Passivation of Co/Al2O3 Catalyst by Atomic Layer Deposition to Reduce Deactivation in the Fischer–Tropsch Synthesis

The present work explores the technical feasibility of passivating a Co/γ-Al2O3 catalyst by atomic layer deposition (ALD) to reduce deactivation rate during Fischer–Tropsch synthesis (FTS). Three samples of the reference catalyst were passivated using different numbers of ALD cycles (3, 6 and 10). Characterization results revealed that a shell of the passivating agent (Al2O3) grew around catalyst particles. This shell did not affect the properties of passivated samples below 10 cycles, in which catalyst reduction was hindered. Catalytic tests at 50% CO conversion evidenced that 3 and 6 ALD cycles increased catalyst stability without significantly affecting the catalytic performance, whereas 10 cycles caused blockage of the active phase that led to a strong decrease of catalytic activity. Catalyst deactivation modelling and tests at 60% CO conversion served to conclude that 3 to 6 ALD cycles reduced Co/γ-Al2O3 deactivation, so that the technical feasibility of this technique was proven in FTS.


Introduction
Nowadays, Fischer-Tropsch synthesis (FTS) has regained interest motivated by the development of two concepts. On the one hand, the biorefinery concept, which is focused on producing chemicals and fuels from biomass sources; on the other hand, carbon capture and utilization (CCU), which is focused on producing these compounds from CO 2 as feedstock [1]. The aim of FTS is to convert syngas, which is produced from biomass gasification or from the reverse water-gas shift reaction, into hydrocarbons through heterogeneous catalysts, usually based on Fe or Co. Fe-based catalysts are more suitable when light hydrocarbons and alcohols are desired, whereas Co-based ones are more selective for middle distillates and waxes.
Despite being discovered more than a century ago, FTS is still facing some economic and technical challenges, which hinder its consolidation as a viable alternative to producing synthetic fuels from renewable sources. In this regard, catalyst deactivation is a major technical issue. According to literature, the main causes of cobalt catalyst deactivation could be: poisoning due to the presence of S, N and alkali metals, reoxidation of cobalt species, coke deposition, sintering of cobalt crystallites, metal-support solid state reactions and particle attrition. Some of these phenomena, such as poisoning by sulfur compounds, can agent. The main conclusion was that ALD depositions reduced the average crystallite size, as they prevented catalyst sintering. Five ALD cycles increased activity and methanation selectivity, whereas at 20 cycles the properties were similar to those of the unmodified catalyst. The authors recommended further work in order to have a broader view of this potential catalyst passivation technique.
The aim of this work is to study the effect of ALD Al 2 O 3 overcoating as a passivation technique on the morphology and the catalytic performance of Co/Al 2 O 3 catalysts in FTS. In the present case, a Co/γ-Al 2 O 3 microcatalyst was used as a reference in order to propose a more stable catalyst for its potential application in the incipient microreactor technology. The relationship between long-term catalytic activity and catalyst properties is discussed. Table 1 lists the main characterization results obtained in the overcoated catalyst samples. At first sight, it is observed that the incorporation of Al 2 O 3 on the catalyst by ALD cycles did not lead to significant morphology changes, neither in textural properties nor in the Co crystallite size. However, important differences in the reduction behavior of cobalt particles were inferred.  Figure 1 shows the nitrogen adsorption-desorption isotherms and the mesopore distribution of the support-γ-Al 2 O 3 -and the catalysts evaluated in the present work. All samples showed a type IV isotherm, according to IUPAC classification, which is typical of mesoporous materials, without noticeable nitrogen adsorption in the micropore range (P/P 0 ≤ 0.03) and a significant increase in the mesopore range. The hysteresis loop observed in all samples (Type H4, according to IUPAC classification) [23] indicated the presence of slit-type pore and was due to capillary condensation in the mesopores. Results evidenced that porosity was induced by the support, rather than by cobalt nanoparticles.

Catalyst Characterization
On the one hand, it is observed that nitrogen adsorption of the cobalt catalysts (161-163 m 2 /g) was significantly lower than that of the support (216 m 2 /g), which evidenced that cobalt particles partially blocked the pores of the support during impregnation procedure, as reported elsewhere [24]. This pore blockage is confirmed by BJH desorption results, also revealing that cobalt particles were deposited uniformly on the support mesopores. On the other hand, there were no differences between BET surface area and pore volume of samples REF and ALD-coated ones, which revealed that atomic layer deposition did not lead to noticeable changes in the porosity of the catalyst.
