University of Birmingham Hydrogen evolution enhancement of ultra-low loading, size-selected molybdenum sulfide nanoclusters by sulfur enrichment

Size-selected ﬁ de (MoS x ) nanoclusters obtained by magnetron sputtering and gas condensation on glassy carbon substrates are typically sulfur-de ﬁ cient (x=1.6 ± 0.1), which limits their crystallinity and electrocatalytic properties. Here we demonstrate that a sulfur-enriching method, comprising sulfur evaporation and cluster annealing under vacuum conditions, signi ﬁ cantly enhances their activity towards the hydrogen evolution reaction (HER). The S-richness (x=4.9 ± 0.1) and extended crystalline order obtained in the sulfur- treated MoS x nanoclusters lead to consistent 200mV shifts to lower HER onset potentials, along with two-fold and more-than 30-fold increases in turnover frequency and exchange current density values respectively. The high mass activities ( ∼ 111mAmg − 1 @ 400mV) obtained at ultra-low loadings ( ∼ 100ngcm -2 , 5% surface coverage) are comparable to the best reported MoS 2 catalysts in the literature. the in uence of sulfur content in the HER catalysis of 2 use of an in sulfur demonstrate that sulfur min) annealing 215 5 in the MoS x nanocluster structure 0.1), by reducing oxygen-containing Mo surface species and converting the amorphous S moieties to crystalline S sites, which also extends the crystalline A consistent 200 mV shift to lower HER overpotential, along with a two-fold increased turnover frequency and more-than 30-fold increase of current density values the cial role of higher S content and crystallinities in the (MoS ) nanoclusters


Introduction
The interest in the hydrogen economy as a potential candidate to replace the current fossil fuel-based energy system [1] has motivated extensive research on environmentally-friendly hydrogen production methods. The hydrogen evolution reaction (HER) taking place at a water electrolyser cathode is a scalable yet energy-efficient route [2] which demands earth-abundant catalysts to be commercially viable. Among them, transition metal dichalcogenides (TMDs) and in particular molybdenum disulfide (MoS 2 ) have stood out in the past decade [3,4]. Their layered structure, analogous to that of graphene, also implies anisotropic properties: only the metallic 1 T phase sites located at the Mo-edge planes of naturally occurring MoS 2 are active for the HER [5,6], whereas the 2 H semiconducting basal planes are almost inactive if no defects are present [7][8][9]. Several strategies have proven to maximize MoS 2 HER activities: [10] triggering the 2H → 1T phase transition in basal planes by chemical intercalation [11][12][13] or stress/ strain effects; [14,15] basal plane activation by incorporation of transition metals [16][17][18][19][20] or other chalcogenides; [21,22] and the fabrication of MoS x nanostructures which are defect-rich [23][24][25][26][27][28][29][30] or have additional S vacancies [31][32][33][34][35][36]. However, the in-operando proven role of S atoms as the HER active sites [37] indicates that sulfur-rich MoS 2+x materials should also present high HER activities [38][39][40][41]. Our recently reported size-selected MoS x nanoclusters, obtained by magnetron sputtering and gas condensation [42], were demonstrated to be sulfurdeficient (x = 1.6 ± 0.1) with low crystallinities. In this article we have evaluated the influence of sulfur content in the HER catalysis of MoS 2 materials through use of an in vacuo sulfur addition treatment previously developed for freshly deposited, sulfur-deficient (MoS x ) 1000 nanoclusters [43]. We demonstrate that sulfur evaporation (5 min) followed by annealing treatment (7 min, 215 ± 5°C) incorporates S in the MoS x nanocluster structure (x = 4.9 ± 0.1), by reducing oxygencontaining Mo surface species and converting the amorphous S 2 2− moieties to crystalline S 2− sites, which also extends the crystalline order. A consistent 200 mV shift to lower HER overpotential, along with a two-fold increased turnover frequency and more-than 30-fold