Silicon PV Text Version

This is a text version of the video Silicon PV, a lecture given as part of the Hands-On Photovoltaic Experience Workshop.

>>Paul: Okay, today we give an overview of cell technologies, the three of us. Me, Miles Steiner, and Joe Berry. We'll cover silicon, III-Vs, and perovskites. We start with silicon. To make these lectures unified, we will – I will give a little bit of maybe more introductory, maybe four slides, five slides, initially, to kind of establish a common ground and, also, to show what, for the PVs, the difference between direct and indirect gap semiconductor, what it actually means.

This is all plan of my talk. So, I'll talk about some fundamental boundaries to photovoltaic conversion, efficiency, and how direct and indirect the semiconductors behave differently, and how we can actually get this super high performance but actually for other different reasons. Specifically, a case of silicon and gallium arsenide is that like a namesake maybe name plate direct and indirect semiconductors. Then I will talk about what is unique about silicon PV. I will basically make a case that silicon PV has established its dominating power in the PV industry because of abundance, knowledge base, and technology, and, actually, indirect gap, as strangely as it might seem.

Then, I will talk about how silicon PV is done as a technology. So, from wafers, I will talk about this upcoming mainstream PV that was actually established very recently and is taking over the silicon PV industry, silicon p-PERC. And then how recent or some of my colleagues also think how silicon PV will evolve beyond this p-PERC technology into the future. Also, in conjunction with this about science problems in silicon PV, these three whales, so to speak – bulk, passivated contacts, and metallization – seem to be maybe the dominant future directions for advanced silicon PV.

So, let's start with a simple question, which is actually not simple at all. How do we get electricity from sun's photons using a semiconductor? When I prepared this presentation, I really – it reminded me of many negotiations, I'm sure, many of you have had with some kind of salesman. You kind of go and you say, "Oh, I need to do this." Then they say, "Well, we have this great product, this spectrum." But there is this print and there's final print and final print. At the end, you've basically got quite little for it big price, because we're trying to advance.

So, okay, I say, "Okay, here is the spectrum. I want to get energy in form of electricity out of it. So, let's use a semiconductor. It has a of forbidden gap. So, any photon that is above the gap will be absorbed. It will produce electron pull. There will be some relaxation process and then I will get electron at the conduction. Vantage and hold at the _____ vantage. Then I can use this energy in external circuit. I will send them out in external circuit to perform work. That can power J times V. So, everything seems fine."

So, immediately, you can see, okay, the first price I have to pay, that there's a gap. So, here, in this graph, I plot actually the power density, integrated power density starting from high energy integrated from here and up, to save my gap. Right? So, you can see that the very highest lower. Then, it's increasing and will integrate all photon energy. The photons I get are this [doors opening and closing] _____ _____ that was curb either per 100 milliwatts per square centimeters, are very real. It's all about PVs. That makes PV so efficient.

So, this is what I could get, the energy I could get. But because I'm using some gap, I'm losing some of the photons. For example, I use silicon. Here, I have 1.4 an easy gap. So, immediately, 19 percent energy is lost because they just go through and there's no way I can use. Then, there's another 33 percent that is lost.

How they are lost? They are lost by this thermalization. The way is done so you can easily calculate it. So, actually, this curve, this black curve is a current. So, if you'll just count – for every photon, you assume you generate one electron-hole pair, which may not be the case for some advanced cell concepts that you guys will develop the next ten years. But right now, this is what happens in a normal solar cell. One photon, one electron-hole pair comes off.

So, if I take this and then I count these pairs and actually integrate them, and I can get current. This is also, you've probably discovered, you know very well. So, again, similar as energy goes up. For silicon, this current is about 45 milliamps per square centimeter. If I integrate it to the right end, I get something maybe closer to 67 milliamps per square centimeter. If I have smaller, wider gap semiconductor – for example, gallium arsenide – then I get less current. Maybe closer to like 31, 32 milliamps _____ _____.

So, now I have this concept. Okay, I have energy, some energy that I put in. Here's my current. I just take this current and multiply it by the gap voltage. I kind of lamely assume, at first, that I can get voltage that is equal to the gap, just multiply, right. This is the curve you get. This is when you multiply current by the gap. You can see that when the gap becomes very small, even though the current is large, but because you're multiplying by such a small voltage, it's a curve of the maximal. So, you can see how much I lose. I lose 19 percent by not absorbing 33 percent by thermalizing. That's only even I just assumed that I can get energy that is equal to the voltage of the gap. That's also not true, actually.

