Loading summary
Transcript1 lines
- [00:02]
A
Gallium nitride is a remarkable material, Dr. Umesh Mishra, a UC Santa Barbara professor and pioneer in GAN transistors and LEDs, said at a retrospective of the WBGSRF program. What that is we'll talk about later, he said. Gallium nitride actually is kind of God's gift to mankind. It just kept cooperating and cooperating, and you're like, what? Why is this material working? Gann's physical problems are legendary. Despite those problems, this grumpy material has changed the worlds of light, radio and power. That's kind of interesting. In today's video, Gallium nitride's journey from a technical flop to a bringer of light to the gates of power. The element gallium was discovered in 1875 by a French chemist named Francois Lecoq de boisboudran. I am 100% certain that I failed to correctly pronounce his name. I beg his forgiveness. The son of a wealthy wine merchant, the Frenchman taught himself science from books and built a chemical laboratory in the second floor of his house, as we would all do if we got the time and money there. He extracted gallium out of a mineral ore sample and identified it as a new element, naming it after France. In 1932, scientists reacted pure gallium with ammonia gas at temperatures of 1000 degrees Celsius. The result was our handsome protagonist, gallium nitride. A hard, chemically stable material, gallium nitride has a direct band gap two, three times wider than that of silicons, a high electric breakdown field, and an electron mobility that favorably compares that of silicon and silicon carbide. Such a combination is pretty rare. In terms of downsides, the material is just average in thermal conductivity. But the key negative trait, one that has defined it for decades, is its manufacturability. We can't make big pure boules of this stuff like we can with silicon. Sad. Gann's breakthrough moment was the blue led. And the dramatic true story has been told several times before, including by some very large channels. If you feel me covering this a bit too lightly, I recommend going to those videos for additional detail. In the 1960s, the iconic American electronics company RCA started playing with various compound semiconductors. Then chairman David Sarnoff wanted to produce an LED that can replace picture tubes in televisions. Using gallium, arsenic and gallium phosphide, RCA scientists produced red and green LEDs. They needed blue to complete the picture. RGB Sometime in 1968 or 1969, an RCA R&D scientist named Herbert Maruska Received a suggestion from his boss, James Tietjen, and about gallium nitride. Gan's bandgap energy corresponds to UV light, making it a compelling candidate for a blue LED. But first they had to grow it. Armed with RCA's rich budget, the team pioneered an exotic method called hydride vapor phase epitaxy. HVPE works by first running hydrochloric acid gas over liquid gallium, then mixing it with ammonia inside a high temperature chamber. The GAN layer is grown on top of a sapphire substrate wafer. Yes, the precious gem. The resulting GAN has many defects, but it was still a big step forward. After RCA published their results, other companies like Philips adopted the technique. The RC18 then encountered a second obstacle. And the LED we know and love needs something called a junction, which refers to a border between two materials. In this case, the required junction is between one material doped to have a plurality of electrons, N type, and another doped to have a plurality of electron holes, P type. Such a junction between P and N types is called a PN junction. And normally the two sides don't much interact. But a voltage run through that P N junction can force its electrons and electron holes to meet and recombine, and in doing so, create light. But nobody knew how to make P type gan. They knew you had to dope it with another element. They used boron to make P type silicon. But what? Lacking one, they tried an alternative arrangement called a metal insulator semiconductor. This is a sandwich of zinc doped GAN and indium surrounding N type GAN with zinc doped gan. This structure produced a green led. Changing the zinc to magnesium produced a nice blue. The MIS is a remarkable achievement and they even patented it. But in terms of commercialization, it was a dead end. The efficiency was too low and it will not scale to a laser. They needed a PN junction, and if they had the time, RCA might have gotten there. But misfortune intervened. Their R and D lab fell into chaos after David Sarnoff's death in 1971, and the company turned a big financial loss with its computer development efforts. Tietzhen and others were forced to cancel the project. After RCA abandoned the project, Bell Labs, Philips, Hitachi and Toshiba all did the same. People largely quit working on gan based blue LEDs, shifting to seemingly more promising materials like zinc selenide. Work on GAN survived only in the form of individual lonely researchers. So we have a twofold problem preventing us from producing a P N junction. GAN based blue led. The first issue concerned growing a nice defect free Layer of GAN As I mentioned, Maruska first grew GAN on top of a sapphire substrate