Two-phase chip cooling with manifold-capillary structures enables 10⁵ COP
10 comments
·May 1, 2025j-pb
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.
crote
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?
ajb
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
Moto7451
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.
hinkley
We don’t bond heat plates to the top of CPUs due to the coefficient of thrermal expansion right? Isn’t that why we need paste?
Because of CoTE was a non issue I would think we could dig heat dissipation channels into the top of chips, maybe Serpinski gasket style. A little proper lapping and you’d get an airtight seal with the heat sink.
Retric
> plug directly into server boards
Because traditional water cooling works better, and air is cheap.
JonChesterfield
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.
null
Easier to access at https://www.cell.com/cell-reports-physical-science/fulltext/...