Optimization of the Composition of PdAuCu Ternary Alloy Nanoparticles for Plasmonic Hydrogen Sensing

Recommendations ABSTRACT: Alloying is a long-standing central strategy in materials science for the tailoring and optimization of bulk material properties, which more recently has started to ﬁ nd application also in engineered nanomaterials and nanostructures used in, among other, nanoplasmonic hydrogen sensors. Speci ﬁ cally, alloying Pd nanoparticles to form binaries and ternaries with the coinage metals Au and Cu has proven e ﬃ cient to mitigate hysteresis in the sensor response, improve response and recovery times, boost sensitivity in the low hydrogen concentration sensing range, and reduce the detrimental impact of carbon monoxide poisoning. However, when surveying the corresponding studies, it is clear that there is a trade-o ﬀ between the sensitivity enhancement and the CO-poisoning resistance e ﬀ ects provided by Au and Cu alloyants, respectively. Therefore, in this work, we systematically screen the impact of the Au and Cu concentration in PdAuCu ternary alloy nanoparticles used for plasmonic hydrogen sensing, to obtain a champion system with maximized sensitivity and CO-poisoning resistance based on an evaluation using the stringent ISO 26142 test protocol. As the main results, we ﬁ nd that the best hysteresis-free and sensitive response combined with deactivation resistance to 500 ppm CO in synthetic air is obtained for the Pd 65 Au 25 Cu 10 ternary alloy system, which also exhibits good long-term stability during operation under severe CO poisoning ber-coupled xed-grating


■ INTRODUCTION
The development of hydrogen safety sensors has accelerated significantly in recent years due to the imminent large-scale implementation of hydrogen-energy technologies. Among the emerging hydrogen sensor principles, solutions based on nanosized Pd and optical readout based on localized surface plasmon resonance (LSPR) are among the most promising and mature ones. 1 Such nanoplasmonic hydrogen sensors rely on visible light-induced collective and coherent excitation of electrons in hydride-forming metal nanoparticlesthe LSPRas a signal transducing mechanism, where the hydrogen sorption and hydride formation processes give rise to a sizable optical contrast, which is proportional to the hydrogen concentration in the sensor surroundings. 2 At the same time, despite their excellent selectivity toward hydrogen gas, the performance of pure Pd-based sensors, in general, and of plasmonic solutions, in particular, is hampered by several inherent limitations, such as hysteresis, slow response/recovery times, and proneness to deactivation by poisoning species like CO. 2−5 As a consequence, numerous strategies have been developed to minimize the impact of these negative effects and the use of coinage metal-Pd alloys has emerged as one of the most promising directions in this respect. 6−10 For example, hydride formation hysteresis can be suppressed by alloying Pd nanoparticles with at least 25 at % Au. 7,11 Interestingly, the corresponding research also has revealed that alloying with Au reduces the hydrogen sorption apparent activation energies, which leads to faster sensor response/recovery compared to pure Pd. 12 Similarly, it has been demonstrated that also a second limitation of pure Pd can be mitigated by alloying since the addition of Cu to Pd significantly improves the resistance against sensor poisoning and deactivation by CO. 8,10 Following this line, in a proof-of-principle fashion, we have recently shown that these two strategies can be combined in a ternary PdAuCu alloy for plasmonic optical hydrogen detection. 10 However, in such a ternary system, the introduction of Cu not only mitigates the CO poisoning effect but also quite dramatically reduces sensor sensitivity due to the shift of the two-phase equilibrium plateau to higher hydrogen partial pressures. 13,14 Therefore, to optimize the performance of this ternary alloy system for application in plasmonic hydrogen sensors, there is an imminent demand for the systematic optimization of the alloy components with respect to the following three criteria: (i) poisoning/deactivation resistance, (ii) hysteresis suppression, and (iii) sensitivity. Therefore, here, we systematically screen the impact of the ternary PdAuCu alloy composition on these three parameters, and we implement for the first time in plasmonic hydrogen sensor evaluation the rigorous ISO 26142 testing protocol recommended by the US National Renewable Energy Laboratory (NREL) to assess sensor poisoning and deactivation under simulated application conditions in synthetic air. 15

