Solar cell technology August 19, 2009Posted by Maury Markowitz in solar.
Tags: solar power
I previously wrote an article about the recent spate of space-based power press releases, noting that I felt they were nothing more than pipe dreams. I noted that the weight/performance ratio of solar cells is too low to make any system practical unless there were utterly unprecedented cost reductions in space launches, on the order of 100 times.
But a reader pointed out that I had skipped over a part of one of the proposals: that the improvement in thin-film PV systems would improve the price/performance ratio from the “other side”. At first I was going to address this point, but as I got to writing I decided to expand this out into a discussion of PV technology in general. I think you’ll find this useful. Enjoy!
When light, a stream of “photons”, shines on a material, the energy in the photons can knock about the electrons in the material. The chance that this can happen is based on two factors – how strongly the material holds onto its electrons, and how much energy is in the photon. As long as the later is more than the former, the electrons can be popped off their atoms, creating a “photoelectron” in a process known as “photoemission”. The basic idea of any solar cell is to collect those photoelectrons to power external electrical devices.
How strongly a material holds onto its electrons can be simplified to a consideration of the sizes of the atoms. In a large atom, like copper, the outer electrons are located so far from the center of the atom that they are only a small jump away from the next atom beside it. That’s why copper is a great conductor – any little push will knock the electrons off their atoms, sending them coursing through the wire. On the other end of the scale is something like plastic, which consists of smallish atoms locked into long chains and separated from each other by large distances. In this case the electrons need a major knock to move about, which is why we insulate our copper wire with plastic coatings.
Between these extremes lie the semiconductors. As the name implies, these materials sort of conduct, at least under certain circumstances. Generally, heating them up causes the electrons to gain enough energy that they can start moving about the material. If they get too hot, too many electrons are energized and the process limits itself. The key measure of the effectiveness of this process is the “band gap”, the amount of energy you need to get an electron to jump to the conductive state.
So why are solar cells made out of semiconductors, and not conductors? Consider what happens when a photon hits an electron in copper and sends it moving. As it flies off, it leaves behind a charged atom, missing one electron that it wants. So when the next photon comes in and hits another electron, it is attracted to that “hole” and moves only as far as that earlier photoemission. There’s lots of tiny currents inside the wire, but nothing comes out the end, the electrons are just moving back and forth within the wire.
What you need is a valve…
Semiconductors have a trick up their sleeve. With a little processing you can make one portion of the material have a constant overabundance of electrons or holes, as you want. It’s very easy to create both types of material in a single crystal, you make a bulk material biased one way or the other, then heat it up and spray one side with a gas. The gas mixes with the crystal and biases the small layer in contact the other way. You’re left with one side of the material having one net charge, and the other the opposite. The area where they meet is called the “p–n junction”, for “positive-negative”.
When photoemission occurs in such a material, the electrons are strongly pulled towards the junction. They get trapped there, on the other side of the boundary. The hole they left behind is pulling on them, but they can’t get back through the junction, so the cell becomes more and more charged as the light falls on it. To correct this imbalance you simply attach a wire from the top of the cell to the bottom, giving the electrons a path back to their holes. They’re so eager to get through that wire that they will happily do some work for us on the way.
The Unbearable Being of Light
The energy of the light is the other part of the equation. White light, like sunlight, is a mix of different colors. Those colors have different amounts of energy – each photon of blue light has about twice as much energy as one of red light. We call this collection of colors, and energies, the “spectrum”. The “visual spectrum” is what you’ll be most familiar with, covering the range from red to violet, but the “full spectrum” keeps going in both directions, down to radio on one end and gamma rays on the other. Yes, radio waves are light, you just can’t see them in your eye.
If we want to produce power from sunlight, we’re going to want to trap as much of the energy in that light as possible. About half of all the energy that hits the surface of the Earth is visible, peaking in yellow. The other half is in the infrared. Ideally you’d like to select a semiconductor that has a bandgap that’s small enough that any of the light coming from the sun will cause photoemission, so one with a bandgap energy close to IR would seem to be right.
