Applying ultrafast transient absorption spectroscopy on photocharged BiVO4

(1)

Applying ultrafast transient absorption spectroscopy

on photocharged BiVO

4 Patrick Kwee,

Vrije Universiteit Amsterdam

June 27, 2016

1 Abstract

Bismuth vanadate is a promising semiconductor which can be used for photo-electrochemical water splitting where water is directly converted into hydrogen which is a fuel. It is im-portant that the conversion of water into hydrogen is performed with a high efficiency. In this study ultra-fast transient absorption spectroscopy (TAS) is used on a time scale from 100 fs to 100 µs to investigate the properties of four similar bismuth vanadate sam-ples (BiVO4) before and after exposing them to a solar simulator. In earlier research,

it is namely found that a so called photocharging treatment where samples are exposed to a solar simulator during a night, would improve the water splitting performance of the samples. In this study we try to find how photocharging affects the photophysical processes which occur on BiVO4 samples. It was proposed in the earlier research that

photocharging changes the surface or the bulk properties of the samples. However we found no differences between photocharged and untreated samples. This indicates that if there is a photocharged effect, it occurs on time scales longer than 100 µs or they occur on the surface. This is because surface contributions cannot be measured using our method. Another possibility is that the photocharging effect fades within 20 minutes, because this is the minimal time necessary to remove a sample from a solar simulator and to start a measurement.

(2)

2 Introduction

Photo-electrochemical (PEC) water splitting is a technique which allows to convert sun-light into hydrogen (H2). Hydrogen can be directly used as fuel and has the potential to

replace fossil fuels which are responsible for climate change. To perform water splitting, a light absorbing semiconductor is necessary. In this study, the n-type semiconductor bismuth vanadate (BiVO4) is investigated. This is a promising material which can

ab-sorb visible light, is non-toxic and can be produced on a large scale as it is composed of earth-abundant materials[1]. Trze´sniewski et al.[1] claimed that applying a treatment called photocharging on BiVO4 samples would improve the performance of solar water

splitting compared to untreated samples. Photocharging is a treatment technique where sunlight (simulated by a solar simulator) is used to illuminate the BiVO4 samples

dur-ing a night with AM 1.5 light. AM 1.5 represents the solar intensity when the sunlight falls on earth in an angle of 48.2 degrees. In this study, ultra-fast transient absorption spectroscopy[5] is used to investigate the photophysical processes of both photocharged and the untreated samples. Ravensbergen et al.[4] used this technique to unravel the dynamics of the charge carriers in bare BiVO4 samples. They found four main occurring

processes with a corresponding time constant. Hole trapping is associated with the time constant of 5 ps, electron relaxation with the time constant of 40 ps, electron trapping with the time constant of 2.5 ns and trap-limited electron-hole recombination occurs on time scales longer than 10 ns. The aim of this study is to find a possible explanation for the improved performance due to photocharging. It is possible that some processes such as hole trapping or recombination are affected by photocharging.

3 Theory

3.1 Photo-electrochemical water splitting

Solid materials consist of a valence band and a conduction band. These are energy levels which determine how well electrons can be transported and thus how well a material conducts. In a metal the bands overlap and in an insulator there is a large energy gap between the bands. In a metal, electrons can move freely which makes it a good conductor while an insulator does not conduct well, as electrons need a lot of energy to overcome the energy gap between the valence and conduction band. Between the metal and the insulator stands the semiconductor, which has a smaller band gap than the insulator. If sufficient energy is absorbed, electrons can be excited to the conduction band which makes a material conduct. There is a so called Fermi energy level[2], which is the energy level at which the probability of being occupied by an electron is 1/2. A n-type semiconductor is donor doped, its Fermi level is closer to the conduction band than to the valence band, because its conduction band contains a large presence of electrons.

For photo-electrochemical (PEC) water splitting, a n-type semiconductor is used as photoanode. Other components necessary for PEC water splitting are a metal counter electrode and an electrolyte[2]. All components are immersed in water as shown in Figure 2. The electrolyte is an aqueous solution which can be a base, an acid or a salt. The metal counter electrode accepts electrons from the semiconductor to form hydrogen molecules and is called the cathode.

(4)

Before electrons can be transferred to the metal counter electrode, a different reaction which is independent of light has to occur. This is a charge transfer reaction where the electrolyte accepts electrons from the semiconductor until the Fermi level of the semiconductor equals the energy level of the electrolyte and this is illustrated in Figure 1.

Figure 1: The light independent charge transfer reaction is shown here[2]. In (a), the semi-conductor and the electrolyte are isolated. In (b) they are contacted and the Fermi level of the semiconductor becomes equal to the redox level of the electrolyte which bends the valence and conduction bands upward.

The valence band and the conduction band at the interface of the electrolyte and the n-type semiconductor are bent upward during the process until an electrochemical equilibrium is reached. A result of this equilibrium is that a space charge layer is formed in the n-type semiconductor which separates the electrons and holes. The holes move towards the interface of the electrolyte and the semiconductor, while the electrons move in opposite direction where they can be transferred to the metal counter electrode by an external circuit. The reason why the n-type semiconductor is used, is thus because the space charge layer separates the holes and the electrons[6].

3.1.1 The reactions in water splitting set-up

The energy which is required to excite electrons in the n-type semiconductor, is provided by photons. Depending on the band gap, energy is absorbed in the visible region, which is between 400 and 800 nm, or in the UV region. Light in the IR region has not enough energy to split water, because for water splitting at least 1.23 eV is necessary if losses are not taken into account. The excited electrons are transferred to a metal counter electrode, and at the interface of this counter electrode and the electrolyte, the reduction reaction takes place. At the interface of the electrolyte and the semiconductor oxidation reaction occur because here are ”holes” (h+) which are positively charged. Four of these holes react with four hydroxide ions (OH−), if the electrolyte is an alkaline electrolyte, and the end products of this reaction are two water molecules (H2O) and one oxygen

(5)

four electrons and four water molecules are converted into 2H2 molecules as shown in the

following equation:

4H2O + 4e− 2H2+ 4OH−. (2)

Water splitting is in competition with another process called recombination. This is the recombination of an electron and a hole and if this happens, the electron can no longer be transferred. The holes have to be mobile as well because they move to the surface to drive catalysis. To increase the efficiency of water splitting, the recombination time should be as long as possible or at least longer than the time necessary to split water.