■ RESULTS AND DISCUSSION
To determine the minimal amount of Cu required to obtain a sizable CO-poisoning protection, we first systematically investigated binary PdCu nanoparticle arrays fabricated using hole-mask colloidal lithography 16,17 (Figure 1a) and comprised of quasi-random arrangements of nanodisks with an average diameter of 190 nm and a height of 25 nm (Figure 1b). The alloy at. % composition was adjusted by controlling the thickness of the evaporated constituent layers. Later, the multilayer nanodisks were annealed at 500°C in 4% H 2 + 96% Ar for 24 h to ensure complete alloy formation through the particles. 10 Specifically, we nanofabricated a series of PdCu alloy samples with Cu concentrations of 0, 1, 3, 5, 7, 10, and 15 at. %. We then evaluated their optical plasmonic response, defined as spectral peak-centroid position shifts, 18 Δλ peak (Figure 1c), to 3 vol. % H 2 pulses in synthetic air (chosen as a representative concentration close to the lower flammability limit of H 2 in air), and in CO backgrounds ranging from 100 to 500 ppm. Here, we highlight that the experiment was designed to comply with the ISO 26142 testing protocol 15 according to which the poisoning gas (CO) is introduced before the test gas pulse (H 2 ) and kept as a constant background in synthetic air ( Figure 1d).
Focusing first on the response of the neat Pd control, we notice the severe impact of CO on the H 2 sorption kinetics, as signified by the Δλ peak signal not reaching saturation within the given 5 min of the H 2 pulse even at the lowest 100 ppm CO concentration ( Figure 2). Also, the sensor does not recover within 5 min of a subsequent synthetic air carrier gas flush with 100 ppm CO. Furthermore, we see that there is a CO concentration dependence of the hydrogen absorption kinetics, with the response being fastest for the 100 ppm CO background. Similarly, we observe faster and more complete recovery of the sensor and a lower CO concentration in the gas. This implies a CO-concentration-dependent poisoning effect, which is reasonable since CO adsorption on Pd is expected to increase at a higher concentration according to, for instance, the Langmuir adsorption model. 19 Similarly, when the CO background is removed again, the sorption kinetics become faster, which indicates that the CO poisoning is not permanent, in agreement with an O 2 -assisted reaction and desorption mechanism. 20 To start the analysis of the impact of Cu concentration in the alloy nanoparticles on the CO poisoning resistance, we first glance over the response curves of the sensors with increasing Cu concentrations summarized in Figure 2. As a clear general trend, we see that the higher the Cu concentration, the more tolerant the sensor becomes to the presence of CO. Interestingly, in 100 ppm CO, as little as 1 at. % Cu is sufficient to remedy the CO poisoning significantly. These very low required Cu concentrations are somewhat surprising at first but can be rationalized on the basis of first-principles calculations, which show that CO binds less strongly on PdCu alloy surfaces compared to pure Pd, due to both an electronic ligand effect and a geometrical (ensemble) effect. 21−24 The ensemble effect in this case means that alloying reduces the number of three-Pd atom sites to which CO binds very strongly and that already rather small amounts of Cu on the surface are enough to the significantly reduce the availability of such sites. 25 Furthermore, the ensemble effect may even be more dominant if any surface segregation would occur since Cu is known to have a tendency to move to the surface. 