Pro/Forums - View Single Post - My micro-channel concept block

Cathar · 09-23-2002, 08:09 PM

Okay, I'll try and clarify where I'm coming from with using such a thin base-plate.

Basically the design is optimised for certain die sizes. In this case the design is optimised for CPU dies up to 15mm x 15mm in size, and as low as 9mm x 9mm. Outside of these ranges the block's performance will be less optimised.

The widths of the channels/walls is just 15mm (17mm if you include the out walls). The baseplate is 1.25mm. The P4/AthlonXP dies are around 11.5 x 11.5 mm on a side, or thereabouts.

A thicker base-plate is actually a hindrance to heat transfer when there's enough micro-channels, walls and surface area to effectively raise the convective heat ratio to be higher than that of copper's conductivity. Meaning that for die sizes greater than 50mm^2 or so, one can actually get away with many microchannels right above the die and a very thin base-plate (no thinner than the channels are wide though) and get away with it.

Like I said, I'm not disagreeing that a thicker base-plate is required for very small heat sources. It is. It's the only way to spread the heat effectively out over a larger surface area.

As a rough rule of thumb with the micro-channel block design, you want your base-plate to be half of the distance from the edges of the CPU die to the two outermost fins for optimal performance. As we get thicker than this, the benefit of the sideways spread of the heat from the copper is overcome by the thermal resistance of the copper itself as we make the copper thicker.

If I was building this block for a 7mm x 7mm CPU die, I'd make the base 3mm thick. For a 1mm x 1mm die, we may as well go open pool, but like I said, CPU's aren't that small (yet). For an 8mm x 8mm die, it'd be 2.5mm thick, and so on.

Now the performance drop-off as we shrink the CPU die is also masked to a large extent by the water-jet coming in straight over heat source. The net result is that the block can quite safely get away with the optimal base-plate thickness for today's CPU die sizes, and still offer close to optimal performance for tomorrow's 88mm^2 CPU die sizes (T'Bred) and 100mm^2 (Barton).

It's really all about trade-offs and optimising for a certain task. Sure, we can build a block that tries to be a jack-of-all trades, and will work well almost regardless of what sizes heat die we stick under it, but that wasn't my design goal. My design goal was explicitly to get the best overall performance for 9mmx9mm to 12mm x 12mm CPU dies and any combination of sizes in between.

I've had people ask me about peltiers in other forums, and the answer is no. This block is not aimed at peltiers at all. That's not its goal. It's aimed at providing close to the best possible performance for straight water-cooling of a CPU die similar is size to those being used today, and planned for the next year.

This is why the channels are low, narrow, and the number of channels isn't that high. All aimed to boost water velocity through the block as low and as close to the heat source as possible, and focuses around maximising the surface area directly above die, and most importantly, cooling the VERY hot dead center of the CPU die.

On of the things I had my simulator do was track the hottest location of the die, and what it's temperature was. The centre of the CPU die is always hotter and in fact by up to 3 or even 4 C over the edge of the CPU die. As we raise the base-plate thickness, we can actually get into the situation where the overall temperature of the die can drop, but the hottest spot actually gets hotter, depending on the location of various channel walls, etc.

There are many balls being juggled here, but I still firmly stand behind by decision to go with a 1.25mm base-plate on the basis of it being optimised for its intended target application.

09-23-2002, 08:09 PM	#12
Cathar Thermophile Join Date: Sep 2002 Location: Melbourne, Australia Posts: 2,538	Base-plate thickness debate Okay, I'll try and clarify where I'm coming from with using such a thin base-plate. Basically the design is optimised for certain die sizes. In this case the design is optimised for CPU dies up to 15mm x 15mm in size, and as low as 9mm x 9mm. Outside of these ranges the block's performance will be less optimised. The widths of the channels/walls is just 15mm (17mm if you include the out walls). The baseplate is 1.25mm. The P4/AthlonXP dies are around 11.5 x 11.5 mm on a side, or thereabouts. A thicker base-plate is actually a hindrance to heat transfer when there's enough micro-channels, walls and surface area to effectively raise the convective heat ratio to be higher than that of copper's conductivity. Meaning that for die sizes greater than 50mm^2 or so, one can actually get away with many microchannels right above the die and a very thin base-plate (no thinner than the channels are wide though) and get away with it. Like I said, I'm not disagreeing that a thicker base-plate is required for very small heat sources. It is. It's the only way to spread the heat effectively out over a larger surface area. As a rough rule of thumb with the micro-channel block design, you want your base-plate to be half of the distance from the edges of the CPU die to the two outermost fins for optimal performance. As we get thicker than this, the benefit of the sideways spread of the heat from the copper is overcome by the thermal resistance of the copper itself as we make the copper thicker. If I was building this block for a 7mm x 7mm CPU die, I'd make the base 3mm thick. For a 1mm x 1mm die, we may as well go open pool, but like I said, CPU's aren't that small (yet). For an 8mm x 8mm die, it'd be 2.5mm thick, and so on. Now the performance drop-off as we shrink the CPU die is also masked to a large extent by the water-jet coming in straight over heat source. The net result is that the block can quite safely get away with the optimal base-plate thickness for today's CPU die sizes, and still offer close to optimal performance for tomorrow's 88mm^2 CPU die sizes (T'Bred) and 100mm^2 (Barton). It's really all about trade-offs and optimising for a certain task. Sure, we can build a block that tries to be a jack-of-all trades, and will work well almost regardless of what sizes heat die we stick under it, but that wasn't my design goal. My design goal was explicitly to get the best overall performance for 9mmx9mm to 12mm x 12mm CPU dies and any combination of sizes in between. I've had people ask me about peltiers in other forums, and the answer is no. This block is not aimed at peltiers at all. That's not its goal. It's aimed at providing close to the best possible performance for straight water-cooling of a CPU die similar is size to those being used today, and planned for the next year. This is why the channels are low, narrow, and the number of channels isn't that high. All aimed to boost water velocity through the block as low and as close to the heat source as possible, and focuses around maximising the surface area directly above die, and most importantly, cooling the VERY hot dead center of the CPU die. On of the things I had my simulator do was track the hottest location of the die, and what it's temperature was. The centre of the CPU die is always hotter and in fact by up to 3 or even 4 C over the edge of the CPU die. As we raise the base-plate thickness, we can actually get into the situation where the overall temperature of the die can drop, but the hottest spot actually gets hotter, depending on the location of various channel walls, etc. There are many balls being juggled here, but I still firmly stand behind by decision to go with a 1.25mm base-plate on the basis of it being optimised for its intended target application.