MOSFET getting too hot or is normal ? LT3750 flyback converter

[moshik3]

I am using the
LT3750 capacitor charger with the
coilcraft **broken link removed** (N=10, 10uH, 10A Ipk) and a
IPB530N15N3 MOSFET D2pak (150vDS, 23miliohm, 9nC, 21A, 84A puls).
Current sense resistor is 8 miliohm, like the LT3750 datasheet suggests for 10A current limiting, all other components similar to the datasheet.

my output capacitor is 500uF 900v. i charge it up to 800v. this is 160 joules, it takes about 8 seconds.
the thing is, that the MOSFET get very hot after a few repetitive charging cycles.
after about 2-3 minutes of repetitive charging (charge to 800v, discharge capacitor with large resistor, in 100ms, and re-charge immediately) it gets so hot that it starts melting its own solder on the PCB ! :evil:

my rough instinct tells me that charging 160 joules in 8 seconds is 20 joules per second which means 20 watts.
if we take a very bad case scenario, of 50% efficiency, the MOSFET will waste another 20watts of power, or almost.
can this case lead to such high temperature that it melts its own solder ?
or i might have some other problem that causes efficiency to be way below 50% ?
is there some formula or thumb rule to predict the heating or power loss in such a setup ?

according to bottom of page 9 in the LT3750 datasheet, the Iavg of the mosfet is 4.34A at 12v which is about 50W.
(N is 10, VOUTpk is 800, Vtrans is 12v)
can it be that it wastes so much power as heat ? it is relatively low Rdson and very low gate charge (9nC)
on the last page (16) of the datasheets, it shows a graph charging 110Joules in just 5 seconds at 12v, with 90% efficiency.
that means that out of the 50W power passing in the mosfet, less than 5W SHOULD BE wasted in heat. this is surely not my case...

I know that i might need to move to a TO220+heatsink setup for higher currents, but that issue make me stop and check, because it might be a bad sign...

Any help would be highly appreciated !! :grin:

- - - Updated - - -

and following is the board layout:

 

[moshik3]

Anyone please ?

 

[godfreyl]

if we take a very bad case scenario, of 50% efficiency, the MOSFET will waste another 20watts of power, or almost.
can this case lead to such high temperature that it melts its own solder ?

Yes, of course. My soldering iron uses less than 20W and it has a much larger heat-dissipating area than the MOSFET.

less than 5W SHOULD BE wasted in heat. this is surely not my case...

Why not? The picture below is from the MOSFET datasheet that you linked. Your board layout looks like the MOSFET has almost minimal cooling, so 5W will cause a temperature rise of about 300 degrees centigrade (5 * 62 = 310).

Even with 6 square centimeters of copper for cooling, the temperature rise would still be 200 degrees (5 * 40 = 200).

 

[BradtheRad]

Are you applying a high enough volt level at the gate, in order to drive it to its minimum ON resistance?

You gave a figure for amps going through it. Did you simultaneously measure volts across the mosfet pins? What value does that give you for resistance?

Can you take readings at other gate voltage levels? It might reveal what gate voltage you need to apply.

It might reveal that the high temperatures have changed the mosfet's operating characteristics. Etc.

 

[moshik3]

The chip drives the gate to VCC-2v.
I tried both VCC of 12v, and VCC of 16v. - no change in heating.

the only thing which i am not sure about and comes into my mind is that the inductive spike caused by the transformer leakage inductance strikes on the Mosftet Vds. (like explained in the datasheet )
can it be that when i set RVout to give 300v the spike is rather small and not causing harm, but when i set RVout to give 800v at output - the spike is so big that it causes such heating of the MOSFET ?

I have measured the primary current (across a 4miliohm resistor between primary and Drain) and i can see some triangular shape forming, but it drops right after the beginning of the triangle, and then continues to rise, and when the MOSFET turns off there is a very big spike downwards (the same amplitude as the MAX triangle height) which means current flows FROM the drain to the Primary.
can that cause the heating ?
it also happens when the output target is 300v.

- - - Updated - - -

i will do another set of measurements later today.
can you please tell me what to scope ? what points and in what ways ?

Thanks !

 

[BradtheRad]

Everything points to a need for a larger heatsink (as godfreyl suggests).

Perhaps even adding another mosfet or two in parallel with the burdened one. (Although paralleling mosfets brings its own issues).

If the spikes only occur once every few seconds then they play a minor part. However if they are in a pulse train then there are measures to reduce spikes. They probably result from switching heavy current through a coil. A diode across the coil is often used, although you may need more robust methods.

What the spikes are liable to do is to change the operating characteristics of your mosfet. Such as increase its 'On' resistance at your applied gate voltage, or prevent complete shut-off at zero gate voltage, etc.