increase of exchange current density values proves the beneficial role of higher S surface content and crystallinities in the (MoS x ) 1000 nanoclusters HER catalysis. Size-selected MoS 2 nanoclusters were produced using a DC magnetron sputtering and gas condensation cluster beam source as shown in Fig. 1 from a 2-inch sputtering MoS 2 target (PI-KEM, 99.9% purity) [44]. The positively charged clusters were accelerated with ion optical electrostatic lenses and then size-selected with a lateral time-of-flight mass filter [45]. A mass of 160,000 amu, corresponding to 1000 MoS 2 units (designated as (MoS 2 ) 1000 ), was selected for depositing onto an amorphous carbon coated TEM grids (Agar Scientific, 200 Mesh Cu) and onto glassy carbon (GC) stubs (5 mm × 5 mm x 3 mm, mirror finish). The loading of the TEM grid samples was approx. 5% projected surface area coverage (i.e., approx. 5% of the surface covered by clusters), while the loadings of the GC samples were 5%, 10% and 20% projected surface area. The clusters were deposited onto amorphous carbon covered TEM grids and GC stubs with an impact energy of 1.0 eV and 1.5 eV per MoS 2 unit, respectively. Sulfur addition was conducted in a sulfur atmosphere created by evaporating sulfur using a home-built in-situ thermal evaporator (5 min). Annealing (7 min, 215 ± 5°C) was performed with an electron beam bombardment heating stage. The temperature was monitored using a pyrometer (IMPAC Pyrometer, IPE 140). Scanning transmission electron microscopy (STEM) images were acquired with a 200 kV spherical aberration-corrected STEM (JEOL 2100 F) in the high-angle annular dark-field (HAADF) mode [46,47].

Experimental
2.2. Physical characterization of (MoS x ) 1000 nanoclusters: X-ray photoelectron spectroscopy (XPS) XPS spectra were recorded using a Kratos Axis SUPRA fitted with a monochromated aluminium source (Al Kα, 1486.69 eV) and a charge neutraliser. Samples were mounted on silicon wafers by use of silver epoxy, and affixed to a sample bar using carbon tape. Wide scans were recorded using pass energies of 160 eV and high-resolution scans were recorded using pass energies of 20 eV and an analysis area of 30 μm 2 . All scans were recorded at < 5 × 10 −9 Torr using an emission current of 15 mA. All high-resolution spectra were corrected to the adventitious C 1 s peak at 284.6 eV, and deconvoluted using the CasaXPS 2.3.18 software, applying a Shirley background correction before individual peak deconvolution. Mo a O b S c is used to refer to the molybdenum oxysulfide species: the superscript a represents the oxidation state of Mo, whilst the subscripts b and c the stoichiometry of O and S atoms in the specific oxysulfide.

Electrochemical characterization
All electrochemical measurements were performed in a conventional 3-electrode electrochemical setup comprising a thermostatted two-compartment cell (295 ± 2 K), the first compartment containing both a saturated calomel reference electrode (SCE, BAS Inc., Japan) and 5 mm diameter, 3 mm thick glassy carbon working electrodes (GC) type 2 stubs (Alfa Aesar, U.K.) modified with as deposited or sulfur evaporated and annealed (MoS x ) 1000 nanoclusters; and a second compartment containing a bright Pt mesh counter electrode (Alfa Aesar, U.K.). All experiments were conducted using a PC-controlled PGSTAT128N potentiostat (Metrohm Autolab B.V, Netherlands). GC samples were polished to until a mirror finish was achieved by use of decreasing size diamond (45-3 μm) and alumina slurries (1-0.05 μm) on a Buehler MetaServ 250 automatic polisher using Trident/ Microcloth polishing pads. All GC samples were immediately tested after nanocluster modification, being transported to the electrochemical cell in a N 2 -saturated sealed container to avoid exposure to air. The nanocluster-modified GC stubs were embedded in a E4TQ ChangeDisk RDE Tip and electrically connected to a E4 Series Rotating Shaft and a Modulated Speed Rotator (Pine Research Instrumentation, USA). No rotation was applied during any electrochemical experiment.
