Strange spikes in switching power supply using Diodes AP63200

[i_like_sparks]

I have designed a small 24V to 5.5V 1.5A switched-mode power supply around the Diodes AP63200. If I set my electronic load to between 155 and 165mA, I'm getting strange voltage spikes of around 250mVpp. At 159mA load they line up perfectly with the line frequency (the attached images are even triggered from AC).



Below ~150mA and above ~170mA load it works fine with <60-70mVpp ripple at 5.5V. For the schematic, I've mostly followed the datasheet/evaluation board:

https://www.diodes.com/assets/Datasheets/AP63200-AP63201-AP63203-AP63205.pdf
https://www.diodes.com/assets/Evaluation-Boards/AP63200WU-EVM-User-Guide-Manual-R1.pdf



I have made two small changes: R10 and D1 are for suppressing large ~40V input voltage spikes at turn-on (maybe from cable inductance?). C17 delays turn-on for 0.5-1s for the same reason. Replacing R10 by a 0 ohm jumper makes no difference. Here is the board:



Things I have tried:

• First I thought it could be a ground loop/ground noise coupling through the probes shield, so I tried various forms of measurement from Dave's video (except the active differential probe which I don't have): single-ended 10x and 1x, two probes and difference function, BNC with 50 Ohm series and 50 Ohm termination. The waveform images were taken with the BNC 50 Ohm setup. But I also noticed that the capacitors create audible noise at exactly the 160 mA range. I disconnected all test gear and still got the noise, so I don't think it is caused by the measurement setup.
• I thought it might have been caused by some oscillation between the SMPS and my active load, so I replaced the active load by a fixed resistor to draw 160mA, same spikey result.
• Then I thought the AC frequency might becoupling into the feedback path somehow, so I lowered the resistor values in the feedback divider by a factor of 100 to 2.7K (R14) and 470 Ohm (R15), same spikes.
• I also tried the suggested feedback capacitor C19 of 100pF from the datasheet, makes no difference.</li><li>I added more (and larger package) input and output capacitors, same result.

What is going on? Is this just a badly designed chip? But I'm still confused that the spikes can be made to line up perfectly with the AC line.

Here are some more pictures. This one is at a smaller load of ~150mA where it works well:



This is at a load that is just a tiny bit smaller, ~157mA. The spikes appear to be much less lined up:



This one is at a larger load of ~170mA, the noise is much better:



Zoomed in on the spikes (top blue curve, taken with difference function):

[Niklas]

What kind of power supply are you using? Battery or something connected to 50 Hz AC?
Have you checked the input voltage to the regulator?
There is no external compensation network on the regulator, so there must be something built in. Recently I have seen a strange behaviour on a Torex regulator with internal compensation. For some narrow input voltage regions, above 95% duty cycle, the load step response changes rapidly. There is also an interaction between the input protection circuitry and the regulator in my application as the load current changes and so does the input voltage seen by the regulator. Do you see something similar as there is not much input buffering capacitance on your board?

[Miti]

I just happen to have a project with the same chip, 12V in, 3.3V out. See attached some screen shots.

[i_like_sparks]

@Niklas: I have tried powering using my bench supply (Agilent E3631A) and using a cheap Ikea "Tradfri" LED power supply, with the same results. No battery yet, but maybe I should try with 3x9V just to be sure.

The voltage at the regulator input looks good, a little noisy but no visible oscillation at line frequency.

I have a suspicion that this might be the exact spot where the regulator switches from PFM mode to PWM mode and it does it in a suboptimal way and goes back and forth rapidly. But it is hard to believe that they wouldn't implement some form of hysteresis there, and it doesn't explain why it would sync up with the line frequency.

I haven't checked the load step response so far, what should I look for and did you find out what the exact issue was?

@Miti: Awesome! That looks much better and overall much more similar to what the datasheet promises. On my board the issue comes up only in a very narrow range of 10-20mA. Do you have any way of sweeping through the load current to see if the spikes come up? I think it is where the pattern changes from PFM to PWM, so I think it should be somewhere between your 200mA (with the large PFM ripple) and 500mA pictures (with the smaller PWM one).

Would you mind sharing the regulator part of your circuit and your inductor/capacitor choice, possibly via PM?

