Red-Ox View of Light-Driven Electrons: From Laser Ablation to Plasmonics

1. Motivation: Solar Fuel

Photocatalytic and photoelectrochemical water splitting store sunlight in the $\mathrm{H}-\mathrm{H}$ bond, but the efficiency of any such device is set by the fate of the electrons and holes created on light absorption. The time-resolved spectroscopy was used to benchmark oxides and to show that performance is governed by three coupled processes on very different timescales: i) ultrafast band-edge and bulk recombination that destroys e-h carriers, ii) the slower spatial separation of electrons and holes, often driven by the space-charge field of a junction or by an applied bias, and iii) trapping at surface states where holes in contact with water either drive oxidation or are lost [1,2,3,4]. A recurring lesson is that water oxidation needs long-lived holes for slow O₂-evolving chemistry and that buying lifetime costs voltage [5]. The intrinsic lifetime is in turn tied to the metal centre, where a sub-picosecond relaxation through metal-centred ligand-field states quenches carriers and explains why high-dielectric $d^0$-materials such as SrTiO₃ hold charges far longer than hematite and perform so well [6,7,8]. It was shown what those design rules yield in practice having extended light harvesting into the visible with (oxy)nitrides and oxysulfides [9,10,11,12,13] and pushed SrTiO₃ to near-unity quantum efficiency by suppressing recombination [14]. The powder photocatalysis was taken out of the laboratory and were made into sheets for above 1% solar-to-hydrogen conversion [15]. The 1 m² panel under natural sunlight [16] and a 100 m² outdoor panel array run safely for months with membrane gas recovery [17,18]. Solar water splitting on metal oxides and nitrides directly advances UN Sustainable Development Goal 7 (Affordable and Clean Energy) by offering a route to renewable green hydrogen, while also supporting SDG 13 (Climate Action) through carbon-neutral fuel production [19].

2. Light as a Trigger for Reduction and Oxidation

Since the discovery of photoelectrochemical water splitting on TiO₂ [20], titania has been the archetypal photocatalyst, and a major route to extending its activity into the visible has been reduction: introducing oxygen vacancies and Ti³⁺ (Ti(III)) centres to make “black” or colored TiO_2-x [21,22]. The central theme of this perspective paper is that, in the processes used to make and to use such materials, light simultaneously induces reduction and oxidation: the absorbed photon energy is partitioned into reducing and oxidising species or carriers. This duality recurs from colloid synthesis to hot-carrier devices.

Pulsed-laser processing of materials in liquids is a versatile, surfactant-free route to colloidal metal-oxide nanomaterials [23,24]. In pulsed laser ablation in liquid (PLAL), a solid Ti or TiO₂ target is ablated. In laser melting/fragmentation in liquid (LML/LFL) an existing suspension is irradiated so that each particle is transiently melted and re-solidified. In both, the reactive aqueous environment and the non-equilibrium quench imprint oxygen vacancies and Ti³⁺, producing reduced TiO_2-x with a blue-to-black coloration [22,25,26,27]. The pulse duration sets the physics: nanosecond pulses act thermally with a large heat-affected zone, whereas femtosecond pulses drive non-linear (multiphoton/avalanche) absorption and optical breakdown before heat diffuses, favoring small, defect-rich particles and reduction of commercial anatase without chemical reductants [25,26]. The color tracks the degree of reduction [22].

The same femtosecond pulses also fragment colloids. It was shown that fs-ablation of gold in water gives bare colloids in two regimes: (i) small (3–10 nm), nearly monodisperse colloids at low fluence and (ii) large particles with a broad size distribution, formed via plasma heating at high fluence [28]. Size is then controlled with the white-light supercontinuum that an intense fs-pulse self-generates in the liquid: spread along a self-focused filament and overlapping the colloid plasmon, it fragments particles in a 3D volume so that they re-coalesce smaller and more stable, tunable by fluence [29]. This method was used to make ultra-pure water-dispersed gold [30], and oxidative fragmentation pushes gold below 3 nm [31]. Comparable down-sizing is reachable by ns-laser fragmentation [32] or by non-laser routes (cavitation, discharge, milling).