Figure 2: The set-up for actual water splitting[3]. Light falls on a photoanode (BiVO4), here

an electron is excited and is transferred to the counter electrode. At the interface of BiVO4 and

electrolyte oxidation reaction takes place and at the interface of the counter electrode and the electrolyte reduction reaction takes place. The aim of this study is to investigate the photoanode thus the power source and the counter electrode are excluded in this experiment.

3.1.2 Characteristics of BiVO4

Bismute vanadate (BiVO4) is a metal oxide photoanode which is composed of VO4 and

BiO8 structural units[4]. It is non-toxic and can be produced on a large scale and consists

of earth-abundant materials. Besides it is stable in an aqueous environment with a pH value close to 7, which is necessary to it use in water. BiVO4 has an energy band gap of

∼2.4 eV which allows to absorb visible light. 2.4 eV corresponds to 520 nm. BiVO4 is

thus a promising semiconductor.

3.2 Ultra-fast transient absorption spectroscopy

Dynamic processes on a bismuth vanadate sample can take place on a pico-second time scale[4], thus a fast technique of measuring is required to investigate these dynamic

(6)

pro-cesses. Ultrafast transient absorption spectroscopy (TAS) allows to measure from 100 fs to 100 µs and is therefore applied in this study. During a single measurement, a pump pulse and a probe pulse are sent through the BiVO4 sample (see Figure 3) and the probe

pulse can be delayed by a value τ [5]. The pump excites the sample and with the probe the absorption spectrum of the excited state can then be determined. The spectrum of the ground state is determined when the sample is unpumped (this is when the pump beam is blocked by a chopper before it reaches the sample) and the spectrum of the pumped sample minus the spectrum of the unpumped sample is defined as ∆A. This can be viewed as a function of time and as a function of the wavelength.

Figure 3: In this figure, it is illustrated how the pump and the probe passes through a sample[5]. The pump which excites the sample is blocked after passing the sample while the probe is not blocked.

3.3 ∆A features for a semiconductor

When measuring ∆A for a semiconductor, three features can be distinguished and these are shown in Figure 4. The features are ground state bleach (1), electron absorption (2) and hole absorption (3). Ground state bleach gives a negative ∆A peak because the difference between pumped and unpumped is then negative. This is the case because a fraction of the electrons is excited while the rest is in the ground state. After excitation, less electrons are in the ground state and thus more ground state absorption is found in the unpumped sample compared to the pumped sample. Electron absorption is measured when the absorption spectrum of the excited electrons (which are in the conduction band) is regarded and this gives a positive contribution. This is because in a pumped sample, more electrons are in the excited state than in the unpumped state. Hole absorption also gives a positive contribution. This is because there are more holes if the sample is pumped compared to when the sample is unpumped. Holes are in the valence band.

(7)

Figure 4: The valance band is indicated in green and the conduction band is indicated in pink. The green dots (a) are excited electrons and the pink dots (b) are the holes in the valence band.

3.3.1 Interpretation of processes on BiVO4 samples using TAS

The processes which occur after excitation are explained by Ravensbergen et al.[4] who applied TAS to BiVO4 samples. According to their results, four main processes occur

(and some compete with each other) and these processes are hole trapping, electron relaxation, electron trapping and recombination. The difference absorption spectrum as shown in Figure 5 is characterized by a negative peak around 440 nm which corresponds to ground state bleach. In Figure 6 the time trace at 440 nm is shown where this negative peak can be observed. Hole trapping is associated with the time constant of 5 ps and is characterized by a positive peak around 475 nm. This contribution is due to excited state absorption. The broad absorption tail beyond 700 nm corresponds to the absorption of free holes and is also associated with the time constant of 5 ps. The time constants 40 ps and 2.5 ns both correspond to electron-hole recombination. This process is in competition with other processes. For the time constant of 40 ps the competing process is the relaxation of an excited electron in the conduction band, while for the time constant of 2.5 ns the competing process is trapping of electrons. Trap-limited recombination is associated with a time constant between 10 ns and 10 µs. Here the recombination of an electron or hole is temporary limited because it is trapped. It should be noted that the time constants 40 ps, 2.5 ns and 10 ns do not correspond to a specific wavelength.

(8)

Figure 5: The spectrum of a sample in air at a time delay of 1 ps. The characteristics are the negative peak at 440 nm, the positive peak at 475 nm and the broad absorption tail at longer wavelengths.

In Figure 5 the spectrum of a sample for a time delay of 1 ps is plotted. This is the spectrum which is expected if the sample is measured in air. Because a pixel in the detector was off during the measurements, the showed example contains a sharp decrease, following a sharp increase immediately after the peak at 475 nm. In Figure 7, the time trace at 475 nm is shown which contains a coherent oscillation. This corresponds to an early trapped fraction of holes at the surface as proposed by Ravensbergen et al.[4].

(9)

recombination. The corresponding time constant is 5 ps. In our study, we compare our data to the data of Ravensbergen et al. for the same delays and time traces and we also view the difference absorption spectrum for a time delay of 1 ps, 100 ps and 10 ns for our samples. The time traces at 440 nm, 475 nm and 700 nm are regarded.

Figure 6: The time trace of a sample in air at 440 nm on a short time scale.

Figure 7: The time trace of a sample in air at 475 nm. From 0 to 3 ps a coherent oscillation is visible.

Figure 8: The time trace of a sample in air at 700 nm. From 0 to 3 ps a exponential decay with a time constant of 5 ps is visible.