But there’s a twist: any energy in the photon above and beyond the bandgap cannot easily be collected, it’s lost. While that electron travels to the bandgap it gives up any extra energy as heat. So if you select a material that’s sensitive to IR, all that extra energy in the blue light is lost, and the overall power production goes down. And it’s a double-whammy, you’ll recall earlier that as a semiconductor heats up it looses its ability to conduct. So not only do we not get that energy out of the cell, the losses are actually lowering the efficiency of the cell as a whole.
The ideal solution is to use a bunch of different cells, each one tuned to collect as much power as possible from a particular color of light. You have one that gets as much of the IR as you can, another for red, another for yellow, and so on. Luckily, photons will only react with materials that are significantly larger than their wavelength, so if you make the layers thin enough you can stack them on top of each other (blue on top, IR on the bottom) and collect all of the light in the stack.
But that’s not trivial, making a single “junction” in a semiconductor is fairly easy, making multiple junctions is much harder. So for cost reasons you generally want to use a single junction. If you run the numbers it turns out that you’re best off selecting a material that becomes sensitive in the red, about where the eye does. Luckily, there’s a whole bunch of semiconductors right in that range.
The semiconductor (“chip”) industry is based almost entirely on the use of silicon (Si) and various alloys. There is an enormous amount of research into silicon’s properties, it’s likely the most studied element in the world.
In order for a chip to work properly, the silicon must be extremely pure. Any flaw in the silicon, mechanical or chemical, renders it useless for chip building. In order to produce the almost-perfect product needed for chips, a raw feedstock of small silicon crystals is melted down and then a tiny “seed crystal” of silicon is placed on the surface and slowly pulled out to produce an enormous single crystal called a boule. The process is extremely expensive and time consuming.
Solar cells were originally made using this same chip-like process. Instead of printing chips on their surface, small lines of lead were deposited in order to collect the electrons being produced by the photoelectric effect. Presto, a solar cell. Not cheap, not terribly efficient, but it was one of the only effective ways to supply power in space. One simple drawback to this method is that the cells have to be circular, you can’t grow a square boule. That means that when you place them together on a plate to make a collector, there’s numerous cross-shaped spaces between them, taking up room. You can cut square parts out of the boules and throw away the rest, or sell the little moon-shaped bits for basically nothing.
As it turns out, the original silicon feedstock is also a useful solar cell material on its own. If the feed is melted down and cooled back into a solid, it can then be processed like a boule. This material, polysilicon, or “pSi”, isn’t perfect enough to be used as the basis of a chip, but it’s good enough for a solar cell, and it’s much less expensive to produce because it skips the “growing” process that is so time consuming. Another upside to this process is that you can make any shape, including squares, so when they are cut into cells they fit together better on collectors, increasing the area efficiency.
The major downside to pSi is that electrons get caught in the gaps where the many small crystals meet, lowering efficiency. Commercial mono-crystal Si solar cells can hit around 22% efficiency, whereas pSi versions are closer to 15%. However, because they cover more of the panel area, panels based on pSi get about 12% efficiency, while the more expensive single-crystal versions are about 15%. As always, heat lowers the efficiency, and a basic rule of thumb is that a pSi panel will be about 8% efficient on a hot summer day.
Until recently, demand for pSi was fairly low. Space applications wanted the best product they could get, regardless of the cost, so pSi simply didn’t make any sense for this field. Commercial terrestrial PV power production was strictly experimental, so there was little sustained demand there. The only major uses were in handheld calculators and similar devices, that had tiny power budgets and were cost sensitive. Most of the pSi in the world was cast-offs from the mono-crystal production runs, pulled out as needed or demand for other products fell off.
Everything changed when oil started heading north. Suddenly it looked like the future was going to be solar powered, and new panel manufacturers started popping up everywhere, especially China. By late 2004 there was a worldwide pSi shortage. To fill this need, by late 2008 a number of new factories dedicated to producing pSi “from scratch” (not as a feedstock for the chip industry) had come online and prices started falling again. I watched this happening while pricing out a system for my home. When I first started looking in 2007, panel prices were the lowest they’ve ever been, about $5 a watt. Now they’re about $3.50, and going down all the time.
