Chia-Jung
Chung‡
,
Chinmay
Nivargi‡
and
Bruce
Clemens
*
Department of Materials Science and Engineering, Stanford University, Stanford, CA - 94305, USA. E-mail: bmc@stanford.edu
First published on 29th September 2015
Magnesium and Mg-based material systems are attractive candidates for hydrogen storage but limited by unsuitable thermodynamic and kinetic properties. In particular, the kinetics are too slow at room temperature and atmospheric pressure. To study the hydride formation kinetics in a controlled way, we have designed a unique ‘nanoportal’ structure of Pd nanoparticles deposited on epitaxial Mg thin films, through which the hydride will nucleate only under Pd nanoparticles. We propose a growth mechanism for the hydrogenation reaction in the nanoportal structure, which is supported by scanning electron microscopy (SEM) images of hydrogenated samples exhibiting consistent results. Interestingly, the grain boundaries of Mg films play an important role in hydride nucleation and growth processes. Kinetic modeling based on the Johnson–Mehl–Avrami–Kolmogorov (JMAK) formalism seems to agree with the two-dimensional nucleation and growth mechanism hypothesized and the overall reaction rate is limited by hydrogen flux through the interface between the Pd nanoparticle and the underlying Mg film. The fact that in our structure Mg can be transformed completely into MgH2 with only a small percentage of Pd nanoparticles offers possibilities for future on-board storage applications.
The Mg hydrogenation reaction is also sluggish on its own and is further restricted due to the presence of a native oxide on the Mg surface. High temperatures of at least 400 °C are needed to crack the MgO layer and dissociate the hydrogen, and even then the absorption/desorption of hydrogen takes place over a period of hours.9,10 The hydrogenation is often catalyzed at lower temperatures by Pd deposited on its surface, in contact with the pristine Mg, enabling dissociation of the H2 molecules into H atoms which then diffuse through and react with Mg. In fact, in our experiments, we did not see a significant rate of hydrogen sorption at room temperature without the presence of the Pd catalyst. The nucleation occurs at Pd–Mg (or the catalyst/additive-Mg interface) followed by growth due to the diffusion of H atoms to the growth front. After nucleation, the kinetics are limited by the slow diffusion of H atoms to the reaction front, since the diffusion of H atoms through the already-grown magnesium hydride is eight to ten orders of magnitude slower than diffusion through a metal.11,12 The formation of a ‘blocking’ layer of MgH2 has been described previously in the literature and hence the diffusion length scale for complete reaction needs to be kept below 30–50 μm.1,13–15
The hydrogenation kinetics can be substantially improved by reducing the diffusion length scale using nanosized materials.10,16–18 The kinetics are also enhanced upon addition of the catalyst or alloying additives. Transition metals or their oxides are the most investigated. A good overview of the research on Mg based materials for hydrogen storage was provided in previous reviews.19,20 However, the connection between the nanostructure and its effect on the kinetic properties of these materials is still unclear. Different studies use a variety of materials, catalysts and geometries, making identification of the underlying mechanisms very difficult.
In order to critically examine the effects of the nanostructure on the kinetics of nucleation and growth, we have designed a Pd nanoparticle ‘nanoportal’ structure, illustrated in Fig. 1(a). This unique structure enables a physical and temporal decoupling of nucleation and growth processes, by allowing the nucleation to occur only under the Pd nanoparticles, followed by growth due to hydrogen flux through the Pd catalyst nanoparticle ‘portals’. As a result, it can be employed to study hydrogenation reactions under a wide variety of carefully controllable experimental conditions. This has potential to provide great insight into the underlying reaction mechanisms.
The nanoportal structure also allows the examination of the reaction using Scanning Electron Microscopy (SEM) and an optical transmission technique based on hydrogenography,21 since the catalyst nanoparticles do not obscure the observation of the underlying phase transformation. We propose a reaction mechanism for this structure and then correlate a kinetic model that we have developed based on the Johnson–Mehl–Avrami–Kolmogorov (JMAK) formalism22 with the experimental observations. The results suggest that the overall reaction rate depends on the flux of hydrogen atoms through the interface between the Pd catalyst nanoparticle and the underlying Mg film.
We found that, even for the thin Mg films studied here, Pd nanoparticles representing a mass fraction of only a few percent can be effective catalysts, increasing reaction rates by orders of magnitude over bare Mg. The investigation of reaction mechanisms with this novel nanostructure can pave the way to rational designs of hydrogen storage systems with better storage properties for future on-board applications.
