Spectroscopic signatures of plasmon-induced charge transfer in gold nanorods

There is continuing intense interest in plasmonic nanoparticle–semiconductor hybrid structures for photovoltaic and photocatalytic devices, utilizing the large absorption cross section of plasmonic nanoparticles and their ability to generate tunable, energetic charge carriers to inject into the adjacent material.^1–6 The predominately proposed mechanism of charge transfer is via hot carriers generated after plasmon decay, followed by their subsequent transfer to the adjacent material.^7–9 However, theoretical work and some experimental evidence also suggest plasmon-induced direct charge transfer (PICT) as a possible mechanism, where electrons involved in the plasmon resonance decay via injection into the adjacent material.^10–13 If PICT is the dominant pathway, it is expected to provide higher charge injection quantum efficiencies than the hot-carrier injection pathway as PICT avoids competing internal relaxation in the metal.^10,11 Thus, quantifying the contributions of PICT and hot-carrier transfer to the overall yield is important to device design.

Many plasmon enhanced photocatalytic and photovoltaic studies rely on completed devices to measure their performance and are carried out on bulk samples.^14–16 Recently, methods for single-particle characterization of plasmonic photovoltaic devices have been developed, utilizing either an electrochemical cell or electron microscopy.^17,18 However, there is still a need for all-optical methods that are capable of predicting device performance at multiple design stages, which is minimally invasive and does not require a completed device.¹³ 

Here, we use dark-field scattering (DFS) and photoluminescence (PL) spectroscopies to identify the mechanisms and material parameters involved in charge transfer from gold nanorods (AuNRs) into the conduction band of an adjacent semiconductor: indium doped tin oxide (ITO) and amorphous titanium dioxide (TiO₂) compared to quartz (SiO₂). These substrates were chosen since ITO and TiO₂ have different Schottky barrier heights, ITO and TiO₂ are widely used and studied semiconductor materials, and SiO₂ is an effective insulator that serves as a standard for no charge transfer. Additionally, our studies offer lower bounds on the spectral signatures of charge transfer efficiency, as the contact area between the AuNRs and the semiconductors is minimal in our sample geometry of AuNRs on a planar substrate compared to overcoated or core–shell AuNRs.

Three different AuNR sizes, nominally 23 × 64, 30 × 75, and 39 × 96 nm and referred to as small, medium, and large, respectively, were used for all measurements. The medium AuNRs were prepared by seed mediated growth according to Ref. 19 and the small and large AuNRs were prepared according to Ref. 20. The AuNR size distributions were published previously.^21–23 The AuNRs were spin coated onto indexed SiO₂, ITO coated glass, and TiO₂ substrates. All substrates were cleaned prior to spin coating by O₂ plasma etching. We made no attempts to remove the surfactants from the AuNRs as that could introduce differences between substrates during O₂ plasma cleaning. While the surfactants are known to create a ∼3.2 nm dielectric shell around the AuNRs,²⁴ previous studies have demonstrated that charge transfer is still observed in electrochemical experiments with the surfactant still present on the AuNR surface.^22,25 The TiO₂ coated substrates were made by atomic layer deposition (Ultratech Savannah ALD) of 12 nm TiO₂ onto glass slides from the tetrakis(dimethylamido)titanium(IV) precursor for 300 cycles at a rate of 0.04 nm/cycle. Substrate indexing was performed by electron beam evaporation of 20 nm Ti through a transmission electron microscopy grid (Ted Pella) shadow mask, resulting in a labeled alphanumeric grid on the substrate.