But here we are. Okay, so let's go further. This is now print becomes finer and finer. [Laughter] So, this is my mega picture. Okay, here I have my gap. Here is my battery, so to speak. This is my voltage. But actually, you can never get cell voltage that is equal to Eg. What you actually do, you get voltage that is equal to separation of these quasi Fermi levels in the cell. That's always less than the gap. Not only that, but it also depends on the illumination intensity. For once on its want, if you concentrate, you can get more voltage, but you only can approach the gap, but you can never reach it.

So, here is the formula. So, voltage at open circuit is equal to this gap minus this. You can see that there is a product of electrons and holes in this formula which means that the more electrons and holes you have, the closer your voltage is to the gap. So, in other words, if you want to make cell more and more efficient, the first thing you have to do is somehow increase that concentration of electrons and holes. Of course, that's easily done by, say, making cell defect free and things like that. You can also see from this picture – this is actually interesting article here to read – that in the cell you can kind of this voltage, but actually your contacts will make it less perfect then actual Voc that you will get out will be actually less than this in flight Voc, the middle of the cell.

We can actually calculate how Voc depends on the gap if you have 1-sun. This is done following formulas that was developed first by Shockley-Queisser. I encourage you to read the article. Basically, I didn't try it here but in the diode decoration, the diode trifactor is assumed to be limited by some black footed radiation absorption processes. But you do that, this is like a fundamental limit. You forget and see how close Voc is to the gap at 1-sun. It's not actually the gap. As a rule of thumb, it's separated roughly by, say, open three, open four voles from this line. So, that's what we can get with 1-sun. If I concentrate it – and this, Miles will talk about that in the future lecture – you can see that actually can get better Voc and the efficiency becomes higher. But at 1-sun, this is what you get.

Okay, so in other words, here, the Voc depends on Eg. Right? It depends on absorbed photon flux through this np and it depends, also, on lifetime of photocarriers, again, through this np. If you have more carriers in the cell, you have higher voltage. That's no doubt about that. If you have no defects in concentration, that photo is Voc.

So, now, okay, so, I have already said, "Okay, fine. I never get the Eg voltage, so at least can I now operate my cell?" So, now there is even final print here. Actually not. It turns out that you have to still pay this tax for fill factor, essentially. That reduces the cell efficiency even more. Basically, think about this. So, voltage depends on credit concentrations, right? So, you build up credit concentrations at the open circuit voltage, everything that comes in eventually convert the _____ _____ _____ _____. Now, I have my external circuit and start drain the carriers to the external circuits. Okay, the intensity goes down, so voltage goes down. So, you cannot have operating cell with the same voltage as open circuit voltage. Simply not possible.

It turns out if you neglect any resistance, losses, and so on, there's a fundamental limit, again. This fundamental limit – I mean this is empirical formula is kind of like a transcendental equation you have to solve. There's no analytic solution for that. But this formula describes that dependence out for significant digits. What you can you see is that this fill factor, which is the power you get out of the cell versus the power you expect to have, which is the product of Jsc and Voc, depends on open circuit voltage only. Actually, open circuit voltage divided kT. That's the only parameter it depends on. This completely fundamental relationship. You can make cells whatever. It will always be there, this relationship. You can see, the smaller the voltage, the more you lose. So, for example, if you have higher band of cells, you are talking about fill factors post 0.9. In silicon, this might be open 85. The lower the gap, the less and less it becomes.

Okay, so now, let's put everything together. So, here is my electrical power that I expect, which is my gap times the current. From there, I lose into – open circuit voltage is here. Then the fill factor and multiplied by these two ratios. This what I actually get. So, my previous purple curve now converts to this green curve. So, essentially, these are spectral voices. These are _____ _____ photon. These are thermalization to the gap. These are these electrical things that happen. So, in other words, chemical potential difference versus the gap energy difference and the fill factor. So, those two things push my curve down. So, this is this typical. This curve is maybe not precisely to the limit that the Shockley-Queisser calculated. But it's close and you can see there's a maximum. This is where the good solar cell materials live like silicon, _____, perovskites, so on. We are talking about, say, 33 percent theoretical efficiency.

