Two-phase chip cooling with manifold-capillary structures enables 10⁵ COP

https://www.sciencedirect.com/science/article/pii/S2666386425001195

by PaulHoule • 1 year ago

22 10 1 year ago

Easier to access at https://www.cell.com/cell-reports-physical-science/fulltext/...

j-pb• 1 year ago

Given that phase change captures a ridiculous amount of energy (it takes roughly 5 times as much energy to go from liquid water to steam than it takes to bring that same water from 0C to 100C). I've always wondered why we don't optimise CPUs for running at ~101C.

And as long as there is material to phase change the tempersture will be fixed at the phase change temperature, so boiling water will never go above 100C at 1bar.

ajb• 1 year ago

It's a lot easier to optimise your phase change fluid for a temperature range than to add another constraint to chip design.

Anyway, apparently fluid in heat pipes work at a range of temperatures not just at the exact temperature of phase change

What I always wondered is why we didn't get flexible heat pipes that plug directly into server boards, leading to some centralised heat exchanger or cooling tower, rather using the air as a transfer medium

Retric• 1 year ago

plug directly into server boards

Because traditional water cooling works better, and air is cheap.

Moto7451• 1 year ago

Heat pipes have a vacuum so phase change happens at lower temperatures than normal Earth atmosphere. This essentially is how you do what you’ve proposed without having the CPU makers change their designs to run hotter. If you boil water at low pressure and 30C you have the same net effect as boiling itself uses huge amounts of heat energy compared to raising the temperature of water one degree. This is the same energy at any temperature level.

On flexibility I think it’s just that vacuum and the wicking structure that limits the materials. I’m not a material science expert by any means but I have to imagine someone has worked on this.

JonChesterfield• 1 year ago

Lots of silicon degradation modes get worse with increasing temperature. Optimising for 100C, laptop style, roughly means running low clocks, low volts, expecting shorter lifespan.

crote• 1 year ago

Thermal resistance, for one. There's always going to be a temperature gradient from the hottest hotspot to the evaporation surface of your cooler. If you want your cooler to be at 100C, your hotspot is going to have to be significantly hotter: can the chip handle running at 150C, or 175C?

librasteve• 1 year ago

i really like this idea … two supportive arguments are (i) you want to run your trannies hot vs. ambient to drive heat flow [they already run at 80-100°C] and (ii) you can use heat pipe vacuum tech to fine tune the capillary transition temp and to produce a lower temp arterial cooling circuit.

Two-phase chip cooling with manifold-capillary structures enables 10⁵ COP

Related Stories