3.4 Photocharging treatment

Trze´sniewski et al.[1] introduced a treatment technique which they called photocharging and with this they measured a larger photocurrent and higher photovoltage. Photocharg-ing means that samples BiVO4 are illuminated by a solar simulator in an open circuit

during a night with AM 1.5 light. AM 1.5 represents the solar intensity when the sunlight falls on earth in an angle of 48.2 degrees. Trze´sniewski et al. found that the photocharging effect was best visible if a PBA PH10 buffer was used, while in a PEC cell, pH7 is usually used. They plotted the photocurrent density as a function of the photovoltage and the catalytic efficiency as a function of the photovoltage. Although an increased catalytic efficiency was found after photocharging, it could be not explain why. The surface mor-phology and the crystal structure were investigated but no differences between before and after photocharging were found in both cases. Trze´sniewski et al. hypothesized that the enhancement after photocharging was due to passivated electronic surface states which would cause an increase of the photovoltage. This is because the presence of the surface states has a negative effect on the performance of semiconductors and passivation of these states would improve the surface properties. Another suggestion is that the enhancement is due to a bulk process and this would explain the reduction of V5+ to V4+ which they observed.

3.5 Hypothesis

It is expected that the samples which are photocharged give a similar outcome as the samples which are untreated, but that the peaks at 475 nm are shifted for example. A shift in the peak of 475 nm in the ∆A spectrum would indicate that there are trapped states at wavelengths other than 475 nm. It is also possible that the signal in the time traces

(10)

at 475 nm and 700 nm decays slower because the recombination time would be longer. Four samples are compared and these samples have a number from 1 to 4. Although the samples are similar, the treatment of samples 1 and 2 is different from that of samples 3 and 4. Sample 1 and 2 are measured in air and sample 3 and 4 in buffer. It is expected that the results of samples 1 and 2 are similar because these are both measured in air. The same goes for samples 3 and 4 which are both measured in the buffer. It is also expected that the data of sample 1 (reference) and sample 2 (reference) are similar to the data of Ravensbergen et al. because in both cases, similar samples are measured in air. The data of sample 3 (reference) and sample 4 (reference) are expected to be similar to the reference of sample 1 and 2, because it is expected that the buffer itself does not affect the absorption spectrum of the samples.

4 Method

4.1 Measurement procedure

One BiVO4 sample is made by applying spray pyrolysis[4] and later divided into four

pieces. The samples are FTO coated glass with 200 nm BiVO4. Two sets of measurements

are done, each on a different day. For the first set of measurements, ultrafast transient absorption spectroscopy is applied on the four untreated BiVO4 samples. Sample 1 and

2 are measured in air, while sample 3 and 4 are measured in a PBA PH10 buffer. As a preparation for the second set of measurements on the next day, all samples (including 1 and 2) were put in the PBA PH10 buffer and then illuminated by a solar simulator during a night (from 6pm until 9am), as it was demonstrated that this would improve the measurements[1]. Before the second set of measurements started, sample 1 and 2 were removed from the buffer in order to measure it in air. Samples 3 and 4 were measured in buffer in such a way that the pump and probe first passed the buffer before it reached the samples. For the analysis of the data, Matlab is used. To investigate the consistency of each set of measurements, multiple scans are performed. Especially for the samples which are illuminated by the solar simulator, it is important to know what happens to the photocharged effect as a function of time and thus to understand the durance of this effect. This is investigated by studying what happens between the first and last scan from which it can be determined if there is a trend. Besides the different scans of the same measurements, the averages of samples 1 and 2 only and the averages of sample 3 and 4 only are compared. Sample 1 and 2 are namely measured under the same circumstances (in air) and so are sample 3 and 4 (in buffer). From this it can be determined if the results for the samples in air and in buffer are consistent or not. At last the samples 1 and 2 are compared to 3 and 4 from which it can be concluded whether or not the measurements in buffer give a similar outcome as the measurements in air. The difference absorption spectrum is regarded as a function of the wavelength and is regarded for a 1 ps, 100 ps and 10 ns delay. Besides the difference absorption spectrum is regarded as a function of time. This is the case when the time trace at 475 nm and 700 nm are viewed. Each

(11)

4.2 Set-up

The set-up as shown in Figure 9, consists of an optical parametric amplifier (OPA) and two lasers which are seeded by a single 80 MHz oscillator which serves as the master clock. This allows the lasers to be electronically synchronized. Both laser systems are amplified Ti:sapphire laser systems. One of the lasers is called the Libra (output power is 4.5 W) and is used to generate the probe beam. The other laser is called the Legend (output power is 3.0 W) and this laser is used to generate the pump beam. The output wavelength of both lasers is 800 nm and the repetition rate of is 1 kHz which means that they give a pulse each 1 ms. The OPA has no specific purpose in our set-up, the output wavelength is equal to the input wavelength which is 800 nm. The output wavelength of the Legend is doubled using a beta barium borate (BBO) crystal and this gives a pump beam with a wavelength of 400 nm. The probe beam is a broadband white beam generated by focussing the output of the Libra on a CaF2 plate. It is a collimated beam

that is focussed on the sample. The intensity of the pump can be varied but for this experiment it is set at 400 nJ/pulse, because in is used earlier research[4] and is more comparable to sunlight than higher intensities.

The polarization of the pump beam is set at 54.7 degrees with respect to the probe beam. This is called the magic angle and avoids polarization and photoselection effects[5]. In the path of the pump there is an optical delay line which can delay the pump by increasing the distance between the mirrors at (1) and (2). This delay line is connected to a computer which controls the positions. We chose to place the delay line in the path of the pump, because the probe is more sensitive to changes in the alignment of the mirrors than the pump. The delays which can be achieved using this delay line are in the range of fs to ns scale and the maximum delay that can be achieved is 3.8 ns[7]. In addition to the delay of 3.8 ns, delay steps of 12.5 ns can be used possible and these delay steps are controlled by a signal delay generator (SDG). The set-up consists of a chopper which blocks the pump at specific moments. At these moments the absorption spectrum of the ground state or unpumped sample is measured such that the difference absorption spectrum between the excited state (or pumped sample) and the ground state can be determined. The buffer solution or electrolyte used for our experiments is PBA PH10 buffer. This buffer is a 0.1 M solution of H3PO4, H3BO3 and CH3COOH (0.1 M each)

(12)

Figure 9: The set-up, adapted from Ravensbergen[7], consists of two synchronized laser sys-tems, the Legend from which the pump beam is generated and the Libra from which the probe beam is generated. The optical parametric amplifier (OPA) in our set-up has no specific purpose, The input and output wavelength of the OPA is 800 nm. The pump can be delayed by varying the distance between mirrors at (1) and (2). The chopper blocks the pump at some moments in order to measure the ground state absorption. At the moments that the chopper does not block the pump, the excited state absorption is measured. The difference between the excited state absorption spectrum and the ground state absorption spectrum is defined as ∆A. The pump is blocked by a beam stopper which prevents the pump to reach the detector.

