Self-made buck-converter problems!

[step_s]

Hi lovely people of EEVblog.
I have tried building a buck-converter with the LM3485 http://www.ti.com/lit/ds/symlink/lm3485.pdf but I'm having some issues.

First of all when no load is on the output, the voltage just flies off, and is some random place between Vout and Vin (always higher than Vout is set to). When loaded down with just 100mA, the output voltage stays without 20mV of the set voltage.
Second of all, the efficiency is only around 70-75% when working with loads between 0.5-1A on the output (haven't tested it higher yet) with 12V in, 5V out. The diode and the inductor gets hot'ish.

The build is on a doublesided PCB cut out with a dremel, topside with components, bottom side for groundplane.
MOSFET http://www.farnell.com/datasheets/1698602.pdf?_ga=1.13462839.1224770061.1458764725
Diode http://www.farnell.com/datasheets/1690201.pdf?_ga=1.71066256.1224770061.1458764725
Inductor 10uH 5.6A http://www.farnell.com/datasheets/1723200.pdf?_ga=1.72769555.1224770061.1458764725
Using a 22uF tantalum capacitor on input and on output.

Hope you can help me with this build

[PChi]

The input capacitor, CIN, on the typical application circuit has to take the full switching current without the benefit of the inductor so it's worth paralleling more 22 uF.

The peak no load output voltage may be too high with lots of ripple because the minimum MOSFET switch on time may be quite long. It appears to use a one shot pulse generator. Can you provide an Oscilloscope trace of the output on no load?

[step_s]

Hi PCHi, thank you for your answer.
I just did some oscilloscope testing, and found out something quite interesting.
When I boot up the device, it works, and holds the voltage well (See no-load working).
But after 30sec or so, it turns of the PWM, and let's the gate of the MOSFET float. . . when the gate floats, the voltage on the output floats, resulting in strange readings.
Then under load (see load working), the PWM is very strange. It seems to be pulsing at around 4MHz, which seems crazy to me!

Anyone have any thoughts?

[Tomorokoshi]

Can you get a trace of the FB pin from when it's switched on?

[step_s]

Hi Tomo.
Do you want a read in the oscilloscope? Cuz I can't see anything other than it following the output, with the same amount of noise as the output.

[Tomorokoshi]

Did you put in Cff as shown in the example? The should be a very low value. Perhaps there is something wrong with it. Remove the capacitor and double-check the feedback resistor values.

The feedback voltage node should be around 1.24 VDC.

[PChi]

It's a puzzler.
The LM3845 data sheet seems to show that the MOSFET gate is controlled so not floating. If the output drifts high could it be the MOSFET drain leakage current though the data sheet seems to show that it should be OK (only a few uA).

I don't understand the On load switching waveform but Hysteretic converters can get confused by noise so perhaps adding more input filtering and checking that the ADJ and FB pins aren't picking up noise might help. Possibly add some temporary filtering to GND might help. I don't like the way the Typical Application Circuit is drawn. I prefer to see traces because with SWMPS there is no such thing as ground.

[SaabFAN]

How does your PCB look like?
I had massive problems with a LM3844 circuit that I built on vero-board and that didn't work properly. It didn't oscillate properly, didn't regulate and blew up one MOSFET.

The solution was to add a ground-plane by putting some adhesive copper-foil underneath the chip.
Maybe you have the same problem: The Return-Paths are too long or have too high impdances, which introduces parasitic inductance and capacitance, which in turn creates oscillations.

[step_s]

@Tomo
The datasheet recommends a 470pF, so that's what I'm using.

@PChi
I have just blown out my mosfet, so is now using http://www.vishay.com/docs/70627/70627.pdf
The whole thing is stable (most of the time) now that I have added a 1K load resistor across the output. It keeps it from floating around.
The crazy thing is now that the frequency of the whole system is 3.5MHz!! This seems crazy high?! It's working pretty well, but is only at aroud 85% efficient (with the other mosfet it was 90%).