Single-particle DFS spectroscopy was performed according to previously published studies using an inverted microscope (Zeiss, Axio Observer.D1m).^21,26 Briefly, incandescent light (Zeiss, HAL 100) was focused by a dark-field oil immersion condenser through the substrate onto a sample region moved by a piezo-sample scanning stage (Physik Instrumente, P-517.3CL). The scattered light was collected with a 50× air space objective [Zeiss, EC Epiplan-Neofluar, 0.8 numerical aperture (NA)], spatially filtered through a 50 µm pinhole (Thorlabs), and either imaged on an avalanche photodiode (APD, PerkinElmer, SPCM-AQRH-15) for obtaining the AuNR position or dispersed through a spectrometer (Shamrock, SR193i-A) and detected with a charge-coupled device (CCD) camera (Andor, iDus 420 BEX2-DD) for acquiring spectra. The raw spectra were corrected for the illumination, background, and collection efficiency by subtracting the background from the raw spectrum and dividing the result by the dark count subtracted incandescent light spectrum.

Correlated single-particle PL imaging and spectroscopy were performed on the same AuNRs as investigated by DFS imaging and spectroscopy according to previously published procedures.^21,26 Briefly, the AuNRs were illuminated in an epifluorescence configuration through the 50× 0.8 NA air space objective with a 488 nm continuous wave laser (Coherent Obis) having a power of 0.155 MW/cm² at the sample plane. The emission was collected by the same objective, transmitted through a 50/50 beam splitter (Chroma), 488 nm notch filter (Semrock), and 496 nm long pass filter (Semrock). The pinhole used in DFS was removed for PL measurements. The PL was either imaged on the APD or dispersed through the spectrometer and detected with the CCD camera as in the case of the DFS measurements. The raw PL spectrum was corrected for instrument efficiency and background according to a previous method.²⁷ Briefly, the background (a PL spectrum of an area with no AuNRs) was subtracted from the raw AuNR PL spectrum and corrected for the wavelength-dependent detection efficiency of the setup.

The quantum yields (QYs) of the individual AuNRs were calculated according to a previously published method.²⁷ Here, the QY is defined as the ratio of emitted photons to absorbed photons. First, the raw photon counts were determined by integrating the photon counts for localized AuNRs within a 10 × 10 µm² scanned APD image. The number of emitted photons was calculated by correcting the raw photon counts for the wavelength-dependent efficiencies of the optics and filters used and photon detection efficiency of the APD. The number of absorbed photons was calculated using the excitation power of the 488 nm laser and the absorption cross section of the individual AuNRs as obtained from finite difference time domain (FDTD) simulations. Details regarding the FDTD simulations are discussed in the section titled Electromagnetic simulations. The energy absorbed was converted into photons absorbed by dividing by the energy of the incident photons (2.54 eV). Finally, the spectrally resolved QY was calculated by multiplying the QY with the integral normalized PL spectrum. The Purcell factor scaled QY was calculated by integrating the QY spectrum at the resonance energy ±3 nm and dividing by the Purcell factor calculated according to the procedure given below. By dividing by the Purcell factor, we removed the resonance energy, linewidth, and mode volume dependencies from the PL QY.

The Purcell factor was calculated according to a previous publication.²¹ Briefly, the Purcell factor can be derived experimentally using the following definition:²⁸ 

where E_res is the scattering resonance energy, Γ is the linewidth, and V is the mode volume. The mode volume was calculated according to the following equation:^28,29

The integral is calculated over all space, where |E| is the magnitude of the electric field determined from FDTD simulations discussed in the section titled Electromagnetic simulations, ω is the frequency, ϵ is the complex dielectric function of gold (taken from Johnson and Christy), and γ is the Drude damping, set to 0.07 eV.^30,31

The absorption cross sections for the QY calculations were determined using FDTD simulations (Lumerical FDTD Solutions) of AuNRs with sizes similar to those measured by scanning electron microscopy (SEM). A total of 302 unique AuNR sizes were simulated for each substrate with ranges of 16–46 nm for the width and 56–115 nm for the length. The AuNRs were modeled according to previously published methods as hemisphere capped cylinders using the gold dielectric function measured by Johnson and Christy³¹ sitting on a semi-infinite substrate of either SiO₂ (refractive index of 1.52), ITO (refractive indices from Ref. 32), or TiO₂ (refractive indices from Ref. 33).²¹ The choice of gold dielectric dataset is not expected to change any trends in the absorption cross section calculation since the 488 nm absorption is far from the AuNR resonances. The simulated absorption cross sections at 488 nm were interpolated with a 1 nm spacing according to a cubic spline. The average absorption cross section of a ±3 nm patch around each experimentally measured AuNR’s size was used as the AuNR’s absorption cross section for QY calculations. The average calculated absorption cross sections had a standard error of less than 10%.

