LM5106: Ringing/oscillation issue

Hi Richard, 

Our expert on this device is currently out of office for the Thanksgiving holiday. Please expect a response early next week. Thank you! 

Vivian

Thanks Vivian.

Hi Richard,

Thanks for reaching out to us.

Did you try removing D4 and D5? If fall time ends up being too slow then you may try reducing the gate resistor from 10 Ohm to 2.2 Ohm or whatever suits/meets the design requirment.

Please let us know and we can take a look into it in more details.

Thanks and regards,

Ritesh 

Hello Ritesh,

Thank you for your quick reply! I tried shorting R5 and R6, of course effectively removing D4 and D5 as well.

In addition, I mounted small HF bypass capacitors (100nF || 1nF) parallel to the large 10uF capacitors, in particular C1 and C12 (which have a self-resonant frequency in the 4-5MHz range).

Unfortunately, this did not make any difference:

The main question is why the positive-going edge is so much steeper at these high (or low) duty cycles than somewhere in the middle range. It would appear that these shorter low-going phases in the 100-200 ns range somehow cause resonance, causing the positive-going edge to rise much faster than would otherwise be the case (see the last image of my first post).

Just for the sake of completeness, here's an image of the relevant part of the PCB:

Here's the same area, now just the bottom layer:

It's a solid ground plane with cut-out traces for the high-side MOSFET gate (left) and the HS signal.

Long ago, I experienced similar problems due to cut-outs in a GND plane, effectively causing resonant loops - but those were much larger, and could be fixed easily by making the cut-outs smaller. These traces are so short that the encompassing GND loop should not produce resonance in the 100MHz range.

But just to make sure, I laid two small GND bridges across both of these traces in the bottom layer. As expected, it did not make any difference at all.

Hi Richard,

By shorting R5/R6 you did effectively removed the diode effect, but it would not help reduce the ringing because now you increased your turn-ON speed as well.

Thus, I would again recommend that if you dont need fast rise/fall time of the power switch then remove those gate diodes, keep the gate resistor. This will slow down the gate and should reduce the ringing.

Apart from switching speed, the gate drive loop inductance and power switch parasitic elements are the key elements contributing to the high frequency ringing. By reducing the gate drive loop we can reduce the ringing but it is iterative process and many a times not feasible (depending on the stage of the project).

My colleague, William, may also add some of his comments here.

Regards,

Ritesh

Hello Ritesh

Shouldn't the diodes only affect the switching off of the MOSFETs? The problem occurs when the high-side MOSFET is switched on, not off - and only the high-side MOSFET, not the low-side one.

Also, I can't possibly make the gate drive loop any shorter - it is as short as it can be. Laying a wire from the MOSFET switch output directly to C11 (bypassing the slightly longer loop to the right) did not make any difference.

Anyway, just to make sure, I restored the 10 Ohm gate resistors and removed the diodes, and as I already expected, this did not change anything. This is what the signal looks like at 95% duty cycle:

And as before, the problem goes away when the duty cycle is moved away from the highest (or lowest) values. This is what 73% duty cycle looks like, with the ringing almost gone and a rising edge that isn't as steep:

It almost looks like that at small (either high or low) duty cycles, so pulses <200 ns or >3.8 us, the LM5106 has a much increased gate drive capability as compared to mid-range duty cycles. So maybe the bootstrap capacitor C11 has something to do with it?

However, first I tried increasing the gate resistors to 47 Ohm to see what that would do. Unfortunately, this destroyed the high-side MOSFET almost instantly, even with 500 mA current limit on the 24V supply.

After replacing the dual MOSFET, the LM5106 and restoring the gate resistors to 10 Ohm (still without the diodes), I increased the dead time resistor R4 to 100K, just to be on the safe side - this should provide some 550 ns of dead time to protect the MOSFETs during experiments; for good measure, I also increased the bootstrap capacitor C11 from 100nF to 1uF, so that gate drive current should not be a problem either.

