ECE Purdue Essentials of MOSFETs/Lecture 5.3: High Electron Mobility Transistors (HEMTs)
========================================
 

[Slide 1 L5.3] So welcome back. You know, throughout this course now, we have talked about one kind of transistor, the MOSFET, I mean the silicon MOSFET. There are all the kinds of transistors. One of the important other types of transistors is called a High Electron Mobility Transistor or a HEMT. 

[Slide 2] So as I mentioned, we talked about MOSFETs, but there are many many transistors. We're going to be talking about one specific transistor. Almost all of these transistors, it turns out are barrier-controlled transistors. Not all of them, but almost all of them are barrier-controlled transistors. That means they operate on the same essential physical principles that a MOSFET operates on. The HEMT is a barrier-control device, so we should be able to understand the operation of the HEMT in the same kind of conceptual framework that we've used for silicon MOSFETs. 

[Slide 3] Now the first III-V transistors that were used commercially were gallium arsenide MESFET. MESFET stands for Metal Semiconductor Field-effect Transistor. There was no insulator because it was very very difficult to produce high-quality insulating layers on III-V semiconductors. III-V semiconductors were of interest because their mobilities were very high, so they were devices that had a lot of promise for high a speed RF applications. So the device would operate like this, it has a source and has a drain, there's a lightly-doped n-type region in between. The metal semiconductor junction is a Schottky barrier. The Schottky barrier has a depletion length by applying a reverse biasing, that Schottky barrier pushing the depletion region down, we can control the current from the source to the drain. This particular device I'm just sketching here is a depletion mode device, it's normally on, and we deplete it to turn it off. So that was an important device, the attraction of gallium arsenide was the high electron mobility. Now the difficulty is that mobility drops as you increase the doping of a semiconductor. There is more scattering due to the ionized impurities that lowers the mobility. For high current or high transconductance in an RF device, we need both velocity or mobility and charge. So we need both of them and that means that we don't get the enormously high mobilities of pure gallium arsenide. We get a lower mobility because we have to dope the channel to get to charge in there. I'll also point out that one of the undesirable features of this device is that because this is a Schottky barrier, we're limited in the voltages that we can apply. We can't apply too much of a forward bias or current would begin to flow. That's a limitation that came about because of the lack of availability of a suitable high quality oxide like SiO2 that we had for silicon. 

[Slide 4] Now there was a remarkably powerful concept that was developed just before 1980. People discovered a way to get carriers in a semiconductor without doping the semiconductor. So let's say we have an intrinsic small bandgap semiconductor, and let's say that we have an heavily-doped wide bandgap semiconductor, and we put these two together to form what we call a heterojunction. When we put them together, electrons are going to flow from the higher Fermi level to the lower Fermi level. So they're going to go from the doped layer, which has all of the scattering due to the ionized dopants into the undoped layer. We now have electrons in an undoped layer. In principle, we don't have the scattering due to the ionized impurities because this is intrinsic undoped material. We should be able to get a high concentration of electrons and a high mobility. That's the high electron mobility part of the HEMT. 

[Slide 5] When we put the two together, we get a band diagram that looks like this. We get some depletion region on the wide bandgap side, the electrons have left that side and have moved over to the small bandgap side. We get a bending down showing that we have more electrons at the interface of this layer. This was discovered in about 1978. This technique for introducing carriers without doping a semiconductor, it's called modulation doping, because this structure, the dopants are here, the carriers are over here. The electrons at this interface were called a 2D electron gas, high carrier density and high mobility. Now, although these devices frequently refer to the electrons here, not as an inversion layer of electrons, but as a 2-dimensional electron gas. The electrons in a MOSFET are in a 2-dimensional electron gas, because the potential well that the oxide silicon interface quantum mechanically confines them. So very similar to what we would see in a MOSFET. 

[Slide 6] This is a more careful energy band diagram of that structure. Here's our n-doped, heavily a wide bandgap layer, where the electrons are, here's our Schottky barrier that we put on the surface. There is some Schottky barrier height, and we have a depletion region between the metal semiconductor junction. So there's a region near the surface that's depleted. The width of that region I'll call Wsurf. We also have a depletion region near the large bandgap, small bandgap side, that occurs because the electrons on the large bandgap side are transferred over, into the small bandgap side, I'll call the width of that depletion region W. Okay, there is a band discontinuity here, which holds the electrons on the small band gap side, and makes it difficult for them to get out, sort of like the discontinuity between the wide band gap SiO2 and the smaller band gap silicon in a MOSFET. Okay, and there is some band bending. That band bending leads to a surface potential. This would be a positive surface potential, just like bending the bands down in a p-type silicon MOSFET. 

[Slide 7] Okay, now when we're designing, these there are some things we have to be careful about. It's this 2-dimensional electron gas in the gallium arsenide that is going to be the channel of our field effect transistor. That's what we want. If we're not careful, there will be an undepleted n-type region in the wide bandgap layer that will be in parallel with that high mobility layer in the smaller bandgap gallium arsenide. We don't want this layer there because it will degrade the performance of the device by having all of those low mobility electrons in parallel with high mobility electrons. So we would have to be careful to make sure that the sum of the depletion width from the Schottky barrier at the surface, the depletion region at the wide bandgap, small bandgap interface, that the two add up to the thickness of this layer so that there is no undepleted parallel conduction path. 

