Buck converter fails instantly when connected. Odd input ringing?

[tinfever]

I'm helping someone debug a basic buck converter that explodes when plugged into a battery pack, and I'm wondering if anyone has seen the ringing/clipping shown in the attached picture before.

The buck converter board takes power from a 16 cell li-po battery pack at ~65V, has an output of 30V at 2A, and uses an TI LMR38025 converter IC. The board works fine when powered from a bench PSU at 60v, but the converter IC explodes and decaps itself instantly when the connector for the Li-po pack is plugged in. This has been reliably reproduced.

I suspect that the inductance of the battery pack wires and the buck input MLCCs (no bulk caps) are ringing up to double the supply voltage, around 130V, which exceeds the rating of the converter IC, so the electricity probably arcs over within the IC and causes the IC to fail.

I attempted to measure the ringing on the scope and found a really odd waveform (see attached). Instead of ringing, the voltage of the input caps goes up to about 110V and is clipped there for 400us, before falling down to slightly below the battery voltage, staying flat there briefly, then ringing and stabilizing to the battery voltage. My best guess for this is that the failed buck IC is clamping the voltage to 110V somehow. Although, when I tried applying 120V DC on the input of the buck, no current was drawn.

To fix this issue right now we are use an NTC inrush limiting thermistor, however this is bad for efficiency since it will burn 0.2 to 0.5W under normal usage. In simulation, a 75V TVS diode is unable to clamp the voltage to a safe level since the current is so high. I'm thinking either a P-FET soft start circuit before the caps or a hot-swap IC might be the only good solutions.

Does anyone know why might cause this weird waveform? Any other theories on what could cause this sort of instant failure? Any other suggested solutions?

Thanks!

[T3sl4co1l]

Standard points apply:
- What is schematic?
- Beware of low ESR / lack of damping: https://electronics.stackexchange.com/questions/713381/correct-placement-of-series-ferrite-beads-to-avoid-dc-disconnect-during-power-cy/713473#713473

Without the input circuit, the square wave shape is strange, but neither here nor there.

Tim

[moffy]

A li-po battery pack can deliver a higher current pulse than your usual PSU, which can then charge any inductance in the leads. If putting an NTC in series solved the issue then you have pretty much proven that. Reduce the inductance, twist the leads together coming from the battery pack, and place an RC snubber across the input to try and absorb the ringing and maybe a TVS as well. The ESR in electrolytics can have a similar effect.

[PCB.Wiz]

....In simulation, a 75V TVS diode is unable to clamp the voltage to a safe level since the current is so high. I'm thinking either a P-FET soft start circuit before the caps or a hot-swap IC might be the only good solutions.

Does anyone know why might cause this weird waveform? Any other theories on what could cause this sort of instant failure? Any other suggested solutions?

There is not much margin from 60V battery to 80V device, so a TVS will struggle.
For testing, you could use a pre-charged large cap and a clamp diode, so at least you do not decap many ICs !

A hot swap or P-FET sounds a good pathway.
P-FET would need gate help, to give faster gate removal so contact bounce or noise still gives a proper soft start.
You might want to include proper current limit inrush too.

Or hot swap controllers like Richtek RT1720GF can work with an external NFET, and manages current and voltage limiting too.

[ArdWar]

Is that waveform actually from input node? That waveform is almost exactly how buck converter switch node looks like during pulse skipping. Except the voltage scaling of course.

[magic]

The buck converter board takes power from a 16 cell li-po battery pack at ~65V, has an output of 30V at 2A, and uses an TI LMR38025 converter IC. The board works fine when powered from a bench PSU at 60v, but the converter IC explodes and decaps itself instantly when the connector for the Li-po pack is plugged in. This has been reliably reproduced.

I attempted to measure the ringing on the scope and found a really odd waveform (see attached). Instead of ringing, the voltage of the input caps goes up to about 110V and is clipped there for 400us, before falling down to slightly below the battery voltage, staying flat there briefly, then ringing and stabilizing to the battery voltage. My best guess for this is that the failed buck IC is clamping the voltage to 110V somehow. Although, when I tried applying 120V DC on the input of the buck, no current was drawn.

Does anyone know why might cause this weird waveform? Any other theories on what could cause this sort of instant failure? Any other suggested solutions?

I would guess that it's a combination of high V and high dV/dt which triggers some sort of latchup in the IC, then it briefly pulls battery voltage low while vaporizing itself, then it's done vaporizing and everything is back to normal, except the chip is gone

You know what to do.

[tinfever]

Thank you for all the great input!

Standard points apply:
- What is schematic?
- Beware of low ESR / lack of damping: https://electronics.stackexchange.com/questions/713381/correct-placement-of-series-ferrite-beads-to-avoid-dc-disconnect-during-power-cy/713473#713473

Without the input circuit, the square wave shape is strange, but neither here nor there.