Of course traditional silicon manufacture has also dropped in price. Many of the production lines set up in the 80’s and 90’s are no longer useful for chip making, but are just as useful as ever for PV. A number of companies, notably AMD, have turned over their older lines to producing boules for solar cell use. In spite of the process being more complex and expensive than pSi, wholesale prices are surprising competitive, and the price/performance ratio is right in line with high-end pSi panels.
The major cost of a pSi solar cell is the Si. So, obviously, if you reduce the amount of Si you use, the cells get cheaper. Unfortunately, that’s not so easy. As the cells get thinner the chance that any inbound photon will interact with an electron in the matrix starts falling. The cells eventually become almost transparent, and are useless for power production. What you’ve seen is what you get; conventional cells are only slightly thicker than the best they can be, and that’s in order to make them stronger so they’re easy to handle during fabrication.
Which is too bad, because in the 70s they realized that cells were much thicker than they should be for optimum performance. A photoelectron produced deep in the cell has to travel a very long distance (for an electron) before it reaches the PN junction where it can be collected. There’s a chance that the electron will interact with “something” on its way – a defect in the chip, a stray atom of an impurity, or a hole left behind by a previous photoemission. In this case the energy is lost, reducing the efficiency. Additionally, the further it travels the more of any extra energy it has is lost – if it was really thin the electrons from those blue photons will still have their extra energy when they make it to the junction.
So not only would a thinner layer be less expensive, it could offer higher efficiency as well, with predictions on the order of 25% maximum, higher than a single-crystal cell could hope for. This one-two punch led to a lot of research in the 1980s to see how to build such a cell. That turned out to be fairly easy, but sadly it also turned out it didn’t work. All of the materials had significant numbers of defects within the cell, which captured electrons with wide abandon, reducing efficiency into the 5% range. Most efforts died off by the mid-90s.
Most efforts. Some teams pressed ahead with their existing systems, while others followed developments in materials coming from other fields. Today there are a number of thin-film solar cell systems available commercially. There are two primary thin-film materials, amorphous silicon (aSi) and cadmium telluride (CdTe).
The former, as the name implies, is different form of silicon that you can think of as micro-poly-silicon. Whereas pSi has mini-crystals a few millimeters in size, in aSi they’re a few angstroms in size (ie, really small). It’s this random layout that makes aSi able to capture light in thin layers; normal pSi looks like a large crystal from a photon’s point of view, and it takes a lot of layers before it hits the atoms and not the spaces between them. In aSi, the atoms are arranged randomly, so the light doesn’t have to go nearly as far before it will hit something.
Unfortunately, when the first aSi cells were being built they quickly discovered a serious problem. In pSi the rough joins between the bits of crystal results in defects that trap the elections – in aSi the entire cell is one big defect. Many research teams tried to sealing off these defects with hydrogen or other atoms, but most of the teams gave up. Stan Ovinski, who one could safely describe as the world’s leading expert on all things amorphous, managed to work this through. However, aSi cells today still offer efficiencies around 10% or less, and are not cost competitive with normal pSi panels. They’ve found a niche in roofing systems because they’re mechanically flexible, but to date the price/performance ratio doesn’t shown any signs of making aSi a universal solution.
CdTe was noticed as a potential solar cell material in the 1960s because it’s band-gap is very closely tuned to sunlight (as is silicon). But there wasn’t a lot of development effort when demand for PV was low. More recently, a number of startups have been using CdTe for production panels. It’s typically printed in thin layers on a glass plate coated with a transparent conductor and backed with a metal film. Notable among the CdTe companies is First Solar, who have been in production for some time and have reached wholesales prices on the order of $1 a watt.
Right now, CdTe is relatively low-cost. However, cadmium is a fairly rare metal, typically produced as a side effect of copper mining. The demand for the cadmium itself is too low to justify production on its own. As a result, its price is extremely sensitive to changes in the price of copper. Moreover, the total production rate is far lower than the demand for solar cells in general – even at full copper production rates there’s nowhere near enough cadmium to build the solar cells to meet total demand. CdTe will always be a “second iron in the fire”, it’s unlikely it will be a primary production method in the future.
