Sub-second and ppm-level optical sensing of hydrogen using templated control of nano-hydride geometry and composition

The use of hydrogen as a clean and renewable alternative to fossil fuels requires a suite of flammability mitigating technologies, particularly robust sensors for hydrogen leak detection and concentration monitoring. To this end, we have developed a class of lightweight optical hydrogen sensors based on a metasurface of Pd nano-patchy particle arrays, which fulfills the increasing requirements of a safe hydrogen fuel sensing system with no risk of sparking. The structure of the optical sensor is readily nano-engineered to yield extraordinarily rapid response to hydrogen gas (<3 s at 1 mbar H2) with a high degree of accuracy (<5%). By incorporating 20% Ag, Au or Co, the sensing performances of the Pd-alloy sensor are significantly enhanced, especially for the Pd80Co20 sensor whose optical response time at 1 mbar of H2 is just ~0.85 s, while preserving the excellent accuracy (<2.5%), limit of detection (2.5 ppm), and robustness against aging, temperature, and interfering gases. The superior performance of our sensor places it among the fastest and most sensitive optical hydrogen sensors.

Our simulation is based on these following assumptions: • Only the shadowing effect and material accumulation are considered. Other physical processes, such as surface diffusion or material penetration, are neglected.
• The deposition at different surface elements happens simultaneously as long as they are directly exposed to vapor.

S4
• The as-deposited film is non-porous and is uniform within each surface element.
In our experimental metal deposition, the substrate holder was rotated azimuthally at a constant rate of 30 rpm to thoroughly cover the top surface of nanosphere. Therefore, in order to mimic this process, we break down the simulation into 3600 steps. We start at = 0° (step index ki = 1), a 0.1° azimuthal rotation of , happens at the end of each step, until a round of azimuthal rotation is completed ( = 359.9°) (step index kf = 3600). In each simulation step, the thickness at each surface element h(θ, φ) is updated, The change of thickness for each step Δℎ is determined by whether surface element ( , ) is directly exposed to the vapor flux or not: • If the surface element ( , ) is under the shadow of other structures (deposited materials on neighboring bead in the previous steps are also considered), then Δℎ = 0. The vapor flux Φ is defined as, where & ' is the projection of S( , ) onto the plane perpendicular to the vapor flux , . With β is the angle between and the surface normal vector # ( ( Supplementary Fig. 1c), & ' can be written as, Combining Supplementary Equations (1)-(3), we achieve, ∆% cos-.
For the simulation of NP / sample, in each step k, we set

S7
Supplementary Figure 2 presents the simulated morphology of NP under different vapor incident angles, θ (at t = 15 nm), which shows a decrease maximum deposited thickness on PS nanosphere when θ increases. We further estimate the surface area and the volume of NP with respect to the vapor incident angle as shown in Supplementary Fig. 3a. While the volume of deposited material gradually decreases when θ increases, we observe a sharp drop of surface area when θ ≥ 70°.
Volume-to-surface ratio (VSR, orange curve in Supplementary Fig. 3b) is consequently calculated using these values in Supplementary Fig. 3a. We also estimate VSR of the NP in the case it is projected into a flat surface, which show about two times large of VSR in comparison with the one on PS nanosphere (pink curve in Supplementary Fig. 3b), regardless of θ. The morphological transition of Pd NP 56 7 with increasing tPd was observed using ultra-highresolution SEM (SU-9000, Hitachi), as revealed in Supplementary Fig. 8. The morphology of NP . contains many sub-10-nm granules, and these granules cover fully the top surface of the polystyrene nanosphere. The size of the granules grows when tPd = 2 nm, and the coalescence via S12 a neck (or bridge) connection between neighboring clusters can be noticed at the thickness of tPd = 3.5 nm. A continuous film is formed at a thickness of 5 nm, as these bridge connections are successively grown (a cross-sectional SEM image and EDS elemental maps of NP can be found in Figure 1d). Once the continuous film is shaped at a thickness of 5-nm, another distinct coalescent mechanism is observed: the size of the clusters increases as the thickness increases, associated with the reduction of the cluster density.
In order to quantify these effects, we define a volume fraction, Vf, such that where Vdep is the hemispherical volume of material deposited based on a deposition of material As shown in Figure 2d in the main text, ln(PAbs/PDes), PAbs, and PDes display a size dependent behavior, where a noticeable transition occurs for tPd < 5 nm. This transition in behavior matches the transition between island-like to film-like morphologies described in Supplementary Subsection 2.2 above. Further, the size-dependent effects can be described by a size-dependent critical temperature, Tc, and an increasing effect of subsurface sites for smaller particles.
The size dependence of hysteresis in nanoparticles, which is quantitively expressed as ln(PAbs/PDes), has been comprehensively analyzed by Griessen et al. and a robust scaling law has been proposed. 4 Based on a simple lattice gas model, the hysteresis was found to only depend on the ratio of T/Tc, where Tc is the critical temperature at which the hysteresis vanishes. The full-spinodal line for hysteresis is given by: where N = R1 − T T U ⁄ and Pus and Pls are upper and lower spinodal pressures, respectively.
Griessen et al. found that the hysteresis behavior of a wide variety of Pd nanoparticles fell between the full spinodal hysteresis line given by Supplementary Equation (8) and 45% of the full spinodal hysteresis value. The size dependence of Tc was found to be well represented by: where A and B are fitting parameters and L is the size of a nanocube in nanometers. Using T = 300 K and Supplementary Equation (8) we can similarly extract Tc values as a function of tPd for the NP films and determine the scaling relationship with tPd. The results assuming a full spinodal hysteresis are presented in Supplementary Fig. 11. Tc varies from 324 -382 K with increasing size S16 and is well fit by Supplementary Equation (9) with A = 387.054 and B = 90.1308. While not presented here, the results for 45% spinodal hysteresis (0.45 is added as a multiplicative factor to the right-hand side (RHS) of Supplementary Equation (9)) show an increase Tc to 341 to 445 K, and are fit by Supplementary Equation (9)  supported assumption that hydrogen in subsurface sites does not transform to a hydride phase:

