As described by the lab, the Sandia cooler predominantly transfers heat across an air-gap (maintained via self-generated fluid bearing) between the base plate and rotating impeller.

As described, they claim that if they can maintain the air-gap to 10um, then they get a net win since the specific (per surface area) thermal resistance is sufficiently low (due to only being 10um), and the transfer surface area is comparatively enormous. Nevertheless, the air-gap still forms the dominant component of their overall thermal resistance.

Reading through their posted development report [0], it's apparent that they were having trouble hitting that 10um target with their designs.

Not exactly a surprise that this is the most difficult part. Not surprising that the "commercialized versions" (see other posts) don't use this mechanism.

[0] https://ip.sandia.gov/techpdfs/Development%20of%20Sandia%20C...

Right. High thermal gradients through ball bearings or roller bearings are tricky, and you don't get much contact area for thermal conduction in the bearing (meaning most bearing applications need to be mindful of cooling). (Pro tip from MIT 2.72: make sure to position bearings such that thermal stress from the shaft undergoing more thermal expansion than the rest of the machine doesn't jam the bearings.)

I guess you'd want a bearing with sodium-potassium alloy (hazmat!) or mercury (hazmat!) or something. Or... make the casing for the shaft into a heat pipe with ammonia or alcohol vapor.

What about old style Babbitt bearings

You're right. Plain bearings provide plenty of contact surface for thermal conduction. Though, thermal expansion will tend to increase clearance and decrease contact area (ignoring differences in thermal expansion between dissimilar materials).

Perhaps the CPU is rotating too? Wait, perhaps I should patent this idea first.

The amazing part of this is that each individual core makes contact with each region of the (obviously stationary) memory drum once per revolution, providing dynamic cache adjacency.

Mel nods in approval: http://www.catb.org/jargon/html/story-of-mel.html

Make sure to incorporate an anti-twister mechanism for extra weirdness points: https://en.m.wikipedia.org/wiki/Anti-twister_mechanism

A linked video claims that the commercial version attempts to transfer heat from one rotating metal surface to another (the the part that connects to the CPU and the bottom of the spinning thingy) separated by an extremely thin layer of air. This is "supposed to" transfer heat as if the metal is touching but "doesn't see to work very well in practice"