Help with buck converter design

[Dijital]

So I am working on an LED project for a friend, and of course I have to take something relatively simple and cheap and make is complex and expensive. Modules are just no fun

TL;DR at bottom if you don't want to read all this

I needed a converter to take 9V and output 3.3V at 4.5A full swing. I chose the LM3150 synchronous buck controller from TI, and used their webench online tool to design it for 6A capacity with a 300kHz Fsw.

www.ti.com/lit/ds/symlink/lm3150.pdf

I laid out the PCB, fingers crossed because I could only do single sided at the time. It didn't work well at higher currents at first, but looking at the feedback pin on the scope it looked like a noisy mess. So I added a 220pf bypass as well as another 220pf bypass on the switching node for the same reasons. That seemed to rein it in, and although the voltage was sagging a bit at 4.5A, it was still fine, efficiency was pretty good (93%), and output ripple was a workable 60mV.

First Board:

Full of confidence, I wanted (and needed) to make some changes for the next revision. Mainly to shrink the switching plane and to put the input caps and source of the lower switching MOSFET closer together, both points that are stressed in the datasheet. Also I wanted to make a number of changes to the lower LED control circuitry.

I made a new board and fully populated it this time because of my aforementioned confidence, and once I put in the jumper leads under the board I can't use a hot plate anymore to assist with hot air soldering. Anyway, to cut to the chase, the performance was abysmal for anything above 3A. At 4.5A was getting 2.3V on the output, and it would slowly rise over the course of minutes to 2.6V. I tried all manner of bypass caps allover the place but was only able to get marginal performance gains at best. A 2.2nF on the feedback pin seemed to work best, but again, only marginal gains. I poured over the board looking for irregularities, but came up dry. Efficiency was still OKish at ~86%.

Second Board:

Well now the board is totally dead. I moved away from playing with the buck to test out the AVR. I just flashed a simple LED blink program, but on the first go also pulsed the EN pin on the buck at 1 Hz by accident (both LED and EN are PORTB). I wouldn't think that that would cause damage, but when I went back to the buck circuit later it was totally dead and it's the only thing I can think of that might have done it. The AVR is powered by a separate 5V linear reg.

I'm not about to give up yet though and just get a buck module (I really don't want to do that). So I am hoping that someone here with more experience could make offer some advice on where to go from here design/troubleshooting wise. At least on paper it seems that the second design should perform better not worse, but it does indeed perform much worse. I recently got some double-sided copper clad, and although I would rather drill holes and run jumpers, I'm not totally adverse to the headache of aligning the two toner transfers. Any specific help or general converter design tips are welcome, thanks!

TL;DR: I made two buck converter designs. The first version performs much better than the second version, despite the designing the second version more in alignment with the design guidelines. I can't figure out why that is, and would like to know how to make it so the next design performs at least as well as the first.

Note: I don't have the first design saved anymore, I wrote over it with the second design. (Doh!)

Bonus unpopulated board images
1st board:

2nd board:

[Phoenix]

Your schematic doesn't match your description which doesn't match your board. Can you please put up consistent details. For example the schematic shows a 100uF output cap - but I can only see a couple of small MLCCs on the output. Your output capacitors are too small...

You say you're putting a capacitor on the switch node - this is a bad idea as it causes large spikes of current through the switches each cycle.

As for what's going wrong - the node to put a scope on is the switching node. It will tell 90% of the story.

[Dijital]

Your schematic doesn't match your description which doesn't match your board. Can you please put up consistent details. For example the schematic shows a 100uF output cap - but I can only see a couple of small MLCCs on the output. Your output capacitors are too small...

You say you're putting a capacitor on the switch node - this is a bad idea as it causes large spikes of current through the switches each cycle.

As for what's going wrong - the node to put a scope on is the switching node. It will tell 90% of the story.

Hmmm, that schematic is indeed the circuit shown, shy of the 2 bypass caps I added on my own. The second design has no bypass on the SW node. Also the output caps are indeed 100uF MLCC's.
https://www.digikey.com/product-detail/en/murata-electronics-north-america/GRM31CR60J107ME39L/490-4539-1-ND/1033313
The LM3150 is actually made for use with ceramic outputs, that's why you see "Rr" and "Cr" in by the inductor in the schematic. They are there to generate ripple for the feedback circuit, since there are no output electrolytics to do it.

Here is the altium sch for the 2nd board:

The lower part of the board is entirely separate with only a trace for EN running up to the buck. The EN pin has it's own internal pull up, so the micro doesn't even have to be on for it to work.The entire lower circuit is powered by a separate 5V linear source connected to the 9V Vin. On the first board it was connected to the buck's output, but for noise reasons and 3.3V reasons I ditched that.


As for the switching node, I do have a scope shot form it.

That ringing is about 62MHz, and it shows up everywhere in the circuit

[Phoenix]

As for the switching node, I do have a scope shot form it.

That ringing is about 62MHz, and it shows up everywhere in the circuit

Geez huge 100uF 6.3V cermaic capacitor for a 3.3V circuit. What's your target capacitance? How much have you accounted for capacitance drop with voltage?

I'd suggest trying some electro caps on the output and closer to the inductor. Might help with the 62MHz ringing. Besides the ring the switch node waveform looks very clean.

Off the shelf modules are also a good idea...

[MagicSmoker]

Here are a few tips:

1. There needs to be a minimum level of resistance* ("Rg") in series with the gate to avoid ringing. The MOSFET datasheet will give you the minimum recommended value of Rg, usually in the section describing switching performance.

2. Ringing at the switching node can either be from the gate ringing (see #1) and/or from the interaction of the junction capacitances and the stray inductance between the two switches. This ringing can be controlled in a number of ways:
    a. slow down the switching by increasing Rg above the minimum (don't go too high or else efficiency will suffer)
    b. add RC damper network across each switch (C is usually 2x-4x the junction capacitance; R is usually chosen empirically in the range of Vo/Io at the low end to sqrt(Lstray/Cjo)*2 at the high end.
    c. reduce Lstray by bringing the two switches as close together as possible and employing a ground plane beneath them.

3. Speaking of ground planes, make the entire bottom layer ground and use vias to connect components and nodes to it. This doesn't require any special alignment of the top and bottom layers when making PCBs at home and it will improve the performance of SMPS circuits tremendously.

4. Beware of the terrible voltage coefficient of High-K ceramic capacitors, in which the effective capacitance drops - sometimes quite precipitously - with applied bias voltage, and that the smaller the package the worse the drop will be. A 6.3V/100uF/X5R capacitor operating at 3.3V might drop down to 60uF in the 1206 (US) package or 40uF in the 0805 package!



* - a ferrite bead can also be used to dampen oscillations, but this is usually only necessary for very fast switches operating above, say, 500kHz; not a recommended approach for those new to smps design.

[Dijital]

Geez huge 100uF 6.3V cermaic capacitor for a 3.3V circuit. What's your target capacitance? How much have you accounted for capacitance drop with voltage?

I'd suggest trying some electro caps on the output and closer to the inductor. Might help with the 62MHz ringing. Besides the ring the switch node waveform looks very clean.

Off the shelf modules are also a good idea...

The calculations were done by webench and the caps were the exact part number recommend. Murata has a nice tool on their website for analyzing their caps, and I don't recall anything sticking out when I went over it a while ago.

The real kicker for me is that all the same components supported a 4.5A load at 3.3V on the first board. I can't figure out what changed to make it fall apart, or what would be a prime suspect to investigate.

As for a module, I'd like to spin at least one more board before going that route. I'm learning a lot through this process and really would like to see it through, there really is no deadline or anything either. We all know how much better something feels when you do it all yourself.