We started out playing around with metal nanowires, and measuring the open-circuit voltage (that is, hook up a volt meter across the device, which nominally doesn't allow current to flow) across those wires as a function of where we illuminated them with a near-IR laser. Because the metal absorbs some of the light, that laser spot acts like a local heat source (though figuring out the temperature profile requires some modeling of the heat transfer processes). As mentioned here, particles tend to diffuse from hot locations to cold locations; in an open circuit, a voltage builds up to balance out this tendency, because in the steady state no net current flows in an open circuit; and in a metal, the way electron motion and scattering depend on the energy of the electrons gives you the magnitude and sign of this process. If the metal is sufficiently nanoscale that boundary scattering matters, you end up with a thermoelectric response that depends on the metal geometry. The end result is shown in the left portion of the figure. If you illuminate the center of the metal wire, you measure no net voltage - you shouldn't, because the whole system is symmetric. The junction where the wire fans out to a bigger pad acts like a thermocouple because of that boundary scattering, and if you illuminate it you get a net thermoelectric voltage (sign depends on how you pick ground and which end you're illuminating). Bottom line: Illumination heats the electrons a bit (say a few Kelvin), and you get a thermoelectric voltage because of that, to offset the tendency of the electrons to diffuse due to the temperature gradient. In this system, the size of the effect is small - microvolts at our illumination conditions.
Now we can take that same nanowire, and break it to make a tunnel junction somewhere in there - a gap between the two electrodes where the electrons are able to "tunnel" across from one side to the other. When we illuminate the tunnel junction, we now see open-circuit photovoltages that are much larger, and very localized to the gap region. So, what is going on here? The physics is related, but not true thermoelectricity (which assumes that it always makes sense to define temperature everywhere). What we believe is happening is something that was discussed theoretically here, and was reported in molecule-containing junctions here. As I said when talking about hot electrons, when light gets absorbed, it is possible to kick electrons way up in energy. Usually that energy gets dissipated by being spread among other electrons very quickly. However, if hot electrons encounter the tunnel junction before they've lost most of that energy, they have a higher likelihood of getting across the tunnel junction, because quantum tunneling is energy-dependent. Producing more hot electrons on one side of the junction than the other will drive a tunneling current. We still have an open circuit, though, so some voltage has to build up so that the net current in the steady state adds up to zero. Bottom line: Illumination here can drive a "hot" electron tunneling current, and you get a photovoltage to offset that process. This isn't strictly a thermoelectric effect because the electrons aren't thermally distributed - it's the short-lived high energy tail that matters most.
It's fun to think about ways to try to better understand and maximize such effects, perhaps for applications in photodetection or other technologies....