26,27 Finally, we also note that it is plausible that Cu alloying also reduces the poisoning effect of sulfuric compounds (e.g. SO 2 and H 2 S) since their adsorption mechanism is similar to the one of CO. 28,29 For a more detailed analysis of the data presented in Figure  2 and to investigate the minimum Cu concentration required to achieve significant CO tolerance, we focus on the sensor response during the first hydrogen cycle in the five different CO background concentrations at hand (Figure 3a−d). This analysis reveals both a decreasing steady state Δλ peak for increasing Cu concentration (see detailed discussion below) and a pronounced CO poisoning effect for H 2 desorption manifested as the deceleration of the kinetics. For absorption, we cannot properly resolve the true kinetics since the gas exchange time constant of the used flow reactor system is similar to the response time of the sensor. Therefore, we used the temporal response of the sensors during (the significantly The alloy composition is adjusted by tuning the thickness of the corresponding layers evaporated through the hole mask. 17 Subsequently, a homogeneous alloy is obtained by annealing at 500°C in 4% H 2 + 96% Ar for 24 h. 10,11 (b) SEM micrograph of a representative Pd 90 Cu 10 (top) and Pd 65 Au 25 Cu 10 (bottom) alloy nanodisk quasi-random array. Scale bar 500 nm. (c) Schematic of the plasmonic sensing readout. The sensor response is defined as the plasmonic peak shift (Δλ peak ) of the Pd(alloy) nanoparticles during the hydrogen absorption/desorption. (d) Schematic of the test protocol used to assess CO poisoning tolerance according to ISO 26142. slower) hydrogen desorption to quantitatively analyze the effect of Cu for increasing CO tolerance. Moreover, the deceleration effect of CO is more apparent in the desorption process because the rate-limiting factor for desorption is hydrogen recombination on the Pd surface, 30,31 which is more susceptible to CO poisoning. In contrast, the absorption process is rate-limited by hydrogen diffusion in the Pd bulk. 32 We define the CO deceleration factor as where t 50 is the time to reach 50% of the full signal for a H 2 concentration change in pure synthetic air. 33 The CO superscript in t 50 CO denotes t 50 measured in the CO background. With this definition, the deceleration factor, f, reflects the sensor tolerance to CO poisoning, where a larger f indicates reduced CO tolerance. The correspondingly derived deceler-ation factors of all alloys for different CO concentrations are plotted in Figure 3e. If we then set the tolerance limit to f = 1, only the 10 and 15 at. % Cu alloys can be considered CO tolerant up to 500 ppm CO. This indicates that the required Cu concentration to promote CO resistance is higher than that we previously reported (5 at. %), 10 as a consequence of the more rigorous evaluation in the ISO 26142 framework used here, and that it strongly depends on the absolute CO concentration in the background. Hence, for our further optimization of the ternary PdAuCu alloy system reported below, we use ≥10 at. % Cu. Finally, we also note that the identified deactivation resistance is quite remarkable, considering that the CO background concentration implemented in ISO 26142 tests of commercial sensors was one order of magnitude lower, that is, 50 ppm. 34 Having identified the necessary minimal Cu content in a Pd alloy for achieving a sizable CO-poisoning resistance in the ISO 26142 testing protocol, we now turn to assess the impact  of adding Au to the PdCu system with respect to obtaining hysteresis-free response and maximized sensitivity. For this purpose, we first measured optical Δλ peak isotherms at 30°C in a custom-built vacuum chamber with optical access. 7 We used samples alloyed with 10,15, and 20 at. % Cu, respectively, to which we added 15, 20, 25, and 30 at. % Au to form the corresponding ternary systems (Figure 4). This analysis shows that for 10 at. % Cu, at least 30 at. % Au is necessary for complete hysteresis suppression (Figure 4a). For 15 at. % Cu, complete suppression is achieved with 25 at. % Au (Figure 4b), and for 20 at. % Cu, only 20 at. % Au is necessary (Figure 4c). Therefore, complete hysteresis suppression is achieved when the sum of Au and Cu is around 40 at. %, which is slightly higher than the necessary concentration in binary PdAu alloys, where complete hysteresis suppression is achieved at 30 at. % Au 7,35,36 This implies that, to some extent in the ternary system, Au and Cu are annulling each other with respect to the hysteresis suppression effect because Au expands the host lattice 37 and Cu contracts it 38 due to larger and smaller atomic radii, respectively, compared to the Pd host.
