Current designs using wide-bandgap semiconductors are driving switching speeds through the roof, with significant impacts across the board. What are the challenges on standard power modules, and how can we actually optimize these standard power modules to be working well enough with silicon carbide?
By Alix Paultre, Editor Power Electronics News (based on a technical paper presented by Stefan Häuser, SEMIKRON)
What is fast switching for us? In an appication space like motor drives, a speed of 15 kilohertz is basically already fast enough. So let’s look beyond that into devices operating at up to 40 kilohertz, especially when we talk about higher power applications. Here, a full silicon carbide solution may not even be necessary, because we know that there are good silicon devices especially in 650 volts, giving good trade-offs so that you can actually select the right chips and also add silicon to get a very good benefit and very good efficiency results. Multi-level topologies lend themselves well to a hybrid silicon carbide solution. So you can keep the switch as an IGBT in the design and just exchange the freewheeling diode to a silicon carbide device. This kind of a solution already can give you up to 50% of power savings, but this is not the focus today.
When we talk about higher switching frequencies of 40 kilohertz and above, that’s where we really should talk about full silicon carbide. Efficiency, modulation accuracy, and in the end, a reduction of size, and hopefully of cost.
So what are actually the challenges when we talk about integration and power modules? Let’s look at systems rated 20 amps and up, and then it also involves commutation inductance. Lower inductance means faster switching, or higher voltages, or higher currents. Thermal performance is key, so we need to imlement new technologies in the power module to make sure that you get the maximum out of the silicon carbide chip.
So let’s talk about stray inductance first. When we look into a system design like the one oin Figure 1, we have a half-pitch module here that obviously has parasitic stray inductances, and when we turn off of a switch, then we see we have a DC bus voltage, but if you measure the voltage on the IGBT you see an even higher voltage. So why is that? Just because of the di/dt, so the change in the current we’ve seen voltage drop over the stray inductances. This occurs in every switching incident, so this has to be really considered.
Figure 1: Stray inductances occur in every switching event
Oscillations, or hysteresis, is another point. If you switch faster, then you will have oscillations coming from the parasitic inductances and capacitances. This over-voltage is, of course, important that you do not exceed the maximum clocking voltage of the device. If you cannot reduce the stray inductances, you have to switch slower. So you lose performance in the end. And of course, on the other hand, you can also lose maximum DC bus voltage, so we have to run the whole system at a lower DC bus voltage just to make sure that there’s enough margin to this breakdown voltage.
In the end, this all comes down to one main part: reducing the stray inductance. When we look at a conventional power module like our SEMITRANS 3 (Figure 2) which is a 62 millimeter module that has been in the market for a long time. If you’ll look at how the parasitic components are actually split up in this power module, you see, first of all, in the connections to the outside to the load or to the DC bus we have a lot of stray inductances coming from the bond wires and the DBC traces.
Figure 2: Note how the parasitic components are actually split up in this power module
So how can we actually optimize something? Clear inductance is directly depending on the area that your commutation loop spans. So clearly, for the DC+, DC- loop, which is always in the conduction loop, you have to arrange that area as small as possible. In Figure 3, where DC+ and DC- are quite far away, it ends up at 25 nanohenry. If you’re able to lead these terminals, and this is really the modular terminals, in parallel, this inductance comes down. It’s a very simple thing to do. In some other modules, it’s not so easy to do. So 11 nanohenry is already quite a decent value.
Another important point is the gate inductance. And there we found actually that a minimum of the gate inductance is not really desirable. We tested a lot of different modules with a lot of different gate inductances and what we found actually is that if you draw the switching losses over the gate inductance you see that there is an optimum, which is not at 0.
Figure 3: If you’re able to lead the modular terminals in parallel, inductance comes down
The reason behind that is, if you have a slight gate inductance, this gate inductance will act as a current boost in the Miller plateau during the switching process. And this really helps. So we see actually an optimum in the range of, let’s say, 15 to 17 nanohenry. If you go lower than this, you need a much more powerful driver to really push the current inside. If you go higher than that, the inductance will just slow it down to complete the switching process.
Let’s talk about thermal performance. In silicon carbide, the current density is much higher, which means coming from the fact that you do not have a lot of switching losses, and also gain in forward losses, and the chips are very small. In the end, you lose thermal performance because the Rth of your chip, the thermal resistance of the complete power module depends on the chip area you put inside.
A very easy way to do that is actually change the ceramic substrate that we are using where we put the chip on, and which gives also an isolation to the heat sink. So there are different materials available if you just take the first one, aluminum oxide which is kind of the standard today, comes with a certain thermal connectivity, a certain thickness, resulting in a thermal performance, which I just note as a kind of indicator.
Now, we have two different materials we can also go for. Silicon nitride, it’s a very stale, robust material, giving a much better thermal performance or thermal connectivity of 90 watt per meter Kelvin. Same thickness because it is so stable you can use it at the same thickness and you have a much better thermal performance. Alternatively, aluminum nitride might be an option. Even better thermal performance, but, unfortunately, aluminum nitride is a bit brittle, so it’s better to use it at a higher thickness, which then, of course, gives away some of the thermal improvement, but in the end, this higher thermal connectivity compensates so these two materials are basically equal.
What happens now if you use one of these materials in the power module, and let’s talk about the SEMITRANS 3 again. 62 millimeter. We started there, it was a very basic version. We took the standard power module, threw out the silicon stuff, put in as much silicon carbide as we could, which means in this case, 12 chips, making it a 500 amp power module. We reached a certain Rth traction case per chip, and then it continues drain current in the module of 430 amps, roughly.
If you then change the ceramic material to aluminum nitride, we can actually reduce the number of chips, reduce the cost of the power module. The Rth per chip is still better, the drain current that we can run continuously is almost the same, but the overall cost of the module comes down. This is a nice, simple way to optimize the module in a way to really optimize also the number of chips that are used and bring down the overall cost of the power module.
Annother consideration is the coefficient of thermal expansion. With every thermal cycling, or power cycling, you have an expansion of all the materials, and unfortunately, all materials in a materials stack up in a power module, they have a different coefficient. And if you choose here again the right materials, you can actually bring these coefficients closer together. Standard to day would be aluminum oxide silicon. You see here already the gap is quite big between these two coefficients. But if you change just the standard module to silicon carbide, the gap gets bigger.
We talked before about the aluminum nitride and silicon carbide, so this combination, and here you’ll see, aluminum nitride CTE is much lower so it’s much closer. It’s a very, very simple way. You don’t have to do anything else with the packaging technology, and not too much with the packaging technology. Just exchange the materials and you will gain or get the same power cycling performance. Confirmed and approved also in our lab.