Transition metal oxides (TMOs) have been attracting growing interest as candidates for nonlinear (NL) optical materials, due to their high optical nonlinearities, ultrafast response times, and low absorption indices [1,2]. TMOs are also the prime candidates for transparent conductive oxides (TCOs), because of their large bandgap and transparent and conducting properties, whose films are utilized in flat panel displays [3], solar cells [4], and electroluminescent devices as transparent electrodes [5].

Being one of the most commonly investigated TMOs, TiO2 in its various polymorphs (anatase, rutile, and amorphous) also possesses remarkable properties, such as the high stability of its excitonic transitions to perturbations such as defects, temperature [6], and excess charge doping [7]. For all optical switching applications, anatase TiO2 thin films exhibit an interesting fast and reasonably large NL optical response at 800 nm [1]. It was, however, noted that the physical properties and, thus, the NL optical response of TiO2 films are strongly dependent on the processing route [8], as well as on the inclusion of metallic nanoparticles (NPs) [9,10].

Measurements of the NL optical properties of TiO2 polymer nanocomposites demonstrated negligible two-photon absorption (TPA) and a negative value of the NL refractive index [11]. This is in line with reports showing no TPA of anatase TiO2 NPs in solution even up to fluences of 300mJ/cm2 [12,13]. On the other hand, large second- and third-order optical nonlinearities of the Au/TiO2 composite films were observed and attributed to the high density of Au particles in nonconductive films, and the strong local field enhancement caused by the localized surface plasmon resonance (LSPR) absorption [14]. In particular, it was reported that Au/TiO2 composites can exhibit a self-defocusing and saturating nonlinearity at optical intensities up to 5.6GW/cm2, while measurements on a pure TiO2 film did not show any nonlinearity [15].

Stable core-shell Au/TiO2 and Ag/TiO2 NPs were reported to exhibit saturable absorption or optical limiting at 532 nm nanosecond-pulse excitation, depending on the applied laser fluence [16]. This behavior was explained in terms of the electronic Kerr nonlinearities and NL scattering in the NPs. The samples were subject to laser fluences up to 20J/cm2 and powers up to 2.8GW/cm2, and showed no appreciable signs of laser-induced damage, making them promising candidates for high energy optical limiting [16]. It has also been noted that Au NP-doped multilayers such as Au/SiO2 and Au/TiO2 films can be used as optical filters, due to their high damage threshold and their larger third-order NL susceptibilities.

It has been established that excitation of the LSPR band of Au NPs deposited on or embedded in TiO2 films leads to injection of electrons from the Au NP into TiO2 [17–23]. Just as for dye-sensitized NPs, probing and time-resolving the electron injection from molecular dyes into TiO2 was performed using visible-to-THz probes [18,21,23,24], X-ray absorption spectroscopy (XAS) at the Ti K-edge [12,25,26], and, more recently, X-ray photoelectron spectroscopy sensitive to the Au 4f- and Ti 2p-orbitals [27]. Overall, the optical-domain pump-probe studies concluded that electron injection is prompt, within the cross correlation of the experiments (typ. <200fs), and takes place with an efficiency reaching 40%–50% [18,23]. It should be stressed that these time-resolved experiments concerned Au NPs deposited on TiO2 surfaces [18,24,27] or embedded in TiO2 films [23], and were all carried out in the linear regime of pump-pulse intensity.

Thus, it appears that the Au NPs can influence the optical properties of the Au/TiO2 system in different ways: electron injection and/or enhanced NL effects (e.g. TPA, saturable absorption) due to the presence of the plasmon field. To weigh these various contributions, it is important to have simultaneously spectroscopic fingerprints of the presence of electrons in the conduction band (CB) of TiO2 and of the Au NP transitions, as well as their temporal evolutions. This has not been the case in any of the above-mentioned studies reporting TPA in Au/TiO2 composite materials.

In recent years, Baldini et al. [6,13,28] carried out a thorough investigation of the spectroscopy and charge carrier dynamics of bare anatase TiO2 NPs using pump-probe spectroscopy in the deep-ultraviolet (3.3–4.6 eV). In particular, they identified the first excitonic transition of the system at ∼3.8eV and showed that it could be used as a marker of electron injection in dye-sensitized TiO2 NPs, by detecting its bleaching due to the blocking of the transition that results from the excess of electrons/holes in the CB/valence band (VB). Further to this, Wang et al. [29–31] used the same tools to investigate the spectroscopy and charge carrier cooling in Au NPs. The novelty of the latter work lies in the fact that they probed the region of interband transitions of the Au NPs for the first time. Thus, in the present case, the deep-UV probe offers the advantage that the characteristic transitions of both the Au core and the TiO2 shell fall in the same spectral region, so that their behavior can be simultaneously monitored.

