Single-nanowire photoelectrochemistry

Transcription

1 Single-nanowire photoelectrochemistry Authors: Yude Su 1, Chong Liu 1,3, Sarah Brittman 1,3, Jinyao Tang 1,3, Anthony Fu 1,3, Nikolay Kornienko 1,3, Qiao Kong 1, Peidong Yang 1,2,3,4* These authors contributed equally to this work. * To whom correspondence should be addressed. p_yang@berkeley.edu Fabrication of the n + p-si single-nanowire device The as-doped p-si single nanowires were thermally oxidized at 1,000 ºC for 1 hour. Then 10 μl of I-line photoresist solution (1:7 dissolved in ethyl acetate) was drop-cast onto the device chip. With the I-line photoresist at the bonding pad scratched to expose the electrode s surface, the chip was baked at 90 ºC for 5 minutes, resulting in ~1 μm of I-line photoresist at the base of the nanowires. Subsequently, BHF (1:10) was used to etch the oxide on the nanowire s upper exposed part, followed by I-line removal in acetone. After that, a Si handle wafer was spincoated with arsenic-containing spin-on dopant (SOD) (Filmtronics, Inc.) and baked at 150 ºC for 30 minutes. Then the device chip was placed on the SOD-coated Si wafer, and annealed at 900 ºC for 4 minutes in an N 2 atmosphere to form a thin n + layer at the surface of the nanowire s upper part. Subsequently, 30 μl of SU-8 solution, which was made by dissolving 0.1 g SU in 2 ml ethyl acetate, was drop-cast onto the device chip. With the SU-8 at the bonding pad scratched to expose the electrode s surface, the chip was baked at 60 ºC, 100 ºC, 150 ºC and 1 NATURE NANOTECHNOLOGY 1

2 200 ºC for 5 minutes, 5 minutes, 5 minutes and 30 minutes respectively, in order to harden the SU-8. The resulting SU-8 layer is ~3 μm thick. Finally, after a quick etching in BHF, the device chip was soaked in the solution containing 0.1 M HF and 0.2 mm K 2 PtCl 6 for 3 minutes to load platinum as the HER catalyst. Scanning electron microscope images of the single nanowires The single silicon nanowires were imaged using a scanning electron microscope (JEOL JSM- 6340F). The significant difference between the as-grown nanowire (Supplementary Fig. 2a) and the nanowire after the PEC measurement (Supplementary Fig. 2b) suggests that platinum nanoparticles (the bright spots in Supplementary Fig. 2b), which serve as the proton reduction electrocatalysts, were successfully loaded on the nanowire s surface. Transmission electron microscope images of the Si nanowire/pt interface High-resolution transmission electron microscope images were acquired with a FEI Tecnai F20 UT with an accelerating voltage of 200 kv. The images show that a thin native oxide layer exists not only before (Supplementary Fig. 3a) but also after the PEC experiments (Supplementary Fig. 3b). It has been reported that such an oxide layer may behave as a recombination center and may impede the interfacial charge transfer 31. However, it also serves as a protection layer to prevent further oxidation 31. In addition, such a native oxide layer also prevents the formation of the ohmic contact at the Si nanowire/pt interface 18. Electrochemical potential calibration of the platinum electrode To reduce the series resistance of the PEC circuit as well as prevent the potential chemical contamination from a standard reference electrode, a platinum wire was used as the quasi- 2 NATURE NANOTECHNOLOGY

3 SUPPLEMENTARY INFORMATION reference electrode in our PEC measurements 32, 33. After the single nanowire PEC measurement, the electrochemical potential of the platinum electrode was immediately calibrated with a Ag/AgCl (0.1M KCl) standard reference through an open-circuit voltage measurement. This potential calibration was carried out in the same solution as the PEC measurement (0.1M aq. K 2 SO 4 adjusted to ph ~ 2 using H 2 SO 4 ). The open-circuit voltage measurement (Supplementary Fig. 4) reveals that the electrochemical potential of the platinum electrode (vs. Ag/AgCl) shows negligible change in one hour. Since the device measurement lasts for about half an hour (for 3 repeated scan cycles), this open-circuit voltage measurement indicates the stability of the platinum electrode s electrochemical potential during the measurement and thus the reliability of the onset potential evaluation. Besides, such potential calibration measurements were also compared every month. The small difference between the measurements in different months indicates the good reproducibility of the potential calibration. Fabrication and PEC measurement of Si nanowire array devices The p-si (boron) wafers (ρ ~ Ω. cm) were obtained from Addison Engineering, Inc. The wafers were thermally oxidized at 1,050 ºC for 8 hours to form a SiO 2 layer with a thickness of ~350 nm. The catalysts for vapor-liquid-solid growth were defined by photolithography, anisotropic plasma etching of the oxide, and subsequent electron-beam evaporation of gold (150 nm). The Si nanowire arrays were grown at 875 ºC for ~40 minutes with SiCl 4 as the precursor and 10% hydrogen in argon as the reducing agent. The resulting Si nanowires are ~15 μm long and ~700 nm thick (Supplementary Fig. 6). The as-grown Si nanowire arrays were etched in BHF for 30 seconds and subsequently soaked in gold etchant (Transene) for 30 minutes at 85 ºC. Then the nanowire arrays were thermally oxidized at 1,000 ºC for 1 hour, followed by etching in BHF for 3 minutes. Subsequently, the Si nanowire arrays were soaked in gold etchant (Transene) NATURE NANOTECHNOLOGY 3