Tim

That link was very informative. I didn't realize that the non-linear aspects of MLCCs could actually make the peak voltage even higher.

Sorry I didn't include the schematic. It isn't my design and I haven't requested permission to publish it from the designer. Although, it is just a basic TI webench implementation of the converter with an input and output connector. There are two 2.2uF 100V 1210 and one 0.1uF 100V 0603 caps on the input.

I would guess that it's a combination of high V and high dV/dt which triggers some sort of latchup in the IC, then it briefly pulls battery voltage low while vaporizing itself, then it's done vaporizing and everything is back to normal, except the chip is gone

You know what to do.

Interestingly, that waveform capture I posted was when I connected the li-po back to a board with a previously blown IC. That waveform is repeatable on reconnecting the same board, albeit sometimes the waveform is different.

Is that waveform actually from input node? That waveform is almost exactly how buck converter switch node looks like during pulse skipping. Except the voltage scaling of course.

Yes, that's the input node. I thought it was odd too. If I have time, I should try removing the blown IC to see if I still get that waveform. That would rule out a measurement error, even though I don't know how that could be.

[Someone]

There are two 2.2uF 100V 1210 and one 0.1uF 100V 0603 caps on the input.

Even if the batteries were soldered onto the same PCB as the converter I doubt I'd include such little bypassing on the input. The evaluation board should be the reference if you aren't calculating all the messy details.

[PCB.Wiz]

Yes, that's the input node. I thought it was odd too. If I have time, I should try removing the blown IC to see if I still get that waveform. That would rule out a measurement error, even though I don't know how that could be.

It's certainly unexpectedly square - usually they are damped ringing, and the ringing is related to the rise time.

You could try adding a resistive load of roughly the expected current, after you remove the IC.

Spice suggests even a constant current feed can drop the overshoot significantly.
A PNP power device + simple base resistor can do that to HFE precisions, but the 60V & low amps region does bump you into larger die PNP parts, > 100V.

[ArdWar]

Yes, that's the input node. I thought it was odd too. If I have time, I should try removing the blown IC to see if I still get that waveform. That would rule out a measurement error, even though I don't know how that could be.

Calling it odd is quite the understatement tbh. Supply rail shouldn't move that much, and if it moves it shouldn't be that square.

How your design even allow that to happen anyway? It implies your supply rail is high enough impedance that it allows switching node load to pull it down, or worse, fed back somehow. Even then your input capacitance still shouldn't allow that high of a dV/dt.

[duak]

Does the battery pack have some sort of Charge Management System? It seems to me that it is not accepting reverse current from the buck converter, as if it has an internal diode to prevent charging.  I wouldn't think a buck converter would reverse its input current but the charge pump and a sneak path from the output may source enough to do so.  Are you able to to scope the input current and check for reverse current?

It occurred to me that if the load is sourcing current, even for a short period of time, the buck converter has to pump the energy on the output side of the circuit back to the input side.  This could happen if the control loop decides the output voltage is too high.  This could happen if the output capacitor is quite large, the load current is low or zero or the output voltage overshot the set point during start up.  The load isn't a battery, is it?

Others have suggested more bulk bypassing and a TVS or something to limit input voltage.

[PCB.Wiz]

Does anyone know why might cause this weird waveform? Any other theories on what could cause this sort of instant failure? Any other suggested solutions?

I slept on this, and decided it was hot plug contact bounce and a light initial load.
See the attached spice plots, that produce a very similar waveform.

To fix this issue right now we are use an NTC inrush limiting thermistor, however this is bad for efficiency since it will burn 0.2 to 0.5W under normal usage.

The power loss is a pain, but it is hard to go past the simplicity of a NTC or power resistor.

I'm thinking either a P-FET soft start circuit before the caps or a hot-swap IC might be the only good solutions.

Nexperia have N-FETS designed for hot-swap, like the PSMN013 series, and with large SOAR, PFETS spec'd for hot swap are less common.

Another fish hook is the simplest MOSFET circuit does not behave well with contact bounce, as the gate C may be still charged, when a fresh edge arrives.

PNP inrush limiters are better behaved, but finding small, low cost PNP parts with a decent SOAR is also a challenge, plus they have base drive power loss.
Audio bipolar parts have good HFE and SOAR, but come in larger packages.

There are LED Linear current drivers with 100mV or even 50mV reference voltages, that could construct an inrush current limiter.

 

[amyk]

Some PSUs with an NTC have a relay that bypasses it after around a second of being turned on, to stop it from continuing to waste power.

[tinfever]

I slept on this, and decided it was hot plug contact bounce and a light initial load.
See the attached spice plots, that produce a very similar waveform.

Wow! Thank you for putting that together!