From: Winfield Hill
Subject: Re: Question about linear derating factor for MOSFETs
Date: 11 Nov 2002 08:14:45 -0800
Organization: Rowland Institute
X-Newsreader: Direct Read News 4.10
John Larkin wrote,
> Winfield Hill wrote:
>> John wrote...
>>> ... we tested a lot of TO-247 power fets to destruction. Bolted
>>> to a solid copper block, it was hard to find one that could stand
>>> 350 watts dissipation for 150 milliseconds (our target), even for
>>> parts rated for a lot higher DC dissipation.
>>> Good parts: IXYS IXTH11P50 P-channel
>>> IR IRFPS37N50A N-channel
>> Yep, 150ms is a long time to a semiconductor. Glancing at the
>> 37N50A's Effective Transient Thermal Impedance curve, fig 11, it's
>> clear that 150ms is nearly the same as infinity to the junction.
>> It's interesting that IRF touts "low gate charge" for the 37N50A,
>> even though the FET's gate capacitance is a staggering 5600 pF.
>> I suppose we're to be thankful it's not 15,000pF. :-)
>> However, one should note that older parts with higher capacitance
>> also had much bigger dies, and therefore lower thermal resistance.
>> One should seek out these older parts for maximum transient-power-
>> handling capability. While newer more compact FETs are appealing
>> for easier fast switching and less switching loss due to their
>> low-gate charge, the are dramatically less suited to non-switched
>> applications where high heat flux is the name of the game. For
>> semiconductor companys, the name of the game is reduced silicon
>> area, more dies per wafer, lower manufacturing cost, more profit.
> yeah, the actual junction thermal time constants are pretty short.
> But there is a (complex) composite time constant that is the junction,
> the small heat spreader in the package, and the copper (or whatever)
> in the heat sink itself.
I have made 2D models of the FET die, the header tab and the immediate
attachment portion of the heat sink, using the appropriate material's
thermal resistance and thermal mass values. I calculated with spice.
This revealed that for the first 50ms or so the heat was spreading
through the FET's junction and tab, and only penetrating into the
initial massive heat-spreading portion of a typical heatsink. So it's
likely a block of copper or aluminum alone could serve for our O.P.
> Too bad the serious thermal simulation software is so expensive.
I agree completely.