LM27762: LM27762 Charge pump doesn't reach negative voltage (why?)

Can you post your schematic, PCB layout and some waveforms (Vin, OUT+, CP, C1+)?

Does this occur on multiple boards or just 1?

You can order the EVM and get a known working circuit.

In the meantime I assembled another board (I already replaced the LM to be sure it wasn't damaged) with only the converter and associated components (so there is no load). Same issue and, in fact, I see horrible ripple on the 3.3 LDO output (about 20mV, not what I would expect from a low noise LDO at no load). As soon as I get an USB stick I'll do some waveform capture. In the old board I even removed the flying capacitor and observed that the charge pump actually put out less than 2.5V. For now here are the schematic and the layout. R14-Q1 and associated components are not on the board (that's part of the output circuit); not pictured is the fact that most of the bottom is ground plane, if that could be a concern. Could be that using bigger package capacitors gave the issue? Also I'm starting to doubt about the capacitor themselves...

Schematic http://imgur.com/a/PmM12 Layout: http://imgur.com/a/1fxmu

Thanks for posting. Yes, kindly post some waveforms when you get them.

Wiring/connecting all the nets/points like this is difficult to do properly. Current from one circuit is drawn across the Vin/GND, which causes a voltage drop at a point sensed by another circuit. Using a GND plane is much better. You might try adding wires to C1/C3 and, using a lab power supply, put 5V there.

Can you share the part numbers of the caps you are using?

Tried to apply the +5V from the lab supply directly on the input caps but didn't help. However I dug out the spec for the capacitor (GRM21BR71C475KE51) and the resonance frequency seem dangerously close to the 2 MHz operating frequency of the converter! Could it be the issue? If it's working near resonance the capacitor is less... capacitive than it should be. Also I don't know why even with the short spring ground probe I see a big disturbance repeating at about 3 kHz everywhere. I think it could be related (capacitor doesn't absorb correctly the charge pump pulse).

Having the self-resonance near the switching frequency means the cap gets the lowest impedance at this point. This is good.

We'll need to look at some waveforms to debug further.

You can also order the EVM for this device to get a known working circuit to test out.

OK. I suppose then the trick is to have the caps near resonance but not over it... I did some tries and modified the input caps more or less like the EVM, now there is a 10µF and a 1µF in parallel. It actually helped, since now at least the positive rail works correctly. I also 'borrowed' a usb stick from a colleague and done some wave shots, taken with an N2872A 1:10 probe with the head 'coil spring' ground clip, ground reference is on the via right above the regulator (there is indeed a ground fill, it wasn't show in the layout screenshot to avoid hiding the traces on the other side). If it helps consumption is about 5 mA

- +5V IN, on the left of the regulator (http://imgur.com/a/eGcHV, AC coupled http://imgur.com/a/D5SD1, AC coupled an zoomed on the transient http://imgur.com/a/VT8Sv)
I think these transients are true (not induced by the clip) since I don't see them in some other points. I see both the 2MHz switching and some other slower transient. Repetition rate could be the one of the PFM mode (remember, the converter is not loaded on this board)

- Input caps, (http://imgur.com/a/jrTuX, AC+zoom http://imgur.com/a/kCdCV) the caps successfully filtered the 2 MHz transient, the slower one remains.

- Regulator input, on the via near the pin (http://imgur.com/a/EC48C), peak mode acquisition, no BW limiter. As in, I can't see better with this scope...

- C1 + side (http://imgur.com/a/VyIQK, zoom on the burst http://imgur.com/a/zdESf), C1 - side (http://imgur.com/a/n21g8, zoom on the burst http://imgur.com/a/TOdLQ) I'm not a charge pump expert but they are at least plausible with the working described in the datasheet. However I don't get why it pumps only to about -2.8V max, and why the PFM kicks in. I would expect something like at least -4.5V, here.

- Charge pump tank capacitor CP (http://imgur.com/a/p4VA4R, zoom on transient http://imgur.com/a/KtXJb). More or less what I would expect from the signals on C1

- V+ output, on the filter cap (http://imgur.com/a/KTvXn, AC coupled http://imgur.com/a/dL1Gs) the +3.3V output is quite fine, at least from what my scope can see. No sign of transients here so the previous ones were not a ground clip artifact, probably.

- V- ouput, on the filter cap, just for completeness (http://imgur.com/a/WbBtx) the negative LDO is starved (the CP doesn't reach it's charge) so I don't expect anything good here.

While the EVM arrives I could eventually try to fit input capacitors nearer to the regulator, on the bottom side. However the malfunctioning is not clear to me. From the datasheet the PFM compares the input voltage with the stored charge on the CP and skip cycles while it's full enough. So, in theory, if the regulator sees a low transient of, say 2 V it could stop charging the CP to about -2 V. However (unless it's a very fast transient I can't see) it doesn't go under, about, 4.1V and the voltage on the flying capacitor isn't spectacular either (I would expect to see almost all the input voltage on C+ on the charging phase)... what's happening?

Thanks for sharing your solution. Yes, the impedance of all capacitors in a switching power supply must be minimized. All your plots show high voltage drops, which implies a high impedance.