Friday, September 9, 2011

Solar Industry Dynamics and Solyndra

The solar cell company Solyndra failed last week, taking with it a $535 million DOE loan guarantee and triggering a press blitz in which every single anti-government pundit attacked solar energy as something that "doesn't work" and putting up Solyndra as everything that's wrong with government intervention. There's a much more subtle picture going on here, however, and the media storm glosses over many of the aspects that made Solyndra's failure inevitable well before the current slump in the solar industry hit.

If you want to forecast the long-term trends in the solar industry, then it always helps to have the end in mind. The end that the world solar industry wants is to drive costs of solar panels down to a level that make them competitive with conventional fossil fuel sources. That's an uphill battle, not least because some fossil fuels (read: natural gas) are becoming much cheaper.

One might expect costs to be driven by economies of scale in the long run - indeed, that's a very important aspect, and as early as 2004 I saw multiple insiders in the business comment that solar startups were particularly risky, as the industry would have to inevitably consolidate. To a large extent, that is still true. However, rather than economies of scale, what's been driving the dynamics of this industry for the last 8 years or so has actually been the cost and supply of raw materials for solar panels.

These days, the cost of the raw materials for a solar panel constitute maybe 35% of the installed cost of an on-grid solar panel home system. Not so long ago, this was closer to 65% (and it was a lot more in absolute terms, too). The raw materials in question are high-purity, solar-grade silicon and some rare earths. The remainder are casings, glass, switches, alternators, and a metering retrofit to allow selling to the grid - many of which have their own supply bottlenecks.

Fundamentally, Solyndra went bust because of raw material supply. Solar PV manufacturers are chained to the price and supply of very high purity silicon. Metallic silicon is refined from essentially sand using a ridiculously energetic arc furnace in the presence of carbon. What results is technical grade silicon, which is pretty much useless except as a carbon capture ingredient in certain specialized types of steel plants. In order for it to be used in solar panels, it must be upgraded to semiconductor-grade silicon, which requires another refining step called the Siemens process. The Siemens process is notable in that the crystal of silicon formed is very nonuniform. The grade is called polycrystalline silicon, or polysilicon, and in lieu of trying to inadequately describe it I'll just show you a picture:
Essentially, it looks like ice that's frozen quickly, or as if a single pane of glass has shattered. By the nature of the process, polycrystalline silicon doesn't form into a single crystal. This has implications for its behavior as a semiconductor.

Polysilicon is used for two things: semiconductors and solar panels. These days, solar takes up more than 50% of global polysilicon supply. However, as I mentioned earlier, using polysilicon directly reduces the output of a solar panel. Crystal alignment ensures a good current. Getting crystals to align requires a long, drawn out refining process called the Czochralski process. This produces a better panel, but costs a lot more. A key to solar economics is which you choose to use, and why. The monocrystalline panel is essentially the solar panel you know from space stations, where performance considerations are paramount:

It is this distinction that is key to why Solyndra failed. Solyndra's technology depended on monocrystalline silicon. When it got that loan guarantee in 2009, the supplies of semiconductor-grade polysilicon were ridiculously tight. Suppliers were used to feeding only the semiconductor industry and were completely unprepared for the new demand from solar panel manufacturers. As a result, polysilicon prices were very high and the relative price of monocrystalline silicon - essentially, the price of polysilicon plus the cost of the Czochralski process - was low compared to polysilicon. Solyndra's competitiveness depended on polysilicon supplies continuing to be tight. That was stupid. Why the USDOE didn't catch that assumption, I'm not sure.

Polysilicon supply capacity increased massively around 2009, but prices were buoyed by high demand; following termination of renewable energy subsidies in Germany and in many other places, prices went way down and the relative price picture for Solyndra looked a lot worse. Even so, it had an advantage in terms of superior products. Ultimately, why Solyndra died was that polysilicon manufacturers managed to cut costs like crazy. This has less to do with general competitive pressure in the wake of the panic of 2008 than it has to do with China. Chinese companies have been literally given free land, no taxes, and guaranteed market positions by government intervention. Rather than the exception, free loans are the norm. So Chinese companies are able to cut costs like crazy and have very little equity in their businesses but still get guaranteed markets.