XRD diffractograms of all samples are depicted in Figure 2. All catalysts showed a main narrow diffraction reflection, identified as the crystalline phase Co 3   On the one hand, it is observed that nitrogen adsorption of the cobalt catalysts (161-163 m 2 /g) was significantly lower than that of the support (216 m 2 /g), which evidenced that cobalt particles partially blocked the pores of the support during impregnation procedure, as reported elsewhere [24]. This pore blockage is confirmed by BJH desorption results, also revealing that cobalt particles were deposited uniformly on the support mesopores. On the other hand, there were no differences between BET surface area and pore volume of samples REF and ALD-coated ones, which revealed that atomic layer deposition did not lead to noticeable changes in the porosity of the catalyst.
XRD diffractograms of all samples are depicted in Figure 2. All catalysts showed a main narrow diffraction reflection, identified as the crystalline phase Co3O4, at 36.84° (ICDD Card No. 00-042-1467). Small reflections of this crystalline phase, at ca. 31.32°, 45.08°, 59.58° and 65.44°, were also detected. No more reflections were observed except some broad ones which were coincident to those of the Al2O3 support. Diffractograms of all catalysts were practically identical to each other, which revealed that ALD deposition carried out at 300 °C did not affect cobalt-alumina interactions. Indeed, cobalt crystallite size of all catalysts was similar (Table 1), taking into account the uncertainty associated with this calculation.   On the one hand, it is observed that nitrogen adsorption of the cobalt catalysts (161-163 m 2 /g) was significantly lower than that of the support (216 m 2 /g), which evidenced that cobalt particles partially blocked the pores of the support during impregnation procedure, as reported elsewhere [24]. This pore blockage is confirmed by BJH desorption results, also revealing that cobalt particles were deposited uniformly on the support mesopores. On the other hand, there were no differences between BET surface area and pore volume of samples REF and ALD-coated ones, which revealed that atomic layer deposition did not lead to noticeable changes in the porosity of the catalyst.
XRD diffractograms of all samples are depicted in Figure 2. All catalysts showed a main narrow diffraction reflection, identified as the crystalline phase Co3O4, at 36.84° (ICDD Card No. 00-042-1467). Small reflections of this crystalline phase, at ca. 31.32°, 45.08°, 59.58° and 65.44°, were also detected. No more reflections were observed except some broad ones which were coincident to those of the Al2O3 support. Diffractograms of all catalysts were practically identical to each other, which revealed that ALD deposition carried out at 300 °C did not affect cobalt-alumina interactions. Indeed, cobalt crystallite size of all catalysts was similar (Table 1), taking into account the uncertainty associated with this calculation.  TPR profiles are depicted in Figure 3. In general, two main peaks were observed in all samples. The first hydrogen consumption (268-283 • C) was due to the reduction of Co 3 O 4 to CoO, followed by the reduction of CoO to Co 0 (481-530 • C) [25]. It is worth noting that sample REF showed a small shoulder at 202 • C, due to the presence of NO x from residual nitrates on this catalyst [26]. As ALD procedure is carried out, residual nitrates on ALD-passivated catalysts vanished during heating. Finally, samples 6ALD-REF and 10ALD-REF presented a small shoulder (620-642 • C) which can be linked to the formation of cobalt aluminate species.
Co3O4 to CoO, followed by the reduction of CoO to Co 0 (481-530 °C) [25]. It is worth noting that sample REF showed a small shoulder at 202 °C, due to the presence of NOx from residual nitrates on this catalyst [26]. As ALD procedure is carried out, residual nitrates on ALD-passivated catalysts vanished during heating. Finally, samples 6ALD-REF and 10ALD-REF presented a small shoulder (620-642 °C) which can be linked to the formation of cobalt aluminate species. Among samples, it is observed that 3ALD-REF did not show important differences in respect to REF (indeed, the peaks showed a slight shift to lower temperatures), whereas the peaks of samples 6ALD-REF and 10ALD-REF shifted to higher temperatures, especially the peak associated with the final reduction (CoO to Co 0 ). Temperature shift to higher values indicates that cobalt species in these samples were more difficult to be reduced, which was directly related to the number of ALD cycles applied on them. In this way, it can be concluded that the higher the number of cycles, the harder the sample is to be reduced.