The electric field components used in the mode volume calculations were determined using FDTD simulations. The electric field distributions at the longitudinal plasmon resonance were obtained using 3D frequency domain field profile monitors around the AuNR. The computational mesh size was set to 1 nm, rather than the default 0.3 nm, to decrease the computational cost of using a 3D frequency domain field profile monitor. As above, the size of each AuNR from SEM was used to find simulation results within a ±3 nm size window.

Correlated SEM micrographs were obtained after performing all DFS and PL measurements to determine the sizes of individual AuNRs. The indexed grids on the samples enabled for single-particle DFS and PL measurements to be correlated with SEM images for the same individual AuNRs. The SEM micrographs were collected on a FEI Quanta 400 environmental scanning electron microscope with field emission gun operated in the low-vacuum mode at a voltage of 30 kV.

We studied the DFS and 488 nm, off-resonant, interband excited PL spectra of single AuNRs from three different size distributions spin coated onto SiO₂, ITO, and TiO₂ substrates (Fig. S1). We assume similar interfacial contacts between the AuNRs and the different substrates, as supported by the overlapping AuNR size distributions for all substrates (Fig. S1). As demonstrated previously, the PL and DFS resonance energies follow a linear relationship with a slight blueshift in the PL resonance compared to DFS [Fig. 1(a)].^34,35 This correlation continues to hold for the AuNRs on ITO and TiO₂, demonstrating that the substrate does not affect this resonance energy relationship between DFS and PL. The scattering linewidth, as extracted through Lorentzian fitting, depends on the resonance energy, size, and substrate, as illustrated in Fig. 1(b). The substrate effect on linewidth is twofold. First, the resonance energy redshifts for AuNRs of the same size but on ITO or TiO₂ compared to SiO₂ [Fig. 1(c)] because of the higher refractive indices of those substrates. This spectral shift results in a linewidth narrowing as the plasmon is moved away from the interband region, as illustrated by the gray shaded area in Fig. 1(b), illustrating the resonance energy dependence of bulk damping based on the complex dielectric function of bulk gold.^30,36 Second, chemical interface damping (CID) leads to a broadening of the DFS linewidth, although this effect is not readily apparent and requires further analysis that removes the bulk and radiation damping contributions to the linewidth, as shown below.³⁷ 

In addition to linewidth changes due to different interfaces, we see a clear PL QY dependence on the substrate for the three example AuNRs presented in Fig. 1(d). We use PL spectroscopy to probe the hot carrier dynamics within the AuNRs under the assumption that PL is due to the Purcell factor enhanced radiative recombination of charge carriers or plasmon emission.^26,38–41 The substrate effect on QY is also twofold. First, the resonance quality factor (E_res/Γ)^28,42–44 affects the QY as it is directly related to the Purcell factor that enhances the radiative recombination of charge carriers within the AuNR or the plasmon density of states.^{21,26,40,41,45} Thus, the substrate modifies the QY through resonance energy shifts and changes in linewidth (due to both bulk damping and CID). We account for these effects by scaling the QY by the Purcell factor (Fig. S2).²¹ Second, the QY depends on the radiative recombination of charge carriers or plasmon emission.^26,38,46 If the excited charge carriers are injected into the semiconductor material at the interface, we expect a reduction in the PL QY.²⁶ 