5 Results

As explained in the measurement procedure, our data will be compared to earlier research done by Ravensbergen et al.. Besides the averages of sample 1 and 2, the averages of sample 3 and 4 and the samples in air and in buffer will be compared. For each set of measurements the ∆A can be plotted as a function of time in picoseconds or as a function of the wavelength in nanometer. In the first case the spectrum is plotted as a function of wavelength for different delays. We compared our samples for delays of 1 ps, 100 ps and 10 ns. In the second case, the time traces for two specific absorption wavelengths are regarded and these are 475 nm and 700 nm as was done in earlier research[4]. All samples are either photocharged or untreated and the samples which are untreated are referred to as ”reference”. In the appendix (chapter 8), the different scans for each measurement are compared. From this it can be concluded that the scans are consistent and similar. In the appendix the data of sample 1 compared to sample 2 is also shown and from this it can be concluded that the measurements in air are consistent. The same conclusion can be drawn

(13)

5.1 Comparison with data Ravensbergen et al.

Here we compare the normalized raw data of the different samples with the normalized raw data of Ravensbergen et al. for delays of 0.5 ps, 20 ps, 1 ns and 0.1 µs. This is done to discuss the similarities and differences between our data, because our methods and used set-up are similar. A difference between in method is however that our measurements are done with an excitation energy of 400 nJ/pulse while the measurements of Ravensbergen et al. are done with a 50 nJ/pulse excitation energy. In Figure 10 the time delay of 0.5 ps is viewed and here the data of Ravensbergen et al. from 2014 is indicated with the black lines. In this figure it can be observed that the recent data contains less noise than the data from 2014 and that the negative peak at 440 nm of the data in 2014 is more negative than for our measurements. Besides the data of 2014 shows an increasing trend in the region >550 nm while this is not the case in our data.

Figure 10: The data from Ravensbergen et al.[4] (from 2014) compared to our data for a delay of 0.5 ps.

While in Figure 10 the negative peak of the data of Ravensbergen et al. is lower than in our data, the negative peak at a delay larger than 20 ps (thus at 1 ns and 0.1 µs) seems to completely disappear. This is shown in Figures 11, 12 and 13. Apart from the negative peak at 440 nm, the lines are similar.

(14)

Figure 11: The normalized data from Ravensbergen et al.[4] (from 2014) compared to our data for a delay of 20 ps.

(15)

Figure 12: The normalized data from Ravensbergen et al.[4] (from 2014) compared to our data for a delay of 1 ns.

(16)

Figure 13: The normalized data from Ravensbergen et al.[4] (from 2014) compared to our data for a delay of 0.1 µs.

In Figure 14, the time trace at 440 nm of our data and the data of earlier research (indicated in black) are plotted on a short scale. The difference absorption spectrum is normalized. In Figure 15, the time trace at 475 nm of our data and the data of earlier research are shown. In Figure 16, the time trace at 700 nm of our data and the data of earlier research are plotted on a short scale. In Figure 14 the black line stays below ∆A=0 while the recent data goes above ∆A=0 as a function of time. In Figure 15 the normalization factor is based on the highest peak which is at a different point for the data of 2014 compared to our data. In Figure 16 it seems that the peak in our data decays faster as a function of time than the peak in earlier data. The differences in all time traces indicate that our samples are different than those used in 2014.

(17)

Figure 14: The time trace at 440 nm where the data from Ravensbergen et al. data from sample 1 and 2 are plotted.

(18)

Figure 15: The time trace at 475 nm where the data from Ravensbergen et al. data from sample 1 and 2 are plotted

(19)

Figure 16: The time trace at 700 nm where the data from Ravensbergen et al. data from sample 1 and 2 are plotted

The general difference between our results and the data of Ravensbergen et al. is that at a time delay of at least 20 ps, no negative peak at 440 nm can be observed in our data while this is clearly the case in the data of Ravensbergen et al.. The absence of the significant negative peak in our measurements can be due a decrease in amplitude of the negative peak at 440 nm or due to an increase in amplitude of the overlapping positive peak at 475 nm. To investigate why there is no negative peak, the time trace at 440 nm of our data is examined and fitted using an exponential. It is expected the time constant τ1 is 0.5 ps, but this gives no converging fit. If a double exponential is

used instead where the first time constant is fixed at 0.5 ps, a converging fit is found. The other time constant τ2 is on average 3.4 ps. In Figure 17 the time trace at 440 nm

of sample 1 (reference) is fitted on a region from 0 to 10 ps with a double exponential. The fits of the other samples are shown in section 8 Appendix and are also fitted using a double exponential.

(20)

Figure 17: The time trace at 440 nm of sample 1 (reference) on an interval from 0 to 10 ps is fitted using τ1=0.5 ps and τ2=3.48 ps.

5.2 Comparing photocharged and untreated samples

In this section our four samples are compared for a time delay of 1 ps, 100 ps and 10 ns. In Figure 18 the normalized spectrum is shown for the four samples. The brown lines correspond to sample 1, the green lines correspond to sample 2, the blue lines correspond to sample 3 and the red lines correspond to sample 4. The dashed lines represent the photocharged samples while the reference is indicated with the solid lines. In Figure 18 the lines corresponding to samples 1 and 2, which are measured in air, almost overlap in the region <500 nm. This is however not the case if samples 3 and 4 (measured in buffer) are compared. In general it seems that the data of sample 3 is odd and this is can be explained by the calibration of the wavelength which is slightly different for sample 3. If the photocharged samples are compared to the untreated samples no significant differences can be observed.