Two conclusions can be drawn from TPR profile analysis. On the one hand, overcoating by ALD may affect the diffusion of gas species (in this case H2) up to cobalt particles, leading to lower reduction degree at a certain overcoating level (10ALD-REF). On the other hand, during reduction the Al2O3 introduced by ALD may interact with cobalt particles to form irreducible species (cobalt aluminates) that are hard to reduce, as reported elsewhere [20]. In order to obtain the FT cobalt active phase (Co 0 ) and avoid the formation of these irreducible cobalt aluminates, 380 °C was selected as the maximum reduction temperature. At this temperature, CoO reduction rate (i.e., the slope of the TPR curve) of samples REF, 3ALD-REF and 6ALD-REF was higher than that of 10ALD-REF, which was closely related to O2 titration results (see hereafter). As cobalt reduction rate was relatively low at this temperature, catalyst reduction procedure should last several hours. Therefore, 380 o C and 16 h were selected as reduction conditions prior to catalytic tests. Detailed reduction conditions are described in the experimental section.
The effect of ALD cycles on reduction properties was partially corroborated through the evaluation of the extent of cobalt reduction (Table 1). On one hand, similar extent of Among samples, it is observed that 3ALD-REF did not show important differences in respect to REF (indeed, the peaks showed a slight shift to lower temperatures), whereas the peaks of samples 6ALD-REF and 10ALD-REF shifted to higher temperatures, especially the peak associated with the final reduction (CoO to Co 0 ). Temperature shift to higher values indicates that cobalt species in these samples were more difficult to be reduced, which was directly related to the number of ALD cycles applied on them. In this way, it can be concluded that the higher the number of cycles, the harder the sample is to be reduced.
Two conclusions can be drawn from TPR profile analysis. On the one hand, overcoating by ALD may affect the diffusion of gas species (in this case H 2 ) up to cobalt particles, leading to lower reduction degree at a certain overcoating level (10ALD-REF). On the other hand, during reduction the Al 2 O 3 introduced by ALD may interact with cobalt particles to form irreducible species (cobalt aluminates) that are hard to reduce, as reported elsewhere [20]. In order to obtain the FT cobalt active phase (Co 0 ) and avoid the formation of these irreducible cobalt aluminates, 380 • C was selected as the maximum reduction temperature. At this temperature, CoO reduction rate (i.e., the slope of the TPR curve) of samples REF, 3ALD-REF and 6ALD-REF was higher than that of 10ALD-REF, which was closely related to O 2 titration results (see hereafter). As cobalt reduction rate was relatively low at this temperature, catalyst reduction procedure should last several hours. Therefore, 380 • C and 16 h were selected as reduction conditions prior to catalytic tests. Detailed reduction conditions are described in the experimental section.
The effect of ALD cycles on reduction properties was partially corroborated through the evaluation of the extent of cobalt reduction (Table 1) In order to have a clear view of the effect of ALD passivation, those images with exceptionally bigger Co agglomerates have been selected. As noticed, an Al2O3 shell grew surrounding the Co polycrystalline nanostructures in ALD-coated samples. Therefore, the passivation of samples led to an Al2O3 shell nanoflake-like morphology. Analogously, high-angle annular dark-field scanning TEM images (STEM-HAADF, Figure 5) highlighted the presence of Al2O3 shell on ALD-coated samples. Images were reported in violet color palette in order to differentiate the species observed from the image contrast and therefore local color. Heavier atoms, in this case Co3O4, appear brighter than lighter compound, Al2O3. From this visual contrast it is possible to quantitatively recognize that the Al2O3 shell in sample 10ALD-REF fully covered the Co core. As expected, the higher the number of ALD cycles, the higher the shell thickness.  Analogously, high-angle annular dark-field scanning TEM images (STEM-HAADF, Figure 5) highlighted the presence of Al 2 O 3 shell on ALD-coated samples. Images were reported in violet color palette in order to differentiate the species observed from the image contrast and therefore local color. Heavier atoms, in this case Co 3 O 4 , appear brighter than lighter compound, Al 2 O 3 . From this visual contrast it is possible to quantitatively recognize that the Al 2 O 3 shell in sample 10ALD-REF fully covered the Co core. As expected, the higher the number of ALD cycles, the higher the shell thickness.
