Item #1 on the National Academy of Engineering's list of the most important engineering challenges of this century:
- Make solar energy economical. Solar energy provides less than 1% of the world's total energy, but it has the potential to provide much, much more.
The detail page claims that the blurb is wrong; only 0.001% of the world's total energy usage is sold in the market, of which a fraction is fossil fuel. Additionally, although it's not mentioned in the page, many of the world's poorest people live entirely off of solar energy. It's just that the other 99.999% of the energy isn't easily used to drive mechanical engines at the moment.
Some of it is captured by plants in sugars and cellulose through photosynthesis, and that is called the planet's "net primary production". Peter Vitousek and others estimated in 1986 ("Human Appropriation of the Products of Photosynthesis") that humans consume as food or firewood 3% of its net primary production (including food for livestock) totaling 7.2 Pg/year of dry biomass, and "co-opt" another 19%, out of a total of 132.1 Pg/year.
7.2 Pg of carbohydrates is about 1.5 * 10^20 joules. 132.1 Pg of carbohydrates is about 27.7 * 10^20 joules. The US DoE EIA International Energy Outlook 2007 (PDF) reports that "world marketed energy consumption" was 447 quadrillion Btu in 2004, which is about 4.7 * 10^20 joules. This figure doesn't appear to include food sales.
A crude calculation (earthradius_equatorial^2 * pi * (1000 W/m^2) *
1 year in units(1) --- gosh, Unix is great!) suggests that the
total solar energy falling on the earth is about 40000 * 10^20
joules per year. If Vitousek's figures are right, that means that
the earth's ecosystem is about 0.07% efficient at converting
sunlight into biomass, and therefore probably not more than 1.5%
efficient at converting sunlight into anything but heat.
This leads to the conclusion that "world marketed energy consumption" is about 17% of the planet's net primary production, assuming that figure hasn't changed much since 1979, but that the planet's net primary production is only 0.007% of the energy received from the sun by the Earth. 17% of 0.007% is indeed about 0.001%, the figure from the original blurb.
At some point we'll have self-reproducing hardware, and the cost of manufacturing the solar cells will cease to be a problem; it will simply be a question of land use. To supply the 4.7 * 10^20 joules per year being currently sold in the market with the 40%-efficient solar cells in some labs today, we'd need to capture 11.75 * 10^20 joules per year of sunlight, which is 0.03% of the surface of the earth, including oceans. Matthias Loster made a lovely visualization of this and put it on Wikipedia.
If you were using 4%-efficient solar cells instead, you'd need 0.3% of the earth's surface area. Off-the-shelf inexpensive thin-film solar panels are about 8% XXX efficient.
In 2005, the National Renewable Energy Laboratory published a FAQ on energy payback times of photovoltaic cells, which explained that at the time, multicrystalline photovoltaic cells produced more energy than had been used to make them in 2-4 years. Here are their references; I haven't read any of them:
E. Alsema, “Energy Requirements and CO2 Mitigation Potential of PV Systems,” Photovoltaics and the Environment, Keystone, CO. Workshop Proceedings, July 1998.
R. Dones; R. Frischknecht, “Life Cycle Assessment of Photovoltaic Systems: Results of Swiss Studies on Energy Chains.” Appendix B-9. Environmental Aspects of PV Power Systems. Utrecht, The Netherlands: Utrecht University, Report Number 97072, 1997.
K. Kato; A. Murata; K. Sakuta, “Energy Payback Time and Life-Cycle CO2 Emission of Residential PV Power System with Silicon PV Module.” Appendix B-8. Environmental Aspects of PV Power Systems. Utrecht, The Netherlands: Utrecht University, Report Number 97072, 1997.
K. Knapp; T.L. Jester, “An Empirical Perspective on the Energy Payback Time for PV Modules.” Solar 2000 Conference, Madison, WI, June 16–21, 2000.
J. Mason, “Life Cycle Analysis of a Field, Grid-
Connected, Multi-Crystalline PV Plant: A Case Study of Tucson Electric Power’s Springerville PV Plant.” Final report prepared for Tucson Electric Power, November 2004.W. Palz.; H. Zibetta, “Energy Payback Time of Photovoltaic Modules.” International Journal of Solar Energy Volume 10, Number 3-4, pp. 211–216, 1991.