Based on the energetic alignment of the AuNR and semiconductor bands (Table I and Fig. 2), we expect to observe changes to the plasmon damping in the DFS linewidth due to PICT. PICT provides an additional plasmon decay channel affecting the scattering linewidth via CID,⁴⁷ where the total linewidth is described by the phenomenological model, Γ = Γ_Bulk + Γ_Rad + Γ_CID. The contributing components to the total linewidth includes Γ_Bulk, which stands for the resonance energy dependent bulk gold damping having a constant intraband contribution of 73 meV and an interband component beginning around 690 nm and increasing with shorter wavelengths. Γ_rad represents the volume dependent radiation damping, and Γ_CID stands for CID.^47–50 Each term, except for CID, can be accounted for from the dielectric function or size of the nanoparticle and can therefore be subtracted from the total linewidth to obtain the contribution from CID only. The CID term is a catchall term for interfacial effects, including resonance energy transfer, charge transfer, and dipole scattering.¹² Here, we calculate the linewidth damping due to CID and relate it to the PICT mechanism of charge transfer into ITO and TiO₂ compared to SiO₂. We exclude an energy transfer mechanism due to the high bandgaps (>3.2 eV) for our materials compared to the 488 nm (2.54 eV) excitation source for PL and <2.2 eV plasmon resonance energy.

We accounted for the resonance energy dependence on the plasmon linewidth by subtracting the bulk damping contribution. We calculated the bulk damping contribution according to $ΓBulk=γb+ϵIBEres3/Ep2$⁠, where γ_b is the Drude damping, ɛ_IB is the imaginary part of the dielectric function due to interband absorption, E_res is the plasmon resonance energy, and E_p is the free electron plasma energy.³⁶ The values for ɛ_IB were calculated by subtracting the free electron Drude model from the imaginary part of the dielectric data [$Imϵ$] from Ref. 30 according to $ϵIB=Imϵ−(γbEP2)/Ephoton3$ using values of 9.3 eV for E_p and 73 meV for γ_b.^30,36 Any negative ɛ_IB values, which have no physical meaning, were set to 0, which only occurred below the interband onset and would lead to artificial linewidth narrowing in the intraband region. The bulk damping contribution to the linewidth is represented by the gray area in Fig. 1(b).⁵⁶ We note that the choice of gold dielectric data and Drude parameters affects the magnitude of Γ_CID but not the substrate dependent trends (Fig. S3). We chose the particular dielectric data published in Ref. 30 to minimize any spectral dependence on the linewidth for AuNRs on SiO₂, which we expect to be spectrally flat after the subtraction of the bulk damping contribution. We accounted for the size dependent radiation damping contribution according to Γ_rad = hκV/π, where we determined κ = 3.34 × 10⁻⁷ fs⁻¹ nm⁻³, consistent with previously reported values,^48,49 by fitting the data for the SiO₂ substrate after subtracting the bulk linewidth contribution. Here, V is the AuNR volume as determined by scanning electron microscopy, κ is a phenomenological constant, and h is Planck’s constant. We attribute the remainder of the single-particle plasmon linewidth, after subtracting the bulk and radiation damping contributions, to CID caused by the TiO₂ and ITO interfaces, as discussed next in Fig. 3(a). The quantum efficiencies of each of the damping processes are calculated and presented in Table II. Our average CID quantum efficiency (ϕ_CID) for TiO₂ is lower than what is observed in Ref. 56, where ϕ_CID = 0.3. We attribute this reduction in ϕ_CID to the presence of surfactant between the AuNRs and the substrate and a smaller interfacial contact between the AuNRs and the TiO₂ due to the sample geometry: AuNRs lying on TiO₂ vs overcoated with TiO₂. Thus, our results represent a lower estimate for CID compared to what should be possible with optimized samples.