attained in 10ALD-REF dropped sharply up to 11.1%. Figure 4 shows low magnification TEM micrographs of samples REF, 3ALD-REF and 10ALD-REF, evidencing the general morphology of the catalyst samples. In order to have a clear view of the effect of ALD passivation, those images with exceptionally bigger Co agglomerates have been selected. As noticed, an Al2O3 shell grew surrounding the Co polycrystalline nanostructures in ALD-coated samples. Therefore, the passivation of samples led to an Al2O3 shell nanoflake-like morphology. Analogously, high-angle annular dark-field scanning TEM images (STEM-HAADF, Figure 5) highlighted the presence of Al2O3 shell on ALD-coated samples. Images were reported in violet color palette in order to differentiate the species observed from the image contrast and therefore local color. Heavier atoms, in this case Co3O4, appear brighter than lighter compound, Al2O3. From this visual contrast it is possible to quantitatively recognize that the Al2O3 shell in sample 10ALD-REF fully covered the Co core. As expected, the higher the number of ALD cycles, the higher the shell thickness.  in Figure 6, where elemental mapping evidences the local distribution of the different atomic species. In the same figure, relative composition analysis to evaluate Al, Co and O relative composition was also reported. For the elemental analysis, O K-edge at 532 eV (blue), Co L-edge at 779 eV (green) and Al K-edge at 1560 eV (red) were used. Although REF presented some Al2O3, the content was much lower than that presented by ALD-passivated samples, and can be linked to the presence of Al2O3 support traces. In any case, both the amount of Al2O3 and the layout around the cobalt particles, evidenced by the relative percentage of this species, revealed, on the one hand, how cobalt particles of ALDpassivated samples were covered by an Al2O3 shell, as well as how the number of ALD cycles affected the thickness of this shell.  Usually, the differences in reduction behavior of Co/Al 2 O 3 catalysts are closely related to important changes in textural properties and/or cobalt particle size [27]. However, nitrogen physisorption and XRD results showed no difference among the samples used in this work, so that it is possible to conclude that ALD passivation only affected the reduction properties of these catalysts by the presence of a thin coating layer of ALD on the top of cobalt nanoparticles, while the rest of the properties were kept unchanged. obtained, which corresponded to hydrocarbons of longer chains (C 5+ ). The maximum relative error observed between experimental species concentrations at equivalent timeon-stream (TOS) was less than 2%. Table 2 compares the activity results of all samples at two different times: 15-17 h (the time at which, once GHSV was adjusted, steady state was attained) and 87 h (otherwise the end of the experiment). In this work, activity was quantified as a function of GHSV and syngas conversion (Section 3, Equations (9) and (10)). Table 2. Experimental results (reaction conditions: 20 barg, 230 • C, 8400 NmL/g·h, H 2 /CO = 2/1 (mol/mol)).

Sample
Time (h) X CO (%) Selectivity (%) First of all, it is observed that, in all cases except 10ALD-REF, CO conversion experienced a significant decrease between the beginning (40.1-43.5%) and the end (34.8-36.7%) of the reaction, which confirmed catalytic activity loss in the respective catalysts. In contrast, sample 10ALD-REF showed initially lower CO conversion than REF (28.1 vs. 43.5%), and then experienced a slight increase up to 30.2%.
As for selectivity, all samples showed, on the one hand, important CH 4 selectivity values, being caused by undesired CO methanation reaction. In particular, high CH 4 selectivity (32.1%) was observed in sample 10ALD-REF, which consequently led to low selectivity to C 5+ (57.2%). As reaction went by, CH 4 selectivity values decreased in all cases, this decrease being sharp in sample 10ALD-REF. This CH 4 selectivity decrease observed in sample 10ALD-REF, together with the increase in CO conversion, suggested that the active phase of this catalyst experienced changes during the time on stream, as explained hereafter. On the other hand, CO 2 selectivity, caused by reverse water gas shift reaction (RWGS), whose value was lower than 1%. As for the selectivity, if results are evaluated in terms of yield to liquid and wax products (Y C5+ ), it is observed that, at the beginning of the reaction, the values of ALDpassivated samples were lower than those of the reference, but then they reached practically the same value (6ALD-REF) or even higher (3ALD-REF), so that it is possible to conclude that ALD overcoating also favored the catalytic selectivity to the desired products.
