High-Performance Nanostructured Palladium-Based Hydrogen Sensors—Current Limitations and Strategies for Their Mitigation

Hydrogen gas is rapidly approaching a global breakthrough as a carbon-free energy vector. In such a hydrogen economy, safety sensors for hydrogen leak detection will be an indispensable element along the entire value chain, from the site of hydrogen production to the point of consumption, due to the high flammability of hydrogen–air mixtures. To stimulate and guide the development of such sensors, industrial and governmental stakeholders have defined sets of strict performance targets, which are yet to be entirely fulfilled. In this Perspective, we summarize recent efforts and discuss research strategies for the development of hydrogen sensors that aim at meeting the set performance goals. In the first part, we describe the state-of-the-art for fast and selective hydrogen sensors at the research level, and we identify nanostructured Pd transducer materials as the common denominator in the best performing solutions. As a consequence, in the second part, we introduce the fundamentals of the Pd–hydrogen interaction to lay the foundation for a detailed discussion of key strategies and Pd-based material design rules necessary for the development of next generation high-performance nanostructured Pd-based hydrogen sensors that are on par with even the most stringent and challenging performance targets.

T o achieve the necessary dramatic reduction of greenhouse gas emissions, alternative energy vectors that replace fossil fuels are imperative. In this respect, hydrogen gas, H 2 , has been identified as particularly attractive since it can be used to generate electricity with water as the only byproduct. 1 Therefore, large investments in a hydrogen economy are imminent all over the globe, as, for example, in Europe, as part of the Green Deal. 2,3 Until very recently, one specific technological challenge related to a widespread use and largescale distribution of hydrogen gas had received little attention in the public debate, but came to broad attention due to a recent accident at a Norwegian H 2 fuel station 4 hydrogen safety. This event was a dramatic reminder of the high flammability of H 2 − air mixtures at H 2 concentrations above 4%, and thus made the importance of robust and fast hydrogen safety sensors for leak detection highly apparent. However, to date, no hydrogen sensor technology exists that can meet all the hydrogen safety sensor performance targets set by, for example, the US Department of Energy (DoE), 5 despite at least a decade of research (Scheme 1). 6−15 In this Perspective, we therefore first critically assess the current state-of-the-art of high-performance hydrogen sensors across all signal transducer platforms. With high performance, here we mean sensors that have been developed with the aim and/or potential to meet the US DoE hydrogen sensor performance targets. This approach thus sets apart our focus of discussion from other more conventional review articles on the topic. 10−16 In the second part of this work, based on the literature survey, we conclude and propose that nanostructured materials based on palladium (Pd) and its alloys are the signal transducer materials known today with the best potential to eventually meet all of the US DoE performance targets for hydrogen safety sensors. To reach this conclusion, we thoroughly discuss the fundamentals of Pd−H interactions to identify the fundamental material properties that intrinsically limit sensor performance. Based on this understanding, we then derive rational material design rules and summarize a selection of existing research efforts that already utilize some of those rules. Last, we propose and discuss future research directions, mainly focusing on the likely most challenging US DoE targets: (i) sensor lifetime, (ii) operation temperature, (iii) absolute operation pressure, (iv) operation in poisoning/deactivating conditions, and (v) at high humidity, in order to hopefully inspire rapid development efforts in these directions.

RESPECT TO THE US DOE TARGETS
Response Time. The sensor response time target probably constitutes one of the most challenging targets to meet, but also one of the most obvious ones to rationalize why it is important from a safety perspective. 17 Thus, response time has been a key topic in many studies, and we have found 553 relevant reports from 1999 to 2020 that claim "fast hydrogen sensors". 18 In terms of operating conditions, they range from room temperature (RT) up to 500°C. The highest temperatures are predominantly reported for oxide-based sensors, where they are needed to enable efficient ionic transport. This wide range of operation temperatures, however, makes direct comparison of different types of sensors somewhat difficult since, at least for the same type of active material, higher operating temperatures will lead to faster response times, as a consequence of the Arrhenius law. Therefore, to allow relevant comparison, and to keep our focus on high-performance sensors that operate at the toughest conditions, we have opted to only include sensors that operate at (or close to) RT in our survey. In this way, we are also directly addressing the power consumption target, which intrinsically is harder to meet for a sensor that needs to be maintained at high operating temperature. Furthermore, to enable direct comparison between different sensors, we use t 90 , that is, the time to reach 90% of the sensor response for the new steady state after a stimulus, as descriptor for response time. Finally, wherever possible, we have used the response corresponding to a hydrogen exposure to 0.1 vol.% (∼1000 ppm ≈ 1 mbar), i.e., the lower hydrogen detection limit in the US DoE performance target. This is important not only from the perspective of identifying this number as the ultimate goal, but also since many experiments 19−23 suggest that the response time depends on both the absolute hydrogen concentration and the amplitude of pressure change to be detected. In other words, a hydrogen sensor generally will respond faster when exposed to a larger concentration change. Hence, this fact has to be taken into consideration when comparing the speed of different sensors, and thus we only list sensors measured at hydrogen concentrations equal to or less than 0.1 vol.% and those reporting maximum 5 s response time when exposed to a H 2 concentration higher than 0.1 vol.%. As summarized in Table 1, only around 10% of the surveyed works (i.e., 58 out of 553) when using the search string introduced above fall into our category, and we make the following observations.