The CID induced linewidth broadening (Γ_CID) has a mean of 3 meV for the AuNRs on SiO₂ [Fig. 3(a), blue] and is negligible, as expected, since SiO₂ is an insulating material for which, based on the energetic band alignments, no charge transfer between the AuNR and SiO₂ can occur (Table I). Thus, SiO₂ serves as a reference case of no Γ_CID due to charge transfer. ITO shows a modest increase in the mean Γ_CID to 7 meV but with a large distribution [Fig. 3(a), red]. The AuNRs on TiO₂ experienced the largest increase in Γ_CID with a mean of 12 meV [Fig. 3(a), yellow]. This value of Γ_CID for TiO₂ is lower than previously measured in Ref. 56, likely because those AuNRs were overcoated with TiO₂, while here the AuNRs are deposited on the surface of a TiO₂ film leading to a smaller surface area contact between the AuNRs and TiO₂. A one-way analysis of variance [F(2167) = 15.15, p = 8.95 × 10⁻⁷] with multiple comparisons test [Tukey honest significant difference (HSD)]⁵⁷ of Γ_CID for AuNRs on the different substrates finds differences between the substrates. The difference in mean Γ_CID values between SiO₂ and ITO (p = 0.055) is slightly above the significance threshold (p ≤ 0.05 is significant), while the differences between TiO₂ and SiO₂ or ITO are significantly different (p = 0.005 and p = 1.3 × 10⁻⁷, respectively). Thus, PICT, as measured by Γ_CID, occurs in TiO₂ and to a smaller extent in ITO. This interpretation of CID due to a PICT mechanism is consistent with observations of free carrier absorption in TiO₂ following plasmon excitation.^58,59

In addition to Γ_CID, we find that the PL QY serves as a spectroscopic signature of electron transfer. In order to properly compare the QY between the substrates, we calculated a scaled QY by dividing the QY at the resonance energy (λ_max ± 3 nm) by the Purcell factor.²¹ The results are summarized in Fig. 3(b). Again, SiO₂ [Fig. 3(b), blue] serves as a standard of no charge transfer with a mean scaled QY of 17 × 10⁻¹³. Scaled QY quenching is visible for AuNRs on both the ITO and TiO₂ substrates with mean scaled QYs of 11 × 10⁻¹³ and 7 × 10⁻¹³, respectively [Fig. 3(b), red and yellow]. We also performed a one-way analysis of variance [F(2167) = 129.9 and p = 9.5 × 10⁻³⁵] between the three substrates for the scaled QY, confirming significant differences. A multiple comparisons test (Tukey HSD) reveals that the quenching on ITO and TiO₂ seen in Fig. 3(b) is highly significant different from SiO₂ (p = 9.56 × 10⁻¹⁰ and p = 9.56 × 10⁻¹⁰, respectively) with TiO₂ and ITO also highly significantly different from each other (p = 9.86 × 10⁻¹⁰). Based on our interpretation of PL, we conclude that there is a reduction in the average number of excited charge carriers able to radiatively recombine occurs when the AuNRs are in interfacial contact with ITO or TiO₂, indicating that the charges are injected into the adjacent material.

Our correlated single-particle measurements enable us to also investigate the effect of the size and resonance energy on the Γ_CID and scaled QY for the different substrates. Figure S4 illustrates the AuNR size dependence of Γ_CID and scaled QY. The inverse effective electron path length (1/L_eff), which is directly related to the surface area to volume ratio via L_eff = 4V/S, has been previously used to describe surface scattering contributions to the DFS linewidth.^49,60 The AuNRs on SiO₂ exhibit no correlation between 1/L_eff and Γ_CID or scaled PL. While previous work has reported a correlation between 1/L_eff and linewidth, the range of AuNR sizes is small here and in the size range where surface scattering is relatively weak.⁶¹ Meanwhile, the AuNRs on ITO and TiO₂ demonstrate a weak negative correlation between size and Γ_CID, where AuNRs with larger 1/L_eff have smaller Γ_CID (i.e., smaller AuNRs show less CID), but no correlation was observed with the scaled PL. The observed correlation between 1/L_eff and Γ_CID for AuNRs on ITO and TiO₂ is contrary to the relationship found in Ref. 61, although the mechanism for CID there was likely due to a change in molecular induced surface dipole moment upon adsorbate exchange.^62,63 Again, the 1/L_eff ranges here are small and fall in the region where radiation damping dominates over surface scattering,^49,61 and the dependence is weak. One possibility for why this negative correlation exists is through substrate induced changes in the radiation damping term κ, which is dependent on the permittivity of the surrounding material.⁶⁴ 

