LM3478MM Sepic problem

Hopefully one is a full common ground plane. Those inductors can pollute into the chip itself but the effect should not be this harsh. So I'd wait for the new transistors. Rule in a SEPIC: The FET must withstand Vin + Vout + Vdiode. Plus lots of spike margin like 50-100%.

Yes, either the FET or it is spike coupling into the LM3478.

Yes, should be fine.

Haven't calculated but should be in the ballpark.

I can read Gerbers but others here might not. Best to post it in pdf. Still, it's nearly impossible to diagnose a switcher across an ocean :-)

If you have the time (lots of it) you'll get there ;-)

What also helps a lot in the beginning is to buy an eval board and stick as closely as possible to their layout. Or at least get the layout for the eval board from an applications note or something.

The LM3478 may not be the easiest chip to start with. I have done a product design with it and had no issues but I remember a guy in a German NG who just could not get it to work reliably. I think he used another chip in the end.

--
Regards, Joerg

http://www.analogconsultants.com/

"gmail" domain blocked because of excessive spam.
Use another domain or send PM.

Those screen shots indicate some rather extreme ringing at a higher frequency than the applied PWM drive. I have seen noise like that on my own

100 kHz switcher. I have tried to measure the MOSFET current with a scope clipped directly across a flat metal strip sense resistor, and there is a lot of high frequency noise. But I also see it when I clip the ground lead of the probe to the tip, which indicated that it is picking up radiated or induced RF. I have read that you must make a special probe tip adaptor to get a good signal.

I had some bad experiences with current mode switching supply controllers, particularly the LT1247 and the equivalent UC1843A. When I tried to get more than about 20-30 watts, it became unstable and I experienced overheating of the inductor(s) and the MOSFET. Of course, as things heat up, even more power is consumed in I2R losses, and a runaway condition can result. I did find that it was important to provide a very good gate drive, especially when I used a MOSFET with low RdsOn and high current and voltage ratings. The gate capacitance and charge caused a lot of switching losses, so I used a 9A gate driver UCC27321 with some improvement.

Now I am using a PIC16HV616 with PWM output for the switching boost converter, and it usually seems stable, but sometimes it tries to regulate at a higher current than it should. But at least everything seems to be more predictable and is easier to measure on a scope. I am only measuring the output current and adjusting the PWM in a software loop, rather than attempting cycle-by-cycle regulation.

The previous current mode converter had such variable waveforms that I could not get a good picture of the operation. The duty cycle was all over the place, and I saw bursts of PWM signals alternating with quiescent periods.

My circuits are unique in that I am regulating an output current into a non-linear load (a string of high power LEDs), so I could not run the circuit at no load (which would be a short circuit). But I only had major problems when I exceeded about 20 watts, and it really gave problems at about 50 watts. Yet it is only a 1" x 2.5" PCB, and I was trying to get better than 85% efficiency. I could easily obtain 75%. It sounds like you have even more severe problems.

Paul

"Joerg" skrev i en meddelelse news:zQPyl.10693$% snipped-for-privacy@nlpi070.nbdc.sbc.com...

Yes, one plane is a full groundplane. I haven't had the luck to build PDF's, CAD is not my area of work, but here are the gerberfiles. The PSU are located in the left side. Note there are two of them, but they are similar.

Well, I got my 100V mosfets and new 100V diodes today, but everything still burns, when loaded :-( I've taken new scope shots

Here the drain, which has changed, but still looks weird.

Here the source, which is so "grassy"

and the source grass up close:

The gate, which has oscillations, I think

The gate, up close:

I almost don't know which was more painful, the actual fingerburn, or the hardhitting fact that the bigger mosfet did not just fix my problem ;-(

I've tried changing R145 to alter the slope, but so far to no avail.

Yes, I know that you are right, but having no more blisterfree fingers, I grasp at every straw ;-)

No, I can definetely see that this is not at all easy and that my approach from the beginning was probably wrong, but I keep the hope alive ;-) I really like this design to run in the end. And learning along the way should make this easier next time ;-)

Best regards Henrik

Doesn't looks like a plane near the switcher though, see below.

Dang :-(

The "grassiness" most likely comes from your probes. The ground clip picks up EMI from the switcher. To really measure well you'll have to make coax probes such as a piece of RG192 or whatever flexible stuff you have around, cut one end, solder a very small 1K resistor to the center conductor, then the coax shield to GND close by. Switch the scope to

50ohms. Ok, you can't measure hi-Z nodes with that but in a switcher you don't have to anyhow. But you can measure ISEN with it (not FA and FB).

Ouch. The rise time is very sluggish, normally the LM3478 is more zippy than that. But it is definitely choking, not operating at its normal

150kHz or so. This means the FET spends much of its time in the linear region and thus gets very hot.

Choking happens mostly when spikes from inductors and such get into feedback paths of the chip, in this case ISENSE, FB and COMP, possibly even FA.

Some files errored on GraphiCode but one issue jumped right at me. At first blush: There is a long (meaning inductive) path from the bottom of the sense resistors to pins 4 and 5 of the LM3478. This needs to be a plane.

The path from R145 to ISEN ist quite long and smack in the magnetic field of L4.

