ringing with mosfets problems

What are the voltage/frequency/power parameters for the circuit?

gate drive is 12V, rail voltage 100V, about 200khz frequency.

Chris

How are you generating the gate drive? Nice non overlapping source or is there any danger that both pull up and pull down FETs are on during the switch over? If that's ok perhaps you need to slug the gate even more? Perhaps 47nF capacitor on the gate and or up the series resistors.

Do you only get this problem with the load connected? If so try an inductors between the output of the bridge and the load.

the problem only happens when theres voltage on the bridge, regardless of load... Even the gate driver IC's when loaded the spike thier own inputs, running from a 45:45 duty so no overlaps... If I "upset" the duty cycle on one side, you see a spike travel across the ground, up the rise edge of the wave and then across the top of the wave, when this spike is "in the middle" of the 12V wave there is no problem, but when you get this spike on the rise edge or near it causes problems... I think all these spikes of various mosfets are just adding up to a larger problem. BTW, the duty would be more like 30:30 at that point, by far from a cure... I could do with a delay line between each stage to spead the spikes across the wave. I have done that already to a point, which works, but wayyy to overcomplicated so its not really a good option. Delaylines are expensive also, was thinking of adding in about 500ns to spead the spikes across the wave a bit more... probably would work though no really a good option....

I'm trying to find some fast gate opto isolators which will run upto 200khz, Agilent make some nice drivers it seems, they are transistor output which I think will help also.

I think the mosfets capacitance is spiking the driver state, trying to pull the outputs to 100V, the IC may even by shuting down, Adding gate resistors helps but it does not solve the problems... I think if I can isolate the mosfets gates from the driver circuit then that should work.

You can look at the squarewave and it would look acceptable, but what you don't see is the spikes on the rise edge, if you "mess about" with the wave adding loads and capacitance etc (not too much tho) you still get the squarewave, BUT you then see that that on the rise its actually switching on and off about 5 times before it actually finishes the rise on the wave, its about 1mhz... hence my posts about filters... it does help also but does not cure it totally. I think if I can isolate the power mosfets and the driver stage then nothing can feedback and hoepfully not effect my waves...

Chris

Chris

--
Can you post a schematic?

Theres nothing to it, just look up the IR2010 example, i'm using 2 of those chips for the H-bridge.

Chris

--
You're the one with the problem, so the least you could do would be to
post a link instead of expecting me to go looking for it.

the

middle"

rise

It sounds like you saying that you have a noise spike that is _asynchronous_ with the switching waveform?

Is that what you are saying?

I've taken some digicam images of the wave, then you can see what I mean.....

wave 1 is the gate driver with no H-bridge voltage, number 2 is with 20V H-bridge voltage..

Which that looks bad, there is another problem which you can't see. on the rise there is about 5 on/off spikes, they don't show up on that image but they are there. Like I say adding filters does help, when you do it breaks up the rise and you then see a lot of notches, a schmitt gate help to clean the last part up of it, though that still does not 100% cure the problem.

The driving wave is say 50:50 duty, if you alter it slowly to say 50:50 -

50:30, a spike then travels along the low side of the wave, up the rise (where the problem is) and then across the top of the wave. It would seem that one wave as it turns on and off is spiking the other wave. It can't be down to cross conduction since 50:30 gives a fair bit of deadtime. I even tried a 12R resistor in the supply rail to see if that changed anything, which it didn't. I will try and take a image of that aswell. I can't see any reason why one wave would effect the other in such a way. The only thing I can think of is when one wave turns on or off, it spikes the supply rail and becasue of that, its effecting the other wave. Its the only common thing between the 2 which I can think of. I do have a linear regulator with a range of caps to smootch and filter noise but it does not help.

There are a lot of problems going on all at once, I think if I isolate the driving wave from the power mosfets and do without the gate driver IC's then that stands the best chance of working, to prove such a thing, latter I will build a Transistor H-bridge and feed that with the squarewaves and see if that effects anything.

Chris

I expected that nobody would bother to look so thats why I didnt post a link.

I can't remember if you said what the load was or why you are using an H-Bridge - power? (eg speaker) or direction control (motor)?

What you describe above sounds slightly wrong to me. Can I check which mode you are driving the bridge? eg Unipolar or Bipolar as shown on this page..

Either way the noise spikes should allways line up with an edge - so how come they move?

is

again a dual channel scope would show that easily. Trigger off the noise spike on the supply and look for a synchronous spike on the gate drive.

then

will

Sounds like a lot of work.

Colin

At the moment im only using a resistor for aload, just keeping the inductive eliment out of it totally. I've tried motors, speakers etc, the load has almost no effect at all. Only the voltage across the bridge is having the effects.

looks like the bipolar..

the noise spikes align to the other cycle, thats what don't make sence really... the spikes do align, just look at any rise or fall and look direct under/over it on the other wave and you can see thats where the spikes are. The spikes move when you alter duty cycle, which is what I did in those 3 images.

I looked into that aswell, I can't remember offhand what the results were, but the supply rail is pretty stable, if I remember right , I dont think any of those spikes aligned to supply noise.

Well ive been trying to find a cure for months now, I've been talking to IRF about it and they don't know what the problem is either.

IMO, I think that when the mosfets have voltage on them, they are spiking thier own gates, in other words trying to pull the gate drive upto the 100V rail, as such the driver is shutting down for a split second, then comming back on again, and bascically I think thats what starting off the ringing.

If you take a squarewave out of a TL494 (or whatever) and feed it direct into a speaker you get a pure tone, though when you add a gate driver in, you get a very minor distortion in the tone and the squarewave on the output of the TL494 develops bad ringing. Its strange that when I feed the TL494 into a 74LS14 buffer, that this does not effect the wave in the slightest. But as soon as you put a gate driver with mosfet inputs on, it all messes up. Like I say I think the capacitance on the mosfets at any point, is causing the ringing, plus also acting as a fraction short to the 100V supply rail which is agrevating the problem. so then you have 2 problems, which will probably interact with each other and thats probably the end result im seeing.

