Need help with a step-down regulator IC

[bfrigon]

Hi,

I don't have much experience with step-down regulators. This is the first time I use one in a project.

It is a ADP2303ARDZ-3.3 from Analog Device. 

Input : 6.5 V
Output : 3.3 V
Load : 280mA

There is noise on the output, about 350 mV peak to peak and it follows the switching frequency (700 kHz).

I did put two 47uF (X5R type) caps on the output as suggested by the datasheet. I was not able to find the exact model of caps they recomended so i used this one :
http://www.digikey.ca/product-detail/en/JMK316ABJ476MLHT/587-3399-1-ND/4157512

I used a 4.7uH inductor (CLF10040T-4R7N) :
http://www.digikey.ca/product-detail/en/CLF10040T-4R7N/445-9298-1-ND/3687735

What could be the cause?

[Skimask]

Without a schematic, or even a picture of your completed board...meh...your fruit juice is expired?

[bfrigon]

I did post a schematic in the attachments (circuit.jpg).

I build a test board with what you see on the schematic, nothing else. The power is supplied by my bench power supply and the load is just a bunch of leds on a breadboard.

[Paul Price]

You are missing an inductor 5uH to 330uH between the two output caps C25 C28, the bigger  the uH in value, the cleaner the output, and using a large value inductor is better in most cases.  if possible make the grounding point connections of all components having a ground very close to each other, and all grounds on one big fat thick PCB trace area. The output ground connection should lead to the output point from the right side of this central grounding point. Using a 22uH inductor should lower the output switching noise to <20mV P-P.

[Zero999]

What sore of capacitors are you useing for C25 and C28? If the answer is ceramic, then their capacitance will be much lower than stated with over half the voltage rating applied.

[bfrigon]

What sore of capacitors are you useing for C25 and C28? If the answer is ceramic, then their capacitance will be much lower than stated with over half the voltage rating applied.

They are ceramic.

in the datasheet, they recommend using X7R or X5R.  What should I use? i'm limitted on board space, it needs to be a 1206 package.

You are missing an inductor 5uH to 330uH between the two output caps C25 C28, the bigger  the uH in value, the cleaner the output, and using a large value inductor is better in most cases.  if possible make the grounding point connections of all components having a ground very close to each other, and all grounds on one big fat thick PCB trace area. The output ground connection should lead to the output point from the right side of this central grounding point. Using a 22uH inductor should lower the output switching noise to <20mV P-P.

I replaced the 4.7uH inductor with a 10uH, there is already an improvement. The noise is now at 125mV p-p.

I don't have the space on the board for a second inductor. Considering that, would a single 22uH still be sufficient to maintain the noise within 20mV ?

[bfrigon]

Try an RC snubber across your rectifier. The output LC (ceramic caps and low inductance) is likely high Q, do you really need a 4+Amp inductor for a ~1W inductor?

If you don't want to get more parts try a lossy inductor in between your two output caps some R may dampen the ringing. You can also try some Tantulm polymer caps Far cheaper then what you paid for the ceramics, (the ones on sale) much more temperature and bias stable as well. Before people freak out they are polymer tants made for switching converters. For 700kHz I would think you would only need 10- 47uF bulk capacitance and maybe a small LC filter to clean up any residual ringing. Sometimes a little bit of esr and/or DCR is a good thing.
Try the RC snubber first though see if you can get good enough without getting more parts.

Thanks, I will try the tant caps.

By the way, I tested it with a 280 mA load for now. But i will eventually need at least 1.5A with 1A reserved for future addons.

[bfrigon]

The original 4.7uH inductor was to small to maintain CCM for a 1W output, your 10uH one increased the DC component and reduced the ripple current. For your power you should really be using about 22uH that will minimze your AC ripple. see pics

Where did you find this calculator ?

[rob77]

http://schmidt-walter.eit.h-da.de/smps_e/smps_e.html

(googled it using phrases from the images posted by AcHmed99)

[Mad ID]

It looks like you are not measuring correctly, i.e. this looks as magnetic coupling to your probe.
Did you use a short ground lead like the one in the picture? Without it the noise will be 10-100x bigger than it is..

[T3sl4co1l]

It looks like you are not measuring correctly, i.e. this looks as magnetic coupling to your probe.
Did you use a short ground lead like the one in the picture? Without it the noise will be 10-100x bigger than it is..

Seconded.  Either your layout, or your probing technique, or both, is bad.  This device and type of circuit requires large, stitched, ground pours for thermal management and carrying ground currents.

Regarding damping (R+C across the diode): contrary to "conventional rule of thumb", you can actually intentionally increase trace inductance around the diode, which will shift the ringing to a lower frequency and higher impedance, where it is easier to snub or dampen.

Tim

[T3sl4co1l]

Tim are you sure about that? Usually the higher the frequency the easier it is to attenuate.

My classic example:

- I designed a 5kW high frequency inverter module.  Multilayer board, "minimized inductance".  It was built with ISOTOP / SOT-227 power modules, and I estimate it had around 15nH loop inductance.  For something about as wide across as your hand, that's not bad at all.

- It rang like a motherfucker, 80% overshoot.  Now, when you're starting with 650V DC link... that's a bit of a problem.  We had to buy 1200V transistors just to do initial testing.

- Ok, so, it's multilayer, we've got connections to the opposite supply rail nearby.  Let's try putting some clamp diodes in there.

8A diodes: toasted.

12A diodes: toasted.

30A diodes: survived.  But the overshoot is still over 10%, and even being generous with stray inductance on the diodes and connections, that's over 40V applied across the die itself.  In just a few nanoseconds.

By the way, the diodes failed mysteriously, barely warm at all.  I suspect electromigration, but I really haven't heard anything concrete about high current failure modes.  (Hot-spotting is unlikely because forward drop is resistive and PTC at those currents.  Maybe there's something spooky like nonuniform forward recovery -- who knows, manufacturers are very quiet indeed about this phenomenon.)

- Later rev, I explicitly added about 100nH of stray to the layout.

Instead of ringing at 60MHz, peaking over 100A, and pounding the ever-loving shit out of the clamp diodes, this brought it down to a manageable 10MHz, the worst case overshoot to 10%, and the peak current down to 20A.

What's more, a proper RCD clamp snubber handled the energy cleanly, dissipating it (of which there is slightly more at high loads, due to the E = 0.5 L I^2 stored in the supply inductance) dissipated safely in a moderately sized resistor (there was under 400W total dissipation under worst case operation -- always over 90% efficiency for a 5kW high frequency module).  At full rated output, the snubber power was a small fraction of losses, so this is a perfectly reasonable sacrifice.

- The moral of the story is, it's not even a sacrifice.  Your inverter necessarily has junction capacitance and loop inductance.  The only reason you're suggested to minimize inductance is because, if and only if you can get the resonant frequency higher than the switching speed, or the impedance sqrt(Lstray/Cj) low enough, then it won't be stimulated by the transients.  In reality, especially with switching speeds so fast now, that this is neither possible nor reasonable, one must match the impedance to the circuit (roughly, switch Vpk / Ipk = Zo), and assign a reasonable frequency for the loop (usually, f_o > 20*f_sw is reasonable without incurring too much losses, while still being reasonable to deal with).

Tim