Recent Advances in the Catalyst Design and Mass Transport Control for the Electrochemical Reduction of Carbon Dioxide to Formate

: Closing the carbon cycle by the electrochemical reduction of CO 2 to formic acid and other high-value chemicals is a promising strategy to mitigate rapid climate change. The main barriers to commercializing a CO 2 reduction reaction (CO 2 RR) system for formate production are the chemical inertness, low aqueous solubility, and slow mass transport characteristics of CO 2 , along with the low selectivity and high overpotential observed in formate production via CO 2 reduction. To address those problems, we ﬁrst explain the possible reaction mechanisms of CO 2 RRs to formate, and then we present and discuss several strategies to overcome the barriers to commercialization. The electronic structure of the catalyst can be tuned to favor a speciﬁc intermediate by adjusting the catalyst composition and tailoring the facets, edges, and corners of the catalyst to better expose the active sites, which has primarily led to increased catalytic activity and selectivity. Controlling the local pH, employing a high-pressure reactor, and using systems with three-phase boundaries can tune the mass transport properties of reactants at the catalyst surface. The reported electrocatalytic performances are summarized afterward to provide insight into which strategies have critical e ﬀ ects on the production of formate.


Introduction
The carbon dioxide concentration in the atmosphere has been rapidly increasing, which has become a major cause of global warming via the greenhouse effect [1]. The global CO 2 atmospheric concentration was approximately 409 ppm in 2014; the concentration had risen for about 120 ppm over the past 250 years and is likely to reach 1000 ppm by the end of this century [2]. To tackle this climate change issue, the Paris Agreement reached a commitment to keep the global average temperature increase to less than 2 • C above the preindustrial levels [3]. The CO 2 reduction reactions (CO 2 RRs) have been proposed as a promising strategy to reduce the greenhouse gas emissions while storing the energy in a value-added chemical [4]. Electrochemical reduction of CO 2 can produce various types of product depending on the material of the electrode, solvent types, local pH on the catalysts surface, and CO 2 pressure [5,6]. Formic acid or formate is a promising CO 2 RR product because of its wide applications, including leather tanning and animal feed markets [7]. Formic acid is also a good hydrogen storage material because of its high hydrogen density of 52 g of H 2 per liter of formic acid [8]. Furthermore, formic acid can be easily decomposed catalytically to CO 2 and H 2 at room temperature [9], and it can be utilized directly as the fuel for a direct formic acid fuel cell without pretreatment [10]. The electrochemical reaction equations and equilibrium potentials (V vs. reversible hydrogen electrode (RHE)) for the reduction of CO 2 to formic acid/formate and the major competing reactions are presented as reactions (1)-(4) [11][12][13]: 2H + + 2e − → H 2 E 0 = 0.00 V The market size of formic acid was about 950 kilotons per year in 2014 and will grow to 1 megaton per year in 2030 [14]. Approximately, 90% of the existing production capacity generates formic acid by methyl formate hydrolysis [15,16]. According to a technoeconomic analysis by Verma et al. using the gross-margin model, for a Sn cathode catalyst with an assumption of 100% faradaic efficiency (FE) for formate production, a gross margin of 30% requires a minimum formate partial current density of 56 mA cm −2 and a catalyst durability of 4000 h at a cell potential of 4 V [17]. These technoeconomic reports suggest that practical implementation is closely tied to improvements in crucial electrocatalytic reaction parameters, such as FE, current density, and overpotential. CO 2 is a stable molecule with an electrophilic carbon center and a linear molecular geometry. Therefore, the first electron transfer to form the CO 2 − intermediate requires a very negative potential of −1.49 V vs. RHE [18]; this thermodynamic barrier makes CO 2 RRs difficult. Hydrogen evolution is a major competing reaction, which lowers the FE HCOOH . Notably, transition metals with relatively strong binding affinities for the hydrogen intermediate tend to show a marked tendency towards the hydrogen evolution reaction (HER) because its limiting potential is more positive than that of the CO 2 RR to HCOOH [19]. Further, mass transport issues limit the partial current density of formate production [20]. The low aqueous solubility and slow mass transfer of CO 2 result in a low CO 2 concentration at the catalyst interface during electrolysis in conventional H-type cells [21], which restricts the partial current density of formate production to low levels. A correlation has been reported between the binding energy of *HCOO, which is the main intermediate, and the activity of the catalyst in formate production [22]. Specifically, consistent with the Sabatier principle, a higher formate partial current density is achieved with a Sn catalyst with a moderate *HCOO binding affinity than with transition metals with much stronger or much weaker *HCOO binding affinities, including Ni, Cu, Pt, Ag, and Au [23]. The nature of the intermediate binding energy can be tailored by synthesizing a bimetallic or doped catalyst to change the electronic structure from that of the pristine catalyst [24][25][26]. Tuning the intermediate binding energy grants control over the selectivity and the overall reaction pathway as a result [27]. Exposing high-energy facets and increasing the number of edge, corner, and grain boundary (GB) sites have also been reported to improve catalyst activity [28][29][30]. These nanostructuring approaches ease the adsorption of the reaction intermediates and intensify the local electric field, which ultimately increase the formate production [28,31]. Meanwhile, the mass transport properties of CO 2 have been improved by raising the pressure of the reaction vessel to increase the CO 2 solubility and by employing three-phase-boundary reactors in which gaseous CO 2 is directly transported to the catalyst surface through a gas diffusion layer (GDL) [32,33].
In this review, we first explain the reaction mechanisms that have been reported for electrochemical formate production. We then proceed from this deeper understanding to ideas for optimizing important electrocatalytic performance parameters, such as FE, current density, and overpotential. The reaction mechanisms and electrocatalytic performances primarily depend on the types of the catalyst materials and the intrinsic natures of their intermediate binding energies. Accordingly, we review the strategies for modifying the binding energies and thereby the catalytic properties of the materials, which can be classified as tailoring the facets, edges, and corners of the catalyst to expose the active sites or adjusting the catalyst composition to tune its electronic properties. Next, the strategies for improving CO 2 mass transport are outlined, which include local pH control, high-pressure reactors, and three-phase-boundary electrodes. Finally, the performances of the reported materials are compared at the end of this review to provide clear insight into which strategies have critical effects on formate production performance.

