Fundamentals of Cadmium Telluride Solar Cells Text Version

This is a text version of the video Fundamentals of Cadmium Telluride Solar Cells, a lecture given as part of the Hands-On Photovoltaic Experience Workshop.

Matt Reese: So I'm talking here about the differences in modules between singulated technologies, and singulated are wafer-based, like silicon and gallium arsenide, and then monolithic architectures, which are the commercialized thin film. There are some examples of thin film technologies, like CIGS, which are done in a singulated manner, but the majority of the ones that are being done at low cost are taking advantage of certain things.

So there's pros and cons for both of them. One of the reasons why you might want to go singulated, other than growing great cells, is that you can sort your cells, bend them, and then put them all in the same module to kind of match performance. But this does entail more work, handling, and then you have failure points as you're putting all the things together.

So in the case of monolithic interconnect, you basically get your cell connections for free by going and being very careful, by having what we call a P1, P2, and P3 scribe. So by being clever in your processing steps as you're growing these layers, you'll grow some of the layers, then you'll scribe down, then you grow more, and then your last layer will connect your cells together. And that will then enable the majority of things to be connected up, where you get all of the connections except for into the junction box is automatic. So you still need to have soldered leaves that are going to go out to your cabling.

So that leads to a lower cost and fewer failure points, so – but a problem with this is that you need to make everything really uniform, and that's challenging, doing uniformity over large areas. So this has actually been one of the things that has been plaguing the CIGS industry.

Okay, so we published a paper a year or so ago, and the first draft I was told read like a manifesto. And if anybody ever tells you that you're writing something that's a manifesto, it's not just if you have a beard you shouldn't do it. But basically, I'm going to give you a condensed and maybe less scary version.

So basically, you guys may be familiar with an experience curve. So this is the average selling price for the entire PV industry. And you see with silicon that the cost of silicon is basically the same as the selling price, and this is why we say it's a no profit industry. It's kind of the who can bleed more as we're expanding out industry.

Now if you look at the CIGS and cadmium telluride, everything shifted over, and so you can go and interpret this data in a number of ways. One way that we chose to do this is that we integrated each one of these points, and so if you integrate your average selling price, then that's a way that you can envision how much it actually – or your cost. Then if you integrate your cost, then that's how much you had to spend to get the amount of learning to build things out. So by integrating all of the runs that you had to do to get to a gigawatt, that tells you what the investment needs to be.

So the minimum investment – so this doesn't include research and development, it doesn't include subsidies, it doesn't include things like the CMOS industry that may have dumped say $100 billion in R&D into things, or all of the scale-up for building their fabs. It also doesn't include for the thin film industry the leveraging of knowledge and how to put a module together. Right? So it's neglecting all of those things.

But if you look at this, the minimum investment to reach cumulative production of a gigawatt was $8 billion for silicon and about $1 billion each for CdTe and CIGS. But if you look at what it costs to get to a dollar a watt, you're talking $155 billion for silicon and then on the order of a couple billion or $0.4 billion for CIGS. So this is one of the reasons why the thin films has really been able to do things very differently.

So here's another way of kind of parsing the data. If you look at things via time, you can see that your curves are going up for the silicon, and there's – it's continuing about the same trend, where there's a very rapid rise for CdTe and CIGS, and then it kind of levels off, yet they're being able to bring their costs down. So the question is how are they doing that?

So silicon, it's not that there's no innovation in silicon, but they are certainly benefiting from having these economies of scale that are driving at least a portion of their savings, whereas the CdTe and CIGS communities are not benefiting in the same way from that. But if you look at the R&D expenditures, it really highlights some of the differences. So for solar, we know that over 90 percent of the PV community, commercial, is silicon. So First Solar is only a few gigawatts a year, yet they are spending more as a company than every other silicon company. So even Sun Power. So Sun Sower is kind of a small but like high end, and they are going and being able to maintain their lead by having this R&D budget, where all of the people who are going and they're low cost silicon, they're not doing that.

So this is explaining how the thin film community is moving, is they're actually going and being innovative and researching.

So here's one other way of looking at it. The cost of First Solar, which is a CdTe company, is actually lower than silicon. The efficiencies of average modules is now just starting to pass multicrystalline silicon, and they're starting to close that gap with single crystal.

So there's also energy payback times. If you look at energy payback times, it's generally being significantly lower for thin films. So this was something that was a number of years ago. I don't know that this is actually the reduction that you would see, but just assuming that there's the same amount of energy that would go into these to make the fleet average of 17.7 percent, that would be a 38 percent reduction in energy payback time. So it's just much less energy intensive, which is one of the reasons why the cost is lower.