Ultimately, all that this extended explanation goes to show is that the issue is very much not black and white. And of course, aside from all of the technology, there's still the possibility of just straight bad management. Some evidence has arisen that the USDOE's due diligence people were duped by Solyndra, hence the recent raid based on misuse of Federal funds.

It also raises the interesting controversy around government support for infant industries. America really has not engaged in wholesale industrial policy since the late 1940s, and in a world where advantages in productivity, infrastructure and technology made any other country's industrial policy an insignificant blip, this was fine. These days, however, industrial policy is making bigger waves. It seems clear to me that China is trying to do some combination of cutthroat competition and clustering externality capture with its rampant subsidies, and that this seriously challenges the Bhagwati-ite view that protectionism in any form will shoot you in the foot.

The obvious solution in my eyes is to immediately refer Chinese support of solar manufacturers to the WTO, as some Americans have with wind power. The problem with this is that (a) this doesn't stop Chinese manufacturers from having access to one of the largest growing markets for solar energy and (b) antidumping tariffs imposed by the US wouldn't be of much use in a global market unless other countries slap on the tariffs as well. For two nations with companies competing to be global scale, the conventional WTO enforcement mechanism just won't work. And I have a realistic idea of how much America is willing to invest in government subsidies for the renewable industry - certainly not the $1 trillion over 10 years that China is giving them - so that's out.

My fear at this point is that when Chinese subsidies run out, it will already have captured the positive clustering externalities from solar manufacturing, by which time hopes for an American solar industry will be dead.


  1. This is a terrific post. Thanks.

    - M.S.

  2. I followed here from M.S.'s post at the Economist and of course, your link Josh/Ah Beng. I appreciate your effort here, though it is slightly depressing. Maybe it is more than slightly depressing.

  3. I don't buy that it was a 2-year silicon supply blip glitch; and there are more Strategic Materials regulations available where these ones are making WTO transparency like getting an eyeful of awful. On the other hand I don't know duping the DOE lately; reckoned it was an internal magazine.

    Semiconductistas knew before 1994 that (in a power co. /refinery furnace-sharing enterprise, in the PRC by default) one could casually seed polysilicon with a desired orientation and arrange for the impurities at crystal edges to not conduct (thus not kill the chance of drawing out photo-current.)

    A glance at the technology makes it seem fragile and low-duty; who's gonna clean a rack of tubes, how do they not pop often, orientation-agnosticism seems a folly; but they looked good performing on the panel market, until they didn't or those ads by natural gas made carbon emissions look fine.

  4. Not certain I understand everything about your comment... but I'll give it a go.

    The two-year supply glitch is still ongoing, so to speak, and it's not going away. Right now, there is currently a large oversupply in the market of the commoditized parts of a solar cell - that is, the module, cell, and the wafers. The impact on the pricing especially of the glut in modules available is very significant. Recall that Solyndra, like most startups, was hemorrhaging cash like crazy and had no alternative source of revenue. It began marketing a product in a severely depressed market environment and its relative pricing picture died because of its choices in module construction. No mystery there.

    The issue with silicon refining was not the conduction. Seeding is necessary for both the Siemens process and Czochralski process but the two are fundamentally different. The Siemens process involves a redox reaction at the seed crystal boundary, which grows the crystal nicely but isn't going to create an ingot that you need. What you've referred to as "casually seed[ing]" to get a single crystal in the Czochralski process is instead far less easy. The production of a monocrystalline ingot requires that the seed crystal be spun, at a constant shear stress - obviously, requiring heavy process control - and requires the melt be within a small temperature range around the melting point of the silicon. Completing a single cylindrical silicon crystal ingot takes between 36 and 40 hours as you slowly draw it out of the melt. This is not a casual process. It's been much improved since the '90s but it's still difficult.

    I'm still not sure what you mean by a "rack of tubes," but there exist voltage gates that prevent irregular current from blocked, shaded or otherwise low-outputting solar cells from damaging the others, but that aspect comes in on the design of the individual panel.