Aproach to ltc3703 killing circuit

[Amper]

Hi!
Im currently working on the Buck converter part of a larger circuit, but im having some trouble with it.

First i took the example circuit from the datasheet and assembled it on a perfboard. I tried to reduce stray inductance as far as i could and generally sorted everything in such a way it should be fine at the low 100kHz it would run for the first tests. Sadly it was killing the high side driver of the ltc as soon as it started switching.Rising the input voltage slowly and limiting the current did not help, protecting the driver section from overvoltage, positive o negative didnt have any effect.

Now i finally have the board with the final high current components, the logic part was rebuilt completely on perfboard again. and stacked on top for testing. I hoped that i just made a simple mistake in assembly, but in this entirely new setup exactly the same is happening.

My Problem is, that the LTC will not survive long enough for any measurements to be made. Every shot costs about 5€, which is starting to be the most expensive part of the projects, i think by now i ruined 10 of them trying to ind a mistake.

Does anyone have an idea on how to approach this? Im not really asking for an exact solution to my problem, more like advice what to do in such a situation.

Hopefully some of you guys can help.

cheers

[MosherIV]

Hi.

The fact that every perfboardbuild has failed suggests that the circuit is not suitable for prototyping on perfoard.

There is a reference board from Analogue devices but it is expensive and I doubt that AD will give one to a hobbiest with no prospect of selling more than a handful of devices.

Your only other appraoch would be to look for cheap chinese PSU that use this device and modify it for your needs, the chinese boards will never live up to the head line specs anyway.

[Amper]

Off cause perf board will limit it a lot, but at 100khz it should still at least run without destroying itself under no load at all. Modification is no option since the entire design is fixed except the buck control part on the second board. The final one will be a 4 layer pcb with many other functions, but making that one as a test circuit will be expensive, even coming from china, so i at least want the thing to not instantly die before i order them.

[ogden]

There is a reference board from Analogue devices but it is expensive and I doubt that AD will give one to a hobbiest with no prospect of selling more than a handful of devices.

They give design files of reference board instead - so hobbyist can replicate PCB.

[Niklas]

A picture says more than 1000 words. Can you post some pictures of the test setup? Wire length, loop sizes for input/output?

If the IC dies right away it is not likely to be from overheating. The switch frequency is not the important parameter, the frequency related to the rise time of the switch sets the operating point.

[Amper]

Here are some pictures. The large board is hard to solder on, since its home made thick copper aluminium and the soldermask doesnt like high temperatures, so please excuse the horrible looks ^ ^

I tried to minimize any long wires. Some are there, but they are mainly supply lines and feedback is held away from current carrying ones. Most important stuff is filtered right next to the pins with ceramics anyways.

[dmills]

A Modern switcher on stacked perfboards? Good luck with that!

Dead bug the thing on a bit of copper clad (as the ground plane), you will likely be much happier.

My random guess would be ringing on the switch node possibly killing the top gate driver, but who really knows, it could as easily be the bottom gate loop causing ground excursions.
If you are only running at 100KHz, then adding some gate resistance (Maybe 5.6 ohms or so) may slow the switching edges enough to make it better behaved at the cost of some efficiency, but really any of those modern parts mean it when they ask for a ground plane.

Regards, Dan.

[Amper]

Probably i will have to make a better board, you are right. Hopefully i can get away with single layer to make it myself. I didnt add gate resistors because my fets already have internal 1Ohm and the ltc3703 does current measurement with the bottom fet, so i didnt want to deviate more from the datasheet than i had to. I will try it though as soon, as i have new parts. Unti then i think i will try to feed a signal from some other source and check, if there is any problem happening directly on my power board. I just finished the gate driver for the 3 output phases of the inverter part, they have even more horrible cabling loops and run of the same 15V supply, but at 100kHz the signals all look beautiful even though it was only meant to do 10kHz. Maybe i will try to throw in an external high side driver next time i try. They will be cheaper to replace 

[MosherIV]

but making that one as a test circuit will be expensive, even coming from china, so i at least want the thing to not instantly die before i order them.

I think you did not understand what I was trying to tell you.

Look for a pre-made PSU based on the same SMPS controller that you want to use.
Then, you can modify the per-build circuit as you need.

[Amper]

Mh yes, i read that the wrong way. Though copying another board doesnt really work either because i have to split it up in power and control part. Otherwise i will either have cooling problems or i would have to put everything onto the single layer aluminium pcb.

Since the pinout by linear is as usual very convenient there is not much variation i can do from what i already did. The only large difference is the groundplane under the circuit, taht could really help much, if it is actually inductance between the wires.

EDIT:

btw, linear is (was?) very generous at least in germany. I once asked them for a sample of the LTM8050 uModule, they usually dont give samples of them, he was basically like "Will 5 be enough? Would you like a demo board with them?" So they sent me somewhere around 150-200 bucks of stuff for free. Maybe i need to ask them again... Hopefully they didnt change since they got sold.