In Figs. 4(a)–4(c), we identify relationships between the plasmon resonance wavelength and Γ_CID for AuNRs on ITO but not SiO₂ and not a statistically significant dependence in TiO₂. The AuNRs on SiO₂ demonstrate no correlation between the scattering resonance wavelength and Γ_CID, consistent again with SiO₂ serving as our standard. The AuNRs on ITO exhibit a negative correlation between the resonance energy and Γ_CID. The observation that AuNRs with shorter wavelength resonances experience larger Γ_CID generally follows a modified Fowler–Nordheim equation,⁶⁵ describing the yield of electron transfer as proportional to $cEres−ϕSB2+b$⁠, where c is a geometric parameter that accounts for enhancement, set to 18 for Γ_CID and 18 × 10⁻¹³ for the scaled QY (obtained by fitting the ITO data), and b is a variable offset. TiO₂ does not exhibit a Γ_CID wavelength dependence since the Schottky barrier is higher for TiO₂ (ϕ_SB = 0.9 eV), and therefore, the excess energy is small, i.e., a shallow regime in the Fowler–Nordheim equation where the variance in the data is larger than any expected trend [Fig. 4(c), shaded region and solid yellow line]. For ITO, a higher excess energy leads to more efficient charge injection into the ITO for higher energy resonances, given the low Schottky barrier (ϕ_SB = 0.2 eV), i.e., a steep regime in the Fowler–Nordheim equation [Fig. 4(b), solid red line].^11,66

For the scaled QY, we again see that SiO₂ exhibits no correlation with resonance wavelength [Fig. 4(d)], thus scaling the QY by the Purcell factor proves to be an appropriate method to account for dielectric and CID effects on the QY. In contrast, for both the ITO and TiO₂ substrates, we see a negative correlation in scaled QY with resonance wavelength [Figs. 4(e) and 4(f)]. This correlation is more pronounced on the ITO substrate [Fig. 4(e)]. We attribute this trend to higher rates of back electron transfer followed by radiative recombination for higher energy plasmon resonances.⁶⁷ Thus, the scaled QY for lower energy resonances is quenched, while the scaled QY for the highest energy resonances is unaffected. This argument is akin to the trend in Fig. 4(b), where a higher plasmon energy results in a higher charge transfer rate in the forward direction, resulting in a loss in plasmon coherence and thus an increase in Γ_CID. For the scaled QY quenching, the trend is due to a combination of forward and back electron transfer. Considering the same small energy barrier in both directions for ITO, the observed resonance energy dependence should have similar impacts for forward and back electron transfer, i.e., back electron transfer followed by radiative recombination occurs at a higher rate for plasmons with higher resonances, resulting in a negative correlation between the resonance wavelength and scaled QY. It is important to note that these observed trends for the PL QY clearly depend on the plasmon resonance even though a plasmon is not initially excited in the PL experiments using 488 nm light. We will discuss the underlying mechanism below.

