SIC451 Buck Converter Instability

[tinfever]

I have a project with two SIC451 buck converters, the output on both of them is oscillating, and I'm trying to figure out why.

The regulators are 12V to 5V and 12V to 3.3V at 15-25A. The design is very space constrained so the switching frequency is set as high as possible at 1.5MHz to minimize the number of output caps and the inductor size.

The oscillation is 1.8v p-p at 100kHz around the output voltage setpoint of 5V. My calculations indicated the minimal output capacitance I used should have been enough to meet my ripple target, taking the DC bias effect into account. I've found that adding more output capacitance makes the regulators stable, although the 5V regulator is still being stubborn. On the 5V regulator I've increased the output caps from 2x 10uF to 11x 10uF 0805 X7S MLCC caps and it will still sometimes start oscillating after a 1A to 20A load step at 2.5A/us. Sometimes it will start oscillating and then stop after a second. Before adding the additional caps, both regulators would start oscillating immediately after turn-on. The 3.3V regulator was stable after increasing the output caps from 2x 10uF to 5x 10uF.

It's odd because the 2x 10uF caps I started with were enough to hit my output ripple target. I can tell because I could connected a 2200uF aluminum polymer across the output to stabilize it, then remove it, and the output ripple would be fine until it started oscillating a second or two later again. The datasheet even says "The device has integrated internal compensation and is stable with any type of output capacitor."

In my initial testing with just 2x 10uF output caps, I found adding the 2200uF 8 mOhm ESR aluminum polymer cap across the output would make it stable, or adding several 10uF MLCCs would make it stable, but a 68uF 250mOhm ESR electrolytic didn't help at all. This implies a low ESR is necessary.

When I power the board from the big high current PSU that will be used in the actual application, rather than my bench PSU, the board seems more stable.

Has anyone seen anything like this before? Is it just insufficient output capacitance? Could this be some weird interaction with my bench PSU and poor input wiring powering the board? Could it be insufficient Pvcc internal gate driver capacitance or something odd like that?



Attachment notes:
In the schematic, you'll see the SIC451 supports differential remote voltage sense and there is an appnote that recommended a filter to reduce jitter. This is the 1R and 2.2uF shown on the 5V output and GND. I have removed the 2.2uF capacitors in case they were causing some sort of feedback issue but there was no improvement. The remote sense lines are currently disconnected so the feedback is coming through the 220R resistors. The 47uF and 270uF caps shown as DNP on the schematic are not populated. The 270uF  aluminum-polymer caps would fix the oscillation issue but they'd have to go on the back of the board due to lack of space, and double-sided SMD load = higher cost. The 47uF caps were 0805 tantalum-polymer caps but their ESR is too high to fix the issue/symptom.

The attached scope shot shows the output starting to oscillate after a load step. This is the 5V regulator with 8x 10uF MLCCs on the output.

In the layout, layers 1 and 3 are shown. Layer 2 is a solid ground plane, not pictured. Pins 11 and 12 are the feedback differential sense traces which snake around the board as a pair, through the previously mentioned remote voltage sense filter and pull resistor, and to the output connector.

[uer166]

When I power the board from the big high current PSU that will be used in the actual application, rather than my bench PSU, the board seems more stable.

Did you scope the input voltage? It's possible than your bench PSU is the one being unstable, since it's feeding another PSU whos' control loop is likely much faster than its' own. Normally you want the upstream converter to have a faster response than any downstream converter, although that's obviously not always possible.

Another options is that you're seeing pulse skipping due to overcurrent during the load step. Add a shunt or current probe to the inductor, and check against the waveform in Fig. 7, it's possible that it simply is skipping multiple PWM cycles while waiting for the current to decrease.

[tinfever]

Did you scope the input voltage? It's possible than your bench PSU is the one being unstable, since it's feeding another PSU whos' control loop is likely much faster than its' own. Normally you want the upstream converter to have a faster response than any downstream converter, although that's obviously not always possible.

Another options is that you're seeing pulse skipping due to overcurrent during the load step. Add a shunt or current probe to the inductor, and check against the waveform in Fig. 7, it's possible that it simply is skipping multiple PWM cycles while waiting for the current to decrease.

Thank you for the suggestions.