[Miti]

Sure, no problem. See attached.
The caps are X5R from JLC, the inductor is LCSC C351092. The schematic shows AP63203 but it's nowhere to be found at any distributor, so I had to use AP63200. If you want I can post my layout as well.

[trobbins]

Did you use the slip on coil type ground prod with the 10:1 probe?  Ie. don't use the typical ground wire and clip, or the sprung hook tip attachement, and connect the ground prod and spiky tip directly across C11/C12.

[Miti]

First observation is that your layout does not follow the manufacturer’s recommendation. The inductor is a bit far from the IC. Second is that if you use 6.3V rated caps at 5V and 25V caps at 24V, probably you end up with half the capacitance that you expect. I always derate my capacitors generously, at least 2x and even more if the size allows it. For example all my capacitor in the attached schematic are rated at 25V.

[i_like_sparks]

@Miti: Thanks for the schematic - I am a bit confused, is this the correct one? For the 63200 (with variable voltage) I would expect a divider network in the feedback path to set the voltage. For the 63203 (fixed at 3.3V) the datasheet doesn't mention C51.

What (physical) size are you using for the input/output capacitors? I read that the smaller ones tend to degrade worse with increasing DC bias voltage (of which there is a lot - the input is at 24V more or less constantly), but I've now replaced them with 1210s, so that shouldn't be too much of an issue.

I will try with higher-rated caps, but I've double the capacity already on the test board just to be sure.

Regarding the layout I couldn't follow the recommendation exactly due to space. But if I look at the schematics of the evaluation board (linked in my first post), the inductor also isn't too close.

@trobbins: The first tries were with a 10x probe with ground lead (without the spring tip), which couples in some noise so is definitely not good for quantitative measurements. But the spikey behavior is the same as with better measurements. For the later measurements (the one in the pictures) I have soldered a piece of RG316 coax with a 2cm wire lead via a 50 Ohm series resistor to the output caps, connected to the scope via a SMA/BNC adapter and BNC cable and terminated with 50 Ohm at the scope.

Another observation I made is that when I go near the inductor with a metal object, the current range where the spikes happen seems to shift slightly down.

I am also a bit puzzled about the strange output waveform (the 170mA picture in my first post). I would expect something much more like a sine wave.

[Miti]

As I said, the intention was to use the fixed voltage version but the stock is zero everywhere and the lead time is 24 weeks. Crazy times for semiconductor industry again. I wanted to start my board quickly, so I ordered the adjustable version and added the rest of the parts. I don’t think the feedback is your problem.
The data sheet gives you a basic recommendation that would make the circuit work, but it is not necessarily the optimal one. That doesn’t mean my schematic is optimal ... and what does optimal means in electronics anyway? 

[i_like_sparks]

Ok. One suspicion that I have is that the issue might by the ground path, which is crossing under the device. I think at least part of it might also be going around the feedback trace. If it's really that it would suck because I'd have to redo the whole thing. I noticed that in the evaluation board they strictly keep the high-power stuff to one side of the IC and all the feedback/low-power stuff to the other.

The confusing part is that if I just replace the 62300 by a 62305 (fixed at 5V and 1 MHz instead of 500kHz), I get a much cleaner waveform. So it might also be related to suboptimal compensation?

[i_like_sparks]

Ok, here are some more observations:

• I'm pretty sure now that the spikes happen exactly in the range where the IC switches between modes.
• The load range where the spikes happen varies only very slightly (a few mA) when varying the input voltage between 12V and 25V.
• If I set the output to 155mA and the input to 24V, I get the erratic behavior. If I then set the input to either 23V or 25.5V, the output is fine (but with very different noise depending on the mode, more noise in the low-power mode)
• The device does have a hysteresis of around 5-10mA load current: it will switch to high-power mode if the load current goes above ~160mA but only switch batch if it falls below ~150mA. BUT this depends strongly on the input voltage! At 12V input, hysteresis is ~10mA. At 18V input, it is ~5mA. Above 22V input, it is virtually 0mA, and at approximately 157mA load the device switches back and forth between modes at 50Hz.
• When I power the device from an 18V battery, I still get the same behavior of intermittent spikes between 152-157mA load synchronized with the mains frequency.