A 2 mM HClO 4 (ACS ≥ 70%, Sigma-Aldrich), 0.1 M NaClO 4 (ACS ≥ 98%, Sigma-Aldrich) solution (pH 2.7) was used in all experiments, freshly prepared with ultrapure water (Millipore Mili-Q Direct 8, resistivity not less than 18.2 MΩ cm). This fully supported, non-coordinating anion-containing, low proton concentration electrolyte was chosen in contrast to the more commonly reported high proton concentration electrolytes in hydrogen evolution experiments (0.5 M H 2 SO 4 , pH ≈ 0.3; 0.1 M HClO 4 , pH ≈ 1) as previous experiments on (MoS x ) y nanoclusters yielded more reproducible electrochemical results, enabling accurate elucidation of the HER reaction kinetic parameters. Acidic electrolytes with lack of a supporting electrolyte (in our case 0.1 M NaClO 4 ) are reported to distort any kinetic analysis due to migration effects of the electroactive species [48].
Nanocluster-modified GC electrodes were preconditioned prior to HER experiments with 10 cycles from −0.045 to −1.645 V (vs. SCE) at a voltage scan rate of 50 mVs −1 to obtain a stabilized performance. HER electrocatalysis measurements were then recorded at a range of voltage scan rates from 2 to 1200 mVs −1 , and electrochemical impedance spectroscopy measurements (EIS) were acquired in the −0.1 to −1.4 V vs. SCE with 100 mV steps, using a frequency range of 10 −1 to 10 5 Hz (voltage amplitude = 10 mV) to apply the iR compensation correction on all HER voltammograms. All HER potentials reported are corrected versus the normal hydrogen electrode (NHE) using the Nernstian shift correction (E NHE = 0.242 V + 0.059 pH). The electrochemical cell was vigorously purged with N 2 prior to any electrochemical experiment (Oxygen-free grade, BOC Gases plc), and a positive N 2 pressure was maintained during experiments. All electrochemical glassware was cleaned overnight by use of a dilute KMnO 4 (ACS ≥ 99%, Sigma-Aldrich) solution in concentrated H 2 SO 4 (> 95% analytical grade, Fisher Scientific) followed by rinsing with

Results and discussion
3.1. Physical characterization of size-selected (MoS x ) 1000 nanoclusters: HAADF-STEM imaging and XPS Fig. 2 shows the aberration-corrected HAADF-STEM images of (MoS x ) 1000 nanoclusters (selected mass at cluster source, 160,000 amu, equivalent to 1000 MoS 2 units per cluster) at 5% projected surface area coverage after deposition on amorphous carbon covered TEM grids. For cluster source schematic and further deposition parameters, see Fig. 1. Fig. 2a and b are acquired at low magnification before and after sulfur evaporation and annealing, respectively. The as-deposited MoS x clusters are rather irregular with poorly ordered structures, and a mean diameter of 5.5 nm is given based on the projected surface area from our previous study [43]. The STEM image of as-deposited MoS x cluster at a higher magnification (Fig. 2c), together with its FFT pattern (inset), show the amorphous feature of the cluster and confirm the absence of extended crystalline order. The clusters have an uneven layered structure revealed by the HAADF intensity line profile, which agrees with previous first-principle simulation studies [49]. Compared with the asdeposited clusters, the sulfurised clusters become larger with a mean diameter of 6.0 nm. This is due to the morphological reconstruction of MoS x clusters with the added sulfur. In contrast to the as-deposited clusters, the sulfurised clusters shown in Fig. 2d and e present rather crystalline structures, which can also be confirmed by their FFT patterns (inset). The sulfurised clusters retain the layered structure with 3-4 layers-thick. The Moirépattern shown in Fig. 2e indicates a misorientation between layers, which can be commonly found in the sulfurised clusters with 3 or more layers. Given that sulfur is long known to sublime at temperatures well below 100°C [50,51], we can conclude that the crystalline structures come from the chemical bond between the added sulfur and the clusters, and that the structural modification into crystalline clusters mainly takes place within the 2D layers.