4 again for 30 minutes at 85 ºC. The silicon nanowire array samples went through the same boron doping process as used with the single nanowires, as follows. First, boron was pre-deposited at the nanowire surface at 750 ºC for 1 hour, with 1% BCl 3 in argon as the precursor and 10% hydrogen in argon as the reducing agent. Second, the boron atoms at the nanowire s surface were driven into the nanowire at 1,000 ºC for 5 hours in vacuum, followed by an hour s thermal oxidation at 950 ºC. As a result, presumably the silicon nanowire arrays have the same radial boron doping profile with the single nanowires in this work. After the boron doping, for p-si nanowire array devices, the chip was directly coated with 5 nm of TiO 2 by atomic layer deposition (Picosun ALD) after the thermal oxide was etched in BHF solution. For n + p-si nanowire array devices, the nanowire arrays were n + doped using a method similar to that applied to the single nanowires. A silicon handle wafer was spin-coated with arsenic-containing spin-on dopant (SOD) (Filmtronics, Inc.) and baked at 150 ºC for 30 minutes. After the thermal oxide was etched in BHF solution, the device chip was placed on the SOD-coated handle wafer, and annealed (rapid thermal annealing) at 900 ºC for 3-4 minutes in a N 2 atmosphere to form an n + layer at the nanowires surface. Then the device chip was soaked in BHF solution for 30 seconds to remove the thin oxide formed during the n + doping process. Subsequently, the n + p-si nanowire array device was coated with 5 nm of TiO 2 by atomic layer deposition (Picosun ALD). Both types of nanowire array devices were sputtered with platinum as the HER catalyst. For the electrode fabrication, ohmic contact to the device chip was made by rubbing Ga-In eutectic on its back side. Then the chip was fixed on Ti foil with conductive silver paint and carbon tape, resulting in good electrical connections. After that, the Si nanowire array samples were sealed using the nail polish. PEC measurements were carried out in 0.5 M H 2 SO 4 under one-sun illumination (AM 1.5G, 100 mw/cm 2 ) with a 20 mv/s sweep rate (single sweep, from positive to 4 NATURE NANOTECHNOLOGY

5 SUPPLEMENTARY INFORMATION negative). Multiple scan cycles were repeated in order to get reproducible PEC behavior. The PEC performance, particularly photovoltage, of the Si nanowire array samples is comparable with the worst single nanowire (Supplementary Fig. 7). This observation applies to both p-si and n + p-si devices (Supplementary Fig. 8). Calculation and comparison of the photovoltage and the energy conversion efficiency The V oc of the Si nanowire array device was determined relative to the RHE potential in acidic electrolyte. For single-nanowire devices, the photocurrent onset potential (V oc ) is defined as the potential vs. RHE where is -0.5 pa, considering the instrumental noise. Here is the difference between the photocurrent and the dark current of the single nanowire device. For both array and single nanowire devices, the V oc drifts slightly among multiple scan cycles, and the median value is selected as the representative to report here. The energy conversion efficiency ( ) is calculated based on equation S1. Eq. S1 Here is the current density at maximum power, is the potential vs. RHE at maximum power, and is the power density of the incident sunlight (mw/cm 2 ). The statistical comparison of the V oc and between the single nanowires and nanowire arrays are listed in Supplementary Table 1. For both p-si and n + p-si devices, the distributions in V oc and of the single nanowire devices are broader than those in the nanowire array devices. Equivalent circuit modeling The photoelectrochemical proton-reducing Si cathode is analyzed under the framework of an equivalent circuit model as applied in photovoltaic devices 35. In such a model, the band-bending NATURE NANOTECHNOLOGY 5

