Nanorod length-dependent photodriven H₂ production in 1D CdS–Pt heterostructures

Colloidal quantum confined semiconductor-metal nano-heterostructures are a promising class of photocatalysts for solar energy conversion.^1–9 In these photocatalysts, the semiconductor domain serves as the light absorber and the metal as the catalyst. Such photocatalysts combine the semiconductor’s superior light absorption and charge transport properties with the metal’s superior catalytic activity and selectivity.^4,10,11 Furthermore, both domains can be independently tuned to enhance the photocatalytic performance of the heterostructure.^5,12–19 Among various semiconductor/metal heterostructures, metal-tipped colloidal semiconductor nanorods^20,21 have attracted extensive interest because they have been reported to have high quantum efficiencies of light driven H₂ generation,^13,22,23 and their charge separation and recombination processes can be examined spectroscopically to advance mechanistic understanding and rational design.^{3,13,22,24–26} It has been reported previously that the light-driven H₂ generation performance of metal-tipped nanorods depends on the pH of the solution,^1,13,23 the nature of the sacrificial electron donors,^3,27 the size of the metal domain,^15,22,28 and the length of the nanorods.¹² Although these studies have reported interesting dependences of the performances on the semiconductor and metal domains, the mechanisms of the dependences are poorly understood. The overall light driven H₂ generation process involves multiple elementary charge separation and recombination steps in the semiconductor and across the semiconductor/metal interface, as well as proton-coupled electron transfer reactions at the catalytic center. The change in the semiconductor or metal domains can often have effects on the multiple competing processes in the overall reaction. As a result of these complexities, the mechanisms for the morphological dependence of the observed H₂ generation efficiencies are not fully understood, hindering the rational design of these photocatalysts.

In this work, we use Pt tipped CdS nanorod (CdS–Pt NR) heterostructures as a model system to study the mechanism by which the CdS rod length affects their light-driven H₂ production. We observe that the quantum efficiency (QE) of light driven H₂ generation (⁠$QEH2)$ initially increases with rod length from 4.6 ± 0.9 to 27.8 ± 4.2 nm, but reaches a plateau value of ∼30% from a rod length of 27.8 ± 4.2 to 88.9 ± 15.8 nm. Because the CdS rod length can affect the rates of the initial electron transfer (ET) from the CdS to the Pt and the lifetime of the charge separated state (with an electron in the Pt domain and a hole in the CdS domain), the length dependences of these rates were directly measured. We propose a simplified kinetics model that considers key processes from the initial photon absorption to the final catalytic reaction and assumes that the overall light driven H₂ generation QE is the product of the QEs of these elementary steps. Using this model, we examine how the CdS rod length affects the quantum efficiencies of these elementary steps and the overall H₂ generation performance.

CdS quantum dots (QDs) and four NR samples of different rod lengths were prepared according to reported literature procedures with slight modifications, and the Pt tips were deposited by thermal reduction as described in SI1.²¹ The as-synthesized CdS and CdS–Pt samples have phosphoric acid as native capping ligands and are soluble in organic solvents. To transfer these samples into an aqueous solution, the surface ligands were exchanged for 11-mercaptoundecanoic acid (11-MUA), as described in SI1. The CdS and CdS–Pt samples are designated as CdS_xnm and CdS_xnm–Pt, respectively, with the subscript x representing nanorod length or x = QD to indicate CdS QD and QD–Pt samples. The first exciton absorption peaks of CdS QDs and NRs, determined by the radial confinement,²⁹ are around 450–470 nm, as shown in Fig. 1(a). The absorption spectra of CdS–Pt can be well fitted to the sum of CdS absorption bands and a broad Pt d-sp interband absorption feature,³⁰ as shown in Fig. 1(a). Transmission electron microscope (TEM) images of CdS and CdS–Pt are shown in Figs. S1 and 1(b)–1(f), respectively. The diameter of the QD is 4.6 ± 0.9 nm. The CdS and CdS–Pt NR samples have four different rod lengths of 13.8 ± 2.3, 27.8.±4.2, 66.6 ± 6.6, and 88.9 ± 15.8 nm, but similar CdS diameters in the range of 4.1 ± 0.4 to 4.7 ± 0.8 nm and similar Pt tip diameters in the range of 3.3 ± 0.7 to 4.0 ± 1.0 nm. Around 70% of the nanorods are singly tipped, 20% are doubly-tipped, and 10% have no tips. The detailed statistics of CdS diameter, rod lengths, Pt diameters, and number of Pt tips are listed in Table S1. Thus, the majority of the samples can be considered nanorods of similar CdS diameters and Pt tip sizes but different CdS rod lengths with a single tip at the end, as schematically shown in Fig. 1(g).