The payback time is a representation of a sort of compound interest rate; a payback time of 2 years is a 50% annual percentage growth, 4 years is 25%. If you spend the produced energy on making new solar cells, those are the actual growth rates of your stock of solar cells (plus a bonus, in the form of compounding the interest more frequently, if you get the new cells into production in less than a year).
So consider Evergreen Solar's current situation. They're a small company with a market capitalization of US$917M as of their last annual report. Their current manufacturing capacity is 15 megawatts per year, and they've contracted to manufacture 125 megawatts in 2009, 300 megawatts in 2010, 600 megawatts in 2011, and 850 megawatts in 2012. Suppose a company of similar size were to invest its 300-megawatt 2010 production merely in making more solar cells, and that it had no non-energy costs. An annual growth rate of 50% --- that is, a 2-year payback --- would look like this:
In [3]: ["%d: %.1fMW" % (2010 + x, 300*1.5**x) for x in range(30)]
Out[3]:
['2010: 300.0MW',
'2011: 450.0MW',
'2012: 675.0MW',
'2013: 1012.5MW',
'2014: 1518.8MW',
'2015: 2278.1MW',
'2016: 3417.2MW',
'2017: 5125.8MW',
'2018: 7688.7MW',
'2019: 11533.0MW',
'2020: 17299.5MW',
'2021: 25949.3MW',
'2022: 38923.9MW',
'2023: 58385.9MW',
'2024: 87578.8MW',
'2025: 131368.2MW',
'2026: 197052.3MW',
'2027: 295578.4MW',
'2028: 443367.6MW',
'2029: 665051.3MW',
'2030: 997577.0MW',
'2031: 1496365.5MW',
'2032: 2244548.3MW',
'2033: 3366822.4MW',
'2034: 5050233.7MW',
'2035: 7575350.5MW',
'2036: 11363025.7MW',
'2037: 17044538.6MW',
'2038: 25566807.9MW',
'2039: 38350211.8MW']
The current 4.7 * 10^20 joules/year being sold in the market is 14 893 719 megawatts, which this curve crosses around 2037, and with the usual 5:1 ratio between peak watts and achieved watts (due to nighttime, solar angle changes, clouds, etc.) you don't reach it until 2041 or 2042. If you start with the 1744 megawatts that Evergreen says Solarbuzz said constituted the global solar power market in 2006, you gain about 8 years.
However, the financial payback time on solar panels is still dramatically longer, which is why they still haven't reached "grid parity" --- costing less per watt-hour than power from the grid. Solar panels still cost US$4-$5 per peak watt at retail. That's about US$20 per average watt, which is 8760 watt-hours per year; that's about US$0.87 of electricity at grid rates. That's a 23-year financial payback, and that doesn't include things like batteries, inverters, wiring, and installation.
Evergreen's annual report suggests some reasons for this: the market is expanding at 42% per year, their own production capacity has to expand by a factor of more than 50 from 2007 to 2012 (this from a company that's already 13 years old). They had 276 full-time employees in manufacturing to reach their 15-megawatt-per-year capacity: 18 employee-years per megawatt. Their new 80-megawatt-per-year facility is expected to require another 410 employees (5 employee-years per megawatt). They are struggling to increase manufacturing capacity fast enough to keep up with demand, and apparently so are their "polysilicon" suppliers, because there's an industrywide shortage of polysilicon; building new polysilicon manufacturing facilities takes several years.
(I'm a little bit dubious about their terminology; I think the company's management may just not be very technical, or maybe not very smart. "Polysilicon" is short for "polycrystalline silicon", and silicon becomes polysilicon at the point that you crystallize it in a polycrystalline form in your furnaces. So the suppliers are supplying Evergreen with silicon; how many crystals are in each piece of silicon they supply is somewhat immaterial, since Evergreen melts the silicon down and crystallizes it in polycrystalline silicon ribbons in their furnaces.)
They report that they had US$58M of product revenues in 2007, with US$53M "cost of revenue", which presumably includes things like manufacturing employee salaries, energy, and raw materials. They spent US$21M on research and development and another US$21M on "Selling, general, and administrative", and US$1.4M on "facility start-up", building a new plant to increase their manufacturing capacity from 15MW to 95MW this year.
