Palladium alloys used as electrocatalysts for the oxygen reduction reaction

Palladium-based alloy materials as cathodes for the reduction of oxygen are regarded as potential substitutes for platinum-based catalysts in fuel cells. In this work, we present a scientometric analysis and critically review the use of Pd alloys for the oxygen reduction reaction (ORR). Through scientometric analysis, publication information, research fronts and hotspots are identified. For the critical review, reaction mechanisms in different media are discussed, with the aid of volcano plots to show the general principles for catalyst modifications to maximise the ORR. Influencing factors, including alloying, structure, strain and ligands, particle size, crystal facets and dealloying are considered with a view to informing the theoretical feasibility to enhance the ORR activity. In addition, Pd-based alloys synthesized by different methods are presented and compared in terms of ORR activities. Future research directions are discussed and possible approaches to mass production for industrialization are also proposed.


Introduction
The oxygen reduction reaction (ORR), one of the key chemical reactions in fuel cells and metal-air batteries, has been intensively investigated over the last several decades. [1][2][3] Research on ORR catalysts can be traced back to the 1970s, where most attention was focused on platinum (Pt) based catalysts. [4][5][6] Pt-based catalysts set the benchmark due to their excellent ORR performance and stability, and are widely used as cathode electrocatalysts in practical fuel cells. 7 However, there are several obstacles for commercialization: (1) the high price and the scarcity of Pt materials; (2) the methanol crossover effect and poor methanol tolerance, which leads to low cell performance and efficiency when using Pt-based catalysts in direct methanol fuel cells (DMFCs), 8 and (3) easy poisoning by impurities, such as CO and chloride ions in electrolytes. In recent years there have been attempts to reduce the amount of Pt catalyst but it has been found that under cathodic conditions Pt can easily migrate and aggregate on the carbon support. Additionally, if the particles are smaller in size then they can dissolve to form Pt cations, which then can deposit on other Pt particles, causing them to grow and deactivate, leading to the use of a large amount of Pt metal. Therefore, research effort has also been directed at the development of alternatives, such as other noble metals with relatively low-cost, non-noble metal materials and even metal-free catalysts. The emergence of these studies has greatly enriched the types of ORR electrocatalysts and paved several pathways for practical applications.
Amongst these, palladium (Pd) based materials have attracted extensive attention as alternative electrocatalysts because of their comparable properties and slightly lower cost compared to Pt. For instance, the average prices of Pd and Pt in the last ten years (from 1 August 2010 to 1 August 2020) are $910.62 and $1230. 96 per troy oz, respectively (www.platinum.matthey.com/prices). Specifically, both Pd and Pt are platinum group metals, and they are adjacent in the periodic table, resulting in similar chemical properties. For example, both are inactive and stable in air and humid environments at room temperature and easily form alloys with other metals. Also, the physical properties (such as colour, appearance, melting point, hardness and ductility) are similar. Pd also has very similar electronic properties to Pt. 9 Catalysts based on Pd were claimed to have higher methanol tolerance than Pt-based catalysts, 10 and be more stable than non-precious metal catalysts in acidic media and highly tolerant towards CO poisoning. 11 Unlike Pt/C, Pd-based electrocatalysts did not show any methanol oxidation peak above 0.7 V. 12 The methanol oxidation on Pt/C caused the onset of the net cathode current to shift negatively and resulted in a significant increase in overpotential. 13 Thus, Pd-based electrocatalysts tend to exhibit better ORR selectivity than Pt-based ones in the presence of methanol.
The study of Pd-based catalysts can be traced back to the first half of the last century. In the early studies, researchers mostly investigated the catalytic oxidation performance, such as methanol oxidation, 14 ethylene oxidation, 15 carbon monoxide oxidation, 16 and formate oxidation, 17 as well as various physicochemical properties [18][19][20][21] and structural characteristics. 22,23 For a pure Pd catalyst, due to its intrinsic properties, it is difficult to achieve catalytic performance for the ORR and stability close to that of Pt. For example, the binding ability of Pd to oxygen is too strong, which hinders the reduction of oxygen. The ORR activity of poly-Pd is about five times lower than that of poly-Pt in HClO 4 solution. 24 In addition, by alloying with other elements (especially transition metals), the ORR activity of Pd based catalysts can be enhanced dramatically 25 and the production cost of the electrocatalysts can be further reduced.
Overall, a large and growing number of studies have investigated the role of Pd alloys as ORR catalysts and strive to enhance the activity to match, or even exceed, Pt-based catalysts. Herein, scientometric analysis and traditional review methods are combined with an aim to provide comprehensive and critical insight and indicate a clear direction for future research in this area. In terms of scientometric analysis, based on the Bibliometrix/Biblioshiny R-package, 26 Citespace 27, 28 and VOSviewer 29 software, annual production, the top 10 sources and countries by publication, co-citation and keyword analysis, and co-keyword networks are discussed. In terms of the conventional review, this work elucidates the development of ORR mechanisms in different media, volcano plots of previously reported ORR electrocatalysts, the theoretical possibility of an enhancement in the ORR activity and the recent advances in Pd-based electrocatalysts.

Scientometric analysis on Pd alloys for the ORR
Before 1970, although there were several reports on the ORR of Pd alloys, 24,30,31 most of them were not exhaustive and systematic. Based on the database in the Web of Science from 1970 to 2020, a total of 400 articles were obtained on Pd alloys for the ORR. As shown in Fig. 1, the first report appeared in 1997, and the studies in this field have grown significantly in the past decade. In this study, Pattabiraman et al. dispersed Pd catalysts on various carbon supports and reported their ORR activity in alkaline media for the first time. 25 Also, amorphous PdP alloy catalysts for the ORR in alkaline media were reported by Podestá et al. 32 In 2004, Savadogo et al. proposed active Pd based alloy catalysts for the ORR in acidic media for the first time and showed that the PdCo alloy (the atomic ratio of Pd to Co is 72 : 28) may exhibit better performance for the ORR than Pt materials. 33 Since then, the relevant research in this area has gradually increased. [34][35][36] For example, Shao et al. synthesized PdFe nanoparticles for the ORR and found that the surface-specific activity of the PdFe alloys is related to the Pd-Pd bond distance: the shorter the bond distance, the higher the activity. 36 Based on the Citespace software, taking advantage of cluster analysis of title terms, emerging trends and new developments in this field can be easily identified, as shown in Fig. 2. Clusters (groupings) could be found and numbered in the size descending order of the cluster size. For example, the size of cluster #0 (Pd nano-particle) is the largest. Each point in the graph represents a node, and the node means keywords or references. For 'tree ring' shaped nodes, the number of citations the article received in a particular year affects the thickness of the nodes; a red ring shows that a particular year denotes a citation burst (a surge of citations). The network indicates the co-citation and co-occurrence of keywords. The network of different types of entities has different meanings. For instance, lines that connect nodes in a cited reference network are co-citation links. Table 1 presents the information about the largest four clusters by size. The silhouette value reflects the homogeneity and consistency of the cluster.
Pd nano-particle (cluster #0) is the focus of research in this field. The study of surface structure (cluster #1) exhibits a good trend for possible future development. Alloyed with platinum (cluster #2), carbon nanotube as the carrier (cluster #4), and applied in alkaline media (cluster #5) are also hotspots.
The keywords summarize the main purpose of the literature to a certain extent, and they can comprehensively interpret its content. Based on the VOSviewer software, co-keyword network visualization maps are shown in Fig. 3 to understand the hotspots and directions in this field. Although there is no specific quantitative data, the direct connection of keywords and the frequency and proportions of the research can be clearly identified from Fig. 3(a and b), respectively. For the topic, ''Palladium (Pd) alloy(s) for ORR'', the keywords highlighted in the figure are: ''nanoparticles'', ''electrocatalysts'', ''stability'', and ''enhancement'' and, through the co-keyword network, the top research directions and underlying problems in this field are revealed, which are namely: (1) Nanoparticles of palladium alloy(s) prepared by various methods are used for ORR catalysis. Catalysts with different morphologies such as nanowire, core-shell and monolayer are widely studied.  (2) The main metals alloyed with Pd are Pt, Co, Au, Fe, Ni and Cu.
(3) Stability and catalytic activity are still the focus in this field. Developing methods to realize comparable performance with Pt is one of the key difficulties in this research.
(4) The pursuit of simpler and more effective synthetic methods, such as facile synthesis and one-pot synthesis.
(5) The combination of carbon material supports, such as graphene, reduced graphene oxide, carbon nanotubes, etc.
(6) In addition to oxygen reduction, Pd alloy catalysts are widely used in formal acid oxidation, methanol oxidation, and ethanol oxidation. The preparation of dual-functional catalysts, which have both oxygen reduction activity and organic oxidation activity, is an interesting research topic.
(7) There is limited research on the mechanism and kinetics, and more in-depth work is needed.