The oxidising face of light is equally important. Optical breakdown of water is radiolytic: it produces reducing species (short-lived $e_{aq}^{-}$, H^.-radical (an atom) as well as H₂) but also strong oxidisers (HO^., H₂O₂), so the very pulse that reduces titania can also re-oxidise it. The same duality appears in plasmonics: decay of a localized surface plasmon yields hot electrons (from the gold $sp$ band) that can reduce, and hot holes from the deep d-band that can oxidise [33,34,35,36].

Building on these ideas, the remainder of the paper follows one thread through three settings: (i) femtosecond laser reduction of anatase and the plasmonic enhancement of its activity; (ii) reduced titania decorated with gold as a photoanode for water splitting; and (iii) the same Au/semiconductor plasmonics enabling sub-band-gap (sub-wavelength) detection on silicon.

3. Results and Discussion from the Electron Transport View

3.1. Laser Reduction of Anatase and Plasmonic Enhancement of Its Activity

Synthesis. Following [37], an aqueous suspension of commercial anatase microparticles ($\sim 40\mu$m) at 15 mg mL⁻¹ was irradiated at the liquid surface with a 515 nm, 230 fs laser (Pharos, Light Conversion) at 200 kHz, using $5\mu$J pulses focused to a $\sim 5\mu$m spot, with periodic agitation (Figure 1); the product settled into color-graded layers.

Fluence, intensity and clamping. For a focal disk of diameter $d=5\mu$m the area is $A=\pi {(d/2)}^2=1.96\times {10}^{-7}$ cm², so $E_p=5\mu$J gives a fluence $F=E_p/A\approx 25.5$ J cm⁻² and, at $\tau =230$ fs, a geometric average intensity $I=F/\tau \approx 1.1\times {10}^{14}$ W cm⁻² (111 TW cm⁻²; for a Gaussian pulse the peak intensity is twice larger); the average power is $1.0$ W and the photon energy $2.41$ eV. This is deep in the optical-breakdown regime (water breakdown $\sim {10}^{13}$ W cm⁻², titania ablation $\sim 1$ J cm⁻²). However, the peak power (∼22 MW) far exceeds the critical power for self-focusing in water ($P_{cr}=3.77{\lambda }^2/8\pi n_0{n}_2\approx 1$–4 MW), so the beam filaments, and Kerr self-focusing balanced by plasma defocusing clamps the in-filament intensity to of order ${10}^{13}$ W cm⁻², largely independent of pulse energy [38,39]. The same self-focusing also generates the white-light supercontinuum. The geometric value is thus an upper bound, with the dose spread along the filament. The superheated, oxygen-deficient plasma and reaction with water (Ti−OH, Ti−H) create the lattice disorder, oxygen vacancies and Ti³⁺ of the bluish product [26,27,37].

Competing oxidative chemistry (ROS). Radiolysis in the fs-laser filament in water also makes oxidisers: HO^. radicals and, by 2 HO· → H₂O₂, hydrogen peroxide, ($E^{\circ }\approx +1.78$ V), which re-oxidise Ti³⁺ to Ti⁴⁺· (here Ti³⁺ denotes Ti(III) in the reduced surface oxide rather than an ion in the solution phase, i.e., Ti₂O₃ + H₂O₂ → 2 TiO₂ + H₂O). Because the Ti(III)-Ti(IV) red-ox couple resides at the surface, reduction is fast and local in the hot bubble, while surface-bound H₂O₂ oxidises. Oxidation and reduction proceed on the same surface, so H₂O₂ is formed at the interface, however could be transported to the surface from the solution. This plausibly explains why the product plateaus at blue rather than black and why reduced titania ages in aerated water, where TiO_2-x can also be oxidised to TiO₂ by dissolved oxygen. Peroxide is useful for oxidative fragmentation [31], and the oxidising channel can be suppressed (inert atmosphere, HO^./H₂O₂ scavengers, lower fluence) to favour deeper reduction. The energetics of the radical channel sets the scale of the oxidising power available: one-electron oxidation of water to the HO^. radical requires potentials as high as $+2.7$ V [40], which makes HO^. the strongest oxidiser generated in the filament. This is the synthesis-stage face of the light-as-oxidiser theme.