[Slide 8] All right, so that's something that careful design would do. Now you'll see, I'm labeling the thickness of this wide bandgap layer as t sub ins, like thickness of insulator. We're thinking of it sort of like the gate insulator of a MOSFET. You might ask why do we need to dope it at all? We don't need to dope the SiO2 in a silicon MOSFET. Well that's because the SiO2 has a very large band gap, we can apply relatively large voltages, pull the bands down in the semiconductor, and create an inversion layer. The band gaps of these heterojunction pairs, the wide band gap pair that goes with the smaller band gap gallium arsenide. For example, are not nearly as wide as SiO2. So we don't have large enough barriers to be able to operate this as a true insulator, and to be able to induce inversion layer charges simply by applying a large gate voltage. So, we really do need to... Because we don't have an insulator that's available, we have a wide bandgap semiconductor with a band gap that isn't wide enough. We really need to dope that semiconductor layer. 

[Slide 9] Okay, now let's talk a little bit about the electrons in the small band gap layer that have high mobility. So one of the things that these layers are grown with sophisticated epitaxial techniques called molecular beam epitaxy or Metal Organic Chemical Vapor Deposition, MOCVD. And these techniques have the ability to grow almost atomically flat interfaces between the small bandgap and wide bandgap layer. That's much different than the interface that you get from an oxidized SiO2 silicon interface. For example, those interfaces have a lot of surface roughness which scatters the electrons. These interfaces have a very small amount of roughness at that interface, which helps promote a high mobility. There's also scattering due to the phonons in the gallium arsenide, or small bandgap layer or whatever it is. But there is in addition to these two scattering mechanisms which are inherently there, there's a third. We have separated the electrons in the small bandgap layer from their dopants in the wide bandgap layer, but these electrostatic charges from those ionized dopants. The electric fields can penetrate for a distance, and the electrons and the small bandgap layer can sense those electrostatic charges, and these remote ionized impurities can actually scatter the electrons in the channel. So for that region, reason will often set back. We'll have an undoped layer to try to set back the dopants so that they're further away from the electrons in the channel, and don't lower their mobility as much. There's obviously a trade-off there, if we set it back too much, it'll inhibit the transfer of electrons from the large band gap to the small band gap. We'll get even higher mobility but, we'll have even less charge, and we need both. So there's a trade-off for transistor design. Okay, so that unset undoped set back layer is one of the features of the epitaxy of these layers. 

[Slide 10] Now if we measure the mobility of the electrons in that 2-dimensional electron gas as a function of temperature, we can see that as we cool down below room temperature, the mobilities go up because we're freezing out the lattice vibrations scattering. We get higher and higher mobilities. There will usually be a peak, and then after the peak, the mobility will turn around and drop, it'll drop because ionized impurity scattering has a different temperature dependence than the phonon scattering. Actually as you increase the temperature, they have a weaker effect because electrons zip past them faster. 

[Slide 11] Okay, so at low temperatures, we can actually get quite high mobilities at very low temperatures in very pure materials. The mobilities can be enormously high, and they don't actually turn around because we have such a small amount of ionized impurity scattering. You're seeing here an example of a mobility in these systems that is extraordinarily high. People have achieved mobilities of over 10 million, okay. And as I said, these are achieved at low temperatures, not where we're going to be operating devices, our hope is going to be that we can achieve mobilities appropriate at room temperature, appropriate to a pure material without the doping. That's on the order of 10,000 or so, for these III-V materials. 

[Slide 12] These are done in sophisticated molecular beam epitaxy or metal organic chemical vapor deposition system. Here's an example of what one of those MBE systems looks like. This particular system has actually achieved the mobility now of more than 35 million for electrons at low temperatures. 

[Slide 13] Okay, so this is an amazing example of a physical effect that was discovered in 1978, and has been very important, and a lot of physics experiments. Two years later, a practical application of this modulation doping was invented by Mimura et al. This is the high electron mobility transistor. So it was very quickly put to use. If you'd like to know more about the history of this device since its invention about 30 years ago, more than 30 years ago, I can refer you to this second paper. 

[Slide 14] So the basic device structure as a transistor then looks something like this. We have a wide bandgap layer, we have an undoped small bandgap layer, we have source and drain contacts, we have a Schottky barrier gate inside, so we have a depletion region around it. More commonly these days, instead of doping that wide bandgap layer, frequently what people will do is something called delta-doping. You'll have an undoped wide bandgap layer, you'll stop and add an atomic, one atomic plane of dopants. This is called delta-doping, and then you'll continue the undoped wide bandgap layer. So the dopants are all located in one plane that is set back a little bit from the interface. Okay, I'll point out now that the material systems that are used now, it's more common to use material systems such as indium phosphide, indium aluminum arsenide, indium gallium arsenide, then aluminum gallium arsenide, and gallium arsenide which was first used. Alright, so there's our 2D electron gas, and ends up being produced in the small bandgap region. 