In order to have a more detailed view about catalyst activity, Figure 7 depicts the evolution of activity of all catalysts studied, which sheds light about their behavior with time on stream (TOS). On the one hand, samples REF, 3ALD-REF and 6ALD-REF exhib-ited similar activity profiles, showing an initial sharp activity drop, followed by a more gradual decline in activity. On the other hand, activity of sample 10ALD-REF increased at the beginning of the reaction, followed by stabilization at ca. 30 h of TOS. Taking into account TPR and O 2 titration results, it is possible to conclude that, at the beginning, cobalt particles of 10ALD-REF were reduced by part of the syngas fed into the reactor, which consequently led to an increase of the catalytic activity. Activation of cobalt catalysts in syngas atmosphere has been reported elsewhere [28]. The high stability observed in this catalyst after 30 h of TOS suggests that, in general, ALD passivation conferred stability to the catalyst in all cases. Moreover, it is observed that, the higher the number of ALD cycles, the more stable the catalyst was.
As for the selectivity, if results are evaluated in terms of yield to liquid and wax products (YC5+), it is observed that, at the beginning of the reaction, the values of ALD-passivated samples were lower than those of the reference, but then they reached practically the same value (6ALD-REF) or even higher (3ALD-REF), so that it is possible to conclude that ALD overcoating also favored the catalytic selectivity to the desired products.
In order to have a more detailed view about catalyst activity, Figure 7 depicts the evolution of activity of all catalysts studied, which sheds light about their behavior with time on stream (TOS). On the one hand, samples REF, 3ALD-REF and 6ALD-REF exhib-ited similar activity profiles, showing an initial sharp activity drop, followed by a more gradual decline in activity. On the other hand, activity of sample 10ALD-REF increased at the beginning of the reaction, followed by stabilization at ca. 30 h of TOS. Taking into account TPR and O2 titration results, it is possible to conclude that, at the beginning, cobalt particles of 10ALD-REF were reduced by part of the syngas fed into the reactor, which consequently led to an increase of the catalytic activity. Activation of cobalt catalysts in syngas atmosphere has been reported elsewhere [28]. The high stability observed in this catalyst after 30 h of TOS suggests that, in general, ALD passivation conferred stability to the catalyst in all cases. Moreover, it is observed that, the higher the number of ALD cycles, the more stable the catalyst was. Regarding sample 3ALD-REF, although initial activity value was slightly lower than that of REF, then this value started to be higher after ca. 25 h TOS. The gap between the activity of this catalyst and REF was getting broader and broader, until maximum gap seemed to be attained at the end of the experiment. Finally, sample 6ALD-REF also showed interesting results. Although the initial relative activity of this catalyst was slightly lower than that of REF, the gap between activities seemed to narrow as reaction advanced. This fact was corroborated by the activity ratio, as quantified above (see Table  2).
As the aim of ALD passivation is to confer stability to the catalyst without affecting activity, results of 3ALD-REF and 6ALD-REF showed that these were the most promising Regarding sample 3ALD-REF, although initial activity value was slightly lower than that of REF, then this value started to be higher after ca. 25 h TOS. The gap between the activity of this catalyst and REF was getting broader and broader, until maximum gap seemed to be attained at the end of the experiment. Finally, sample 6ALD-REF also showed interesting results. Although the initial relative activity of this catalyst was slightly lower than that of REF, the gap between activities seemed to narrow as reaction advanced. This fact was corroborated by the activity ratio, as quantified above (see Table 2).