Characterisation. The product separated by density/color into light-blue, dark-blue (laser-ablated “LA-mid”) and dark-blue/black (“LA-top”) fractions (Figure 1), the darkest staying suspended longest [37]. X-ray photoelectron spectroscopy showed the signatures of reduction in every fraction: the O 1s lattice-oxygen peak shifted up by $0.2$–$0.3$ eV with a new ∼532 eV oxygen-vacancy component, and the Ti 2p doublet broadened and shifted to lower binding energy, indicating Ti³⁺ alongside Ti⁴⁺; the darker LA-top fraction carried the higher defect density. X-ray absorption at the Ti K-edge (XANES/EXAFS) was, by contrast, indistinguishable from anatase, so the reduction is surface-confined while the bulk remains crystalline anatase, as expected from a commercial-anatase feedstock. The visible-to-IR absorption responsible for the blue color (Figure 2(a)) is the optical fingerprint of the Ti³⁺/vacancy states, and a CIE-1931 evaluation of it returns a blue transmitted chromaticity (similar color as in Figure 1). Under visible light the colloid degraded ∼$79\%$ of methylene blue in $3.5$ h (pseudo-first-order $k_{\mathrm{app}}=7.1\times {10}^{-3}$ min⁻¹) with enhanced dark adsorption relative to the parent anatase; the low yield of the darkest fraction (∼5–6 mg) was, however, insufficient for N₂-adsorption (BET) surface-area analysis [37].

Plasmonic enhancement. Reduction of titania broadens its absorption, but the trapped carriers recombine readily. Decorating the reduced oxide with gold nanoparticles adds a route to harvest visible light and to separate charge. Gold nanorods on titania show transverse/longitudinal plasmon extinction bands (∼680 and ∼1000 nm, Figure 2(b)) that can be tuned to overlap the substrate absorption. Plasmon decay then injects carriers across the metal-oxide interface, the mechanism of plasmon-induced charge separation first demonstrated for Au/TiO₂ [33] and now central to plasmonic photocatalysis [34,35,36].

3.2. Reduced Titania with Gold as a Photoanode for Water Splitting

The Ti(III)-rich titania becomes the n-type substrate of a plasmonic photoanode (Figure 3). Stoichiometric anatase is a semiconductor with an empty conduction band $3.2$ eV up; the Ti³⁺/vacancy donors make it n-type and conductive, set up the depletion field/Schottky barrier, and add sub-band states that receive the injected electron, all prerequisites for a working electrode [20,41,42]. Gold (20–50 nm) adds a plasmon at ∼520-550 nm and a Schottky junction (${\varphi }_B\approx 0.9$-$1.1$ eV): $sp$ hot electrons above ${\varphi }_B$ inject into the conduction band while Au serves as a low overpotential cathode [34,42]. A $1.1$ eV (1100 nm) photon just lifts an $sp$ electron over ${\varphi }_B$, but the hole it leaves near $E_F$ cannot oxidise water. Only an interband $d\to sp$ excitation ($\ge 2.4$ eV) leaves a deep, oxidising d-band hole (Figure 3(b))—again the electron/hole (reduce/oxidise) duality.

Plasmonic photocatalysis and the surface-normal field. Decorating titania with plasmonic Au (or Ag) turns it into a visible-light photocatalyst, and action-spectrum analysis established that the photo-catalytic rate tracks the localized-plasmon absorption band of the metal—direct evidence that the chemistry is driven by plasmon-induced charge transfer into the oxide rather than by heating [33,43]. Which part of the plasmonic near-field does the injecting is not arbitrary. For a metal nanoparticle sitting on a substrate it is the depolarised field component normal to the interface (the $E_z$ component) that couples charge across the metal-semiconductor junction. For asymmetric spheroidal gold or silver nanoparticles on a dielectric substrate we showed that the frequency- and polarization-dependent optical response is governed by this perpendicular component, and that it is the surface-normal $E_z$ field, not the in-plane field, that is active in the measured action spectra [44]. Physically, $E_z$ is concentrated in the few-nanometer presurface region of the oxide where the band bending is strongest, so an oscillation perpendicular to the interface most efficiently drives hot electrons over the Schottky barrier into the conduction band [41].