Reaction Pathways
The ability of a catalyst to adsorb and desorb specific reaction intermediates determines the main reaction pathway, along with the overpotential that is needed to overcome the energy barrier to that pathway. Therefore, a deep understanding of the CO 2 RR mechanism will give researchers the necessary knowledge to design more efficient catalysts [34,35]. In this section, we will discuss four reported reaction pathways for HCOOH/HCOO − production: (1) CO 2 insertion, (2) an O-bound intermediate, (3) a C-bound intermediate, and (4) a bicarbonate intermediate.
The first probable mechanism for HCOOH production is CO 2 insertion into the metal-H bond and formation of the *HCOO intermediate (Figure 1a) [36]. HCOOH is subsequently obtained via *HCOO reduction. This mechanism occurs on Pd-based catalysts. The strong hydrogen binding affinity of Pd results in hydrogen absorption into the Pd lattice and the formation of β-phase PdH x , if the Pd catalyst is held at potentials of <0 V vs. RHE [37]. The surface hydride will reduce the CO 2 and form the *HCOO intermediate. Formate has been produced by this reaction mechanism at a very low potential of 0 V vs. RHE in HCO 3 − solutions [38,39].
The second reaction pathway, which involves an O-bound intermediate, starts with the first electron transfer to CO 2 and the formation of a weakly adsorbed CO 2 − radical intermediate (Figure 1b).
Formic acid is produced from the transfer of protons and electrons from the reaction between the CO 2 − radical and proton donors like water, bicarbonate, and hydronium ions [36,40]. Li [19,42]. This group produces formate with high selectivity overall, although Sn and In produce a small amount of CO.
In the C-bound-intermediate pathway, the CO 2 − radical is formed and then bounded to the catalyst surface via the C atom ( Figure 1c). The CO 2 − radical next reacts with H + to form an adsorbed *COOH.
HCOO − /HCOOH can then be produced from *COOH. According to Sullivan et al., the *COOH form of the intermediate for HCOOH production is unstable and tends to either decompose into M-H + CO 2 or lose OH − through a nucleophilic attack to form the M-C = O + intermediate [43]. However, appropriate metal catalysts can stabilize the *COOH intermediate by an isomerization process shown in (5)-(7) that forms a more stable *HCOO intermediate and ultimately releases HCOO − . The number of proposed alternative reactions indicates that reactions will occur in parallel in the majority of systems [40,44].
Cu-and Ru-based catalysts produce formate via the C-bound-intermediate pathway because of their basic natures and because the bond formed by electron transfer from CO 2 to unoccupied metal orbitals is stabilized by back-donation from the d orbitals of the metal atom. Those metals are also able to produce other products, such as alcohols or hydrocarbons, depending on the reaction conditions [40,45]. A theoretical report by Yoo et al. stated that Cu(211) can produce formic acid at low overpotentials via the *COOH pathway due to its higher (more positive) limiting potential compared to the *HCOO pathway. Cu-based electrocatalysts have been reported to produce formate as the main product when applied to the anionic membrane of a membrane electrode assembly (MEA) electrolyzer [46]. The same report stated that applying high pressure resulted in formate becoming the main product of the reaction on the Cu catalyst.
The bicarbonate-intermediate pathway starts with a reaction between adsorbed *OH and CO 2 to form the adsorbed bicarbonate (CO 3 H*) species (Figure 1d) [47,48]. The CO 3 H* intermediate reacts with H + and receives an electron to form *HCOO and *OH. HCOOH is released after HCOO − has reacted with another H + and received another electron. This reaction pathway has been reported to occur at the Bi-Sn interface with an onset potential of −0.7 V vs. RHE in 0.5 M KHCO 3 . The FE HCOOH reaches 96% at −1.1 V vs. RHE [47]. PdSnO 2 and SnO x surfaces were also reported to produce formate via the bicarbonate-intermediate pathway [49,50]. This reaction mechanism is likely to occur on catalysts with relatively high *OH binding strengths.
The intermediate binding affinity of a catalyst is associated with its product selectivity, the reaction overpotentials, and the mechanism, which implies that catalytic performance can be enhanced by controlling the intermediate binding affinity. In the following sections, we will discuss several strategies in detail, including tailoring the facet, edge, and corner sites of the catalyst and controlling the catalyst composition.
Catalysts 2020, 10, x FOR PEER REVIEW 4 of 21 reaches 96% at −1.1 V vs. RHE [47]. PdSnO2 and SnOx surfaces were also reported to produce formate via the bicarbonate-intermediate pathway [49,50]. This reaction mechanism is likely to occur on catalysts with relatively high *OH binding strengths. The intermediate binding affinity of a catalyst is associated with its product selectivity, the reaction overpotentials, and the mechanism, which implies that catalytic performance can be enhanced by controlling the intermediate binding affinity. In the following sections, we will discuss several strategies in detail, including tailoring the facet, edge, and corner sites of the catalyst and controlling the catalyst composition.  [36]; (b) CO2 to HCOO − via O-bound intermediate [36]; (c) CO2 to HCOOH via C-bound intermediate [51,52]; and (d) HCOOH production from CO3H* [47]. Color codes: black, C; red, O; white, H; grey, metals.

