We’ve all heard about how wide-bandgap semiconductors are changing things, and how they will bring about a new revolution in power systems design. That’s all well and good, but it’s time to get specific and talk directly about the benefits and how they can be implemented.
By Alix Paultre, Editor Power Electronics News (based on a technical paper presented by Steve Colino, Efficient Power Conversion)
In this case we are going to discuss Gallium Nitride (GaN), a Wide Bandgap Hetero Junction materials technology. What’s really cool about GaN is it’s really a fundamentally new conduction mechanism to take advantage of. The GaN that everyone’s talking about is still very far from theoretical limits. For comparison, the HEXFET was designed in 1977, and 40 years later it’s bouncing off of some of its theoretical limits. With the current pace of development, it will take us less than 40 years to get to theoretical limits on GaN.
Most of the GaN that you’re seeing for power conversion is GaN on silicon. GaN on silicon is a less expensive wafer than a silicon carbide wafer, although it does add to the challenges. Enhancement-mode GaN starts out with a not quite standard lattice structure .
On top, we’ve got an Aluminum Nitride Isolation Layer. On top of that we grow a very thin layer of GaN. To put it into perspective, or chips are about 0.7 millimeters thick, with only a couple of microns of GaN, so it’s a very, very thin layer of GaN that gets grown. GaN is a pretty good insulator, so to coax it to be a conductor, we put down aluminum gallium nitride and have the spontaneous polarization giving the 2 Dimensional Electron Gas.
To make it an enhancement mode device, there’s a couple of different ways to deplete the electrons that are used, but there’s a certain area that gets depleted and then that’s where our gate goes and then we put our terminals and then the super structure with dielectric and metal layers to collect the current. It is a purely lateral device, so that’s one of the limitations that we’re going to have. For a given RDS(on) and voltage capability, GaN has a much lower capacitance, it’s a smaller device, and it’s lower inductance, depending on your packaging. A major value of GaN is in the switching speed of the device.
For example, EPC’s latest device, the EPC2045, is our fifth generation 100 Volt device. Looking at, everything that we look at is in the about performance and fast switching. If you want to include the QRR and the size of the device is also, it’s not quite one tenth, but it’s not far from one tenth of the size of the comparable silicon device.
At EPC we continue to stick with the wafer-level packaging, because it has got some significant advantages. The figure below shows it’s a single interface to the PCB, so it’s a very, very simple structure.
In addition, the active area of the device is intimate with the PCB, so we get some very good inductance cancellation throughout the whole high frequency loop. If you look at the cross section and again, these are roughly to scale, the wafer-level package is very intimate with the board and you can kind of see the clip that goes up and over that’s going to add inductance and it’s also going to pull the conduction path away from the PCB. With it being intimate, we’re able to do some interesting things with the layout.
When we integrate the high side with the low side to a half bridge, that reduces to about 200 picohenries, and that’s energy that you’re putting in and taking out every time you’re switching, so the loss component is tremendous. The overshoot and the ringing is tremendous, so we have a very, very small overshoot, very little ringing and that’s a big component to the EMI. The best way to combat EMI is to generate less of it. Also the wafer level package are RoHS 6 of 6, we don’t have any lead, we don’t need any exceptions and it’s got a great thermal path. We’re also MSL-1, there’s two layers of protection. One is a glass layer and then there’s a polyimide layer, so it is an MSL-1 device.
What does eGaN technology have less of, that’s really what we’re looking for is reduction in parasitic elements and where does it impact the waveform? I did some work on some class-T audio and it’s really taking a look at all the different parasitics that are shown in the figure below.
If you look at common source inductance this is the one that’s the absolute worst because the DiDt is going to get higher and higher you hope, the DiDt gives a V, which takes away from the gate drive voltage. If you look at a MOSFET loop inductance versus the wafer level package inductance, the overshoot and ringing is a lot less and you can really make a good estimate based on the ring frequency.
Thermally, these devices are dual side cooled. In many cases we do reduce heat and we’d prefer to not use a heat-sink, but really using a heat-sink we can get a lot more power out of it. The sides of the devices are quite substantial in area, so to maximize heat transfer, gap filler in between is very, very useful. Then a heat-sink can go on top and then the heat-sink would be compressed onto the device, giving really an incredible thermal capability.
So what can you do? There are a lot of people that go and say, “Okay I’ve got 100 kilohertz converter, and if I use your parts I get a 1% increase in efficiency, but your parts are more expensive.” Where GaN has gotten traction is where customers really try to think differently about their applications.
When it comes to output power, for example, one demo uses a Gen 5 200-Volt device with Kapton as an interface, and a heat-sink glued down with RTV around the edges, with temperature sense leads underneath. We rounded up to 100° C and we got good correlation between the temperature sensors and the thermal imager on the back of the board.
The figure above shows an efficiency curve where we take the current and step it from nothing to 34 amps. In this example we use 140 Volt to 28 Volt converter, but it really could be anything. We step it up to 34 amps at 200 and 250 kilohertz. If you take a look at it for a motor drive where the duty cycle is going to be about 50%, if you increase the duty cycle you’re basically shifting the losses from one device to the other device. It should be about the same. If you look at what a 50% duty cycle with 140 Volts at 34 amps, that’s about two and a half kilowatts of power that’s being delivered by something that’s about this big. It’s really incredible, the power that you can get out of these devices with a little bit of heat-sinking.
Using a limited, simple heat-sink structure, but with dual-side cooling really gives you power, and instead of putting energy into the board, its taking the power out of the board. Usually your power dissipation limit is thermal, now it’s board temperature that’s going to limit how much power you can get out of it.
Other things you can do with high frequency performance are applications like envelope tracking. With 4G and 5G, you’re getting higher and higher peaked average ratios. With a constant power supply your PA is a constant power dissipator. At high output power, the power’s going to the antennae, at low power you’re just burning the heat in the PA. What if we can get a frequency that’s high enough that we can follow the power envelope and now you’re going to reduce the power dissipation. The average efficiency doesn’t go up by 1 or 2%, it goes up by tens of percents in these things.
GaN can also give the notebook computer and tablet guys a lot more space to work with, either for reduction or for reduction in weight or for reduction in thinness. GaN is getting up to about 86% efficiency at 10 amps, and this is one where it’s ultimately going to be a two phase implementation at ten amps per phase. This is one where it’s just an extreme size reduction that again gives the capability.
When it comes to a 48 Volt to 12 Volt bi-directional controller, in one direction the high side’s the control switch, and in the other direction, the low side’s the control, so you really don’t have room to trade off. You need a good switching device for both positions.
In wireless power transfer, 6.87 megahertz and 13.56 megahertz is a highly-resonant power transfer, so one of the things you can get is you can actually get power into the body. Now what does that mean? That means that somebody who needs a heart pump or a nerve stimulator doesn’t need wires going into them.
On the consumer end of things, with what you can do with the highly resonant area, is you can create a desk with a constant magnetic field over the desk and simply place different items that have the receive coils in them to power it. It’s not just wireless charging, it’s really wireless power that you can get with the resonant conversion.
GaN gives fast, low loss switching, which really gives the designer a whole different toolbox in order to work with, in solving his power conversion problems. Power conversion costs money, takes space, dissipates heat, and doesn’t deliver value. GaN can make it cost less, be smaller, and deliver value.