I have no idea what's really going on. From the last measurement it is clear that the noise must be creeping in either through the electronic load (but it also did occur when I loaded with a resistor), or through the probe setup. Unfortunately I don't have a differential probe so can't really measure without a ground connection. What I did found is that when I didn't attach the ground lead and did an open ended measurement on the GND pad of the output capacitor, the ground noise became much worse when I decreased (!) the load, which didn't happen with the output.

[trobbins]

Given the sensitivity to mode change is related to mains frequency, then it would seem appropriate to continue to exclude mains power equipment from your test setup - ie. don't use the active load, and use a battery power supply, as there may be noise flowing through the ground traces due to ground/earth loops.  Obviously a battery powered scope would help, or a laptop and soundcard running scope software.

Once you have a clearer awareness, then retracing the ground/earth connection loops, and seeing if they can be mitigate would seem worthwhile.

Also you may need to be more pedantic with the feedback layout - the takeoff point for R14/C19 should be between or after C11/C12 (and preferably in the C12 pad), not on the inductor side (where noise current flow is at its highest).  Also R15 should have as isolated a return to pin 4 as possible (and preferably direct to the pin 4 pad), whereas your layout has it on the way to C11/C12.  The feedback signal has to be as 'clean' as possible for the IC to do its job - the same goes for any other control signal that may be affected (although in this case the only other control input is the enable and that is unlikely to be affected by noise on the input side).

[i_like_sparks]

Thanks trobbins, that makes sense. I did have a suspicion that the GND connection for the divider resistor R15 could be the culprit, because it is not only close to the GND pins of the output capacitors, but also half way to the GND pins of the input capacitors. As a test I cut the ground plane above R15 so that it had its own connection to the ICs GND pin, but that didn't change anything.

I'm currently re-routing everything to be as close to the evaluation board as possible: https://www.diodes.com/assets/Evaluation-Boards/AP63200WU-EVM-User-Guide-Manual-R1.pdf

One thing I was wondering is why in their layout on page 7 (which looks  a bit different to the datasheet) the feedback trace takes a huge detour and runs under the inductor. Is it to keep it away from the noisy ground currents?

Also it looks like they have removed the soldermask above the input/output cap GND vias, I guess it is to improve resistance? I'm wondering if that really helps or if it's just cosmetic.

[trobbins]

The feedback trace layout may have been some designer musing - it seems a little tenuous as to reasons - like keeping it away from the IC and from the feedback pin connected parts and node (to minimise capacitive noise coupling).

The removed soldermask may have been to inspect the vias better.  It may also have been to see if they could fill in the vias with solder to lower the resistance between top and bottom, and/or improve thermal conduction.

The proto board layout does inherently separate IC gnd pin currents that relate to power, versus control.  It also situates the inductor as far from control parts as practical - and that Wurth inductor range supposedly has some inherent shielding.  If your inductor is getting hot, you may need to enhance cooling with some vias to the rear pour.

[i_like_sparks]

Ok, thanks for all the input. This is my second attempt. I had to rotate one of the connectors and move everything to the top. The feedback path should be much cleaner now. Thermally it's probably worse, but let's see how it goes.

[i_like_sparks]

While waiting for the new boards I've tried Mitis suggestion and replaced the capacitors by higher-rated ones (50V input caps for the 24 rail and 16V output caps for the 5V rail) and larger package. That decreased the peak-to-peak noise by about 30% (!), so more capacitance = definitely better.

More confusingly, I also tried a "better" inductor - larger package, higher-rated (>3A for supply with 1.5A max), and smaller DC resistance (29mOhm). With everything else the same, this increased the switching noise by around 25%.

I also noticed a "ringing" at 5MHz in the switching output, which appears to precede the edges in the output voltage.

[i_like_sparks]

So here is the update. With the new layout, the general shape looks much less bouncy and more like what you would expect from a switching regulator. The sensitivity with respect to line frequency is also gone. BUT the spikes at a specific load are still there. I have attached a few pictures for 73mA (1+2), 100mA (3) and 168mA (4+5) load. In the 168mA picture the spikes are clearly visible and go up to ~225mVpp noise.

Below ~428mA load I also still get the "ringing", see the last two pictures: 151mA with strong ringing, 1.5A with no ringing.

I think next would be to check if adding some capacitors to the feedback circuit helps? Or does anyone have any other ideas?

