snipped-for-privacy@netfront.net ---
Not necessarily. We're within a factor of two, which isn't bad for thermal stuff like this.
I was assuming roughly 150K/w per square inch of surface, maybe too conservative, but it depends hugely on orientation, convection paths, stuff like that. Lacking the computational tools to do this sort of thing, I like to build mockups and test them. Cardboard. Duct tape. Dummy load resistors.
John
The circuit in question would be inside a biggish laser, and they probably have cooling water available. A small copper block could have some water flowing through it, through plastic tubes, and we could bolt the fets to the block. You can buy things like that, "cold plates", or make it. We could arrange for the voltage to be, say,
-1000 volts for short pulses at very low duty cycle, tens of ns, so the conductivity wouldn't be a big problem.
Of course, we still don't have a circuit that works, so cooling is sorta academic.
John
snipped-for-privacy@netfront.net ---
Oh, yes, of course. I only use my rule-of-thumb to get a "feeling" of where I am at in the approach. I can then decide how much attention it needs.
John
Not sure what exactly the problems are but just as an idea, could LDMOS and an RF transformer work? PolyFet supplies SPICE models for their LDMOS parts which is really practical for this sort of stuff. NXP doesn't for theirs, for reasons I'll never understand.
-- Regards, Joerg http://www.analogconsultants.com/ "gmail" domain blocked because of excessive spam. Use another domain or send PM.
snipped-for-privacy@netfront.net ---
Hi,
The power will be limited based on max operating current as well as PCB temperature. The 50watt number is pretty conservative based on some ltspice sims I did for the switchers. I think I will settle on a 2layer design that could be built in FR4 or aluminum and derated if necessary. The board is meant to go into a standard ATX PC case (5.25" bay) so there is some cooling available with case fans. In the sims I have the boost running at 800kHz (12VDC to 13.5V at 800watts) and the buck at
350kHz (30~90V to 13.5V at 400Watts). The high frequency is why I'm using SMT mosfets instead of heatsinked TO-220s.Here's the SMT fet used on the boost converter:
The junction to PCB thermal resistance is far lower than the junction to top of case for these SMT fets, so that is the main reason I was thinking about an aluminum PCB.
Not for a rocket, this is an alternative energy project "ATX power tower" :)
cheers, Jamie
--- Posted via news://freenews.netfront.net/ - Complaints to snipped-for-privacy@netfront.net
800kHz is a bit high for that kind of power. But you are probably restricted by the size of the magnetics. The proverbial rock and the hard spot. What about making the catch diode an active rectifier? That ought to take a ton of heat out of the equation.
Is that a sync buck? (it should be)
2mohms is nice but it'll probably burn north of 10W between Rdson and switching losses. Maybe use more of these?And here I thought this would be one of those hotrod projects :-)
-- Regards, Joerg http://www.analogconsultants.com/ "gmail" domain blocked because of excessive spam. Use another domain or send PM.
We need to put several kilovolts into an E/O modulator crystal, with rise/fall times of a few ns. One common way to drive these things is to makeseries stacks of mosfets with transformer-coupled gate drive. That gets ugly for lots of reasons. I'm playing with using something like an 800 volt beast of a mosfet and a step-up transformer. I'd love to use a tapered transmission line for voltage gain (I've been wanting to find a use for that for years now) and/or a distributed amplifier (which I've played with, without much luck, except to make oscillators when I wanted amplifiers.)
Avalanche transistors are probably out for several reasons.
The LDMOS parts all look to be low voltage devices.
Motorola used to make some screaming mosfets, the TMOS7 parts, but they are discontinued. If you go to the OnSemi site, and click on a mosfet datasheet, you get a Sanyo part!
IR seems to stop at around 300 volts. Wimps.
ST makes some nice fets. STD6N95K5 is stunning: 950 volts, 1 ohm, and
*** 30 PICOFARAD *** drain capacitance.Who else makes high-voltage power mosfets?
I wish the GaN fets ran at higher voltages, and weren't so expensive. We blew up almost $1000 worth of Nitronex parts one afternoon.
John
snipped-for-privacy@netfront.net ---
-- I think your board size (and lousy connector density) is going to waste a lot of money, in AL core. Better channel the heat into heatsinks. RL
al
ower
eo make
seem
t 15to
sth
Think about a heat pipe - the heat transfer can be pretty amazing. When we used one for the Peltier cooler, we had to check each one that came in to make sure that there wasn't any air inside the heat pipe, by measuring thermal resistance at fairly low temepratures when the vapour pressure of water isn't that high - provided that there isn't any uncondensable gas inside the pipe, the water vapour can move at pretty close to the speed of sound and shift at lot of heat even at low vapour pressure.
For the first batch or two we ended up sending back one in three, before the supplier cottoned on and set up his own test rig.
-- Bill Sloman, Nijmegen
Vishay?
