The figure to the right shows the solar spectrum as a function of energy (rather than the typical wavelength.) From this figure it is clear that a 2.3 eV photocatalyst will not be able to absorb much light. Thus this approach is limited to about 10-15% efficient photocatalysts. One way to get around this low efficiency is instead of using a single photocatalyst, use 2 photocatalyst in tandem. By using 2 photocatalysts in series their combined photovoltage would only need to match that of the single photocatalyst. What this means is that you can use lower band gap photocatalysts, and thus harvest more light. Obviously if you use 2 photocatalysts that means the 2 photocatlysts basically need to share the solar spectrum. Even so, modelling has shown this approach could realistically give efficiencies up to 25-30%. While this is a more complex system, the increased efficiency can potentially outweigh the increased cost due to the complexity. Many researchers have modelled 2-photon water splitting and they almost all come to basically the same conclusion. Optimally one of the photoabsorbers should have a band gap around 1.0 eV and the other one around 1.7 eV. These numbers vary slightly depending the assumptions for reaction kinetics and photocatalytic properties. Nevertheless John Turner's group has the record for 2-photon water splitting (or any water splitting for that matter) at 12.3%. Furthermore their device was not completely stable and used relatively rare materials. Thus there is major room for improvements in this field.

My approach is to try to optimize the individual parts of a 2-photon water splitting device and then eventually combine them all together. To the right is a scheme of one potential way to design a water splitting device. The top figure shows the overall physical design. The advantage of a design like this is that the hydrogen and oxygen are separated in-situ, thus eliminating a potential explosive mixture. One interesting issue with this design is that the proton needs to crossover from the anode to the cathode in a manner that does not allow the hydrogen or oxygen gas to also crossover. This means that either porous, holey, or wired photocatalysts need to be used and that an ion exchange membrane needs to be used. The middle figure shows how this device is designed optically to maximize light harvesting. The first photoabsorber has a large band gap and absorbs all the large energy photons. The small energy photons pass through this material and are then absorbed by the 2nd photoabsorber, which has a low band gap. The bottom figure is the energy diagram for the water splitting device. In this design the large band gap material needs to have a low valence band to allow for water oxidation. The conduction band of the large band gap material must be significantly high so that it can transfer its photogenerated electron to the valence band of the small band gap material. The small band gap material must have a relatively high conduction band to allow for hydrogen evolution. The dotted lines in this figure is the material's intrinsic Fermi level. Since the Fermi level's must align due to thermodynamics, it is essential that the photoanode is an n-type semiconductor and the photocathode is a p-type semiconductor. It should be noted that in this design that the oxygen is produced on the side closest to the solar irradiation and the hydrogen is produced on the backside. However you could just as easily have the hydrogen produced on the font side and oxygen on the backside. I have recently been part of a group that has analyzed which side to produce the hydrogen. The results of our analysis can be found here.

Simplifying the Experiments

Rather than try to build a 2-photon water splitting device all at once, it is possible to break this device down into 2 separate photoelectrochemical reactions: 1) Photoanodic O₂ evolution and 2) Photocathodic H₂ evolution. The great thing about photoelectrochemical reactions is that one can use a potentiostat to control the potential (i.e. Fermi level) of the semiconductor. Thus we can see how both the photoanode and photocathode perform at different potentials. From the energy diagram above it can be seen that eventually both the photoanode and photocathode will need to be at the same potential for overall water splitting to occur. Thus by plotting both the photoanode and photocathode current as a function of voltage on the same graph, the intersection between the 2 curves will be the current at which the device will operate at. The figure below shows this general concept. Current can be converted to amount of hydrogen via Faraday's constant and dividing by 2 since 2 electrons are needed per mole of H₂.

Just as a reference, 16 mA/cm² corresponds to about 20% solar to hydrogen efficiency.

By only focusing on one photocatalyst at a time, it is much easier to analyze and then optimize the catalyst. Thus in almost all the work I am a part of, we simply focus on either the photoanode or photocathode and look to optimize that reaction. When we get to the point where the crossover between the photoanode and photocathode is respectably high, we will then put together the total device. We are not at that point yet.

Our Research-Past

One important point that really complicates things is issues relating to ohmic losses due to ionic transfer. This issue is well known in the electrolyzer field, and a recent (report shows it can also be a major issue for photocatalytic water splitting. Basically neutral pH has low ionic conductivity even if you add a large amount of ionic species. However highly acidic or highly basic solutions can take advantage of water's ability to dissociate into H⁺ or OH^-, and can be much more conductive. This allows for ohmic losses to be greatly reduced. However these extreme pH's one must go to also causes stability issues with many photocatalysts. Simply put, few photocatalysts are stable in extreme pH conditions. Many photocatalysts also struggle to stay stable in either the highly reductive potentials needed for hydrogen evolution or the highly oxidative potentials needed for oxygen evolution. Since stability is a function of both pH and potential, Pourbaix diagrams are quite useful in looking for thermodynamically stable materials.