6 and subsequent generation of photocurrent is modeled as a standard diode junction with an apparent ideality factor n, along with the reverse saturation current density j s and photocurrent density j ph. In addition, the electrochemical reaction is considered as a series resistance R s, which is dictated by the classic Butler-Volmer equation 36. In summary, the governing equation for current density j in the photocathode is: 1 Eq. S2 Here we assume that the differences in PEC performance of individual single nanowires, in particular the photovoltage (V oc ), originate from the differences in n. This is reasonable given the variation of unintentional dopants that might exist, as noted in previous literature 24, 25. Moreover, we assume that among all of these single nanowire devices, V oc follows a truncated normal distribution, which means that the value of V oc follows a normal distribution within the range that its photovoltage (V oc ) is physically reasonable, i.e. 0,1.1 :,, 2 0, 1.1 Eq. S3 Here A is the normalization coefficient, which in practice is close to the one for a normal distribution. The upper bound of 1.1 is chosen based on the 1.1 ev band gap of silicon. The above assumptions lead to the relationship between n, an implicit parameter, and V oc, a value that has been reliably measured experimentally in this report: 1 Eq. S4 Subsequently, n follows the similar distribution as V oc does: 6 NATURE NANOTECHNOLOGY

7 SUPPLEMENTARY INFORMATION Eq. S5 The ensemble array devices, each of which is composed of a large number of single nanowire devices linked in parallel, can be considered mathematically as the probability-weighted integration of every possible configuration of single nanowires. Therefore, the j-v characteristics of an ensemble device could be written as:. Eq. S6 At open circuit condition (j = 0), the above expression transforms into:. Eq. S7 Here, V oc,array is the observed open-circuit potential of the array devices. Therefore, a correlation between V oc,array of ensemble devices with the distribution of V oc in single nanowire devices is established (Supplementary Fig. 9). With known statistics of V oc in single nanowire devices, it is possible to calculate the theoretical V oc,array of an array devices, assuming that the measured statistics from single nanowire devices faithfully represent the properties wires within the array samples. Fabrication and PEC measurement of planar n + p-si The p-si (boron) wafers (ρ ~10-30 Ω. cm) were obtained from Addison Engineering, Inc. The wafer s surface was n + doped using a method similar to that applied to the single nanowires. A silicon handle wafer was spin-coated with arsenic-containing spin-on dopant (SOD) (Filmtronics, NATURE NANOTECHNOLOGY 7

8 Inc.) and baked at 150 ºC for 30 minutes. Then the device wafer was placed on the SOD-coated handle wafer, and annealed (rapid thermal annealing) at 1,000 ºC for 1.5 minutes in a N 2 atmosphere to form an n + layer at the wafer s surface. Subsequently, the wafer was conformally coated with 5 nm of TiO 2 by atomic layer deposition (Picosun ALD). After that, the wafer was diced into small chips (1cm 1cm), which were subsequently sputtered with platinum as the HER catalyst. For the electrode fabrication, ohmic contact to the chip was made by rubbing Ga- In eutectic on its back side. Then the chip was fixed on Ti foil with conductive silver paint and carbon tape, resulting in good electrical connections. After that, the planar Si samples were sealed using the nail polish. PEC measurements were carried out in 0.5 M H 2 SO 4 under one-sun illumination (AM 1.5G) with a 20 mv/s sweep rate (single sweep, from positive to negative). The resulting PEC characterization shows that the current density and the corresponding surface photo-generated electron flux of the planar n + p-si are -20 ma/cm 2 and 1,240 electrons/ (nm 2 s) at 0V vs. RHE (Supplementary Fig. 10). Quantification of Flux wire Because of the well-defined nanowire geometry, the single nanowire can be considered to be a cylinder whose length ( ) and diameter ( ) were characterized precisely through the scanning electron microscope images. As a result, the electron flux normalized to the geometric crosssectional area ( ) and the electron flux through the nanowire s actual surface ( ) can be defined by equation S8 and equation S9, respectively. 4 Eq. S8 8 NATURE NANOTECHNOLOGY