S17
where the superscripts denote the nanoparticle or bulk concentrations and the subscripts denote the concentrations at the boundaries of the α and β phases. 5 As described in the main text, the hysteresis can be explained by the thermodynamics of an open, coherent two-phase system. According to Schwarz and Khachaturyan, the absorption and desorption plateau pressures may then be calculated by: where the volume of one hydrogen atom in Pd, Ω = 2.607 Å 3 , the shear modulus of Pd, Gs = 47.7×10 9 Pa, the Poisson number ν = 0.385, the change in lattice constant, ε0 = 0.063, and Boltzmann constant k. 6 Using ! j klm = 0.008 and ! n klm = 0.0607, Supplementary Equation (11) gives ln(PAbs/PDes) = 0.62, which is in excellent agreement with the experimentally measured values for tPd ≥ 5 nm as described in the main text. By combining Supplementary Equations (10) and (11) and using the experimentally determined ln(PAbs/PDes) values, the fraction of subsurface sites, ct, can be calculated for each tPd value. The values are < 0.10 for tPd ≥ 5 nm but rapidly increase for tPd < 5 nm ( Supplementary Fig. 12). The volumes of the bulk and subsurface regions in the film-like hemispherical patchy caps (tPd ≥ 5 nm) may be respectively estimated as: where Vtot is the total volume, the radius of the PS bead template is R = 500 nm, and h is the p == q = −r − Δr + 3T # A II P II F + sta u . (16) θb and θss are the H coverages of the bulk and subsurface sites, E0 is the gain in energy during the hydride formation at R → ∞, ΔE > 0 is the energy difference between the bulk and subsurface sites, γ is the surface tension (0.2 eV/Å 2 for Pd), and Ω is the partial volume of hydrogen in the hydride phase (Ω = 2.607 Å 3 ). The last terms on the RHS of Supplementary Equations (15) and (16) are the surface tension term, and for the Pd patchy system r = R + tPd, with R = 500 nm. At equilibrium, the chemical potential of H2 molecules in the gas phase is:

S20
where m is the mass of a H2 molecule and ℏ is the reduced Planck's constant. Note that this expression takes only the translation partition function into account, and the minor rotational and vibrational contributions are ignored. The average coverage of the bulk and subsurface sites is given by: By setting 〈 〉 = 0.5, using the value for h determined above for (h = 0.4nm), selecting ΔE = 2.6 kJ/mol H, and using the experimentally measured absorption plateau pressures at T = 300 K, On the other hand, θss is higher than expected (θss × 0.607 = 0.44 H/Pd, typically), and thus could indicate higher H capacity in surface shell sites facilitated by the polymer substrate. Similar effects were seen in Pd nanocubes covered with a metal organic framework. 8 It is interesting to note that the subsurface site effect is more pronounced in the Pd patchy particles than it is in many other Pd nanoparticles. This is because the surface curvature changes much more slowly than the volume to surface ratio in the Pd patches. Hence, Wadell et al. observed a relatively constant enthalpy of hydride formation for decreasing particle size in their spherical nanoparticles.
In addition, we note that the hydrogen absorption and desorption isotherm pressures above the systems. 13 Note that VSR will go as length L or radius r in nanocubes and nanoparticles, which are traditionally used in power law fittings. In the Pd hemispherical caps on 500 nm PS beads VSR will be close to tPd for small thicknesses (< 20 nm), but will begin to deviate for increasing thicknesses. The maximum optical change in the NP sample is much greater than that of the control Pd thin film sample ( Supplementary Fig. 14). This can be attributed to diffraction within the hexagonal Pd patchy arrays, which does not occur in a thin film. Transmitted light at certain wavelengths is diffracted, partially localized inside the PS nanosphere causing the so-called partially-localized surface plasmon resonance (LSPR), and then is amplified through interactions with the Pd hemisphere cap. 14,15 This LSPR is suppressed when Pd turns to Pd hydrides, inducing significant redistribution of the local electric field. 14 This enhanced light-material interaction emerges in the far-field as local extrema on ∆T spectra. To confirm this hypothesis, FDTD calculations were performed using NP morphology generated by a home-built MATLAB program, 1 and the calculated results are in excellent agreement with the experimental transmission spectra  Fig. 15c).
We observe an partial-localized enhanced electric field spot located under the Pd NP at ∆T(λ) peak (at green circle, Supplementary Fig. 15c), and a whispering-gallery modes inside nanosphere at ∆T(λ) dip (at red circle, Supplementary Fig. 15c). On the other hand, no enhanced electric field can be observed when it is out of resonance (at yellow circle, Supplementary Fig. 15c). are extracted at normalized ∆T = 0.5 (as denoted in Supplementary Fig. 16b). (c) PAbs, PDes, and (d)