Okay, so I'm going to now do about ten slides or so where it's kind of the history of CdTe, a lot of the innovations that have been made over the years, and I'm borrowing heavily from some slides that Colin Wolden at the Colorado School of Mines shared with me.

So this is outlining the efficiencies, the record efficiencies in CdTe, over time. So the first thin film solar – CdTe solar cell was a cad-sulfide CdTe solar cell. It was invented by Dieter Bonnet and Ravenhorst in the early seventies. So it's using the standard thin film structure, where we're growing on glass, we're using a transparent conductor, so in – it's been tin oxide is the dominant one, and then the emitter layer is cad-sulfide, the absorber is CdTe, and then there's a back contact.

So this structure started out at a relatively low efficiency, and it really – this early work kind of identified a lot of the challenges that the community was working on. Kind of how do you dope it well, how do you do – is it good to have an abrupt junction at the emitter, or do you want it to be graded? And then like this is a polycrystalline technology, so dealing with the grain boundaries, are they good, are they bad? And then making an ohmic contact to the device is actually very challenging. The work function is about 5.9 EV or so. Does anybody have a sense of – who can tell me what the work function of gold is? Anybody? Yeah?

[Background voices]

Matt Reese: No, it's 5.1. Yeah. So there are not a whole lot of metals – so gold is typically used as a high work function metal. It's one of the highest. And there's obviously reasons why we don't want to use that in manufacturing. It's very, very difficult. So you want something that's on the order of 5.9, 6 EV or so. It's very challenging to contact this. So that is a problem that I'll talk about a little bit more.

Okay, so the first wave of improvements. These basically were highlighting the importance of impurities. So we started introducing copper, chlorine, and oxygen, so adding some copper resulted in doping the devices somewhat and improving the carrier concentration. This is a very defective material. I'll talk about that a little bit more. And that was a way of improving efficiency. And then chlorine was another defect that was added which has some passivating properties. And then oxygen also has some passivating aspects to it.

So the second wave was really the introduction of this cadmium chloride stuff. This was introduced by Ametek, so Peter Meyers was integral in that. And so what you typically do is you throw your CdTe as fast as you can at the glass, and then you end up with this very defective structure, and then you try fixing it. And that's really what the cad-chloride is going to be doing.

So there are different ways in which you can do it. You can dip it in a solution of cad-chloride and then heat it up. You can just evaporate a thin layer and then heat it up. Or you can just put it in an ambience where it's hot. But like you typically – your growths of the CdTe might on the order of 500 to 600 Celsius. The annealing conditions are at least 400 C for cad-chloride.

Now the question is what is it doing? So it results in this recrystallization and grain growth. If you look at CdTeluride and you wanted to – so if you look at the phase diagram here, you have to go to over 1,000 degrees C to start going and kind of melting things. However, if you introduce cad-chloride, you can form a eutectic, which drops that significantly. And if you inject oxygen into it, it actually drops it a little bit further.

So going and doing this then allows you to basically dissolve those bad grains and make them better. So here's some EBSD, which Joe has a pet peeve about people who talk about SEM images, and then will just say that's a grain. One of the ways you can get around that is you can do electron backscatter diffraction, where this actually tells me what the facets are, so I do know these are grains.

So here would be as deposited material. These are grown on different – cad-sulfide or MZO, and then by going and doing that cad-chloride treatment, you can start seeing that it grows as you increase the temperature. So this is a way of kind of getting around – so it's not that we made a single crystal afterwards, but we repaired a lot of the defects.

Something else that happens, it doesn't just do one thing, is there's some passivating properties. Here is some work a while back where you can see devices that – this is as grown, and then cad-chloride treated, and you can see that in the as grown cases, that the lifetime is very low. So this is less than 100 picoseconds for the as grown, and then we can go an increase it by an order of magnitude by doing the cad-chloride process and get ourselves up into the nanosecond range.

And so there's clearly passivation, and then you see – so that's changing the lifetime. So this is minority carrier lifetime as measured by time resolve photoluminescence, or TRPL, and that it correlates with the increases in voltage. Okay? So I can clearly say that there's a passivating result for that lifetime.

So the chlorine is then segregating out at the grain boundaries, and so this is a compositional image, and then this is a photoluminescence image. So I don't know how well you guys can make it out, but there are dark lines in between the grain boundaries on this one here, whereas this is after cad-chloride, and they kind of get washed out. So then you can see that the grain boundaries are not recombining as much.