Nanostructural Engineering
The geometrical features of a catalyst can greatly affect its CO2 electroreduction performance. The exposed high-energy facets, edges, and corners will significantly enhance the catalytic activity towards CO2 reduction [28,31,53,54]. The catalytic activity of a nanostructured catalyst is usually proportional to the number of under-coordinated sites [31]. Sites such as high-index facets have abundant under-coordinated atomic steps and exhibit enhanced catalytic activity compared to lowindex nanoparticles [29,53]. Moreover, a large number of edge and corner sites enhances catalytic activity by facilitating a strong electric field at the catalyst surface to ease the adsorption of CO2 [28,31]. Klinkova et al. reported the relationship between surface atom coordination and the catalytic performance of Pd nanoparticles (NPs) in CO2 electroreduction to formate [53]. The authors first conducted density functional theory (DFT) calculations on Pd(111), Pd(110), Pd(100), and Pd(211) surfaces and a Pd19 cluster. As shown in Figure 2a, the authors found that the ∆Gformation of CO* increased from −0.83 eV on Pd(111) to −0.65 eV on Pd(211), while the ∆Gformation of *HCOO decreased from 0.28 eV on Pd(111) to −0.09 eV on Pd(211). Therefore, the formation of *HCOO is made more favorable by incorporating higher-index facets. After obtaining these predictions from the DFT calculations, the authors then synthesized {100} plane-enclosed nanocubes (NCs), {110} planeenclosed rhombic dodecahedra (RDs), NPs with mixed low-index facets, and branched NPs (BNPs) enclosed by high-index facets. The synthesized NPs are shown in Figure 2c-g. Electrochemical characterization revealed that at the beginning of electrolysis, the total geometric current densities  [36]; (b) CO 2 to HCOO − via O-bound intermediate [36]; (c) CO 2 to HCOOH via C-bound intermediate [51,52]; and (d) HCOOH production from CO 3 H* [47]. Color codes: black, C; red, O; white, H; grey, metals.