9 SUPPLEMENTARY INFORMATION 4 Eq. S9 The recorded photocurrent of individual nanowires would reach saturation when negative enough bias was added, indicative of a light-activated process limited by the number of incident photons. Correspondingly, the saturated is systematically quantified as a function of the length (L) and diameter (D) of individual nanowires. The 3D curve (z axis: ; x axis: L; y axis: D) is illustrated, and the projected views are displayed in Supplementary Fig. 11. In general, decreases as the diameter becomes smaller or the length becomes longer. Estimation of PEC current density as a function of Si thickness The theoretical PEC current density as a function of Si thickness (here the nanowire s length) needs to be calculated to normalize. The calculation is carried out based on silicon s optical properties in water as well as the spectrum of simulated sunlight (AM 1.5G). In detail, the calculation processes are as follows: first, the reflection percentage of the incident radiation ( ) at the Si-water interface is expressed by equation S10. Here and are the refractive indices of silicon and water, respectively. And and are the angles that the reflected and refracted rays make to the normal of the interface, respectively. Eq. S10 Considering that the vertically grown silicon nanowires were illuminated from above, and can be regarded as approximately 0. As a result, equation S10 can be simplified to equation S11. Eq. S11 NATURE NANOTECHNOLOGY 9

10 Next, the incident radiation that is not reflected will be divided into two parts: absorption and transmission. For each wavelength, the percentage of power absorbed can be determined by the Beer-Lambert law, as described in equation S12, where is the percentage of absorption, is the wavelength-dependent absorption coefficient in Si and L is the thickness of Si. 1 1 Eq. S12 Assuming that the internal quantum efficiency is 100%, then the absorbed power (, W/m 2 ) produces a current density (, ma/cm 2 ) that can be determined by equation S13, where is the light velocity, is the Planck s constant and is the wavelength of the incident photon. Here the absorbed power ( ) is defined by the one-sun illumination spectrum (AM1.5G) multiplied by from Eq. S Eq. S13 As a result, the PEC current densities were calculated for each wavelength and summed over all wavelengths in the solar spectrum for different Si thicknesses (Supplementary Fig. 12). Practically, the measured current density of single-nanowire device may deviate from the theoretical calculation for three reasons: 1) the practical quantum efficiency of Si nanowires may not be 100%, leading to a decreased current density; 2) there may be a slight misalignment ( 3º) between the incident light and the axis of the single nanowire, resulting in an increased absorption and measured current density; 3) the resonances supported in individual nanowires can increase the absorption at specific wavelengths 37. Correlation between the effective reduction of electron flux and 10 NATURE NANOTECHNOLOGY

11 SUPPLEMENTARY INFORMATION To better explain how functions to dilute the electron flux, the effective reduction of electron flux ( ) is introduced by equation S14, where is defined by equation S15. Eq. S14. 1 Eq. S15 is the ratio between the estimated photogenerated electron flux of a planar Si counterpart (with thickness L) and the measured. of the singlenanowire devices are plotted vs. in Supplementary Fig. 13. The black line in Supplementary Fig. 13 outlines when equals to. Overall, follows the linear trend with roughness factor of the individual nanowires. Overpotential estimation In the PEC process, the electrocatalysts should have a certain turnover frequency (TOF) to handle the photo-generated electron flux at the surface of the photoelectrode. To reach a certain TOF for a specific electrocatalyst, an electrochemical overpotential is necessary based on the Butler-Volmer equation. Typically, a large photo-generated electron flux will require a large TOF and thus a large overpotential. Because of the well-defined nanowire geometry, can be quantified precisely. Take the specific single-nanowire device mentioned in Fig. 4a (main text) for example. reaches 13 electrons/ (nm 2 s) at 0V vs. RHE. In comparison, for the planar n + p-si, is 1,240 electrons/ (nm 2 s) at 0V vs. RHE. As a result, loading electrocatalysts on the Si-nanowire will lead to a lower TOF requirement and a reduced NATURE NANOTECHNOLOGY 11

12 overpotential compared to loading electrocatalysts on the planar-si. Supplementary Table 2 lists several reported typical electrocatalysts with quantitative TOF measurements, and summarizes the electrochemical overpotentials needed for these electrocatalysts to reach different TOFs. Here we assume that all the reported electrocatalysts cover the surface of semiconductor photoelectrodes, and the density of the surface catalytically active atom is atom/cm 2 or equivalently 10 atom/nm 2 for all listed electrocatalysts. So for the single-nanowire photoelectrode, the photo-generated electron flux of 13 electrons/ (nm 2 s) will correspond to a 1.3/n s -1 TOF (n being the number of electrons for a particular catalytic reaction) requirement on the electrocatalysts. In comparison, for the planar-si, the photo-generated electron flux of 1,240 electrons/ (nm 2 s) will correspond to a 124/n s -1 TOF requirement on the electrocatalysts. As a result, our calculation results show that the nanowire geometry can reduce the necessary overpotential by about 40 mv for platinum 38, one of the best proton reduction electrocatalysts. 39, For more earth-abundant but less active proton reduction electrocatalysts, like the MoS 40 2, the nanowire geometry can reduce the overpotential by more than 100 mv, indicating a large benefit. For more complicated and sluggish reactions, like the CO 2 reduction 41, the reduced overpotential due to the nanowire s large surface area can be much more significant. 12 NATURE NANOTECHNOLOGY