The photo-catalytical H₂ production measurements using 11-MUA capped CdS–Pt nanocrystals were carried out under continuous 405 nm LED light illumination (power density of ∼100 mW/cm²) for ∼1–2 h. The aqueous solution pH was tuned to 12 with KOH, and L-cysteine (3 mg/ml) was added as a hole scavenger.³¹ The absorbances of all samples at 405 nm were controlled to be in the range of ∼0.3 to 0.4. The generated H₂ amount was quantified by gas chromatography Further details of the measurement are provided in SI2. Figure 2(b) shows the H₂ production amount over time. The linear time-dependence indicates stable H₂ production rates over time. Three measurements were taken for each sample to obtain the H₂ generation rate and its error bar. The slopes of linear fit (i.e., H₂ generation rate, $kH2$⁠) first increases with rod length, then stays similar for longer rods. The internal quantum efficiencies of H₂ generation (⁠$QEH2)$ are calculated as $QEH2=2kH2khv,CdS*100%$⁠. k_hv,CdS is the photon absorption rate by CdS NR at 405 nm in the CdS–Pt samples. Although both Pt tips and CdS NRs absorb at 405 nm, photons absorbed in Pt do not lead to H₂ generation due to the fast cooling of carriers in Pt. In addition, our previous work has shown that CdS nanorods without tips also have a negligible contribution to H₂ production.³¹ Thus, in calculating the internal quantum efficiencies of H₂ production (⁠$QEH2)$⁠, only the photons absorbed by tipped CdS NRs are considered, following the procedure described in SI2. As shown in Fig. 2(c), $QYH2$ of tipped nanorods increases from 0.2% ± 0.0% to 28.9% ± 0.4% when rod length increases from 4.6 ± 0.9 nm (QD, the rod length of the QD sample is taken as its diameter) to 28 ± 3 nm, and then remains nearly constant, within the range of 30% ± 10%, for longer rod lengths (67 ± 7 and 89 ± 15 nm).