So, and this is what you encourage you to work on, somehow to fix this problem, because that's really where the main thing is. I mean we are working here. Specifically, we actually not even work in here because this is – as I said, this is very fundamental. This is only fundamental. In reality, we are down here. We are basically now, in solar cell technologies, we are down here and working our way up to that green curve. That's what we are doing, really. All this is outside our limits. But I encourage to actually examine those limits.

So, for example, if sun was a laser, I mean, disregarding the fact that we don't want that because we would be dead, but if sun was a laser, this is what happens. There is an interesting paper from Zach Haldron's Group where they plotted so-called spectral efficiency. That's basically what it is. You can see that if you come very close to the gap with a laser, then you get the cell efficiencies which are, say, 60 percent for gallium phosphide, about 50 plus percent for gallium arsenide. For silicon, it's also quite high like definitely above like maybe 40 percent or something. Perovskite, maybe 40 or something percent.

This is what you would have if you would have cells and with the monochromatic light very close to the gap. Of course, when the light is higher up, the gap, then, of course, you lose this in thermalization. So, the efficiency goes down. Similarly, to this, you see how these losses depend on the bandgap. The lower the bandgap, the more losses. You can see that you kind of draw a line, you can see how this goes down. Because electric losses are different from different bandgaps.

Okay, so to conclude this, let's kind of come to the real world and look at the cases that I use for the actual PV. So, you have, say, compared gallium arsenide and silicon. What's different and what's common? Direct bandgap has this band structure, as you all know, and this photon we can easily excite electron-hole pair. So, there's a very strong optical absorption. As a reverse process, there's also a very strong photoluminescence. In indirect gap, because you need photons involved in absorption luminescence. There's quite like weaker, about ten times weaker optical absorption. Also, very much weaker photoluminescence like four orders of magnitude weaker than direct gap.

Because there is this strong photoluminescence, the dominant recombination is III-Vs with a radius of recombination. That is if you somehow get rid of all defect. You will still have – this is fundamental. In silicon, since you have forbidden photoluminescence, that's not a problem. But you have auger recombination. For practical, so now we have to come into the practical side, you can see this is absorption depths for gallium arsenide and silicon. Absorption depth is just anywhere _____ _____ _____ _____ _____. So, for gallium arsenide, you can see that with one micro layer, you can capture all of the above bandgap photons. You don't need anything thicker than that. For silicon, here you see that it kind of continues to rise. To capture all photons that are even like close to the gap, it's best to have a thicker wafer like 180 microns. That's what industry uses now. Right?

Here is little bit of aberration of this. So, this is lifetime, due to fundamental process. The fundamental process being auger recombination and radiative recombination. For silicon, a radiative recurrence _____ _____ therefore, the lifetime associated with that is very long. This is auger lifetime. It has this _____ independence of docking concentration. For silicon, it's well-studied and well-known. So, this is actually – this black curve is close to the truth. For auger, I looked at some literature values. A little bit scattered and wore with wider range.

So, I couldn't find it. But approximately, since order, tend to 30, which is very close to silicon. So, actually auger recombination with gallium arsenide is also not far from silicon. So, I don't know what to say. But radiative recombination is much, much stronger. So, the lifetime is talking about that tenths of nanoseconds, hundredths of nanoseconds, maybe, that will be concentration. So, very strong.

Therefore, fundamentally, III-Vs at whatever happens, efficiencies are limited by these radiative processes. Miles will talk about photon recycling that actually allows us not to lose all those photons. I mean we think, "Okay, reclamation. It happened. It's lost." That is not true. If you capture that photon, you can recycle it again. For auger, it's not the case. So, for III-Vs in the silicon, limited by auger. We do some simple calculation.

I won't get into the details but, basically, if you equate the generation rate to get this np product – np product, I remind you, that is what determines the Voc. Then for luminescence pros, very easy to do, because this is just luminescence. Radiative recombination goes as this V times np. So, if I take two microns later, against np product, six times _____ 30 is and with intrinsic carrier concentration for gallium arsenide I plug in here, I get Voc of 1.8 or difference between gap and Voc, 0.32. So, that's really close to the actual – what people have. So, very, very best cells. Right?

For silicon, it's different. Now I have to use auger recombination to set the limit. How do I do that? So, I use high-injection limit, which will always be the case in high – in a good silicon cell. So, basically, I write the same thing but now it's auger recombination. I take 200 microns wafer and I again calculate np product. I get dealt end from here among and then get it squared. So, actually, np product is higher than this. Quite much higher. So, one would assume a higher voltage from silicon than from gallium arsenide. But it's not the case because the depth of gallium arsenide is wider. So, intrinsic carrier concentration is much, much lower for order of magnitude lower than the silicon.