13 SUPPLEMENTARY INFORMATION Supplementary figures: Supplementary Figure 1. The complete fabrication process, starting from the original SOI substrate, and resulting in the individually addressable single-nanowire photoelectrode. For p- type devices, step 9 is skipped and the process directlyy goes to step 10. NATURE NANOTECHNOLOGY 13

14 Supplementary Figure 2. The scanning electron microscope (SEM)) images show the significant difference between the as-grown nanowire (a) and the nanowire afterr PEC measurement (b). This difference suggests that the platinum nanoparticles, which appear as bright spots in the image, were successfully loaded on the nanowire s surface during the electroless deposition NATURE NANOTECHNOLOGY

15 SUPPLEMENTARY INFORMATION Supplementary Figure 3. High-resolution transmission electron microscope (TEM) images of the Si nanowire/ /Pt interface. There is an amorphous SiO 2 layer at the interface before (a) and after (b) the PEC measurement. NATURE NANOTECHNOLOGY 15

16 Supplementary Figure 4. Electrochemical potential calibration of the platinum electrode. The electrochemical potential of the platinum electrode is immediately calibrated with Ag/AgCl standard reference after the single-nanowire photoelectrochemistry measurement. The open- circuit voltage measurement lasted for 1 hour, twice thee time for a typical device measurement time. 16 NATURE NANOTECHNOLOGY

17 SUPPLEMENTARY INFORMATION Supplementary Figure 5. I-V characteristic of a single-nanowire device in the dark. The current starts to increase faster at approximately 0V vs. RHE, indicative of a proton-reduction reaction. NATURE NANOTECHNOLOGY 17

18 Supplementary Scheme 1. Schematic illustration of the band diagram for (a) a p-si nanowire and (b) an n + p-si nanowire in contact with acidic electrolyte. After contact, the Fermi level (E F ) of the Si will equilibrate with the Fermi level (E F ') in the acidic electrolyte, leading to the band bending in the Si. For the case of an n + p-si nanowire, the n + /p buried junction will significantly increase the built-ithe photo-generated electrons can tunnel through such a barrier, given the thin depletion layer (~5 nm) resulting from thee high arsenic concentration (~10 20 /cm 3 potentiall (ϕ bi ). Despite the upward band bending at the n + -Si/electroly te interface, ). 18 NATURE NANOTECHNOLOGY

19 SUPPLEMENTARY INFORMATION Supplementary Figure 6. SEM images of Si nanowire arrays. (a) Low-magnification and (b) high-magnification SEM images show that the Si nanowire arrays are vertically grown on a p-si substrate oriented <111>. NATURE NANOTECHNOLOGY 19

20 Supplementary Figure 7. Characteristic of the normalized J vs. V. The n + p-si nanowire array samples are compared with the n + p-si single-nanowire samples. The normalized J is defined as J/J saturated d, where J satu urated is the saturated photocurrent density. Multiple n + p-si nanowire array samples were characterized to yield reproducible PEC performancee (dashed curves in the plot). Generally, the normalized PEC characteristic of the array device iss comparable with the worst single-nanowire device. 20 NATURE NANOTECHNOLOGY

21 SUPPLEMENTARY INFORMATION Supplementary Figure 8. J-V characteristic comparison between the single-nanowire devices and nanowire arrays devices. Both (a) n + p-si and (b) p-si samples are compared. Generally, the photocurrent onset potential of the arrayy devices iss comparable with the worst single-nanowire devices. NATURE NANOTECHNOLOGY 21