I just have a gate driver on its own been driven with a squarewave now, only when you connect the gate driver IC does it start ringing, you dont even need to load the gate driver IC for it to start. So this is why I think its just down to the mosfet inputs. The only cure I can really see is just to isolate the TL494 output via a opto isolator, then feed that output into the gate driver inputs. That way the gate driver can't effect its own driving wave. I did try that, but as I just found out I am going to have a job to find some 200khz opto isolators, the ones I have sure don't work at that speed... I've been checking out opto stuff all day, theres some nice opto gate drivers (with transistor output) but they are only like 25khz... even though using opto's could cure the problem, looks like I can't use them :-( I can't really think of anyother easy way to isolate things, other than place a normal 74LS buffer between the 2.... even so I then have to do something with the gate drivers to mosfets isolation.......... now where did I put those headache pills......

Chris

its

The only other thing I can think of is the power and ground layout... I know you think thats ok but follow this line of thought.... The gate of the MOSFET is like a capacitor and it takes current to charge/discharge that capacitance. The current is supplied by the gate driver IC. Is it possible that you are getting ground/power bounce (not at the MOSFET) but before the gate driver? The cure might be better decoupling capacitors on the gate driver?

It could be power bounce, though the TL494 does not take too much power, neither does the gate driver, I wouldn't think that the gate drivers alone would be enough to spike the supply rail. I guess those few pf at 200khz could be enough to spike the rails. I've had something like 100UF,

47UF,10Uf,100NF,10NF,1NF, connected across everything supply wise, 1NF across the mosfets does *something* but it just seems to mash up the ringing.

I'm even running the TL494 on a 5V rail, and the gate drivers on a 12V rail, both running though a linear regulator which is smack next to the IC's. I'm using the entire bottom layer of the PCB for ground, everything goes though the board to that ground, aparently so ive read thats the best grouding idea...

I've even had resistors in series with the supply lines to see is that dampens the ringing any... which it doesn't.. I could try some small chokes in the supply line to see if that does anything.....

I've tried everything I can think of.... one thing which is a bit of a puzzle also, is that if you touch the mosfet it alters the ringing slightly, im not sure why that is, probably altering the load on it somehow..... and yes I have tried screening the whole thing also....

Only other thing I could try is spliting the supply rails up and running each output on its own seperate supply rail, such as running each driver IC on its own 12V rail, and the outputs of the TL494 I think you can tie them into whatever rail you want., so I would end up with 3 supply rails for the TL494, and 2 for the gate drivers..... what fun this stuff is..... whos brill idea was it to use mosfets in the first place.....

chris

>

Presumably the edges are a lot faster than 200KHz though?

I've had something like 100UF,

rail,

and there are good HF decoupling caps close to those IC's?

If you think one half of the bridge is spiking the other half - What happens if you remove the FETs from one half of the bridge?

Its hard to say, it looks like there are 2 spikes on the supply rail, though when you have a 50:50 duty cycle you can't see them, but at 50:30 you get something like a 1V spike which is bad. I didnt notice that until just. Even though the wave looks a bit crappy, the waveform does not change much, but the supply does.... it looks like a supply problem now, so I will do was I was thinking yesterday and used seperate regulators. That makes 6 of them in total.. a lot of fun....

Chris

Have you considered doing a Spice simulation of your circuit? 6 months ago I had a job to redesign a 10A PWM'd generator field-coil driver, where the original kept blowing up it's output power darlingtons.

I LTspice'd the original, and at first glance it looked all right... Ok, the output transistors did run slightly hot at 10A, but not outrageously so. Then I added-in reasonable values for the supply cabling, etc, just 10-100 milliohms or so.

In the simulation, at about 5Adc output something nasty started to happen. The output ON pulse first acquired a small notch in it. Just a few mA higher it became an actual double pulse, and a few mA above that the whole thing locked up, output transistors motorboating, out of control, huge dissipation, probable meltdown.

Fortunately I had some oscilloscope photographs of a running system and, sure enough, signs of that notch were present on one photograph, at about 6Adc output. So this gave confidence in the LTspice simulation.

Probing around the rest of the simulation showed that an earlier stage was being unexpectedly modulated by the power supply ripple, with the modulation reaching a peak effect about 2/3 into each ON period. At higher output currents that local positive feedback loop took over.

LTspice indicated that the problem could be cured with an impractical number of low ESR capacitors on the main supply rail. But also showed that the far easier solution was to simply improve the decoupling on the sensitive stage.

No breadboarding, the redesigned circuit went straight to pcb layout, and worked ok first time.... *cough*, well nearly first time, but we don't mention that perp because no re-layout was required. :)

Intelligent simulation can be a very a useful design tool.

--
Tony Williams.

Thanks for the little story, I have not done a simulation of it, its normally easier just to build the thing and see what happens... I do think supply noise is being modulated onto the driving waves, causing false turn on's etc... Mosfet's shouldn't be getting warm switching a few amps, but they get red hot with just 100mA load.. I'm designing a new PCB, every single rail is going on its one regulator now, can only help!

Chris

That's what I thought, for years. Then downloaded (the free) LTspice, and started to compare what it did on circuits where I knew what would happen. Surprisingly accurate.

Simulation is handy for power output stages because the components can be 'overstressed' without meltdown. A breadboard would just be giving off smoke in this state, but measurements can continue with a simulator.

Simulation is *cheap* compared to re-laying out a PCB in order to try something different.

--
Tony Williams.