Electrical – DC-DC buck regulator MP2359 sometimes damaged when input voltage is high

buckdc/dc converterpower supplyswitching-regulator

I use MP2359 (datasheet pdf) to convert high voltage into 3.3V. The input range of voltage is from 4.5V to 24V.

At low input voltages, things seem to be OK. However when I input a voltage of 21V (MP2359 allows 24V max), it sometimes failed to work when start up, the internal MOSFET seem to be punctured, and the output voltage goes to 21V as the same as the input voltage terribly, parts at the output side burned.

Here is my circuit:

schematic diagram

The output current requirement is low, about 20mA.

The input capacitor is MLCC 50V 10uF, and the output capacitor is 100uF 6.3V MLCC, they are from Samsung in 1206 package.

1N5819 is in SOD323 (0805) package, and L1 is in 0805 package.

Here is my layout:

PCB layout, without components

PCB layout, with components

So what's going on?

Best Answer

A Ceramic input cap on its own can be a very risky proposition, please tell me there is something of the few hundred uF with some ESR on VBUS somewhere reasonably close?

That 10uF input MLCC that probably has no ESR to speak of forms a high Q L/C network with the supply lead inductance that can cause a short term overvoltage on startup as it rings down I would not be that surprised to see 30+V on that node for a few microseconds if you connect a stiff power supply via a few feet of wire. The cure is a bog standard 100uF electrolytic across that 10u ceramic, to provide a lossy component to damp the resonance (An alternative is a big MLCC with a small value series resistor), or even a few tens of ohms or so in series with VBUS (You are only drawing a few mA, so the power will be tiny).

21V with a 24V abs max is really uncomfortable, I would have picked at least a 36V max input switcher if using a 21V supply, just to deal with startup transients. For reliability running anything above 75% of rating is not usually a good idea (sometimes you have no choice, but...).

Now your layout sucks, the datasheet will have a suggested layout, and at these speeds you really want to follow it, or at least respect the high dI/dt loops..

Feedback would be better from the VDD net and not directly from the end of L1, minor detail.