So suppose they manufactured 15MW in 2007, as their annual report suggests. That would mean they got paid US$3.87 per watt on average, which is more or less in keeping with the US$4-$5 the panels cost at retail. They spent US$3.53 of that on their actual manufacturing costs. They explain:
The main purpose of our Marlboro facility [where all of their manufacturing currently takes place] is to develop and prototype new manufacturing process technologies which, when developed, will be employed in new factories. As such, our manufacturing costs incurred in Marlboro are substantially burdened by additional engineering costs and also reflect inefficiencies typically inherent in pilot and development operations.
Elsewhere they explain that they use about 5 grams of silicon per watt; metallurgical-grade silicon costs about US$0.77 per pound , or US$0.0017/g.
They don't break out the costs of the silicon they buy from their suppliers, which might cost considerably more than the metallurgical-grade silicon it's made from. It appears that they have made prepayments and cash loans of, as I read it, about US$50M on a set of multi-year silicon supply contracts, although they only list US$23M in their "prepaid cost of inventory" line item; and elsewhere they say, "we have silicon under contract to reach annual production levels of approximately 125MW in 2009, 300MW in 2010, 600MW in 2011, and 850MW in 2012", for a total of 1875MW; and they say, "We believe future enhancements to our technology will enable us to gradually reduce our silicon consumption [from 5g/W] to approximately two-and-a-half grams per watt by 2012."
So suppose those 1875MW are to be made at an average of 3.5 g/W; that's 6600 million grams of silicon. And suppose the US$50M represents about half of the total price of that silicon; that would give us US$0.015 per gram of silicon. That's more or less in line with the raw silicon cost I estimated above for metallurgical-grade silicon --- it's higher, and by a factor of only about 2 --- which gives me some confidence that my guesstimate that the cost of the silicon is not yet a significant factor in the cost of the solar cells.
However, note that securing one of these long-term silicon supply agreements required selling about 15% of the company to the silicon supplier. The restrictions on that stock "will lapse upon the delivery of 500 metric tons of polysilicon to the Company", so we can guess that the total contract with that supplier is in the neighborhood of 1000-2000 million grams.
Also, they list $629M in "raw materials purchase commitments" among their "contractual cash obligations". This, plus the prepayments, is perhaps a ceiling on the amount of payment they may have committed to for the silicon; US$679M for 6600 million grams of silicon would be US$0.10 per gram, which would raise the cost of silicon above from US$0.008 per watt to US$0.35 per watt. (It's possible that they have other raw materials purchase commitments, say for silane or hydrofluoric acid.)
In 2006, Evergreen and EverQ bought US$8M worth of silicon from REC, who I think was their sole silicon supplier at the time (unless DC Chemical was also a supplier?). During that year, they had $102K of sales. In 2007, Evergreen bought US$3M worth from REC, which is US$0.20 per watt if they produced 15MW.
Let's assume that their per-employee cost of labor on the factory floor is about US$120 000 per year. At 18 employee-years per megawatt, that's about US$2.2M per megawatt, or about US$2.20 per watt.
If we assume that the NREL numbers are applicable to their manufacturing, then each peak watt of panels required about 4kWh of energy; let's assume that costs US$0.10/kWh. So, per watt, we have:
revenue US$3.87
gross profit US$0.34
electricity US$0.40
raw silicon US$0.008
labor US$2.20
OTHER US$0.92
At present, they're also spending about half of their revenue on research and development. (That's part of why they're still losing money.) We can expect that the cost of labor per watt will decrease substantially in their 80MW non-pilot facility: 5 employee-years per megawatt would be US$0.60 per watt.
They also have been spending on the order of US$50M per year on capital expenditures, mostly equipment and facilities improvements. They report that their "fixed assets, net" are worth US$115M, including US$53M of "laboratory and manufacturing equipment", US$14M of "leasehold improvements", and US$67M of "assets under construction". They seem to expect that constructing the first 80MW/y production line in their new facility will cost around US$100M, although they don't really break it out that way in the report. That's about US$1.25/watt/year.
A capital cost of US$1.25 per watt/year of manufacturing capacity does not unavoidably contribute much to the cost per watt; after all, you can in principle amortize it over an arbitrary number of years. However, in an industry with a 42% annual growth rate, almost all cells will necessarily have come out of factories built within the last year or two, so it probably adds US$0.60/watt or more to the cost of the cells.
EverQ, a separate company that Evergreen owns a third of, had operating revenue of US$194M, cost of goods sold of US$160M, "other expenses" of US$27M, and assets of US$556M. I wish I had handy EverQ's manufacturing capacity numbers.
