Gate driver for big IGBTs

[TMaxElectronics]

I thought I'd make a post about a little project I've been working on for a while

It's a gate driver made specifically for gates of large IGBTs (like the one Steve gave away a while back). Nothing special at the moment, but I do want to include de-saturation detection at some point, and maybe even measure dissipated power (accurately measuring the current would be an issue for that though). The goal is more or less to just be able to source much more current than integrated gate driver ICs could, with a convenient isolated logic level input for control from any old FPGA/CPLD/whatever.
I'll probably make a github repo and document it properly at some point so people can replicate it if they want to.

The low and high voltage sides are bridged because I fried the last digital isolator I had by soldering it in wrong way round...


The driver itself is powered from a 10W flyback converter with 18V and 4V (winding ratios didn't work out for 5V) output. Well at least the transformer is dimensioned for 10W, but the sot23 mosfet would probably get a little cooked running at that power level continuously I'm not even sure how much power I'll actually need for driving the IGBTs in the end though, so it might actually be quite overkill
I've wound the transformer on one of the cheap aliexpress bobbin and core sets, which I really like. It's regulated from a feedback winding and driven by a pic12f1572 running a pid loop. The primary mosfet is extra-low gate charge and seems to be happy running from the PICs pin directly even at 250kHz. With the pic it would also be quite easy to make the drive voltage adjustable via RS485.
At 500mA current on the 18V rail I got a surprisingly high efficiency of 91% with the mosfet reaching a marginal 95°C according to the thermal camera.

For the output stage I'm using a SO-8 mosfet with one P and one N channel mosfet, driven by a normal gate driver. That seems to work fine, except for around 100ns of cross-conduction when the output switches off... might need to add some diodes there.
I have two different resistors for charge and discharge at the moment even though I don't know if I'll need them, because the big drivers I've seen do the same. My plan is to handle dead times in the driver logic so if its for that its not needed.

I ran a burn in test with a 470nF cap on the output (with additional 1 ohm 5W gate resistors). Things got a little toasty drawing 7W but it happily drove the cap at 50kHz.
I'd say the performance is decent, except for the overshoots. My guess is that those are from the inductance of the wire wound power resistors (measured as ~1uH ).




And finally here's the performance driving an actual IGBT (IXXN110N65C). The ones that kept blowing up in my coil



What's your guys verdict? What sort of test can/should I run on it to see if it works ok in all circumstances?
And what do you think about using SMA as the output connector? It should be very low impedance and is nice and robust.

[davekni]

In general it is a great idea.  I did something somewhat similar for my DRSSTC using 40 TO247 IGBTs - four isolated supplies and discrete gate drivers.  I'm using 0-18V with a pair of TO220 FETs for each leg (10 parallel TO247 FETs), as they have no internal gate resistors.  Is 18V/-4V appropriate for your bricks?  Some bricks are specified for +-15V.  That is due to their internal gate resistors without parallel diodes.

I don't directly monitor Vce, but do estimate power and die temperature using current sense and simulated Vce.  There's more details in the first post too, but here's one post describing that:
https://highvoltageforum.net/index.php?topic=798.msg5332#msg5332

Here's a schematic of one switch of my DRSSTC isolated gate drive, including resistors to reduce shoot-through current:


BTW, you can get ~2x power dissipation from an SOT23 FET by adding copper area to the drain terminal ECB node.

[TMaxElectronics]

Is 18V/-4V appropriate for your bricks?

Driving to negative gate voltages might be interesting. At the moment the ~4V is only logic supply. I also don't really aim to drive any specific brick, the driver should be general purpose.

I don't directly monitor Vce, but do estimate power and die temperature using current sense and simulated Vce.

Well that way de-saturation can't be detected though, only assumed right? Apparently there's a pretty simple circuit that just uses a diode and an opamp that can do voltage sensing during the on-state.
Then I'd only need a few logic gates and a seperate slow-discharge transistor and that'd be it.

[TMaxElectronics]

So I just decided to test the IGBT driver a little more thoroughly before ordering some PCBs and hooked it up to some IGBT bricks I had laying around.

Powerwise it had no issues with any IGBTs I tested with, including an old cm300dy-24h with a lot of gate capacitance (even managing a decent gate waveform at 500kHz... not that that IGBT would be usable at that frequency )
The propagation delay of it is 50ns for switch-on and 35ns for switch-off which I think is perfectly acceptable.