The first key point to note is that in terms of readout principle, electrical sensors (resistance-based), which comprise two electrodes connected to a transducer element that changes resistivity upon interaction with hydrogen, are most abundant among the systems with fast response according to our definition. The typical transducer materials for this resistivebased electrical sensors are Pd, metal oxides (e.g., SnO 2 , TiO 2 , In 2 O 3 , ZnO, MoO 3 , and WO 3 ) or a hybrid of Pd and a metal oxide. This is no surprise, since this type of sensor is most mature and has a simple, yet effective, construction. Interestingly, however, from a commercial hydrogen sensor market perspective, they are not the most common ones available. 81 The second key observation is that in terms of active material, the majority of fast sensors employ Pd and its alloys in various forms. The third and maybe most striking finding is that all fast hydrogen sensors employ some sort of nanostructured transducer element(s). This development has been enabled by the parallel advances in nanoparticle synthesis and nanofabrication, and it was triggered by Favier et al.'s seminal work on ultrafast Pd nanowire array electrical hydrogen sensors from 2001, 40 the first sensor to achieve millisecond response time to 5 vol % H 2 ( Figure 1a).
However, if we take the US DoE's most stringent response time target (<1 s at 0.1 vol % H 2 ), even to date, still only a few works may have or do have reached this target at RT. For example, Lee et al.'s porous Pd@CPPy conducting polymer 24 and Zhang et al.'s SnO 2 @graphene 27 hydrogen sensors may have reached the 1 s target (Figure 1b). Specifically, Lee et al. and Zhang et al. recorded 4.5 and 2 s response times, respectively, at extremely low H 2 concentrations of 20 ppm (∼0.002 vol %) and 100 ppm (∼0.01 vol %) in air. Thus, as discussed above, if measured at a higher concentration of 0.1 vol %, these sensors may, in principle, respond faster than the 4.5 and 2 s reported at those low ppm and thus meet the US DoE target. However, no explicit measurement is presented by the authors. Similarly, there are some works that report response times close to the 1 s at the 0.1 vol % limit. For example, porous PdPt thin film (5 s at 1000 ppm), 44 Pd ultrathin films (0.07 s at 2 vol %), 34 and Pd nanowire arrays (0.075 s at 5 vol %). 40   deactivation by other molecular species is critical to consider when developing hydrogen sensors.
Cross-Sensitivity and Deactivation by Poisoning Gases and Humid Conditions. The presence of other molecular species than H 2 in the sensor environment may interfere with or completely disable the response of a hydrogen sensor via either cross-sensitivity or poisoning/deactivation. Cross-sensitivity here refers to how sensitive a sensor is toward unwanted stimulus by another species than the to-be-measured one and is thus related to selectivity. A perfectly selective sensor only responds to the species that is to be detected (H 2 in this case), while being completely inert toward other species. Practically, a sensor selectivity test is carried out by measuring sensor response upon exposure to pulses of different species. On the other hand, poisoning is sensor deactivation by one or multiple species that themselves do not induce a sensor signal but prevent H 2 detection due to, e.g., surface blockage. For certain systems, both phenomena can also occur at the same time. To this end, CO, NO x , and sulfuric compounds are known to poison Pd-based hydrogen sensors. 19,58,82,83 In the US DoE targets (Scheme 1), cross-sensitivity is a main factor to determine the sensor accuracy while poisoning/deactivation is to sensor lifetime, accuracy, and response time.
In Table 2, we summarize studies from the same pool of fast hydrogen sensors surveyed above, 18 which have investigated the effect of interfering/poisoning species. It is clear that many studies performed selectivity tests, but only a few investigated poisoning/deactivation. Again, we see Pd as the main transducer material, due to its inherent excellent selectivity toward aliphatic/alcohol hydrocarbon species, such as CO, CO 2 , C 2 H 5 OH, and CH 4 . Furthermore, we note that testing sensor selectivity is particularly important for oxide-based sensors, because unlike Pd, they do not have inherent selectivity toward hydrogen gas. Therefore, oxide sensors usually employ a Pd coating/capping to improve the selectivity, as, for instance, shown for Pd-capped SnO 2 nanorods 84 and TiO 2 nanotubes. 85 Therefore, understanding the limiting factors of Pd−hydrogen interactions is equally important for this class of sensors.
With regard to sensor poisoning/deactivation, we found that this aspect is much less addressed than selectivity, despite its high relevance for real applications. For example, in the case of Pd-based sensors, species like CO, NO x , sulfuric acid, and H 2 O strongly interfere with hydrogen detection due to strong adsorption on the Pd surface, where they prevent H 2 dissociation and further absorption. 164−167 CO adsorption, for example, leads to (much) slower response times. 19,40,41,58,82,168 This, in turn, can also cause incorrect sensor readings that underestimate hydrogen concentration if sensor saturation is not achieved within the period of exposure.
