Case Study: SMPS fails at elevated voltage

[T3sl4co1l]

I remembered this case from a little while back, and thought it might be useful food-for-thought.

An engineer brings you a (failed) PCB, noting: the SMPS fails at elevated voltage.

More specifically, it's a POL buck converter:
- Nominal 24V input; the regulator is rated to 50V
- The load is modest and stable: MCU, LEDs, super basic device-control stuff; well below the regulator's current rating
- Input current is also as expected
- Failure occurs as soon as input voltage is raised to, say, 27-28V
- Nothing gets hot (or, well, I suppose the chip might've briefly when it failed, we'll never know)

As your response, give your immediate follow-up questions -- what you think is the minimum additional information required to reach a confident conclusion.  Additionally, if you think you have the solution, give it.


My response: having access to.....a bit more information than listed here -- I pondered for a few minutes, then provided a single component solution.  The engineer applied the fix and found it a complete success.

Tim

[ConKbot]

Go check it out myself because my coworkers have a knack for doing no troubleshooting at all even at the highest level. "Xyz peripheral isn't working"  examine the board and see a BGA regulator knocked off from rough handling. So yes, said peripheral isn't working, nor is any other peripheral, just the micro core supply. I'm not asking for full troubleshooting just a voltage rail check before it shows up on my bench.

[PeteH]

Regulator itself fails... What's the boost voltage rating, where is it being fed from, output voltage? (If applicable)

Assuming it has temperature shutdown.... Leads to it being EOS failure....

Monolithic?
If it's a controller, are the FETs appropriately rated?
Is it synchronous? - diode rated for current and voltage? Is the diode breaking down?

Inductor saturation is something ... But not likely if you're not already getting heating at 24V due to the ripple...pk to pk doesn't change much at 27 unless it's a knee, unrealistic hard saturation....

Input was continuously raised to 28 or set to 28V and hot plugged? Undamped input ?

Assuming that you had to ponder for a solution for a few minutes eliminates most obvious problems...

[WattsThat]

OP has the ultimate advantage. You were handed the board in question. We know only what you tell us.

You’re asking for assumptions. Foul. I cry most foul.

I guess I could play along and imagine it is an interview question. But then I’ll be 65 this year. My brain refuses to go into that mode.

[oPossum]

Enable input tied to converter supply in rather than logic level (or whatever it's allowed maximum is).

[bdunham7]

My followup question:  Does the circuit fail at elevated voltage if  you test it at full output?

If not, I propose reducing the output capacitor size.

[Phoenix]

Ring on the switching node caused by reaching discontinuous conduction?

[T3sl4co1l]

Go check it out myself because my coworkers have a knack for doing no troubleshooting at all even at the highest level. "Xyz peripheral isn't working"  examine the board and see a BGA regulator knocked off from rough handling. So yes, said peripheral isn't working, nor is any other peripheral, just the micro core supply. I'm not asking for full troubleshooting just a voltage rail check before it shows up on my bench.

The classic IT exchange: "my computer doesn't work!" "did you try turning it on?" ... "is it even plugged in?" "oh."

I guess I could play along and imagine it is an interview question. But then I’ll be 65 this year. My brain refuses to go into that mode.

Deny everything -- also a safe strategy, especially at retirement age.

OP has the ultimate advantage. You were handed the board in question. We know only what you tell us.

True, this is rather unfair.  I will reveal some more background:

The regulator was MP2456:
https://www.monolithicpower.com/en/documentview/productdocument/index/version/2/document_type/Datasheet/lang/en/sku/MP2456/document_id/6942/
Buck, 1.2MHz, 50V, 500mA, external diode, precision EN threshold for easy UVLO (the divider resistors were set appropriately, for something like 12V turn-on I think).

Input had an electrolytic capacitor; the ESR suppresses inrush overshoot.

Inductor of appropriate value and saturation current.

I don't think load range was tested, but probably it would've gotten worse as load current goes up. (Hard to test when the reg is blowing...)

Client has used this regulator and associated components, many times.   What was different on this new board?

Tim

[uer166]

First thing I'd look at is the schematic and then the layout, of which we have neither. This would be a lousy interview question.

[uer166]

Even though this is a low value discussion I'll bite: since it's an asynchronous converter check the schottky diode voltage rating and makes sure it doesn't break down causing the FET to supply unlimited current before the inductor.

[brabus]

Bootstrap MLCC voltage rating?

[Dulus]

From the datasheet:

4 EN On/Off. Pull EN above 1.35V to turn the device ON. For automatic enable, connect to VIN using
a resistor. Note that make sure the sink current of EN pin not exceed 100μA.

Enable resistor not big enough? 

[PeteH]

The layout is always different.
FB routed next to switch node.... Weak ground connection...

[T3sl4co1l]

First thing I'd look at is the schematic and then the layout, of which we have neither. This would be a lousy interview question.

It's not an interview question, it's to consider the possibilities -- and you've just found one.

The layout is always different.
...
Weak ground connection...

Ding ding ding!

You look at the layout.  You find the ground return trace to the regulator is maybe 5mm.  The VIN trace is about 20mm to the nearest capacitor.  Both traces are 10 mil width.  The board is 4 layer with inner planes, but the regulator extends out past the edge of the planes for some reason!?

What do you tell the engineer to do?

Tim

[capt bullshot]

Solder an appropriate capacitor from VIN to nearest GND plane. 20mm to the next cap is too far, causing inductive kickback from the trace into the regulator.