The result was quite unexpected:

Not only does switching on the high-side MOSFET still cause the same ringing as before, and with the same weird edge behaviour as before, there is now also a very strange half-sine-like waveform preceding the actual switching on.

I have no idea what to make of this, except that it must have something to do with the LM5106 supply or boost circuit. And oh, increasing the duty cycle from here eventually destroys the high-side MOSFET again, probably because of excessive power dissipation due to being half switched on for a lot of the time. Something clearly is very wrong, but I can't figure out what it is.

So I'm pretty much at a loss here. I think I tried everything that I could, but all my changes so far either did nothing or even blew things up. The strange thing is that I used the LM5106 before on several occasions in almost the same configuration, except that I used single MOSFETs then, with a larger input capacitance, a larger inductor and a rather different PCB layout (one of which caused ringing too, but that was easily fixed).

Does anyone have any ideas what may cause this strange behaviour?

Hey Richard,

I am a colleague of Ritesh's and wanted to add some insight here.

How are you taking these measurements to see the ringing?
In situations where we see excessive ringing, we advise using a tip and barrel probe at the pins of the driver as well as setting the oscilloscope to AC coupling and full bandwidth mode. This allows you to capture the ringing accurately to ensure what is happening in the circuit is what you are seeing.

Are you able to add a bootstrap resistor and see if this improves this ringing and overshoot? One possibility is that you are seeing overcharging of the bootstrap capacitor in the low duty cycle scenario. That doesn't answer why this occurs during the high duty cycle scenario too but it could provide insight on if the bootstrap capacitor is overcharged.

Let me know what your findings from my questions are and if you have any further questions.

Thank you,

William Moore

Hi William,

Thank you for joining this rather tricky puzzle session!

I am using the standard oscilloscope hook probes with ground clips, with small bits of shellac-insulated transformer wire soldered to the actual measuring points. All the ground clips are clamped onto one thick bit of GND wire soldered to one of the GND connections shown in the PCB image above.

This is what it looks like from above:

Yes, I know that a tip and barrel probe method gives much cleaner results, but since I have so far failed to realize the electronics engineer's dream of growing several extra sets of hands, I can only measure one signal at a time that way. And even then, manipulating the controller input and the oscilloscope takes quite a bit of juggling.

Anyway, I changed C11 (C-bootstrap) back to 100nF again, and put a smallish bootstrap resistor in series. At 10 Ohm, nothing changed, with the oscilloscope picture still identical to the last one. With R-bootstrap=100 Ohms, the ringing diminished somewhat, but everything else remained the same, including the strange hump during what should be the high-side MOSFET's dead time:

Now I could of course reset the dead time to some more sensible value by decreasing R4 to 10K again - this would no doubt eliminate the hump, but then I'm back to square 1. Then again, at least things should no longer blow up then, and I could also measure the ringing frequency more precisely then - although I don't know if that is very informative.

The main question that I think I need to answer is what causes the edges of the high-side drive to become much steeper at very low and high duty cycles, leading to oscillation, while returning to a far more gentle slope in between. And this is probably also related to this mysterious hump - which looks a bit like part of a 2-3 MHz signal, and that is roughly the same as the resonance frequency of the 10 uH inductor and the combined MOSFET output capacitance (380 pF).

So maybe these short pulses (both at high and low duty cycles) cause the MOSFET switch + inductor to resonate at a few MHz, driving current back through the top MOSFET source capacitance to the gate, forcing a steep turning-on? Then again, I never saw anything like this happen before, not even in my more Frankensteinish creations, with neighbours complaining about TV interference ...

Another observation: I measured the RF ringing (see below), and introduced extra capacitance from the MOSFET switch output (HS) to GND to see if that made a difference.

