl 2 ) evolution is another technologically relevant reaction, which might be worth coupling with CO 2 reduction (also called chlor-syngas process, see Fig. 21) [144]. Importantly, the redox potential of chloride oxidation matches with that for the water oxidation; therefore, this approach does not lead to an increased cell voltage [145]. In this case, however, important precursors of some commodity chemicals are formed on both the electrodes. As these products are all in the gas phase, it is easy to separate them from the aqueous electrolyte during a subsequent technological step. This concept is very similar to the so called oxygen T a g g e d P depolarized cathode chlor-alkali cells, where chlorine is formed on the anode, while oxygen gas is reduced on the cathode [146]. Plants operating on this concept have been in operation for years; and therefore the infrastructure and technological know-how are readily available. Finally, while H 2 oxidation at the anode is not a valueadded approach, it allows for gas feed on both sides, which can be beneficial in certain instances [44,147].
T a g g e d P We are also convinced that concentrated efforts need to be devoted to scale-up and scale-out, to achieve reactor sizes which are at least similar to industrially used water electrolyzers (2 MW). It is worth emphasizing that conclusions drawn for electrochemical cells offering very low current densities are not necessarily valid for those with high currents. Consequently, analyzing electrodes/cells under conditions which are far removed from those which are necessary for practical applications, is a futile exercise. Finally, we hope that the proposed benchmarking protocol will provide insightful guidelines to researchers involved in this endeavor and will lead to more comparable results.