-- Failure does not prove something is impossible, failure simply indicates you are not using the right tools... nico@nctdevpuntnl (punt=.) --------------------------------------------------------------
Of course! Siliconix and such. Thanks.
John
If there is system water in the laser, and there probably is, I can just flow water through a cold plate. Much simpler than a heat pipe. I wouldn't even need a radiator... just take in cold water and give back warm water.
John
power
make
seem
Yes, but those are screaming fast. Making a "few nsec" transition happen with ordinary FETs could turn out to be a challenge.
Check out IXYS, see if they'll let you have tasters of these. IME the switching times in FET datasheets are often not too meaningful:
Ouch. Drown it at Zeitgeist with the guys, it'll only increase the total tab by 10-20%, would be in the noise :-)
-- Regards, Joerg http://www.analogconsultants.com/ "gmail" domain blocked because of excessive spam. Use another domain or send PM.
te:
ional
ee power
in
the
gh
e to make
't seem
ld
It
heat
m.
ut
ch
as 15
er to
yers
e a
ith
with
Heat transfer by evaporation and condensation is actually more effective than heating up a bunch of liquid water in one place and then moving it somewhere else. Water vapour moves faster than liquid water, and the heat is being shifted as latent heat of condensation, which shifts a lot more heat per gram.
If you probelm is getting heat out of a confined space, the heat pipe can probably move a good deal more heat through the same cross- sectional area.
Transfering heat out of a surface by letting it evaporate water gets around the boundary layer problem that you get when transferring heat into a fluid that is cooling by forced convection.
It is an interesting inter-disciplinary exercise to get your head around it - you need to know some thermodynamics and some fluid dynamics before it starts making sense.
-- Bill Sloman, Nijmegen
This is a challenge. AFAIK a feasible Miller slope of a single FET is slower by 100 times or so. You are pushing the physical limits.
May be a comrade Marx generator approach?
That should be very special transformer.
A vacuum or a gas discharge device?
This device is still about 50 times slower then needed, isn't it?
Vishay, Siliconix, IXYS, Avago, Hitachi, Tochiba.
Vladimir Vassilevsky DSP and Mixed Signal Design Consultant
ote:
tional
re
he power
in
n the
igh
ke to make
n't seem
uld
It
heat
rm.
out
ich
as 15
per to
ayers
rke a
with
k with
eat
ed
vishay, onsemi and fairchild has a few in ~1000V range
-Lasse
Yup, coax-based transmission-line types. Easy to make.
I don't really understand mosfet datasheet rise/fall times. People routinely get speeds 5 or 10x what datasheets suggest. All you need to do is drive the gate really hard. We get 60 volt pulses into 50 ohms with 2 ns rise/fall times, through a transformer, from a couple of
2N7002s.Win Hill has done a lot of this sort of stuff, using fairly ordinary fets to switch a kilovolt in a few ns. Maybe I'll pester him for suggestions. But I shouldn't distract him from The Third Edition.
I once used a pair of the TMOS7 parts, NTP15N40's, to switch 400 volts and 50 amps in a few ns. Had to add gate resistors to slow them down. Pity they don't make them any more.
John
power
make
seem
One problem with a heat pipe is that the return water path is a wick, which will be slow. Another problem is that you still have to dump the heat somewhere, radiators and fans and such. In an open-loop water system, I don't have to deal with the heat. It's not my problem.
Are heat pipes reliable these days? I'd read about problems in the past. I've never used one myself.
1 GPM of water flow has an effective theta of 0.0037 K/W. That's an impressive number. Which is why electric water heaters are an expensive way to take long, hot showers.The wall-plug power of the laser in question is around a megawatt. They wouldn't begrudge me a liter per minute of water.
John
In the language of zurgs it is called "...psst..BAM!"
Rgate, Lgate, Vgs limit, Igate limit, power dissipation in the gate limit.
I don't think the slew rate of a naked FET could be higher then 1e10 V/s. The 1e9 V/s is easy and it is a very practical value.
DirectFETs from IR are neat; I had a good experience with them. The actual switching speed was limited by recovery of the body diode.
Vladimir Vassilevsky DSP and Mixed Signal Design Consultant
The really big problem is dI/dT into the source lead inductance, which messes up the relative gate voltage and forces you to use huge gate drive voltages. Pity that mosfets have the drain on the chip/case, and not the source. Gaasfets are source-on-chip, which makes them a lot easier to drive hard.
This is 50 volts in 2 ns, which is 2.5e10, into 50 ohms through a 1:1 transformer, using 2N7002s:
ftp://jjlarkin.lmi.net/HV_mosfet_pulse.jpg
We have another box that does 100 volts in about 1.5 ns, which is about 7e10. 1000 volts in 10 ns should be possible, which is 1e11. Maybe even 5 ns. You just have to bash the gates hard.
They are appealing because of the low source inductance. But they don't seem to be available in high voltage.
I'll have to look at the DEI parts. They have super-low inductance.
John
ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.