I went to this year’s second Everhart Lecture yesterday by Josh Spurgeon, who is working with Harry Atwater and Nate Lewis, trying to develop cheap, scalable solar cells. As with most of the Everhart Lectures, it was a very well presented talk. Unlike many of them, it was directly relevant to a real-world problem: how can humanity continue to utilize on the order of 10TW of power, without changing the composition of the atmosphere (see Nate Lewis’ excellent presentation for more information). The ultimate solution to that problem will almost certainly involve directly capturing incident solar energy, because the potential resource available is both vast and relatively concentrated, when compared to other sources of renewable energy. But solar has two very serious problems today: it is expensive (both in absolute terms on a per watt installed basis, and in an up-front capital expenditure sense), and it is not available when the sun isn’t shining. Whatever the solution looks like, in order to scale up to 10TW, it needs to use only earth-abundant, non-toxic materials. In semiconductor photovoltaics then, silicon probably has an unassailable lead. It’s the second most abundant element in the Earth’s crust, and it’s about as toxic as sand (though silicon semiconductor fabrication has serious toxicity associated with it and certainly needs to be made closed-loop). Exotic materials like cadmium-telluride, and copper-indium-gallium-selenide (CIGS) are unlikely to scale to tens of terawatts, simply because of the limited availability of elements like indium and tellurium. Additionally, owing to the vast silicon microprocessor industry, we are much better at micro and nano-scale manipulation of silicon than any other material on Earth (ignoring for the moment biological systems).
Spurgeon’s work so far has been primarily focused on cost reduction, which is a little bit unusual for an academic researcher. A lot of what has so far made photovoltaics expensive is the fact that in order for them to be efficient (i.e. to convert a high proportion of the sunlight into electricity), they require very high quality materials: large, nearly perfect crystals of silicon, as are used in the microprocessor industry. This requirement, it turns out, is actually a side-effect of the simple planar geometry which all solar panels to date have assumed. That is, you have a sheet of the semiconductor, which is used to absorb incident photons traveling roughly perpendicular to the surface of the sheet. The electron-hole pair generated by the absorption subsequently also migrates perpendicular to the surface of the sheet, until they each reach their respective conductive collecting layer and are swept away into the grid to run your iPod, blow-dryer, or whatever (if it’s the electron). The problem is that the ideal length-scales for these two processes (photon absorption and electron-hole diffusion) are different. The absorbing medium needs to be thick enough that the photon is absorbed – the thicker the better, up to a point – but the further the electrons and holes have to diffuse, the more likely they are to recombine and thus be wasted, reducing the cell’s overall efficiency. Impurities and crystal defects or grain boundaries act like traps for the migrating charge carriers, and increase the chances that a given photon absorbed will only generate heat, instead of useful current. In the planar geometry, you’re forced to make the electron diffusion distance the same as the photon absorption distance, which is relatively large. In order to get high efficiency then, you need nearly perfect silicon crystals, hence the expense.
Spurgeon’s approach is to use a few of the tricks of the chip fabrication trade (and a few of his own) to change the geometry of the problem, and separate the photon absorption and charge carrier diffusion lengths, allowing each of them to be independently optimized. Instead of a simple plane of silicon, he creates what looks like a carpet of silicon wires, each on the order of a micron wide, but tens or hundreds of microns long. Photons are absorbed along the axes of the wires, but the charge carriers migrate radially, meaning that the quality requirements for the material are much lower. However, the wires are grown by vapor-liquid-solid deposition on an expensive, near perfect single crystal silicon wafer. To make the wires cheaper than the wafer, it has to be re-usable, so after growth they’re embedded in a silicone polymer goo, and sliced off the wafer with a decidedly low-tech razor blade. Not only does this allow re-use of the growth substrate, it also transforms the microwire array into a flexible and reasonably durable film that can be easily manipulated with tweezers in lab. The whole process is designed to lend itself to roll-to-roll processing, which would allow scalable (and thus cheap) manufacturing of the resulting PV cells. At the moment, the efficiency of the cells seems limited primarily by the density and pattern of the microwire arrays, and the speed with which the generated current can be transported away from the cells by the liquid charge carrier they’re working with at the moment (any finished product would be solid-state, and not face this limitation).
All in all, the system seems elegant and plausibly very disruptive (in a good way) to the PV market. I can’t believe he got it all working in the course of a PhD. One has to wonder how much of that is luck, and how much is skill. My impression is that a lot of the distribution of success and failure in graduate research is due to the statistics of small numbers, and that like Fermi’s “great generals”, a fair number of bright burning young research stars are just plain lucky (in addition to being very smart and hard working). At this point, we could all use some luck when it comes to global energy supplies.
One piece of context that I wish he’d touched on was first brought to my attention by Harry Atwater in his NRG talk two years ago (streaming video). Today only about 50% of the cost of a PV installation is due to the solar cells. The rest is for the structural elements and power electronics. This means that, assuming we’ve already pretty well cost optimized the mechanical and electrical systems, the best we can hope for with cost reductions in PV cells is a 50% reduction in cost, which isn’t enough for solar to compete directly with fossil fuels (it needs more like a 90% cost reduction in the absence of emissions regulations, and other sensible re-arrangements of our electricity systems). Alternatively, the efficiency of PV cells could be raised, generating more power per unit area, and thus per dollar invested in structure. The best commercial cells today are something like 25% efficient, and DARPA has a running challenge for 50% efficient PV cells (which some R&D groups are starting to get close to, and, um, why does DARPA have this challenge? Why isn’t it DoE for civilian consumption?). I wasn’t able to glean from the talk what kind of upper bound in efficiency the microwire PV system might have. At the same time, if these things can be made cheap enough, then the structure can kind of just disappear. A lot of the motivation for an expensive optimizing structure is the cost of the solar cells. If they were cheap, we’d just roll it out on every roof, and design future roofs with more optimal solar exposure, allowing the structure we were already building for architectural purposes to double as support for the power generating carpet.
The video of the talk will be up here at some point hopefully soon. I wish they’d just set up a Caltech iTunes U, or YouTube Channel already. This research and more from the Lewis lab is also covered in Caltech’s painfully paper and PDF quarterly Engineering and Science, (Vol. 71, No. 2, 2008) in an article entitled Rods and Stones (PDF, worth it for the pictures).