Buck Converter, High Output Ripple - MLCC Inductance?

[tinfever]

I'm working on a buck converter and the output ripple is larger than I was expecting from my calculations and simulations. On the output, I'm seeing the attached waveform, with sudden changes in the output voltage when the high-side turns on and off. Is this being caused by the inductance of the output capacitors?

Also, I'm seeing high current consumption at no-load too. Roughly 110mA input at 12V when the 3.3V output is unloaded. Does this sound excessive? I'm running the buck converter IC at the highest frequency possible due to space constraints, so I'm sure that isn't helping.

For context, this is using a SIC451 running at 1.5MHz, 12V input, 3.3V output. Output caps are 3x 10uF 0805 X7S, effective capacitance of 9.2uF under DC bias and ESR of 4.9mOhm at 1.5MHz. Waveform shown is at no-load. CH1 is the output, CH4 is the switching node.

[PCB.Wiz]

I'm working on a buck converter and the output ripple is larger than I was expecting from my calculations and simulations. On the output, I'm seeing the attached waveform, with sudden changes in the output voltage when the high-side turns on and off. Is this being caused by the inductance of the output capacitors?

Yes. Add some series R and L to your simulation models, and tune until you get a similar waveform.

Also, I'm seeing high current consumption at no-load too. Roughly 110mA input at 12V when the 3.3V output is unloaded. Does this sound excessive?

Buck converters often have two choices at no load. Sometimes set by a pin, sometimes an order code variation.

They can run fixed frequency, where the current recirculates, for lowest ripple but higher quiescent power.
Or they can run in variable frequency (rare pulses) at no load, for lowest power, but higher ripple

[moffy]

The waveform steps look like ESR or just resistance, no ringing and an exponential transition.

[jbb]

Hold on… if this is buck converter output, then the current won’t change instantly - there’s an inductor between the switch and Cout.

What will change is di/dt.  So stray inductance Lstray could well generate those voltage steps. Stray inductance is a bit tricky and will depend on the component and layout and probing.

[moffy]

Hold on… if this is buck converter output, then the current won’t change instantly - there’s an inductor between the switch and Cout.

What will change is di/dt.  So stray inductance Lstray could well generate those voltage steps. Stray inductance is a bit tricky and will depend on the component and layout and probing.

Could be the case, if so then lower value 1uF caps with a higher resonant frequency would help.

[kimballa]

I don't know how much I would worry about the output waveform cleanliness at no load. If there's no load, there's nothing for the signal to bother. What does it do when you have a dummy load like a power resistor comparable to your design requirements?

Testing at something like 10mA (just power an LED for testing) might be more realistic.

Also, what you have circled is not ripple. The large sinusoidal shape of the CH4 waveform is the ripple, it looks about 450-500mV based on eyeballing that graph. That could be OK depending on the output voltage and your application, or not. The part you have circled (the HF spikes) is the switching noise, which is a separate problem.

That can be reduced with an additional LC filter on the output side, with a 3db frequency <= 1/10 the frequency of the pulse (zoom in and measure it with the cursors on your scope). You may need a separate RC snubber section after that to dampen its own response.

Also, TI application report SNVA871 may be a helpful read.

[moffy]

Hold on… if this is buck converter output, then the current won’t change instantly - there’s an inductor between the switch and Cout.

What will change is di/dt.  So stray inductance Lstray could well generate those voltage steps. Stray inductance is a bit tricky and will depend on the component and layout and probing.

Just an observation, but the dv/dt on the charge cycle looks pretty linear which means the current going into the capacitors is reasonably constant, the discharge looks a bit more non linear, so it looks like for the charge cycle the current is reasonably flat therefore half a square wave.

[David Hess]

Make sure that your oscilloscope is not picking up common mode noise which will display spikes like that.  This can be done by connecting the probe tip to the same point as the ground lead connection.

[ArdWar]

I tend not to think too much on zero load output waveform. Weird things happen there, and parasitics contribute proportionally larger effect on the waveform.

The waveform looks like the square switching waveform superimposed on the usual triangle "inductor" waveform. It isn't exactly something unexpected, but if you're that concerned you might want to try other model of inductors and see if things change. That *might* be nonlinearity caused by slight inductor hysteresis. May also explain the unexpected high quiescent power consumption (besides the inherent inefficiency of using forced PWM at low load).

I'd put layout problem pretty low on the list here, simply because it's hard to fuck up a power module-style layout unless you're doing it intentionally.

[tinfever]

Make sure that your oscilloscope is not picking up common mode noise which will display spikes like that.  This can be done by connecting the probe tip to the same point as the ground lead connection.

I think you nailed it. It was measurement error. I was reading this app note from Ricktek, AN079 (https://www.richtek.com/~/media/AN%20PDF/AN079_Output%20Ripple%20Measurement%20Methods%20for%20DC-DC%20ConvertersII_EN.pdf) where they recommend soldering a twisted pair across the output caps and connecting the scope probe through that to increase the distance from the noisy inductor. This made a big difference. I'm now getting much closer to what I'd expect.



I'm thinking this is radiated noise though, since I don't actually see much when connecting the probe tip and the ground lead connection together at the output. However, if I bring the probe very close to the output caps where I was probing, but don't actually touch anything, I get this:



I have an old Tek P6046 I bought on eBay a while ago but haven't used. I guess this would be the time to use it. I'll have to do some reading to make sure I don't damage it.

Before I discovered the measurement issue, I was playing with the simulation and found I could get a similar waveform by adding 500pH and 5mOhm inductance/resistance in series with all of the output caps. I think it actually makes sense that parasitic inductance could cause this waveform because the changing di/dt direction across the caps would cause a corresponding changing in voltage. Just not the issue in this case I suppose.

[David Hess]

Make sure that your oscilloscope is not picking up common mode noise which will display spikes like that.  This can be done by connecting the probe tip to the same point as the ground lead connection.

I think you nailed it. It was measurement error. I was reading this app note from Ricktek, AN079 (https://www.richtek.com/~/media/AN%20PDF/AN079_Output%20Ripple%20Measurement%20Methods%20for%20DC-DC%20ConvertersII_EN.pdf) where they recommend soldering a twisted pair across the output caps and connecting the scope probe through that to increase the distance from the noisy inductor. This made a big difference. I'm now getting much closer to what I'd expect.

I have some RG179 pigtails which plug directly into my oscilloscope probes for these types of measurements.  The extra capacitance and loss of bandwidth are usually irrelevant when testing switching power supplies.  They provide a low impedance ground connection to the probe with a minimum of loop area for picking up magnetic flux, and the coaxial shield prevents electrostatic coupling.  Twisted pair works works well also.

A short length of coaxial cable plugged into a x1 or x10 probe works better in my experience than a longer length of cable, so I do it this way unless I am working with 50 ohm signals.