22 Supplementary Figure 9. Comparison of photocurrent onsett potentials (V oc ) between ensemble and single nanowires. Two sorts of Si nanowires with different radial doping profiles, p-si (blue) and n + p-si (red), are considered here. Thee statistically averaged ensemble results (square dots) are compared with the ones from the corresponding single nanowires (circular dots). Assuming a truncated normal distribution, the distribution function of V oc from each case is plotted (solid lines). Furthermore, such a distribution n function of V oc from single nanowire measurement allows for a theoretical prediction of the V oc for ensemble nanowire arrays (dashed lines), which is calculated based on the equivalent circuit model. The predicted ensemble V oc matches well with experimental ones, and both are close to the values of the worst-performing single nanowire devices. Consequently, a correlation of V oc between the ensemble and single nanowires is established. 22 NATURE NANOTECHNOLOGY

23 SUPPLEMENTARY INFORMATION Supplementary Table 1. Comparison of the photocurrent onsett potential (V oc ) and energy conversion-efficiency (η) between different types of Si wire samples. All the Si wire samples are synthesized by the VLS mechanism. NATURE NANOTECHNOLOGY 23

24 Supplementary Figure 10. The PEC characteristics of n + p planar Si. The PEC measurement was carried out in 0.5 M aqueous H 2 SO 4 solution (200 mv s -1, single sweep from positive to negative potential). Under one-sun illumination (100 mw/cm 2, AM 1.5G), the current density and the correspondin ng photogenerated electron flux are -20 ma/cm 2 and ~1,240 electrons/ (nm 2 s) at 0 V vs. RHE. 24 NATURE NANOTECHNOLOGY

25 SUPPLEMENTARY INFORMATION Supplementary Figure 11. The projected 3D view of saturated as a function of (a) or (b). NATURE NANOTECHNOLOGY 25

26 Supplementary Figure 12. The calculated theoreticall PEC current density as a function of silicon thickness under sunlight illumination (AM 1.5G).. 26 NATURE NANOTECHNOLOGY

27 SUPPLEMENTARY INFORMATION Supplementary Figure 13. Correlation between the effective reduction of electron flux and. NATURE NANOTECHNOLOGY 27

28 Supplementary Table 2. Overpotential calculation. The electrochemical overpotentials are calculated and compared between loading electrocatalysts on Si-nanowire surface and planar-si surface. The results show that the overpotential can be reduced if loading electrocatalysts on the nanowire s surface. And this is especially significant for less active electrocatalysts. 28 NATURE NANOTECHNOLOGY

29 SUPPLEMENTARY INFORMATION Supplementary References: 31. Dai, P. et al. Solar hydrogen generation by silicon nanowires modified with platinum nanoparticle catalysts by atomic layer deposition. Angew. Chem. Int. Ed. 52, (2013). 32. Kasem, K. K. & Jones, S. Platinum as a reference electrode in electrochemical measurements. Platin. Met. Rev. 52, (2008). 33. Belew, W. L., Fisher, D. J., Kelley, M. T. & Dean, J. A. A cell design for minimizing ir error in controlled-potential polarography of high specific resistance solutions. Instrum. Sci. Technol. 2, (2008). 34. Yuan, G. et al. Understanding the origin of the low performance of chemically grown silicon nanowires for solar energy conversion. Angew. Chem. Int. Ed. 50, (2011). 35. Green, M. A. Solar cells: operating principles, technology, and system applications (Univ. New South Wales, 1998) 36. Bard, A. J., Faulkner, L. R. Electrochemical methods: fundamentals and applications, 2 nd ed., John Wiley&Sons, Inc., (2000). 37. Kempa, T. J. et al. Coaxial multishell nanowires with high-quality electronic interfaces and tunable optical cavities for ultrathin photovoltaics. Proc. Natl. Acad. Sci. U.S.A. 109, (2012). 38. Kita, H., Ye, S. & Gao, Y. Mass transfer effect in hydrogen evolution reaction on Pt singlecrystal electrodes in acidc solution. J. Electroanal. Chem. 334, (1992). 39. Kibsgaard, J., Jaramillo, T. F. & Besenbacher, F. Building an appropriate active-site motif into a hydrogen-evolution catalyst with thiomolybdate [Mo 3 S 13 ] 2- clusters. Nature Chem. 6, (2014). 40. Shin, S., Jin, Z., Kwon do, H., Bose, R. & Min, Y. S. High turnover frequency of hydrogen evolution reaction on amorphous MoS 2 thin film directly grown by atomic layer deposition. Langmuir 31, (2015). 41. Kim, D., Resasco, J., Yu, Y., Asiri, A. M. & Yang, P. Synergistic geometric and electronic effects for electrochemical reduction of carbon dioxide using gold-copper bimetallic nanoparticles. Nature Commun. 5, 4948 (2014). NATURE NANOTECHNOLOGY 29