If you were following the electronics world in the 1980s, you’ll remember that the “next big thing” was chips made out of Gallium Arsenide, GaAs, instead of silicon. GaAs has many more electrons available in its structure, and as a result it can move them about much easier. In the chip world this was a major bonus, because it meant that it could switch on and off faster, making chips run at far higher speeds. At a time when the traditional CPU ran at around 5 MHz, GaAs chips were running at 500. But as it turned out GaAs had some very serious downsides, and as they were being slowly solved, billions of dollars were being poured into silicon research. Any advantages was eroded away through massive R&D.
But GaAs was by no means useless even if it didn’t go into common chips. The high switching speed not only made it useful for chips, but all sorts of other electronics as well, especially radios. Most high-end systems used by the military and comsats use GaAs components for major portions of their analog sections. The GaAs market remains small, but its an entrenched niche that shows no signs of disappearing.
GaAs’s main advantage, those extra free electrons, makes it useful as a solar cell as well. Each photoelectron produced is one less electron available for the next photon to excite. If the production rate is low enough this isn’t a problem, but eventually the sunlight will run out of electrons to use. In silicon this happens at light levels just past bright noontime sun – on the Earth. In space, where it’s always noon and there’s no atmosphere, there’s more sunlight than traditional silicon cells can use. That makes GaAs solar cells very useful in space applications, where you have a set amount of space to work with and the power-to-weight ratio is critically important.
One can build a solar cell using GaAs in the same fashion you do with Si. In practice, however, GaAs solar cells go way beyond traditional Si design. Remember that the maximum amount of power you can produce is a function of both the brightness and the color of the light. Since price is no object, the majority of GaAs solar cells consist of three or four cells, each one tuned to a different color of light and stacked with red on the bottom to blue on the top. This is possible only if the cells are extremely thin, but that is possible when you’re using GaAs. It’s expensive to make these compound cells, hundreds or thousands of dollars each, but they deliver about 30% efficiency and weigh less than a silicon cell of the same size, let alone the same power. For space applications there’s nothing that comes close.
Although they’ve only been used in space applications so far, there’s been some recent interest in GaAs cells for power generation down here too. The amount of sunlight that these cells can handle is far higher than noontime in space, its as much as 1,000 times higher. Better yet, their efficiency has a broad response curve that makes them get better at higher concentrations, between 50 and 1,000 times, depending on the particular cell. That means you can make a larger solar panel without using more cells. Instead of covering a panel with cells you use a single cell and a large collector, like a mirror or lens, and concentrate that light onto the cell. With a 1,000 to 1 ratio of concentrator to cells, even the high price of the individual cells can be offset by the low cost of the collector.
The downside to this approach is that the sun moves during the day. If you’ve ever used a small telescope, you’ll remember how hard it is to keep a planet in view as the Earth rotates. That telescope might have had 50 or 100 power, and here we’re talking about 10 times that magnification. In order to keep the panel focusing the sunlight on the cell, a complex mechanical system has to be used to closely track the sun. These trackers are so expensive that, to date anyway, concentrated solar using GaAs hasn’t been competitive. The only projects using it are funded by the companies (Emcore primarily) as demonstrator systems.
As the prices of pSi continues to fall, their price/performance benefits are falling further behind the competition. It seems highly unlikely that GaAs cells will be a major factor in the future.
Conventional pSi is around 12% efficient, and mono-Si around 16%. aSi and CdTe are both around 9%. That means that in order for a thin-film collector to product the same amount of energy as a conventional pSi cell, you’ll need more of it. Where does this break even? Given that both thin-film solutions are backed on a sheet of metal, it’s probably a loosing proposition compared to the blanket solution used on the Hubble. But my theorizing isn’t work a hill of beans, so I’ll simply note that the extremely competitive space power systems that already exist – the comsats – all use GaAs. There’s not one thin-film deployment in space.
The proposal doesn’t say they’ll use thin-film cells now, it says they might use them after some unspecified improvements make the price/performance ratio improve. But as you can see from the arguments above, this is extremely unlikely to happen in the foreseeable future. CdTe is limited by production concerns, and aSi has already pretty much maxed itself out after decades of development.
So, unconcentrated multi-junction GaAs remains the best price/performance solution for space applications, now and in the foreseeable future. As I pointed out in my earlier article, that performance is not even remotely good enough to consider commercial space based power.
I stand by my comments: space power in the short term is a pipe dream.