A specific shortcoming of the handful of studies that do investigate the effects of sensor deactivation/poisoning is that, except for the work by Hayashi et al., 96 none of the tests executed in the works presented in Table 2 follows the protocol suggested by ISO 26142, since all studies applied premixed H 2 and poisoning gases. The ISO 26142 protocol, however, suggests exposure to poisoning species prior to a H 2 pulse to test the poisoning effect, since this is closer to a scenario in a real setting. 169 Furthermore, although explicitly mentioned in the Table 1. continued a CNT = carbon nanotubes, CPPy = 3-carboxylate polypyrrole, GO = graphene oxide, MEMS = microelectromechanical systems, NB = nanobelts, NG = nanogaps, NP = nanoparticles, NR = nanoribbons, NS = nanosheets, NT = nanotubes, NW = nanowires, PANI = polyaniline, PMMA = poly(methyl methacrylate), POSS = polyhedral oligomeric silsesquioxanes, PPy = polypyrrole, PTFE = polytetrafluoroethylene, PUA = polyurethane acrylate, rGO = reduced graphene oxide. b LoD = limit of detection.   When it comes to the dynamic range, the US DoE demands hydrogen sensors to be able to detect hydrogen concentration changes from 4 vol % (the lowest H 2 gas flammability limit in air) down to 0.1 vol % (Scheme 1). Overall, this requirement is essentially reached by all the surveyed hydrogen sensors, unlike response time and gas selectivity/ poisoning discussed above. For instance, as shown in Table 1, all sensors exhibit detection limits well below the target of 0.1 vol %. As state-of-the-art, detection limits down to single-digit ppm, or even ppb, have been demonstrated. 19,24,27,32,33,170 These sensors are also capable of detecting hydrogen concentrations up to 4 vol %, although the response is usually not linear across this range of concentrations and rarely measured for both increasing and decreasing concentration. The former limits the concentration range across which the sensors exhibit high sensitivity (i.e., a significant change in sensor readout per H 2 concentration change), while the latter creates history-dependent sensor readout (i.e., hysteresis) and thus reduces accuracy within a certain pressure range. Both these aspects are very pronounced for pure Pd transducers, but strategies to alleviate these issues have been established, as we discuss in detail below.
Lastly, regarding sensor power consumption, the most stringent DoE target is 1 W (Scheme 1). A number of the fast Pd-based sensors listed in Table 1 exhibit power consumption as low as 1−100 nW 40,78,171 and none beyond 1 W. We note that all of them are electrical sensors. On the other hand, no works related to optical sensors report power consumption. However, examining available components (e.g., LEDs and photodiodes 172,173 ) reveals that similar low power consumption as for electrical sensors can be expected. 174,175 Hence, reaching the US DoE power consumption standard seems feasible, irrespective of readout principle and transducer material.
Lifetime, Accuracy, Operation Temperature, and Absolute Operation Pressure. Operational lifetime is by definition the expected useful life of a sensor under operating conditions, while accuracy is the relationship between the sensor readout and the actual H 2 concentration. 17 Assessments of both aspects are normally first carried out at a prototype or product level. 176 In the surveyed fast sensors, 18 these two aspects have therefore not been addressed explicitly. Nevertheless, some efforts have been reported related to improving sensor lifetime and accuracy. For instance, to improve lifetime, deactivationresistant sensors have been researched intensively, e.g., by alloying and polymer coatings. 19, 58 For the accuracy aspect, the main strategy has been the development of stable and deactivation-resistant transducer materials with low cross-sensitivity toward other analytes. To this end, sensor stability is usually examined by exposing the sensor to a large number of hydrogen cycles or by intermittent testing over a long period of time. 58,91,112 To the best of our knowledge, no specific recommendation for the minimal number of cycles exists, but we would recommend at least 50 cycles for such tests, ideally more.
In terms of operation temperature, for obvious reasons, fast sensors operating at high temperature (up to +85°C, according  179 an optical sensor based on carbon nanotubes (−120°C), 180 a Pt-doped WO 3 film (−50°C), and a tapered fiber optic solution using a Pd thin-film as signal transducer (−196°C). 181 All these works have in common very long response times on the order of minutes, with the exception of Bevenot et al., who utilized the local heating generated by the high-power laser diode used as light source in their Pd thin-film tapered fiber optic hydrogen sensor, which successfully enhanced the response time to 5 s for 4 vol % H 2 at −60°C. 181 Hydrogen sensor response variance at different absolute (atmospheric) pressures is another aspect that has not been scientifically addressed so far, since all studies we have surveyed have carried out their experiments at 1 atm (∼101 kPa). This is surprising since hydrogen sensors in, for example, automotive applications are very likely to be operated at varied altitudes. 7 To date, altitude or similar tests are first carried out at the prototype level and have a strong influence of the absolute atmospheric pressure on sensor response, even when the hydrogen partial pressure is kept constant. 176,182 We therefore advocate such tests taking place already at an earlier stage, when the active transducer materials are developed.
As an intermediate conclusion, it is clear that existing hydrogen sensors fulfill several of the US DoE performance targets, i.e., detection limit, selectivity, and power consumption. At the same time, a number of highly challenging targets remain unreached and relate to (i) response time, (ii) performance under poisoning/deactivation conditions, and (iii) in high humidity, operation at (iv) low/high temperature, and at (v) different absolute atmospheric pressure, where the last target (v) remains completely unaddressed. In other words, hydrogen sensor performance at realistic application conditions, rather than idealized laboratory environments, has rarely been addressed. In terms of active material, (nanostructured) Pd is most widely used due to its intrinsically high selectivity toward hydrogen combined with the potential for fast response (cf., Tables 1 and 2). Hence, to develop hydrogen sensors that are able to satisfy all of the stringent US DoE requirements and to alleviate the shortcomings of Pd as active material, one has to understand the fundamentals of the H−Pd interaction, as well as the opportunities offered by nanostructuring. Therefore, in the second part of this Perspective, we discuss these fundamentals and identify design rules for ultrafast, highly sensitive, poisoningand humidity-resistant, and hysteresis-free hydrogen sensors.