Nanostructural Engineering
The geometrical features of a catalyst can greatly affect its CO 2 electroreduction performance. The exposed high-energy facets, edges, and corners will significantly enhance the catalytic activity towards CO 2 reduction [28,31,53,54]. The catalytic activity of a nanostructured catalyst is usually proportional to the number of under-coordinated sites [31]. Sites such as high-index facets have abundant under-coordinated atomic steps and exhibit enhanced catalytic activity compared to low-index nanoparticles [29,53]. Moreover, a large number of edge and corner sites enhances catalytic activity by facilitating a strong electric field at the catalyst surface to ease the adsorption of CO 2 [28,31]. Klinkova et al. reported the relationship between surface atom coordination and the catalytic performance of Pd nanoparticles (NPs) in CO 2 electroreduction to formate [53]. The authors first conducted density functional theory (DFT) calculations on Pd(111), Pd(110), Pd(100), and Pd(211) surfaces and a Pd 19 cluster. As shown in Figure 2a, the authors found that the ∆G formation of CO* increased from −0.83 eV on Pd(111) to −0.65 eV on Pd(211), while the ∆G formation of *HCOO decreased from 0.28 eV on Pd(111) to −0.09 eV on Pd(211). Therefore, the formation of *HCOO is made more favorable by incorporating higher-index facets. After obtaining these predictions from the DFT calculations, the authors then synthesized {100} plane-enclosed nanocubes (NCs), {110} plane-enclosed rhombic dodecahedra (RDs), NPs with mixed low-index facets, and branched NPs (BNPs) enclosed by high-index facets. The synthesized NPs are shown in Figure 2c-g. Electrochemical characterization revealed that at the beginning of electrolysis, the total geometric current densities increased from 15 to 22 mA cm −2 at a −0.2 V overpotential. All geometric current densities decreased during the reaction due to CO poisoning of the catalyst surface. The measured rate of decrease of the current density was in agreement with the calculated results in which BNPs, which have the highest-index facets, have the lowest rate of decrease of the current density ( Figure 2b). Kim et al. reported the correlation between the number of edge and corner sites and the electrocatalytic performance of Bi NPs [31]. Sharper Bi nanoflakes had greater numbers of edge and corner sites. The authors used the electrodeposition pulse current (PC) method to synthesize nanostructured Bi nanoflakes. Figure 3a,b shows the samples fabricated with between 1 and 9 pulse cycles at 20 mA cm −2 . The samples at 1 and 3 pulse cycles showed FEs of 89% and 90%, respectively, and were not perfectly converted to Bi nanoflakes. The sample at 6 pulse cycles was completely converted to Bi nanoflakes and achieved an FE of over 100%. However, the sample at 9 pulse cycles showed a decrease in FE to 97% because Bi layers were covering the edge and corner sites, transforming the structure into a tripod shape. The prospects of electrostatic field intensification at the edges and corners were explored using a COMSOL Multiphysics simulation. A hexagonal Bi nanostructure was used as a representative three-dimensional structure with an edge length of 500 nm at various thicknesses ( Figure 3c). Reducing the thickness from 500 to 10 nm enhanced the electrostatic electric field intensities at the edge and corner 3-and 5-fold, respectively. A twodimensional Bi nanostructure was simulated to investigate the effect of sharpening the corners ( Figure 3d); reducing the angle from 70 to 5° enhanced the electric field 2.5-fold. The electric field affects the overall reaction rate by increasing the local concentration of K + , which enhances CO2 adsorption by lowering the free energy for the adsorption of intermediates [28,31]. Kim et al. reported the correlation between the number of edge and corner sites and the electrocatalytic performance of Bi NPs [31]. Sharper Bi nanoflakes had greater numbers of edge and corner sites. The authors used the electrodeposition pulse current (PC) method to synthesize nanostructured Bi nanoflakes. Figure 3a,b shows the samples fabricated with between 1 and 9 pulse cycles at 20 mA cm −2 . The samples at 1 and 3 pulse cycles showed FEs of 89% and 90%, respectively, and were not perfectly converted to Bi nanoflakes. The sample at 6 pulse cycles was completely converted to Bi nanoflakes and achieved an FE of over 100%. However, the sample at 9 pulse cycles showed a decrease in FE to 97% because Bi layers were covering the edge and corner sites, transforming the structure into a tripod shape. The prospects of electrostatic field intensification at the edges and corners were explored using a COMSOL Multiphysics simulation. A hexagonal Bi nanostructure was used as a representative three-dimensional structure with an edge length of 500 nm at various thicknesses ( Figure 3c). Reducing the thickness from 500 to 10 nm enhanced the electrostatic electric field intensities at the edge and corner 3-and 5-fold, respectively. A two-dimensional Bi nanostructure was simulated to investigate the effect of sharpening the corners ( Figure 3d); reducing the angle from 70 to 5 • enhanced the electric field 2.5-fold. The electric field affects the overall reaction rate by increasing the local concentration of K + , which enhances CO 2 adsorption by lowering the free energy for the adsorption of intermediates [28,31]. Catalysts 2020, 10, x FOR PEER REVIEW 6 of 21 Two-dimensional materials were reported to have higher selectivity and intrinsic activity than bulk materials because of the high exposure of the active facets. Additionally, these materials have high surface areas compared to bulk materials [55][56][57][58]. For example, Han et al. reported ultrathin Bi nanosheets (NSs; BiNS) with an exceptional FE of almost 100% and a larger partial current density than that of commercial Bi (Figure 4a,b) [56]. The ultrathin BiNS were prepared by in situ topotactic transformations of BiOI NSs. To support the study, DFT calculations were conducted via the computational hydrogen electrode methodology. These simulations were performed on the Bi(001) facet because it was the predominantly exposed facet of the BiNS (Figure 4c). The CO2 reduction to formate was initiated by protonation of the C atom to form the *HCOO intermediate in a mildly endothermic reaction (+0.49 eV). The second proton-coupled electron transfer to HCOO − was an exothermic reaction (−0.17 eV). Ultimately, HCOO − was spontaneously released from the catalyst. Protonation of the O atom to form COOH*, which was the intermediate for CO production, had a significantly higher energy barrier (+1.16 eV). The energy barrier for H adsorption into Bi(001) was also too high (+0.95 eV). The production of formate on Bi(001) was accordingly more favorable than the evolution of CO and H2. Two-dimensional materials were reported to have higher selectivity and intrinsic activity than bulk materials because of the high exposure of the active facets. Additionally, these materials have high surface areas compared to bulk materials [55][56][57][58]. For example, Han et al. reported ultrathin Bi nanosheets (NSs; BiNS) with an exceptional FE of almost 100% and a larger partial current density than that of commercial Bi (Figure 4a,b) [56]. The ultrathin BiNS were prepared by in situ topotactic transformations of BiOI NSs. To support the study, DFT calculations were conducted via the computational hydrogen electrode methodology. These simulations were performed on the Bi(001) facet because it was the predominantly exposed facet of the BiNS (Figure 4c). The CO 2 reduction to formate was initiated by protonation of the C atom to form the *HCOO intermediate in a mildly endothermic reaction (+0.49 eV). The second proton-coupled electron transfer to HCOO − was an exothermic reaction (−0.17 eV). Ultimately, HCOO − was spontaneously released from the catalyst. Protonation of the O atom to form COOH*, which was the intermediate for CO production, had a significantly higher energy barrier (+1.16 eV). The energy barrier for H adsorption into Bi(001) was also too high (+0.95 eV). The production of formate on Bi(001) was accordingly more favorable than the evolution of CO and H 2 .
Grain boundaries (GBs) enhance the activity of catalysts by providing highly active reaction sites [59]. Feng et al. pioneered the quantitative correlation analysis between the density of GBs and the catalytic activity of catalysts [60]. They found a linear correlation between the GB surface density and the specific activity for CO 2 RRs on vapor-deposited Au NPs on carbon nanotubes. Li et al. reported the catalytic performance of Bi NPs/Bi 2 O 3 NSs with abundant GBs. The catalyst, which was obtained by facile hydrothermal synthesis, was compared to low-GB-density Bi NSs/Bi 2 O 3 NSs. High-resolution transmission electron microscopy (HRTEM) images of the catalysts have shown the GB density difference between the two catalysts (Figure 5a,b). The Bi NPs/Bi 2 O 3 NSs exhibited a superior catalytic performance to that of Bi NSs/Bi 2 O 3 NSs (Figure 5c,d). The Bi NPs/Bi 2 O 3 NSs also showed a higher formate partial current density of 24.4 mA cm −2 , an FE HCOOH of >90% over a broad potential range, and over 24 h of catalyst durability. Grain boundaries (GBs) enhance the activity of catalysts by providing highly active reaction sites [59]. Feng et al. pioneered the quantitative correlation analysis between the density of GBs and the catalytic activity of catalysts [60]. They found a linear correlation between the GB surface density and the specific activity for CO2RRs on vapor-deposited Au NPs on carbon nanotubes. Li et al. reported the catalytic performance of Bi NPs/Bi2O3 NSs with abundant GBs. The catalyst, which was obtained by facile hydrothermal synthesis, was compared to low-GB-density Bi NSs/Bi2O3 NSs. Highresolution transmission electron microscopy (HRTEM) images of the catalysts have shown the GB density difference between the two catalysts (Figure 5a,b). The Bi NPs/Bi2O3 NSs exhibited a superior catalytic performance to that of Bi NSs/Bi2O3 NSs (Figure 5c,d). The Bi NPs/Bi2O3 NSs also showed a higher formate partial current density of 24.4 mA cm −2 , an FEHCOOH of >90% over a broad potential range, and over 24 h of catalyst durability.