So, when I plug them into formula, actually I got kind of similar different between gap and the open 35 electron volts and IVoc. This is like nobody has reached it yet, but it seems very like something that actually can be done for silicon or 770 millivolts. So, you can see how these two types of materials play completely differently by fundamental processes, one being radiative, one being auger with that Voc. That's what are these limits for technology of the cell.

Here, now, let's talk about silicon PV. So, what's unique about silicon in photovoltaics? So, I already said that silicon is neatly absorbing, limited by auger. Being neatly absorbing has advantage because use wafers instead of thin layers. Turns out that actually that's paradoxically because that's a huge advantage for silicon PV. In fact, that's what makes silicon PV to kick so far at least. These thick wafer, about 0.2 millimeters are easy to process. Technology base already developed. I see those two.

Another advantage of wafers, you can access both sides. You can train to _____ them. You can handle them. Last, but not least, these wafers allow huge throughput because you can process them as batches. You can put wafers in the huge _____, thousands of them, and process for let's say two hours. The effective throughput is huge, basically. That makes silicon so cheap.

So, as to efficiency, this is theoretical efficiency calculated using this order limit 29.4 percent, which is here. This is where we are now. It's connect the recent record, 26.7 percent cell. So, actually, pretty close. This is where gallium arsenide is today. It's altered device is 28.8 percent. The cell, we're waiting for the cell. These curves actually – I will not discuss them. Miles will discuss. But this is what photon recycling would allow us to do, to capture those seemingly lost photons that are emitted as a result of recombination and recycling them and kind of – it's like electron-hole pair photon, electron-hole pair photon. Like this, basically. Then, if you do that with a very high photon lesson sealed, you can actually increase that theoretical efficiency. In silicon, it's not possible. Silicon has auger. No, it doesn't have any photon recycling. It doesn't have any photons to recycle.

So, this is where silicon is. Dominates the PV technology market at 95 percent of the rest of PV. These are these module price curves versus cumulative production for crystal silicon at the end of 2016 was produced 300 though gigawatt peak. I think current number is somewhere that would be about 400-gigawatt peak. Price is going down with technology advancements and production yield. Currently, the module price is below – in some cases, below $0.30 per watt.

This is how technology vies. This looks like it comes – crystalline silicone comes – mono silicon and multicrystalline silicon – multicrystalline silicon, despite being less efficient, actually dominates the market. Monocrystalline silicon constitutes maybe a third or even less than that. Then there are thin films which is basically cadmium telluride and N-type has some share, but little. N-type may be think some call NLG, and also _____. So, these are super-high efficiency cells, more higher efficient then defect but they are more expensive. Therefore, they sort of photo _____ more like a niche.

So, this is how silicon is made. So, we start with Polish silicon, which is actually produced from very clean sand and with some production processes and vilification conversion to soil and gas and so on. Then, this very pure Polish silicon is grown into ingots. This is ingot for multicrystalline silicon. Basically, it's a box, quartz box where silicon is molten and then it's frozen and then _____ and broken apart, breaks apart. Anyway, or it can be pooled like this. Then you make wafers. Wafers are then processed in solar cells. Mainly using batch processes but some processes are in-line.

For example, chemical processes that require access to one side with chemicals are usually in-line. So, have very high throughput like automated chemical tool, which you can see some laboratory _____ example of the _____ lab here at NREL, the clean room. Then they're made into modules. Current efficiencies are like SunPower record modules are 24.1 percent efficient. Panasonic HiT cell modules are 21.6. The cells are quite efficient, actually. The mainstream p-type, we are talking with modules actually approaching 20 percent now.

This is how silicon is grown. This photon process is a very nice, very clean process. It results in very little oxygen and silicon. Unfortunately, it's too expensive or ten-times more expensive than these other ones. So, it's not used in industry. Industry uses Czochralski process for monocrystalline silicon, essentially pulling the seeded – this pull out of the silicon melt. Multicrystalline silicone, it's essentially crucible, which is essentially a quartz box where silicon is molten. Yes?

>>Audience: Do you do all these processes here at NREL or do you just buy your ingots and whatnot from a producer?