Fundamentals of Pd−H Interactions and Their
Implications for Hydrogen Detection. Pd enables an essentially barrierless hydrogen molecule (H 2 ) dissociation into chemisorbed hydrogen atoms (H) on its surface at ambient conditions ( Figure 2a). Once these atoms have been formed, they rapidly saturate the surface and diffuse into interstitial lattice sites in the subsurface region, and finally into the bulk of the system at hand. Upon diffusing, the H atoms face an energy landscape (Figure 2a), which is characterized by energetically more favorable subsurface sites compared to bulk interstitials. Consequently, subsurface sites can be assumed occupied, regardless of the hydrogen concentration in the bulk. To this end, the extension of this subsurface hydrogen layer has been proposed to be between 0.3 and 1 nm. 183−186 Furthermore, it has been shown that the presence of hydrogen in the subsurface layer leads to the generation of lattice strain, which can influence the thermodynamics of the sorption process in nanoscale systems, such as nanoparticles.
Since the surface is the first and last contact of a hydrogen molecule upon interaction with Pd, it plays a key role in the sorption processes. Hence, any modification of the physical and chemical properties of the surface, such as impurities and strongly adsorbed molecules, or atomic rearrangement due to, e.g., refaceting or elemental surface segregation in an alloy, will to a certain degree affect the sorption processes by changing the overall energy landscape. 19,187−194 Furthermore, engineering the surface-to-volume ratio (SVR) of nanostructures provides a route to modify the sorption kinetics, where smaller structures Figure 2. Palladium−hydrogen interaction. (a) Energy landscape encountered by a hydrogen molecule, H 2 , upon interaction with a Pd surface. In the first step, the H 2 molecule dissociates on the Pd surface. In the next step, the formed hydrogen atoms, H, diffuse into the subsurface region and occupy subsurface interstitial lattice sites. Subsequently, H diffuses interstitially further into the bulk. (b) Schematic of the different stages of Pd hydride formation. In the low hydrogen pressure regime, H is highly diluted in a solid solution (α-phase), locally expanding the Pd host lattice. Increasing the equilibrium concentration of H in the lattice, as a consequence of a hydrogen pressure increase in the environment, eventually creates sizable attractive H−H interactions via strain fields and electronic interactions that promote the formation of hydride (β-phase) nuclei. The growth of the β-phase then continues until the entire Pd host is transformed, and it is accompanied by significant expansion of the lattice. (c) Schematic of pressure−composition isotherms of the Pd−hydrogen system and the corresponding phase diagram. The equilibrium plateau pressure, at which the αand β-phases coexist, is temperature dependent and different for hydride formation and decomposition, due to hysteresis. The width of the plateau and the width of the hysteresis shrink for higher temperatures until they eventually vanish at the critical temperature, T C .

ACS Sensors
pubs.acs.org/acssensors Perspective generally exhibit faster response. 19,23,195 However, for particles larger than 5−10 nm, this is not the consequence of an altered energy landscape due to, e.g., lattice strain, 196 but rather due to shorter diffusion lengths for H to reach the core of the structure. 197,198 Once the surface is saturated with hydrogen, which for Pd occurs at very low pressures, 199 H species start occupying interstitial sites of the Pd host lattice to form a solid solution, which is known as α-phase (Figure 2b). In this regime, H is highly diluted and H−H interactions are very weak. Thus, the H/Pd ratio in the system increases proportionally to the square

ACS Sensors
pubs.acs.org/acssensors Perspective root of pressure according to Sieverts' Law. 200 Upon increasing the hydrogen pressure, the H-concentration increases proportionally, such that attractive H−H interactions (both electronic and via lattice strain fields 199 ) become appreciable. This eventually leads to the nucleation of the hydride (β-phase) at the two-phase equilibrium pressure (the "plateau"), where both α+β phases coexist as the system undergoes a first-order phase transformation (Figure 2b). Since the β-phase has a larger lattice constant than the Pd host (4.03 vs 3.89 Å), 201,202 this process is accommodated by significant lattice expansion, and the corresponding lattice strain is the origin of hysteresis between hydride formation and decomposition at constant temperature ( Figure 2c). 183,184,203,204 From a sensor perspective, the processes outlined above quite dramatically change both the electronic and optical properties of Pd and thus constitute the mechanism of hydrogen detection based on Pd, and explain Pd's intrinsically high hydrogen selectivity. 8,11 Specifically, for electrical hydrogen detection, the hydrogen absorption induces higher resistivity 21,205 or, in the case of discontinuous Pd nanostructures/films, expands their volume, which in turn forms new electrical contact points within the film/structures, resulting in reduced resistivity. 40,41 For optical hydrogen detection, the optical contrast generated due to hydrogen absorption into the host is measured (i) as a change in transmittance through a thin Pd film where it obeys the Beer− Lambert Law, 206,207 (ii) as a shift of the surface plasmon resonance (SPR) frequency of a thin Pd film, 77 or (iii) as a spectral shift of the localized surface plasmon resonance (LSPR) wavelength of Pd nanoparticles or nanostructures. 208,209 Also, indirect optical readout schemes based on inert plasmonic nanoantennas adjacent to a Pd nanostructure 184,210,211 or solutions based on nanostructured perfect absorbers 212,213 have been reported. They all have in common that the optical response is linearly correlated with the H/Pd ratio of the system. 207,209,214 Another aspect of importance for hydrogen sensing with Pd is the two-phase coexistence plateau and the hysteresis between hydrogen absorption and desorption in this regime because it creates a number of problems. First, since the phase transformation to the hydride phase, and thus the generation of a large optical contrast or electric conductivity change, occurs in a very narrow pressure range, sensitivity on either side of the phase transformation is rather low. This is problematic, since at room temperature, hydride formation in Pd occurs at ∼20−30 mbar, 199 which means that sensitivity below (and above) this pressure will be low. Second, hysteresis renders the sensor signal to depend on the history of the hydrogen pressure, i.e., on which branch of the hysteresis loop the sensor is located at a specific point in time. This may create ambiguity in the sensor reading, which severely hampers its accuracy.