@SaabFAN
I have made the project on a real copper plated double layer PCB board with groundplane on the backside, so this should be optimal. I have had the problems you are describing myself when messing around with lazy setups

[PChi]

Thanks for the update. It reads like it is working OK.
It's never going to have a stable pulse width just by using Hysteretic control.
The data sheet is very clever stating 93 % but at only 4.5 V input. As it uses an external diode (not bypassed with a MOSFET) the efficiency reduces as the input to output voltage differential increases.
The real world measured efficiency sounds reasonable.

[step_s]

@PChi
Would you consider this normal conditions? (look at picture, taken from the gate of the MOSFET)

According to the datasheets calculation, the inductor is what sets the frequency, but this calculation is very far from correct, and that is a bit worrying to me. . .

[PChi]

No I wouldn't consider it normal.
All the little wobbles are going to be strongly dependent on the positioning of the probe ground lead. Also some Oscilloscopes aren't very good at rejecting common mode noise and the trace can be misleading. Also can you increase the sample rate? At the moment it's only 20.8 ns, sometimes the display can be misleading. If it's possible switch the Oscilloscope display to only display the sample points without the interpolation/filtering creating a line whicgh can give a false view.

My guess is that the PGATE pin is having a hard time driving the MOSFET capacitance. I  presume it should be being pulled down to 0V. By the time it's turned it on the PGATE pin goes high again.

My interpretation of the data sheet page 11 is that the output capacitor ESR also affects the frequency.
The reality is that everything is going to alter the approximate switching frequency including layout so wont be stable.

[step_s]

@PChi
Have tried connecting the probes to different places on the ground plane, the input PSU and the output, but the readings seem to be the same.
The oscilloscope is a cheap hantek model, so I'm afraid the sample rate isn't really going higher :/
Have included a picture without lines. The green is the gate, and the yellow is the switching note.

It looks like the gate driver is working itself too hard, and from my calculations, driving a 6nC gate at 3.5MHz will cause the chip to work at around 0.5W, which is pretty high (it also gets very hot).
Could try to replace the chip again, to see if it is what's causing the problem.

The gate pin isn't suppose to go to 0V, since it's around Vin - 5V, in order to driver higher voltages, and still protect the gate.

I'm already using a 22uF tantalum cap on the output, together with a 220uF electrolyctic. What would i do to change the ESR?

[PChi]

I admit to only glancing at the data sheet and failed to understand that the PGATE voltage was limited. Thanks for pointing that out.

My guess is that the 220 uF electrolytic capacitor will have a higher ESR than the tantalum 22 uF capacitor so removing the tantalum will increase the ESR but will that not increase the switching frequency? you could just add a few mm of PCB track or wire in series with the capacitor as well to add some ESR.
My guess is that the noise on all the LM3485 pins will all affect the comparator so it's trial and error time.

If anyone else has any ideas please contribute.

[step_s]

Seems like I have tried everything now. Have changed to chip, and the same result. Different inductor (22uH), same result. More capacitors on the output, same result. Stille running at 3.5MHz. Now it's even more unstable, it would seem.
Seems to me like the LM3485 is too much work as a controller :/

Cheers on the input everyone

[P_Doped]

A few comments about how TI expects this device to be used:

Fig 22: The controller works by switching when the waveform it sees at the FB pin goes up and down by 10mV

8.2.2.4: The controller wants you to have high ESR so the output voltage waveform (AC coupled) looks like a triangle wave.  If you have a low ESR cap, they want you to add a resistor to make the net impedance of the "output capacitor" branch resistive at the switching frequency.  The larger the ESR, the higher the switching frequency.  ESR is proportional to 1/f.  You could take 2 of the same caps and if you put them in parallel, the switching frequency should halve.

With the 12V->5V step down, the 10mV controlled at the FB pin should be 10mV*(12V+5V)/5V = 34mV peak-peak at Vout.

If your Cff is too large, that will pass the Vout ripple directly to the FB pin, so the ripple seen at the output will be only 10mV peak-peak and the switching frequency will be higher.   

Once you have that, you can set the switching frequency by changing the inductor.  You want to 1/2 the frequency, double the inductor: L is proportional to 1/f