[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: DRSSTC thoughts...



Original poster: "Jim Lux" <jimlux-at-earthlink-dot-net> 

 > Yeah exactly.  They also didnt strive to minimize stray L inside the
 > bricks either.  Greg Leyh wrote an article on this... cant remember
 > where to find it though.  As we push these IGBTs harder and harder the
 > little things like stray L inside the IGBT will add up quick!  And
 > heck, we'd be doing something bad if we dissipated over 50W per IGBT
 > (unless it was a REALLY big system).  Even with making 11 footers, my
 > IGBTs stay cold to the touch.


But guys (and gals), you're designing RF power amplifiers now.. maybe you
need to add some C to the system to cancel the reactance of the L.  This is
the whole matching thing and, aside from "minor details" like thermal
management, it's what makes it a challenge (speaks someone whose office mate
just fried some DEI/Ixys kilowatt FETs).  Not having looked into the design
problems, I can't say whether it's even feasible to match these devices in
that way, but dealing with package parasitics has been an aspect of power
amplifier design from the days of the first Audions.


 > Yeah, that certianly is impressive.  And it seems that you could take
 > more control of problems with stray L.  Then again, how to minimize
 > inductance on a layout consisting of so many individual devices is
 > quite a feat!

Do you actually need to minimize it, or can you go towards a "distributed
amp" kind of topology, using the parasitic L with appropriate networks to
make a transmission line.  Don't switch them in parallel, but stagger the
gate signals and drain connections.

 >
 > Oooh, now THAT can get interesting.  Talking to Jimmy Hynes, the idea
 > came up to make our own bricks with some dies, but instead of
 > designing for lots of power, design it to withstand very high
 > temperature transients.  That is, put heatsinking on BOTH sides of the
 > die to remove heat even faster.  Or even just some sort of small
 > aluminum block on top of the dies to wick away heat faster... then IT
 > can be dealt with via forced air cooling.  I wonder if they come in
 > 1200V versions :-).

Or consider other substrates to bond the die to (AlSiC or CVD diamond?).
Thermal management is going to be the key.



 >  >
 >  > Also note that the primary cap strings could be distributed between the
 >  > IGBTs to force current sharing like the OLTC does if that would help:
 >
 > Not sure if there would be much reason for that one, though if we were
 > paralleling a bunch of small IGBTs it may somehow work.

If you're going for a distributed amp kind of architecture, then it's a
natural development.  Think in terms of the stacked Blumlein pulse
generators.