As our recent analysis work has shown, finding thermodynamically stable materials for 2 photon water splitting is quite difficult. However there is a way around this stability issue. Using protection layers dissociates the stability issue from the photocatalyst issue. Thus, in effect, we can have a great photocatalyst that doesn't need to be stable as long as it has a valid protection layer.

Our first foray into protection layers was trying to protect a Si photocathode. With a band gap of 1.1 eV, Si is almost a perfect small band gap material. However Si oxidizes to the insulating SiO₂ very easily, which makes it extremely difficult to use in water splitting applications. Our approach was to sputter on a thick enough layer of titanium metal to protect the Si, but yet thin enough to allow light to pass through. This approach was relatively successful, but the Ti absorbed light and did fall off after some time. To resolve this issue we used a thick sputtered TiO₂ layer instead. (Importantly we still kept in a small amount of Ti so the TiO₂ didn't oxidize the Si.) TiO₂ was successful for a couple of reasons. First it is naturally n-type which means charge transfer is primarily via electrons going through its conduction band. Secondly the location of the conduction band is almost perfectly aligned with that of the H₂ evolution potential. Thirdly is that TiO₂ tends to naturally have a high doping level, which increases its conductivity. Since the protection layers was on the order of nanometers, the TiO₂ was conductive enough to not cause any noticeable ohmic resistance.

The Si with sputtered TiO₂ worked perfectly for 3 days with no stability issues. The Si was only irradiated with small energy photons (to accurately portray its use in a 2-photn device), thus there was no photoexcitation of the TiO₂. At a later date in collaboration with the Gratzel group we used atomic layered deposition of TiO₂ and tested the device for 30 days with no noticeable stability issues related to the Ti or Si. However in these works we were irradiating from the side of the Si facing the electrolyte.  However for a 2-photon tandem device the light will actually be irradiating from the backside of the small band gap Si (See cartoon above showing 2-photon design).  Irradiating light from the backside is not trivial because the backside needs to absorb light and the front side needs to do catalysts, thus there is nowhere to put an electrical contact to test how well the device is performing.  Furthermore when the Si is irradiated from the backside the light absorption is on the opposite side from the catalysts thus making charge transfer slightly more complicated.  However we resolved both the electrical contact issue as well as found the optimal compromise between light absorption and charge transfer and published our results.

There was also another property of TiO₂ in which we were not fully taking advantage of: its transparency. TiO₂ has a band gap of 3.2 eV (for anatase) and by looking at the above figure which plots photons vs photon energy, it should be apparent that TiO₂ absorbs very little of the solar spectrum. However, if you review the scheme of a 2-photon device (see above figures), you will realize that since Si is a small band gap material, the protection layer will be on the opposite side of the incoming light. Thus TiO2 transparency means little on the backside of the device. It would be much more interesting if we could protect the large band gap material because in that case having a transparent protection layer is the utmost importance. While the above figures show H₂ being produced on the small band gap and O₂ being produced on the large band gap, we thought if we switch the reactions around (i.e. H₂ produced on the large band gap material) we could take advantage of TiO₂'s transparency.

As a model system we used GaP as a large band gap material to evolve H₂. We showed that this material could achieve H₂ evolution without a TiO₂ protection layer, but it corroded rapidly. Once we added the TiO₂ protection layer we mitigated corrosion issues. Unlike Si, GaP is not a highly researched material and thus it is not easy to make p-n junctions to maximize photovoltage. However from this work we found that the p-GaP and n-TiO₂ actually form a p-n junction, which helps to boost photovoltage.

With the photocathode protection layer under control, we moved onto looking at protection layers for the photoanodic O₂ evolution reaction. With the idea of using the large band gap material to evolve H₂, that leaves the small band gap material to evolve O₂. Thus we looked at protecting our Si with protection layers. We initially thought that Ti could not be used as a protection layer since we showed that at anodic potentials, the TiO₂ prevented any charge transfer due to band bending effects. However we realized that since Si is on the side opposite from the incoming light, light absorption is of minor concern. This allowed us to deposit a thin Ir/IrO₂ protection layer. Given that IrO₂ is the only stable O₂ evolution catalyst in acid, this allowed us to produce a combined protection layer/O₂ evolution catalyst all in one. We also took this same approach for protection layers in basic conditions. In this case we used a Ni/NiO for a protection layer/O₂ evolution catalyst. Since NiO is inherently not a very good O₂ evolution catalyst, we cycled our device in a ferrous solution. This allowed us to produce Fe intercalated NiO, which is one of the best O₂ evolution catalysts.

A provocating result by the Lewis group, led us to rethink TiO₂ as an anodic protection layer.  From this we discovered that as long as TiO₂ was covered by a film of catalyst, the TiO₂ would interact with the film and not the electrolyte.  Evaporation of our Pt catalyst onto TiO₂ gave us a Shottky barrier and thus prevented charge transfer, however sputtering Pt onto TiO₂ gave us an ohmic contact and thus this ohmic contact allowed us to use TiO₂ as an anodic protection layer.  (It should be noted that Gary Hodes was the first one to discover that evaporated Pt-TiO₂ form a Schottky barrier whereas sputtered Pt-TiO₂ forms an ohmic contact.  While he presented this at conferences in the 1980's, unfortunately he never published this.)  We also went further than just putting on films of catalysts, and put on islands of Pt catalysts.  While the majority of the TiO₂ was in contact with the electrolyte, locally their was a  TiO₂-Pt interface which allowed for the ohmic contact to dominate.  Thus even with Pt islands TiO₂ could still work as a anodic protection layer.  It was only when Pt nanoparticles were used did the TiO₂ band bending dominate and thus prevent any anodic reactions from taking place. While our results were very useful from a applicability standpoint, to understand exactly how mixed barrier heights interact (i.e. ohmic contact vs. band bending) I refer the reader to an excellent article by the Lewis group.