So there's a little bit more to this second wave refinement than just introducing cad-chloride. There is also some refinement on the back contact, getting the copper conditions just right, because making that contact, if you do a graded back contact, can help. And there was also going and playing with the front contact, so ways of improving the optical collection. So people are thinning the cad-sulfide layer, using alternate TCOs that might be higher quality, better glass, etcetera. So this kind of previews the efforts to enhance current.

And then there was about a decade where you don't really see the efficiency move. However, in this decade you started actually seeing companies decide to step in. So there are actually a number of companies that were working on it here, and they start actually selling products in this decade.

There are a variety of ways in which one can deposit this material. There are companies that – some of them are still around even – that – and these – they're underneath the different deposition techniques that they're using for the absorber. One of the things that you can do with cadmium telluride, if you're doing a vapor phase, is it has something called congruent sublimation. So if you heat it up, both species come out at the same rate, which is actually not the case for a lot of materials. So that actually makes it a lot easier for you to keep your stoichiometry about where you want.

You can play games a little bit in the stoichiometry by changing your temperature, where you may shift yourselves a little bit more towards the cadmium side or the tellurium side, but in general, you're going to maintain the stoichiometry that you want.

With all of these, though, so they're chemical and vapor phase approaches. The one that for – solar is using vapor transportation deposition method, in which you have a solid source, you heat it up, and you flow a carrier gas through it. And that has been proven to be highly scalable, and – but if you look back around the year 2000, there were several companies that were competing, and at the time, First Solar was not a clear leader, and the module efficiencies on these were relatively low.

But they stuck with it, and there are various reasons why companies may not have made it, and we don't really have time to go into that. So if we kind of fast forward to today, this is a picture of a First Solar factory, and they can go from glass to panel in 3.5 hours, and they can make 18 percent efficient modules. So they are a multi-gigawatt scale company, and they are ramping up to about seven gigawatts per year, where I think somewhere in the 2020 timeframe, or maybe it's a little bit past, they're planning on hitting that. And they're doing some of the manufacturing in the United States. They're building out an over a gigawatt factory in the Ohio area.

Okay, so third wave. This is around where I start in the CdTe group, as Joe mentioned. I can't really take credit for all of these. You may notice they're companies. But clearly, there might be a correlation.

Now there's kind of the exploration of new architectures, and so Marcus Gloeckler at First Solar highlighted that there was an enormous amount of work to get from here to here. Over 70 million current voltage curves and 40,000 samples analyzed via much more complicated characterization to try understanding what's going on. And the result of these can largely be described in three big changes.

The first one is the introduction of this alternate back contact, where instead of just putting metal directly at the back, there's the insertion of a zinc-telluride layer. As I mentioned before, the work function of cadmium telluride is very high, and the cadmium telluride itself is lightly doped. So by going and putting zinc tel in as opposed to pulling things down and having this loss here, you can actually band match with the zinc-telluride, and you have very small – so there's still a little bit of a barrier. This is not a perfect contact. But it is very good, and you can really minimize that loss, and there's a clear increase in voltage. And these are actually module efficiencies as shown by First Solar.

And one of the other things that this results in is improving the stability. So stability in cadmium telluride is related to this copper defect, and I'm going to talk about that a little bit more in a moment, where you want just a little bit, but it can move around in the device, and then it can accumulate in different places where you don't want it, can end up in the wrong place, in the lattice, and it moves really fast. So there's problems with light stressing and temperature. But moving to the zinc-telluride contact has actually improved both of those, and you can kind of almost think of it as a sponge.

Okay, so entitlements. People know what I'm talking about with entitlements? Joe? Okay. I was thinking millennials, but fine. So there's this detailed balance limit, oftentimes referred to as the Shockley-Queisser limit. For cadmium telluride, the band gap is about 1.5 EV, so we really should hypothetically be able to get over 31 percent efficiency. And I'm going to just mention briefly that what I'm calling cad cell tel could have a band gap. When you start alloying it, you can drop it say 1.4 EV-ish, and if you were to do that, now you've improved your efficiency to 32.9. And we've shown these different graphs, but you can kind of break this down, not just into efficiency, yeah, like there's a ideal band gap and such, but you can start thinking, well, if we have our current and our voltage, so the current is relatively easy to understand. Was this covered already in silicon and three fives? Yes? No? You pick this out, 1.5 EV, you throw away about half of your photons, and that works out pretty well.