using halide vapor phase epitaxy. But it was far from ideal. There is a mismatch between GAN and sapphire's crystal structures. If you try to grow your GAN layer on top of bare sapphire, this lattice mismatch can cause little islands to form then leading to crystal clusters that kind of look like mold. In the 1960s, scientists growing gallium arsenide had the idea of buffer layers, a layer grown on top of the original substrate to absorb those lattice mismatches and provide an even foundation for additional crystal layers. This is the idea that Hiroshi Amano applied to GAN using a new tool that did not exist in the 1960s called metal organic chemical vapor deposition. He deposited first on top of the sapphire an aluminium nitride buffer layer at a relatively low temperature of 600 degrees Celsius. The result was a smooth layer of GAN. The second major issue was the P type GAN. As it turned out, the RCA team in the 1970s had been on the Reich track. Magnesium was indeed the right dopant. It just needed more work to be activated. The ammonia used in Maruska's HVPE method had significant amounts of water impurities. This led to hydrogen atoms interacting with the magnesium, creating traps that prevented them from electrically activating. Amano and colleague Isamu Akasaki addressed this in 1989 by running a low energy electron beam over the magnesium doped GaN, creating a weak P type GaN. Thus, in 1989 was demoed the first PN junction based blue LED. It was reported as being not very bright. I guess we have something in common as most of its emission was in the UV spectrum. After this first breakthrough, Shuji Nakamura of Nichiya Chemical Industries in the late 1990s changed the E beam annealing method to a more uniform thermal annealing1 to produce a better performing P type layer and a brighter blue. Over the next few years, they tuned the GAN material by alloying it with indium nitride to eventually produce a practical high brightness blue led. The Candela class double heterostructure blue led. This is the one used today all over. Once you have a blue led, you can use that to produce a blue laser. Such lasers can be used in disc readers, projectors and some forms of interferometry. The real boom was the white led. To achieve it, they put the blue LED inside a semi transparent chamber cover coated with a special phosphor made from yttrium aluminium garnet or yag doped with cerium. When the blue LED inside lights up, some of its blue light passes through to the outside. The rest is absorbed by the yag phosphor, which then emits a greenish yellow light. Voila. It mixes together to make white light. Amano, Akasaki and Nakamura won the 2014 Nobel for their efforts. They deserve all the praise. Usually it looks easy with hindsight, but with this one, even hindsight shows this took some serious belief and technical skill. Yet we should also acknowledge all the ideas, techniques and new tools that helped them make their breakthrough. In a 1995 profile about Nietzsche and forthcoming blue toy, Wired magazine dubbed the blue laser the Holy Grail, the closest the semiconductor biz has to pure sex. Might be the first time I have ever said those words on this channel. The article then teases a future where the far more efficient, virtually unbreakable and long lasting white LED replaces the old tungsten filament lamp. HP's George Crawford is quoted as saying, there's no reason you can't replace tungsten lamps in a wide variety of applications with with LEDs. They will replace a lot of the lighting market. I don't know how long it'll take, but I think it will happen. Article author Bob Johnstone says in the next sentence that this view sounds extreme, which is a bit amusing to read today because that is indeed what happened. The first blue LEDs cost about $15 per unit at release, but new competition from the American firm Cree and their GAN on silicon carbide chips helped reduce those prices. Companies in Japan, Taiwan, and the United States rapidly entered the white LED market. The first white LEDs produced by Nichia and other Japanese companies, put out a bluish light that was cool and maybe a bit clinical, suitable for flashlights, car headlights and sign illumination. But for general home lighting, consumers wanted a light that was a bit warmer. Such an LED would not arrive until 2002, gradually opening up the massive home lighting market as their brightness and cost structures improved. Another market that emerged were mobile devices. Phones use LEDs as backlights for their displays, keypad light sources and lighting for cameras. In 2005, mobile contributed 52% of the white LED market. Today, white LEDs hold 80 to 85% of the general lighting market, saving the world immense amounts of electricity. The LED boom massively grew the ecosystem of gallium nitride fabrication and research and in a way, helped pave the road to Gan's second revolution. Radio frequency RF chips help devices transmit and receive wireless signals. When transmitting, they upconvert the digital baseband signal by mixing it with a carrier frequency and then boost the signal power so that it can reach the tower. When receiving a signal, the reverse happens. For a long