It did however have a serious problem with shoot through of the p- and n-channel mosfets. I added this network to fix this, since I didn't really find any other solution while googling:


I also noticed that the gate waveform looks very weird when driving an skm200 brick: left is without a load, right is with 3ohm@12V; Yellow = input, purple = output (before gate resistors), blue = gate, cyan = collector-emitter voltage

Shouldn't it be a capacitor like charge curve if no load voltage is applied?

And can anybody recommend some connectors for the gate voltage sense line? I would like to make the driver easily replaceable if something goes wrong which includes having connectors for everything

[davekni]

I added this network to fix this, since I didn't really find any other solution while googling:

Another option is to reduce supply voltage of the driver chip.  I run my driver chip from 12V and the FETs from 19V.  If your FETs are designed for even lower gate voltage, reduce farther.  R/D delay works fine too.

Shouldn't it be a capacitor like charge curve if no load voltage is applied?

Looks much like my CM600DY-13T waveforms.  The initial step is due to brick internal gate resistors.  Then gate capacitance charge, Miller plateau, and final R/C gate charging.

[TMaxElectronics]

Another option is to reduce supply voltage of the driver chip.

I saw that in the GDT driver from the UD2.X/UD3, but since I want to be able to switch the transistor on continuously that wouldn't work. I also just noticed that the gate drivers from an old NdYag flashlamp driver just doesn't care about shoot through and connects both gates together... so a little bit probably wouldn't be the end of the world anyway. For my circuit it was just bad enough to cause excessive ringing with the (admittedly poor) PCB layout and gate capacitance.

The initial step is due to brick internal gate resistors

Interesting... I didn't know the SKM200s had any until i looked it up and in deed they do have 5ohm.
Still thought the miller plateau only occurs when the collector voltage is >0V, and you get the dV/dt across the gate-collector capacitor. But obviously its there even without that

I finished the layout yesterday and should have time to thoroughly test the finished version in a few weeks

[davekni]

Still thought the miller plateau only occurs when the collector voltage is >0V, and you get the dV/dt across the gate-collector capacitor. But obviously its there even without that

You are correct.  The small Vce charge was enough.  Most of the Miller charge is at the low-voltage end due to non-linear capacitance.  If you short Vce, the plateau should go away.

[TMaxElectronics]

A miracle has happened and I finally had some free time so I decided to test the driver circuit a little again. I made some PCBs for them a while ago, and included space for expansion boards (like desat-detect shown here).


I've kept the same PIC based flyback power supply, in which I have quite a lot of trust now. It did kill quite a few mosfets initially because of non-zero reset values in some registers but that's fixed and I haven't had any issues since then. I even included a duty-limiter algorithm that prevents core saturation based on the ratio of input voltage to output voltage. The regulating loop is a PI-controller (no D term) with load compensation (for non-linear feedback voltage under load).
If there is interest I'll post the code here.
The power supply also easily stands up to the task of driving even the massive 3600A IGBTs I got from Steve a while back at 100kHz without any problems.

One thing I didn't consider when choosing usb-c for the connections is how they can be easily mixed up. In the beginning I had one port for every driver, but I changed that to one port per half bridge by stacking two drivers on top of each other.

I've been running the bridge from up to 350V (no more, since I can't regulate the pwm yet), and so far there were no problems from common mode voltage on the output. I did however drive the IGBTs at only 10kHz so the stress on the drivers was minimal.


And finally, I also tested the desaturation detector. I didn't have the courage to go for a direct low impedance short of a half bridge, and limited the current with an inductance of ~35uH. But I did still push up to a thousand amps and managed to kill the IGBT when the detector was disabled. For supply I used the bus from a semikron power module, modified to only have the caps in parallel instead of parallel-series strings. With the delay and trigger threshold properly adjusted it managed to detect and shutoff both a switch on into a high current, as well as overcurrent during on-time. The trigger to gate-off delay was less than 500ns, so I believe it should actually be able to protect half bridges from direct bus shorts. But I don't have enough half bridge modules to verify that claim


There wasn't much effort put into cost optimisation yet (even including a 2€ RS485 transceiver for dynamic voltage adjustment over serial, which I won't ever do), so the prototypes cost around 25€ each. But I reckon with a little bit of optimisation I could get that to 20€ or even 15€. Quite a bit more than GDTs, but I think the desaturation detection, as well as PWM capability makes it worth it. Oh and of course GDTs need drive circuitry aswell

I'll write up a little more documentation on the drivers and create a github repo with all the data needed to build your own, if anybody is interested