The overall light-to-H₂ production from CdS–Pt NRs consists of multiple charge separation, charge recombination (CR) steps, and catalysis at the Pt surface, which will be discussed below. To understand the length-dependent H₂ production discussed above, transient absorption (TA) spectroscopy was utilized to measure the rod length dependences of the initial electron transfer rate from the CdS rod to the Pt tip and the lifetime of the resulting charge-separated state in CdS–Pt heterostructures. We first discussed the rod length dependence of ET kinetics from the CdS NR to the Pt tip, measured following previous reports.^1,3,24 TA studies of CdS and CdS–Pt NRs were conducted under the same pH and hole scavenger conditions as the light-driven H₂ generation measurement described above using 400 nm pulse laser excitation. The details of the TA measurement are provided in SI3. The TA spectra of selected CdS (CdS_QD, CdS_28nm, and CdS_89nm) and the corresponding CdS–Pt samples are shown in Figs. 3(a) and 3(b), respectively, and the spectra of the remaining samples are shown in Figs. S2 and S3. These spectra show similar features: (a) a bleach of the 1Σ exciton band (XB) centered at 450–470 nm due to the electron state filling at the conduction band (CB) edge, which reduces 1Σ exciton absorption; (b) a photoinduced absorption at ∼420 nm that is attributed to the red-shift of a higher energy transition due to biexciton interaction.²⁴ The slight time-dependent red-shift of the XB peak position in CdS_89nm spectra is likely due to its ununiform rod diameter, which leads to exciton localization into the bulb region.³² As shown in Fig. 3(c), the XB decay in free CdS NRs shows similar kinetics, independent of the rod length. Because the XB feature reflects the population of photogenerated CB edge electrons in the CdS NR, its decay in bare CdS reflects the recombination of photogenerated electrons with the holes in the NR.²⁴ The XB bleach recovery in Fig. 3(c) can be fit to a stretched exponential function A(t) = A₀ exp[−(t/τ)^β], with τ = 15.2 ± 0.3 ns and β = 0.42 ± 0.01, corresponding to an average exciton lifetime of 45.7 ± 0.9 ns. The TA spectra of CdS–Pt NRs [Fig. 3(b)] show similar spectral features as those in CdS NRs. However, their XB decays on a much faster time scale [Fig. 3(d)], which can be attributed to ET from the CdS rod to the Pt tip.²⁴ The kinetics of XB decay in CdS–Pt NRs can be fit by a multiexponential function, as described in S3.3 and Table S3. From the fit, the XB decay half-life time is determined to be 0.7 ± 0.1 ps in CdS_QD-Pt and increases to 1.8 ± 0.3, 23.4 ± 3.6, 52.9 ± 7.9, and 170.2 ± 29.5 ps for CdS_14nmPt, CdS_28nmPt, CdS_67nmPt, and CdS_89nmPt rods, respectively, monotonically increasing with the rod length, reflecting decreasing ET rates.

Previous studies have shown that the presence of holes in the nanorod shifts the exciton transition energy and leads to a derivative like TA feature, indicative of the formation of the charge separated states and this feature can be used to follow the lifetime of the charge separated state.^3,24 Interestingly, with MUA ligand capped CdS–Pt NRs under the light driven H₂ conditions described above, such a charge separated state signal was not observed (Fig. S4). To get insights into the length dependence of the CR time of the electron in the Pt tip with the hole in CdS in aqueous solution, we carried out a study on phosphoric acid capped CdS–Pt NRs dissolved in toluene, where the hole recombination displays an observable feature in the TA spectra. The TA spectra of CdS–Pt from 10 ns to 200 µs are shown in Fig. 4(a) for CdS_QDPt (upper panel), CdS_28nmPt (middle panel), and CdS_89nmPt (lower panel) and in Fig. S5 for CdS_14nmPt and CdS_67nmPt. After the completion of ET from the CdS to Pt on a <1 ns time scale, a second derivative feature of the exciton band is observed beginning at 1–10 ns. This derivative feature has been attributed to the trapped hole in CdS, which shifts the CdS exciton absorption band.²⁴ The decay of this derivative feature has been attributed to the recombination of the electron in Pt with the hole in CdS.²⁴ A comparison of the recombination kinetics of CdS–Pt at different rod lengths [Fig. 4(b)] shows that the CR lifetime increases with the rod length. With the rod length increasing from 4.6 ± 0.9 to 143.8 ± 2.3 and 27.8 ± 4.2 nm, the amplitude-weighted average lifetime of the charge separated state increases from 4.7 ± 0.4 to 25.7 ± 3.3 µs and 149 ± 34 µs, after which the average lifetime increases negligibly at longer rod lengths. This length-dependent trend is consistent with the observed $QYH2$ in Fig. 2, implying CR time may be the determining factor for the length-dependent $QYH2$⁠. For 11-MUA capped CdS–Pt in aqueous solution under light-driven H₂ generation conditions, ET from the CdS to Pt does not lead to the charge separated state signal discussed above, as shown in Fig. S4. This is likely caused by fast hole transfer to the surface 11-MUA ligand or large dielectric screening by water, which reduces the Stark effect signal of the CdS exciton band.