XPS measurements were acquired from molybdenum sulfide clusters deposited onto amorphous carbon TEM grids to investigate the degree of sulfur incorporation. The high-resolution Mo 3d and S 2p spectra of the as-deposited molybdenum sulfide nanoclusters reveal a complex surface composition (see Fig. 3a). The Mo spectra (Fig. 3, top row) could not be solely deconvoluted into the Mo 4+ 3d 5/2:3/2 spin-orbit doublet characteristic of MoS 2 materials (binding energies of ∼229.8 and ∼232.9 eV, respectively). Two additional doublets were needed, ascribed to Mo a O b S c (∼231.5 and ∼234.6 eV, see Experimental for Mo a O b S c definition) and Mo 6+ (∼233.1 and ∼236.2 eV) oxidation states reported in molybdenum compounds such as molybdenum oxysulfides [52] and MoO 3 [53]. Analysis of the Mo 4+ : Mo a O b S c : Mo 6+ relative percentages (at. %) from the XPS photoemission intensities yields a relative ratio of 53.8:25.2:21.0 at. %, corroborating the significant proportion of oxidized molybdenum species at the nanoclusters. The S spectra (Fig. 3, bottom row) were deconvoluted using two 2p 3/2:1/2 spin-orbit doublets related to the S 2− (∼161.3 and ∼162.5 eV) and S 2 2− (∼162.6 and ∼163.8 eV) oxidation states consistently reported for amorphous MoS x thin films and nanoparticles [ 54,55], yielding a S 2− / S 2 2− relative ratio of 20:80. The broad S signal centered at ca. 170 eV is ascribed to SO x yspecies [56], The XPS intensity ratio between the S-containing Mo species (Mo 4+ /Mo a O b S c ) and the S 2− / S 2 2− species yields a close-to-stoichiometric but still S-deficient ratio (1:1.9 ± 0.1), similar to that found in our previous investigations [42,57]. Likewise, high-resolution XPS spectra on the sulfur-evaporated and annealed (MoS x ) 1000 nanoclusters (Fig. 3c) [43]. Hence, it is concluded that the best methodology to produce S-enriched MoS x nanoclusters with enhanced crystalline order is by the adoption of sequential sulfur evaporation and thermal annealing.

Electrocatalytic activity to the hydrogen evolution reaction: influence of sulfur enrichment
The hydrogen evolution activity of the as-prepared and sulfur-enriched (MoS x ) 1000 nanoclusters was evaluated in a 3-electrode electrochemical setup, by recording linear sweep voltammograms between 0 to −1.2 V (scan rate = 50 mV s −1 ) in a 2 mM HClO 4 /0.1 M NaClO 4 aqueous electrolyte (normalized vs. NHE and iR compensated, for further details, see Experimental). The low proton concentration in the electrolyte used ([H + ] ≈2 × 10 -6 mol cm -3 , pH ≈ 2.7) is responsible for the diffusion decay peak profile in Fig. 4a and b, analogous to that found with our previously reported magnetron-sputtered nanoclusters. [57,58] The as-prepared samples present onset potentials, |η onset | for current densities of |j| = 0.05 mA cm -2 , of ca. 690 mV, which are ∼60 mV positively shifted compared to the recorded |η onset | for bare glassy carbon. This confirms that even at ultra-low loadings MoS 2 effectively catalyzes the HER. The peak half-maximum overpotentials (|η half max |) and current densities (|j half max |) metrics previously used to describe the HER catalysis of magnetron-sputtered nanoclusters [57] are found to be ca. 810 mV and 0.31 mA cm -2 , respectively (see Table S1 ESI).
These are in good agreement with the results obtained for (MoS 0.9 ) 300 nanoclusters, which presented a higher cluster loading (ca. 3.5 μg cm −2 ) but equivalent surface coverage given the smaller cluster sizes (∼20%). [57] Interestingly, such ultra-low loadings of size-selected MoS x nanoclusters used in the present work (5% coverage: ∼84 ng cm −2 , 10% coverage: ∼168 ng cm −2 , 20% coverage: ∼335 ng cm −2 ) already present HER activities comparable to those of (MoS 0.9 ) 300 nanoclusters with loadings higher by 1 order of magnitude. Despite both smaller dimensions (∼2.6 nm) and higher loadings, the Sdeficient Mo:S ratio and cluster overlapping upon random surface landing can then explain the (MoS 0.9 ) 300 nanoclusters' reported performance. After sulfur incorporation, all (MoS x ) 1000 nanoclusters exhibit remarkable improvements in their HER performance. A consistent 200 mV shift in the HER |η half max | was found independently of the sample loading (see. Fig. 4a-b).
