Dynamics of Photo‐Induced Surface Oxygen Vacancies in Metal‐Oxide Semiconductors Studied Under Ambient Conditions

Abstract Surface‐enhanced Raman spectroscopy (SERS) is a powerful analytical technique commonly used in the detection of traces of organic molecules. The mechanism of SERS is of a dual nature, with Raman scattering enhancements due to a combination of electromagnetic (EM) and chemical contributions. In conventional SERS, the EM component is largely responsible for the enhancement, with the chemical contribution playing a less significant role. An alternative technique, called photo‐induced enhanced Raman spectroscopy (PIERS) has been recently developed, using a photo‐activated semiconductor substrate to give additional chemical enhancement of Raman bands over traditional SERS. This enhancement is assigned to surface oxygen vacancies (V o) formed upon pre‐irradiation of the substrate. In this work, the exceptional chemical contribution in PIERS allows for the evaluation of atomic V o dynamics in metal oxide surfaces. This technique is applied to study the formation and healing rates of surface‐active V o in archetypical metal‐oxide semiconductors, namely, TiO2, WO3, and ZnO. Contrary to conventional analytical tools, PIERS provides intuitive and valuable information about surface stability of atomic defects at ambient pressure and under operando conditions, which has important implications in a wide range of applications including catalysis and energy storage materials.

. Raman spectra of MBA-AuNP on TiO 2 after UV irradiation (blue solid) and over time showing the enhancement changes from PIERS towards the average SERS (red dashed). Additional sample kept under vacuum after UV irradiation for 2 hours (black dotted) measured just after release from vacuum, shown to have comparable enhancements to the initial PIERS spectra as opposed to average SERS. Hence, exposure to air results in decay of PIERS enhancement. be seen for most samples, however noticeably fluctuations in intensity can be seen over short periods of time. Initial decrease in intensity between 0-5 minutes was assigned to laser induced degradation dominating over vacancy induced affects. Nearly all samples shown tend to decrease towards the end of the measured time (particularly > 15 minutes), which was attributed to subsequent healing of induced vacancies. Figure S4. A sample of the calculated rate of vacancy formation using Equation 2, determined through deconvoluting experimental data. The rates shown here correspond to the measured data presented in Figure 3B and Figure 3C in the main text. Note, as described in the main text under no UV light exposure the change in effective vacancy concentration is only due to vacancy healing and therefore can be determined using Equation 1.

Experimental Sections Substrate Synthesis
Metal oxide (MO) films were grown using aerosol assisted chemical vapour deposition, see Methods, under specific conditions to produce flat thin films. Resultant films were annealed under ambient conditions at to create the desired crystalline phase and remove defects formed during film formation. Characterisation of the MOs was performed using Xray diffraction (XRD), Figure S8, confirming that the films were produced in the desired phase. From SEM images NP were sized at 40 nm in diameter. AuNP appeared most distributed on TiO 2 with much less clustering. Large clusters were often found on ZnO films, as can be seen in Figure S9f. WO 3 films were found to cause AuNP to cluster less than ZnO but more than TiO 2 films.

Calculating oxygen vacancy lifetimes
All Raman spectra of MBA were found to have 2 significant Raman bands at a frequency shift of around 1060 cm -1 and 1575 cm -1 . A series of Raman spectra over time for AuNPs-MBA (AuNP functionalised with MBA) were taken on each metal oxide substrates after pre-irradiation and under constant UV irradiation. Reference spectra were taken before UV irradiation. Raman band intensities over time for these 2 bands were calculated using Matlab from the series of spectra, with the assistance of "findpeaks" function. The changes in band intensities were then normalized with respect to the SERS baseline. An average over the relative changes in enhancement were taken over all measured positions on each respective sample and then plotted against time.
The resultant plot was computationally fitted using Matlab's "Curve Fitting" toolbox.
We refer the reader to the main manuscript where an explanation and justification for the type of functions used for analysis is explained in depth. Samples without UV irradiation were initially fitted to an exponential function in the form of where t is time from initial measurement and a, b and c are fitting parameters. An average of each data for each MO was used to determine the average decay due to laser induced affects.
This function describing laser induced signal decay was labelled L.
Band intensities were then determined using the same method above for pre-irradiated samples. The data was processed using the same method, however the functions were fit to where P is the measured PIERS data and V o represents the changes in intensity due to oxygen vacancy healing. As described in the main text, Equation S2 represents the case where, as no UV irradiation is present, no additional vacancies are induced, and therefore the change in effective induced vacancies is only due to vacancy healing. The contributions to the decay from laser photobleaching effects are independent to the photo-induced enhancements, hence the decay can be fit using a linear combination of these. V o was then determined using calculated L values and was fit to an exponential function as described above with L. Parameter b is defined as the rate of each process and related to the time constant for the exponential decay. From this value, an estimation of the vacancy healing rate and lifetime was determined. An average for the calculated lifetime between each MBA peak, 1060 cm -1 and 1575 cm -1 , were taken for each MO. The reported data reflects an average of the lifetime estimations across all measurements for each substrate material.
Series of Raman spectra during UV exposure were initially fit to an exponential function of the same form as Equation S1 . This was then analysed as described in the main text using Equation 1 and Equation 2 to calculate rates for V o formation, V o + .

Extrapolated and shifts in induced Vo concentrations
It is important to note that the out of the three fitting parameters used above only 'b' has a physical interpretationthe rate. This is due to the normalisation process used to compare each respective process on each respective metal oxide. Hence, the resultant curves may be translated without affecting the rate of decay or increase respectively. Data was presented relative to the SERS baseline background. Hence, where no UV irradiation was used the decay in the measured SERS enhancement, due to laser photobleaching, was set so EF=1 when time=0. For samples pre-irradiated with UV, at long time periods where all the induced surface vacancies are healed the enhancement returns to the SERS baseline. However, photobleaching of the sample will also have occurred during measurements. Hence, at large t the enhancement would be comparable to the decrease in enhancement just due to laser photobleaching and so was shifted in y accordingly. For samples under UV irradiation, at time=0 no vacancies have been induced, and therefore the enhancement seen in purely due to the SERS baseline. Where the relative EF=1, the CE=0 (due to V o ). Hence, where CE and the effective vacancies are displayed in Figure 3D, at time=0 and later times after the UV lamp was turned off CE=0. V o values were shifted in time to be 10 minutes after V o + values. This corresponded to the average time samples were under ambient conditions after the initial long UV exposure before measurements were taken, required due to experimental constrains (i.e. deposition of AuNPs-MBA and aligning and focusing samples with the Raman laser). Data was also extrapolated between V o + and V o points. Extrapolated data was found using