ORR mechanisms
The ORR mechanisms on noble metal surfaces, such as Pd and Pt, have not been completely understood despite the intensive studies over the past few decades. 37 In general, due to the similarities of Pd and Pt, the ORR processes on these two electrocatalysts are basically similar, that is, this reaction starts with the adsorption of molecular oxygen on the catalyst surface, and then two mechanisms have been generally accepted: one is a direct four-electron mechanism where O 2 is reduced directly to water without the production of H 2 O 2 ; the other is a multistep four-electron mechanism where O 2 is reduced to H 2 O 2 and then further reduced to water. Depending on the pH value of the electrolyte, the electrochemical reduction of oxygen follows different pathways. Specifically, in acidic solution, the reaction is: while in alkaline solution the overall reaction is: In acidic media, the ORR process undergoes the following steps: 38 Pathway a: OOH ads -O + OH ads (8) O + H + + e À -OH ads (9) OH ads + H + + e À -H 2 O OOH ads + H + + e À -H 2 O 2 (11) For four-electron mechanisms, as shown in Fig. 4(a), after the oxygen gas is transferred to absorbed oxygen (eqn (3)), there are two common pathways. One is denoted as O 2 -diss-gas 39 (Pathway a): the absorbed O is formed through the O 2 dissociation step (eqn (4)), and then combines with a proton and an electron to produce absorbed OH (eqn (5)), followed by the reduction with protons and electrons to form H 2 O (eqn (6)).
The other one is called HOO-form-gas 39 (Pathway b): the intermediate superoxide is formed by an oxygen molecule, a proton and an electron (eqn (7)). After the breaking of the O-O bond, the absorbed O and OH species are produced (eqn (8)) and then reduced to H 2 O by combining with protons and electrons (eqn (9) and (10)). Also, the intermediate superoxide combines with a proton and an electron to form hydrogen peroxide, and then becomes absorbed OH species by a cleavage reaction (eqn (11) and (12)).
According to the general ORR mechanism in alkaline solutions proposed by Anastasijevic et al., 40 it consists of the following steps: As can be seen from Fig. 4(b), the first step is that O 2 À is formed by the absorbed oxygen molecule with an electron (eqn (13)) and the intermediate O 2 À has been detected by surfaceenhanced infrared reflection absorption spectroscopy with attenuated total reflection in the ORR in an aqueous solution at pH = 11, as reported by Shao et al. 41 Then the formation of species HO 2 À is followed by the reaction of O 2 À with water and an electron, as shown in eqn (14). Finally, the HO 2 À is further reduced to OH À , as shown in eqn (15) and (16). Another possible pathway is that absorbed O species are formed by the dissociation of O 2 (eqn (17)), and then these species are reduced to O À (eqn (18)), followed by further reduction to OH À (eqn (19)). According to Kinoshita's work, a four-electron pathway appeared to be predominant for Pd electrocatalysts. 7 However, Kim et al. researched oxygen reduction on bare Pd in 0.1 M LiOD solution and found that about one third of the O 2 reacts at a bare Pd surface and is reduced directly to OD À via the four-electron pathway, while the rest is reduced to deuterium peroxide via the two-electron pathway. 42 Therefore, ORR catalysts are expected to have an optimal balance between both the cleavage and the reduction of oxygen.
Furthermore, representative studies focused on the ORR process and other possible mechanisms were proposed, such as the oxygen hydration mechanism, 43 peroxide mechanism, 44 aquoxyl mechanism, 44 etc. While alloying with various elements, the mechanisms on Pd or Pt alloys for the ORR 45-50 are more sophisticated and need further investigation. Advanced characterization techniques need to be applied to understand the mechanisms. For example, Wang et al. used shell-isolated nanoparticle-enhanced Raman spectroscopy (SHINERS) to illustrate the ORR processes that occur on the surface of bimetallic Pt 3 Co nanocatalyst structures. 49 They found direct spectroscopic evidence of *OOH, which suggests that the ORR undergoes an associative mechanism on Pt 3 Co in both acidic and alkaline media.
With the aid of DFT calculations to study the ORR mechanism, there were more attempts in the field of Pt-based catalysts to study the ORR mechanism, 50-55 and relatively fewer studies for Pd-based catalysts. 55,56 Except for the similarities between the two, there are differences in the ORR mechanisms for Pd and Pt electrocatalysts. For instance, the enthalpies and barriers for each step are different and the ORR process varies under specific conditions. For example, Ford et al. indicted that, while the activation barriers for the O-O bond scission steps differ by ca. 0.1 eV on Pd and Pt, the hydrogenation steps (including O 2 + H and OH + H) are kinetically less active on Pd than on Pt, suggesting that overall Pd is less effective than Pt. 44 Sha et al. considered that, under conditions with high H ad , another mechanism producing HOOH is possible, but this is unlikely under ordinary fuel cell operating conditions. 43 Additionally, Ou et al. pointed out that, in the presence of hydrated protons, the mechanism of the ORR on the Pd (111) surface only involved the O 2 molecule dissociation mechanism, whereas the mechanism of the ORR on the Pt(111) surface involved the dissociation mechanism of both O 2 molecule and OOH species. 55 As for the entire four-electron ORR, the protonation of adsorbed O atoms to form OH was the slowest step, and it was thus the rate-determining step (rds) for both the Pd (111) or Pt (111) surfaces. Such an rds finding of the ORR explained why Pt-and Pd-based electrocatalysts showing weaker bonding with atomic oxygen had higher ORR activities.
In addition, there is relatively limited research focusing on the ORR conducted in neutral media (pH = 7). 57,58 However, microbial fuel cells (MFCs) operating under neutral conditions show promising application prospects. The low concentrations (10 À7 M) of H + and OH À have negative effects on the ORR kinetics, resulting in high overpotentials. 59 Precious metal catalysts (Pt, Pd, etc.) tend not to be the first choice due to the high cost compared to the low power output produced. 60 More breakthroughs are needed in this field. Hence, methods to enhance the catalytic activity and the factors affecting the ORR activity are described in detail below.

Volcano plots of ORR catalysts
The theoretical description of electrocatalytic phenomena is extremely challenging. In this regard, density functional theory (DFT) calculations have been extensively used to correlate the ORR activity with certain parameters of electrocatalysts. The experimental evidence suggests that the ORR activity is related to the strength of the metal-oxygen (M-O) bond, which corresponds to an adsorbed oxygen atom as the ORR intermediate. 61 Metals with very negative oxygen binding energy easily cleave molecular oxygen but subsequently form strong inert oxides and have an inappropriate cathode potential. Metals with very positive oxygen binding energy, for example, noble metals such as Au and Ag, have a high barrier for the dissociative adsorption of oxygen, so they also tend to be poor ORR catalysts. 62 With the deepening of our understanding in this area, researchers have linked the physical quantities representing the ORR activity of catalysts with some other factors and represent findings in so-called volcano plots. The development of volcano plots of the ORR activity against various factors is shown in Fig. 5. It is noted that these factors, including the d-band vacancy value, the incipient -O or -OH adsorption potential, the oxygen binding energy, the d-band center, the degree of alloying of the electrocatalysts and the adsorption strength for certain surface reactions, can account for various ORR catalytic performances. All these factors influence each other in determining the best conditions for efficient catalytic behavior for the ORR and, therefore, it can be noted that better catalytic activity can be achieved under the optimal conditions of these factors.
In the 1970s, Appleby showed a volcano plot of the current density, i, in the ORR vs. d-band vacancy values of the electrode metal. 63 Pt, Pd and PtRu alloys were at the peak position of this curve, as shown in Fig. 5 the relative DG ads (V) (Fig. 5(b)). 37 DG ads is the incipient -O or -OH adsorption potential from anodic cyclic voltammetry. These similar volcano relationships can also be observed in the overpotential Z-DG ads plots at constant i for the ORR in acid solution for platinum group metals, Ag, Au and their alloys. In 2004, Nørskov et al. presented a method for calculating the stability of reaction intermediates of electrochemical processes based on electronic structure calculations. 61 Trends in the ORR were plotted as a function of the oxygen binding energy. As shown in Fig. 5(c), in terms of pure metal, Pt has the best activity towards the ORR and Pd occupies the second place. The model explained the reason that Pt is the best elemental cathode material and alloying can be used to improve the activity of Pd. Later, Stamenkovic et al. studied the Pt-metal system in detail and described that the variations in the electronic structure determine the trends in the catalytic activity of the ORR across the periodic table. 64 The authors showed that Pt alloys involving 3d metals (such as Ti, V, Fe, Co, and Ni) are better catalysts than pure Pt because the electronic structure of the Pt atoms on the surface of these alloys has been modified slightly. At first, the catalytic activity is related to the adsorption energy of oxygen DE O . DE O should be moderate, not too high or too low (the Sabatier principle). It is well-established in a number of studies that surface bond energies correlate with the average energy of the d-states on the surface atoms to which the adsorbate binds (the d-band center). [69][70][71][72][73] It is difficult to measure the oxygen bond energy; however, the d-band center is accessible by experiments such as synchrotron-based highresolution photoemission spectroscopy. 74 The measurement of the d-band center provides a powerful tool to directly correlate the variations in the ORR activity with the changes of the surface electronic structure. As shown in Fig. 5(d), the activity predicted from DFT simulations is shown in black, and the measured activity is shown in red. Although there is a certain deviation between the two, the overall trend is generally consistent.
Shao et al. reported a volcano-type dependence of activity on the energy of the d-band center of Pd monolayers, with Pd/Pt(111) at the top of the curve, by using DFT calculations, as presented in Fig. 5(e). 65 The authors considered that, due to the strong surface segregation of Pd at high temperature, the downward shift of the d-band center of the Pd that constituted the alloy surface could lead to the high ORR activity. Later, Shao et al. established the volcano-type dependence of the ORR activity on the binding energy of oxygen ( Fig. 5(f)) by combining experimental data and DFT calculations, and predicted that the Pd overlayer on a Pd 3 Fe(111) alloy might have the most active performance. 66 In 2007, Suo et al. described the ''volcano'' relationship between the ORR activity and the degree of alloying, shown in Fig. 5(g), and noted that the catalytic activity correlates with the  68 Dashed lines are based on a simple Arrhenius relation; the solid line is obtained with the consideration of lateral interactions between adsorbed reaction intermediates. The finite interactions between different reaction intermediates may also shift the optimal adsorption energy and the corresponding maximum activity. In Fig. 6, the development of volcano plots and their use to understand the ORR activity and the breakthroughs in the application of Pd (alloys) for the ORR are presented, which clearly shows that the combination of theories and experiments has greatly contributed to the rapid development of this field in recent years. Since 1997, when dispersed Pd-based catalysts on carbon for the ORR in alkaline media were first reported, 32 there have been increasing numbers of novel Pd alloys with superior catalytic performance. From realizing comparable performance to Pt to far exceeding Pt towards ORR performances, 75 research on Pd-based catalysts is constantly delivering breakthroughs. It is believed that Pd-based catalysts could deliver greater achievements in the near future.

Methods to enhance the ORR activity
There are many factors that contribute to the enhancement of the ORR activity. In this section, commonly used methods to enhance the ORR performance are concluded.

