fet gate drive traces differential vs ground plane

I always run a twisted pair out. Doesn't quite apply as I'm always wiring things point to point, but a similar approach on PCB will minimize inductance and stray currents (you wouldn't want to connect the gate drive to somewhere upstream in series with the source current, even if it's on a ground plane).

If I were drawing a PCB, I think I'd run traces on one side out from the gate and source. If the board is thin, running them on opposite sides would work, too.

Tim

-- Deep Friar: a very philosophical monk. Website:

Do you mean "source" where you say "drain"?

Put the driver as close to the fet as possible. If the source is grounded, just use the ground plane for the return. If not, run a fat trace/pour from source back to the driver on one layer, and a fat gate drive trace on another layer.

Don't forget a small series resistor at the gate. If the distance is non-trivial, I'd match the trace impedance to the resistor, ballpark

20 ohms maybe?

A differential pair just encourages ringing.

John

Tim, you're a beast! If your runs were things point to point, but a similar approach on PCB will minimize

;-)

I often breadboard gate drives, or put them on seperate PCBs, so I use a bit of twisted pair to run to the bridge:

I don't have any PCB that I think is the right dimensions to do a really good bridge, so I build it in space with copper strap and loads of bypass. I do actually have some pretty stocky PCB, but it's 'only' 0.015" copper (around 10oz.?), and pretty widely spaced (0.2" I think). I don't think I can bypass that as well, at least not in the canonical parts-inserted-in-PCB method. Seems to work well, although I need it quite a bit slower than the OP in this thread wants (in fact, full speed (80ns) edges seem to be detrimental to it..).

Tim

-- Deep Friar: a very philos>> I always run a twisted pair out.

Okay, that's decent.

Hmmm, FETs or IGBTs? A little squirrely for FETs I'd think, but maybe not as bad for a compound device like an IGBT (what's their Crss like?).

Oh my! You need those bus-bars, yes, fine, and fat copper there, but a separate, local, thin-foil PCB for the gate driver might be nice. You know, if you ever re-do.

Those twisted pairs... Oh, I suppose you've tested it and the local damping resistors save the day, plus the slow switching speed... but it's awfully easy to make FETs sing that way.

Yeah, alright. Grrr.

Cheers, James Arthur

It's perfectly okay to beef up traces by soldering on bus bars. I've seen some ecl stuff with 3/8 x 1/2" solid copper bar for Vee, soldered on plain old FR-4. Works fine.

That might be an easy construction technique--you could augment traces by soldering lengths of heavy bare solid copper wire on top of 'em. Cheap, too.

Cheers, James Arthur

IRG4PC50UD is what I'm using. C's in the low nF range. I don't have the datasheet at hand (I left it downstairs).

The gate drives are printed, so-to-speak... pad-per-hole to be exact. I certainly could shorten the wires from them, but that would restrict my setup right now, which I still want to be flexible.

I don't have a clue if the resistors do anything, I was just told to put them there. ;-) I'm sure on more than one occasion, they at least saved the gate drives from destruction.

There was one occasion when I saw oscillation;

Snubbing solved that, reducing dV/dt. I suspect capacitive coupling across the gate drive coupling transformer, since a similar signal appeared at the oscillator side. The period is approximately the propagation delay of the gate drive.

The gate drive is pretty fast, around 200ns rise/fall time at the gate, which is about right given the total gate charge and resistor. I think it's

50ns when unloaded.

Tim

--
Deep Friar: a very philosophical monk.
Website: http://webpages.charter.net/dawill/tmoranwms

No, I meant you could integrate them all on a local board, right at the IGBTs. Like this:

Q1 Q2

+-------+ +-------+ | O | | O | |-------| |-------| | | | | | | | | | | | | +-------+ +-------+ | | | | | | | | | | | | +---------------------------------+ | | | | | Driver | | Board | | | | | | | | | +---------------------------------+ | | | | | | | | | | | | +-------+ +-------+ | | | | | | | | | | | | |-------| |-------| | O | | O | +-------+ +-------+

Q3 Q4

Make a single-layer board, SMT [1], with the parts and IGBT leads all soldered to the top. Then it's easy to mod, change IGBTs, or even change driver boards. Bottom side = GND.

[1] Old-style parts surface mount easily too, and it saves drilling.

That's clean electrically and easy to work on. You could still fly the heavy stuff with airborne copper, or you could run PCB traces, bolstered with strap or bar or such. That's what I meant.

For later. You know, when you start mass production. :)

Cheers, James Arthur

Could also be the leakage inductance of the transformer.

200nsec isn't "pretty fast" ;-)

--
Regards, Joerg

http://www.analogconsultants.com/

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