Nanosolar claims an energy payback time of one month and a per-watt cost of 30 cents with their copper indium gallium diselenide thin-film cells, in a November 2007 article on Celsias, although they had expected a cost in the sixties of cents per watt in a July 2007 interview. In the Celsias article, they also say they plan to reach 430 megawatts of production per year in 2008.
In the interview, CEO Martin Roscheisen also says:
...it is clear we are going to be manufacturing capacity limited for about as far out as we can see. There’s presently really only two truly scalable solar markets in the world — Germany and Spain — and we do a lot there. Being a scalable market is today as much about feed-in-tariffs as about the administrative framework; tomorrow, with grid-parity PV systems, it is primarily about the latter.
As I said before, Evergreen is experiencing an industrywide polysilicon shortage; however, the raw material silicon is extremely abundant, being the principal component of one of the most common minerals in the terrestrial crust.
However, the materials used in copper indium gallium diselenide (CIGS) thin-film cells like Nanosolar's are somewhat less abundant. Copper has been a precious metal since the Bronze Age, but indium, gallium, and selenium are all fairly rare.
As a point of comparison, after years of rapid increase, silver prices averaged US$13.40 per troy ounce in 2007, according to the USGS's silver report. That's US$430 per kilogram. About 20 700 tons of silver were mined in 2007.
Indium, by contrast, cost US$795 per kg in 2007, and averaged an even higher US$918 per kg in 2006, and only 510 tons were refined in 2007, making it 40 times rarer than silver and 85% more expensive. The USGS claims, "Thin-film ... CIGS solar cells require approximately 50 metric tons of indium to produce 1 gigawatt of solar power," which still makes it a tiny fraction of the total cost. (I am assuming the USGS is referring to peak watts at one sun, i.e. in direct sunlight without lenses or mirrors, and not average output or solar-concentrator output.) That's US$0.04 of indium per watt, so the price of indium would have to increase by a factor of 75 to increase the cost of thin-film cells by US$3 per watt. That would be about US$60 000 per kg. I think grid parity is somewhere around US$1 per watt, which would be around US$20 000 per kg.
At higher prices, you would expect new low-concentration sources of indium to become economic to refine, which would be nice, because current world indium production is only enough for about 10 gigawatts of CIGS per year. It's difficult to predict what kinds of improvements could occur and how much they could increase indium production. However, we can get a little bit of a clue by looking at the last several years. In 2002, indium cost only US$130 per kilogram, so we've already experienced a dramatic price increase, driven by dramatically increased production of LCD displays, which use indium tin oxide for thin-film transparent electrodes. So how much did indium production increase when the price increased by a factor of seven over four years? It increased from 335 tons to 510 tons. [XXX check that. probably slightly wrong.]
So, although it's error-prone to predict, the evidence suggests that indium production capacity will prove quite difficult to scale up over the next several years, which could limit CIGS thin-film solar cells to a small fraction of the overall energy market.
Gallium is only slightly more expensive than silver, at US$460 per kg. Supplies of gallium are even more limited than those of indium; the USGS report estimates world primary gallium production capacity at 184 metric tons per year, and actual production at 80 metric tons per year, making it 250 times rarer than silver. In the absence of the LCD demand that has caused indium's price to skyrocket over the last several years, its price has remained relatively constant 2002-2007 even as imports have more than doubled. This would seem to suggest that gallium's production could be increased considerably more easily than indium, but I suspect that this is not the case, as I explain below.
The gallium prices are stated for extremely pure gallium, with less than 0.1ppm impurities, because this is what is needed for its largest-volume use, high-performance integrated circuits made of gallium arsenide, largely for RF components in cell phones. The USGS also reports some information on "low-grade" 99.99% pure gallium:
Prices for low-grade (99.99%-pure) gallium increased in the first
half of 2007 from $300 to $350 per kilogram at the beginning of
the year to about $500 per kilogram by midyear. Producers in China
claimed that there was a shortage of supply, which was the
principal reason for the increase in prices. Some were offering
gallium at prices as high as $800 per kilogram, but little
business was completed at this price level.
The reason I think gallium production will hit limits similar to indium production is that indium and gallium are chemically very similar, and they are both primarily refined from trace amounts (50 ppm or more, at present) found in zinc ores and bauxite, and consequently they are found as impurities in zinc. So I think it is unlikely that there are large amounts of easily recoverable gallium hiding somewhere without corresponding amounts of indium accompanying them.