This is the HF ringing without the extra capacitance, as measured on the IN(!) signal, so the bottom trace in the above oscilloscope images:

And this is the same IN signal, now with a 470pF ceramic capacitor from HS to GND:

The amplitude is somewhat lower, but the frequency has essentially not changed. Which means that the HF ringing is not dependent on the capacitance at this point, but seems to come from the LM5106.

However, the 'hump' does appear to be notably wider, so this lower-frequency resonance does stem from the LC combination at the MOSFET switch output:

The puzzling thing is that literally nothing I tried so far made any difference to the core problem. I tried increasing (and diversifying) capacitors on the +24V and +12V supply lines, introduced extra GND bridges across the two bottom layer traces, and tried different gate resistors, a bootstrap resistor and several small capacitors at different places just to see what would happen. During all this, the RF ringing stays a solid 100 MHz signal, only slightly suppressed by capacitors, but not changed in any other way, let alone eliminated.

My experience tells me that the most likely cause is still the PCB geometry, but I have no idea how that can be improved - I kept everything as close together as possible, with a closed GND plane at the bottom except for two short traces.

Hey Richard,

It looks like you could be experiencing a "quiet" switching condition where the power train is soft switching or mostly soft switching. This occurs where the drain of the MOSFET transitions or mostly transitions from low to high during the dead time.

For the ringing on the waveforms, it looks like hard switching is occurring when you turn off the low side FET. This occurs when the body diode starts conducting (switch node stays low) and when you turn on the high side, the switch node dV/dt is a lot higher due to forcing the body diode of the low side FET off.

For softening this hard switching, you can increase turn on resistance of the high side FET or utilize a FET with faster body diode recovery.

Let me know what your findings on this are and if you have any further questions.

Thank you,

William Moore

Hello William,

First of all, and once again, my thanks for your continuing support and input. I really appreciate it, even though so far, it hasn't brought me any closer to a solution.

I don't quite understand how this 'quiet switching' or 'soft switching' works; I'll see if I can find more information on that.

William Moore said:

For the ringing on the waveforms, it looks like hard switching is occurring when you turn off the low side FET.

I'm afraid I don't share your analysis here - quite the contrary: as far as I can see, waveforms do not show any relationship between switching the low-side MOSFET and ringing in the high-side MOSFET. Here are two traces that I showed earlier, one with a short dead time and a high duty cycle, and the other one with a very long dead time and a low duty cycle:

Trace #3 (magenta) is the low-side MOSFET gate signal. In either case, switching this gate on or off hardly causes any visible ripple in the other signals. (The slight ring in trace #4 upon switching off the low-side MOSFET is an HF artefact picked up through the ground loop, and disappears when using the tip-and-barrel method.) However, when the high-side MOSFET turns on, this causes a strong ringing on all lines.

Also, when switching the low-side MOSFET causes the ringing, then shouldn't this happen all the time, regardless of the duty cycle? Because there are no problems at all when the duty cycle is somewhere between 25% and 75%, like here:

It is quite strange to see the high-side MOSFET drive signal changing from very steep (and ringing) to more gently sloping with increasing duty cycle, and back to very steep again at the end - I could try shooting a video of it, if that would give you people a clearer picture of what I see happening.

About the body diode: reverse recovery time is given in the datasheet as 29 ns (starting from 19 A, with dI/dt = 100A/us), which is faster than in most other MOSFETs that I used without any problems. So I don't think that a slow body diode is causing this. But I'll first dive into the mentioned phenomenon of quiet/soft switching, to see how that works. Do you happen to have any links to development notes about this?

Anyway, trying another MOSFET or increasing its on-resistance would require creating and ordering a new PCB in order to accommodate the new device, and that would take at least a couple of weeks, or more likely even a month, what with the holidays coming up.

BTW, I could duplicate the circuit and physically send it to you for testing, if this is something that you do - I still have a couple of boards in stock. Yes, this is something that I really should figure out for myself, but I must admit that this problem has me stumped.

But thank you again, regards,

Richard