SMPS Troubleshooting Help: Buck Circuit, audible noise, shutting off, sawtooth out

Hi,

I am using an LTC3853 which is a triple buck controller in my circuit. I have x2 5v outputs and 1 3.3v output. Vin is about 12v nominal, +/-3v on a bad day.

The one 5v output works perfectly. The second 5v output and the 3.3v output not so much. They make an audible click sound, and the inductor gets hot. The output waveform is a choppy wave slowly getting higher and higher then eventually just resetting my entire board. The layout is pretty decent and on a 4 layer PCB with 2oz copper. The main difference I see between my first 5v output and the other 2 outputs are the max current designed for. I designed the 5v that works around 1.5A maximum output. The 3.3v rail is in the 8A output range and the second 5v rail is in the 15A range. The FETs are all the same but should be able to handle that (did for ease of BOM selection).

I have changed the switching frequency, and the audible noise changes frequency as well. But it still exhibits the same problem from 200KHz to 700KHz (loose ranges).

I have changed the inductor from a big beefy inductor to a little one. The little one just makes one big "click" sound, overcurrents my powersupply, then resets the board. I changed it to another big inductor in a different inductor family that was a powdered core and stays clicking longer but gets burning hot and resets.

I should mention when I say "reset" I mean the stable 5v rail dies which then resets my microcontroller which controls the enable pins for the other outputs so it turns on waits a few seconds and tries to start back up again in a viscous cycle.

I am not sure where to troubleshoot from here.

I feel the PCB layout is adequate since I have 0 load. I would expect a crazy non-regulated waveform for a poor layout. But my layout is pretty decent with all feedback connections being kelvin connections and running together and away from switching paths.

It is hard to pinpoint where the noise is coming from, but I feel it is coming from the inductor. The fact that it gets really hot when in this sawtooth of death mode tells me that it is continuously trying to pump in the full amount of current and then nothing (or a very small amount) is being stored in the FET-off stage. Perhaps why it is slowly climbing, then the overcurrent protection of the LTC3853 activates and shuts it off?

I have desoldered all output capacitors and measured between the output and ground and got an open circuit (no load, no connection, as I would expect). I put on various known good capacitors (thinking the ones I had on there were insufficient to store the energy during the off-bits) and there was no marked effect.

Does anyone have any clues or suggestions on how to proceed from here? I have put the schematic of the 5v rail below, which is almost identical to the working rail with the only difference being the inductor itself.

**broken link removed**

Thankyou

First you should change your setup so that each supply can be activated independently from the others. While testing each supply, look at the Ith pins and the actual inductor current to make sure they track each other. Also try adjusting the soft start settings on the supply. It sounds like you may be tripping the OVP repeatedly, which would definitely cause trouble, and increasing the soft start time would help prevent that (lowering your control loop bandwidth may help also).

Also make sure there aren't any shorts on the QFN packages. They can be very tricky...

mtwieg said:

First you should change your setup so that each supply can be activated independently from the others. While testing each supply, look at the Ith pins and the actual inductor current to make sure they track each other. Also try adjusting the soft start settings on the supply. It sounds like you may be tripping the OVP repeatedly, which would definitely cause trouble, and increasing the soft start time would help prevent that (lowering your control loop bandwidth may help also).

Also make sure there aren't any shorts on the QFN packages. They can be very tricky...

Click to expand...

I can activate the 3.3v and 5v separately from the 5v that powers the MCU. The only non-controllable output is the low power 5v output which powers the control logic. And when it is on by itself, it can run stable indefinitely. Now in testing I have the large delay and then "enable" signal going to the LTC3853 from the controller for just 1 output.

I'm not sure I have a way of measuring the inductor current unless I solder it up in the air and wrap a loop around the airwire. I have tried to look at the voltage difference across the output resistor, but my oscilliscope is not isolated so I need to do a math function on 2 inputs, and I am not sure how much I trust that.

I will try the softstart. Should just be a capacitor change since it is a low current charger.

The over-voltage protection on the LTC3853, or on my bench supply? I am pretty sure my supply can take it and I have tried 2 different kinds with similar results. It should be good for 30A at 12v. Looking at VIN it stays fine even when the inductor gets hot and squeals. I wasn't aware the 3853 had OVP, I thought it only had current protection.

I am not sure how to lower my control loop bandwidth with this chip, I figured all of that was internal and not adjustable. I have no external opamps or anything of the sort.

The QFN package was my first guess, but I have double checked under the microscope for connectivity and no shorts on the visible pins. The only thing I cannot see is the exposed pad. I did take it completely off using a heatgun, cleaned it all up and put it back on and the same behaviour. I also cleaned it in a sonic cleaner just to be sure as well.

NVergunst said:

I can activate the 3.3v and 5v separately from the 5v that powers the MCU. The only non-controllable output is the low power 5v output which powers the control logic. And when it is on by itself, it can run stable indefinitely. Now in testing I have the large delay and then "enable" signal going to the LTC3853 from the controller for just 1 output.

Click to expand...

I'm saying you should eliminate the control logic and operate the shutdown/enable pins manually. Remove the microcontroller or whatever if necessary.

The over-voltage protection on the LTC3853, or on my bench supply? I am pretty sure my supply can take it and I have tried 2 different kinds with similar results. It should be good for 30A at 12v. Looking at VIN it stays fine even when the inductor gets hot and squeals. I wasn't aware the 3853 had OVP, I thought it only had current protection.

Click to expand...

The 3853 has input UVLO protection, and output OVP protection, which triggers when the output goes over the nominal voltage by 7.5%. The block diagram of the device suggests that there is some hysteresis on there, but I can't tell how much.

I am not sure how to lower my control loop bandwidth with this chip, I figured all of that was internal and not adjustable. I have no external opamps or anything of the sort.

Click to expand...

The control loop bandwidth is adjusted by the components on the Ith pins. Those pins are the outputs of transconductance amplifiers, which function as error amplifiers for each channel. you should read the datasheet section on loop compensation.

The QFN package was my first guess, but I have double checked under the microscope for connectivity and no shorts on the visible pins. The only thing I cannot see is the exposed pad. I did take it completely off using a heatgun, cleaned it all up and put it back on and the same behaviour. I also cleaned it in a sonic cleaner just to be sure as well.

Click to expand...

solder bridges can easily hide under the body of the package, making them invisible. You need to continuity check each pin to the neighboring pins, and to ground. You DID put down a test pad/via for each pin, right??