Alloying
There is a growing recognition of enhancing ORR performances by the effect of the alloying process. The ORR activity of Pd can be dramatically enhanced by alloying with other metals, such as Pt, 76 Co, 33 Fe, 36 Cr, 77 Ni, 78 Rh, 79 Ag, 80 Au, 81 Cu, 82 Mo, 75 Ti, 83 Sn, 84 Y, 85 Mn, 86 W, 87 Zn, 88 Ru 89 and Ir. 90 Fernundez et al. proposed a simple thermodynamic model to interpret the enhanced activity on alloying. 91 This model involves the combination of one metal that will easily break the O-O bond of O 2 (forming adsorbed atomic oxygen) with another metal that will easily reduce the adsorbed atomic oxygen. The analysis of the Gibbs free energies of these two reactions guides the selection of combinations of metals that can produce alloy surfaces with enhanced activity for the ORR. In their experiment, PdCo/C (10-30% Co) electrodes exhibited remarkable activities for the ORR, close to that of carbonsupported Pt. For Pd with fully occupied valence d-orbitals, Balbuena et al. argued that alloying with transition metals, such as Co with unoccupied valence d-orbitals, reduced significantly the Gibbs free energy of both the first charge-transfer step, and the steps involving the reduction of intermediates. 92 In addition, intrinsic Pd surfaces bind oxygen too firmly to allow the efficient removal of the adsorbed reaction intermediates, which affects the ORR activity. For example, in Adzic et al.'s work, the volcano-type dependence of the ORR activity on the binding energy of the oxygen and the d-band center of the noble metal overlayers was established. 66 They noted that alloying Pd with some transition metals could lower the d-band position of the noblemetal overlayers and therefore may improve the activity significantly by inducing strain and electron redistribution between the substrates and the overlayers. Henkelman et al. investigated the effect of the alloy composition in Pd/Cu nanoparticles on the ORR activity. 93 The activity enhancement is due to the difference in how the two metals respond to a shift in their d-band centers. For instance, charge transfer from Cu to Pd raises the d band of Cu and lowers that of Pd, resulting in stronger binding between oxygen and

View Article Online
Cu and weaker binding between oxygen and Pd. Henkelman et al. also pointed out the charge redistribution between the core and shell in Pd-based core-shell nanoparticles, which became an important factor for lowering the d-band center to promote oxygen reduction. 94 Moreover, the binding energy of the dissociated O 2 molecule is linearly related to the average d-band energy of electrons in the Pd shell. Onana et al. argued that the enhancement of PdCu alloys was attributed to an optimal d band property that makes OOH dissociative adsorption easier. 95 Recently, Crooks and co-workers also demonstrated that the activity of the individual sites on the surface of the nanoparticles can influence the overall activity. 96,97 For example, alloying Pd with Au weakens the O-binding energy on Au x Pd (300Àx) alloys, which results in increased ORR activity, but in Pd x Ir (100Àx) the effect is opposite as the O-binding energy increases at the active sites containing Ir. 96,97 Additionally, Ou revealed that an ideal Pd-based bimetallic alloy catalyst for the ORR should possess simultaneously negative alloy formation energy and negative surface segregation energy of Pd. 98 The alloy formation energy of Pd with transition metals M can be mainly determined by their electron interaction, which could be the origin of the negative alloy formation energy for Pd-M alloys. The surface segregation energy of Pd is primarily determined by the surface energy and the atomic radius of M. M with a smaller atomic radius and higher surface energy would tend to favour the surface segregation of Pd in the corresponding Pd-M alloys.
Furthermore, recent studies found that the ORR activity was related to transition-metal dissolution. For example, Han et al. found that the amount of transition-metal dissolution from PtM (Pt-metal) nanoparticles increases when Pt is alloyed with more negative V dissolve transition metals despite their strong alloy-formation energy, where V dissolve or the dissolution potential is the thermodynamic potential of the dissolution of the transition metal (M 3 M n+ + ne À ) at pH = 0. 99 Moreover, the increase of transition-metal dissolution and the decrease of V dissolve were correlated with the ORR activity of PtM nanoparticles. Among all PtM nanoparticles examined, PtFe nanoparticles were found to have the highest ORR specific activities, roughly three times better than that of Pt nanoparticles before voltage cycling. Similar to PtM, PdM is expected to have an analogous effect.

Structure, strain and ligands
Many studies describe the role of the structures and morphologies of electrocatalysts in the enhancement of the ORR. There have been many efforts to manipulate the structure and shape of catalysts during the synthetic process to fabricate metallic nanoparticles. Structure or shape-controlled metallic catalysts have exhibited improved electrochemical activities because of the exposure of a particular surface, which is favourable for electrocatalytic reactions. Hong et al. synthesized PdPt alloy nanocrystals (NCs) with hollow structures such as nanocages with porous walls and dendritic hollow structures. 76 They found that the type of surface facet plays a crucial role in determining the ORR activities of PdPt NCs. PdPt nanocages prepared from octahedral Pd NC templates exhibited the largest improvement of the ORR performance. Duan et al. prepared nanoporous PdCr alloys with uniform ligament dimensions and a controllable bimetallic ratio. 77 Specifically, the nanoporous Pd 75 Cr 25 alloy displayed the highest specific kinetic activity with a value of B0.24 mA cm À2 at 0.9 V, which is more than three times higher than that of Pd, and also higher than that of Pt/C (0.15 mA cm À2 ) catalysts. The Pd 75 Cr 25 alloy exhibited a higher mass activity (0.16 A mg À1 ), which was nearly 1.4 times that of Pt/C, and 3.1 times that of nanoporous Pd. The improved overall ORR performances of the PdCr alloy could be ascribed to its excellent structural integrity and continuity as well as the appropriate changes in the Pd electronic structure induced by alloying with Cr. By DFT calculations, the authors clarified the reasons for the enhanced ORR performance. The decrease of electron backdonation from the Pd 4d orbital to the 2p* orbital of O usually results in the downshift of the d-band center of Pd, generating a weaker metal-O bond, which was proposed to be the possible cause for the higher ORR activity of PdCr alloys than Pt/C and Pd catalysts. 100,101 Core-shell nanoparticles often exhibit improved catalytic properties due to the lattice strain created in these core-shell particles and the substrate material provides an additional advantageous modification of the electronic structures of surface atoms. Chen et al. reported the synthesis of bimetallic PdAu nanoparticles with a core-shell construction. 102 The sufficient lattice strain imposed by the Au core and Ag removal, which could tailor the d-band center of the Pd shell, accounted for the enhanced ORR performance of the core-shell Au@Pd nanoparticles. Also, Suo et al. noted that the key to improving Pd-based catalysts for the ORR is alloying Pd with elements of smaller atomic size to form a ''Pd-shell/alloy-core'' structure to take advantage of the lattice-strain effect and to prevent the disadvantageous surface-ligand effect. 67 Xiong et al. also studied the surface structure and the strain in PdPt core-shell nanocrystals. 103 From a geometrical phase analysis, they found that most of the Pt shells in the icosahedral Pd@Pt nanocrystals are dominated by compressive strain at a specific size, while compressive and tensile strains co-exist in the Pt shells of the octahedral Pd@Pt nanocrystals, which might be responsible for their different ORR properties. Furthermore, Yang et al. recently prepared three bimetallic PdZn nanoparticles for the ORR and found that the catalytic performance followed the order of Pd@Zn_core-shell 4 PdZn_ordered c PdZn_disordered, which can be attributed to the effects of different surface structures of the catalysts. 88 Firstly, from the structural perspective, the well-defined core-shell structure and ordered structure can provide more predictable control over geometric and structural effects for catalysis optimization. Secondly, stronger electronic interactions exist in Pd@Zn_core-shell over PdZn_ordered and over PdZn_disordered. Such electronic interactions could make the d-band center of Pd downshift, weakening the adsorption of the oxygenated intermediates.
He et al. prepared AuM (M = Pt/Pd) alloyed flowerlike-assembly nanochains (FANs) for an enhanced ORR. 104 The authors View Article Online discovered that the enhanced performances of AuM FANs are mainly ascribed to the interconnected porous bimetallicalloyed structures, which provide an enlarged electrochemical surface area (ECSA) and more available active sites for the ORR, promote O 2 diffusion and electron transport, and suppress Ostwald ripening. Furthermore, the highly ordered structures and tight interconnection of AuM FANs facilitate the mass transportation of reactant molecules and increase the electrochemically utilized Pt/Pd atoms. Conformal deposition of Pt as ultrathin shells on facetcontrolled Pd nanocrystals offers a great opportunity to enhance the catalytic performance while reducing the mass loading. Wang et al. proposed such a system by depositing the active metal as shells of only a few atomic layers on nanocrystals made of another metal, together with an optimized surface structure. 105 Owing to lateral confinement imposed by twin boundaries and thus vertical relaxation only, the platinum overlayers evolve into a corrugated structure under the compressive strain. Brandiele et al. described the facile synthesis of Pd 3 Y alloy nanoparticles and confirmed that a strong ligand effect due to the introduction of Y in the Pd lattice affords better catalytic activity through electrochemical characterization. 85 Wang et al. describe a facile method for the preparation of Pd-rich Pd x Co alloy nanoparticles supported on carbon, using an adsorbate-induced surface segregation effect. 106 The electronic properties of Pd were modulated by alloying with different amounts of Co, which affects the ORR activity. Xiao et al.'s work offered compelling evidence that the surface strain in Pd-based alloyed nanoparticles can be readily tuned by adding Co and Fe elements to achieve optimal electrocatalytic performance. 107 Moreover, Ham et al. showed that the electrochemical activity of Pd 3 Co alloy catalysts towards the ORR can be enhanced by adding a small amount of Ir. 90 Their study highlighted that the enhancing effect is attributed to the synergetic interplay between the surface electronic structure modification due to underlying Ir atoms and the compressive strain caused by the Pd 3 Co substrate. DFT calculations showed that the PdIrCo ternary alloying leads to a noticeable reduction in the DOS peak intensity near the Fermi level and a downshift in the d-valence band center, compared to the monometallic Pd(111) surface. They further noted that the addition of Ir to the Pd 3 Co alloy causes a slight increase in the activation energy for O-O bond breaking but it significantly decreases for O and OH hydrogenation, leading to improved ORR activity. The ORR activities of typical Pd-based electrocatalysts with different structures mentioned in this section are compared in Table 2.

Particle size and crystal facets
The particle size of catalysts is also a complicated element in determining the ORR activity and many studies focused on the relationship between the size of Pd nanomaterials and the catalytic performance. For example, Zhou et al. noted that the Pd nanoparticle size has a strong impact on the ORR in acidic solutions and an optimal average Pd particle size is in the range from 5.0 to 6.0 nm. 108 Meanwhile, Jiang et al. conducted similar studies in alkaline solutions and pointed out that the ORR activity of the catalysts varies with the Pd particle size and carbonsupported Pd catalysts with average Pd particle sizes from 3 to 16.7 nm are highly active for the ORR. 109 It is noted that the stronger adsorption of OH on smaller particles would block the active reaction sites, so the particle sizes are not as small as possible.
In regards to the particle size effect of PdM alloys, the number of related studies is relatively limited. Castegnaro et al. revealed that, in alkaline electrolyte, for bimetallic nanoparticles consisting of highly crystalline nanoalloys with a size of about 5 nm, the charge transfer involving Pd and M atoms affects the activity of the catalysts. 110 Furthermore, they noted that different compositions may induce different valence band structure and the materials whose d-band center is closer to the Fermi level will have higher ORR activity.
Crystal facets are also crucial aspects, since relevant research found that the ORR activity depends strongly on the orientation of the Pd surface. 10 For instance, Kondo et al. studied catalytic behaviours for the ORR in acidic electrolytes on low index planes, the n(100)- (111) and n(100)-(110) series of Pd, and concluded that the (100) lattice plane is a possible active site for Pd and the ORR activity does not depend on the step structure. 111 Specifically, the specific activity has the following order on the low index planes of single crystal Pd at 0.90 V vs.

