Horrible load regulation with MPQ8623


I am using MPQ8623 for 12V → 5V conversion. Everything works great except for the fact that the output voltage changes significantly with load, as measured at the output capacitor:

0 A - 5.14 V
1 A - 4.82 V
2 A - 4.64 V
3 A - 4.54 V
4 A - 4.46 V

My layout is a close copy of the example from the datasheet (p. 19), the only difference being that the inductor is rotated 90 degrees clockwise. The inductor is 2.2 uH, 9.5 A (FXL0630-2R2-M cjiang (Changjiang Microelectronics Tech) | C167218 - LCSC Electronics) and the resistor divider for the FB pin has been suitably adjusted (20.5/4.42 kOhm). MODE is set to Forced CCM 1.1 MHz. Imax is set to 6.3 A with a 4.7 kOhm resistor. Softstart is ~9 ms (100 nF). PGOOD is floating. PCB is 4-layer, with both internal planes being ground.

The multimeter shows 0.9 V on the FB pin, although on the scope the value is pretty noisy. Vin is stable 12 V from a linear bench PS.

What could be the reason for the observed behavior? This obviously is not acceptable, so either I am able to correct it, or I need to find another buck. :confused:

UPDATE: I thought that perhaps induction was affecting the feedback signal on its way to the resistor divider, but comparing its both ends shows 1-2 mV difference (which should result in approx. 5-10 mV of difference in Vout). So probably it’s not that.


I removed the Cff capacitor from the feedback circuit and… the problem vanished: the Vo is very stable no matter the load.

I would however not like to lose the ability to react to fast transients. What could be the reason for Cff capacitor causing issues like that?

Hi milekspam,

That is a very strange issue.
What value Cff are you using?
How long are your feedback traces?
Are they closer to the chip or the load?

Vinh Tran

Hi Vinh,

Thank you for your reply. The Cff was 470pF. I have since replaced it with 220pF and redone the layout of AGND to use copper area rather than traces, and the situation significantly improved. There is still some minimal deep on heavy loads, so I’ll probably eliminate Cff altogether from my design - I don’t need to handle rapid transients in any case.