As the aim of ALD passivation is to confer stability to the catalyst without affecting activity, results of 3ALD-REF and 6ALD-REF showed that these were the most promising samples. After practically 90 h of experiment, 3ALD-REF exhibited slightly higher activity than the reference, which proves that this technique can be used to increase the resistance of Co/Al 2 O 3 to deactivation. In order to compare quantitatively the deactivation rate of samples REF, 3ALD-REF and 6ALD-REF, deactivation modelling was carried out using a generalized power-Llw expression (GPLE) proposed by Bartholomew [29,30]: where a stands for the relationship between the reaction rate at any time and the initial value (W cat refers to catalyst mass): k d is the deactivation rate constant. f (P i ) stands for a power-law expression of the deactivation rate dependence on the concentration (partial pressure) of reactants, products and/or poisons. Previously, it was stated f (P i ) = P H2O P CO +P H2 ≈ F H2O F CO +F H2 as a criterion for estimating the deactivation rate of cobalt catalysts in FTS [31]. F H2O was calculated based on the oxygen balance between the reactor feed and the reactor outlet. Moreover, it is assumed that activity did not decline up to any steady state, since this value could not be obtained accurately. The exponent d is the order of deactivation, so that the following expressions can be obtained: 1st order : a(t) = exp(−k d · f (P i )·t). Linear fitting : ln((a(t)) = −k d · f (P i )·t (4) 2nd order : a(t) = 1/(k d · f (P i )·t + 1). Linear fitting : 1/a = k d · f (P i )·t + 1 (5) Table 3 lists the linear fitting results for the three orders of deactivation above described. Modelling of the experimental results suggested that catalyst deactivation can be fitted to a 1st or 2nd order of GPLE, although the accuracy of the fitting was not enough to draw a clear conclusion about this point. What is more important is the fact that, whether deactivation obeys a 0th, 1st or 2nd order, deactivation rate constants (k d ) of ALD-passivated samples were lower than those of REF sample and the deactivation rate constant was directly related to the number of ALD cycles applied. Lower k d means less deactivation rate, so that the benefit of ALD passivation on reducing catalyst deactivation in FTS was confirmed.

Catalytic Performance at Accelerated Deactivation Conditions (X CO ≈ 60%)
Although GPLE fitting and k d estimation give a first estimation of how ALD-coated samples would behave with respect to REF catalyst, further experiments at accelerated deactivation conditions were carried out in order to have a complementary experimental view of this behavior. Samples REF, 3ALD-REF and 6ALD-REF were submitted to tests at higher CO conversion level (≈60%) with the aim of evaluating their behavior in this so-called environment of accelerated deactivation. The higher the CO conversion, the higher the H 2 O concentration, one of the most important deactivation agents in FTS [3,26]. Results are listed in Table 4. In addition, 10ALD-REF was not evaluated at these harder conditions due to the lower catalytic activity previously observed.
Interesting facts can be remarked from results shown in this table. On the one hand, it should be noted that, using these conditions, the relative activity of sample 6ALD-REF was becoming higher and higher, whereas that of 3ALD-REF remained constant, although slightly higher than 1, i.e., higher than the activity of sample REF. On the other hand, the yield of ALD-passivated samples was lower than that of the reference in the starting point, although then these values started to increase. Table 4. Experimental results (reaction conditions: 20 barg, 230 • C, 4800 NmL/g·h, H 2 /CO = 2/1 (mol/mol)). Although these results seemed to give a different optimum value, it should be noted that reaction conditions in these tests were different from those performed at regular conditions. Indeed, the strong decrease of GHSV required in experiments at accelerated deactivation conditions suggested a possible change in the rate controlling step, then confirmed by experimental procedure (results not shown). In any case, FTS tests altogether confirmed that a small number of ALD cycles conferred stability to the catalyst.
The ALD passivation procedure was carried out directly in the catalyst (3 g) in an

ALD Passivation Procedure
A microcatalyst (Co/γ-Al 2 O 3 , 16.3 wt.% cobalt) was used as a reference. This catalyst was synthesized by incipient wetness impregnation of the cobalt precursor onto spheres of γ-Al 2 O 3 (500 µm), following the procedure described in [26].
The ALD passivation procedure was carried out directly in the catalyst (3 g) in an R200 Picosun atomic layer deposition equipped with a POCA 200 powder coating system. Trimethylaluminum (TMA, Sigma Aldrich, Germany), 663301-25G, electronic grade) and H 2 O precursors were introduced and purged with N 2 (99.999%, Linde, Spain) in successive cycles, following the sequence TMA-N 2 -H 2 O-N 2 . The time sequence of each cycle was 1.5-60-1.5-60 (s). The process was performed at 300 • C. Table 5 lists the samples prepared for the present study.

Sample Characterization
Textural properties were determined from N 2 -physisorption (adsorption/desorption) isotherms recorded by a TriStar II 3020-Micromeritics sorption analyzer. Prior to the measurements, the samples were degassed at 90 • C for 1 h, and then at 250 • C for 4 h on a FlowPrep 060 (Micromeritics, Norcross, GA, USA). Brunauer-Emmett-Teller (BET) method was used to calculate the BET surface area for a relative pressure (P·P 0 −1 ) range between 0.05 and 0.30. Furthermore, Barrett-Joyner-Halenda (BJH) method applied to desorption branch of the isotherm was used to determine the mesopore volume.