Because of their chemical similarity, they are substitutable for one another in some semiconductor applications.
I believe CIGS contains equal numbers of atoms of indium and gallium, but I think the gallium is somewhat heavier. XXX I need to look at a fucking periodic table.
Selenium is also only found in trace amounts in the Earth's crust. I don't know how much it costs or how much is being mined.
Silicon solar cells are made from silicon, arsenic, boron XXX, and aluminum --- some of the most common elements on Earth. However, their processing XXX
Everything above --- costs per watt, factory production capacities in watts, materials per watt, etc. --- is about solar cells in "one sun", i.e. the intensity of sunlight that naturally reaches the surface of the Earth, which is about 1000 W/m². Silicon photovoltaic cells can theoretically turn up to 31% of that into electricity, but the less expensive polycrystalline cells in common use are only about 12% efficient, with even lower efficiencies of 9-12% or so for thin-film cells and 6% for amorphous silicon cells. There are more expensive "multijunction" non-silicon cells available for sale now that are 34% efficient, and 41%-efficient cells in laboratories that will presumably reach production soon; and there are quantum-dot and photonic-crystal approaches that could reach 60% in theory. (Some of these numbers are from the NREL report cited earlier, while others are from the National Academy of Engineering page cited earlier.)
However, these more-watts-per-unit-area approaches are very expensive per watt, so they are currently mostly only used in space missions --- to power satellites and the like.
Most types of photovoltaic cells continue to work in higher-intensity light, even working at higher efficiencies [XXX]. If you have mirrors that cost less per square meter than your solar cells, you can use mirrors to gather the same amount of sunlight onto a smaller area of expensive solar cell, for a lower overall system cost. This sort of thing is called a "solar concentrator", and there are some very-large-scale systems that don't even use photovoltaic cells at the focal point, instead using heat engines like an old-time locomotive, which can be more efficient at sufficiently high temperatures.
One experimental project uses a balloon, half of aluminized mylar, half of transparent mylar, to make a concave reflector for a small photovoltaic panel. In photos, it looks like it generates about "100 suns", or 100 times the normal intensity of sunlight. This means that "one watt" of solar panels, rated according to normal sunlight, can produce 100 watts or a little more [XXX confirm this], with the aid of a square meter or so of aluminized mylar, which costs on the order of US$2, and can be recovered abundantly from garbage in many areas. However, I suspect it needs some special cooling [XXX check this].
This kind of setup could theoretically be quite inexpensive and sturdy, but there are difficulties. Your hundred-sun system will suddenly become a zero-sun system if it's not pointed fairly accurately at the sun, so it requires control motors to follow the sun across the sky; this adds to the cost, and also reduces reliability. Your balloons will eventually deflate, and you have to reinflate them. And on cloudy days, your hundred-sun system is, at best, a one-sun system. So most of the production solar concentrators I've heard of have been large-scale thermal generators.
If your 1m² concentrating mirror cost US$5, your 100cm² 12%-efficient photovoltaic cell cost US$5, your motors and control system cost another US$20, and your cooling cost another US$20, you'd have a US$50 system producing about 120 watts, or about US$0.42 per watt. If you could upgrade to 24% efficient cells that cost another US$10 (I have no idea if this price is realistic), you'd have a US$60 system producing about 240 watts, or about US$0.25 per watt --- even though the solar-cell component of the system cost 50% more per watt, the system as a whole cost less per watt. In this way, photovoltaic concentrator systems can economically take advantage of more expensive photovoltaic materials, as long as the solar cells themselves are a small part of the cost of the system.
You would think that this kind of technology would have been adopted wholesale long ago, since it would appear to cost dramatically less per watt than fossil-fuel plants, not even counting the cost of the fuels. So there must be some difficulties that have prevented it from achieving the kind of efficiencies I've suggested above, at least scalably.
There are various experimental systems working on this principle: Solient's (see also the Technology Review article),
In summary: photovoltaic solar concentrators could, in theory, provide electrical generating capacity for US$0.05--US$0.50 per watt with current technology, and I don't know of any practical reason this potential won't be realized. But I also don't know why it hasn't already been realized, say ten or fifteen years ago, and there must be a reason; and maybe that reason still applies.