View Article Online
In contradiction to the above results, the high catalytic activity of Pd nanorods as compared to spherical Pd nanoparticles has been attributed to the prevalence of Pd(110) facets. 113 Related oxygen reduction studies on Pd nanocubes with a preferential (100) surface orientation have been published. 114,115 For example, Shao et al. indicated that the ORR activity of Pd nanocubes enclosed by (100) facets was one order of magnitude higher than that of Pd octahedra enriched with (111) facets. 114 They have demonstrated that the ORR activity was strongly dependent on the surface structure of Pd nanocatalysts with exposed (100) facets being much more active than (111) facets.

Dealloying
Dealloying has generally been known as the selective removal of a less noble component for a given bimetallic alloy. Dealloying is an effective and crucial strategy to control and modify the surface electronic structure and the chemical composition of an alloyed electrocatalyst, thus enhancing the activity and stability. Much of the related research has been focused on Pt-based catalysts synthesized by dealloying towards the ORR. 116,117 For Pd based catalysts, relatively limited investigations have been conducted. These Pd based ORR catalysts include PdCo, 118 PdCu, 119 PdZr 120 and a dealloyed PdAg core Pt monolayer shell electrocatalyst. 121 Yang et al. investigated the ORR activity of electrochemically dealloyed PdCu 3 thin films and noted that the ORR activity enhancement is due to the compressive strain in the Pd overlayer in the dealloyed Pd-Cu films. 122 Also, the researchers found that the dealloyed structures and the ORR activity are dependent on the nature of the noble component of the alloy. Gunji et al. synthesized an electrochemically dealloyed PdCu 3 intermetallic compound for the ORR. 119 They indicated that, after electrochemical dealloying, the oxygen binding energy of PdCu 3 was lower than that of the Pd(111) structure, and the longer distance between oxygen and the catalyst surface explained the weaker binding of oxygen, which led to the superior ORR activity.
In Lu et al.'s research, the electrocatalytic activity was enhanced by the dealloying process, which partially leached out the inactive transition metal species (Mn and Ni) on the surface and generated Pd-rich surfaces on the nanoparticles. 86 The Pd-based dealloyed catalysts exhibited only a slight degradation in ORR activity in alkaline media, which could be reversed by repeating the dealloying process. Mondal et al. demonstrated electrochemical dealloying-assisted Pd-based catalysts (Co x Cu y Pd z ) for the ORR and indicated that the elemental composition and dealloying-induced lattice strain and the change in the electronic structure due to the downshift in the d-band center of Pd control the overall performance of the alloy electrocatalysts. 123 However, the dealloying effect of electrocatalysts for the ORR can be controversial. Lee et al. reported a comparison of alloyed and dealloyed AgPdPt nanoframes for catalyzing the ORR and concluded that the mass activity of the ternary catalysts was higher than that of the dealloyed ones with less Ag. 124 Thus, to understand the dealloying effect of PdM, more studies including both experimental and theoretical approaches need to be conducted. For example, addressing questions of how to conduct systematic research in this area and to provide a general design for these systems is required.
The enhancement of the ORR activity can be attributed to many components, like alloying, structure, strain, ligands, particle size, crystal facets and dealloying. Most previous studies use theoretical methods to verify their correctness by establishing simple models, and the surface atomic structure of the alloy could be a very important factor for predicting the activity. However, bonding between metal atoms may also result in more complicated alloy properties. Catalytic reactions can also take place at surface defects, which is difficult to simulate and verify. The formation of intermediates (such as hydroxyl) could also dominate the kinetics, or these could bind irreversibly to cleaved oxygen. These factors need to be further examined.

PdM (Pd-metal) nanoparticles obtained by different preparation methods
There are many methods reported in the literature to prepare PdM alloys and they could be divided into chemical reduction reactions, electrochemical methods, dealloying and so on. The synthesis methods play a vital role in the morphology of catalysts, which significantly affects the performance of the catalysts. In the following chapter, the preparation methods of PdM nanoparticles for the ORR will be discussed in detail.

Chemical reduction reactions
The chemical reduction method is the most common method for synthesizing alloys. In general, metal precursors are mixed with surfactants, reducing agents and/or capping agents in a solvent (aqueous phase or organic phase), and, by controlling the reaction conditions, alloy catalysts are prepared.
6.1.1 Hydrogen (H 2 ). As one of the most common reducing reagents, hydrogen is cheap, clean and relatively simple to use. However, considering the nature of H 2 , there is a certain degree of operational risk.
Ramanathan et al. synthesized Pd 3 Ni/C catalysts by a co-precipitation method from a mixture of Pd(NO 3 ) 2 Á2H 2 O and Ni(NO 3 ) 2 Á6H 2 O salts in an aqueous solution. 125 Pd-Ni hydroxides were formed by slowly adding sodium hydroxide (NaOH) solution and hydrogen was bubbled as a reducing agent at 80 1C, which was further annealed in a reducing atmosphere (10% H 2 + 90% Ar) to reduce the hydroxides to the metallic form at about 300 1C for 2 hours. Similarly, Wang et al. prepared PdCu alloys by a reduction reaction in a hydrogen atmosphere. 126 They pointed out that the catalyst with the best ORR performance was the alloy nanoparticles with a Pd-Cu molar ratio of approximately 1 : 1.
Dai et al. reported PdW alloy catalysts for the ORR by reducing palladium(II) chloride and ammonium tungstate with H 2 in a tube furnace, 87  two-fold that of Pd/C; however, it was lower than that of commercial Pt/C. In addition, according to recent reports, PdCoMo catalysts, 34 carbon-supported PdPtFe alloys, 127 core-shell Pd-Co@Pd/C nanoparticles, 106 Pd 3 M (M = Fe, Ni, Cu, Co) alloys, 128,129 bimetallic Pt surface-enriched PtPd(x) nanoparticles, 130 and core-shell fct-PdFe@Pd nanoparticles 131 were prepared by using H 2 to reduce metal precursors.
6.1.2 Sodium borohydride (NaBH 4 ). Sodium borohydride (NaBH 4 ) as the reducing agent is often used in aqueous phase reactions with the advantages of low cost and a fast reaction rate. Generally, high temperature or high pressure reaction conditions are not required; therefore, the energy consumption of the preparation is relatively small.
Yang et al. synthesized three types of PdZn with different surface architectures by using NaBH 4 as the reducing agent. 88 Through heat treatment under a 0.1 MPa Ar/H 2 (10%) atmosphere at 600 1C for 5 h, the disordered PdZn alloys turned into ordered PdZn alloys, and the mass activity (MA) and specific activity (SA) of ordered PdZn were markedly higher than those of disordered PdZn. Core-shell PdZn was also obtained through a slightly different reaction process and had better ORR performance. The TEM images of these disordered PdZn, ordered PdZn and core-shell Pd@Zn are presented in Fig. 7(a-c).
Fernández et al. prepared carbon supported PdCoAu electrocatalysts by a reverse microemulsion method using sodium dioctylsulfosuccinate (AOT) as the surfactant, heptane as the oil phase and sodium borohydride as the reducing reagent. 83 Remona et al. synthesized PdPt alloys by a microemulsion method using NaBH 4 as the reductant at room temperature. 132 Neergat et al. prepared carbon-supported PdFe electrocatalysts (Pd to Fe ratios = 1 : 1, 2 : 1, and 3 : 1) by a co-reduction method at 80 1C in alkaline media (pH = 10) with sodium borohydride (NaBH 4 ) as a reducing agent and without any stabilizing agents, 12 as shown in Fig. 7(d). Similarly, Ramos-Sánchez et al. reported that mesoporous carbon-supported nanoparticulated PdNi 2 (Fig. 7(e and f)) exhibited higher catalytic activity for the ORR, which was synthesized by the reduction of metal chlorides with NaBH 4 in aqueous media. 133

View Article Online
The preparation of carbon-supported PdCo alloys, 34,135 PdV alloys, 136 PdFe alloys, 12 and PdAu nanoparticles 137 also used the NaBH 4 reduction method.
Recently, some PdM alloys were fabricated via the cyanogelreduction method with NaBH 4 as the reducing agent. A schematic illustration is presented in Fig. 7(g). Firstly, a cyanogel of two metals is formed, and then reduced with NaBH 4 . Liu et al. reported the preparation of PdNi nanocorals with hierarchical porosity materials through the formation of a Pd-Ni cyanogel followed by chemical reduction with NaBH 4 solution. 134 Xu et al. prepared polyallylamine (PAA)-functionalized PdCo alloy nanonetworks by a functional molecule assisted cyanogelreduction method using a NaBH 4 /PAA mixture as the reductant. 138 A similar strategy was used to prepare polyethyleneimine (PEI)functionalized PdNi alloy nanostructures, 139 and reduced graphene oxide supported PdNi alloy nanocrystals. 140 6.1.3 Polyol. Polyol as the reducing agent is often used in solvothermal reactions with the advantages of low-cost and low toxicity. Various polyols are reported to be used in the synthesis of PdM alloys.
Zhao et al. reported a polyol reduction method using ethylene glycol as the solvent as well as reducing reagents to synthesize carbon-supported PdNi nanoalloy electrocatalysts. 78 After the heat treatment at 500 1C, the degree of alloying (atomic percentage of Ni in the alloy) was increased from 0 to 13.3%. They found that, with increasing the heat-treatment temperature (or Ni content), more Ni atoms get into the Pd lattice, resulting in higher lattice strain, weaker M-O ads bonding, and higher specific activity. However, the ECSA decreases significantly with temperature, and thus the mass activity shows a volcano relationship with temperature. PdPt catalysts were also synthesized by using a modified polyol process and ethylene glycol (EG) was used as both a solvent and a reducing agent. [141][142][143] A similar process was used to synthesize PdIr alloy networks. 144 Jang et al. presented an ultrasound-assisted polyol process without any added surfactant, pH adjuster, or stabilizer to prepare PdCo nanoparticles with two different structures (coreshell structure and bimetallic alloy nanoparticles, as shown in Fig. 8(a and b)). 145 The difference between the preparation methods of the two structures is whether the Co precursor is firstly added to prepare cobalt seeds or Pd and Co precursors are added to the reaction at the same time. The Pd 4 Co coreshell structure shows dramatically enhanced ORR activity (its mass activity was 173 mA mg À1 and its specific activity was about 480 mA cm À2 ). In addition, PdM (M = Ni, Fe, Co) nanoparticles were obtained through the reduction of the metal precursors, and 1,2-hexadecanediol was used as a reducing agent and oleic acid and oleylamine as capping agents. 146 Octahedral Pt-Pd nanoparticles were synthesized by reducing metal precursors with glycerol as a reducing agent and showed good ORR activity in buffer solution. 58 6.1.4 Ascorbic acid (AA). Ascorbic acid (AA) is another widely used reductant in the preparation of PdM alloys and is often applied in solvothermal reactions. Besides, Huang et al. found that ascorbic acid not only acts as the reductant but also works as a weak acid to remove heteroatoms, 147 thus obtaining catalysts with abundant defects.
Luo et al. reported PdMo bimetallene (Fig. 8(c)) for the reduction of oxygen with the highest mass activity (16.37A mg PGM