Effects of Nanostructuring. In the limit of (ultra) thin films, nanostructures and nanoparticles, the SVR increases and factors like subsurface or low-coordination sites start to play an increasingly important role. 193,198 A second aspect is the impact of lattice strain, induced either by surface tension effects in the sub-10 nm particle size range 195,215−219 or by the formation of a subsurface hydride layer in (larger) nanocrystals, 183,184,203 or along grain boundaries of polycrystalline nanoparticles. 220,221 These strain effects directly affect hysteresis and render it particle size dependent. 183,184,203,216−218,222−224 Similarly, in thin film systems, hysteresis can be suppressed by reducing the film thickness down to a few nanometers, due to clamping effects at the Pd−substrate interface. 225−228 Another interesting effect observed in ultrasmall Pd nanoparticles, as well as in thin films with nanosized grains, is an apparent increased and decreased H solubility in the αand βphases, respectively. 218,229 Together, these two effects give rise to a narrowing of the two-phase coexistence plateau and have been explained by an increasing subsurface-to-bulk site ratio (nanoparticles) 223,224 and by the abundance of grain boundaries (thin films), 229 based on the fact that subsurface and grain boundary sites are likely to be fully occupied before the bulk, due to their favorable energetics.
As a final important trait of Pd nanostructures in general, and of Pd nanoparticles in particular, we note that the consequences of lattice expansion upon hydrogenation are not as severe as for bulk systems, where it is the main cause for embrittlement, cracking, and peeling when bound to a substrate, which in turn leads to rapid sensor aging and failure. Added to the fact that hydrogenation-induced defects are reversible during the reverse phase transformation in Pd nanoparticles, 230 nanostructured Pd hydrides promise potential for the improvement of sensor durability and thus sensor lifetime.

■ RATIONAL DESIGN OF NANOSTRUCTURED
PD-BASED HYDROGEN SENSORS Accelerating Response Times. In the quest to increase sensor speed, several different strategies have been tested and reported in the literature. The first one is based on the effect that increasing the SVR leads to faster sorption kinetics 198 (and thus response/recovery time) because (i) a larger surface area accelerates the hydrogen atom flux into the Pd and (ii) a smaller volume reduces diffusion path length to the core, as well as the total number of H atoms that need to be supplied to reach the new equilibrium state. This effect was first quantified in a fundamental study by Langhammer et al. for small (<5 nm) Pd nanoparticles. 195 In the context of sensors, this effect has been demonstrated on two examples using different nanostructures and sensing principles, namely, Pd nanowires with electrical readout 23 and PdAu alloy nanoparticles with optical readout (Figure 3a). 19,231 In both studies, clear proportionality between higher SVR and faster response time across all hydrogen concentrations (0.01−10 vol %) was found. By setting aside other factors influencing the kinetics, this finding suggests nanostructured Pd as the concept of choice for the design of ultrafast hydrogen sensors.
A second strategy to improve hydrogen sorption kinetics in nanostructured Pd is related to engineering its morphology, such as faceting and grain boundaries. For the former, a number of recent in situ investigations have shown that vertices of colloidal Pd nanocrystals act as hydride nucleation sites. 234−236 Building on this insight, a recent study by Johnson et al. 193 employed single-crystalline Pd nanocrystals with three different types of faceting, and thus different number of vertices, but with similar surface-to-bulk atom ratios achieved by adjusting size, i.e., octrahedral, cubic, and truncated cubic shape, which possess 6, 8, and 24 vertices, respectively (Figure 3b). By means of in situ XRD measurements, they showed that truncated cubic nanoparticles respond fastest when exposed to 5 vol % H 2 , followed by the cubic and octahedral particles. Due to very similar surface-to-bulk-atom ratios of all particle types, they conclude proportionality between response time and number of vertices, highlighting the role of nanocrystal shape in Pd hydrogenation and thus for sensor applications of single crystalline Pd nanoparticles. For polycrystalline nanostructured Pd on the other hand, morphology engineering offers a handle to improve  (Figure 3c), due to significant recrystallization and corresponding reduction of total grain boundary length (40%) and defect density (30%), and average grain size increase from 20 to 49 nm. 232 However, we also note that other reports conclude an opposite effect of grain boundaries, i.e., that at very low H 2 concentrations grain boundaries act as traps for H, which was reported to decelerate kinetics. 240,241 Beyond size reduction and morphology engineering, alloying of Pd provides yet another strategy to tackle the response time challenge, since characterization of the sorption kinetics of nanostructured Pd-alloy systems consistently reveals faster response than for the pure Pd counterparts, as first reported for PdNi alloy thin films 242 and PdAu alloy nanoparticles (Figure 3d). 231 Specifically for the latter system, systematic measurements revealed that increasing the Au concentration in the alloy results in a proportional acceleration of the hydrogen absorption process. These findings are confirmed in multiple studies using PdAu alloys in various forms, 19,243,244 as well as for other alloys, e.g., PdAg 245 and even the ternary system PdAuCu. 58 To this end, a recent DFT study by Namba et al. revealed that alloying Pd with Au lowers the energy barrier between surface and subsurface sites, which is responsible for the observed accelerated absorption (Figure 3d). 191 This is in good agreement with experimental works that investigated the apparent activation barriers for hydrogen sorption in PdAu 19 and PdCu nanoparticles. 58 A fourth way to tailor the energy landscape for hydrogen sorption to accelerate response times is through interfacing the Pd surface with another material, such as a polymer or a metal− organic framework (MOF). This effect was first reported by Ngene et al. in 2014, who upon sputter-deposition of a 15 nm thin PTFE layer onto PdAu thin films found significantly faster hydrogen absorption. 