Here we investigate the response of Au/TiO2 NPs in aqueous solution using pump-probe TA spectroscopy upon resonant excitation of the LSPR of the Au NP core in the visible, while the probing is done using the deep-UV broadband continuum. By detecting the excitonic transition of anatase TiO2 when excessive electrons/holes are present in the CB/VB [13], we get insights into the excitation mechanism of the system. We compare these results with LSPR-excited core-shell Au/SiO2 NPs and above the bandgap (BG) excited bare TiO2 NPs. The BG of SiO2 is too large (7.5 to 9.6 eV) for any electron injection from the excited Au core to take place, or it would require absorption by three to four visible photons, which is a very low-probability event. We observe that at pump energies resonant with the LSPR of the Au core, efficient TPA of anatase TiO2 promotes electrons from the VB directly into the CB by simultaneously absorbing two green photons. Even at higher pump fluences, no TPA is observed in the case of bare TiO2. We also find that the excess electrons/holes generated in the CB/VB have a significantly longer lifetime than in bare TiO2, offering a strategy to store charges. The experimental setup and procedures are presented in the appendix.

Figure 1 shows the absorption spectra of bare anatase TiO2 and bare Au NPs, as well as core-shell Au/SiO2 and Au/TiO2 NPs; the inset zooms into the region of the LSPR band of Au/SiO2 and Au/TiO2 NPs. The bare anatase TiO2 NP shows the well-known absorption with a wing extending down to ∼3eV, due to scattering. The shoulder at ∼3.8eV corresponds to the first excitonic transition of anatase TiO2, as previously discussed in detail [6]. The spectra of ∼25-nm-diameter bare Au and Au/SiO2 NPs are almost identical with both exhibiting the LSPR band at ∼2.35eV, and at higher energies, the region of interband transitions, discussed in Refs. [29,30]. In the case of Au/TiO2 core-shell NPs, the LSPR band shifts to the red by ∼0.37eV and exhibits a red wing extending down to ∼1.5eV spectrum. Its spectrum is similar to that of ∼10-nm-diameter Au NPs embedded in a TiO2 film [23]. The red shift of the LSPR in Au/TiO2 compared to Au/SiO2 is fully accounted for by the dielectric effects [23,32].

Figure 2 shows the time-energy deep-UV transient absorption (TA) maps and the corresponding TA spectra at different time delays of the three samples: bare anatase TiO2 NPs excited at 4.0 eV above the BG, and Au/SiO2 and Au/TiO2 core-shell NPs excited into the Au LSPR band at 2.0–2.4 eV. The TA of bare anatase TiO2 NPs [Fig. 2(a)] is identical to what was reported in Refs. [6,28]. The TA spectra of bare anatase TiO2 at 1, 100, and 800 ps time delay [Fig. 2(b)] clearly exhibit a negative (bleach) signal over the entire probe range with a band centered at ∼3.83eV, which represents the first excitonic transition of anatase TiO2 [6]. The bleach signal is due to electrons/holes generated by BG excitation into the CB/VB of TiO2. Figure 2(c) shows the TA map for Au/SiO2 NPs excited with a broadband pulse (2–2.4 eV) at the LSPR, and Fig. 2(d) shows the TA spectra. These results are close to those we recently reported for the case of bare Au NPs of identical size [29,30]. The TA maps and spectra for Au/SiO2 NPs exhibit three main features: two negative ones at ∼3.8eV and ∼4.4eV, and a positive one at ∼3.4eV. Briefly, they were identified in Refs. [29,30] using the band structure diagram of bulk Au, as interband transitions from the 5d valence band to the 6sp band above the Fermi level at the X and L symmetry points for the two negative ones, monitoring the population of electrons and holes, above and below the Fermi level, respectively. The positive band was identified as a transition between valence sub-bands (i.e. an intraband transition), monitoring the population of holes. The overall conclusion was that charge carrier cooling is slow (∼150ps) and commensurate with the time scale of thermal lattice cooling (∼250ps). The fluence dependence of the 3.4 and 3.8 eV bands at a time delay of 50 ps is shown in Fig. 3, and it clearly reflects a one-photon excitation process.

Figures 2(e) and 2(f) show the TA maps and spectra for the Au/TiO2 NPs excited by a 2.0–2.4 eV broadband pulse. Compared to Figs. 2(c) and 2(d), some notable differences but also similarities arise. First, the ∼3.4eV positive signal (due to holes in the Au core) is shorter-lived in Au/TiO2. Second, a band shows up centered at ∼3.8eV (the early time signal at this band is due to the Au NP, but it dramatically decays within ∼10ps, as observed in the Au/SiO2 case). Thereafter, the bleach signal persists up to the longest time of our temporal window. Figure 4 compares the 800 ps transients for the three samples. For the Au/SiO2 sample, the signal has all but died away, except for the weak remnant of the positive feature <3.6eV. However, the signal for bare anatase TiO2 and Au/TiO2 NPs exhibits the same bleach feature at 3.8–3.83 eV. Given that the equivalent band due to the Au core has fully decayed, we conclude that this is the bleach due to electrons/holes in the CB/VB of the TiO2 shell in Au/TiO2 NPs [13,28]. This is further supported by Fig. 5, which shows the time profiles of the 3.8 and 3.4 eV bands for both core-shell NPs. Finally, the features >4.1eV in Au/SiO2 do not show up in the case of Au/TiO2, due to the fact that the Au contribution to the transient is shaded by the strong absorption of TiO2 in this region.