>>Paul: Yeah. No, we don't do. We do have a crystal seeded puller, actually, but it's not used. It's – to run this process requires – this is a big project. You cannot just – and also, it's good that it's running like all the time. So, yeah, n-type of p-type, so you see how silicon is sensitive to impurities. It turns out that p-type, which is like mainstream, it's actually quite sensitive to impurities. Even tiny amounts, say 10 to 11 or a cubic centimeter, can actually severely degrade cell efficiency. So, that's why these industrial processes are typically assembled in a cleanroom-type environment even for industry.

N-type is less sensitive to impurity, especially to iron. But n-type is a little bit harder to draw. For reasons, maybe, I'll discuss this with the silicon team later. It's actually – there are other reasons why. P-type has some unique property, one of which you can form Baxter for steel pipe, just because the zinc aluminum and then using _____ _____ epithets, which is not the feature of n-type, unfortunately. So, that's why [door closes] it is less used.

Basically, these are steps that are done to produce cells that, essentially, dopant diffusions are like key. Then cleaning and saw-damage etched, and texture. The junctions are made by diffusion right now in industry. But the next generation maybe will change that. Finally, metallization.

This is one of the enabling – this is list of like enabling processes for silicon PV. It's quite unique. Actually, it might be that some of them will be, eventually, adopted for other cell technology because some work really so good. So, of course, wafering, this is unique to silicon and not important for thin films. Texturing, you can produce texturing like this to capture light by double-bounce and, also, perhaps the long wavelengths like – by just very simple chemical etch process which is very highly selective to crystalline facets. If one doesn't etch, then one zero-zero etch very quickly. So, it produces.

Then, using aluminum for back surface steel in p-type silicon. That's also a very simple process and very enabling, actually, for silicon PV. Then, of course, enabling gettering as a byproduct. So, cell is then, at some point, brought to high temperature. That high temperature is spent many things at once. These things are like forming off the aluminum back surface field, fighting through the paste, and also fascinating to the hydrogen released from the nine track. So, passivation is a big part of silicon PV.

This inexpensive metallization by screen printing – now people talk about different types of metallization, the chemical plating and all this. It's all good. But screen printing is so elegant and cheap. It just beats everything. So, essentially, imagine you put the screen print paste on top of layers which is completely isolate nitrate. Nothing, no current, causes _____. Then you fire it through some special sequence that they counted disintegrate. Magically, there's a chemical reaction that paste eats through the nitrate, contact silicon and you have a really good contact. So, no need of like disorder of things, placing carbon mess. It just happens automatically.

This is how we see maybe the evolution of silicon PV. It used to be very recently, like three years ago, dominated by this very simple structure, aluminum back-surface filled cell where the front was screen printed paste, passivated, diffused phosphorus and so on. But the back was simply aluminum paste. So, you form this back field by just – you dope it by aluminum, simply by this high-temperature step forms and pick tick essentially silicon, liquid silicon aluminum alloy. Then cool down. Liquid phase epitaxy happens and you have this doped region, have this very highly doped P plus region with aluminum. But that compilation is very strong.

So, p-PERC cell, which is now the use standard for industry, it has now – the back is passivated mostly but maybe couple percent area is actually aluminum back side of the spill cell. That's enough for collecting current. So, now cell efficiency suddenly goes up quite a lot from, say, 18 percent here to 23 percent about here. Very big advancement. Then, selective recharge. Just put dopants under this degree to minimize recombination from the metal. It's another advancement.

So, the roadmap says we have to reach 24 percent. It might be that the roadmap will be reached faster than we think. It's already 23 percent up with the audit. But this is not the next generation. So, these are cells which do not require dopant diffusion which has so-called isolated contact with an extremely high passivation of silicon vapor. Silicon vapor is very high lifetime and these contacts are put on very gently and metallization is done very gently. So, this is a HiT cell, and these are variants of like high-temperature analogs of HiT cell where passivation is instead of being done by amorphous silicon is done by tunneling oxide. So, for example, here, you see the tunneling oxide which is very good passivation barrier for silicon vapor but being open only about 1.5 nanometers. So, it allows tunneling. So, current actually can go through. So, you have post-passivation and transport. This is, for example, to try excel by ______ 23 percent was done that way.

This is the role of p-PERC. I will not repeat that. Coming back to here. So, anyways, this is once. Maybe – okay, my computer froze. Oh, now it goes. Yeah, anyway. So, these are the examples of passivated examples. I don't know. For some reason it died. That probably means I should finish. [Laughter] From this, I – basically, I had about maybe three slides that showed interesting scientific directions. Those scientific directions related to the passivated energy selective context, new materials for these passivated contexts. That, we believe, is a very interesting, huge field.