To gather further insight about the HER kinetics and electron transfer properties, Tafel slope analysis and electrochemical impedance spectroscopy (EIS) experiments were carried out before and after sulfur enrichment of (MoS x ) 1000 nanoclusters. Tafel plots of the cathodic linear sweep voltammograms (|η| vs. log|j geom |, Fig. 4c) show Tafel slopes in the 143-154 mV dec −1 range for all (MoS x ) 1000 nanocluster samples irrespective of both loading and sulfur modification, similar values to the one found for bare GC (≈154 mV dec −1 ). This indicates that the sulfurisation treatment does not modify the mechanism under which the HER operates: for slopes close to b ≈ 120 mV dec −1 this is the Volmer mechanism, its rate-limiting step being the electroadsorption of monoatomic hydrogen [59]. Previous reports on amorphous MoS x catalysts have reported Tafel slopes of b ≈ 40 mV dec −1 (Volmer-Heyrovsky rate-limiting step), significantly lower than the ones obtained for the as-deposited amorphous (MoS x ) 1000 nanoclusters. Two main factors are responsible for this: the electrolyte pH and the inherent morphology or the clusters. Recent investigations by Dubouis et al. on electrodeposited, amorphous MoS x materials have shown that the HER mechanism (and consequently the Tafel slope) is pH-dependent: [60] for pH ≤ 1, the hydronium cation electroreduction governs the proton reduction with pH-independent Tafel slopes of b ≈ 40 mV dec −1 ; at higher pH values the lower proton concentration leads to mass transport limitations which ultimately result in the proton electroadsorption (i.e. Volmer rate-limiting HER step, b ≈ 120 mV dec −1 ) dominating the HER. Alternatively, the 40 mV dec -1 Tafel slopes reported on amorphous MoS x are well known to arise from the [Mo 3 S 13 ] 2cluster-based structure and the different sulfur moieties entailed [61,62]. The pH ≥ 1 used for our electrolyte along with the trigonal prismatic coordination as found in 2 H-MoS 2 for our size-selected MoS x nanoclusters [42] support the ca. 143-154 mV dec −1 Tafel slopes obtained. Electrochemical impedance spectroscopy (EIS) Nyquist plots were fitted with a simplified equivalent circuit model based on the recentlyused linear transmission model [63,64] for amorphous/porous MoS x structures (see Fig. S1 ESI for further details) [65,66]. Unlike the Randles circuit conventionally used to physically describe the HER on TMD materials, this circuit not only accounts for the charge transfer resistance (R ct ), but also for the contact resistance between the nanoclusters and the glassy carbon electrode interface (R c ). Such information is of physical relevance given the layer-dependent HER catalysis of TMDs and their inherently high through-plane resistance [67][68][69][70][71]. At −1.1 V vs. SCE (∼−0.7 V vs. NHE), a significant decrease in all EIS resistance components was found after the combined treatment of sulfur evaporation plus annealing on the (MoS x ) 1000 nanoclusters (Fig. 4d, Table S2 ESI): R ct (∼1240 vs. ∼1180 Ω, 5% coverage; ∼6060 vs. ∼840 Ω, 20% coverage), and R c (∼4640 vs. ∼3250 Ω, 5% coverage; ∼12,420 vs. ∼6820 Ω, 20% coverage). We postulate the extended crystalline order of the sulfur-enriched nanocluster structure to be the governing factor. This can be supported by both the FFT analysis of the nanoclusters imaged by HAADF-STEM and the high-resolution S 2p XPS results. The former shows, after sulfur incorporation, that the (MoS x ) 1000 nanocluster FFT pattern changes from a diffuse ring