We performed a similar analysis for the interband region of the PL spectrum at 525 ± 5 nm (Fig. S5). Here, the QY and not the Purcell factor scaled QY was used since the longitudinal plasmon resonance does not overlap with this spectral region and the transverse plasmon is weak and less sensitive to local refractive index changes; thus, Purcell factor enhancement is not considered.²⁶ We expect the interband QY (QY_IB) to be unaffected by the TiO₂ substrate because the primary electrons generated through 488 nm interband excitation lack the energy to overcome the interband barrier height (ϕ_IB) between gold and TiO₂ (hν < ϕ_IB, where hν = 2.54 eV is the photon energy and ϕ_IB = 3.1 eV; Table I). Indeed, AuNRs on TiO₂ and SiO₂ substrates have nearly identical values for QY_IB with means of 1.25 × 10⁻⁸ and 1.23 × 10⁻⁸, respectively (p = 0.98). Interestingly, AuNRs on ITO experience an enhancement in QY_IB with a mean of 2.36 × 10⁻⁸ (p = 5.6 × 10⁻¹⁰). We attribute this enhancement to interband excited charge injection since 488 nm interband excitations have a high enough energy to overcome the interband barrier height for ITO (hν > ϕ_IB, where ϕ_IB = 2.4 eV; Table I), followed by either fast back electron transfer and then radiative recombination or potentially direct radiative recombination from an electron at the bottom of the ITO conduction band to the d-band hole formed in gold. The latter mechanism implies that effectively a larger density of electronic states is available for PL through the participation of electrons in ITO and explains the increased QY. This observation and interpretation, however, merits future investigations.

The results of these studies lead us to propose several distinct mechanisms for charge transfer between AuNRs and ITO or TiO₂, as schematically illustrated in Fig. 5. For AuNRs on SiO₂, the plasmon energy is much smaller than the required energy to reach the SiO₂ conduction band and thus no charge transfer occurs [Figs. 3(a) and 5(a)]. Additionally, when AuNRs on SiO₂ are excited with a 488 nm laser, the interband excited electrons even after Auger scattering lack the necessary energy and therefore no charge transfer occurs [Figs. 3(b) and 5(e)].^68,69 These properties allow SiO₂ to serve as a standard for which to compare the Γ_CID values and scaled QYs of TiO₂ and ITO.

For AuNRs on TiO₂, the band alignments suggest that plasmon resonance energies above 0.9 eV can inject electrons from the AuNR into the TiO₂ conduction band [Fig. 5(b)]. We observe PICT spectroscopically via an increase in the Γ_CID compared to AuNRs on SiO₂ [Fig. 3(a)]. Additionally, we measure no resonance energy dependence for Γ_CID [Fig. 4(c)], although this observation may be due to either being in a shallow region of a Fowler–Nordheim dependence or is similar to the lack of an energy dependence observed in Ref. 10. When the AuNRs on TiO₂ are excited with a 488 nm laser, the interband excited electrons lack the energy to overcome the Schottky barrier, similar to the AuNRs on SiO₂ [Fig. 5(e)]. The inability of the initially excited d-band electrons to inject into the TiO₂ conduction band is supported by the lack of change in QY_IB compared to SiO₂ (Fig. S5). Unlike the AuNRs on SiO₂, there is a large scaled QY quenching at the plasmon resonance [Fig. 3(b)], indicating that while the excited d-band electrons are unable to directly overcome the Schottky barrier, Auger scattered electrons can overcome the Schottky barrier. These secondary electrons can have a maximum energy of E_f + hν but are not necessarily dependent on the plasmon resonance energy. However, we nevertheless observe a weak resonance energy dependence for the scaled QY [Fig. 4(f)] and, therefore, postulate that the plasmon resonance must play a role in interfacial charge transfer although we are exciting off resonance at 488 nm. This resonance energy dependence, which is more pronounced for ITO (discussed below), can be described in two ways: (1) Auger scattered electrons have an energy distribution controlled by the density of electronic states modified by the plasmon, and thus, the Auger scattered electron energy maximum is E_f + E_res [Fig. 5(f)]; or (2) charge carrier scattering generates a plasmon that can directly inject an electron with energy E_f + E_res through the PICT pathway [Fig. 5(g)].^35,40,41,46 The resonance energy dependencies of these mechanisms can lead to the observed trend in Fig. 4(f), where higher energy resonances result in electron injection higher in the TiO₂ conduction band, therefore enabling more efficient back electron transfer followed by PL.