À1
) so far in 0.1 M KOH solution at 0.90 V. 75 This activity was 77.9 and 327.4 times higher than that of commercial Pt/C and Pd/C catalysts, respectively. The value of the specific activity was 11.64 mA cm PGM

À2
. They used ascorbic acid (AA) to reduce [Pd(acac) 2 ] and Mo(CO) 6 in oleylamine at 80 1C for 12 h to obtain this material. In acid electrolytes, PdMo bimetallene also exhibited better ORR activity than Pt/C, but its stability was poor for practical applications. Zuo et al. prepared PdCuCo anisotropic structure catalysts by using ascorbic acid (AA) to reduce metal ions. 148 2 ]) in oleylamine (OAm) with ascorbic acid (AA) or benzoic acid (BA) serving as the reducing agents. 149 Through the surface treatment protocol by adding a mixed solution of H 2 O 2 and sulfuric acid to remove the surface Ni and Cu atoms, the treated catalysts (PdCuNi-AB-t/C) exhibited a mass activity of 0.45 A mg À1 Pd in alkaline medium at 0.90 V vs. RHE. It is noted that, before the acid treatment, the MA of the catalysts was 0.10 A mg À1 Pd under the same conditions. Besides, different reducing agents resulted in various catalytic activities; the combination of BA and AA was better than sole BA or sole AA in this research. Feng et al. prepared PdNi icosahedra (Fig. 8(d)) for the ORR with a mass activity of 0.22 A mg À1 Pd and a specific activity of 0.66 mA cm À2 Pd at 0.9 V vs. RHE in 0.1 M KOH solution. 150 Pd(acac) 2 and Ni(HCO 2 ) 2 Á2H 2 O were used as precursors, ascorbic acid (AA) was selected as a reducing agent, and oleylamine (OAm) and 1-octadecene (ODE) were applied as solvents and stabilizers. PdPt alloy nanodendrites 151 were also prepared by using AA to reduce metal ions. Yan et al. reported the synthesis of RhPd alloy nanocrystals for the ORR by using an extremely slow injection method 79 (Fig. 8(e and f)). This synthesis involved the simultaneous injection of Na 3 RhCl 6 and Na 2 PdCl 4 using a syringe pump at a rate of 2 mL h À1 into ethylene glycol (EG) with ascorbic acid (AA) and KBr serving as reducing and capping agents, respectively. The Rh 8 Pd 92 alloy octahedra exhibited high mass activity with a value of 0.18 mA mg À1 in terms of the equivalent Pt cost.
6.1.5 Other reducing agents. In addition to the above common reducing agents, other reductants might have excellent and unexpected effects in different synthesis reactions. However, some reagents are expensive, highly toxic, volatile, and not suitable for mass production. In Zhang et al.'s work, a route to fine tailoring of the PdCu nanocrystal morphology by controlling the concentration of the reactants was introduced. 152 By reducing copper(II) acetylacetonate (Cu(acac) 2 ) and palladium(II) acetylacetonate (Pd(acac) 2 ) with formanilide in the presence of oleic acid and oleylamine, different PdCu nanocrystals, such as nanocubes, truncated nanocubes, cuboctahedra, irregular polyhedra, and branched tetrapods, could be obtained, as Fig. 8(g-i) demonstrates, wherein the PdCu nanotubes had superior catalytic performance with mass activity 0.130 mA mg À1 and specific activity 0.310 mA cm À2 in acid media. PdPb alloys were prepared by a similar method. 154 The hierarchical Pd 4 Fe nanoflowers (NFs) ( Fig. 9(a)) in Lian et al.'s work were prepared by reducing palladium acetylacetonate with Fe(CO) 5 and oleylamine at 120 1C. 155 Fig. 9(b)) for the ORR by a modified impregnation-reduction approach. 156 It included a mixture of Pd and Pb precursors with carbon black in THF, co-reduction of the Pd and Pb precursors, and heat treatment. Potassium triethylborohydride (KEt 3 BH) and lithium triethylborohydride (LiEt 3 BH) were used as reducing agents because they both had fast reduction kinetics.  158 Moreover, if cetyltrimethylammonium bromide (CTAB) was added, polycrystalline PdAu alloys were prepared. PdPt popcorn-shaped catalysts were obtained via a hydrothermal method involving the co-reduction of K 2 PtCl 4 and PdCl 2 in an aqueous solution containing PVP and NaI. 159 The authors found that if NaI was absent, octahedra (E20 nm) and very small particles (E3 nm) were generated for Pt and Pd. The bio-synthesis of palladium nanocubes (PdNCs) was realised using pine needle extract as the reducing agent and CTAB as the capping agent. As eco-friendly and readily available biomass, pine needle extract avoided the use of highly polluting chemical reducing agents. 160 1-Naphthol ethanol solution as the reductant and structuredirecting agent was applied to fabricate tree-like Pd x Ag y nanocrystals 153 (as shown in Fig. 8(j)). Chemical reduction reactions can be used to prepare metal alloy catalysts with various morphologies and properties, and a reasonable selection of reducing agents cannot be ignored. However, the reaction temperature ranges from room temperature to several hundred degrees Celsius, and the types of reagents and experimental methods used in the reaction vary greatly. To obtain a catalyst with excellent performance, it is very important to explore the experimental conditions.

Electrochemical methods
Electrochemical methods are another way to prepare alloys. Compared with traditional chemical reduction methods, electrochemical methods are environmentally friendly without organic reducing agents, surfactants or high temperature, and provide a rapid and effective route to prepare hollow nanostructures. However, to accurately control the structure, these methods often require complicated multistep operations and specific equipment, such as pulsed electrodeposition, which may not be available in many laboratories. 6.2.1 Electron reduction. The electron reduction method does not require chemical reducing agents, protective chemicals or dispersants. Moreover, this method can be used to synthesize alloys with small particle diameters when the precursor is a mixture of different metal salts. 161 Liu et al. reported PdPt alloys with a particle size around 2.6 nm for the ORR synthesized by electron reduction at room temperature with argon glow discharge as an electron source, and without any chemical reducing agents, protective chemicals or dispersing agents. 161 The Pt 1 Pd 1 /C catalyst ( Fig. 9(d)) has a larger MA of 0.488 mA mg Pt À1 in 0.5 M H 2 SO 4 solution at 0.85 V vs. RHE. The kinetic current j k of the Pt 1 Pd 1 /C catalyst was 0.976 mA cm À2 . Liu et al's group also prepared PdAu alloys by using this electron reduction method. 162 6.2.2 Underpotential deposition (UPD) and electrodeposition. Underpotential deposition (UPD) is an electrochemical surface process that involves depositing one (or two) monoatomic layers of metal on the electrode surface with an applied potential that is more positive than its equilibrium potential. 163 This method has been extensively studied over decades and could accurately and reproducibly control the amount of foreign metal atom adsorption on the substrate; 164 therefore, it is very useful for surface modification with the aim of improving the functional properties, such as catalytic activity and selectivity. 165 Electrodeposition has been widely used in the preparation of metal alloys. In contrast to physical deposition methods, it does not require a vacuum environment. 163 Some types of alloys can be hard to prepare by thermal methods but are easier to obtain by electrodeposition, such as an alloy composed of both lowmelting volatile constituents and high-melting metals. 166 Electrodeposited alloys have enhanced properties and possess special properties. 166 However, there are limited reports on the application of these methods to prepare Pd alloys for ORR application.
Betancourt et al. reported a strategy to prepare PdAg alloys for the ORR using a UPD method. 80 They have used underpotentially deposited Cu and subsequent galvanic displacement to deposit atomically dispersed Pd to achieve precise tuning of the electronic properties. The Pd@Ag/C catalysts had a higher normalized Pd mass activity with a value of 878 mA mg À1 in 0.1 M NaOH at À0.1 V vs. Hg|HgO. In addition, Gobal et al. synthesized PdCu alloys by an electrodeposition method. 167 6.2.3 Galvanic replacement reaction. The galvanic replacement method provides a one-step and universal route to prepare hollow nanostructures of noble metals (Pd, Pt, and Au) on a large scale, and the morphology, void space, and shell thickness of these hollow structures can be controlled by solid templates. 168,169 Xu et al. prepared nanotubular mesoporous PdCu catalysts through a galvanic replacement reaction using dealloyed nanoporous Cu as both the template and reductant. 170 This material exhibited a half-wave potential of 0.840 V for the ORR, superior to the commercial Pt/C (0.825 V) in 0.1 M HClO 4 solution. PdFe@PdPt/C was also prepared by the galvanic replacement reaction between PdFe/C alloy nanoparticles and PtCl 4 2À . 127 6.2.4 Pulsed electrodeposition (PED). Pulsed electrodeposition (PED) is an advanced type of electrodeposition. By controlling the interface supply and electrochemical reaction, the deposition performance can be well-controlled, 171 which allows accessing a non-equilibrium synthetic environment, thereby promoting the synthesis of metastable ordered intermetallic nanoparticles.
Wang et al. presented a pulsed electrodeposition process to prepare PdBi alloy electrocatalysts for the ORR. 172 Pd 31 Bi 12 ordered alloys (Fig. 9(e)) were grown directly onto carbon supports by deposition from an aqueous electrolyte containing ethylenediammenetetracetic acid (EDTA), Bi(C 2 H 3 O 2 ) 3

