Switching losses in half bridge

That means that all the drain voltage swing comes from a nearly zero ohm source... the other fet.

That driver is connected to the gate through 10 ohms. The gate is connected to the other fet through the drain to gate capacitance and the only thing limiting the current is the drain to source resistance and the rate of change of the voltage. Which do you imagine will be stiffer?

I think your turned off fet is turning back on during the drain voltage swing.

The transformer inductance is swinging the drain voltage during the dead time, as the gate voltage drops, but its impedance is high enough to lose the tug of war with the gate driver during the slew. This is a weird case where less load causes more loss because the slew doesn't take place during the dead time.

IGBTs tend (tend...) to have slower switching times than mosfets, which would support John's contention of miller effect switchon during transition from on fet to the other.

Which igbts were you using?

Cheers

PeteS

You should be paying attention to Ciss, along with gate-drive current that determines the switching speed. Your new "lower-capacitance" part actually has nearly 2x the gate capacitance, 4300pF vs 2400pF. I'd look to improving your FET driver scheme, what are you using now?

I have a fairly classic half-bridge 'forward converter style' ( but no feedback loop - it runs open-loop ) SMPS with a Vbus =~ 320V and operating frequency ~ 100kHz.

You may recall I posted about his a while back - notably in connection with the UC3525.

It's working and delivering load but I've got to the point where I need to look closely at various aspects of its operation and refine them. In particular for example I've been looking at switching losses.

I'd noticed that my mosfets were running warmer than I'd like. Part of the problem was that some snubber Rs were actually contributing to the problem, so I took a step back, removed them completely and made some measurements from new.

I can run the bridge into a simple 33kohm load which just gives me a basic load so that I can look at the no-load switching waveforms. With this puny load I saw a heatsink temp of ~50C on the mosfets.

I assumed switching losses on account of the Coss. A quick calc showed a calculation on that basis to be in the right ball park with the observed temp rise, so I substituted some other mosfets I had handy with about 60% of the original Coss value.

To my surprise they actually ran about 1C hotter !

Anyone got any bright ideas why that should be ?

Mosfets are Infineon SPW20N60C3 ( originally )

and Fairchild FQA28N50 ( substituted )

Ton is ~ 4us and Tdead is ~ 1.5us ( big but I have a trimpot on it ) btw.

Graham

Not visibly. You mean as in dragging the output voltage towards Vbus/2 ? That point has intruiged me in its own right.

I don't recall looking at it recently but it damn well should be ! It's got an IR2013 driver and about 10 ohms Rg.

Erk ! I don't think so. You're thinking of the 'Miller capacitance ' effect ? The IR2013 ought to keep that sorted.

No - lol. No neighbours. The office is in an industrial park. No adverse signs locally either.

Incidentally when driving the transformer - there's no noticeable temp rise. And the output waveform looks truly lovely. Nice slow transition from 'rail to rail' almost perfectly matched to the dead time. Lmag ~ 900uH btw.

Graham

You've given me some interesting stuff to chew over there.

Curious since I would describe this as a 'plain vanilla' application. I reckon the app notes are rather less helpful than they might be.

Graham

For a while I was previously using some IGBTs in place of the mosfets. I don't recall this dissipation with the 33k load with them. Does that make any sense ?

Graham

To fully understand and analyze their circuit, you need to know their transformer's leakage inductance, and a few other things.

OK, here's a rough tutorial on calculating MOSFET power-switching losses. You started with Coss, which has to be discharged each cycle. This is a dissipated energy of 0.5 C V^2 = 4.25uJ for one 20n60's 83pF of "energy-related output capacitance." A rep rate of 100kHz puts this loss at 0.85 watts, not very significant.

Another very relevant MOSFET spec is related to gate capacitance Ciss, the Qgd parameter, the Gate-to-drain charge. This is the gate charge required to move the drain voltage from the on to off voltages, during which time the FET may continue to conduct at up to the maximum current (depending on the load inductance, which includes the transformer's leakage inductance). For the Infineon 20n60 Qgd is 33nC and for the 28n50 it's 52nC, which is 58% higher.

The IR2113 has the capability of sinking about 2A of current, but you've added 10 ohms, so the gate current will be closer to 0.5A. Since t = Q/i we can calculate 33nC/0.5A = 66ns for the '20n60's switching time, and 104ns for the '28n50. If the transformer's primary current is 2A at turnoff, we can calculate 0.5 * 320V * 2A = 320 watts during the switch-off. The turn-off duty ratio is 66ns/10us = 0.66% which times 320W = 2 watts shut-off dissipation. If the transformer current is continuous, add another 2W for turn-on switching loss. We're up to about 5W, without counting ohmic losses. However, given these FET's low 0.25-ohm Ron values (at elevated Tj), which given a 50% duty cycle corresponds to only 0.5W at say 2A, it appears this will be minor.