Then there's not actually a good scientific reason why you shouldn't be getting almost all of your current out, so in the detailed balance assumptions. There's practical reasons why you might not be able to, but scientifically, you should be getting just about everything out, right?

Okay, so when we go and we look at what cadmium telluride has been doing historically in this third wave, we're starting out down here in the 25.1 per – _____ per square centimeter or so, and we're seeing it go up, and then we pass the detailed balance limit for cadmium telluride of 1.5 EV. So clearly, there's something going on here, right? These were certified measurements. And we're getting up much closer to this 1.4 EV.

So if we look at the quantum efficiency, so this black curve is a cadmium sulfide CdTeluride device. Its efficiency is about 16.4 percent certified. And you can see where its band gap is here. YOU can see that there's a lot of loss associated – so with cadmium sulfide, we have what we call parasitic absorption, so it will actually absorb light, and you think, well, it's a semiconductor. Who cares? But it turns out that it's fine as an N type conductor, not so good as a P type. You get the wrong carriers in it, you lose them.

We're losing current in that, so we actually want to thin it down as much as possible, remove it, if we can, and then you can also see kind of fringe patterns from the TCO that we grow on this device.

Okay, so in this red curve, you see the elimination of cad-sulfide in the device. So it could be something like – so this particular one is MZO, so there are other strategies that could be used, but now this does not have parasitic absorption. You've improved your current. But you have not actually improved it past this detailed balance limit.

In this blue curve, you can see that the band gap is now different. We're absorbing light out and collecting more photons over in this region, and that's because there's an alloy now. So a cadmium telluride device is no longer just cadmium telluride. There's some selenium at the front that's been alloyed to get some more current out of it. It has some other things as well going on. But if we assume this detailed balance limit of 1.4 EV, which maybe it's 1.38, I don't know exactly, we're getting almost 95 percent of the current entitlement in the devices, which is pretty darned good.

So the fact that we're at 22 percent out of this 32, it's not current voltage. There's this loss that you have with voltage, so it's – VOC is really describing the quasi-Fermi level splitting. There's limits theoretically of what you can do based off your band gap, because there's a certain amount of charges that you're going to be injected, based off of the intensity of your light, what you can actually photo-dope, right? So that's going to limiting what that voltage loss you can get – just how close you can push things to the band edges.

But clearly, we should be able to get fairly high voltages. So now let's look at where we are entitlement-wise.

It looks like things weren't really moving very much. There's a little bit of an uptick, but it doesn't look very impressive at all. Because the detailed balance limit for cadmium telluride – remember, that's 1.5 EV, so it's this higher line, because we inverted between current and voltage. But we know that these last devices have selenium in them, right? So while there is an uptick, we've dropped our detailed balance limit, so it's actually a lot more impressive than it appears, the way this is plotted right now.

But let's now talk about why we have this historic gap, because this is the major problem that we have in cadmium telluride, is getting the voltage out of our devices. And it's associated with our minority carrier lifetime, which is in the low nanosecond range, our doping, so we're not actually intrinsic enough to be an NIP, because we can't fully deplete, so we need to actually put some things in, and then we have a recombination at our interfaces.

Okay, this – so these rectangles – I don't know how well you guys can see them, they're gray – are highlighting where we live today in terms of the hole density. So we're kind of in maybe the – we're 10 to the 14, mid 10 to the 14 range, and the lifetime is generally 10 nanoseconds or less.

And if you look at – so then this column here is a low front interface recombination velocity, and this is a high one _____ 10 to the 5th. And if you look at these, the difference between them is very, very small. It doesn't really matter what your interface recombination is, as long as you are low carrier concentration and low lifetime. But as soon as you start stepping outside of that box, you start seeing differences. This is what we've been trying to do, is to get ourselves outside of that box.

Several years ago, we were trying to figure out, well, what is the source of our short lifetimes? Is it because we have point defects, or is it extended defects that's our problem? So we did a series of correlated microscopy where we looked – so we'd fib a sample, and then we would look at that marker, and then we would do EBSD, so that we could actually know which facet, which type of – so this is the grain interior, and then we would have different coincident site lattice grain boundaries. So is everybody familiar with what a sigma III versus sigma V or whatever is? So I see some yes, some no.

You can think of it as imagine you have this nice perfect crystal right next to another one, but they don't quite line up. So it's describing the angle. And a sigma III would be every third atom is going to line up again, and the sigma V would be lining up every five, etcetera. Okay? And those would be called coincident site lattice. And then you also have non-coincident site lattice. Those would be what a lot of times people think of as grain boundaries, where you just have like this boulder that was grown here, and a boulder that was grown here, and they don't have to actually line up. So those would be another type of grain boundary.