Structural properties were studied by powder X-ray diffraction (XRD). The wideangles data were obtained on a XRD D8 Advance A25 diffractometer (Bruker, Billerica, MA, USA) using Cu Kα radiation (λ = 1.5406 mm), a voltage of 40 kV, a current of 40 mA and a step size of 0.05 • (with 3 s duration at each step). The acquisitions were done in a range from 20 • to 80 • . The software X'Pert HighScorePlus was used to identify the crystalline phase using the standard powder XRD files published by the International Centre for Diffraction Data (ICDD). The average crystal size was calculated using the Scherer's equation: dCo 3 O 4 = (Kλ/βCosθ), where λ is the X-ray wavelength, β is the full width of the diffraction line at half maximum (FWHM) and θ is the Bragg angle.
Reducibility of the catalysts was analyzed by temperature programmed reduction method (H 2 -TPR) on an Autochem (Micromeritics, Norcross, GA, USA). In each run, 150 mg of fresh calcined catalyst was used for the analysis. The H 2 -TPR measurements were conducted using 12 vol.% H 2 /Ar at a flow of 50 NmL·min −1 in the temperature range of 35 to 800 • C and a heating ramp of 10 • C·min −1 . The extent of cobalt reduction was studied using O 2 titration experiments, which were conducted in the same apparatus as was used for the TPR experiments. Firstly, samples were reduced once H 2 -TPR results have been discussed (see below). After reduction, sample was kept at 380 • C in He and held for 1 h to desorb any chemisorbed H 2 . Calibrated pulses of O 2 were then added into the continuous He flow until further consumption of O 2 was not detected by the TCD located downstream from the reactor. The extent of reduction was calculated assuming stoichiometric reoxidation of Co 0 to Co 3 O 4 .
The morphology of the catalyst was visually analyzed by high-resolution transmission electron microscopy (HRTEM) together with scanning TEM (STEM). Investigation was performed on a field emission gun FEI Tecnai F20 microscope (Thermo Fischer Scientific, Waltham, MA, USA). High angle annular dark-field (HAADF) STEM was combined with electron energy loss spectroscopy (EELS) in the Tecnai microscope by using a GATAN QUANTUM filter (Thermo Fischer Scientific, Waltham, MA, USA). For TEM analyses, sample particles were crushed, seeing the large particle size, and the obtained powder was dispersed in pure ethanol. Then, a few drops of the above suspension were deposited on a C-coated TEM Cu grid.

Catalytic Tests
Fischer-Tropsch catalytic studies were carried out in a laboratory scale fixed-bed reactor (Microactivity Reference, PID Eng&Tech, Spain). In each run, samples (0.5 g) were diluted in 5 g of SiC 46 grit (Sigma-Aldrich) to minimize the temperature gradients throughout the catalytic bed. Hydrogen (99.999%, Linde), carbon monoxide (99.5%, Linde) and nitrogen (99.999%, Linde) were used in each catalytic run. Firstly, samples were reduced in 100 NmL/min of hydrogen at 380 • C and atmospheric pressure, with a hold time of 7 h and a heating ramp of 1 • C/min. Then, the system was cooled to 150 • C under hydrogen atmosphere. Once this temperature was attained, 200 NmL/min of the reactant mixture was introduced (H 2 /CO ≈ 2 (v/v) and 5% N 2 (v/v) as internal standard) and pressure was raised to 20 barg. This step was kept for 6 h, during which the concentration of the reactant mixture was analyzed and taken as a reference for calculations (blank experiment). Afterwards, temperature had risen to 230 • C to begin the reaction, keeping 200 NmL/min of reactant flow during 16 h. Finally, the GHSV was adjusted to a specific value depending on the set of the experiments. In a first set of experiments, referred to as "regular conditions", GHSV was adjusted to 8400 NmL/g·h to obtain an intermediate level of conversion. At this GHSV, CO conversion reached approximately 50% over REF. A second set of experiments, referred to as "accelerated deactivation conditions" was also performed, in which 1 g catalyst was diluted in 5 g of SiC 46 grit, and GHSV was adjusted to 4800 NmL/g·h to attain, approximately, 60% CO conversion over REF.