characteristic of highly amorphous materials to a well-defined set of diffraction spots ranging from single sets ascribed to aligned MoS 2 layers along the (100) plane (intralayer spacing: 0.25 nm) to dual sets related to misoriented stacking layer arrangements [43]. The high-resolution S 2p XPS data monitoring the S 2− / S 2 2− intensity ratio, which serves as a descriptor of the degree of MoS x crystallinity, reveals an increased S 2− relative content after the sulfur evaporation treatment: 75:25 vs. the 20:80 found in pristine nanoclusters. Thus, the sulfur evaporation and annealing not only incorporates sulfur into the nanocluster structures but also converts the characteristic amorphous MoS x /MoS 3 S 2 2− moieties [ 41,55,72,73] to S 2− as found in crystalline MoS 2 [74]. From these findings we can conclude that the sulfur evaporation and subsequent annealing of (MoS x ) 1000 nanoclusters results in an overall improvement in their charge transfer properties. A previous report on polymorphic MoS 2 (a system which resembles the non-crystalline nature of our asdeposited nanoclusters) revealed that electron hopping only occurs between metallic 1T domains bounded by semiconducting 2H regions, and therefore is limited [75]. On a separate note, it is also noteworthy to explore which are the potential HER active sites in our MoS x nanoclusters. For amorphous MoS x , terminal S 2 2− [76], bridging S 2 2- [37] or unsaturated Mo IV centers (i.e. S vacancies) [40] have been proposed as moieties responsible for hydrogen evolution, reaching no unambiguous consensus to date. For the as-prepared (MoS x ) 1000 nanoclusters, the presence of terminal/ bridging S 2 2− as found in our S 2p XPS spectra seems to indicate they might participate in the HER along with the well-established TMD unsaturated S 2− active sites [5,77]. In the case of our Senriched (MoS x ) 1000 nanoclusters, the almost total conversion of the partially-oxidized Mo a O b S c and S 2 2− species to Mo 4+ and S 2− as found in crystalline MoS 2 and subsequent HER enhancement lead us to believe that the main HER actives are the unsaturated S 2− moieties.

Evaluation of figures of merit and catalyst benchmarking
Further catalyst benchmarking by turnover frequency (TOF) and exchange current density (j 0 ) analysis also demonstrates the HER enhancement observed. For 5% surface coverage, as-deposited (MoS x ) 1000 nanoclusters present TOF ≈ 3.0 H 2 s −1 and j 0 ≈ 8.8 × 10 -10 A cm -2 at |η half max | = 825 mV, whereas for an equivalent |η half max | the sulfurmodified (MoS x ) 1000 nanoclusters sample exhibits TOF ≈ 6.1 H 2 s −1 and j 0 ≈ 2.8 × 10 -8 A cm -2 . At 20% surface coverage, similar enhancements can be found (TOF ≈ 1.4 vs. 0.8 H 2 s −1 at |η half max | = 814 mV; j 0 ≈ 5.2 × 10 -8 vs. 7.9 × 10 -10 A cm -2 ). The two-fold increase in TOF and more than 30-fold increase in j 0 indicates improved per-site activities and active site densities: positive shifts in onset potential values under given HER kinetics (i.e. same Tafel slope values) have been related to higher densities of active sites [11]. This, along with the onset potential shift, significantly surpasses the HER enhancement (ca. 70 mV at |j half max |, see Fig. S2a ESI), found after S-edge site doping with Ni in (Ni-MoS 2 ) 1000 nanoclusters (3-fold increase in j 0 but lower TOF after doping) [57], indicating that the synergistic effect of sulfur enrichment and improved crystallinity prevails over a S-edge activation strategy on as-deposited MoS x nanoclusters.