And we compared this using cathodoluminescence, so the CL, and you see the grain interior, so that's what you're normalizing everything. And you see the sigma III, which is a fairly shallow one, is a relatively benign fault. But then the other types are actually not necessarily so good. But cadmium chloride seems to passivate them at least somewhat.

We also have lifetime data, so this is TRPL data, versus grain size, and there are various treatments and on different substrates, so – but you can see a general trend. As you increase your grain size, the lifetime goes up. That's indicative that grain boundaries are probably not the best thing, right? But maybe there are certain ones that aren't so bad, and maybe you can passivate it.

There have been various people who have argued what the grain boundaries actually look like. Some people are putting forward that it's a downward band bending, some that it's upward, some that there's a barrier. I'm not going to go ahead and get into that argument. Some people think that they're actually beneficial, but there's certainly data that would argue that maybe it's not.

Here is what we're talking about in a way – a general way that you might think about defect tolerance. You can think about the ionicity of a particular type of compound as you go from being primarily covalent bonds to primarily ionic bonds. If we look at these point defects, so this is just how the lattice is disturbed, we've had different theories along the way. And there was something back in the early 2000s where they used the basis and said, hey, cadmium rich is really bad. You should avoid that. Stay in the tellurium rich. And then we went ahead and were doing experiments and they kind of readjusted their basis, said, actually, it was backwards.

That's what these arrows are. With the new instruction from DFT, we'd say that you have a fair number of defects that can happen in the tellurium rich, but cadmium rich seems to not have them as much.

That's just the intrinsic cadmium telluride semiconductor, but then we start going and trying to fix things by putting chlorine, copper, and then we're not even going to get into oxygen. You have all of these defects are possible as well. Notice a bunch of them are in the middle here, and that's not good.

What we did to try to take a step back for the defect chemistry, so what we've been doing for 40 years or so was doing this copper-based defect chemistry to dope things, and you'd have chlorine and oxygen to try to passivate stuff, and there's kind of a limit, because you end up with a compensating defect chemistry. Does everybody know what I mean when I say compensating? I see some nodding heads, and I see some – okay. No.

So what that means is if you put something in, it can – you want it to be a dopant, and you want it to sit in a particular place in the lattice, and that's going to be good. But there's maybe another place – so in the place of copper, it has more than one oxidation state. That other – those other – it's not just one other oxidation state. Those other oxidation states can allow it to sit in other places in the lattice, which are not necessarily good.

In the compensating defect chemistry, you may put it in to do one thing, and you want it to say be an acceptor, but there are other ways in which it can sit where it's a donor, and they cancel it out. Okay? Then you basically just put defects in, and you didn't really get much out of those defects, and this is kind of why we're stuck in that 10 to the 14 range.

By going and using single crystals of cadmium telluride, we dope them using either a tellurium rich defect chemistry, which then kind of can encourage things to sit on a particular site, or cadmium rich, and then as grown, and you can see the lifetime of these. And so notice that the lifetime is a little bit lower in the case of say copper, but there are other defect chemistries that have a significantly higher lifetime. So that tells us that there's hope. Here's an example of that copper compared to some of those others.

Now one of the other things that you can do is you can look at the carrier concentration, right? So just how effective were you able to put things in? And notice we were able to actually get the dopants up into this mid 10 to the 16 region when we're in the copper, even though it's a lifetime – it's not great for a lifetime perspective. You can actually get decent ones with this phosphorous, or group five doped lattice, but if you go and you just leave it out, like after having put it into lattice and activated it, you just do a shelf life study, or if you just put it in temperature, you start seeing that the copper quickly starts compensating. It goes into another place in the lattice, even in a single crystal. So it's not just the fact that it's a polycrystalline system. This is an inherent problem with copper in cadmium telluride, whereas the group five dopant in this case is actually very stable over time and temperature.

This shows you just some of the different activation levels. So SIMS is telling you how much of something is compositionally present, where Hall is the electrically active. So in copper, you can – so these are initial – but this shows you what you would more typically get in polycrystalline. So you have maybe one out of a thousand, one out of ten thousand atoms that is where you want it to be, so it's not a very great dopant, whereas if you start going to these group five dopants, you can actually get very high activation even in polycrystalline systems.

Going and looking at this activation – and activation, by the way, I used that word, I don't know if I defined it, is you have something physically present, so you count the number of atoms, and then you say how many of those are electrically active. So if half of them are electrically active, then you are 50 percent activated. Okay? S you want high activation.