Dealloying
Nanoporous materials obtained by the dealloying method have been proved to represent a particular class of multifunctional catalytic nanoarchitectures with interconnected nanoscale skeletons and voids, 174 which are especially beneficial for electron and mass transport during electrocatalytic processes. The advantages of dealloying are simple and high-yielding preparation methods without any assistance from organic solvents. 6.3.1 Chemical dealloying. The chemical dealloying method generally uses chemical reagents (such as acids and alkalis) to obtain materials with different structures and improved properties.
Zhang et al. showed that nanoporous (NP) PdCu alloys (as presented in Fig. 9(f and g)) synthesized by dealloying from ternary PdCuAl source alloys exhibited superior ORR activity and higher catalytic durability compared with the Pt/C catalyst. 82 PdCuAl alloy foils were made by refining pure Pd, Cu, and Al at high temperature in an arc furnace, followed by melt-spinning under an Ar atmosphere. The nanoporous metals were prepared by dealloying the alloy foils in 0.  173 as exhibited in Fig. 9(h and i). NP-PdNi showed higher specific activity with a value of about 0.21 mA cm À2 at 0.9 V and the mass activity of NP-PdNi is about 0.15 A mg Pd

À1
. PdTi alloys 175 and PdCe 176 were also prepared by a similar method. In Begum et al.'s study, Pd nanonetworks (Pd-Net) without any support material were prepared by using 0.1 M HCl etching of PdZn alloys. 177 The introduction of CTAB plays a crucial role in the structure and morphology of the catalysts. Pt-Net had a higher ECSA and ORR performance than those of commercial Pt/C and homemade Pd nanoparticles in alkaline media.
6.3.2 Electrochemical dealloying. The difference in chemical potential between the elements in alloys can lead to selective etching of the more active components and it has been regarded as a productive and controllable route to fabricate nanoporous structures. 178,179 Through electrochemical dealloying, the structures and compositions of PdM alloys often change, which sometimes can greatly enhance the catalytic activity. However, the electrochemical dealloying method is not favorable for the control of the bimetallic ratio, which is easily affected by the applied potential, corrosion time, etc.
Sun et al. reported a process for converting colloidally synthesized ordered intermetallic PdBi 2 to ordered intermetallic Pd 3 Bi nanoparticles under ambient conditions by electrochemical dealloying to significantly enhance the ORR activity, 180 as shown in Fig. 10(a and b) 129 The dealloyed Pd-M nanoparticles had a core-shell structure with a Pd 3 M-core and a Pd-shell.

Other methods
In addition to the above synthesis methods, other methods can also be used for the preparation of PdM alloys. These methods are not widely used in the synthesis of Pd-based alloys, although they are extensively applied in the preparation of other materials, such as carbon materials, other metal nanoparticles (Au 181  Laser-irradiationinduced melting is an effective method to prepare bimetallic alloys with tunable size distributions. 183 Brandiele et al. reported a process to synthesize Pd 3 Y alloy nanoparticles by a robust laser assisted method in pure organic solvents. 85 Laser ablation of the bulk Pd 3 Y target was performed with the set up sketched in Fig. 10(c); ethanol, acetone and n-hexane were chosen as the solvents and the obtained PdY alloys had different morphological structures and ORR activities. From TEM images ( Fig. 10(d-f)), the large mass of the samples is composed of spherical nanoparticles (NPs) and, in the PdY-A and PdY-H samples, a core-shell structure. Among the three types of PdY catalysts, PdY-E displayed excellent ORR performance in 0.1 M H 2 SO 4 solutions, with an SA of 0.575 mA cm À2 and MA of 146 A g À1 .
6.4.2 Sputtering method. The sputtering method has many advantages such as process constancy, dependability, plasma production at low temperature, high quality and uniform products and industrial practicality with desired chemical composition. 185 Lee et al. synthesized PdM alloys (M = Co, Ni, Cr) by an rf sputtering method. 186 Sputter deposition was performed where 1 cm 2 Pd sheets were put on each transition metal target under an argon atmosphere. It is noted that the order of the ORR activity of various PdM alloys is different in the presence or absence of methanol. Specifically, in the absence of methanol, the order was Pd-Cr 4 Pd-Co 4 Pd-Ni 4 Pd, while, in the presence of methanol, the order was Pd-Ni 4 Pd-Co 4 Pd-Cr 4 Pd. Also, PdCu alloys were prepared by using magnetron sputtering equipment with a Pd and a Cu wafer as targets. 95 6.4.3 Pyrolysis synthesis. Pyrolysis is a rare way to prepare Pd alloys. It is relatively simple to operate without the addition of surfactants, additives or capping agents. However, it seems difficult to precisely control the morphology and particle size of synthetic materials. Furthermore, it needs high temperatures and the energy consumption is relatively high. PdFe alloys were prepared by thermochemical synthesis. 187 It included the adsorption of solutions of iron and palladium precursors on a carbon support, evaporation of solvents and subsequent pyrolysis. It is noted that the presence of basic metal particles (in this case, iron) can depress the agglomeration degree of palladium.
6.4.4 Ultrasonic synthesis. Ultrasonic synthesis offers a simple route to prepare materials by introducing energy into a chemical system through acoustic cavitation from the formation, growth and implosive collapse of bubbles within a liquid. 188 Distinctive properties of noble metal nanoparticles by ultrasonic synthesis have been achieved with a narrower size distribution, larger surface area, and much smaller size. 189 Karuppasamy et al. reported the preparation of reduced graphene oxide (rGO) nanosheet-supported PdAg nanoparticles by using a simultaneous ultrasonic probe irradiation method. 189 The mass activity of the PdAg (1 : 1)/rGO alloy was 1.01 mA mg Pd À1 , and the specific activity was 806 mA cm Pd
In short, various synthetic methods of PdM alloys are summarized in Fig. 11. Among the synthetic methods of Pd-based alloys, the most commonly used method is chemical reduction, while electrochemical methods are regarded as simple and environmentally friendly routes to prepare nanomaterials with a tunable size distribution and chemical composition and desired morphology. Moreover, the number of other methods has gradually increased, and various new methods have also appeared to be used in the synthesis of Pd-based alloys with excellent ORR performance. In addition, during the synthesis of Pd-based alloys, the following concerns need to be noted.
(1) The effect of heat treatment. (a) Change the structure of alloys. In Yang et al.'s work, the disordered type is changed to the ordered type via heat treatment, which increases the ORR activity. 88 Also, Maiti et al. mentioned that, through hightemperature treatment at B500 1C in an Ar/H 2 atmosphere, phase transformation of face-centered cubic (fcc) to facecentered tetragonal (fct) PdFe alloys was achieved, which is due to the movement of core Pd atoms to the surface of the alloy as Pd has higher adsorption enthalpy of H than Fe, resulting in the formation of fct-PdFe@Pd structures. 131 (b) For core-shell structures, heat treatment at a low temperature can facilitate core-shell structure formation, which plays a key role in enhancing the ORR activity of PdCo alloy catalysts. 190 (c) For PdFe and PdCu alloys, annealing at elevated temperatures enabled both Pd enrichment on the surface and an increase in the ECSA, thus enhancing the ORR performance. 128 (d) The degree of alloying can be increased to a certain extent, which is conducive to the improvement of the catalytic activity, but the post-treatment temperature should not be too high, which may lead to an increase in particle size and a consequent decrease in the electrochemically active surface area. 78,83,135,191 But there are some exceptions; in the case of the Pd 60 V 40 sample, heat treatment at 300 1C decreased the ORR performance of the as-prepared sample. 136 (e) Remove the organic reagents (such as surfactants, capping agents, etc.). 192 Therefore, these issues are clearly complicated and further experimental and theoretical work is required to fully understand the effect of heat treatment on the electrocatalytic activity.
(2) The effect of the CO reagent. Through the comparison of Fig. 12, it is noted that a CO reagent is used in the synthesis process of many Pd-based alloy catalysts with extremely high catalytic activity, such as PdMo bimetallene 75 and Pd 4 Fe nanoflowers. 155 Besides, Ahmad et al. used W(CO) 6 to synthesize W-doped Pd nanocubes. 193 At 0.9 V RHE , the specific activity of 1.2% W-doped Pd nanocubes/C was 1.18 mA cm À2 , which surpassed most reported Pd-based catalysts for the ORR in alkaline media.