(21)

(22)

(23)

Figure 20: The normalized spectrum for all samples for a time delay of 10 ns.

The time trace at 475 nm on a short time scale is shown in Figure 23 and here a difference between the normalized data of sample 3 and the other samples is also visible. No differences between the photocharged and untreated samples can be observed. In the normalized time trace at 475 nm on a long scale or long semi-logarithmic time scale (see Figures 22 and 21), the lines of sample 3 and the other samples seem to overlap thus on a long time scale there is no significant difference between the samples. Besides because the normalized data of sample 3 (reference) almost overlaps with the normalized data of sample 3 (photocharged) it can be assumed that sample 3 is not odd because of photocharging effects. The same assumption can be made when studying the time trace at 700 nm as shown in Figures 24, 25 and 26 on different time scales. In the normalized time trace at 475 nm on a long semi-logarithmic time scale, at 104 _{ps there}

(24)

the measurements, thus we assume that it is not due to photocharging effects.

(25)

(26)

(27)

(28)

(29)

Figure 26: The time trace at 700 nm for all samples on a short time scale.

6 Discussion

When comparing our data to the data of Ravensbergen et al. we found in our data that the expected negative peak at 440 nm was not present or at least lower than in the earlier research. This can be explained by a decrease of ground state bleach or an increase of the overlapping absorption of trapped holes. Ravensbergen et al. showed that the hole absorption peak grows on a 5 ps constant. That the 440 nm kinetics in our data can be largely fitted with a timeconstant close to 5 ps could indicate that in our samples more holes are trapped than in the samples used by Ravensbergen et al. and this could be further investigated.

(30)

In general we observed no significant differences between the different samples or between the untreated and the photocharged samples if we either studied the (normalized) spectrum for different delays or the time trace at 475 and 700 nm. The odd data points are likely due to an issue in the software used for the measurements and not due to a photocharging effect. The reason why the peaks at 475 nm for sample 1 (both reference and photocharged) at 1 ps delay are higher than the peaks of sample 2, is likely because sample 1 was measured at a thicker spot than sample 2. We found that by moving the sample a few mm and thus by changing the measured spot, the height of the peaks changed significantly.

Trze´sniewski et al. hypothesized that the photocharging effect could be due to chang-ing bulk properties or due to changchang-ing surface properties. The changchang-ing bulk properties would explain the reduction of V5+ to V4+ while the changing surface properties would cause the electronic surface states to be passivated which would explain the increase in photovoltage. It is possible that photocharging effects are on the surface, because these contributions cannot be measured using our methods. To investigate if there are changes on the surface due to photocharging, we suggest to investigate BiVO4 samples

using Kelvin Probe Force Microscopy[9] (KPFM) before and after photocharging. This is a non-contact method which makes use of an atomic force microscope (AFM) and this could give information about the surface potential for the untreated and photocharged samples. If the photocharged effect is a bulk process, we did not observe it using TAS. This could indicate that the differences between the processes of photocharged and un-treated samples are possibly only visible on time scales longer than 100 µs. From other research done by Ma et al.[10] it was namely concluded that water splitting occurs on the ms-s time scale. It is also possible that photocharging sustains for a shorter pe-riod than 20 minutes. The time between the moment that a sample was removed from the solar simulator until the start of the measurement was namely at least 20 minutes. However it was claimed by Trze´sniewski et al. that the effect can be sustained for at least several hours, thus the hypothesis that the photocharging effect only sustains for a shorter period than 20 minutes would not be consistent with their results. A suggestion for further research (besides the use of KPFM) would be to investigate photocharged samples on time scales longer than 100 µs. Besides photocharging effects on different n-type semiconductors such as TiO2, Fe2O3 or WO3 could be investigated.

7 Conclusion

We viewed the (normalized) difference absorption spectra for different delays and for corresponding time traces at 440 nm, 475 nm and 700 nm. The behaviour of the untreated samples is similar to the behaviour of the photocharged samples. Our data is similar to data from earlier research except for a negative peak at 440 nm in the difference absorption spectrum. From our measurements we conclude that if there is a photocharging effect we have not found it using TAS. Further research is required to investigate how photocharging

(31)

TiO2, Fe2O3 or WO3.

References

[1] Photocharged BiVO4 photoanodes for improved solar water splitting, B. J.

Trze´sniewsk and W. A. Smith, J. Mater. Chem. A, 2016, 4, 2919-2926, DOI: 10.1039/C5TA04716A.

[2] Towards highly efficient bias-free solar water splitting, F.F. Abdi, PhD thesis (2013). Delft University of Technology. ISBN 978-90-6464-693-5.

[3] Electrochemistry: Photocatalysts in close-up J. Hofkens and M.B.J. Roeffaers Nature 530, 36-37 (2016) doi:10.1038/530036a

[4] Unraveling the Carrier Dynamics of BiVO4: A Femtosecond to Microsecond Transient

Absorption Study, J. Ravensbergen, F.F. Abdi, J.H. van Santen, R.N. Frese, B. Dam, R. van de Krol and J.T.M. Kennis, J. Phys. Chem. C, 2014, 118 (48), pp 27793-27800 doi: 10.1021/jp509930s.

[5] Ultrafast transient absorption spectroscopy: principles and applications to photo-synthetic systems, R. Berera, R. van Grondelle and J.T.M. Kennis, Photosynth Res (2009) 101:105-118, doi: 10.1007/s11120-009-9454-y.

[6] Photoelectrochemical Hydrogen Production, R. van de Krol and M. Gr¨atzel, Springer: New York, 2012, doi:10.1007/978-1-4614-1380-6

[7] Photophysics of solar fuel materials, J. Ravensbergen, PhD thesis (2015). Vrije Uni-versiteit Amsterdam. ISBN 978-90-5383-155-7.