We finally proceeded to benchmark the performance of our (MoS x ) 1000 nanoclusters with recently-reported MoS 2 -based catalysts from the literature. (Table S3 ESI) However, the ultra-low loadings utilized in this report preclude quantitative comparisons based on the HER metrics commonly cited (|η| at 10 mA cm −2 and |j geom | at 200 mV). It is well known that these metrics are heavily affected by the catalyst loading (for loading-dependent HER see Fig. S2b ESI) [54,[78][79][80][81], catalyst layer thickness [67,82,83] and TMD morphologies [25,41,84]. Instead, we normalized all previous |j geom | reported values by mass activity (mA mg -1 ), a metric widely accepted in the noble metal electrocatalysis community (see Table S3 ESI) [85,86]. The mass activities found for (MoS x ) 1000 nanoclusters at |η| values as low as 400 mV (close to the HER onset) are, after sulfur evaporation and annealing, comparable with the best reported MoS 2 catalysts at 200 mV tested using a high proton concentration electrolyte. The values obtained are ca. 110 mA mg -1 at 5% coverage and ca. 70 mA mg -1 at 20% coverage (see Table S1 ESI). For |η half max |, mass activities are in the 1000 mA mg -1 range: for 5% coverage, ca. 3620 mA mg -1 (pristine) and ca. 4010 mA mg -1 (sulfurised); for 20% coverage, ca. 980 mA mg -1 (pristine) and ca. 1040 mA mg -1 (sulfurised). This highlights the remarkable activities of the sulfurised (MoS x ) 1000 nanoclusters obtained at very low loadings.
The electrochemical stability of MoS x electrocatalyts is also an important feature for evaluating prospective long-term HER performance. A preliminary comparison of the very first cathodic HER cycle recorded during our preconditioning step with the final pseudo-stationary LSV reported (11th real HER cycle, as shown in Fig. 4a and b) reveals clear differences in stability before and after sulfur evaporation and enrichment (Fig. S3 ESI). For 20% surface coverage, as-deposited and S-defficient (MoS x ) 1000 nanoclusters present an extraordinarily high activity on their first cathodic polarization scan (|η half max | ≈ 380 mV) which dramatically decays shown by a 415 mV overpotential shift at the 11th scan (Fig. S3a). This indicates that, despite of their high activity, the edge/defect-abundant nature of amorphous MoS x nanoclusters also confers them a high electrochemical instability.
Remarkably, the S-enriched crystalline (MoS x ) 1000 nanoclusters present a dramatically enhanced stability (Fig. S3b): although their initial activity is not as high as the amorphous nanocluster counterparts, |η half max | is modified less than 30 mV. We believe that the improved crystallinity and subsequent minor presence of dissolution-prone undercoordinated Mo sites after S-enrichment mitigates electrochemicallyinduced MoS x leaching yielding higher stabilities.

Conclusions
In summary, the initially sulfur-deficient (MoS 1.9 ) 1000 size-selected nanoclusters obtained by magnetron sputtering and gas condensation and deposited onto glassy carbon substrates have been successfully sulfur-enriched, by sequential application of sulfur evaporation and annealing, for HER applications. This treatment has been shown to induce extended crystalline order, compared with the initially amorphous nanocluster morphology, plus the incorporation of S 2− moieties at the (MoS x ) 1000 nanocluster surface to yield Mo 4+ /Mo a O b S c : S 2− / S 2 2− ratios of 1: 4.9 ± 0.1 instead of 1:1.9 ± 0.1. The annealing step is found key to reducing fully the oxygen-containing Mo species to Mo 4+ and maximizing sulfur incorporation at the nanoclusters surface. A consistent positive shift in the HER |η onset | was found irrespective of sample loading of S-enrichened (MoS x ) 1000 nanoclusters (approximately 200 mV), whilst the Tafel slope remained unaffected by the sulfur treatment (ca. 145 mV dec -1 ). The 2-fold and more than 30-fold increases in TOF and j 0 values, respectively, surpass the HER enhancements previously reported after S-edge site activation by Ni in (Ni-MoS 2 ) 1000 hybrid nanoclusters. The results illuminate the critical role played by S-enrichment and crystallinity in MoS x nanocluster hydrogen electrocatalysis: creating higher densities of proton-acceptor S sites and lower charge transfer resistances, as well as conferring higher electrochemical stabilities. Nanocluster benchmarking by mass activity emphasizes the remarkable performance of S-rich (MoS x ) 1000 size-selected nanoclusters at the ultra-low loading level (83.78 ng cm -2 , 5% surface coverage): 110.5 mA mg -1 at 400 mV overpotential, and 4010.5 mA mg -1 at |η half max | = 652 mV. These results are comparable to the state-ofthe-art MoS 2 -based catalysts, reflecting the significant activities of sizeselected MoS x nanoclusters obtained at ultra-low loadings, resembling previous enhancements reported for noble metals [87][88][89].