View Article Online
Some researchers have specifically studied the role of the CO reagent. For example, Wei et al. reported the effect of CO treatment for PdCoAu catalysts. 194 The authors found that the CO treatment can modify the PdAu and PdCo alloy ratios in the bulk structure and result in the formation of Pd-Au alloy enriched on the outer shell, thereby increasing the ORR performance. The PdCu/C catalysts heat-treated in CO have a higher alloying degree and electrochemical activity towards the ethanol oxidation reaction compared with those treated under O 2 and H 2 . 195 The CO-confinement effect, which is the strong binding of CO on the metal (111) facets, is used to induce the formation of 2D nanomaterials. 196 Therefore, more in-depth research in this aspect needs to be done.
(3) The effect of solvents. In chemical reactions, the role of the solvents is often less noticeable, but its impact cannot be ignored. The solvent will affect the morphology and structure of Pd-based catalysts. 159 Zheng et al. found that, during the synthesis of both Pt and Pd nanoparticles using H 2 as the reducing agent at room temperature, ethylene glycol (rather than the commonly used water) used as the media can minimize the aggregation of particles during their formation due to the high viscosity, thereby controlling the particle size. 130 Also, it is found that the solvent played significant roles in controlling the synthesis of intermetallic Pd 3 Pb square nanoplates. 154 (4) The effect of doping (a) To improve the selectivity. Pd-based alloys also have high catalytic activity for the methanol/ethanol oxidation reaction. To avoid the poisoning problem induced by the alcohol crossover effect, achieving the selectivity of Pd-based alloys towards the ORR is highly desired. The surface chemical functionalization of metal nanocrystals was used in Pd-based alloys to enhance the ORR selectivity, such as polyallylamine (PAA)functionalized PdCo alloy nanonetworks. 138 (b) To improve the methanol tolerance. For instance, the essential role of Rh in obtaining high methanol tolerance capacity was reported in Rh-doped PdAg nanoparticles. 192 (c) To improve the ORR activity. The surface decoration induced structural change could directly influence the electronic structure, thereby improving the catalytic activity, such as for Pt submonolayer decorated PdAu/C nanocatalysts. 137 Besides, polyethyleneimine (PEI) as a dopant was reported. 139 Specifically, -NH 2 groups on the modified PEI polymer can been protonated into -NH 3 + in an acidic medium, which will increase the interfacial H + concentration on the Pd surface and consequently accelerate proton-coupled electrocatalytic reactions (such as the ORR and HER). 197 (5) The effect of reductants. Calderon et al. studied the effect of five different reductants (methanol, ethanol, n-propanol, formaldehyde and ascorbic acid) on the ORR activity of PdPt catalysts. 198 Different reductants can tune the size and surface Pt content, thus influencing the properties of the catalysts. In addition to several common reducing agents, the introduction of other new reducing agents may bring surprises, such as pine needle extract as the reducing agent, 160 1-naphthol ethanol solution, 153 urea slowly degraded to NH 3 during pyrolysis 199 and a borane tert-butylamine complex. 200 (6) The effect of additives. The introduction of some additives can affect the morphology and properties of the catalyst, such as EDTA, 158,177 arginine, 201 glycine, 202 NaI, 159 amines, 196 etc.
The effect of carriers. In this review, we did not conduct a detailed analysis of the role of the carriers. But a suitable carrier will greatly enhance the catalytic performance of Pd-based alloy catalysts, such as ordered mesoporous carbon (OMC), 143 defective mesoporous carbon, 203 nitrogen-doped graphene, 199 a CoFe 2 O 4 -Vulcan composite, 200 carbide-derived carbons (TiC, Mo 2 C and SiC), 204 etc. The issue that carbon-support corrosion leads to a decrease in the stability of catalysts is worthy of further investigation. 205 (8) Other considerations. For Pd-based alloys prepared with different metals M or different synthesis methods, the ratio of Pd to M at the optimal activity is uncertain. For example, the View Article Online maximum activity for the ORR was observed at an alloy composition of B60 at% Pd for three types of PdM alloys (M = Co, Ni, Cr). 186 Meanwhile, in Liu et al.'s investigation, Pd 70 Co 30 /C alloys had the highest catalytic activity. 206 As for PdCu alloys, a molar ratio of approximately 1 : 1 showed the optimal alloy composition. 95,126 Using a combination of multiple synthesis methods to synthesize highly active ORR catalysts is a method to control the structures of the targeted materials. For instance, in You et al.'s work, three catalysts with PdNi, PdNiCu and PdCu cores and a PdIr shell were fabricated by a polyol method and galvanic replacement. 207 Chen et al. reported a process to prepare nanoporous PdCe (NP-PdCe) nanocubes by melt spinning combined with chemical dealloying. 176 Carbon supported PdCu alloyed catalysts were prepared by using a twostep process involving the synthesis of Cu nanoparticles, followed by galvanic substitution of Pd on Cu nanoparticles. 208 Focusing on efficient bifunctional catalysts, for example, PdIr alloys 144 as efficient bifunctional catalysts for oxygen reduction and oxygen evolution reactions, tree-like PdAg nanocrystals towards anodic formic acid oxidation and cathodic oxygen reduction 153 were reported. Additionally, Pd alloyed with novel elements is worthy of attention, such as PdTe, 209 PdCe, 176 etc.
Finding suitable ORR catalysts used in proton exchange membrane fuel cells (PEMFC) is a key concern, and ideal catalysts should meet the requirements of good reliability, durability and stability in acidic electrolytes. 210 One of the main issues for Pd-based catalysts is Pd dissolution under the operating conditions including low pH, high temperature, high potential and frequent start-stop cycling. 10 Compared with Pt, Pd is much more easily dissolved.
In acid solution, the main pathways for Pd dissolution include direct dissolution (eqn (20)) or oxide film formation (eqn (21)) and a subsequent chemical reaction (eqn (22)). 211 Pd -Pd 2+ + 2e À Pd + H 2 O -PdO + 2H + +2e À PdO + 2H + -Pd 2+ + H 2 O In the direct dissolution pathway, the equilibrium potential of Pd dissolution is 0.2 V lower than that of Pt. In the chemical dissolution pathway, the Pd 2+ equilibrium concentration is B5 orders of magnitude higher than that of Pt 2+ . Thus, Pd is much less stable than Pt in an acid medium. 212 It is noted that Pd/C was not regarded as an ideal catalyst for the ORR in acid solution due to the gradual degradation during potential cycling. 213 However, some Pd-based alloys which had good stability in acid media were reported. 148,173 In addition, Ou et al. compared the ORR mechanisms on the Pd(111) and Pt (111) surfaces in acidic solution by DFT, 55 and revealed that the adsorption and dissociation processes of O 2 molecules more easily occurred on the Pt (111) surface and that the serial protonation of the dissociative product to form H 2 O molecules could also occur more easily on the Pt(111) surface than on the Pd(111) surface, indicating that Pt can serve as a better ORR electrocatalyst than Pd.
In general, Pd-based catalysts are still not comparable with Pt-based ones in terms of ORR activity and stability in acidic electrolytes. More efforts need to be made to achieve new breakthroughs.

Electrochemical analysis
Electrochemical analysis determines the specific performance of electrocatalysts, which is particularly important for Pd-based electrocatalyst optimization. Accordingly, typical methods of electrochemical analysis, including determination of the electron transfer number (n) and electrochemically active surface area (ECSA), durability evaluation protocols and standardization of electrochemical analysis processes are concluded in this section, as well as the performance of representative Pd-based ORR electrocatalysts.

Electron transfer number (n)
The electron transfer number per O 2 molecule (n) is one of the key parameters for the ORR, which can provide information on both the oxygen conversion efficiency and the mechanisms, thus helping to evaluate the performance of electrocatalysts. The rotating ring-disk electrode (RRDE) method 214 88 alloys. These studies showed that the n values are near to four, indicating excellent performance with limited amounts of H 2 O 2 generated in the ORR process for these Pd-based electrocatalysts. However, according to recent studies, the n values obtained by the RRDE (n RRDE ) and KL (n KL ) methods have discrepancies, 42,224,228,229 and the n values from the RRDE method are believed to be more accurate. The KL plots are often not linear, and the n KL values sometimes exceed theoretical limits. Masa et al. proved that the KL plots are affected by the coverage of electrocatalysts on the working electrode and demonstrated that the surface area ratio (total electroactive surface area to geometric area of the rotating disc electrode (RDE) surface) should be taken into account when inferring electrocatalytic effects on the basis of KL analysis of RDE data. 230 Zhou et al. pointed out that, from a theoretical viewpoint, the ORR is neither a single-step nor a one-way reaction, and therefore does not fulfil the assumptions from the KL method. 231  with the RRDE method were also presented in detail by Zhou et al.'s study. 231 For example, the collection efficiency (N C ) of the RRDE decreases significantly with the catalyst loading. Moreover, when the electrode surface is rough, N C also decreases dramatically with o. In addition, the widely applied RRDE method with a Pt ring is not suitable for H 2 O 2 collection in alkaline electrolytes because the oxidation of H 2 O 2 on Pt is not a masstransfer-limited process. A properly biased Au ring rather than a Pt ring is more suitable. As a result, the RRDE method with a properly biased Au ring is recommended to determine n values for the ORR in alkaline electrolytes, supplemented by the calibration of the collection efficiency. Therefore, as for Pd-based electrocatalysts, the best possible way is that the RRDE with the relevant calibration is the major approach, and the KL method is an auxiliary verification.

Electrochemically active surface area (ECSA)
The ECSA of ORR electrocatalysts is a key parameter to evaluate mass activities and specific activities. It can be determined by CO-stripping (oxidation of pre-adsorbed CO monolayers) and the H upd method (integrating the underpotentially deposited hydrogen (H upd ) regions from cyclic voltammetry). 232 In previous work, the ECSAs of PdZn 88 and PdAu 157 alloys were estimated by CO-stripping measurements and the ECSAs of PdFe 155 and PdPt 161 alloys were calculated by the H upd method, respectively. It is noted that, unlike Pt, Pd could absorb a certain amount of H 2 , resulting in the fact that it is impossible to quantitatively differentiate between H upd and absorbed hydrogen (H abs ). Therefore, it is hard to obtain the exact ECSA by the H upd method. Recent studies reported that the ECSA of a Pd-based catalyst is measured by Pd oxide reduction analysis. 233,234 The Coulomb charge of Pd-based catalysts is calculated by using the chemically adsorbed oxygen reduction peak instead of the hydrogen adsorption/desorption peak in Pt-based catalysts. The ECSAs of PdCo, 224,235 PdPb, 154 and PdAg 233 alloys have also been measured by this approach. In addition, the use of underpotentially deposited copper (Cu UPD ) is feasible; 236 however, the major disadvantage of this method is that the metal ions might largely affect catalytic reactions. The ECSAs of PdFe alloys 12 were determined by this method.
Shao et al. studied the ECSA measurements of Pt-and Pdbased nanoparticles in detail and demonstrated that the ECSAs follow H UPD o CO stripping o Cu UPD and Cu UPD is the most accurate method. 237 In Luo et al.'s work, the ECSAs of PdMo catalysts were determined by H upd , CO stripping, Cu upd and Pd oxide reduction, and the ECSAs determined from the Cu upd method were used for further analysis. 75 Some research pointed out that one of the most suitable techniques to determine ECSAs is differential electrochemical mass spectrometry (DEMS), 238 which allows in situ measurements of supported catalysts under controlled mass flow conditions, but this technique is not available in most labs.

Durability test
Durability is regarded as a key factor for ORR electrocatalysts in practical applications. Durability tests in different studies of Pd-based electrocatalysts varied greatly, and even a small proportion of studies did not provide specific durability performances. Standard durability test protocols are summarized.
7.3.1 Start/stop cycle and load cycle. The Fuel Cell Commercialization Conference of Japan (FCCJ) proposed several methodologies for evaluating the durability of the membrane electrode assembly (MEA) such as catalysts and membranes in 2007. 239 The proposed potential cycle tests consist of two protocols; the start/stop durability test and the load cycle durability test were revised in 2011. 240 Table 3 shows the durability test conditions, protocols, and diagnostic tests for electrocatalysts using a half-cell (RDE) and MEA, respectively. Many studies were reported using the load cycle protocol to evaluate the durability of electrocatalysts. [241][242][243] 7.3.2 Accelerated stress test (AST). The US Department of Energy (DOE) proposed an accelerated stress test protocol in 2007 with the aim of determining the durability and performance of current fuel cell components without the need to test over many years. 205 The associated DOE accelerated stress test protocols and performance metrics are presented in Table 4.