[8] Electron-Phonon Coupling Dynamics at Oxygen Evolution Sites of Visible-Light-Driven Photocatalyst: Bismuth Vanadate. N. Aiga, Q. Jia, K. Watanabe, A. Kudo, T. Sugimoto, and Y. Matsumoto. J. Phys. Chem. C, 2013, 117 (19), pp 9881-9886 DOI: 10.1021/jp4013027

[9] Deconvolution of Kelvin probe force microscopy measurements - methodology and ap-plication, Machleidt T., Sparrer E., Kapusi D., and Franke K.-H., Meas. Sci. Technol. 20, 084017 (2009).10.1088/0957-0233/20/8/084017.

[10] Dynamics of photogenerated holes in undoped BiVO4 photoanodes for solar wa-ter oxidation, Y. Ma, S.R. Pendlebury, A. Reynal, F. Le Formal and J.R. Durrant, Chemical Science, 2014, pp 2964-2973, DOI: 10.1039/c4sc00469h.

(32)

8 Appendix

8.1 Comparing the different scans of the same measurement

In our measurements it can be observed that the time between the first and the last scan of the same sample does not affect the outcome systematically for sample 1, 2, 3 and 4. This is because for each graph where a delay between the pump and the probe for a specific sample is investigated, there is a peak in ∆A around 475 nm in the spectrum which does not depend on the moment of the scan (except for sample 2 reference and sample 3 reference).

8.1.1 Sample 1

In Figure 27, the spectrum of sample 1 (reference) for a time delay of 1 ps is plotted. For all samples and the shape of the spectrum is similar. It shows a negative peak around 440 nm which corresponds to ground state bleach[4], a positive peak around 475 nm and a broad absorption tail which corresponds to absorption of charge carriers. Around 480 nm a sharp peak is shown which is because this pixel in the detector was off. From Figure 27 it can be observed that all scans give similar results. The peak at 475 nm is slightly higher for the first and second scan (scan0 and scan1) compared scan2 and scan3 but this difference is not significant. Because the spectra almost overlap, it is possible to take the average of the scans. In Figure 28, the spectrum of sample 1 (photocharged) for a delay of 1 ps is plotted for different scans. The height of the peak at 475 nm for the scans is independent of the moment at which the scan is done. The heights do vary more significantly than for sample 1 (reference), especially scan2 is above the average of the other scans. It is still possible to take the average of these scans.

(33)

(34)

Figure 28: The spectrum of sample 1 (photocharged) at a time delay of 1 ps.

The time traces corresponding to the spectra at 475 nm and 700 nm for scans of both sample 1 (reference) as sample 1 (photocharged) are investigated. Figure 29 shows the different scans for the time trace of 475 nm for sample 1 (reference). It appears that t0 shifts to a negative value, as scan0 is at t0 = 0 (where t0 is expected) while

scan1, scan2 and scan 3 have a t0 with a more negative value than the previous scan

(t0,scan0 > t0,scan1 > t0,scan2 > t0,scan3). On average the value of t0 is shifted by -0.1 ps.

The shift also occurs for the time trace at 700 nm as shown in Figure 30, but here the shift of t0 on average is +0.03 ps. The time traces of the different scans for 475 nm

(35)

(36)

(37)

(38)

Figure 32: The time trace of sample 1 (photocharged) at 700 nm.

In all cases considering sample 1, the average of the different scans can be taken and by doing so for both reference and photocharged, the behaviour of the spectra as a function of time (using delays) can be investigated. This is shown in Figure 33 where the red lines correspond to sample 1 (reference) and the green lines correspond to sample 1 (photocharged). The characteristics of the raw data from Figure 33 are partly consistent with Ravensbergen et al.[4] who found that the peak of the spectrum at a delay of 20 ps was greater than that at a delay of 0.5 ps and that the peak for a longer delay then decreases. In our data the peak of sample 1 (photocharged) is larger than the peak

(39)

Figure 33: The spectra of sample 1 for different time delays: 1 ps (solid line), 100 ps (dotted line) and 10 ns (dashed line). The red lines represent the averages of sample 1 (reference) and the green lines represent sample 1 (photocharged).

8.1.2 Sample 2

The comparison of different scans for sample 2 is done in the same way as for sample 1. As shown in Figure 34, the scans corresponding to sample 2 (reference) at a delay of 1 ps have a similar shape but do not overlap. It seems that the peak at 475 nm decreases after each scan which can be due to a decreasing the intensity of the pump during the measurement. From Figure 35, it can be observed that the peaks corresponding to sample 2 (photocharged) at a time delay of 1 ps are consistent, because the data almost overlap. It is thus possible to take the average of the scans.

(40)

(41)

When regarding the time trace of sample 2 (reference) at 475 nm which is shown in Figure 36, it can be observed that the value of t0 is consistent but not at t0 = 0. On

average t0 is moved 0.15 ps to the right. A similar observation can be made for the time

trace of sample 2 (reference) at 700 nm which is shown in Figure 37, but here the time axis is shifted 0.55 ps to the right. Figures 38 and 39 show the time traces for sample 2 (photocharged) at 475 nm and 700 nm and these figures are similar to Figures 36 and 37 which correspond to sample 2 (reference). In Figure 36, t0 is moved 0.15 ps to the right

(42)

(43)

(44)

(45)

In Figure 40, the average spectrum for the three time delays (1 ps, 100 ps and 1 ns) are plotted for sample 2 (reference) and sample 2 (photocharged). This figure is partly consistent with Ravensbergen et al. because the peak of sample 2 (photocharged) for a delay of 100 ps is higher than the peak for the delay of 1 ps. This is however not the case for sample 2 (reference). It is possible that sample 2 (reference) was measured at a thicker part than for sample 2 (photocharged). This is because if Figures 34 and 35 are studied again, it appears that the peak of all scans for sample 2 (reference) are higher than the peaks for sample 2 (photocharged), except for scan2. Another explanation might be that the light intensity during the measurements of sample 2 (reference) was higher than during the measurements of sample 2 (photocharged).