A practical corollary, relevant both to the photoanode here and to the silicon detector below, is that the geometry should be chosen to maximise the surface-normal field: nanoparticles or nanorods with their long axis (or asymmetry) out of the surface plane, oblique or unpolarised illumination, and arrays engineered to concentrate $E_z$ at the contact all enhance the injecting channel, whereas purely in-plane (transverse) polarization couples poorly to across-barrier transfer even when strongly absorbed. The match between the photo-response action spectrum and the plasmon resonance is therefore a signature that the normal-field charge-transfer channel, not mere plasmonic heating, dominates the activity.

Importantly, the practical voltage of water oxidation is not the formal $1.23$ V. Being a four-electron reaction, oxygen evolution on real surfaces cascades through intermediates, each with its own potential: the two-electron peroxide stage (2 H₂O → H₂O₂ + 2 H⁺ + 2 e^-, $E^{\circ }\approx +1.77$ V) sets the realistic onset [45,46]—H₂O₂ reappearing, now as the oxidation intermediate. Noteworthy, the one-electron route is steeper still: formation of the HO^. radical requires a potential as high as $+2.7$ V [40]. The fewer electrons transferred per step, the higher the price per electron: $1e^{-}$ ($+2.7$ V, HO^.) >$2e^{-}$ ($+1.77$ V, H₂O₂) > $4e^{-}$ ($+1.23$ V, O₂). On the energy ledger of Figure 3(b), only the deep d-band holes (∼$+3$ V vs. NHE) are energetic enough to open the radical channel, whereas holes near $E_F$ of gold cannot drive even the four-electron reaction. A single $1.1$ eV photon therefore cannot split water unaided. In the biased cell of Nishijima et al. [42] (Figure 3(c)) an anodic bias ($+0.4$ V, SCE) and a Pt cathode supply the deficit: injected electrons exit through the circuit, biased holes oxidise water at the TiO₂/Au anode, and H₂ evolves at Pt. Following Juodkazytė et al. [47], the cathodic step runs through the adsorbed dihydrogen cation, H₃O⁺$+e-\to \left(\begin{array}{c}\hfill errortypeceH2+\end{array}\right)ad+$OH^- then $\left(\begin{array}{c}\hfill errortypeceH2+\end{array}\right)ad+e-\to$H₂, the H-H⁺ bond paid for by H⁺ hydration so that the discharge is depolarised to ∼0 V (RHE). This mechanism of water splitting is consistent with the simultaneously occurring mass and charge changes upon HER [47]. Fully plasmon-driven (unbiased) water splitting has been demonstrated when charge separation is engineered, e.g., in the autonomous device of Mubeen et al. [48].

3.3. Plasmonic Detection on Silicon Below the Band Gap

The same EBL-defined plasmonic antennas used to read out wavelength and polarisation on glass (Figure 4) can be placed on silicon to build detectors. Here the metal-semiconductor physics turns from chemistry to photocurrent.