S4. Plateau pressures extraction
ln(PAbs/PDes) extracted at different wavelength positions. We observe a slight variation of these values at different wavelengths; however, the general thickness-dependent trend is conserved.

S5. Sensor accuracy calculations
The sensor accuracy (A) is calculated by the following equation: 16 where PAbs and PDes are the pressures reading (in µbar) during hydrogen absorption and desorption, respectively.

S6. The calculation of void coverage
Supplementary Figure 17. The calculation configuration for void coverage.
The void coverage percentage V (black area) between PS nanosphere (radius of R) can be calculated as: In general, incorporating Ag or Au into NPs does not change the overall shape of ∆T(λ) (at † ‡ h = 1000 mbar), but it reduces the absolute value of ∆T(λ) ~50 % in comparison to that of pure Pd NP with the same deposited thickness (Supplementary Fig. 19). The drop of ∆T(λ) magnitude is more significant than the amount of Pd atoms replaced by Ag or Au (20%), which can be explained by the reduction in the limiting solubility of H in the PdAg or PdAu system. This reduction is due to a decreasing number of available electron states in the d-band of the Pd electronic structure, which is induced by the introduction of Ag/Au atoms. 17 In contrast, at very low pressures ( † ‡ h < 1 mbar), the Pd80Ag20 and Pd80Au20 alloy NP sensor shows about 3-and 1.8-times enhancement in sensitivity in comparison to that of pure Pd NP ( Supplementary Fig. 19c), respectively, which Shaded areas denote the periods where the sensor is exposed to hydrogen.

S37
Shaded areas denote the periods where the sensor is exposed to hydrogen. Note that the ∆T/T response data presented in Fig. 5b  The Pd80Co20 NP sample shows a good stability without the sign of degradation, upon >300 of (de)hydrogenation cycles. The first 100 cycles (1/1 minute of loading/unloading with 2% H2 in synthetic gas) of 3-weeks old and 10-months old sample are summarized in Supplementary Fig.   27a. While a noticeable reduction of sensor signal due to aging can be seen in 10-months old sample ( Supplementary Fig. 27b), upon cycling, the signal is recovering back to that of a fresh sample.
The sign of the aging effect can be seen by the degraded performances of the sensor over the times. In particular, the response time t90 (at † ‡ h = 1 -100 mbar) increases about ~2 times and ~3 times after a period of 4-weeks and 10-months, respectively ( Supplementary Fig. 27c). In addition, the long-term ∆T/T responses show a significant reduction both in vacuum mode and flow mode (Supplementary Figs. 27d and e), just after 4-5 weeks in air. We ascribed the degradation of the sensor to a small trace amount of poison gases exist in the ambient air, (e.g. CO) (see the deactivation test with CH4, CO2, and CO in Figure 6). However, we note that the response time and LOD of the sample still are <2.5 s (at 1 mbar) and <10 ppm, respectively, over a >10-month period. The degradation of the Pd80Co20 NP sample upon long-term storage in air can affect to the accuracy, sensitivity, and response time of the sensor. Hence, we look for a solution to mitigate this process. Inspired by the work of Nugroho et al., 18 we coated the Pd80Co20 NP sample with S38 a ~50-nm layer of polymethyl methacrylate (PMMA) (by spin-coating of PMMA dissolved in acetone, more details can be found in Methods section), which has been demonstrated to effectively block the poisonous species. In this test, we store the PMMA-coated and uncoated samples in an identical condition and measure their sensing metrics.

S39
The sensing performances of Pd80Co20 NP /PMMA sensor are summarized in Supplementary   Fig. 28. After storing in air for 6-weeks and underwent >200 cycles of (de)hydrogenation with 2% H2, we observe very little variances in response time over the pressure range of 1-100 mbar (t90 are <1.5 s (at 1 mbar)) and insignificant reduction of sensor signal upon exposure to pulses of very low H2 concentration (100 -2.5 ppm). Clearly, the degradation of the sensor performance is significantly slowed-down with a polymer coating layer. S42 S12. Sensing metrics of state-of-art optical hydrogen sensor (at room-temperature) Pd strip -20 10 n.a. 22 PdY film 6 -1000 n.a. 23 Pd/SiO2/Au 3 -5000 n.a. 24 Pd/Au film 4.5 --n.a. 25 Supplementary Table 1. Sensing metrics of state-of-art optical hydrogen sensor (at room-temperature). *not addressed.