Then, future field, which I discussed with some of you during lunch about how to make a great defect in develop. I mean when the cells become so efficient, like about 23 percent, everything matters. It all becomes very, very important. One would think that silicon, bulk silicon is dull and there's just the diamond like this and all this. But because it contains different impurities, which do matter, especially oxygen, you can have, for example, oxygen precipitation in n-type. That limits cell efficiency, say, from 25 percent and you go down to 22 percent. It's a big deal.

Another example is boron oxygen conflicts in the deep edge seizing material that limits p-PERC long-term stability. Again, how that boron oxygen conflicts react with hydrogen or maybe different types of oxygen dimers that makes this actually stable in long-term. That's a field of advanced study. Then, in multicrystalline silicon, there is another degradation mold involved which happens at elevated temperatures. That still is unknown, actually, what causes it. Most likely, some specific impurities to be debated.

So, yeah. So, those are the directions. Also, of course, we are working on tandems. That's very interesting. We can make tandems that are very efficient. Adele is leading that project. Record tandems with III-Vs and also, we hope, at some point, also, the silicon. So, I will end here and if you have some couple questions I can maybe answer them.

>>Audience: So, I just have a general question. One of the good things about silicon solar cells is that they're relatively cheap to make, but if we're going to get into that regime of more complicated structures, as you add more and more structuring, more layering onto these processes, does it make the overall solar cells more and more expensive to make? Then it kind of negates the advances that you make in efficiency with the cost of manufacturing and it's just a –

>>Audience: Hold it. So, have we stipulated that they're low-cost because they're cheap to make?

>>Audience: My guess, yeah _____ _____.


>>Paul: Low-cost because they cost little.

>>Audience: Yeah, I mean –

>>Paul: You can buy them cheaply. That's what low-cost means.

>>Audience: I mean I didn't see all of Paul's presentation, but he did indicate that, basically, at parts per million and lower levels of contamination that there were impacts to lifetime, right, which ultimately led to impacts in performance.

>>Audience: But, ultimately, the price to buy a silicon solar cell is low.

>>Paul: Okay, let me try to answer your question. So, basically, what you are saying is right. But oftentimes, technology doesn't work the way we think as a physicist, actually, oftentimes. It is, for example, not at all clear why silicon, being so sensitive to impurities and indirect and so on, is actually so cheap. Really you can buy panels which are cheap. Possibly you can make panels cheaper with something. But silicon panels are already very cheap.

If you look at – if you talk with some of industry, they care about things that we think are completely irrelevant. Like, for example, silver content in the paste or led-free paste, things like that. Right? So, all these things are actually very important. Similarly, if you add complexity, if you make tandem, right, it actually – well, that costs quite a lot. But depending how you make it, how you make the standard, it might be – we have debated it a lot actually.

It seems that maybe making tandem within one technology – for example, if you can make just perovskite tandem, right, not put on anything else, just perovskite tandem, that appears, perhaps, to be a better way than to make tandem coupled with something else. Because you are putting two technologies together which are very different and that can cause a problem. However, there might be some unexpected benefits. You never know. But why I encourage in the beginning of this lecture to work about these lost photons, it might be that there exists a way without making complicated structures, just by maybe some nanostructured material or something. Self-organized quantum dots, some kind of cascade something that actually allows to utilize that thermalization energy. So, it's wide open. I mean 50 percent of light is lost. That's my challenge to you.

>>Audience: Why were the diffusion links for the indirect bandgap materials so much longer than diffusion links for the direct bandgap material? Can you elaborate on that more?

>>Paul: Yeah. So, I showed the graph that showed the lifetimes. Right? So, because silicon cannot emit photons, there's no radiative recombination. The diffusion links limited by – there's only auger recombination with fundamental limits. This is why the diffusion links with silicon can be so long. I mean silicon – it's not uncommon to have millimeter long diffusion links. This is absolute – you cannot get them in direct – there's no way, because you will have radiative recombination, essentially. So, therefore, inevitably, if you go direct gap, you have to use thin layers. If you go with silicon, in principle, you can use both thin or thick layers, but just because there's advantages to using thick layers, why not use them. It would be disadvantage, actually, for silicon, to use thin films. You'd have to capture the life and it's just problems.

>>Audience: More questions for Paul?

[End of Audio]