The relevant levels (vacuum-referenced) are Si $\chi =4.05$ eV, $E_g=1.12$ eV ($E_C=-4.05$, $E_V=-5.17$ eV) and Au $\varphi =5.10$ eV ($E_F=-5.10$ eV), giving an electron Schottky barrier ${\varphi }_{Bn}={\varphi }_{Au}-{\chi }_{Si}\approx 1.05$ eV (Schottky-Mott; ∼$0.8$ eV with Fermi-level pinning) and a small hole barrier ${\varphi }_{Bp}\approx 0.3$ eV (Figure 5). After contact the Fermi levels equilibrate and the n-Si bands bend over a depletion width W, leaving ${\varphi }_{Bn}$ unchanged. Plasmon decay in 20–50 nm Au gives hot electrons up to $\sim \hslash \omega$ above $E_F$; those exceeding ${\varphi }_{Bn}$ are emitted over the barrier into Si and collected by the built-in field. Crucially, this works for ${\varphi }_{Bn}<h\nu <E_g$ (∼$0.8$–$1.12$ eV, $\lambda \approx 1.1$-$1.5\mu$m), i.e., below the silicon gap, internal photoemission, the basis of Au/Si Schottky hot-electron detectors for the near-infrared [35,36,49]. The injected carriers are $sp$-band electrons: the filled d-band (∼$1.8$-$2.4$ eV below $E_F$) supplies energetic electrons only via interband ($d\to sp$, $\ge 2.4$ eV) excitation, which places the electron just above $E_F$ (below the barrier) while leaving a deep d-band hole (Figure 5(d)). In the sub-band-gap window only the $sp$ channel operates, the mirror image of the titania case, where the d-band holes did the oxidation chemistry.

Polarization and wavelength sensitivity. Because a nanorod’s longitudinal plasmon scales with its aspect ratio, an array of rods is intrinsically wavelength- and polarization-selective, and this selectivity is carried directly into the photocurrent. In the titania photoanode of Figure 2 the Au-nanorod pattern shows distinct transverse and longitudinal extinction bands (here near 680 and 1000 nm) whose relative weight follows the incident polarization; the plasmon-assisted photocurrent measured by Nishijima et al. accordingly peaks at the wavelength and polarization that maximise the longitudinal resonance, so the same electrode effectively reports both the colour and the polarization of the sub-band-gap light it converts [41,42]. Transplanting the identical lithographic patterns onto silicon turns this into a hot-electron photodetector with built-in spectral and polarization discrimination: rods of different length resonate and therefore inject over the Schottky barrier at different wavelengths, while rods of a given length respond only to light polarized along their long axis, the orientation that also maximises the surface-normal injecting field discussed above for the photoanode. An array interleaving several rod lengths and orientations thus provides a set of photocurrent channels, each tuned to a particular wavelength-polarization combination, realising on a single chip the filterless wavelength-polarization read-out sketched in Figure 4. The price is the familiar one of internal photoemission, a modest quantum efficiency traded for operation below the silicon band gap and for intrinsic spectral and polarization selectivity.

4. Conclusions

The thread running through these three settings is that light simultaneously drives reduction and oxidation. In femtosecond processing of an aqueous anatase suspension the breakdown plasma reduces the surface to Ti³⁺/TiO_2-x while the radiolytic ROS (HO^., H₂O₂) re-oxidise it; intensity clamping (∼${10}^{13}$ W cm⁻²) together with this red-ox competition caps the reduction at “blue,” leaving the bulk anatase intact and the Ti³⁺ confined to the surface. Decorating the reduced oxide with gold tunes its visible–near-IR absorption and, through the surface-normal ($E_z$) plasmonic field, injects carriers across the interface, so that the photoactivity exceeds what reduction alone provides. The same Ti³⁺ donors make the oxide a conductive n-type photoanode on which $sp$ hot electrons drive reduction and d-band holes drive water oxidation.

A small applied bias completes the cell, with hydrogen evolving at the platinum cathode through the adsorbed (H₂⁺)_ad intermediate. Finally, the identical Au/semiconductor physics transferred to silicon yields a hot-electron photo-detector that responds below the silicon band gap (∼$1.1$–$1.5\mu$m): $sp$ electrons are emitted over a ∼$0.8$ eV Schottky barrier while the d-band again supplies the oxidising holes, and patterning the gold as nanorods of chosen length and orientation renders the photocurrent wavelength- and polarization-selective without filters. In every case it is the partition of the absorbed plasmonic energy into $sp$ electrons and d holes - into reductants and oxidants - that determines what the device can do.