Simulated Die-IHS-WB
 

Simulated:
Silicon die: 10 mm sq., 0.8 mm thick, 100 W distributed uniformly on the bottom.
TIMa: 35 W/(cm^2*K), based on Arctic Silver's specs
Copper IHS: 32 mm sq., 1 mm thick
TIMb: 20 W/(cm^2*k), based on ph and Cathar data
Copper Waterblock: 64 mm sq., 6 mm baseplate, h(eff)=24k over a 50 mm sq. on the top
Ambient (water): 300 K

The darn thing crashed just after the sim run, but before it ordered and grouped the data like I had wanted. 253,798 data points are sitting in no particular order.

I have a map of what's where, but it'll be a pain to assemble the bits into coherent datasets. I'm planning on these maps: temperature of the bottom of the die, heat flux at both TIM joints, temperature at top of the block (where it transitions to abstracted convection).

What do you want to see?

Top view of the 10 mm square thermal interface between the silicon die and the copper heat spreader. Heat flux density perpendicular to the plane of the interface is shown.
http://pages.sbcglobal.net/water.gro...-tima-flux.gif

I think these maps are a good choice.
Look forward to them.

I take it that this h(eff)=24k is equivalent to my h(bp/Fin Interface); mentioned here .As such I have no quibble with the value.

In my opinion,Billa and Incoherent have the only data from which an estimate of a TIM interface's Conductance (w/m^2*K) can be calculated.
Yes, pH's and Carhar's results can be modelled using values calculated from Billa's and Incoherent's data, but they are not the source of the values.

. I suspect that AS's 35 W/(cm^2*K) Thermal Conductance is the Conductance of the TIM material and not of the TIM interface (My 10 W/(cm^2*K) was chosen to allow for Zero(0) Contact Resistance where as AS's 35 W/(cm^2*K) require Contact Resistance adjustment)

Apart from the above quibbles any simulation will be interesting and educational

I take it that the "ticks" are 0.1mm and this is the spreading predicted within the 0.8mm of Si ?
On initially reading of objectives was not certain that you were including a Si layer( I easily confuse "top" and "bottom")

Is it just me, or does the scale of the graph in relation to the colors/heat-flux appear to be inverted? It appears to be telling me that the very hottest sections of the IHS are at the very corners?

Think they are correct .
Think maybe confusing Heat-Flux and temperature.
Note there is possibly Typo in Heat-Flux units
Heat-Flux should be w/m^2, unless it is h(eff) being shown and not Heat-flux.
Nomenclature and units are a pain.

Dammit! fixed the units on the graph.

Yep, same sort of thing. It abstracts the top .5 mm of the baseplate plus the pins/fins. It still needs some work, but accounts for the net convection with less overhead.Something to fight over when I've had some sleep.

I guess I could have included some explanatory text with the picture....I'll insert some in a bit. Done!

What it's saying is that a significant portion of the heat leaves the die at the corners/edges, where the IHS is a bit cooler. I hadn't thought the silicon of the die would show much lateral heat flow, but there it is, a concept in need of pondering and experiments.

The underside of the die, where all the transistors and their interconnects live:

[img] Image superceded! See below[ /img]

New and improved, now with numbers by the axes!

Ambient(cooling medium's temperature)?

Edit: Would also like Cu and Si Conductivities used - for Waterloo predictions of the system. Please.

I "get it" now. I just got a little confused as to what I was looking at was representing. Basically saying that it's easier for the heat to leave at the edges of the die than in the middle, and I wholly agree - matches up well with my own models.

Now featuring temperatures for the top of the die!
http://pages.sbcglobal.net/water.gro...w-die-temp.gif

Ambient (water) is 300K; I used 385 W/mK for Cu, 148 for Si.

Last bit before I head out: The temperature difference between the underside and topside of the die.
http://pages.sbcglobal.net/water.gro...ie-delta-T.gif
Quite reasonable, the corners shed heat easier.

A DeltaT of ~3c seems a liitle low for a 0.8x10x10mm slice of Si
My simple "DeltaT= Resistance x Flux" puts it at ~ 5.4c.

My play with Waterloo predicts a "Die Underside" of ~ 330.4k
(As always, with my beermat calcs, subject to checking)

Edit: All sums done using your parameters

Same system without IHS (using 20 W/(cm^2*k), TIM) gives a "Die Underside" of ~322.5k.
This copper waterblock has: h(IHS eff) = 25,8122 w/m^2*c and h(die eff) =58,396 w/m^2*c

This is why I like you guys.

Hmmm, I'll have to pick through my parameters to check for errors.

Error
Should be ~5.4c
Dunno why, but calculated for a !mm slice.
Have corrected "Die Underside" temps in accordance.

Found an typographical error in the scripting of the die, apparently a zero is not a worthy substitute for a decimal point. Net effect was 13% increase in thermal conductivity in the z direction, and 64% laterally. Yeesh.

Crunching again, pictures to be updated.

Groth, do you think it would be possible to plug in some numbers to check the flux density through the cross section of a narrow heat channel? Narrow as in a flux block, replacing the IHS with a 10x10x10mm block sitting on the die. I would like to see if the flux block numbers are valid with regards to power density and would be curious to see what your model says.

Can do. I'll build a flux block after I re-run the 'IHS vs not' with the corrected die. Varying degrees of non-uniform heating are on the agenda, too -- be interesting to couple them with the flux block.

Very much so, for the interpretation of "Flux-block on cpu " data.
With present uniform heating, a Flux-blockr replacing the 0.8x10x10mm Silicon die and eliminating the IHS may possibly describe a "die simulator" situation.

Revised to : 326.6K
Double counted C/W(wb bp +TIMb + I/h(eff bp))

Does not effect the no IHS sums and prediction remains at ~ 322.5k

Oh woe. Wish had kept out of this.

Great. Look forward to it.

Hoping I've finally got it right :

http://www.jr001b4751.pwp.blueyonder.co.uk/Groth4.jpg http://www.jr001b4751.pwp.blueyonder.co.uk/Groth5.jpg

Face it, we're addicted to numbers. :cry:

Added Waterloo predictions for a 1mm baseplate version.