Localizing the nucleation sites results in an ability to study reaction kinetics with unprecedented control. By controlling the density of nanoportals we can vary the nucleation density and examine new growth behaviors. For example, when the distance between nucleation sites is large compared to the film thickness (true for all the samples discussed here) the hydride growth will occur as two-dimensional lateral growth for most of the reaction processes, since vertical growth will rapidly be limited by the film thickness (Fig. 1(b) and (c)). By following this growth process we can examine the limiting mechanisms as well as explore nucleation behavior.
Fig. 3 shows SEM images of the hydrogenated Mg films grown on sapphire and oxidized Si substrates, respectively, with an areal Pd nanoparticle coverage of 1.4%. Clearly the structural difference in the parent Mg phase, as shown by X-ray diffraction, altered both the extent of the hydrogenation reaction and the morphology of the hydride. For the epitaxial Mg films, the hydride regions were nearly circular, while showing some overlap of neighboring regions. These phenomena indicate that the hydride reaction rate was roughly independent of the in-plane crystallographic direction within a grain. However, the hydride regions on the non-epitaxial Mg films were more irregular in shape and much smaller. It appears that the shape of the hydride region in the non-epitaxial Mg films was influenced by the underlying Mg grain structure, coupled with a variation in the difficulty with which the growing hydride was able to cross the different grain boundaries between adjacent grains. The shape of the hydride regions appears to mimic the grain structure of the parent Mg phase, indicating that the hydride growth is impaired during transit across grain boundaries. Further examination can explore further the connection between this growth impediment and grain boundary orientation, but it is clear that the small angle grain boundaries present in the sample grown on sapphire offer much less interference to hydride growth.
We note that for both epitaxial and non-epitaxial samples, the spacing between the hydride regions is much larger than the spacing between the Pd nanoparticles, and that there are Pd nanoparticles over unreacted Mg regions of the film, or more than one Pd nanoparticle on each nucleated hydride patch. This implies that only a fraction of the Pd nanoparticles are effective in nucleating a hydride region. We suspect that the failure of many nanoparticles to result in hydride formation is due to the lack of direct contact between the Pd nanoparticle and the Mg film, perhaps due to oxidation in the imperfect vacuum of the deposition environment, or oxidation of the interface subsequent to nanoparticle deposition and exposure to air.
It is also notable that the hydride regions in a given film at a given stage of reaction are approximately the same size, which indicates that each hydride region begins to grow at approximately the same time. This result indicates that nucleation takes place early in the reaction process, and that once the effective nucleation sites are used up, no further nucleation occurs. Finally, we note that the spacing between hydride regions is much larger than the film thickness, so that much of the transformation occurs via two-dimensional growth in the plane of the film after the hydride spans the thickness of the film, which is consistent with our predicted growth mechanism.
To extract the time-dependence of the hydride growth behavior we used SEM to observe the size of the hydride patches in samples with the same nominal nanoportal areal density (∼1.7%) but exposed to hydrogen for varying reaction times. The average radius (extracted by image analysis using ImageJ25) as a function of reaction time is shown in Fig. 4. We see that the growth rate dr/dt decreases greatly with time. Fitting to a growth law r ∼ tm leads to unsatisfactory fits and to a value of m ∼ 0.3, which is incompatible with growth limited by either diffusion through the growing hydride or by flux through the nanoparticle–film interface,15 both of which would produce a r ∼ t1/2 behavior for two-dimensional growth (see ESI†). At longer times, the observed growth rate decreases more rapidly than the r ∼ t0.3 of the fitted function, leading to the conclusion that the effectiveness of the nanoportals might be decreasing with time. We speculate that this decrease may be due to further oxidation of the nanoparticle–film interface, resulting in a decrease in the effective nanoportal area. This possibility is discussed more fully below. Note that these observations of nanoparticle diameter are taken in the time and nucleation density regime before coalescence occurs, so the decreased growth rate is not due to the impingement of growing hydride regions.