This idea of re-thinking of old concepts came up again when we wrote our recent prespective on H₂ evolution catalysts. (See accompanying video here).  One thing we realized is that while we have made impressive strides in H₂ evolution catalysis with the discovery of MoS₂, CoP, and others, they are still not as good as Pt.  We then thought how much Pt would you need for H₂ evolution if you wanted to base the entire world's energy on the H₂.  The first thing we did was model how little Pt you could use and then tested it experimentally.  (This is much harder than it sounds).  From our results one could argue that about 15% of the current worldwide Pt production could be enough to support 1TW worth of devices per year.  However  we  did not determine stability of the Pt due to contamination issues.  Thus if the devices are only stable for 1 week, we will not have enough Pt to meet our energy demands, however if they are stable for 20 years, that is 20 TW worth of energy we can get.  We also compared the Pt to other catalysts to look at their ability to upscale using current raw material production rates.  Surprisingly some of these 'earth abundant' catalysts currently have bigger problems scaling up than Pt.  However one should be very careful when using this data due to the assumptions involved.  Our paper goes into much more detail about the limitations of this analysis.

Our Research-Present

The diagram on the right shows our current approach towards 2-photon water splitting. It is a fluid approach, thus things could change tomorrow depending upon results in the laboratory. In this design the greatest bottleneck is our large band gap photoabsorber. We need a photocatalyst that can provide 1.1-1.3 V open circuit voltage with a band gap that is sufficiently small to absorb a large amount of the solar spectrum. Amorphous Si, Perovskite solar cells, Organic solar cells, and certain III-V semiconductors may be able to meet this requirement, but each has their issues. We are currently investigating these approaches and are working hard to optimize this area. On all other parts of the device we have a viable solution to at least some degree. However there is room for improvement in all these areas. The TiO2 protection layer can be improved by using a material with a slightly higher band gap. TiO2's 3.2 eV band gap still means it can absorb about 3% of the light. A higher band gap material could potentially allow for that extra 3% of light to reach the photocatalysts. Currently we deposit our TiO2 via ultra-high vacuum techniques, thus we are looking for cheaper techniques that can be done at atmospheric pressure.

Connecting the large band gap material to the Si is not necessarily a simple task. An ohmic contact must be made to prevent a Shottky diode, but this must be done without damaging the surface of either material otherwise defects will produce recombination sites, which hurts the photovoltage. While there is nothing fundamentally impossible, the technical details can be challenging. Currently we are investigating and optimizing this issue with regards to our individual system.

World record Si solar cells have an open circuit voltage of 740 mV and the ones constructed in our lab have only reached 580 mV. Due to engineering constraints from water splitting, we will not be able to reach 740 mV, but we should be able to produce results significantly better than our current wafers. Any voltage we save here is less voltage we need for our large band gap material. Thus while our Si photovoltage is among the world leaders in the field of photoelectrochemistry, we still are working on improving this.

The metal protection layer goes hand-in-hand with the oxygen evolution catalyst. This is because a metal needs to be deposited onto the Si to prevent oxidation of the Si, and metal oxides are the best (only ?) oxygen evolution catalysts. Thus having the metal and metal oxide be the same material mitigates interfacial issues between the protection layer and catalyst. Since there are many catalysts that are fighting to claim the role as best oxygen evolution catalyst, there are many different materials we need to test in our devices. However all of these still have around 300-350 mV in losses at photocatalytic water splitting currents, thus switching catalysts will likely only produce marginal improvements. Recently it has been discovered (1,2) that there is a fundamental scaling relationship between two of the oxygen evolution intermediates, which fundamentally limits the catalytic efficiency. Our O₂ evolution group (led by Ifan Stephens and Ib Chorkendorff) is working on finding ways around breaking this scaling relationship, which should lead to much lower losses. Any efficiency breakthrough they make will be quickly disseminated to us where we then apply it to our Si and test for photoanodic O₂ evolution.

One overarching issue that we are keeping in mind is the ionic transfer. While it is essential that we get the electron from the anode to the cathode at the proper energy level, it is also essential that we get the ions to transfer from one side to the other (without the deleterious effects of gas crossover). Thus we are looking at ways to construct our catalyst/protection layer/photocatalyst devices in a way that allows facile ionic transfer to minimize ohmic losses. Most probably a membrane will be needed to separate the anode and cathode, but we are not currently working on that technology. Currently Nafion is an excellent material for proton transfer (if done in acid) and there are some up and coming materials for hydroxide transfer (if done in base).