Semi sporadic issue with photovoltaic driven H-bridge

[b_force]

I have been working on a pretty tricky problem the last couple of days with a H-bridge driving a DC motor.
This H-bridge contains one of those photovoltaic MOSFET drivers for the high side.

The issue is that the circuit seems to completely short every once in a while.
This is semi sporadic, although going through a couple of steps it seems to be somewhat repeatable.
The weird part is that it doesn't happen when the bridge changes switch states, but when the motor is just already going into a certain direction.

I haven't designed this circuit, a friend of my asked to help.
There is no PWM being used, the motor just switches direction after like a second or two.

There are a couple of things that seems to be already suspicious to me, but maybe some people have direct experience with these kind of things.

The impedance of the output of the photo driver seems to be rather undefined.
At this moment it totally depends on the input impedance of the MOSFET.
Also I am more familiar with this kind of circuit with a additional push-pull buffer, since the driving current of the photo driver is so weak (couple of uA).

There is no resistance between the dual MOSFETs, potential risk of oscillation (although I haven't really been able to measure this).

In general it wouldn't be my first pick for a H-bridge, since there is no isolation needed.
So the choice is either try to get this working, or completely redesign a new circuit.

Any thoughts or ideas are welcome!

[moffy]

Just a few points:
1. No snubbers across the FETs to absorb inductive transients.
2. The photvoltaic drive is very slow, ms maybe, driving the FETs, good chance of the wrong FETs being on together and you create an effective short across the supply, depends upon your dead band.
3. As you stated no gate resistors to prevent oscillations.
Use a proper high side driver or a long dead band(period when all devices are off) when changing state i.e. a. Turn off all devices b. Wait until everything is off c. Turn on new devices
Slow turn on of the photovoltaic chip can lead to excessive switching losses of the FET and possible failure.

[b_force]

Just a few points:
1. No snubbers across the FETs to absorb inductive transients.
2. The photvoltaic drive is very slow, ms maybe, driving the FETs, good chance of the wrong FETs being on together and you create an effective short across the supply, depends upon your dead band.
3. As you stated no gate resistors to prevent oscillations.
Use a proper high side driver or a long dead band(period when all devices are off) when changing state i.e. a. Turn off all devices b. Wait until everything is off c. Turn on new devices
Slow turn on of the photovoltaic chip can lead to excessive switching losses of the FET and possible failure.

Thanks,

Yeah, I basically thought about all these things.
The problem is, all those things will result in a failure when the direction is being switched.

The main problem is that the short occurs BETWEEN those switching states.

[Wolfram]

Solve the known problems before you try to analyze the failure too deeply. Circuit problems can manifest in weird and unpredictable ways, and not addressing out a known issue with the circuit because you think it would cause a different failure mode can be counterproductive.

If you want to look deeper into the problem, measure voltages and currents in the circuit (switching node voltage, motor current, upper device gate voltage keeping in mind that its source is floating) during failure. At the moment, there's not enough info to give advice beyond what has already been given, and the points you addressed yourself.

[shakalnokturn]

No experience whatsoever.

Is decoupling sufficient?
How much motor noise gets back to the H bridge?
Do you know if the shoot through occurs on both sides at once or is it the high-side that isn't meant to be conducting that latches-up?
What are Vcc and control voltages?

I'd suspect a noise related problem, possibly depending on motor load or back EMF...
Very vague idea: Depending on design layout and noise sources (again) could the opto's tiny leakage capacitance slowly build up enough charge on the high-side gates?

[b_force]

Solve the known problems before you try to analyze the failure too deeply. Circuit problems can manifest in weird and unpredictable ways, and not addressing out a known issue with the circuit because you think it would cause a different failure mode can be counterproductive.

If you want to look deeper into the problem, measure voltages and currents in the circuit (switching node voltage, motor current, upper device gate voltage keeping in mind that its source is floating) during failure. At the moment, there's not enough info to give advice beyond what has already been given, and the points you addressed yourself.

Thanks,

Yeah that is already the idea. But it's always worth a shot to ask if people are very familiar with certain circuits.

But maybe I would just advice to scrap the whole thing all together, and go for a very simple and straight forward NMOS/PMOS bridge instead.

[fourtytwo42]

In my opinion the high side gates are floating when off so liable to pickup any noise going that will likely be considerable if this is driving a motor.
You don't show the low side drivers, could they be just as bad ??
Using gate series resistors of such high value is unusual as is the lack of any gate/source resistors, given the schematic indicates poor design maybe the layout is full of issues too.