Standardization of electrochemical analysis
Cyclic voltammograms (CV) and linear sweep voltammograms (LSV) on RDEs are extensively collected to evaluate the ORR performance of Pd-based catalysts. These methods allow studying the material in a large potential and time-scale domain, presenting information on the thermodynamic and kinetic behaviors of the electrocatalysts. However, due to the differences in test methods and conditions, it is hard to compare the ORR performances of electrocatalysts using MA or SA values. To better compare the ORR activities of different electrocatalysts, more standardized and reasonable guidance needs to be proposed.
Mayrhofer et al. scrutinized the thin-film rotating disc electrode (TF-RDE) method for investigating the activities of high surface area electrocatalysts. 244 They pointed out: (a) Normalized activities have to be determined. The normalization can be based on the noble metal loading (mass activity) or surface area (specific activity). (b) The active surface area has to be evaluated by applying a proper method that also considers the capacity of the support. (c) The electrode thickness, the diffusion limited current of the ORR and the current under the specific potential need to be limited.
Garsany et al. studied the impact of film drying procedures on RDE characterization and found that films prepared with the rotational drying method are more reproducible than films prepared with the stationary method. 245 Takahashi et al. emphasized the importance of ink preparation for the RDE method. 246 In different acidic electrolytes, the ORR activity varies. The catalytic activity of Pt/C followed the trend HClO 4 4 H 2 SO 4 4 H 3 PO 4 and the durability of Pt/C in H 2 SO 4 was found to be lower than in HClO 4 . 246 Pd is less studied in this aspect, but it can be predicted that Pd may exhibit similar trends to Pt. Therefore, it is necessary to unify the types and concentrations of the electrolyte during the test.
Moreover, the ultimate estimation of the electrocatalyst catalytic activity towards the ORR should be performed in MEAs. To a certain  75 Hence, the MEA test is strongly recommended for Pd-based electrocatalysts in future work.

Summary of the electrocatalytic performance of representative Pd-based catalysts
As Tables 5 and 6 and Fig. 12 show, regarding the ORR activity, Pd-based alloy materials had better performance in alkaline electrolytes than in acidic electrolytes, and some of them, such as PdMo bimetallene and Pd 4 Fe nanoflower materials, even exceeded that of Pt-based catalysts. They exhibited excellent ORR activity in alkaline solutions and also had good activity in acidic solutions. However, in acid solution, the stability of Pd alloy catalysts was often not as good as that of Pt-based materials. Indeed, Pd is more susceptible to oxidation at more negative potentials than Pt and is more easily poisoned by anions such as ClO 4 À and other oxygen-containing species as compared to Pt. 10 Pd-based catalysts with different element   compositions and morphologies and structures had obvious activity differences. Although previous work has studied the relationship between morphology and activity, there is still no exact direct relationship between material composition, structure and performance. Methods to obtain the expected structure and electrocatalytic performance through experimental preparation are still a challenge. Other factors, including the electrochemically active surface area (ECSA), half wave potential (E 1/2 ), Tafel slope, onset potential (E onset ), etc., are also factors that affect the performance of electrocatalysts and could not be ignored. Many catalysts reach high mass activities (MAs) by increasing the specific activity (SA) at the expense of a reduced ECSA. In the potential region of kinetic control, low ECSA is counterbalanced by high specific activities. At higher overpotentials (usually used in actual systems), the reaction rate is not limited by kinetics, but by mass transfer. 248 Therefore, it is necessary to obtain materials which combine a high mass activity with a high ECSA. In addition, E 1/2 is another common parameter that allows comparing the ORR activity of catalysts, especially between platinum group metal (PGM) catalysts and PGM-free catalysts. A positive shift in E 1/2 indicates a high ORR activity. 249 However, there is an issue that E 1/2 increases with RDE catalyst loading. Thus, comparisons between catalysts by E 1/2 are biased towards studies that used higher loading, and the loading dependence complicates the comparison between studies of different loadings. 250 The Tafel slope is derived from the microkinetic model, and thus represents the reaction mechanism and surface kinetics. 251 Different electrocatalysts show distinct rate-determining steps, such as the dissociation of O 2 or desorption of OH. 251 In general, the smaller the Tafel slope, the faster the ORR kinetics. In the case of cathode Pd-based catalysts, the decrease in onset potential and the increase in current density detected during the ORR process are considered to be key parameters for evaluating the catalyst performance. 252 How to achieve it is the goal that researchers are striving for.

Conclusions and perspectives
Recent studies on Pd-based alloy catalysts for the oxygen reduction reaction are summarized in this review. There is a trend of rapid development and electrocatalysts are being developed that show excellent activities for the ORR, which often even exceed the performance of Pt-based catalysts, especially in alkaline media and in the presence of methanol. We note that theoretical research often only focuses on one or a few factors, while actual experiments are very complicated. Therefore, the development of related theories and the combination of theory with experiments can help us to predict development trends and prepare high-performance materials more efficiently. The alloys formed from palladium and various elements may have good catalytic activity. In addition, Pd alloys with different morphologies and structures can be obtained by different synthesis methods, and their catalytic performance is quite different. Regarding the mass activity and specific activity towards the ORR, PdMo bimetallene, PdBi alloys and Pd 4 Fe nanoflower materials seem to be the most promising cathode electrocatalysts; however, other parameters, like stability, and catalyst activity under actual industrial conditions rather than laboratory conditions, are more important factors in determining whether commercialization can be achieved. In addition, some other issues for mass production need more attention.
(1) In terms of reaction thermodynamics and kinetics, more in-depth research is needed to pave the way for future industrialization. The establishment of universal reaction kinetics and thermodynamic models could be achieved through more advanced machine learning techniques for building up correlations to produce high-performance Pd alloy catalysts.
(2) Regarding catalyst deactivation, studies need to focus on deactivation mechanisms, especially for precious metal catalysts. The study of the deactivation mechanism can better understand the catalytic process, thereby helping to extend the service life and understand how to regenerate the catalyst.
(3) As for recycling, it needs to be determined if there is a facile way to regenerate deactivated catalysts. If so, how can this be achieved with efficient circulation and still maintaining high activity and stability. If not, whether the abandoned catalysts can be processed for other applications needs to be studied.
(4) At present, little research has been reported on the life cycle assessment (LCA) of ORR catalysts (not just the Pd or Pt series), so it is difficult to evaluate the environmental impact of these electrocatalysts during the production process. For example, the production of a certain catalyst using low-cost raw materials under harsher reaction conditions is a dilemma. Through LCA, the advantages and disadvantages can be obtained, thus deciding whether this catalyst is worth producing.
In short, the research on Pd-based electrocatalysts towards the ORR is moving in the direction of low cost, high activity, stable performance and long lifespan. Researchers need to analyze and solve the above problems to promote the industrialization of this material set, as demonstrated in Fig. 13.

Conflicts of interest
There are no conflicts to declare.
Appendix A. Supplementary data 1. Methodology

Data collection
The data was obtained from the Web of Science Core Collection (WoSCC) using the advanced search ''TS = (''Palladium alloy*'' or ''Pd* alloy*'' or ''Palladium catalysts'' or ''Pd* catalysts'') AND TS= (''oxygen reduction reaction'' or ''ORR'')''. English was the only chosen language, and the type of documents was restricted to the article. The timespan was from 1970 to 2020. For citation indexes, the data was only selected from Science Citation Index Expanded (SCI-EXPANDED) -1970-present. As a result, 400 documents were identified on 6 July 2020.

Scientometric analysis methods
1.2.1 Biblioshiny. Biblioshiny, which provides the analysis, presentation and manipulation of data, is a user-friendly interface for bibliometrix. 26 The collected data was extracted in plain text format from WoS. Then RStudio version 1.3.959 was used to export the interface of biblioshiny by bibliometrix under R version 4.0.1 and the data was uploaded to the biblioshiny package. In this work, the extracted data was analysed in terms of annual production, top 10 sources and top 10 countries by production via biblioshiny.
1.2.2 CiteSpace. CiteSpace is free software based on the Java environment developed by C. Chen for detecting and visualizing trends and patterns in the literature. 27 The data were analysed through CiteSpace version 5.6.5R to visualize the network. This work concentrated on the period from 1997 to 2020 and used 1 year slices.
1.2.3 VOSviewer. VOSviewer is a freely available computer program for constructing and viewing bibliometric maps, which is especially focused on displaying large bibliometric maps in an easy-to-interpret way. 29 The data was analysed via VOSviewer version 1.6.15 to obtain the network visualization and density visualization map based on co-keywords. Fig. 14(a) describes the top 10 sources of these 400 articles. Amongst them, Electrochimica Acta was the most productive source with 37 articles, which is intended for studies in the field of electrochemistry. In terms of average citations, Journal of the Electrochemical Society (J ELECTROCHEM SOC) was the highest journal, because three highly cited articles (Lee K, 2006; 35 (189), which was almost four times that of the USA (ranking in second place). But the average citation number was less than that from the USA. Although the number of publications is not high, Japan had the largest average citations. The number of MCP articles to the total number of articles (so-called MCP ratio) could reflect the extent of inter-country collaborations to a certain degree. Estonia and Japan had higher MCP ratios, while Mexico and India had lower ones. Table 7 displays the top 10 references with the strongest citation bursts sorted by strength. The centrality value indicates the importance of the position in the network. High values of centrality identify potentially revolutionary publications. 256 Sigma is a hybrid metric, used to measure the structural centrality and citation burstness of a cited reference, suggesting the scientific novelty of a document. 257 From Table 7, 8 in 10 are from cluster #0 (Pd nano-particle).

View Article Online
It opened the curtain for the follow-up study of Pd alloy catalysts. In addition, in Gasteiger's article, 255 the authors quantified the activities and voltage loss modes for MEAs (membrane electrode assemblies), and provided benchmark oxygen reduction activities for electrocatalysts by using two different testing procedures to establish the relative merit of candidate catalysts. Its centrality was the largest (0.04) and the Sigma value was the second largest (1.40). Besides, it is noted that Shao MH was the first author of three references in Table 7, suggesting that he/she has made outstanding contributions in this field.