Bimetallic Compounds
Controlling the composition of bimetallic compounds is an important strategy for tuning the electronic behavior of nanomaterials to enhance their selectivity and catalytic activity [24,62,63]. Bimetallic compounds can change the electronic structure of a catalyst, which affects the intermediate

Bimetallic Compounds
Controlling the composition of bimetallic compounds is an important strategy for tuning the electronic behavior of nanomaterials to enhance their selectivity and catalytic activity [24,62,63]. Bimetallic compounds can change the electronic structure of a catalyst, which affects the intermediate binding energy, which in turn controls the overall reaction pathway [27]. Many previously reported bimetallic compounds have shown enhanced catalytic activity compared to the corresponding pristine-metal-based catalysts, such as Ag-Sn [63], Cu-Bi [24,64], In-Sn [41], Sn-Cu [65,66], Pd-Ni [67], and Bi-Sn [47].
Wen et al. synthesized a Bi-Sn catalyst that had better selectivity for formate than a well-known Sn catalyst [47]. As-synthesized Bi-Sn on carbon fabric (CF) exhibited a dramatic current increase in CO 2 -purged electrolyte (Figure 6a). At −1.14 V vs. RHE, Bi-Sn/CF had a superior FE HCOOH of 94% ± 2% compared to Sn/CF, which had an FE HCOOH of 78% ± 2% (Figure 6b). To support the experimental data, the authors conducted periodic DFT calculations on the reaction pathways to CO and HCOO − from the adsorption of CO 3 H*. As shown in Figure 6c,d, the binding energy differences between the *COOH and *HCOO intermediates (∆E 1 ) on Sn(101) and Bi-Sn(101) surfaces were 0.55 and 0.81 eV, respectively. The binding energy differences between the CO* and HCOOH* intermediates (∆E 2 ) on Sn(101) and Bi-Sn(101) surfaces were 0.43 and 0.85 eV, respectively. The larger ∆E of the Bi-Sn catalyst indicates greater selectivity for formate, which agrees with the experimental results. In summary, tuning the composition of bimetallic compounds can help with the optimization of the intermediate binding energy of the catalysts to achieve higher catalytic selectivity and performance.

Doping Materials
Doping is another strategy to enhance the catalytic activities of electrocatalysts. The dopant affects the electronic properties of the host [66,68,69], which may allow the binding energy for a

Doping Materials
Doping is another strategy to enhance the catalytic activities of electrocatalysts. The dopant affects the electronic properties of the host [66,68,69], which may allow the binding energy for a particular intermediate to be designed to facilitate the desired reaction pathway [70].
Pd-based catalysts are known to have almost zero overpotential to activate CO 2 RRs to formate but also have a parallel CO pathway that competes with and may even deactivate formate production [37,71]. Bei Jiang et al. successfully increased the selectivity of a Pd-based catalyst for the production of formate with a high CO tolerance by doping the catalyst with boron [72]. The authors compared Pd-B/C to a control sample of Pd/C. The FE CO increased as the potential became more negative on Pd-based catalysts. However, the FE CO of Pd-B/C was much lower than that of Pd/C at any given potential (Figure 7a). The FE HCOOH of Pd-B/C reached a maximum value of 70% at −0.5 V vs. RHE, but Pd/C had an FE HCOOH of only 4.8% at the same potential. Furthermore, Pd/C had an FE HCOOH close to zero at potentials lower than −0.7 V vs. RHE, whereas the Pd-B/C catalyst had an FE HCOOH of 15-30% under those conditions. To gain a greater understanding of the promoting effect of B doping on formate production on Pd, the authors conducted DFT calculations on Pd(111) with and without B doping. Figure 7c,d present the Gibbs free energy diagrams for the intermediates of the CO and HCOOH pathways. The CO production pathway has a lower energy barrier than the HCOOH production pathway on the Pd(111) surface. In contrast, HCOOH production has the lower energy barrier on Pd(111)−4B, making it the more favorable pathway. The effects of B doping on the subsurface of the Pd lattice interstices include unique electronic behavior owing to partial electron transfer between Pd and B as well as a downshifted d-band center for the surface Pd. Thus, the new electronic structure leads to a more negative adsorption energy for *HCOO than for *COOH.