[b_force]

In my opinion the high side gates are floating when off so liable to pickup any noise going that will likely be considerable if this is driving a motor.
You don't show the low side drivers, could they be just as bad ??
Using gate series resistors of such high value is unusual as is the lack of any gate/source resistors, given the schematic indicates poor design maybe the layout is full of issues too.

Yes, I did check the low side as well, exactly for the reasons you just mentioned.

Apparently that was all just fine after measuring.
Keep in mind that we don't use a high frequency PWM signal, but just a simple on/off signal in the order of magnitude of seconds.

Maybe I can show some graphs next week.

[fourtytwo42]

Apparently that was all just fine after measuring.

Were talking source impedance seen by the gate here, the high side is abysmal, so what have you measured, if just voltage that will tell you nothing, what is the low side driver ?

[Benta]

Several issues (I've worked a lot with photovoltaic couplers, also in this application):
1: Forget the gate resistors R5 and R8. They have no use.
2: Add gate discharge resistors (a few 100kohms) from gate to source (one per side is enough).
3: The couplers you've selected are not very good, very old and obsolete. Use something like Vishay VO1263A instead.
4: Insert a lot of dead time (millisecond range). Try calculating the voltage rise time of 20 uA into 4 nF capacitance and you'll see what I mean.

[shakalnokturn]

I assume the R5, R8 resistors are only for layout considerations, used as SMD jumpers.

Interesting that you recommend gate discharge resistors, I would have thrown 1MOhm in for the sake of testing.
Is their purpose to improve switching fall time, or do the gates just tend to float up to VGSth because the opto PV side has no internal "off" resistancance to pull them low?

[Benta]

Is their purpose to improve switching fall time, or do the gates just tend to float up to VGSth because the opto PV side has no internal "off" resistancance to pull them low?

Mainly for turnoff. But without light, the PV part is just "off", leaving the gate susceptible to noise.
If you check out PV optocoupler application notes, you'll often see a second transistor optocoupler between gate and source to speed up turnoff and keep the gate at zero when inactive.

[b_force]

I assume the R5, R8 resistors are only for layout considerations, used as SMD jumpers.

Interesting that you recommend gate discharge resistors, I would have thrown 1MOhm in for the sake of testing.
Is their purpose to improve switching fall time, or do the gates just tend to float up to VGSth because the opto PV side has no internal "off" resistancance to pull them low?

I have tried a 1Mohm as well, still same problem.

I have to ask about R5 and R8, he was mentioning something like being able to use an alternative part with a similar foorpint or something.
Deadtime shouldn't be an issue, it seems to be in order of magnitude of 100ms or more (I have to double check and measure that again)

We are at a point now that I would rather recommend redoing the whole circuit I think.
Since fast switching is not a goal here I probably would even just go for a very simple NMOS/PMOS circuit probably.

He has two prototype boards. What's weird is that with one the problem seems to be a lot worse.
Which really makes me believe it's some kind of noise issue like Benta was mentioning.
Heavily dependent on parameters of certain parts, aka some kind of undefined situation.

[b_force]

Is their purpose to improve switching fall time, or do the gates just tend to float up to VGSth because the opto PV side has no internal "off" resistancance to pull them low?

Mainly for turnoff. But without light, the PV part is just "off", leaving the gate susceptible to noise.
If you check out PV optocoupler application notes, you'll often see a second transistor optocoupler between gate and source to speed up turnoff and keep the gate at zero when inactive.

As far as I am aware of, this is only with the ones that don't have active circuitry inside.
This particular one has all of that build in.

I have been searching around, but can you please provide some links to those applications notes about noise?

[CaptDon]

We used those PV devices to drive mosfets on GE locomotives for many years now. It is mandatory to have gate to source resistors (as another poster mentioned) to assure complete turn-off in a sensible amount of time. The PV's really have no discharge path so you are bound to see shoot-through failures!!!  Get something like 470k across gate to source!! And drive the LED's hard, all you have is photon coupling so want lots of them!

[b_force]

We used those PV devices to drive mosfets on GE locomotives for many years now. It is mandatory to have gate to source resistors (as another poster mentioned) to assure complete turn-off in a sensible amount of time. The PV's really have no discharge path so you are bound to see shoot-through failures!!!  Get something like 470k across gate to source!! And drive the LED's hard, all you have is photon coupling so want lots of them!

Thanks,

I tried 13mA as well a 1Mohm g-s resistor, still same issue.