I've gone back to a 5x 10uF MLCC configuration on the output of the 5V regulator (down from 11x 10uF), so now the regulator starts oscillating immediately after startup, and I've done further testing.

(All scope shots are under no load other than quiescent current from some on-board components, 50mA max.)

This scope shot shows the oscillation starting after the startup ramp finishes. I'm not sure why it would be able to ramp fine and then only oscillate afterwards. Powered from bench PSU.
(Yellow = V_out, Green = V_in, Blue = I_in)


Same as above but zoomed in.


Switching node (green) and V_out (yellow). It seems to stop switching when the output voltage gets too high, which I think makes sense as the datasheet mentions (page 9, Pre-Bias startup) "If the sensed voltage is higher than VSET, control logic prevents HS and LS FET from switching."


Adding a 1000uF electrolytic or 2200uF aluminum-polymer capacitor to the input yielded only a minor improvement in reduced amplitude of oscillation to 1.4V p-p from 1.8V p-p and the oscillation amplitude was less consistent. Similarly, connecting the high current power supply for this application (which has its own on-board bulk capacitance) directly to the board reduced the oscillation amplitude further to 1.1V p-p. (pictured is high current PSU and no added capacitance)
(Yellow = V_out, Green = V_in)


Moving the 1000uF electrolytic to the output made it sort-of stable, in that it sometimes oscillates for less than 1ms after startup before stabilizing to the normal <30mV p-p ripple I'd expect. Sometimes it starts up without oscillating at all.
(Yellow = V_out, Green disconnected)


From all of this, I'm not sure it is an upstream power supply control loop issue since it oscillates at no load and seemingly regardless of the amount of input capacitance I add. Even if that is the issue, I can't modify the upstream power supply so I have to work around the issue somehow.

The fact that adding output capacitance makes a dramatic improvement, makes me think the regulator or my implementation is likely at fault. I'll check the inductor current waveform next.

[uer166]

Your Vin is obviously oscillating severely and either doesn't have a low enough impedance for a converter of this power level, or it itself has a control loop interaction with your converter.

Your buck is likely fine, get a lower impedance source such as a battery and test with it, I bet it'll be stable in all loading conditions.

[tinfever]

Your Vin is obviously oscillating severely and either doesn't have a low enough impedance for a converter of this power level, or it itself has a control loop interaction with your converter.

Your buck is likely fine, get a lower impedance source such as a battery and test with it, I bet it'll be stable in all loading conditions.

I see your point. The input voltage certainly isn't steady. The high current PSU I was testing with in the fourth screenshot in my last post was a 12V 62.5A power supply with 3600uF of electrolytic caps on its output. It was directly connected to my board with an edge connector. I'd think it would be hard to get lower impedance than that, at least at DC and low frequencies. I guess it's possible both my bench PSU and this high current PSU are having control loop interactions with my converter, but how could I fix that? Lower the switching frequency on my converter to lower the speed of its control loop? Add a filter on the input of the converter to sort of isolate it from the input PSU?

I see someone else's design using the same IC, half the number of input caps that I have, and connected over two feet of wire to the input PSU, but they have much more output capacitance, in both MLCCs and bulk. They are probably running at a lower switching frequency for sure. I'm trying to figure out why that design is stable enough to put into production but mine obviously isn't.



I soldered a wire loop in series with the inductor to attach a current probe, and somehow the regulator is now magically stable at startup, still with the 5x 10uF output caps, powered from the bench PSU. I have no idea why. Extra resistance or inductance of the wire providing more feedback for the controller? Effectively reflowing the converter IC fixed some partial short under the IC?
(The current probe only has a bandwidth of 2.5MHz vs the 1.5MHz switching frequency, which isn't ideal.)
(Green = V_out, Blue = inductor current)


Testing with a fast load step (1A to 19A at 2.5A/us) still starts the oscillation, although it goes away when I remove the load. A slow load step (0.001A/us) doesn't start it. Some of the drop in V_in shown here is due to the small wiring from the bench PSU dropping about 1V.
(Yellow = V_out, Green = V_in, Blue = inductor current)


I started testing powering the board from a li-ion battery pack with bizarre results. The pack was three cells at 10.8V total. I think I might have been hitting the UVP on the converter IC so I'll do more testing at a higher pack voltage, although I don't think this is causing the odd perturbations shown. I could be doing something wrong with the probing here. I need to think about it more.
(Yellow = V_out, Green = V_in, Blue = inductor current)

[uer166]

Add a filter on the input of the converter to sort of isolate it from the input PSU?