[Slide 15] Now why delta-doping? Some people have discovered that there are some good benefits of this delta-doping. One of them is that you can if you do it correctly, you can get a higher channel charge this way, by introducing a lot of dopants and that single atomic plane. It generally modifies the electric field in a way that it increases the breakdown voltage of the gate electrode, and it tends to be able to put the gate electrode closer to the channel. That as we know from MOSFETs suppresses 2-dimensional electrostatic effects, and it also gives us more gate control over the charge in the channel that gives a higher transconductance of the channel. 

[Slide 16] This particular transistor was first called a HEMT, a High Electron Mobility Transistor. You'll see several other names that were given to this by groups that we're developing their own versions of this device elsewhere in the world. But the most widely used name continues to be HEMT, referring to this type of a transistor. 

[Slide 17] This is an example of what a more realistic structure might look like. You know, the channel lengths are comparable to minimum silicon channel lengths these days. There will be a large band gap substrate, it might be indium phosphide or gallium arsenide. Then there will be various buffer layers to do the epitaxy on top of that. Indium gallium arsenide is a wider bandgap layer. The red layer here is a smaller band gap channel, like the indium gallium arsenide has a smaller band gap. The more indium we put in the channel, the higher the mobilities would be. So the highest performance devices now have very indium rich InGaAs channels. Then above that will be a wider bandgap indium aluminum arsenide layer which we think of as sort of like the insulator of a MOSFET. It'll be delta doped in order to get the dopants in the channel. There will be some lower band gap layers on the top in order to facilitate ohmic contact to the source and the drain. The gate itself has a dimension of 25 nanometers or so, and then it mushrooms out into a larger cross section because the gate resistance is important for RF applications of these devices. And there will be metal contacts on the source, and the drain will need low contact resistances just as we do for... Just as we do for silicon MOSFETs. So that's what a structure looks like. 

[Slide 18] The point out, it involves some very sophisticated growth, epitaxial growth of a number of different types of layers in order to produce these. But the techniques have been developed over the years to do that and produce very high quality materials with very low defect to densities. 

[Slide 19] There are many applications of MOSFETs. I'm sorry, of HEMTs, they were initially explored for digital logic, but then their benefits for RF analog quickly became apparent. They're especially good at low-noise amplifiers in the micro or millimeter-wave RF spectrum. So they're used in satellite communication and radar astronomy, and other applications, and also in cell phones themselves. They're also used in some applications as millimeter power amplifying devices. 

[Slide 20] So this is a little bit dated sketch that shifted compares the cutoff frequencies versus year, shows you the evolution of this technology over the years. There continues to be a horse race between this device and the heterojunction bipolar transistor which we'll discuss next. But the main point I want to illustrate here is that, a, you can produce monolithic millimeter-wave integrated circuits with small numbers of these transistors that have important applications, you know, tens or hundreds of transistors, not millions or billions as in silicon MOSFETs. And the other point is that the cutoff frequencies that can be obtained in these III-V transistors are significantly higher than you can achieve in silicon MOSFETs. And that's their benefits when you need very high frequency RF devices. 

[Slide 21] Now we can also, because if we draw an energy band for this diagram, for this device, it's very similar to a MOSFET, it's operating principle is a barrier control device, or we're controlling an energy barrier with a Schottky barrier. Now instead of with an MOS structure, we can then analyze this, in fact it can be analyzed with our virtual source model very carefully. We can compare the measured characteristics to the ballistic characteristics in the red line. This is an example of a transistor that operates very close to the ballistic limits. So it's much better to assume that this is a ballistic device when you take the initial look at it and try to understand its I-V characteristics than a traditional device dominated by scattering. 

[Slide 22] Okay, so near ballistic operation for these devices. Now what about three-five MOSFETs? There's actually been a lot of progress recently in learning how to grow insulating layers on a III-V semiconductors like gallium arsenide with low enough interface state density that they can actually perform as high quality MOS transistors with a III-V substrate. So for a summary of where that work stands, I can refer you to a paper here. 

[Slide 23] So, first of all, I need to thank two of my colleagues for helping me put together this lecture, and then let's summarize some key points. We've had a very quick look at a transistor that you should... You should know that it exists, you should know how... Quick look at how it works and what its applications are. So these are an important technology for high a frequency RF applications. This device and the heterostructure bipolar transistor that we will be discussing in the next two lectures, have both achieved terahertz speeds. So they're remarkably a high frequency devices. They operate in exactly the same barrier controlled mode that the silicon MOSFET operates in. So they are well described by our virtual source model. And you should also be aware of the fact that they operate very close to the ballistic limit, much much closer than silicon MOSFETs do. 

[Slide 24] Okay, so we've seen another type of transistor in this lecture. There is a different type of transistor. The very first transistor that was invented at Bell Labs is something called a bipolar transistor. A bipolar transistor operates in a little different way than a field effect transistor does. But bipolar transistors are barrier controlled transistors also, and modern bipolar transistors which are usually heterostructure bipolar transistors also have important RF applications. The next lecture will be a brief review of PN junctions in order to set the stage for our discussion on heterostructure bipolar transistors.