246 XPS analysis revealed the formation of Pd−CF x bonds at the PdAu/PTFE interface, which was proposed to create favorable adsorption sites for hydrogen. Further insights on the mechanism of the enhancement effect were later presented by Nugroho et al.,19 who revealed by DFT calculations on the example of Pd and PdAu alloy nanoparticles that the presence of the PTFE layer, via bond formation, reduces the activation barrier for H to migrate from a surface site to a the subsurface site (Figure 3e). Interestingly, in the same study a similar acceleration effect was also observed for another and chemically different polymer, PMMA, although with slightly lower barrier reduction. These two examples hint at the genericity of the effect, which appears to be governed by bond formation between the Pd surface and the polymer, with specific barrier reduction depending on the polymer chemistry at hand. However, we also note that the polymer thickness has to be considered when designing such a coating. When it is too thick, the response time may deteriorate due to the increasing contribution of molecular H 2 diffusion through the polymer to the overall response. 247,248 Similar to polymer coatings, also Pd coated with MOFs has been reported to feature superior hydrogen sorption kinetics. For example, Koo et al. found an improvement by a factor 20 when coating their Pd nanowires with a thin ZIF-8 layer and testing their sensor in air. 20 They put forward the filtering of O 2 by the MOF as the reason because O 2 is known to catalytically react with H 2 on the Pd surface to form water and thus "compete" with hydrogen sorption. However, we argue that this conclusion cannot explain similar acceleration effects observed when other Pd@MOFs were exposed to hydrogen in N 2 environment or even in pure H 2 . For example, earlier work by Li et al. showed that a Pd nanocube@HKUST-1 system responds faster compared to its uncoated counterpart in pure H 2 . 249 A followup DFT study by Nanba et al. then revealed that the MOF coating leads to an increased diffusion rate of hydrogen due to hydrogen adsorption destabilization induced by the Cu atom in the MOF. 250 They also propose a steric effect to take place, which creates a new hydrogen diffusion path through a Pd 5 Cu octahedral site. Hence, their findings hint at a similar generic enhancement effect for MOFs as for polymer coatings. As a final comment, we note that coating with thin polymer or MOF layers provides an effective way to accelerate the Pd-based sensors' response not only through modification of absorption energy landscape but also via filtering of unwanted species that may poison or deactivate the sensor, as we discuss in detail below. 251 To start the discussion of the last reported strategy for response time enhancement, we remind ourselves that hydrogen sorption in Pd is an activated process and therefore depends on temperature. Thus, heating a Pd-based sensor, for example, using an external heater, can accelerate its detection speed. This strategy is particularly important when the sensors are to be used in a low temperature environment. With a hydrogen absorption activation energy of 60−80 kJ/mol H 2 for pure Pd, 19,58,192,196 a mere 10 K temperature increase from RT will roughly double the sensor speed. This approach has been demonstrated by Yoon et al. by positioning the active Pd on a Pt microheater, which locally heats the active sensing area up to 150°C ( Figure  3f). 233 On the downside, however, heating will also reduce the absolute amount of hydrogen absorbed into the system, and therefore lower sensor signal amplitude for a specific hydrogen pressure change, meaning that also the limit of detection will be reduced.
From the above summary of different factors that affect Pdbased sensor response times, it is clear that the surface state of Pd is particularly important. Therefore, as the last point, we want to highlight here an aspect of high relevance when using colloidal Pd nanocrystals for hydrogen sensors, namely, that they inherently come with a surfactant (or similar) coating, applied during the synthesis in solution to both promote/block growth of specific surface facets and to prevent aggregation. To this end, in recent studies, the impact of the presence of the widely used ligands cetyltrimethylammonium bromide (CTAB), cetyltrimethylammonium chloride (CTAC), and tetraoctylammonium bromide (TOAB) has been investigated in detail. As the key conclusion, it was shown that their presence in the Pd surface increases response times significantly, as a consequence of an increased apparent activation barrier for hydrogen absorption. 192,194 At the same time, in line with the accelerating effect of a polymer coating discussed above, Stolaś et al. showed that using the polymeric colloidal stabilizer poly(vinylpyrrolidone), PVP, instead accelerates the kinetics. 192 Similarly, post-synthesis removal of surfactants from the nanoparticles has been shown to amend the negative effects of their presence on the Pd nanoparticle surface. 252 Suppressing Hysteresis. The inherent hysteresis between hydride formation and decomposition in pure Pd significantly  58 PdNi, 242 PdTa, 254 and ternary PdAuCu 58 and PdCuSi alloys. 255 Depending on the atomic radius of the alloyants, two types of pressure−composition isotherms exist. First, for alloyants with an atomic radius larger than Pd, e.g., Au, hysteresis shrinks symmetrically and disappears at room temperature (RT) at ca. 25 at % Au, since the critical temperature, T C , at this concentration has been lowered to below RT (Figure 4a). 256 Mechanistically, this can be understood as the Au atoms, which occupy Pd lattice sites, slightly expanding the Pd host. Thereby, they reduce the straininduced energy barrier created upon hydrogen sorption, and thus hysteresis. 256,257 To this end, Wadell et al. have shown that alloying Pd with 25 at % Au reduces sensor uncertainty to below 5% across the 1−1000 mbar H 2 pressure range, 231 which is on par with the corresponding performance target by the US DoE. Similarly, alloying with metals with smaller atomic radius than Pd, e.g., Cu 58 and Ni, 242 also suppresses hysteresis at ca. 25 at % alloyant concentration. However, due to a now slightly contracted Pd lattice that will increase the strain-induced energy barrier to form the hydride, hydride formation in this case occurs at higher pressures compared to neat Pd hydrogenated at the same temperature (Figure 4a). 