That'll probably make it even less stable, see: .

The waveforms with the li-ion pack are interesting. The input is stable, but output has occasional oscillations that do look like control loop instability. Have you considered doing a loop magnitude/phase measurement to get the system phase/gain margin? I think it's a linear control loop so it should be valid. It's internally compensated though, so not sure what you can do to fix it if that's the issue.

[SiliconWizard]

I have looked at the datasheet and don't understand the way you have connected Vsen+ and Vsen-. That doesn't look correct to me.

[tinfever]

The waveforms with the li-ion pack are interesting. The input is stable, but output has occasional oscillations that do look like control loop instability. Have you considered doing a loop magnitude/phase measurement to get the system phase/gain margin? I think it's a linear control loop so it should be valid. It's internally compensated though, so not sure what you can do to fix it if that's the issue.

I've done some more testing powering the board with two three-cell Li-ion battery packs in parallel to reduce the input impedance as much as possible, including adding a 2200uF aluminum polymer cap to the input of the buck converter. The converter definitely still oscillates under these conditions. I think this rules out the input supply as the cause of the oscillation. Perhaps this regulator is just not designed to be stable with this little output capacitance? Or maybe it needs a higher inductor ripple current?

Unfortunately I don't have an injection transformer to attempt to measure the system gain and phase margin.

I have looked at the datasheet and don't understand the way you have connected Vsen+ and Vsen-. That doesn't look correct to me.

There are few different things going on with Vsen+ and Vsen-. The datasheet shows them directly connected to the output, however they have an appnote (https://www.vishay.com/docs/66843/anpowersupply.pdf) which suggests adding an RC filter to improve transient response and jitter when using long lead for the remote voltage sense, which applies to me. This is where the 1R + 2.2uF components come from. My calculation of the values could be suspect though. The board also needs to be able to operate with the remote sense wires disconnected, which is what the 220R resistors allow. If the remote sense wires are disconnected, the feedback will come through those resistors, however if the remote sense wires are connected, that signal will dominate. Vishay told me that the internal impedance between the Vsen+ and - terminals was 33k +/- 30%, indicating the 220R resistors should induce minimal error.

To rule out these parts as potential causes of the oscilation, I removed the 2.2uF caps very early on. Then I just now removed the 220R resistors and shorted on their terminals, with no change. The remote sense wires have always been left disconnected.

I've attached a schematic of the current circuit configuration for clarity.

Dual li-ion battery pack powering board at roughly 12.3V. 1A to 12A load step at 2.5A/us. Some oscillation of output.
Yellow = Vout, Green = Vin, Blue = inductor current


Exact same configuration, nothing changed, sometimes it will be stable after a load step like the last image, sometimes it does this


Same configuration but with 2200uF aluminum-polymer cap on input. Vin doesn't undershoot as much, but oscillation can still occur.


Zoomed in on start of oscillation same li-ion battery + 2200uF input configuration. I think you can almost see how the converter control is unstable, where it skips a pulse, then undershoots, then overcompensates, then skips pulses for longer so it undershoots further, etc.


Now stable with the 2200uF cap is moved to the output. I cannot get this to oscillate. Pictured is a 1A to 25A load step at 2.5A/us. Those small wiggles during the recovery from the undershoot still concern me greatly. Does this mean it is still borderline unstable?

[Someone]

There are few different things going on with Vsen+ and Vsen-. The datasheet shows them directly connected to the output, however they have an appnote (https://www.vishay.com/docs/66843/anpowersupply.pdf) which suggests adding an RC filter to improve transient response and jitter when using long lead for the remote voltage sense, which applies to me. This is where the 1R + 2.2uF components come from. My calculation of the values could be suspect though. The board also needs to be able to operate with the remote sense wires disconnected, which is what the 220R resistors allow. If the remote sense wires are disconnected, the feedback will come through those resistors, however if the remote sense wires are connected, that signal will dominate. Vishay told me that the internal impedance between the Vsen+ and - terminals was 33k +/- 30%, indicating the 220R resistors should induce minimal error.