Mass Transport
The catalytic properties of CO2RRs depend on the nature of the catalyst and the local concentrations of the reactants at the catalyst interface. The CO2RR to formate has been reported to be the first order with respect to the CO2 concentration [37,51]. This means that the rate of formate production is limited by the CO2 concentration but could be greatly enhanced by the mass transport control. Therefore, the mass transport control of CO2 in CO2RRs is another crucial aspect of achieving good catalytic performance. Further, substantial hydrogen evolution can proceed from various

Mass Transport
The catalytic properties of CO 2 RRs depend on the nature of the catalyst and the local concentrations of the reactants at the catalyst interface. The CO 2 RR to formate has been reported to be the first order with respect to the CO 2 concentration [37,51]. This means that the rate of formate production is limited by the CO 2 concentration but could be greatly enhanced by the mass transport control. Therefore, the mass transport control of CO 2 in CO 2 RRs is another crucial aspect of achieving good catalytic performance. Further, substantial hydrogen evolution can proceed from various proton-donating species, such as bicarbonate and water, thereby lowering the CO 2 RR selectivity [21]. Controlling the local pH at the catalyst surface can affect the chemical equilibria of proton-donating species and, therefore, the HER rate. In this section, we discuss mass transport control strategies that are relevant for CO 2 RRs: (i) tuning the local pH at the catalyst-electrolyte interface; (ii) using high-pressure reactors; and (iii) employing three-phase-boundary electrodes and reactors.

Local pH Control
During electrolysis, the local pH near the electrode surface tends to be higher than that of the bulk region due to OH − production from both the CO 2 RR and the HER [74,75]. The increase in local pH will change the concentrations of proton-donating species, which is beneficial for inhibiting the HER. However, a high local pH can also decrease the CO 2 concentration at the electrode surface because of the equilibria between CO 2 , HCO 3 − , and CO 3 2− , which might lead to a low CO 2 RR partial current density ( Figure 9) [12,13,76,77]. Thus, the local pH value should be targeted to suppress hydrogen evolution while maintaining the CO 2 concentration near the electrode. Luo et al. reported the enhanced CO 2 RR performance of porous-structured-Zn (P-Zn) owing to an enlarged surface area and a strengthened local pH effect [78]. Upon comparing Zn foil to P-Zn, the enlarged surface area of P-Zn led to a 10-fold higher total current density, and the local pH effect reduced the HER rate. The buffering capacity of the electrolyte also has an important role in controlling the local pH. Ma et al. evaluated the CO 2 RR performance of a flat Ag electrocatalyst in 0.1 M K 2 HPO 4 , 0.1 M KHCO 3 , and 0.1 M KClO 4 to identify the correlation between the electrolyte buffering capacity (K 2 HPO 4 > KHCO 3 > KClO 4 ) and the local pH effect [79]. As the buffering capacity decreased, the FE for the CO 2 RR increased because a lower buffering capacity allows the high local pH to be maintained, thereby suppressing the HER.
ts 2020, 10, x FOR PEER REVIEW 1 sed because a lower buffering capacity allows the high local pH to be maintained, the ressing the HER.

igh-Pressure Reactor
ccording to Henry's law, increasing the partial pressure of the CO2 can increase ntration of the dissolved CO2. One of the pioneering efforts by Todoroki et al. in this regard ctrochemical reduction of CO2 at a pressure of 60 atm, which achieved an FEHCOOH of n and a large current density of 200 mA cm −2 [80]. Komatsu et al. reported that the FEHCOOH icantly as the partial pressure of CO2 increased on Cu catalysts [46]. These results indicated

High-Pressure Reactor
According to Henry's law, increasing the partial pressure of the CO 2 can increase the concentration of the dissolved CO 2 . One of the pioneering efforts by Todoroki et al. in this regard was an electrochemical reduction of CO 2 at a pressure of 60 atm, which achieved an FE HCOOH of nearly 100% and a large current density of 200 mA cm −2 [80]. Komatsu et al. reported that the FE HCOOH rose significantly as the partial pressure of CO 2 increased on Cu catalysts [46]. These results indicated that an elevated CO 2 partial pressure would improve the rate of HCOOH formation by increasing the solubility of CO 2 in aqueous solutions. However, a recent report by Ramdin et al. found that CO 2 pressures of over 40 bar while using a bipolar membrane (BPM) had several disadvantages, including formate crossover through the BPM that decreases the FE HCOOH by reoxidation of the as-synthesized formate on the anode. Another major disadvantage of this technique is the significant pH drop caused by the high concentration of CO 2 in the solution, which favors hydrogen evolution [81]. Thus, to obtain excellent performance in a high-pressure reactor, it is preferable to use an ion-exchange membrane with a low product crossover rate and to maintain the pH value.