258,259 From a sensing perspective, this leads to a lower sensitivity in the low hydrogen partial pressure regime. 209,255 Another way to suppress hysteresis is through size reduction of the system to the regime below 10 nm. This was explicitly demonstrated by Langhammer et al., who systematically measured optical pressure−composition isotherms of Pd NPs of different size across the range between 1.8 and 8 nm and found that hysteresis vanishes for a size below ∼3 nm at 30°C (Figure 4b). [215][216][217]224 Mechanistically, this can be understood through the so-called metal@hydride core@shell model. In this model, during the absorption (desorption) process, the hydride (metal) formation induces expansion (contraction) at the surface, while the metal (hydride) core shrinks. The volumetric difference between the two phases and the distortional deformation formed during the phase transformation lead to a mechanical stress at the interface of the two phases, which causes hysteresis. In sufficiently small nanoparticles, the corresponding strain energy is reduced and hysteresis therefore disappears. 224 Similarly, also for a thin Pd film, hysteresis suppression can be achieved by reducing the thickness down to 5 nm as shown by Lee et al. 227 The suppression is due to the clamping effect by the support, which becomes more pronounced in an ultrathin Pd layer. This clamping restricts the Pd lattice such that it is energetically (too) costly to form the β-phase. Furthermore, the clamping effect depends on the support material, 226 and it can be intensified, for example, by a Ti buffer layer grown between the substrate and the Pd film. 225 As the last strategy to reduce or completely avoid hysteresis in a sensor application, local heating of the active sensor area to above T C can be applied to retain the system in the extended solid-solution state. For this purpose, Fisser et al. have provided a guideline for the corresponding hydrogen pressure and operational temperature required to retain a Pd thin film in the α-phase (Figure 4c). 260 For bulk Pd, T C is 295°C, while for small nanoparticles it decreases proportionally to their size. 203 As previously mentioned, however, limiting the sensing operation to the α-phase leads to a reduced limit of detection.
Protection against Poisoning Species and Humidity. Any surface at ambient conditions is covered by multilayers of molecular species present in air, including H 2 O. Since the state of the surface is critical for any Pd-based hydrogen sensor, this is problematic. Among molecular species present in ambient air, CO chemisorbs strongly on Pd surfaces and thus effectively blocks it for H 2 dissociation (other species with similar effect are NO x and SO x 83,167 ). This, in turn, significantly slows, or even completely prevents, hydrogen absorption into Pd. Therefore, finding ways to eliminate such poisoning effects is very important to ensure long-term reliable sensor operation. To this end, it is well-known that alloying Pd with Cu reduces the affinity of Pd surfaces to CO. 261−265 Hence, this concept has been successfully used in hydrogen separation membrane technology, 263,266 and since rather recently also in hydrogen sensing. Mak   hydrogen sensor, however, with the main aim to reduce O 2 interference. 267 Then, Darmadi et al. presented a systematic study of optical PdCu alloy nanoparticle sensors with focus on CO poisoning, which showed that incorporating as little as 5 at % Cu to Pd effectively eliminates CO poisoning even at 5000 ppm CO in air (Figure 5a). 58 Interestingly, they also found that the added Cu works synergistically when combined with another alloyant, i.e., a ternary PdAuCu alloy resulted in a CO-resistant and hysteresis-free sensor. Unfortunately, however, with regard to other poisoning gases such as NO 2 , this alloying strategy turned out to be insufficient. 58 An alternative and potentially more broadly applicable strategy is the use of the molecular filtering properties of polymers or MOFs, when applied as coatings to a Pd-based active sensor surface. The efficiency of this approach is then directly dependent on the selectivity of the "filter" and the diffusivity of H 2 within it. If the latter is low, even if good protection is obtained, sensor response time will be dramatically increased. As an example for a polymer system, PMMA exhibits high H 2 permeability, while diffusivities of CO and NO 2 are very low. 269 Therefore, PMMA coatings have been used successfully to protect hydrogen sensors from these species. 19,82,168,247,248 For example, Nugroho et al. have demonstrated that a PMMA coating as thin as 30 nm is sufficient to enable the operation of a PdAu alloy optical hydrogen sensor in 30,000 ppm CO 2 , 1000 ppm CO, and 100 ppm NO 2 mixed in synthetic air (Figure 5b). Furthermore, bulk-processed Pd@PMMA nanocomposite hydrogen sensors have been demonstrated to exhibit exceptional long-term stability at ambient conditions. 247 Finally, PMMA also possesses excellent selectivity toward O 2 , 251 a trait that is shared with MOFs that have started to find application in hydrogen sensors as molecular filtering and protection layers, e.g., ZIF-8. 20,82,270 Finally, we propose that employing the concept of a filtering coating may also hold the key to the development of highly humidity-tolerant hydrogen sensors and thereby tackle this so far widely unaddressed scientific challenge and resolve the corresponding DoE target. As one rare effort in this context, PTFE coatings have been utilized for humidity protection due to the inherently high hydrophobicity of fluorinated polymers, 268 and Mak et al. have utilized a 30 nm PTFE coating on an optical fiber sensor to demonstrate efficient hydrogen detection in oil. 267 As an alternative, alloying also may provide a means to increase the humidity resistance of Pd-based H 2 sensors, as demonstrated for the PdAu system, however, only up to 60% RH. 112 Last, locally heating the region where the active transducers are placed also provide a potentially effective strategy to reduce the detrimental impact of humidity, as demonstrated of the examples of heated Pd@Si nanowire 98 and Pd@graphene 119 hydrogen sensors.