To rule out these parts as potential causes of the oscilation, I removed the 2.2uF caps very early on. Then I just now removed the 220R resistors and shorted on their terminals, with no change. The remote sense wires have always been left disconnected.

Positioning of and frequency response back through the feedback path becomes critical with increasing switching frequency. The routing in your first post is far from ideal and you've not shown the whole layout of that.

[tinfever]

Positioning of and frequency response back through the feedback path becomes critical with increasing switching frequency. The routing in your first post is far from ideal and you've not shown the whole layout of that.

Sorry. I've attached an image of the entire layout of that section, highlighting the feedback path for Vsen+ and -. The remote voltage sense wires would go to pins 8 and 17 on the bottom connector. If those aren't connected, the feedback comes from pin 6 for the Vsen+ and comes from a via to ground for Vsen-, both through the previously installed 220R resistors shown at the bottom of the image.

You mention the layout isn't ideal. How could I improve it? Some non-idealities are due to space constraints (like the odd upper input capacitor trace or the bypass cap positioning on pin 24). I know the feedback path is somewhat close to the switch node on the lower 3.3V converter but I once tested the 5V converter with the others disabled and the oscillation was still present.

Part of my thought process on the feedback traces was that the when the remote voltage sense wires are connected, they'll be two feet (60cm) long and will just be generally bundled together, not even specifically twisted, so I figured any non-idealities on the PCB would be moot compared to the non-idealities induced by those wires.

Tomorrow, I'll cut the feedback traces and run a twisted pair directly from the converter IC input pads to the output MLCCs and see if there is any improvement.

[tinfever]

Positioning of and frequency response back through the feedback path becomes critical with increasing switching frequency. The routing in your first post is far from ideal and you've not shown the whole layout of that.

I cut the feedback traces and ran a twisted pair for Vsen+ and Vsen- directly to the output caps but the issue remains.
(Photo of mod attached. Not the prettiest soldering. The twisted pair running out of frame is used for output ripple measurements on the oscilloscope.)

Oscillation starting after 1A to 11A load step at 2.5A/us with new feedback twisted pair. Powered by external high current PSU.
(Yellow = Vout, Green = Vin, Blue = Inductor current)


Added 2200uF aluminum-polymer to output. As usual, this makes it stable even with a 1A to 20A load step, although Vout looks awful suspicious when recovering from the undershoot.

[VEGETA]

so you made it stable with big 2200uf polymer at output? weird that low esr makes it stable

i wanted to suggest that you add a snubber circuit to the output, a more simple solution is a low value high esr elec. cap... something like 10uF small smd footprint to ensure high esr, it acts as a snubber and will stabilize it.

add these everywhere, like input and output, 1 for each. please try this with or without big bulky cap.

another thing you can do is put multiple smaller value polymer caps in parallel which gives less esr than 2200u... this could save you money since that big cap is expensive.

i would also suggest a small cap (1nF or less) across the feedback network but looks like it is not the case here, it does not use feedback resistors.

[youngda9]

Take a look at the typical application circuit on sheet 1 of the datasheet, linked below.  Output ground is defined as P ground and is shown separate from the A ground.  You have them both tied together.  You then have a 2.2uF cap in parallel with a 220ohm resistor from the VSENS- bin to BOTH the P and A grounds.

If you haven't already, I would try shorting across this 220ohm resistor (R308) which will tie all of the ground together.

Maybe try and do the same with the upper VSENSE+ resistor as well to remove the remote sense feature at this time in order to get it running right.  Simplify it and then add the more complex features.


https://www.vishay.com/docs/77863/sic450_sic451_sic453.pdf

[youngda9]

What is the part number of the inductor in the circuit?  Looking at the datasheet, the part uses Rds-on sensing to monitor the inductor current, and all compensation is done internally which is interesting.  I was going to suggest doing a bode plot to check for stability...still a good thing to do since output cap ESR and output inductor size will affect the loop.

[jkostb]

I think problem is caused by wiring between your lab supply combined with ceramic capacitors and switching power supply.
If middlebrook stability criterium is not satisfied you typically get these oscillations. I think your problem is at input side.

Could you do an experiment where you replace all input ceramic capacitors with an electrolytical capacitor. Electrolytical capacitor has higher ESR than the multiple ceramic capacitors.