To explore the effect of oxidation of a nanoportal on the hydride growth rate, we considered both the interfacial and diffusive flux limited encroachment of the oxide under the nanoportal. This is illustrated schematically in Fig. 5. This leads to a simple expression for the nanoportal area and hence the flux of the H atoms through the nanoportals as a function of time, which in turn can be used to obtain an expression of the dependence of the radius of an individual hydride patch as a function of time. Balancing the nanoportal flux and the hydride growth in a time increment dt for mass conservation, we find:
jHAnpdt = cH(2πrh)dr |
![]() | ||
Fig. 5 The portal pinches off under the Pd nanoparticle. On the right is an exaggerated top view showing a decreased area of the portal and hence a lowered H flux. |
Next we consider growth of the nuclei as a function of time. A nucleus that begins growth at time τ and grows as a two-dimensional disk will have area at time t given by:
Aτ(t) = c[(t − τ)1/2]2 = c(t − τ) |
If we add up all the area of growing hydride regions, ignoring impingement and counting phantom nuclei that are present in already transformed regions we find the extended area:
A(t) = A0(1 − e−Ae(t)/A0) |
Comparing this model with the observed transformed fraction f = A(t)/A0 behavior we can investigate the nucleation behavior and growth limiting mechanisms. To explore nucleation behavior we first fit the short-time (t ≤ 700 s) experimental data using a power law growth behavior (r ∼ tm). We find relatively good agreement with the optical behavior as shown in Fig. 7, with an exponent of 0.2 ≤ m ≤ 0.5 which is roughly consistent with the fits we obtained for the radius behavior discussed above. We find that we are unable to extract meaningful nucleation times τN for samples with low nanoparticle density, and that the nucleation time we extract for the sample with high density is short (∼54 s), but comparable to the reaction time of around 100 seconds. This indicates that the nucleation is complete in the low nanoparticle density samples before significant reaction takes place, while for the high nanoparticle density samples there is nucleation and growth occurring during the reaction. The quality of the fit decreases if we include longer reaction times, indicating that the power law growth behavior is not good for a long time. This is consistent with the hydride radius behavior discussed above and with the idea that the nanoportals become less effective with time, perhaps due to pinching off by MgO growth under the nanoportal.
![]() | ||
Fig. 7 Model fits to observed short-time behavior using power-law growth behavior. The solid lines are fits to the data and the best fit to the parameter m is shown. |
To explore this we incorporated the hydride growth behavior associated with the decreasing nanoportal area due to the oxide encroachment discussed above (see ESI,† for the model). The oxide encroachment was modeled as both interfacial as well as diffusive flux limited growth, leading to a decreasing area of the nanoportal, and hence a slowdown of the growth of the hydride patch from the expected r ∼ t1/2 behavior. The resulting fits for the interfacial flux limited model are shown in Fig. 8. As can be seen, the model is able to describe the behavior fairly adequately over the entire reaction time. The diffusive flux limited model shows somewhat similar behavior. This model was considered only for the low coverage samples for which the oxide encroachment before reaction completion was significant. The fast reacting sample essentially reacted fully before encroachment became appreciable.
![]() | ||
Fig. 8 Model fits to observed behavior incorporating portal pinch-off and an additional leak flux through MgO, for the low coverage samples. The solid lines are fits to the data. |
Using this approach, it was again found that the nucleation time is short compared to the reaction times, and hence nucleation is effectively instantaneous. This results in hydride patches of about the same size, as observed in the SEM. The time of nanoportal pinch-off varies across samples with different nanoportal densities. It is also presumably affected by the amount of time it was exposed to air before hydrogenation and the specific variation of nanoparticle sizes across samples, which is hard to completely account for in the modeling. However, for a similar nanoparticle density, it was found that the time for the nanoportals to completely pinch-off was comparable for the fits to the transmission measurements (∼350−1340 s depending on interfacial or diffusive flux limitation) and the radius measurements (∼516−950 s). These times are very sensitive to the initial radius of the nanoportals, which, of course, have a size distribution that is not included in the modeling.
Although the portals eventually pinch off, the reaction is still seen to proceed slowly, indicating that the MgO layer on top of Mg and under the nanoparticle might be slightly permeable to H atoms dissociated by Pd, through cracks or defects. This is likelier after appreciable nucleation and growth, as opposed to at the beginning of the reaction where there is no significant nucleation yet. This is because of the growing hydride having a 32% larger volume,8 compared to Mg, that might contribute to a decreased integrity of the oxide layer on top. A constant flux term was incorporated into the fits to account for this behavior, leading to much better agreement at long and short times, as shown. A better engineering of the Mg–Pd nanoparticle interface could help avoid this slowdown and pinch-off. A potential way would be to accelerate the nanoparticles towards the sample by means of a voltage bias during deposition so as to have a more intimate contact with the underlying Mg film – to prevent oxidation or cause more Mg–Pd intermixing. Investigations to verify this hypothesis are ongoing in our lab.
The amount of Pd used in this study is much less than the conventional Pd/Mg thin film structures. With only a small percentage of Pd nanoparticles, the Mg film can be fully transformed into hydride. The separation between the nanoparticles is less than 30–50 μm of a blocking layer, thus enabling full conversion. Therefore, the storage capacity is significantly increased with this nanoportal structure compared to Pd thin film catalyst structures. This makes it promising for future on-board storage applications. We hope to do further work to apply this novel structure to understand and design other hydrogen storage systems.
Footnotes |
† Electronic supplementary information (ESI) available. See DOI: 10.1039/c5cp04515k |
‡ These authors contributed equally to this work. |
This journal is © the Owner Societies 2015 |