So we were able to demonstrate that we can do this in the case of phosphorous, and you can still get pretty good lifetimes out. And when you compare this kind of whole density versus lifetime, as you go, you don't dope something more, you do lose lifetime, so there is a limit to how much you would actually want to put into something from a lifetime limiting perspective. But we're kind of on the same level as gallium arsenide in this case, which has a very similar – it's indirect, similar band gap and what not, and people think of as a very high quality material.

So once we took these single crystals and then made them into devices with a polycrystalline interface and standard back contacts, we were able to improve our voltage in cadmium telluride to above a volt, where historically we'd been kind of in that 850 millivolt, which is then supporting what we've been saying.

Now if we want to put all of this together, in the case of polycrystalline materials, we need to get a lifetime increase, we need to get a carrier concentration increase, and we also need to reduce the front interface recombination, which – and if you don't do all three of those, you can get some boost, but you will never get all of your voltage possibility.

So you can reduce the carrier velocity, the recombination velocity, but you can also just limit how sensitive you are to it by going and playing with your interfaces. I've only got a couple of slides more.

By going – normally in the cad-sulfide device, you have what's termed a cliff, where you could instead put in a little spike to kind of push things away from recombining right where you don't want them to recombine. And by going and playing games with this, even with a relatively high recombination velocity, you can then start getting out into that 25 percent range and higher voltages.

So these are kind of some of the things that people have demonstrated, but we're trying to put all of those pieces together to kind of move past that 22 percent into kind of the region that we have not lived. So there have been – not just in single crystals. We've now been starting to demonstrate things that have higher lifetimes in both cadmium telluride, and it turns out that selenium actually has a passivating role, so you can get to higher, because a lot of those tellurium defects can now be replaced with selenium, and it moves it out.

We've been able to get to reasonable doping, and sometimes we can get decent activation in polycrystalline material, and we're working on strategies to improve the interface recombination. So really, voltage is our major problem, but I figure that I should mention tellurium scarcity, because this is something that people sometimes throw at you. It's like, well, there's just not enough tellurium.

And in general, if someone tells me that you've got a scarcity problem, I figure that they probably have something that's not as good that they're trying to go and sell, and they're going to try blaming that. But basically, tellurium comes as a byproduct of copper, in general, because no one really wants to mine it. But there are a couple of mines which are just tellurium mines in the world, but people are not going and mining waste dumps, and no one's trying to generally mine things at the moment. Getting actually estimates of how much tellurium – like if you pull up the USGS reports, they admit that there's just limited data on it because not that many people care. First Solar is using about half of the global production, which is one of the things that they have to be very careful about, is if you're using half of the world's production and you want to double your production, and you don't want to drive the prices up, you need to start managing that supply chain and asking people to start mining more of it.

So they are being very careful, and they are ramping up from kind of the two gigawatt, and I don't know where they are, maybe three or four now, and they're aiming for seven, and they have this figured out. So if you want to get to a terawatt, that would require about 100,000 metric tons. There's estimated reserves, which means people absolutely know that it's present, of about 25,000, assuming that you're losing 50,000 percent of it. But there's a lot of admission that people – like in Latin America, they are saying that there's like almost none, even though there's major copper mines, just because the stuff isn't publicly available. So there's likely a lot more than these low numbers.

Okay. Questions?

[Applause]

Matt Reese: Yes?

Audience: Can you talk about like end of life cycle of tellurium, or cadmium?

Matt Reese: Sure. So early on, First Solar realized that people were concerned about this, so they ended up setting up a program where every time they sold a module, they said, when you're done with it, give it back to us. We will recycle it. We will take care of all of that. And this is actually one of the things that they're like, oh, we now control all of that supply, that we can go and make it into devices again. Yeah, it's actually viewed as a good thing. Does that make sense?

Audience: Sure. Yeah.

Matt Reese: Yeah.

Audience: I was wondering if you knew like more like what – how easy it is to recover, like how much of it you get back, and –

Matt Reese: Oh, well, so the way we make the modules is you're growing on glass, and then there's an encapsulant on the back, and then there's glass again. So you basically can – if you can rip apart your glass, for instance, by ashing – so like a way of doing it – I'm not telling you that this is the way they are doing it – but a way is all you need to do is heat things up sufficiently that you're going to burn off the polymer, and then you have easy access to your device. And then it can be dissolved in acid. So it's not – there are potentially better ways of doing it, but conceptually, it's not difficult to do.

[End of Audio]