Mosfet Failure in offline 2-switch Forward converter

Thread Starter

michael hemingway

Joined May 24, 2018
Hi All,

I'm hoping there's some expertise that can quantify the dimensions of my ignorance. This is a high power (2kw) offline 2-switch forward converter project. Obviously this is not the natural topology for applications at this power level but I had read this was possible though clearly not the right option commercially. The board is based on a UCC28513 combined controller with integrated gate drivers for both boost pfc stage and PWM for DC-DC converter. The board is built and seems to function correctly up to the main transformer. The pfc stage correctly regulates at 385V.

The issue is with the forward converter section. The two switches are driven using a double dual duty transformer coupled gate drive circuit. So the gate drive signal from the controller is capacitively coupled through a 1:1:1 pulse transformer which is rectified providing both power and signal to a pair of ucc27424 (4A peak gate drivers). For what it's worth, without the mosets populated the gate drive signals they generate look good on the scope.

Looking at the figures you might see the layout is, let's say, suboptimal. All, or most, of the high dissipation components where aligned to use a single large heatsink. I realise that the respective grounds of the two gate drive circuits are high current paths, with switching occurring at 100 KHz going from conduction to freewheeling. I just couldn't see a way of avoiding this.

Long story short, when turned on detonation occurs. The main victim is the PWM current sense resistor (30W, 0.068 Ohm). Seems to me that the mosfets fail short circuit, a few microseconds later the primary winding saturates, essentially shorting the 385v bus through the current sense resistor. The transformer is sound, saturating at around 16 A (500 uH primary). I don't think that's the issue (it was before, lol).

My feeling is this is a high dV/dt or dI/dt issue. The gate drive current loops are tight but the power loop is clearly quite big. This seemed unavoidable during the design. Both Mosfets and freewheel diode are good quality 600v with current rating with healthy margins. In my naïve enthusiasm and lust for efficiency, I've used 1 ohm gate drive resistors (anti-parallel diodes for turn off). I'm going to slow the turn on with higher values (got some 20 Ohm 1206s to hand). I've killed some of the finest mosfets money can buy. Its broken my heart. I'm using cheaper ones (higher Rdson) until I know what the problem is.

Obviously there's a lot of information missing that might help and can be provided, but anything that anyone could add would be helpful. Thanks in advance.


Joined Feb 8, 2018
First, your assessment of how the FETs fail is probably bang-on. :D Excitin' innit? I've blowed up a few. Wait til you blow up one of the filter caps. (in all seriousness, I recommend blast shields when working on this sort of thing - heavy acrylic or polycarbonate if you need to see what you are doing, plywood if not)

Your layout looks pretty good. I frequently say that in switchers, everything is in conflict with everything else. You want all the parts crammed into the tiniest area possible, preferably zero, but that is in conflict with heat management. 2 kW is rather a lot for a forward converter, but if you don't mind the big chunk of ferrite and can handle the currents - why not? One advantage over something like a half-bridge is that you don't have to worry about unbalancing the voltage on series-connected filter caps if you use current mode control.

My first suspicion is the gate drive supply voltages. I don't see an undervoltage lockout on the gate drivers. Unless you are managing this with some other mechanism, the possibility of insufficient drive for the gates is very real.

One ohm is perhaps a little low, but I doubt if it is causing any problem related to failure. Without looking at the FET spec's I'd probably be using something in the 5 ohm range. Many of the new FETs have delightfully low gate charge, so you might do well with even higher. When I was doing this sort of thing (never again!), I frequently wished for an instrument that would display all the power path voltages and currents from AC in to DC out and compute and display the powers and efficiency to high resolution. It would make quantifying the effects of messing with gate drive resistance and the like a lot easier. Taking a small efficiency hit can sometimes be quite worthwhile if it improves something else.

One thing to beware of with really fast switching: you can put some fearsome current spikes into the heatsink via the capacitance between the drains and the sink. There is no easy fix for this. You can add intermediate shields of copper foil connected to ... somewhere, but that degrades your thermal interface. Thickish ceramic insulators can reduce the capacitance, but they aren't the nicest things to deal with and the high thermal conductivity types are expensive - especially boron nitride.

Thread Starter

michael hemingway

Joined May 24, 2018
Thanks Ebp,

on the underdrive condition, The ucc27424 has dual enable pin which are joined. The En pins have a pair of 100k pull-up resistors. This did cause a small issue as originally these were similar but different gate drive ICs. R28 has, I forget, some value which make the En threshold of 2.4v when the vcc reaches around 6v. This gate drive scheme need a few cycles before it charges the gate drive circuit. Thinking about it, I'll probably change the resistors for the Enable divider just to make sure they are almost fully charged.

On the current sense transformer, I knew it was the right option, its just I didn't want two parts of the circuit to have been designed by me, freestyle. It felt like there was some doubt about the gate drive part. If and when this works the next project will be a 3kw full-bridge ZVS type. I feel more comfortable now and doing the job right with current sense transformers.

I feel the instrumentation pain. I'm testing with a safety(isolation) transformer so I can float it and attach a SINGLE probe to the scope. So it's a case of choosing my signal wisely. I'm going to make a full set of active differential probes before I get involved in this madness again.

The heatsink issue is one I was aware of. There's a decent EMC filter for the line but I just didn't know where to start to try and quantify the effect between switching elements. The two stages (boost PFC and PWM) are synchronous. Not sure whether that will make it worse or better. Oh god I know about the ceramics. I'm a research physicist and have had to use Double copper clad aluminium nitride to make microwave laser mounts. The thought of combining this nightmare with that one is enough to make me consider getting a proper job and a

Thanks a lot though. Just not being called a moron has given me the energy to blow a few more things up. I'll keep you posted.