To provide a different perspective of the data shown in Figure 4, we replot the same isotherms but for a constant Au concentration in each panel ( Figure 5). Starting with 15 at. % Au and three Cu concentrations of 10, 15, and 20 at. %, we clearly see how the addition of Cu elevates the plateau pressure of the ternary system in the same way as observed for PdCu binaries (Figure 5a). 10 The same trend can also be found for ternaries based on 20, 25, and 30 at. % Au (Figure 5b−d). Simultaneously, increasing the Cu concentration also induces a significant reduction in Δλ peak for a given hydrogen pressure change (Figure 5), which is equivalent to a sizable reduction of sensor sensitivity and thus limit of detection.
In summary, the data presented in Figures 4 and 5 indicate that in PdAuCu ternary alloys, Au mainly contributes to the closing of the hysteresis gap, whereas Cu leads to an elevation of the plateau region to higher hydrogen partial pressures. Projected onto the hydrogen sensor application, this means that there exists an optimal ternary alloy composition which is the best compromise between the pros and cons of alloying with Au and Cu, with respect to (i) eliminating the hysteresis gap and (ii) maximizing the magnitude of Δλ peak per hydrogen partial pressure change. Criterion (i) is achieved for alloys with at least 25 at. % Au, and criterion (ii) sets an upper limit to both Au and Cu concentrations since increasing them generically reduces the overall sensor sensitivity because Δλ peak per hydrogen partial pressure change is reduced. As the result of these selection rules, the Pd 65 Au 25 Cu 10 ternary alloy system emerges as the best compromise and thus our champion system. It is worth mentioning that the slight gap between the absorption and desorption branches observed for the Pd 65 Au 25 Cu 10 alloy in Figure 5c is a measurement artifact rather than an inherent hydride hysteresis gap. This is corroborated by the absence of the α−β phase transition along the curve, which is a strong indication of hysteresis-free hydride formation.
As the last step of our analysis, we therefore investigate the poisoning resistance of Pd 65 Au 25 Cu 10 by exposing it to more than 50 five-minute pulses of 4 vol. % H 2 in the constant 500 ppm CO background in synthetic air carrier gas at 30°C, over a duration of 12 h (Figure 6a,b). Evidently, the sensor exhibits excellent stability and deactivation resistance, as manifested in essentially constant response and recovery times (Figure 6c) during the entire experiment. Furthermore, response/recovery times are identical to those of the sensor operating in pure synthetic air (no CO background) both before and after CO exposure and in 500 ppm CO. The Δλ peak amplitude ( Figure  6d), on the other hand, shows a slight drift, which may indicate a minor gradual change that is likely due to alloyant surface segregation within the first few atomic layers, induced by CO and O 2 . 39−42 The overall nanodisk shape and arrangement on the surface, however, remain unchanged even after the longterm test (Figure 6f), and also the Δλ peak amplitude still lies within the ±20% tolerance limit set by ISO 26142 (Figure  6e). 43 This further corroborates the very good stability and resistance to deactivation of the Pd 65 Au 25 Cu 10 ternary alloy system.