(46)

Figure 40: The spectra of sample 2 for different time delays: 1 ps (solid line), 100 ps (dotted line) and 10 ns (dashed line). The red lines represent the averages of sample 2 (reference), and the green lines represent sample 2 (photocharged).

8.1.3 Sample 3

The different scans for sample 3 (reference) are shown in Figure 41 and these are similar but do not exactly overlap. It seems that the peak at 475 nm decreases after each scan. This is not the case for the scans of sample 3 (photocharged) as shown in Figure 42. The graphs in this figure show similarities with the scans of sample 3 (reference).

(47)

(48)

The time trace of sample 3 (reference) at 475 nm is shown in Figure 43 and in this figure, it can be observed that the time axis of all scans is moved to the right. On average this shift is 0.24 ps. Figure 44 shows the time trace corresponding to sample 3 (reference) at 700 nm and here the time axis of all scans also moved to the right. On average t0 is shifted 0.7 ps. Figures 45 and 46 are the figures corresponding to sample 3

(photocharged) at 475 nm and at 700 nm. They are similar to Figures 43 and 44 but the t0 values are shifted by respectively 0.15 ps and 0.63 ps.

(49)

(50)

(51)

(52)

Figure 47 shows the average spectrum for the three time delays (1 ps, 100 ps and 1 ns), plotted for sample 3 (reference) and sample 3 (photocharged). This figure is consistent with Ravensbergen et al. because the peak for a delay of 100 ps is higher than the peak for the delay of 1 ps. It can be observed that the peaks of sample 3 (photocharged) are higher for a delay of 1 ps and 100 ps compared to the peaks of sample 3 (reference) at the same delays, but not for a delay of 10 ns.

(53)

8.1.4 Sample 4

The spectrum for the different scans of sample 4 (reference) for a delay of 1 ps is similar, but the height of the peak at 475 nm varies for each scan. The spectra are shown in Figure 48. The scans shown in Figure 49 represent sample 4 (photocharged) at a time delay of 1 ps and these scans are similar to the scans of sample 4 (reference).

(54)

(55)

The different scans shown in Figure 50 correspond to the time trace of sample 4 (reference) at 475 nm and these scans are consistent. An overall positive shift of t0 can

be observed and the average shift is 0.12 ps. The different scans corresponding to the time trace of sample 4 (reference) at 700 nm are also consistent and the average shift is 0.62 ps. These scans are shown in Figure 51. Figures 52 and 53 show the time trace of sample 4 (photocharged) at respectively 475 nm and 700 nm and the figures are similar but have different shifts of t0. These shifts are on average respectively 0.15 ps and 0.67

(56)

(57)

(58)

(59)

The spectra of sample 4 can be compared by taking the averages of the different scans. This is done for the three time delays (1 ps, 100 ps and 1 ns) and this is shown in Figure 54. The red lines correspond to sample 4 (reference) and the green lines correspond to sample 4 (photocharged). The red lines are consistent with Ravensbergen et al. as the peak for the 100 ps delay is higher than the peak for the 1 ps delay. This is not the case for sample 4 (photocharged), where the peaks decrease if the delay is increased. The peaks shift to lower wavelengths if the delay is increased. It is interesting to note that the peaks of sample 4 (reference) and sample 4 (photocharged) around 475 nm for the 1 ps delay overlap but the other peaks do not. The green dashed line which represents sample 4 (photocharged) for a delay of 10 ns is below the red dashed line which represents sample 4 (reference) for a delay of 10 ns.

(60)

8.2 Time trace at 440 nm for all samples

As mentioned in the section ”Discussion” the time traces at 440 nm for the other samples are shown here for a region from 0 to 10 ps. Figure 55 corresponds to the time trace at 440 nm for sample 1 (photocharged). The line is fitted using the time constants 0.5 ps and 3.91 ps. Figure 56 corresponds to the time trace at 440 nm for sample 2 (reference). The line is fitted using the time constants 0.5 ps and 3.475 ps. Figure 57 corresponds to the time trace at 440 nm for sample 2 (photocharged). The line is fitted using the time constants 0.5 ps and 3.229 ps. Figure 58 corresponds to the time trace at 440 nm for sample 3 (reference). The line is fitted using the time constants 0.5 ps and 3.15 ps. Figure 59 corresponds to the time trace at 440 nm for sample 3 (photocharged). The line is fitted using the time constants 0.5 ps and 2.748 ps. Figure 60 corresponds to the time trace at 440 nm for sample 4 (reference). The line is fitted using the time constants 0.5 ps and 3.956 ps. Figure 61 corresponds to the time trace at 440 nm for sample 4 (photocharged). The line is fitted using the time constants 0.5 ps and 3.303 ps.

(61)

Figure 55: The time trace at 440 nm of sample 1 (photocharged) on an interval from 0 to 10 ps is fitted using τ1=0.5 ps and τ2=3.91 ps.

(62)

(63)

(64)

(65)

(66)

(67)

8.3 Comparing samples 1 and 2

Sample 1 and 2 are both measured in air and it is expected that the results of these two are consistent with each other. The averages of the different scans for samples 1 and 2 are shown in Figure 62 where sample 1 (reference) is indicated with the brown lines, sample 1 (photocharged) is indicated with the green lines, sample 2 (reference) is indicated with the blue lines and sample 2 (photocharged) is indicated with the red lines. The solid lines correspond to a delay of 1 ps, the dotted lines correspond to the a delay of 100 ps and the dashed lines correspond to a delay of 10 ns.

(68)

Figure 62: The spectrum for samples 1 and 2 for a time delay of 1 ps, 100 ps and 10 ns.