Three-Phase Boundary
Despite the significant progress that has been made, there are still limitations for CO 2 mass transport that set the upper limit of current density at tens of milliamperes per square centimeter [82,83]. Therefore, utilizing a gas diffusion electrode (GDE), which was originally developed for fuel cell applications, for direct CO 2 delivery to the catalyst surface using water vapor as a carrier was proposed [84]. An article by Delafontaine et al. reported that in a conventional H-type cell, the concentration of CO 2 in aqueous solution was 0.038 M. Direct delivery of humidified CO 2 to the flow cell resulted in a modest increase in the relative saturated CO 2 concentration to 0.041 M. In the same article, the diffusion coefficient of aqueous CO 2 was reported to be as low as 0.0016 mm 2 s −1 in CO 2 -saturated 0.1 M KHCO 3 [85,86]. Delivering humidified gaseous CO 2 with a GDE dramatically increased the CO 2 diffusion coefficient 10,000-fold to 16 mm 2 s −1 (Figure 10a) [86]. The ease with which gaseous CO 2 reached the catalyst resulted in a high CO 2 availability and a subsequent increase in the CO 2 RR partial current density. As shown in Figure 10a, the GDE system is composed of a diffusion medium and a porous catalyst layer [87]. The diffusion medium is usually a hydrophobic carbon layer consisting of a macroporous carbon fiber layer and a microporous carbon powder layer. The diffusion medium serves many purposes, such as providing a porous medium through which CO 2 can diffuse to the catalyst layer, mechanically supporting the catalyst layer, and providing conductive pathways for the flow of electrons. Commercial diffusion media are usually treated with polytetrafluoroethylene (PTFE) to provide hydrophobicity. An ideal GDE remains hydrophobic throughout the process to prevent the electrolyte from leaking into the gas chamber. The catalyst is deposited on the microporous layer (MPL) to form the active catalytic site. The catalyst particles are usually mixed with a binder to hold the particles together and provide ionic conductivity. throughout the process to prevent the electrolyte from leaking into the gas chamber. The catalyst is deposited on the microporous layer (MPL) to form the active catalytic site. The catalyst particles are usually mixed with a binder to hold the particles together and provide ionic conductivity.  Del Castillo et al. reported the performance of an Sn/C-GDE using a membrane reactor [89]. Toray paper (TGP-H-90) was used as the carbonaceous support. The MPL was formed from Vulcan XC-72R and PTFE (40:60) that was air-brushed onto the carbonaceous support and sintered at 350 • C for 30 min. The catalyst ink consisted of Sn NPs and a Nafion solution (70:30) and was sprayed onto the MPL to achieve a Sn loading of 0.75 mg cm −2 . The Sn/C-GDE catalyst reached 70% FE at a current density of 150 mA cm −2 .
The MEA reactor is a common electrochemical flow cell for CO 2 RRs (Figure 10a) [85]. In a traditional batch-type cell, the ions must diffuse in the bulk electrolyte to reach the membrane. This kind of system has a relatively high diffusion resistance, and the reactant availability in the counter electrode is accordingly limited. In contrast, in an MEA reactor, the membrane is sandwiched between the cathode and anode to directly transfer ions between the electrodes (e.g., H + transfer between the anode and cathode). Consequently, the current density is markedly increased relative to traditional flow cells.
Another notable flow cell design is the microfluidic cell system proposed by the Kenis group (Figure 10b) [90,91]. This reactor has a thin channel of less than 1 mm through which the electrolyte flows between the anode and cathode [85]. CO 2 gas is supplied to the electrocatalyst through the GDL and reacts at the three-phase interface to form the product. A polymer electrolyte membrane (PEM) is not needed for a gaseous-product-type reactor. This reactor relies on the diffusion of the gaseous product to separate the oxidation and reduction products (Figure 10c, top), avoiding the use of high-cost membranes [85]. The microchannel enables laminar flow of the electrolyte, which eliminates the need for a membrane but still allows ionic transport between the electrodes. However, a PEM is still needed for liquid-product-type microfluidic cells to prevent product crossover (Figure 10c, bottom). Recently, Deng et al. reported the performance of a metal-organic framework (MOF)-derived carbon-nanorod-encapsulated bismuth oxide catalyst in both an H-type cell and a microfluidic cell system. Catalysis using an H-type cell resulted in an optimal FE HCOOH of 92% and a partial current density of 7.5 mA cm −2 at −0.9 V vs. RHE. As predicted, the microfluidic cell system exhibited a dramatic performance increase compared to the H-type cell, including a current density of 200 mA cm −2 at −1.1 V vs. RHE. Overall, these results indicate that the barrier of low CO 2 solubility has been tackled by using GDL-based three-phase-boundary reactors.