time, RF transistors were niche items, in part because they handle signals in the gigahertz frequency range or higher, and in part because until the 1980s, the markets they served were in the shadows. Defense items like radars, radios, etc. It would not be until the rise of satellite television in the 1980s that any substantial civilian markets existed in RF. Since then, we of course saw the emergence and rise of mobile phones, which massively upscaled the RF chip industry. The blue LED needed a P type gan, which wasn't achieved for many years, and luckily this wasn't necessary for RF. In the late 1970s, Bell Labs, following a trail of work on semiconductor superlattices started by Leo Esaki and Raphael Tsu at IBM, demonstrated a way to produce a thing called two dimensional electron gas. Two D E GS are a state of matter where electrons can freely and very easily travel in two dimensions, but are constrained in the third. You can imagine something like a sheet. Bell Labs then conceptualized the niche transistor structure now known as the high electron mobility transistor, or HEMT. HEMs are field effect transistors with a source drain and a gate made from a heterojunction, a junction between two different semiconductor materials. When exposed to a high enough voltage, the heterojunction forms a layer of 2D electron gas and electrons can travel through it at high speeds, allowing the transistor to switch very fast. The first actual hemps were produced by Japan's Fujitsu in 1979, though a team at Thomson CSF in France came in a close second. It used heterojunctions made from undoped gallium arsenide and N doped aluminium gallium arsenide. Then, in the early 1990s, it was discovered that we can make hemps with N type gan and aluminium gallium nitride heterojunctions. This caught the attention of the US defense industry. In the 1990s, the US Navy sought new radars sensitive enough to track more distant targets and extend the ranges of their missiles. This needed more power, but they had maxed out the power levels of their gallium arsenide chips. Something else was needed. Fortunately, the Navy was incubating potential replacements. Throughout the early 1970s and 1980s, the Office of Naval Research had been underwriting fundamental research on several wide bandgap semiconductors, trying to find a suitable candidate. Much of this work was championed by a director there named Max Yoder. Gann emerged as one of the major contenders. After the ONR de risked the the space, they drafted the commercial vendors to take it home. In 2002, DARPA, the military's RD arm, kicked off the Wide bandgap Semiconductors for Radio Frequency Applications program. At the time, early GAN chips were riddled with defects. Some likened it to mold or Swiss cheese, which to be precise is mental cheese. The WGS program sought to improve this as well as the whole industrial ecosystem. For rfgan, it was theoretically possible but seemingly impossible to achieve in practice. And remarkably, the ecosystem achieved it. The WBGS RF program ran for a decade plus across three phases. They improved GAN epitaxial quality, built longer lasting high performance hemps, and produced monolithic microwave ICs with record performance. Considering the challenge and how they went all the way from materials to transistors to entire weapons systems, I dub it one of the most successful commercialization projects in DARPA's history. Certainly outranks the Internet in my book. GAN is now a backbone defense technology. In the 2010s, Raytheon and other defense contractors began integrating GAN based RF electronics to replace older gallium arsenide. This includes radars for tracking ballistic missiles or jamming systems. On the consumer side, GAN helped usher in the mobile phone revolution with roles powering cellular infrastructure like base stations for millimeter wave, they're generally not used for handsets due to their high costs compared to silicon. So we have now gone from light to radio. Now for us to take our ring of GAN to the gates of power. Electronics like your phone charger or EV have inside them power semiconductors power transistors gate the energy being sent to light bulbs, phone batteries or motors. So much modern technology today requires the precise control of power. For instance, power being fed to electric motors needs to be adjusted up or down to to do precise work. Or more simply, power fed to LEDs can be varied to adjust brightness levels. It is important to note here that while there is some overlap, RF and power semiconductors have different needs. Transistors for RF chips focus on high frequency and efficiency in turning DC power to RF power. Power transistors on the other hand are high voltage switches. Frequency is less a concern. They should be capable of handling high voltages without breaking down. They should also allow a lot of current to flow through with the least resistance and heat. And for safety reasons it is best that the transistor is default off rather than on. So like a gate held up by gravity that suddenly loses its power it automatically closes as a failsafe. These are called enhancement mode transistors. Since the days of the revolutionary thyristors, we