[Weston]

I think you are mainly just seeing the converter enter different operating modes and different loads

From your first two pictures you can see the converter is running in pulse frequency mode, the converter is running for a few cycles and you see the output voltage increasing as the converter supply power and also some switching noise. When the converter turns off the output voltage begins to decrease and the switching noise vanishes.

In the third picture the converter is also running in pulse frequency mode and then runs in PWM mode for a bit. The converter turns off, output voltage drops, converter turns on and rapidly ramps back up to the set output voltage, and then runs in PWM mode for a bit.

Unsure about waveform 4, it looks like it could be some aliasing of the sampling rate / displayed samples of the scope and the switching converter. If real, it could imply something is up with the internal compensation and your input / output capacitance values.

Your last two waveforms show the converter operating in discontinuous vs continuous conduction mode. At low currents (before the converter enters pulse frequency mode) the converter will turn off the synchronous rectification switch when the current hits zero to prevent it from conduction reverse current. This causes the switching node to ring at the frequency of the inductor + the switching node capacitance.

Not sure about your first revision, but these seem like relatively typical waveforms for a converter that can operate in PWM and PFM mode. If you need lower switching noise consider switching to an IC that does not have a PFM mode.

[uer166]

I concur that what you see is normal. In order of pictures: PWM CCM, PFM mode, and finally PWM DCM (I assume it still modulates using PWM when inductor current is discontinuous.

If you don't want large ripple cause by PFM, don't use a PFM chip. If you don't want ringing caused by DCM, use a synchronous buck in forced CCM mode.

[i_like_sparks]

Thanks Weston and uer166 for the explanations, that makes more sense now and I also have an intuition now where the oscillations come from. I'm ok with the ripple, it's for LED lighting so not critical.

The reason I initially posted (and why I'm still puzzled) are the "excursions" in picture 4 (168mA). These are not always as benign as in 4 but can go up to ~100mV either side as shown in picture 5 (168mA zoomed out) and then take a long time to settle back. If I take a single capture and zoom in, the spikes all look like picture 4. This is also visible in all horizontal settings. I'm using peak detect mode, AC coupling and 20MHz BW limit.

Meanwhile I have tried adding the 100pF compensation capacitor across the top resistor in the voltage divider as suggested in the datasheet, but that didn't change anything.

[KT88]

I agree with the conclusions of Weston and user166 that we see normal operation of the AP63203.
It often makes sense to tweak the burst frequency to a higher rate. This can be achieved by reducing the size of the output cap. Tthe cap depletes faster and requires faster charge-up.
But there are limits where you run into stability issues. Ideally you simulate it or play with the values at the actual PCB.
This often frees space (and budget) for a second filter stage which in sum can score much better EMI results that just one cap.

Cheers

Andreas

[i_like_sparks]

Thanks Andreas, will try that too. One important remark: All the pictures and the spikes only happen with the AP63200 (i.e., the variable voltage version). I have also tested the AP63205 which has the network built-in, on the same layout with 0 Ohm jumpers, and it does NOT produce the spikes as in pictures 4/5. It also runs a lot cooler, maybe because for some reason it runs at twice the switching frequency (1MHz vs 500kHz) of the fixed-voltage version.

[KT88]

Switching losses scale linearly with the switching frequency.
Concerning the noise with the external feedback variant, it is likely linked to the trace on the bottom side that goes into the feedback divider. It should be routed the other way arond the part. The way you did it, it forms a secondary transformer winding coupled to the switching node related traces. If you delete that trace an run a wire the other way you would probably see an improvement. You could also try it without C19 if you have that populated right now...
The observed noise could potentially trigger the MOS driver partially multiple times contributing to the heat dissipation - but that's just a hunch...

Cheers

Andreas

[i_like_sparks]

Hi, in my original variant (see my first post) I had a direct feedback trace connection as suggested in the datasheet. In the revised design of which I also posted the board layout I moved it to as far away from any large currents as possible, as done in the official evaluation board. If it's really the layout then both are wrong and I'm not sure what to do more. But the layout DOES work with the fixed-voltage versions.

edit: C19 is the compensation cap. The datasheet proposes to use 100pF for MLCC output caps. I have tried with and without, it doesn't make a difference. Even with 100nF I get the same spikes.