After reaction, the reactor effluent passed through two consecutive traps. In the first one (100 • C and 20 barg), waxes were trapped together with some H 2 O, whereas the second one (10 • C and 20 barg) retained the liquid hydrocarbons and remaining H 2 O. The whole system (traps excluded) was placed in a hot box at 150 • C to avoid possible condensations on the tubing. The gas effluent (light hydrocarbons, CO 2 and the remaining reactant mixture) was analyzed online with a gas microchromatograph equipped with a TCD (490 microGC, Agilent Technologies, Madrid, Spain), which was previously calibrated using different cylinders and N 2 as standard.

Evaluation of Catalytic Performance
CO, H 2 and syngas conversion (X i ) were calculated using the following formula: where F is the molar flow and i refers to CO, H 2 or the sum of both species. F i was evaluated using the concentration of the N 2 as internal standard in the effluent stream. Selectivity (S i ) to gas products (CO 2 and light hydrocarbons) with respect to CO was calculated as follows: where δ refers to the stoichiometric factor of CO with respect to that of the product i (CH 4 , CO 2 , C 2 H 6 , C 2 H 4 , C 3 H 8 , C 3 H 6 , C 4 H 10 ). Selectivity to liquid and wax products (S C5+ ) was obtained considering a carbon balance of 100%: The activity of the catalyst was quantified as a function of the GHSV and the syngas conversion, according to the following equation: being syngas conversion calculated as follows: where F i,in the molar inlet flow and F i,out the molar outlet flow. i refers to H 2 and CO. Besides, the yield to liquid and wax products (Y C5+ ) was calculated using the formula: To compare the ALD-passivated samples, relative activity values were evaluated at any time with respect to those of the reference catalyst (A ref and Y C5+ref ).

Conclusions
The results obtained in this work gave two main conclusions. Firstly, it was proven that ALD passivation is a promising technique to increase the resistance of Co-based catalysts against deactivation in FTS. Secondly, the differences between ALD-coated samples in such a narrow range (3-10 coatings) make it necessary to carry out exhaustive control of the number of passivation cycles, which is one of the main features of this technique.
Characterization results revealed that ALD did not affect either textural properties or cobalt crystallite size. The main differences were observed in TPR and O 2 titration experiments, as the number of ALD cycles affected catalyst reduction properties. In this sense, 61.4, 56.5 and 65.4% of Co contained in samples REF, 3ALD-REF and 6ALD-REF was respectively reduced, whereas only 11.1% of that in sample 10ALD-REF did. As for TEM results, it was clearly observed that a shell of Al 2 O 3 (ALD passivating agent) was formed around cobalt particles, and the thickness of this shell was closely related to the number of ALD cycles applied on the catalyst.
Catalytic tests at regular conditions (X CO ≈ 50%) confirmed that sample 10ALD-REF resulted in being overcoated, so that in the first stages of the experiment, syngas role was reducing Co particles that resulted in not being reduced in the previous reduction step. Consequently, an initial relatively low CO conversion was observed, which started to increase slightly as reaction went by. Conversely, samples 3ALD-REF and 6ALD-REF showed promising results, as they reflected the double objective of ALD passivation, that is, conferring stability without affecting catalytic activity. Besides, ALD-coated samples were more selective to hydrocarbons of longer chain. Fitting experimental results to a GPLE function corroborated that these samples were more resistant to deactivation than REF, as the k d values of ALD-passivated samples were lower than those of REF. In order to confirm this promising estimation, experiments at accelerated deactivation conditions (X CO ≈ 60%) were carried out, confirming that 6ALD-REF catalyst showed better results in terms of activity and stability, whereas 3ALD-REF performance overlapped that of sample REF.
Therefore, the results presented in this work confirmed ALD as a promising technique for cobalt catalyst to gain resistance against deactivation. Specifically, 3 to 6 ALD cycles would be sufficient to increase catalyst stability, whereas a higher number of cycles (10) led to blocking of cobalt particles, avoiding transport of reactants up to catalyst surface. Therefore, the technical feasibility of passivating a Co/γ-Al 2 O 3 catalyst by atomic layer deposition (ALD) to reduce deactivation rate during FTS has been confirmed. However, the positive effect observed was very small and more effort should be addressed in order to further increase the stability of the reference catalyst. Further studies dealing in depth with the causes of catalyst deactivation and how ALD prevents them are recommended.