The overall light-driven H₂ generation process in CdS–Pt NRs involves multiple forward steps: hole transfer (HT) from the CdS to surface ligands (L), ET from the CdS to Pt, hole removal from the surface ligand to L-cysteine in solution or hole scavenging (HS), and water reduction (WR) at Pt. Each of these steps competes with a backward charge recombination process. Figure 5(a) shows the schematic energy levels of electrons and holes involved in these steps, and Fig. 5(b) shows a simplified kinetics model that considers key intermediates and their competing forward and backward processes.³¹ As shown in Fig. 5(b), the absorption of a photon generates an exciton in L-CdS-Pt NRs (L = 11-MUA), which undergoes fast trapping of the VB hole on a sub-picosecond time scale to form L-CdS*-Pt.^24,33 In L-CdS*-Pt, the transfer of the trap hole (k_HT) to the MUA ligand to form L⁺-CdS⁻-Pt competes with the electron–hole recombination (k_CR1) process within CdS* to determine the hole transfer QE $(QEHT=kHTkHT+kCR1)$⁠. QE_HT is estimated to be nearly 100% because k_HT, ∼ (100 s ps)⁻¹, is much larger than k_CR1 (∼10 s ns)⁻¹.^1,3 k_HT and k_CR1 [Fig. 3(c)] are length independent, giving rise to a length independent QE_HT. In L⁺-CdS⁻-Pt, ET from the CdS to the Pt tip (k_ET) forms a charge separated state (L⁺-CdS-Pt⁻), which competes with rod-length independent CR (k_CR2) [Fig. 3(c)]. Because k_ET is dependent on the rod length, the ET QE (⁠$QEET=kETkET+kCR2)$ is rod-length dependent and can be determined directly from the transient absorption kinetics shown in Fig. 3(d). In L⁺-CdS-Pt⁻, hole transfer from the oxidized surface ligand to the solution hole scavengers (with a rate constant of k_HS) forms a long-lived L-CdS-Pt⁻ species that can proceed on to the slow water reduction steps, and the HS step competes with the rod length dependent CR (k_CR3) to give rise to rod length dependent hole scavenging QE (⁠$QEHS=kHSkHS+kCR3)$⁠. Finally, water reduction QE (QE_WR=) should not vary with the rod length as this step involves only the electron in the Pt particle and the hole in the hole scavenger. Although water reduction is a multiple step process consisting of two-proton reduction, we have assumed that there exists a rate limiting step that controls its QE.

Because QE_HT and QE_WR are rod-length independent, in Eq. (1) we have grouped the product of the rod length independent QEs into a constant C(=QE_HT ∗ QE_WR).

A previous study of light driven MV²⁺ reduction using mercaptopropionic acid MPA capped CdS QDs provided an approach to estimating the k_HS via static radical generation yield.³⁵ Under the conditions of pH 7 and 3-mercaptopropionic acid (MPA) as hole scavenger, the quantum yield of MV²⁺ reduction was determined to be 12.61% ± 1.55% and the radical lifetime was ∼5.1 µs (or k_HS + k_CR3 = 0.19 µs⁻¹). Because of the unity QEs for both electron and hole transfer steps under those conditions, the radical generation QEs are determined solely by QE_HS. Using Eq. (3), k_HS and k_CR3 can be determined to be 1/(40 µs) and 1/(5.8 µs), respectively. Because the hole scavenging rate should not be affected by the presence of the Pt tip, we assume that this k_HS value can be used as an estimate of the hole scavenging rate in the CdS–Pt samples under light driven H₂ generation conditions, although it should be noted that k_HS depends on both the identity and concentration of the hole scavengers as well as the identification of the NR surface ligand layers.^36,37 As an estimate of the charge recombination rate in L⁺-CdS-Pt⁻ in water, the measured values of k_CR3 of CdS⁺-Pt⁻ in an organic solvent [Fig. 4(d)] are used. Although in the former, the hole is localized on the thiol group of the surface ligand, it resides at a surface trap site (likely the surface S atom) in the latter.

