3
\$\begingroup\$

I'm trying to use a Semtech 5V 3A buck regulator, and there is obviously something wrong with the circuit, because whenever I try to pull current from it, the IC catches fire...

I've done buck regulators before and the schematic looks fine to me, but maybe I'm missing something obvious. I depopulated U15 and hooked up resistors to the output to draw current, and when ~100mA were being drawn the voltage started to drop until eventually the regulator caught fire. Any thoughts on causes?

Buck Regulator Schematic

Below are the different parts with digikey part #s:

  • U2: TS30013-M050QFNRCT-ND
  • L2: SRN6045TA-4R7MCT-ND
  • C4: 1276-1104-1-ND
  • C6, C8: 1276-6504-1-ND
  • C2: 1276-6736-1-nd

The full project can be found here: https://github.com/UWARG/ZeroPilot-HW/releases

Thanks!

EDIT: Below is a screen shot of the pcb layout. You can also look at the github link if you are familiar with KiCad.

enter image description here

EDIT #2: Below is an oscilloscope screenshot of the input on CH2 and output on CH1 with a 200 Ohm load (25mA). There isn't substantial noise for either signal at any frequency, no matter how much I zoomed in or out. As you can see, the output voltage has dipped to under 5V under a 25mA load.

enter image description here

5V_INT_OUT node with no load:

enter image description here

5V_INT_OUT node as it started to smoke:

enter image description here

Hopefully those oscilliscope images help, I managed to catch one of the screenshots half way through the buck regulator failing, using a 200mA load.

\$\endgroup\$
  • 2
    \$\begingroup\$ I have never seen an IC actually catching fire. Maybe some smoke, but not fire. Are you sure it's a genuine part? \$\endgroup\$ – Eugene Sh. Nov 15 '17 at 14:34
  • \$\begingroup\$ We need to see your PCB layout. Do you have a ground plane connected to the exposed metal on the bottom of the part? Is the input source stiff enough? (The schematic says VBATT so I assume it's some form of battery, but hey). Is the duty cycle on the switching node stable? Did you try fitting the optional catch diode (DCATCH)? \$\endgroup\$ – Adam Lawrence Nov 15 '17 at 14:34
  • \$\begingroup\$ I added the PCB layout. I bought it from Digi-key so I would assume the part is genuine. I was testing with VBatt = 7.5V \$\endgroup\$ – hmnbvcxz Nov 15 '17 at 15:06
  • \$\begingroup\$ Stick a scope on the top of C2 and see if Vbat is still 7.5V under some load. Also, can we see a dual trace picture of both ends of C4 (SHORT ground leads please)? \$\endgroup\$ – Dan Mills Nov 15 '17 at 15:32
  • \$\begingroup\$ How about a scope trace of the switching node (5V_INT_OUT) please show a compete cycle and use a SHORT ground clip, not one of those useless three inch bits of wire with a croc clip. \$\endgroup\$ – Dan Mills Nov 15 '17 at 18:25
2
\$\begingroup\$

Check C4 and the connections around it carefully. This sounds like the integrated switch FET is ending up in linear mode, and therefore dissipating a lot of power. C4 is part of the charge pump that creates the gate voltage for this internal N-channel FET. If this charge pump is not working, then the FET can't be fully turned on.

I once accidentally connected the charge pump cap to the wrong side of the inductor, and got similar symptoms. Mine weren't quite as spectacular, but the output voltage was regulated at low currents, and the chip got hot.

Added

When debugging a switching power supply, the first thing to do is to look at the waveform going into the inductor. That is very diagnostic, and various subtle things can be seen from its details.

Show us the waveform over at least one complete pulse with a relatively light load. That will tell us a lot. When capturing this trace, make sure to ground the scope probe properly to a solid ground node near the switching chip.

Scope trace

This is a rather annoying scope trace. You've got plenty of vertical room, yet have the gain set to 5 V/div. There is very little detail per cycle because there are too many cycles on the screen. And the worst is that the zero volt are in random places not lined up with a division, making it more work than necessary to see the voltage.

If you want someone to spend their free time looking at your problem, show some respect, by taking some care and paying attention to the details.

I'm outta here.

\$\endgroup\$
  • 10
    \$\begingroup\$ "Check the C4". That's some high quality explosive. \$\endgroup\$ – Harry Svensson Nov 15 '17 at 14:54
  • \$\begingroup\$ I checked and don't see anything wrong with it. C4 is connected just as the buck regulator datasheet says to, and it looks to be the correct value. The capacitor I used is 0.022µF ±10% 50V Ceramic Capacitor X7R 0603. The data sheet asks for 15-200nF capacitor, so this should be fine. \$\endgroup\$ – hmnbvcxz Nov 15 '17 at 15:14
  • \$\begingroup\$ What is VBat? The part specified 4.5V minimum and I would not be surprised to see funky behaviour if you are pushing that limit. The layout has significant plane area on the switching node, usually not a great idea. Is the inductor end of C4 actually soldered, I ask because it has no thermal relief, so if reflowed it may be dry, same for the Gnd end of C2 and the switching nodes on the chip (You might want to look into thermals, they make soldering easier and more reliable). \$\endgroup\$ – Dan Mills Nov 15 '17 at 15:29
  • \$\begingroup\$ I was testing Vbatt = 7.5V using a bench power supply. I just took a look under the microscope and with a multimeter and the solder joint for the inductor and C4 look fine. Why is a large area on the switching node a bad idea? \$\endgroup\$ – hmnbvcxz Nov 15 '17 at 15:35
  • \$\begingroup\$ It forms a capacitor with the ground plane which creates a resonance which can be somewhere you really wish it wasn't, if you end up with too much Q you can even blow the sand due to excessive ringing on this node. \$\endgroup\$ – Dan Mills Nov 15 '17 at 15:51

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.