I'd do two things to improve the scene. First, I'd add try adding Schottky diodes across the 10-ohm resistors to allow for the fastest- possible turn-off to help the MOSFETs. Then I'd say your FETs are too large: bigger is NOT better, because it means more capacitance. Yep, I'd consider replacing these pigs with something 2 to 4 times smaller.

Hey Poo, This sounds like "Electronics 101" again. Why don't you give us a schematic and photos of gate and drain voltages and currents. This is a simple problem that can be fixed if given the correct data. Cheers, Harry

It's an IR2013 with Rg = 10 ohms. I see what you're saying. I'll take a closer look at that tomorow.

Graham

Yes. Noticeably so. I'll drop them back in again on Monday for comparison.

Yup. I follow you.

Infineon SGW30N60.

Cheers, Graham

Correction - IR2113.

Well.... there is some truth in what you say but I was just mulling it over and concluded that I've never ever seen an app note that covers the clever 'itsy bitsy' stuff. What I'm experiencing must happen thousands upon thousands of times to other engineers new to this discipline. I'm sure this is a part of the reason why SMPS design is regularly seen as a bit of a black art by many. It's not actually inherently complicated rather more not so well explained in detail.

In a similar vein - trying to get my head round transformer design, I came across no end of reams of info with complicated equations yet when I saw the light there's about 2 or 3 design decisions to make and Bob's your uncle ! I now can see transformer design as nearly trivial but it took ages to be able to see through the forest obscuring the core info.

I was indeed thinking of posting some pics but fancied checking here first to see if I'd missed something. For example I'd now want to check my Vgs waveform again.

If you're curious, there's a very similar circuit that's available online. Indeed it's from a competitor's unit. The product is highly mature btw and well regarded for its reliability amongst other things.

Page 5

I wouldn't say I actually 'copied' it but it's close enough for rock 'n roll.

One notable difference is their inclusion of a 'sort of' resonant tank in series with the transformer. L6 is ~ 2uH btw and serves also as the primary of a CT.

I'd get my colleague to make me a pdf of the schematic were it not for my embarrassment over his idea about how a schematic should look ! :-(

Cheers, Graham

Why are you using a SG3525, it does nothing for your circuit but complicate it. Just go to the IRF website and get a Half Bridge Driver with internal oscillator and fixed dead time, typ IR2153.

No, sorry, it simply doesn't work that way, it's the winding method that makes all the difference. For example, if a winding is broken into two parts, one part under the other and the other over, you'll get a dramatically lower leakage inductance. Conversely, if one winding is on half a form with an insulating spacer, and the other wound separated from the first, you'll get a dramatically higher leakage inductance. And etc, for various scheme to further enhance the effect. Leakage inductance can be and is often tuned in SMPS designs. A bit extra may be added externally, to reach Nirvana.

As it happens, a high leakage inductance can be very beneficial for low-loss resonant switching. So, you really do need to know the transformer's leakage inductance in order to properly evaluate a design. You could buy one and measure it, or you could learn the physics and calculate your own optimum level and how to wind one customized for your design. At no extra manufacturing cost.

Several of us have written much about this subject in these pages, all searchable by Google. No offense. :>)

You'll have to deal with the case where the rectifier caps are dead zero and tight coupling transfers this short circuit to your half bridge. The extra inductance (leakage or external allows the current limit a low enough rate of change of current for it to function, safely. The full output voltage, full load may not be the instantaneous worst case for the half bridge.

It's a convenient and relatively inexpensive method of generating the required waveforms and includes a shut down input and a power-on ramp function.

I really want control over the dead time. Certainly at this stage of the hame anyway.

I did look at the 2153 or one of its cousins but didn't choose it for the above reasons.

I'd just ben musing over gate drive. I imagine there's some benefit to be had from driving the gate negative to turn off.

Please point me to an app note or design brief that gives me that 2005 lok if you can ! ;-)

Actually, I haven't managed to break it since I fixed an issue caused by an early oversight. The overcurrent shut down seems quite effective.

Graham

I know what turns it's got and the core so I can certainly calculate Lmag and take an informed guess at Lleak too. I expect them to be pretty close to my own TX actually.

What 'other things' did you have in mind ?

Graham