It appears that the peaks at 475 nm for both sample 1 (reference) and sample 1 (photocharged) are on average higher than the peaks for sample 2 (reference) and sample 2 (photocharged) for a delay of 1 ps and 100 ps. This is not the case for a delay of 10 ns, where the peaks of sample 1 (reference) and sample 1 (photocharged) are between the peaks of sample 2 (reference) and sample 2 (photocharged). In the appendix, the average spectrum for the three time delays is plotted for sample 1 only and sample 2 only. These are shown in Figure 33 and Figure 40 respectively. The lowest peak as shown in Figure 62 for a delay of 10 ns corresponds to sample 2 (reference). The peak corresponding to sample 2 (photocharged) is thus higher than the peak for sample 2 (reference). Because

(69)

the left and to the right while at the same height and thus changed the position of the spot at which was measured. This would explain why the peak at 475 nm, corresponding to trapped holes is higher. However it does not explain why for both sample 1 and 2 (reference), the peak decreases faster as a function of time than for the photocharged samples.

In order to understand why this appears to happen, we look at the time trace at 475 nm. The time traces at 475 nm and 700 nm for samples 1 and 2 are plotted in Figure 63 on a semi-logarithmic scale. The brown lines represent the data of the time traces for sample 1 (reference) and sample 1 (photocharged) at 475 nm and the green lines represent the data of the time traces for sample 2 (reference) and sample 2 (photocharged) at 475 nm. The blue lines correspond to the time traces at 700 nm for sample 1 (reference) and sample 1 (photocharged) and the red lines to the time traces at 700 nm for sample 2 (reference) and sample 2 (photocharged). In the figure, it can be observed that the data of samples 1 and 2 is similar for both the 475 nm time trace as the 700 nm time trace and that the time traces at 475 differ from those at 700 nm as expected. The data of sample 1 (photocharged) at 475 nm is above the data of sample 1 (reference). This is also shown in Figures 64 and 65. This explains why the peak at 475 nm seems to decrease slower for the photocharged sample 1 compared with the reference.

In general the data of sample 1 is above the data of sample 2 which is consistent with the data of Figure 62. At 10 ns however, the data of sample 2 (photocharged) is above the data of sample 2 (reference) and overlaps with the lines corresponding to sample 1. This explains why the peak at 475 nm corresponding to a delay of 10 ns in Figure 62 seems to decrease slower for the photocharged sample 2 compared to the reference. It is likely that this data point does not correspond to the right time point in the time axis and that it thus moved. This is due to an issue in the software used for the measurements. In the time trace at 475 nm for samples 1 and 2, no significant contributions of photocharging are found.

If the time traces at 700 nm as shown in Figure 63 are regarded, it can be observed that the shapes of the blue lines (corresponding to sample 1) and the red lines (corresponding to sample 2) are similar. This is the case if the reference is compared to photocharged sample and if sample 1 and sample 2 are compared. In the time trace at 700 nm for samples 1 and 2, the contribution of photocharging is also not found. When regarding the time traces at 475 nm and 700 nm for samples 1 and 2 on a relative shorter time scale than in Figure 63 which are shown in Figure 65 and Figure 64, the same observation can be made.

(70)

Figure 63: The time trace at 475 nm and 700 nm for samples 1 and 2 on a semi-logarithmic scale.

(71)

(72)

Figure 65: The time trace at 475 nm and 700 nm for samples 1 and 2 on a short time scale.

8.4 Comparing samples 3 and 4

Sample 3 and 4 are both measured in buffer. In Figure 66 where the spectrum is plotted for three delays, it can be observed that the peaks of sample 3 (photocharged) at 475 nm are higher for a delay of 1 ps and 100 ps, but not for a delay of 10 ns compared to sample 3 (reference). Sample 3 (photocharged) is indicated with the green lines and sample 3 (reference) is indicated with the brown lines in this figure. This is also the case for sample 4 where sample 4 (reference) is indicated with the blue lines and sample 4 (photocharged) with the red lines. The peaks seem to decay faster as a function of time in both cases for the photocharged sample in buffer and this can thus be due to photocharging. To investigate if this is the case in buffer, the time trace at 475 nm is regarded.

(73)

Figure 66: The spectrum for sample 3 and 4 for a time delay of 1 ps.

The time traces at 475 nm and 700 nm are shown in Figure 67 on a semi-logarithmic scale. Here sample 3 at 475 nm is indicated with the brown lines, sample 3 at 700 nm is indicated with the blue lines, sample 4 at 475 nm is indicated with the green lines and sample 4 at 700 nm is indicated with the red lines. The data corresponding to photocharged samples have the same colors as their reference but are indicated with the dashed lines. In the time trace at 475 nm as shown in Figure 67, it can be observed that the dashed brown line corresponding to sample 3 (photocharged) is above the solid brown line which corresponds to sample 3 (reference) except for the region around 104 ps. The observed decrease of the peak at 475 nm in Figure 66 which seemed to be faster for the photocharged sample 3 than for the reference can be explained by the delay of 10 ns (104

ps) which is viewed. If a time delay of 105 ps would be regarded, it would be expected that the peak at 475 nm would be lower for sample 3 (reference) compared to sample 3 (photocharged). If the time traces at 475 nm for sample 4 (green lines) are regarded, it

(74)

can be observed that the data of sample 4 (photocharged) is above the data of sample 4 (reference) except for the region around 10 ns. This explains why it appears that the peak decreases faster for photocharged sample 4 compared to the reference. It is likely that the data point in this region is due to the same software issue as for sample 2. In the time trace at 475 nm for samples 3 and 4, no convincing contributions of photocharging are thus found. The same conclusion can be made for the time trace at 700 nm for sample 3 and 4 as the corresponding lines show no significant differences on any time scale. The time traces at other time scales are shown in Figures 68 and 69.

If the normalized spectra for different time delays is viewed, which is shown in §8.3 ”Comparing samples 1 and 2”, no significant differences between the reference and pho-tocharged samples 3 and 4 can be observed. The normalized data of sample 3 differs from that of sample 4 which is likely due to the used normalization factor.

Figure 67: The time trace at 475 nm and 700 nm for samples 3 and 4 on a semi-logarithmic scale.

(75)

(76)

Applying ultrafast transient absorption spectroscopy on photocharged BiVO4