Buck converter using AOZ2153 chips

[fairyheavyindustries]

Hey folks,

Doing a 24V->5V DC buck converter using the AOZ products, specifically the AOZ2153EQI-30 chip.  A 3x3 QFN, that can handle 8A power output.  The design seemed easy enough, and I managed to get some boards printed out at oshpark.  I've been struggling to get the second revision of the board to work.  The symptom is that I get a slow rise of output voltage - over many seconds - to about 1.3V.  The Vcc on the chip never produces any voltage.  A load connected causes the output voltage to collapse to 0.  Basically I think the output cap is being charged somehow, and essentially I'm just seeing leakage of some sort?

The chip specs just seem so easy, and although this is my first design, I have a science background and have successfully built 2400 LED/600 watt medium scale projects before. 

I've included my schematic, and an image of the board layout.

Any suggestions?

ETA: as a side note, I've been doing my reflow via hot plate.  It's very possible I've exceeded the heat recommendation, but also I've been informed that these packages are more durable than it seems.  I've decided to get a reflow oven - the chinese $200 specials because I'm also teaching a SMD class to high schoolers and I've already burnt my FR4 boards and I don't need that smell and nonsense in the future.

[T3sl4co1l]

Hm, I'd like to see a stack of vias right by the power GND pads.

Checked for shorts?  Especially on EN and SS?  All pins and components connected?

Any guess if you fried it, if not in soldering then from powering up?  The 28V rating isn't much above the 24V supply and if you hot-plugged that thing into a bench supply, that lone ceramic cap on the input is going to overshoot severely.

The single cap on each side seems optimistic too, and also the small sizes.  For a 5V 8A supply I'd expect 2-4 x 1206 or 1210 sized caps?  A TVS at the input would be wise to handle inrush voltage surge, but a higher device rating altogether (or a lower nominal supply voltage) would be much more comfortable.

Tim

[fairyheavyindustries]

Why the vias?  From the data-sheet, it seems like they want the agnd and pgnd to be separate mostly.  so the bottom layer is my agnd, and there is a plane for power-ground on the top.  Connected at the input pin.  My understanding is there isn't supposed to be substantial power flow.

As for the caps, well, I was following the recommended application design on their data sheet (I only pretend to know what I'm doing).  My understanding is most of the magic is in the inductor?

The EN/SS short seems like a strong possibility.  I'll check - I did a stainless steel stencil (round 1 was a polymide stencil and that was a disaster) and I think the solder paste game is on point.  We shall see though.

Re: 24/28V, I had current limited my bench supply to 0.02 A, and that gave it a long slow startup slope - there's a small inrush to like 5V or so, then a second to ramp up to 24V.  No rush over 28V, as per my scope.

I'm thinking damage from reflow, like I said I'm getting the IR oven in soon and I'm going to try that.  It's quite possible the heat the board from below is dumping too much heat and prevents the chip from cooling down fast enough.

[T3sl4co1l]

Easy enough to do a board like this with a hot air machine, too -- something to consider.

Shorts and bad joints and stuff, are all a good reason to inspect everything with a magnifier (10x loupe or better), to probe for shorts (commercially, a flying probe tester is used -- CNC machine carrying a few probes over the circuit, pricey!), that sort of thing.

Tim

[fairyheavyindustries]

my leading theory is that my hot plate reflow method damaged some of the more sensitive parts of the AOZ chip.  My understanding is there's a linear regulator in the chip to produce a 5.3V Vcc for 'analog functions' as per their datasheet.  The power transistors are supposed to be robust, but the 'control' side is likely more fragile.

Since I'm not even getting Vcc power out, it seems like these chips are all not responding in the same way.

I have checked the shorts and the like.  For example, I show the requisite 2-4V on "EN" (I have 2 boards, with different values for the EN line), but the Vcc line which is nearby isn't showing any voltage.  Nor is the "SS" slow start pin which lies next to the EN and Vcc showing any voltage either.

I am going to try again with a T962 modded reflow oven.  I decided to jump on the oven because (A) I burnt FR4 and that shit's horrid and (B) I'm teaching a SMD design & build class to high school students and we need something more sane than a hot plate. 

[fairyheavyindustries]

Ok, so I just got a board to work.  I bought a T962, modded it, calibrated it, and built a board.

My prior guess was I was thermally damaging the chip during reflow.  That guess has been totally validated!

The problem is the hot plate method doesn't have precise temp control.  The PCB board carries over residual heat.  Doing the 'heat it up until it reflows' was too much for the little sensitive chips.

[SiliconWizard]

Well, if you don't have the patience to straighten wires on your schematics, I would question your attempts at hand-soldering small QFNs with thermal pads.
 

[fairyheavyindustries]

I think it's a matter of not caring, rather than 'not having the patience for it'.

Besides which Eagle is poorly designed software that has nonsensical default hotkeys.  Why is 'move' ^M? You can't do a split-keyboard/mouse style, where everything you commonly use needs to be entirely on the left hand.  I can't believe autodesk tricks people into paying for their crap UX!

Anyways, no need to be a jerkwad insulting people trying to do things!