Using the estimated values of k_HS(=1/40 µs) and k_CR3 [Fig. 4(d)], we calculate the $QEHSL$ as a function of rod length following Eq. (3). As shown in Fig. 6(b), the trend of the calculated $QEHSL$ agrees well with the trend of $QEHSL$ estimated from $QEH2$ and QE_ET, according to Eq. (1). Although the calculated $QEHSL$ values should be considered a rough estimate because of the approximations involved in estimating k_HS and k_CR3, this analysis provides useful insight on how the competition of these processes leads to the CdS rod length dependent $QEHSL$ and $QEH2$⁠. Previously, it was reported that the H₂ production yield increases with the rod length in CdSe@CdS–Pt NRs from 20 to 60 nm, and increasing the charge separation distance may account for this improvement.¹² This result is consistent with our trend from QD to 28 nm, but not for the longer rods. Our analysis shows that the loss caused by the competition between charge recombination and hole scavenging depends on the hole scavenging rates. Under different hole scavenging conditions (the identity and concentration of the hole scavenger), the length dependence may change because of the variations in the rates of these competing processes.

The mechanism of electron transfer from a quantum confined semiconductor to an electron acceptor has been previously investigated in colloidal quantum dots via a shell-thickness dependence study,³⁸ which reports that the electron transfer rate from a CdSe/ZnS quantum dot to anthraquinone decreases exponentially with the shell thickness, consistent with the electron density that extends from the CdSe core to the ZnS surface. This indicates that in the short-distance regime, the electron transfer mainly undergoes tunneling pathway. However, the tunneling mechanism predicts an exponential decrease in electron transfer rate, which may be too slow in the long-distance regime. In conjugated molecular systems, a transition from tunneling mechanisms to activated hopping/diffusion has been reported.^39,40 In quantum confined nanorods, hole trapping to the surface states or surface ligands happens at sub-picosecond²⁴ to 100 s ps¹, respectively, which may reduce the electron–hole Coulombic binding. Weakening the binding energy may facilitate electron hopping along the nanorod and make hopping a preferential mechanism for ET. Several studies have already reported hopping-mechanism-dominant long-range electron transfer.^1,41 For example, the exciton localization in a CdSde@CdS nanorod can be well-expressed by solving the boundary-confined diffusion equation. Similarly, electron transfer from CdSe@CdSe to the Ni tip was also reported to follow the diffusion model with a diffusion constant of 1.6 cm²/s.⁴² 

D is the diffusion constant of an electron. L is the total rod length. This series is asymptotic to a single exponential because higher-level terms (n > 0) vanish quickly with both decay rates increasing and amplitudes decreasing as ∼(n + 1/2)². The prominent term in the series is the n = 0 term. However, as shown in Fig. 3(d) and Table S3, the measured charge separation kinetics cannot be described by single exponential decay functions. This non-exponential behavior has been reported widely in colloidal nanocrystals,^3,24,32,44 although its origin is not fully understood. In our system, it may be attributed to the broad distribution of electron diffusion constants caused by trap states or rod lengths. Equation (4) predicts that the rate of electron transfer should scale with D/L². If we assume the distributions of diffusion constants are independent of rod length, the charge separation kinetics of NRs of different lengths should agree with each other if they are plotted against a scaled time t*/L². To test this model, we rescaled the data shown in Fig. 3(d) against the scaled delay time T = t*/L². L is initially set as the rod length. To get better scaling, L is slightly varied around the average rod length, as shown in Table S6. The rescaled TA kinetics are plotted in Fig. S6, which shows that for long nanorods, the scaled kinetics agree well with each other. This suggests that the length dependence of the charge separation kinetics in long nanorods is consistent with the hopping mechanism predicted by Eq. (4). For QDPt and CdS_14nmPt, a faster decay is observed at an early time compared with longer rods. This may indicate that in this short-distance regime, electron tunneling may be the dominant ET mechanism.