Reported Performances
We have summarized the reported performances in terms of the strategies for enhancing catalytic performance to provide insight into the effects of these strategies ( Figure 11) [10,47,53,61,72,92,93]. The results of the strategy of tailoring the exposed facets can be seen in the difference between Pd catalysts with low-index and high-index facets reported by Klinkova et al. The Pd BNPs with high-index facets have an almost identical FE HCOOH and a 1.5-fold larger formate partial current density compared to the low-index-facet Pd NCs. High-density-GB Bi/Bi 2 O 3 (HDGB) had an FE HCOOH of almost 100%, whereas the FE HCOOH of low-density-GB Bi/Bi 2 O 3 (LDGB) was 87.5%. The formate partial current density of HDGB was enhanced almost 1.6-fold compared to that of LDGB. The effect of using bimetallic materials can be seen in the performances of Sn/CF and Bi-Sn/CF. The bimetallic Bi-Sn/CF had an almost 20% higher FE HCOOH and a nearly 3-fold increase in formate partial current density compared to Sn/CF. The FE HCOOH enhancement can be attributed to an interaction between Bi and Sn that modified the electronic structure of the catalyst in a way that increased the intermediate selectivity, but the improved formate partial current density was because of the nanostructuring effect where Bi-Sn/CF had greater active site exposure. The effect of dopants can be seen in the difference between the Pd/C and Pd-B/C catalysts. Boron insertion resulted in an enlargement within the lattice and a partial electron transfer between B and Pd. These phenomena altered the catalyst electronic structure, which increased the formate selectivity almost 15-fold to 69.3% on Pd-B/C. A two-dimensional SnO 2 material showed a higher formate selectivity of 85% compared to bulk Sn foil and had an FE HCOOH of 63%. The current density of two-dimensional SnO 2 also increased to 14.75 mA cm −2 from the 3.11 mA cm −2 of bulk Sn foil at −1.01 V vs. RHE. Moreover, the application of SnO 2 NSs to GDEs successfully enhanced the formate partial current density from 14.75 mA cm −2 on SnO 2 NSs to an incredibly high 471 mA cm −2 on the SnO 2 GDE.
Catalysts 2020, 10, x FOR PEER REVIEW 15 of 21 performance. Furthermore, even though Bi-and Sn-based GDEs have resulted in higher current densities by orders of magnitude, new catalyst design strategies are required to reduce the overpotential to close to the equilibrium potential of formate production.

Conclusions
Researchers have made efforts to develop new catalysts and extrinsically enhance catalytic systems. However, the inert nature of CO2, the low solubility of CO2 in aqueous solutions, and the slow mass transport of CO2 result in the CO2RR system for formate production still having low selectivity, high overpotentials, and limited formate partial current density. Several strategies have been employed to address these challenges. Tailoring the binding affinity of a catalyst for a key intermediate was reported to enhance the selectivity and reduce the overpotential of the catalyst. This tailoring can be achieved in several ways, including exposing the high-energy facet, edge, and corner sites and controlling the composition of the catalyst. Attempts have been made to address the low CO2 solubility and slow mass transport of CO2 onto the catalyst surface by optimizing the local pH at the catalyst surface, using a high-pressure reactor to increase the CO2 solubility, and employing a GDE to directly transport gaseous CO2 to the active sites. However, the high-pressure environment makes the product cross over through the ion transport membrane towards the anodic side of the reactor, which depresses formate production. The loss of hydrophobicity in the GDE during electrolysis also decreases the CO2 concentration on the catalyst surface owing to the electrolyte flooding the gas diffusion medium. These issues must be addressed in future studies in addition to stabilizing the catalyst. Machine learning based on DFT calculations is expected to accelerate the identification of efficient catalysts satisfying the requirements of low overpotential with high selectivity and stability. We believe that the minimum requirements for technoeconomic feasibility Overall, optimizing the electronic structure of the electrocatalyst can enhance the formate selectivity, and the use of GDEs is the finest strategy for increasing the formate partial current density. However, achieving a low overpotential along with a high FE HCOOH and a fast formate production rate remains a challenge. The reported Pd-based catalysts exhibited low overpotentials, but the stability issue raised by CO poisoning must be addressed for further optimization of catalytic performance. Furthermore, even though Bi-and Sn-based GDEs have resulted in higher current densities by orders of magnitude, new catalyst design strategies are required to reduce the overpotential to close to the equilibrium potential of formate production.

Conclusions
Researchers have made efforts to develop new catalysts and extrinsically enhance catalytic systems. However, the inert nature of CO 2 , the low solubility of CO 2 in aqueous solutions, and the slow mass transport of CO 2 result in the CO 2 RR system for formate production still having low selectivity, high overpotentials, and limited formate partial current density. Several strategies have been employed to address these challenges. Tailoring the binding affinity of a catalyst for a key intermediate was reported to enhance the selectivity and reduce the overpotential of the catalyst. This tailoring can be achieved in several ways, including exposing the high-energy facet, edge, and corner sites and controlling the composition of the catalyst. Attempts have been made to address the low CO 2 solubility and slow mass transport of CO 2 onto the catalyst surface by optimizing the local pH at the catalyst surface, using a high-pressure reactor to increase the CO 2 solubility, and employing a GDE to directly transport gaseous CO 2 to the active sites. However, the high-pressure environment makes the product cross over through the ion transport membrane towards the anodic side of the reactor, which depresses formate production. The loss of hydrophobicity in the GDE during electrolysis also decreases the CO 2 concentration on the catalyst surface owing to the electrolyte flooding the gas diffusion medium. These issues must be addressed in future studies in addition to stabilizing the catalyst. Machine learning based on DFT calculations is expected to accelerate the identification of efficient catalysts satisfying the requirements of low overpotential with high selectivity and stability. We believe that the minimum requirements for technoeconomic feasibility of this pathway to formate production can be achieved via the combined approaches of catalyst development and broader system-level strategies.