■ FUTURE PERSPECTIVES
The recent acceleration in the large-scale global implementation of hydrogen energy technology has dramatically expedited the need for fast, selective, robust, and long-lived hydrogen safety sensors. In this respect, our assessment of state-of-the art hydrogen sensor technology has revealed a significant gap between currently available sensor performance and the performance targets set by stakeholders like the US DoE. This gap is, however, steadily closing, as a consequence of intense research activities in the field of hydrogen sensors during the last 15 years. For example, detection limits are now in the low ppm or even ppb range, and sensor response and recovery times have developed rapidly with the first reports of subsecond speed in the 1 mbar pressure regime, which is on par with the corresponding DoE target. This development has predominantly been enabled by nanostructured Pd and Pd alloy transducer materials, which,  95 Cu 5 (right) nanoparticle hydrogen sensors to three 4% H 2 pulses, followed by 9 pulses of 4% H 2 + 0.5% CO in synthetic air. Alloying 5 at % Cu to Pd is sufficient to suppress the CO poisoning effect. Adapted from ref 58 Intermittently, the sensor was exposed to humid air and dipped in water, as marked. The PTFE layer provides significant protection as proven by unchanged response after each exposure to humid conditions. Adapted with permission from ref 268. Copyright 2012 The International Society for Optics and Photonics.

ACS Sensors
pubs.acs.org/acssensors Perspective irrespective of the readout principle, offer intrinsically very high selectivity toward hydrogen thanks to the hydride formation process and rapid response due to short diffusion paths. It is therefore likely, and also our personal opinion, that clever engineering of Pd and Pd-alloy material properties at the nanoscale holds the key to even further push sensor detection limits and response times. At the same time, alternative material solutions are highly welcome since the foreseeable broadening of hydrogen sensor application conditions in the wake of the largescale deployment of hydrogen energy technologies most likely requires increasingly tailored sensor solutions, which will be hard to provide on a single material platform alone. For example, hydride forming metals like Mg, 268,271 Ta, 272 V, 273 or Hf 272,274 may offer sensors with significantly wider dynamic range and unprecedented detection limits, however, at the cost of significantly longer response and recovery times, due to (orders of magnitude) slower hydrogen diffusion in these materials.
Other highly important metrics that have been much less addressed by the field are operation temperature, absolutepressure range, sensor lifetime, and operation in poisoning/ deactivating conditions and in high humidity. However, specifically for the last two points, interesting developments on the basis of polymeric or MOF coatings have been reported, where the molecular filtering properties of such layers have been demonstrated to prevent molecules larger than H 2 to reach the Pd surface, and thus hinder poisoning species from blocking the surface. Interestingly, if combined with nanoscale Pd transducers, the presence of such layers has also been reported to enhance response time, due to strong interactions between the metal surface and the coating. Furthermore, there are convincing indications that such hydrogen sorption kinetic enhancement effects are generic to both polymeric and MOF coatings, which fuels the hope that specific selection and optimization of coating materials in this respect may enable unprecedented response times. Here, however, the current lack of fundamental understanding of these effects hampers the identification of corresponding material design rules and calls for research efforts both by experiment and theory.
As a further key message to the community, we identify the almost complete lack of effort to design hydrogen sensors that can operate at high humidity levels up to 95% RH. This is critical in view of the fact that in the context of, for example, fuel cells, high humidity levels occur in the exhaust, and that hydrogen production sites at or close to, for example, offshore wind farms or offshore hydrogen pipelines will experience vast humidity fluctuations. Mechanistically, it is likely that the concept of coatings, as well as alloying, may hold the solution to this problem, but experimental demonstrations are widely lacking. Furthermore, the long-term stability of coatings or surface segregation effects in alloys are factors that need to be critically addressed, ideally already at the early active sensor material development stage, and not only at the sensor device prototype or product level, as commonly done today.
Finally, we want to highlight the need for more standardized hydrogen sensor characterization already at the research level to, for instance, enable more straightforward comparison of sensor performance, as well as to guarantee that sensor performance is measured at conditions relevant for targeted applications. This means the assessment of detection limits, operation temperatures, response and recovery times in (synthetic) air rather than in inert gas or even vacuum environments, and in the presence of trace molecular species found in the atmosphere, such as CO 2 , NO x , CO, SO x and H 2 O. It also means the assessment of response and recovery times over a wide(r) range of hydrogen concentrations since these performance parameters are strongly concentration dependent. We also advertise the adaptation of the ISO 26412 protocol for the assessment of poisoning/ deactivation resistance of hydrogen sensors. The manuscript was written through contributions of all authors. All authors have given approval to the final version of the manuscript.

Notes
The authors declare the following competing financial interest(s): C.L. is co-founder of Insplorion AB that develops plasmonic hydrogen sensors.

■ ACKNOWLEDGMENTS
We acknowledge financial support from the Swedish Foundation for Strategic Research framework project RMA15-0052, from the Knut and Alice Wallenberg Foundation Project 2016.0210, and from Swedish Energy Agency project 49103-1. Parts of the TOC and figures use free-licensed resources from Freepik and Macrovector (www.freepik.com). (2) The European Green Deal: hydrogen is a priority area for a clean and circular economy. https://www.fch.europa.eu/news/european-greendeal-hydrogen-priority-area-clean-and-circular-economy (accessed Aug 1, 2020).