Similar to charge separation, charge recombination in nanorods and 2D nanoplatelets has been described by activated diffusion of holes.^1,43 It was shown previously that electron–hole recombination in a CdS nanorod with a bulb region shows a power law decay ∼t^−1/2, indicative of unbiased 1D hole diffusion.⁴³ However, as shown in Fig. 4(b), for phosphonate acid capped CdS–Pt nanorods, a power-law recombination kinetics was not observed. Furthermore, the recombination kinetics appear to be independent of the rod length in longer rods, which is not consistent with the activated diffusion model. For CdS–Pt nanorods, the spatial distribution of the trapped holes is not well known, which may complicate the analysis of the charger recombination mechanisms. In this regard, CdSe@CdS–Pt nanorods may offer a better candidate for this study because of the preferential localization of the holes in the CdSe core.

In this work, we use CdS–Pt NR heterostructures as a model system to study how the average distance between the semiconductor and the metal, controlled by the rod length, affects charge separation, charge recombination, and H₂ production. We found that by increasing rod length from 4.6 ± 0.9 nm in quantum dots (QDs) to 89 ± 15 nm in 1D nanorods (NRs), the charge separation half-life time increases from 0.7 ± 0.1 to 170.2 ± 29.5 ps, and the charge separation efficiency QE_ET decreases from 92% to 81%. The amplitude-weighted average lifetime of the charge separated state increases from 4.7 ± 0.4 µs in CdS_QDPt to 149 ± 34 µs in CdS_28nmPt, and the lifetime does not increase further in longer rods. The H₂ production quantum efficiency increases from 0.2% ± 0.0% to 28.9% ± 0.4% when the CdS NR length increases from 4.6 ± 0.9 nm (QD) to 28 ± 3 nm NR, and remains at ∼20% to 30% for longer rods, following the trend of charge separation lifetimes. We propose a model that assumes that overall H₂ generation quantum efficiency can be considered as a product of four key steps: $QEH2=QEHT*QEETL*QEHS(L)*QEWR$⁠. According to this model, the observed length dependence of the H₂ generation efficiency is a combined result of the length-dependent charge separation, $QEETL$⁠, and hole scavenging, QE_HS(L) quantum efficiencies. The former is a result of the rod length dependent electron transfer from the CdS NR to the Pt tip, and the latter is caused by the rod length dependent charge recombination, which competes with hole transfer to the scavenger. However, because the rod length dependence of $QEETL$ is weaker than that of QE_HS(L), the latter and ,hence, the rod length dependent charge recombination time, is the key factor that controls the overall light driven H₂ generation quantum efficiencies. Our findings provide key insights into how the dimension of the semiconductor domain affects the performance of the photocatalysts and guidance on how to rationally design these materials for efficient light driven H₂ generation.

Synthesis procedure, TEM images, statistics of rod and Pt size, photocatalytical H2 production measurements, transient absorption measurements, and kinetics fitting details.

This material is based upon work supported by the U.S. Department of Energy, Office of Science, Office of Basic Energy Sciences, and Solar Photochemistry Program under Award No. (DE-SC0008798). The transmission electron microscope images were taken at the Robert P. Apkarian Integrated Electron Microscopy Core (IEMC) at Emory University.

The authors have no conflicts to disclose.

Yawei Liu: Conceptualization (lead); Data curation (lead); Formal analysis (lead); Writing – original draft (lead); Writing – review & editing (lead). Wenxing Yang: Formal analysis (supporting); Methodology (supporting); Writing – original draft (supporting); Writing – review & editing (supporting). Qiaoli Chen: Investigation (supporting); Writing – original draft (supporting); Writing – review & editing (supporting). Zhaoxiong Xie: Funding acquisition (supporting); Supervision (supporting).

The data that support the findings of the study are available within the article and its supplementary material.