The main observation here is the long-lived bleach band in Au/TiO2 NPs at the position (∼3.8eV) of the first excitonic transition of anatase TiO2. Excess electrons/holes in the CB/VB could either originate from the electrons transferred from the VB of TiO2 or the injected electrons from Au NPs. The presence of the intraband transition of Au at ∼3.4eV is indicative of holes in the Au NP core, due to electrons that have been removed from below the Fermi level upon LSPR excitation. They may end up in the CB of TiO2 as expected in the scenario of plasmon-induced electron injection into the TiO2 shell, typically within hundreds of femtoseconds [18,23]. However, the time scales of the band around ∼3.4eV and ∼3.8eV are not commensurate, implying that hole decay in the Au core is not due to a back-electron transfer from TiO2, but rather by the kinetics of charge carrier cooling just as in Au/SiO2 NPs [29,30]. Of course, the decay time of the 3.4 eV band is shorter here, which may point to an influence of the TiO2 shell. This is not surprising considering the way the latter affects the spectral features of the Au core (Fig. 1). In summary, it appears that the ∼3.8eV feature consists of three contributions: one at very early times, due to charge injection from the Au core; the second at early to somewhat later times due to the response of the Au interband transitions (close to that of Au/SiO2 NPs); and a third one that persists up to the longest time delay, which is due to the bleached TiO2 excitonic band.

Further insight into the origin of the population mechanisms in the CB of TiO2 is given by the fluence dependence of the various bands at short (10 ps), intermediate (50 ps), and long (800 ps) time delays (see Appendix C). Figures 6(a)–6(c) show log–log plots of the signals of Au/TiO2 at ∼3.8eV at these time delays for pump fluences ranging from 0.16 to 0.98mJ/cm2. The slopes of the signal are 1.32±0.20, 1.55±0.66, and 1.83±0.21 at 10, 50, and 800 ps time delay, respectively. While the latter points to a predominant two-photon process, the former two point to a regime intermediate between one- and two-photon absorption. Further, these slopes reflect an increasing weight of the TPA-induced signal as a function of time. This can be rationalized by noting that the ∼3.8eV signal contains the contributions of the interband transition of the Au NP, the electron injected via LSPR excitation, and the excitonic transition of anatase TiO2. As already mentioned, the first contribution, which dies away in ∼250ps, has a linear dependence on the pump fluence (Fig. 3). The second contribution, also linear with pump fluence, dies away much faster (<50ps) according to Refs. [23,27]. Therefore, at 10 and 50 ps delays, it is present [as seen in Figs. 2(c) and 2(d)], while at 800 ps it has vanished. In the latter case, we are left with a predominant TPA contribution that involves the TiO2 shell. This process is of course present from the earliest times onwards but is overlapped by the above one-photon contributions. Figure 7 shows the value of the ∼3.8eV signal slope as a function of time. We assume a maximum value of 2 at infinite times, considering that all signals due to the Au core have died away. The biexponential fit of the data point yields values of τ1≈50ps and τ2>2ns with uncertainties of a factor of 2. Bearing in mind the latter, it is fair to say that the first component decays on comparable time scales to its counterpart in Au/SiO2 NPs, as is actually the case with the 3.4 eV band [Figs. 2(c) and 2(d)].

Finally, the fact that three contributions are at play is visible from Fig. 8, which shows the change of profile of the TA spectra at 10 ps time delay for increasing fluences under visible (2.0–2.4 eV) excitation. Whilst for Au/SiO2 the TA profile does not change with pump fluence [Fig. 8(b)], for Au/TiO2 clear changes are observed with the growth of a blue shifted bleach band implying that the TPA-induced TiO2 excitonic bleach is taking over the overall bleach contribution with increasing pump power [Fig. 8(c)]. Note that for bare TiO2, there is no effect on the TA spectra over the same range of fluences, implying no signal indicative of charge carriers in the VB/CB. To further make the point about the two-photon process causing the bleach of the 3.8 eV band, Fig. 9 shows the log–log plots of the signal due to the ∼3.4eV band of Au/TiO2 NPs at 10 ps and 50 ps, respectively. Its slope is close to 1 (bearing in mind the large error bars), which is reasonable since this band is due to holes in the Au core caused by the LSPR excitation [29]. Note that the 3.4 eV band is at the limit of our detection, and therefore its signal at 50 ps is weak, which explains the relatively large uncertainty of its slope.