 

[mtwieg]

As others have implied, that MOSFET has almost nothing to cool it. As it is, it can probably only dissipate 1-2W without exceeding its SOA. You need a FET with a better heatsink.

For a flyback converter like this, the FET shouldn't dissipate more than maybe %10 of the input power. If it's dissipating more, then something is wrong with the circuit.

Normally you should have some sort of snubber circuit on the flyback primary, otherwise the primary leakage energy will be dissipated in the FET. However it seems the datasheet for your controller doesn't mention using one (perhaps because it would interfere with the voltage feedback). So you can expect that leakage energy to be dissipated in the MOSFET as well.

 

[moshik3]

Hi,
thanks, the spikes are the one i suspect. as i said the problem comes when the output voltage is above 500v mainly so it might be the limit that start generating really big spikes (?)
The datasheet doesnt show any snubber, although the datasheet of LT3751 (a similar part, identical application) does show the use of a snubber.
i tried different values of R and C, but none reduce the problem.

I tried paralleling 2 MOSFETs, with and without individual gate resistors, but each of them is getting hot like before. sometimes one even hotter than the other.

can you help me determine if "something is wrong" in the circuit ? at ~20 watts output power, it should dissipate 10% as you say, which is 2W. - i dont think it is the case.

let's concentrate on the clues we got:
WHAT can cause increased heating while the output voltage is raising over 500v ? (up to 800v)
bigger spikes ? we named it, what else ?

isn't one behavior of flybacks - that the higher output voltage, the higher duty cycle is ?
that would mean that the MOSFET is ON much longer than when the output is at 300v...
am i wrong ?

what measurements can i do to check things that can go wrong ?
I only have the DPSCOPE which shows quite well the charging waveform at about 1us cycles

I thought of trying and ordering a 110v TVS (transient voltage suppressor) and connecting across the primary. (Vout=800 divided by N=10 is equal to 80v. plus Vin=~15v, total ~95v.
while the Vds of the mosfet is max. 150v, and the TVS is 110v,
should it eliminate any spike that can cause the MOSFET to go mad ?

 

[crutschow]

The transistor you are using has an ON resistance of close to 50mΩ, not the 23mΩ you stated so the efficiency is likely worse than you think.

That transistor can only dissipate perhaps two watts without a heatsink. If it dissipates much more than it can indeed get hot enough to melt solder (and fry the device). So you need to mount the transistor on a good heatsink.

 

[BradtheRad]

You are encountering the typical problems (overheating, spikes) that crop up when dealing with high currents and high voltage.

I am only slightly acquainted with these problems. I fried a few mosfets trying to construct my own inverter.

I observed spikes which must have been caused by switching the transformer coils. The spikes got worse with greater current flow.

I tried slowing the mosfet transitions to soften the spikes, but this made them spend more time in linear drop mode. Temperature rose.

I attempted to add the right kind of snubbers. I tried different locations. I could not figure out what it would take.

Between the overheating and the spikes, I decided to give up.

If you manage to succeed at this project, then you've accomplished something on the level of a commercial manufacturer.

For the time being, you'll have to build for robust-ness. Mosfets are hardy in some ways, yet they are fragile in other ways.

Solve one problem, find another. Steps in the path of troubleshooting.

- - - Updated - - -

I tried paralleling 2 MOSFETs, with and without individual gate resistors, but each of them is getting hot like before. sometimes one even hotter than the other.

This is unequal load distribution. To divide the load evenly it is common to install equalizing resistors in the current path.

Even if it may reduce power throughput, it's important to divide heat. If 2 mosfets don't do the trick, consider using 3.

Another way to obtain equal current distribution, is to fine adjust turn-on volt levels at the gate of each mosfet, while reading current flow through each mosfet.

High power output requires high power input. The current path must be as low resistance as possible.

isn't one behavior of flybacks - that the higher output voltage, the higher duty cycle is ?
that would mean that the MOSFET is ON much longer than when the output is at 300v...
am i wrong ?

The greater the duty cycle, the greater the output V. Up to a point.

Frequency of operation also has a lot to do with it. The 'On' time must be long enough to let current flow ramp up to the required level, in order to get the desired output volts.

Does the IC manage all this automatically? You may want to do your own experimenting in this area.

what measurements can i do to check things that can go wrong ?
I only have the DPSCOPE which shows quite well the charging waveform at about 1us cycles

You still need to verify whether the mosfet is turning fully off and fully on. Read volt level across the mosfet both when it is off and when it is on.

Install a low ohm resistor in the current path. Just high enough ohms to get sensible readings. Likewise read volt level across it when the mosfet is off and when it is on.

Calculate values for current when on and off. Calculate mosfet resistance values when on and off. These are chief indicators of what is going on.

When you say 1uS cycles, I think you mean time increments? It would not be possible to run the converter at 1 MHz.