made power transistors out of silicon. But while great for digital logic, silicon's relatively low avalanche breakdown field of about 0.3 MVC is a limitation. It means a strong enough electric field can cause the material to inadvertently switch from an insulator to a conductor not desired. GAN's avalanche breakdown field, on the other hand, is about 10 to 11 times higher than silicon's. Its wider bandgap allows a GAN power switch to handle higher voltages than a silicon one can, or handle the same voltages more efficiently, allowing us to make the device physically smaller and lighter. So over the years, power devices based on GAN hemps have replaced silicon in certain situations. As I mentioned, GAN isn't used in phone handsets for cost reasons, but many people still personally interact with it through their phone chargers. Power adapters have long used silicon chips. And while GAN enabled adapters have existed since about 2010, it was not until later in the decade that these smaller adapters entered the market. Produced by Navitas Semiconductor and made popular by mobile phone OEMs like Xiaomi, Samsung and Oppo, such adapters, along with other consumer items like audio amplifiers and computer power supplies, are the most visible example of gan's strengths in the power semiconductor space. Low to medium with a range between 30 to 650v. Despite the success that GAN has had in that low to medium space, it has struggled to climb out of that. That high ground of high voltage. Applications like EVs, data centers, HVDC and trains is held by silicon carbide. And historically, GAN has struggled in this market segment because silicon carbide has good vertical mosfets and GAN does not. The HEMT is a lateral device, meaning that its source gate and drain are all on the same plane. The current flows horizontally across the surface plane from source to drain through the 2D electron gas layer below the aluminium GAN layer. This works well for high frequencies, but a lateral structure is challenging. To scale up to higher voltages, raising a transistor's breakdown voltage requires proportionally raising the distance between the source and drain. But going laterally, we can only extend that so far. Again, lateral HEMT with a breakdown voltage of 1200 plus volts would need a channel length, which is the distance between source and drain, of 16 micrometers. Small for a pollen grain, but it makes the chip hilariously long and is antithetical to scaling vertical. Transistors have their sources and drains on opposite sides of the wafer. So the power current flows through the bulk of the device. This means raising the breakdown voltage can be done by easily thickening the distance between the source and drain. So why is this hard for gan? There are a few issues. One is that the most economical way to grow GAN is on top of another substrate like Sapphire silicon carbide or silicon that won't fly for a vertical transistor. Because then obviously it wouldn't be a GAN vertical transistor. Now bulk GAN crystals are possible. I read an interesting article about some made by a Polish company called Amono. Unfortunately they went bankrupt. A casualty of the difficulty of turning cool science into a profitable technology. Though their core tech was later acquired by the state. Nowadays we have this thing called GAN on Gan with the GAN substrate grown via. And it all comes full circle here. Hydride vapor phase epitaxy. The same method used by RCA so many years ago. Issues remain, but we are making progress here. And then two. These are MOSFETs. So thus returns the old P type GAN demon. Yes, we have P types good enough for LEDs. But its magnesium dopants require higher energies to activate, so only a small percentage of them do. What Amano and Akasaki did years ago was good enough for the blue led, but not good enough for a proper vertical mosfet like what silicon carbide can do. Though there is one niche vertical structure called the current aperture vertical electron transistor or cavet that uses the P type in a superficial way. This might finally be changing. In October 2025, Onsemi debuted a vertical GAN transistor made in the United States built on a GAN on Gan substrate. They say that it is capable of handling 1200 volts. I look forward to seeing it clash with silicon carbide in the high voltage world. Let them fight. Gallium nitride is indeed a wonderful material. How many things can do so well at three different fields? It is unlikely that GAN will ever get as big as silicon, financially or culturally. Silicon still owns the digital logic space and GAN is not changing that. GAN is most likely best suited for its niche spaces in RF and power. But those spaces are becoming increasingly important and valuable. Worth tens of billions of dollars and growing fast. And they're also key to providing performance in this new AI enabled high tech world in things like robots, EVs and renewables. For some reason, I cannot quite quit these wide band gap materials. So I will continue to cover this topic more in the future. Alright everyone, that's it for tonight. Thanks for watching. Subscribe to the channel. Sign up for the Patreon and I'll see you guys next time.