basically, it's like any other themal problem ..
From that datasheet , it says:
Power Rating @ 25 ?C Case Temperature = 35 W
So you can get 35W from it only if the case stays at 25 ?C.. But that won't happen without getting rid of the heat generated, using a heatsink.
The datasheet also says the thermal resistance is Rthjc 3.7 ?C/W, and the maximum temperature is 155 ?C, although it's not clear if that is case temperature or junction temperature. Lets assume junction temp and be safer.
so the 25 W that you want will make the thermal difference across the junction-case be 25 W *3.7 ?C/W = 92.5 ?C. That's going to be
the temperature rise above ambient temperature at 25W. So if the ambient air temperature is 25C, the case temperature will be 117 ?C at 25W.
That's hot, but still within spec. ... you would think .... however, there's a gotcha.. there's a derating curve.
At 117 ?C the maximum power has to be derated by 40% , so it can't really handle 25W at 117 ?C, it can only handle max*40% = 35*40% = 14W max at that temperature. So it's clear from the datasheet that you can't operate at 25W without a heatsink.
So, what is the maximum temperature at 25W? figure it out using the derating curve => 25W is what percentage of 35W? about 72%
Now look at 75% derating, find the temperature, that's the maximum temperature you can allow at 25W. It's about 85 ?C. So shoot for 80 ?C and you'll be safe.
So without a heatsink, you'll be at 117 ?C, but you need the case to be at 80 ?C. You need to get rid of 117-80 = 37 ?C at 25W.
37 ?C/25W = 1.48 ?C/W . So you need a heatsink for the dpak that can at least achieve 1.5 ?C/W. That will keep your case temperature around 80 ?C.
Also keep in mind the TCR value. any temperature rise on the device above 25C means it's value will change, +-100PPM / ?C
If that's important you need to consider the change in resistance due to this heating. If you need less change in resistance at that power level, you need a really big heatsink, to get rid of 117-25 = 92 ?C .... 92 ?C/25W = 3.7 ?C/W.. not surprising, this is the Rthjc from the datasheet. If you have a heatsink that can get rid of that temperature rise, your junction stays at 25C and the resistance is stable.