Can't see all the parts designators but it seems C3 is connected to the top art plane and that's not via'd to GND anywhere close.

Best would be to borrow a FET-probe, connect the shield as close as possible to pin 4, and measure ISEN, FB and COMP. Short meaning a few millimeters. Often scope probes have the shield run to the tip end so you can use that to ground it (the alligator clip lead won't work). Nothing should be longer than a few millimeters. Now the scope would show what the LM3478 "sees". For this type of work there are special sockets that can be soldered onto the board and the probes stick into them.

If you do these kinds of projects a lot it really pays off to have a decent FET probe. Tektronix and Philips make good ones. Sanioty check for measurements: Touch the GND plane with the tip and there should be almost nothing on the scope screen with the switcher running.

A quick thing you could try: The inductors aren't very good ones. Take both out and hook them up via 3cm of twisted pair wire (thick enough for

10A so it doesn't burn up) and let them ride higher above your board. This isn't a "fix" but gets their magnetic fields a bit away from the board, to see if there is an improvement.

To be honest I think this layout may not work. It would be good to spend the time and build up one of those regulator again, on a piece of copper clad. Just a bare piece of FR-4 with a copper layer. Run everything directly to that plane and the rest as close to the LM3478 as possible. Move L4 1-2cm away. When you get this to work you can pick better inductors of the shielded kind and do the layout again with a solid plane structure.

--
Regards, Joerg

http://www.analogconsultants.com/

"gmail" domain blocked because of excessive spam.
Use another domain or send PM.

[snip]

I think you have other problems but maybe try a resistor at the gate.I don't think that's resonant oscillations (I've never seen them like that before) I could be wrong though but a small value resistor in the gate wouldn't hurt. You might also want to put a Schottky on the current sense pin to clamp reverse voltages. If the chip doesn't have leading edge blanking you also may want to place an RC filter on current feedback, sometimes this is still recommended with an IC that has LEB internally.

Also maybe swap out the controller if you have a spare that would at least eliminate it as the problem.

All those data sheet application schematics are usually bare bones so they can say look how few components you need. Sometimes they work sometimes (likely) you will need to modify it.

I'm by no means an expert but that's been my experience from some of my previous experiences in SMPS's.

Good luck ;)

Not sure about that grounding path for the LM3478..

Best regards, Spehro Pefhany

--
"it\'s the network..."                          "The Journey is the reward"
speff@interlog.com             Info for manufacturers: http://www.trexon.com
Embedded software/hardware/analog  Info for designers:  http://www.speff.com

I was going to say that the scope signals were noisy, until I saw the drain waveform. Still, even when things are normal, you'll need to watch scope probe ground lead length. Please apply a minimum load when scoping signals - this may avoid other problems that aren't currently of interest (ie heat build-up and death are the real concern here).

You're going to have to clean up the current sensing signal, as the

160mV threshold of the controller won't tolerate a lot of noise. It appears to be reacting to drive noise, as is. R145 won't add much in the way of slope compensation if it is below 1K.

You might also move C11 to the other side of R145 (and reduce it to =100K.

Higher frequency driven oscillation shouldn't be possible if current sensing is latched, and is unlikely if the integrated driver's impedance is in excess of 15R, as per spec.

May need a larger decoupling cap on Vin.

Nat Semi seem to be pretty optimistic in the functional diagram for this device, showing internal analog level shifting on the current sensing signal between pins 4 and 5. I'll believe it when I see it.

RL

Try not to save iterated waveform plots with the same name as previous plots. It's going to confuse things.

Giving the name date and time stamps is the simplest ie gate03280947 is gate plot at 9:47AM on the 28th.

RL

By engine company 89 and two ambulances ... ? :-)

--
SCNR, Joerg

http://www.analogconsultants.com/

"gmail" domain blocked because of excessive spam.
Use another domain or send PM.

You posted a similar thermal problem with the same IC and a similar power train in Sept of last year.

How was that resolved?

RL

"Joerg" skrev i en meddelelse news:5Nrzl.14540$8 snipped-for-privacy@flpi147.ffdc.sbc.com...

Haha ;-) No, actually. By following the suggestions of that time, I switched to using inductors that were better for the job at hand and by picking a MOSFET that did not challenge the driver as much, I was able to get that design rolling. Even without making a single modification to the PCB's. Apparantly now, my luck has changed.

Today was familyday, but tomorrow I'll go in and try all the suggestions that have been set forward since my last post friday evening. Hopefully I'll get something running then. When I go tell the bossman I need extra time and money for redesigning and prototyping this one all over, I would feel a lot more comfortable if I could add a "but now I know how-to" ;-)

Thank you all, who has kindly replied. I value this alot.

Best regards Henrik

don't=20

and=20

then=20

stomach=20

=20

I=20

two=20

large=20

for=20

them=20

european=20

positive=20

is=20

=20

and=20

Well IEEE 315 came to mind, but that is still US centric, a google turns up IEC 60617 for electronic graphical symbols for the rest of the developed world. Asia might have even more standards. .

Hey, Henrik, I hope we didn't scare you away ;-)

Does it work better by now?

-- Regards, Joerg

"gmail" domain blocked because of excessive spam. Use another domain or send PM.