■ CONCLUSIONS
We have systematically assessed the optimum Au and Cu alloyant concentrations in ternary alloy PdAuCu nanoparticles to enable hysteresis-free and sensitive plasmonic hydrogen  sensor operation in a CO-rich environment according to the ISO 26142 testing protocol. We have found that the minimum Cu content required to provide deactivation protection in the 500 ppm CO background in synthetic air is 10 at. % and that the required minimum amount of Cu depends on the CO concentration in the background, where a higher CO concentration requires a higher Cu content. Moreover, we observed that, in the ternary PdAuCu system, the hysteresis suppression effect is predominantly driven by the Au component and that the minimum amount of Au required for obtaining a linear, hysteresis-free, and yet sensitive response is 25 at. % Au in combination with a Cu concentration of 10 at. % < Cu < 20 at. %. Consequently, we have identified the Pd 65 Au 25 Cu 10 alloy as the champion system that constitutes the best compromise in terms of hysteresis-free response, sensitivity, and CO deactivation resistance. This system also exhibits very good long-term stability during operation under severe CO poisoning conditions, where it shows no sign of deactivation after more than 50 hydrogen pulses in the 500 ppm CO background after more than 12 h on stream. For further quantitative assessment of PdAuCu ternary alloy hydrogen sensor performance in comparison to other types of hydrogen sensors, we refer to a recent review. 1 In a wider perspective, this study may provide important insights into other applications of Pd alloy systems and their hydrides, such as in hydrogen separation membrane technology or heterogeneous catalysis, where alloying is a widely employed strategy for performance optimization with respect to, for example, hydrogen permeability or the catalytic reaction rate under COrich conditions. ■ METHODS Sample Fabrication. The alloy nanodisk arrays were fabricated using layer-by-layer deposition through a hole mask 16 onto 1 × 1 cm glass substrates (Borofloat, Schott Scandinavia AB). Deposition was followed by annealing at 500°C for 24 h under a flow of 4% H 2 in Ar. 17 A more detailed description of the nanofabrication procedure can be found in our earlier work. 17 Deactivation and Poisoning Tests. The deactivation measurements were carried out in a quartz tube flow reactor with optical , with and without the 500 ppm CO background, as depicted in (b). (c) Sensor response and recovery times (t 50 ), and (d) Δλ peak extracted for each cycle and plotted versus cycle number. (e) Δλ peak normalized with the response to the first hydrogen pulse for (i) the initial H 2 cycles in pure synthetic air, (ii) the H 2 cycles in the 500 ppm CO background in synthetic air, and (iii) final H 2 cycles again in pure synthetic air. The error bars correspond to standard deviation of the cycles in each category. The red band denotes the ±20% tolerance set by ISO 26142. 43 (f) SEM micrograph of the nanodisk array after the long-term deactivation test. Scale bar denotes 500 nm.
access for transmittance measurements (X1, Insplorion AB) at atmospheric pressure, using synthetic air as the carrier gas, which has a time constant for complete gas exchange of ∼10 s. 44 The gas flow rate (335 mL/min) and gas composition were controlled by mass flow controllers (Bronkhorst ΔP). The sample inside the flow reactor was illuminated by white light (AvaLight-Hal, Avantes) through an optical fiber equipped with a collimating lens. The transmitted light was then analyzed using a fiber-coupled fixed-grating spectrometer (AvaSpec-1024, Avantes). The LSPR peak descriptors were obtained by fitting a 20th order polynomial to the LSPR peak in the measured optical extinction spectra. The measurement temperature was kept at 30°C.
Pressure−Δλ peak Isotherm Measurements. The isotherm measurements were performed in a custom-made vacuum chamber setup with optical windows reported earlier. 7,17 The absolute hydrogen pressure in the chamber was monitored using two capacitive pressure gauges with different ranges (MKS Baratron). Optical transmittance through the sample was enabled using quartz vacuum viewports mounted on the vacuum chamber and by using a fibercoupled, unpolarized white light source (Avantes AvaLight-Hal) and a fixed grating fiber-coupled spectrophotometer (Avantes SensLine AvaSpec-2048XL). The pressure inside the chamber was controlled using microbar-precision leak valves. The temperature was controlled with a heating coil wrapped around the chamber and a temperature controller (Eurotherm 3216N) in a feedback loop manner, where the sample surface temperature inside the vacuum chamber was continuously used as the input. We performed all of our experiments at constant 30°C. The LSPR peak descriptors were obtained by fitting a Lorentzian to the LSPR peak in the measured optical extinction spectra.