Coming back to the ∼3.8eV signal, one may ask whether the early time one-photon absorption signal could be due to injection of electrons from the Au core [18,21,33], which then decay on a ∼250ps time scale. This might be excluded because: first, it is not supported by the shape of the transients in Fig. 8(b); second, its decay is comparable to what was reported for the same band in Au/SiO2 NPs [29,30]; and, third, as was shown by Ratchford et al. [23] in the case of Au NPs embedded in TiO2 films, the injected electrons have a fast decay (∼2ps) followed by a long-lived (on the window of 20 ps of their time scan) weak component. We can therefore conclude that the early time signal is largely due to the response of the Au core. This does not exclude a TiO2 response due to injected electrons, but their contribution is short-lived. The bleach signal at 800 ps time delay, when the signal of the Au core has died away, is therefore due to the TiO2 excitonic bleach only. Its slope of 1 .83±0.21 reflects the fact that the electrons/holes in the CB/VB of TiO2 have been generated via a TPA process. Again, we stress that the same range of fluences does not show any signal in bare TiO2 NPs, which would be due to electrons/holes in the CB/VB [Fig. 8(a)]. This is in line with previous studies, which used fluences up to 200mJ/cm2 of a 10 ps/532 nm laser [12], i.e., two orders of magnitude larger than in the present case, showing no evidence of two-photon excitation. In Ref. [13], femtosecond pulse excitation at either 400 nm or 550 nm with fluences up to 53mJ/cm2 also showed no sign of multiphoton excitation. On the other hand, in the case of dye-sensitized TiO2 NP, the fluence dependence of the ∼3.8eV band due to injected electrons remained in the linear regime up to 336μJ/cm2 using 550 nm excitation. These observations lead us to conclude that the occurrence of multiphoton excitation of TiO2 in Au/TiO2 core-shell NPs is due to the presence of the Au core and is mediated by the resonance excitation of its LSPR.

We estimated the enhancement factor of the TPA of TiO2 in Au/TiO2 NPs, as described in Appendix D. Based on our experimental TA intensity (we took an average ΔA value of ∼0.3 at 3.8 eV, over the 10–200 ps range), from Fig. 2(e), we calculate that in Au/TiO2 NPs, the enhancement factor of the third-order NL two-photon process is ∼108–109. This is consistent with the results of surface-enhanced Raman scattering (SERS) [34,35] utilizing LSPR of nanostructured metallic NPs (e.g. Au, Ag, Cu) to amplify Raman signals of molecules at their surface with enhancement factors of 106–1011. It is also reported that the quantum yield of fluorescence in Au nanorods compared to a bulk metal surface is enhanced by a factor of more than 1 million [36,37]. In addition, simulations show that the enhancement of the molecular TPA cross section by the LSPR may reach many orders of magnitude, depending on the distance of the NPs and the |E|4 local field [38,39].

This huge enhancement of the TPA cross section in the case of core-shell Au/TiO2 NPs should be contrasted to the linear dependence reported in the case of Au NP charge injection into anatase TiO2 [17–23]. This suggests that the architecture of NPs deposited on the surface of TiO2 does not favor LSPR enhancement of TPA, which is likely caused by the limited contact interface with the NP in this architecture [18,24,27]. The case of Au NP embedded in the TiO2 film [23] is close to the core-shell architecture (although the Au NPs are densely stacked), but the upper fluence used in their work is close to the lowest in ours. It is also interesting to note that while on the short time scale (<20ps), the decay of the signal due to injected electrons is faster in the Au/TiO2, but leaves a weak long component, we observe that for the long time case, the excitonic bleach survives for significantly longer time than in bare TiO2 NPs [Fig. 5(b)]. The short time decay is probably due to injected electrons that go back to the Au NP, as also suggested by Borgwardt et al. [27], while the long-time component in their work may correspond to the one we observe here.

This paper reports on two remarkable observations: first, TPA in TiO2 occurs, and is initiated already at mild fluences; and, second, a longer-lived excitonic bleach of TiO2 compared to the bare NPs, implying that the presence of the Au core “blocks” the electron–hole recombination in the TiO2 shell. It is not clear what mechanism could be behind this behavior, as the charge carrier dynamics in the Au core is over, while the bleach of the excitonic band goes on up to the time limit of our detection. Nevertheless, that electrons/holes in the CB/VB are present for a long time deserves further investigation as it opens perspectives in terms of charge storage in a TMO, with possible applications in the field of TCO and photocatalysis.

Acknowledgment. The authors thank Dr. L. Mewes, Dr. M. Puppin, and B. Bauer for helpful discussions. L.W. acknowledges support from the China Scholarship Council (CSC).