Invoking a plasmon in this way, effectively corresponding to the reverse process of plasmon decay, has been suggested before and becomes especially necessary for the ITO substrate.^40,41 For AuNRs on ITO, the PICT pathway is available for plasmon resonances with energies above 0.2 eV [Fig. 5(c)]. Similar to AuNRs on TiO₂, we observe PICT via an increase in Γ_CID compared to AuNRs on SiO₂ [Fig. 3(a)]. Unlike for AuNRs on TiO₂, on ITO we observe a pronounced resonance energy dependence in Γ_CID [Fig. 4(b)]. We assign this observation to being in a steep region of a Fowler–Nordheim dependence compared to AuNRs on TiO₂. For the AuNRs on ITO, we also observe quenching of the scaled QY compared to SiO₂ [Fig. 3(b)], with a magnitude smaller than for TiO₂ while now following strong plasmon resonance energy dependence [Fig. 4(e)]. When excited with a 488 nm laser, the initially created interband excited electrons have enough energy to directly overcome the barrier. If charge transfer were to only occur through these primary electrons, we should not expect a dependence on the plasmon resonance energy. However, because we clearly observe a resonance energy dependence, the same mechanisms of charge transfer leading to scaled QY quenching for AuNRs on TiO₂ must be present for AuNRs on ITO, i.e., Auger scattered electron injection into the ITO conduction band either through plasmonic modification of the density of electronic states [Fig. 5(f)] or scattering induced plasmon excitation followed by PICT [Fig. 5(g)]. We, therefore, conclude that it is not just the primary interband excitation generated electrons, which are injected into the semiconductor, but secondary charge carrier scattering events mediated by the plasmon must also play a significant role to explain the observed plasmon resonance dependence of PL quenching on ITO and TiO₂.

Finally, while we expected QY_IB to be quenched due to charge transfer for AuNRs on ITO, we observed an enhancement of QY_IB compared to AuNRs on SiO₂ (Fig. S5). We hypothesize that this enhancement is either due to a larger density of electronic states leading to direct radiative recombination of an electron at the bottom of the ITO conduction band with the d-band hole formed in gold or fast back electron transfer followed by radiative recombination.

We have demonstrated the ability to spectroscopically detect charge transfer through single-particle DFS and PL spectroscopy. Specifically, we identified scattering linewidth broadening and scaled QY quenching as spectroscopic signatures of charge transfer following both plasmon and interband excitation. For TiO₂, we observed PICT and Auger scattered charge transfer with low electron back transfer rates. For ITO, which is a degenerate semiconductor with a conduction band close to the Fermi energy of gold, we observed three mechanisms of charge transfer: resonance energy dependent PICT, Auger scattered charge transfer, and interband excited electron transfer. These processes had higher electron back transfer rates than TiO₂ because of the lower Schottky barrier between ITO and gold. These two complementary spectral signatures of charge transfer have utility in predicting the performance of plasmonic photoelectric and photocatalytic devices,^1,71,72 without needing to have a completed device, thus allowing evaluation of the device’s individual components, such as a hole acceptor in the absence of an electron acceptor. Further studies into these mechanisms and spectroscopic signatures include expanding the range of AuNR sizes to further characterize predicted dependencies, expanding the resonance wavelength range, and comparing other metal-oxide semiconductors possessing different barrier heights and carrier mobilities. Investigating other spectral signatures, such as changes to the absorption spectrum, could provide additional insight into plasmon-assisted charge transfer mechanisms.

See the supplementary material for supplemental Figs. S1–S5.

This material is based upon work supported by the National Science Foundation under the NSF Center for Adapting Flaws into Features (Grant No. CHE-2124983). S.L. acknowledges the Robert A. Welch Foundation for support through the Charles W. Duncan, Jr.-Welch Chair in Chemistry (C-0002). This work was conducted, in part, using resources of the Shared Equipment Authority at Rice University.

The authors have no conflicts to disclose.

The manuscript was written through the contributions of all authors. All authors have given approval to the final version of the manuscript.

The data that support the findings of this study are available from the corresponding author upon reasonable request.