The Fanless Spinning Heatsink

An anonymous reader writes "There's a fundamental flaw with fan-and-heatsink cooling systems: no matter how hard the fan blows, a boundary layer of motionless, highly-insulating air remains on the heatsink. You can increase the size of the heatsink and you can blow more air, but ultimately the boundary layer prevents the system from being efficient. But what if you did away with the fan? What if the heatsink itself rotated? Well, believe it or not, rotating the heat exchanger obliterates the boundary layer, removes the need for a fan, and it's so efficient that it can operate at low and very quiet speeds. That's exactly what the Air Bearing Heat Exchanger, developed by Jeff Koplow of the Sandia National Laboratories, has developed. It's even intrinsically immune to the build up of dust and detritus!"

Comments

Re:I'm curious...

[canajin56]

Yes, a layer of air does form between the heat spreader base, and the base of the rotating heatsink. This is called an air bearing. It's extremely thin, and for that reason an excellent thermal conductor even though it's conducting heat poorly. You see, it has a surface area of 100 cm squared, but it is less than 0.03 mm thick. So, heat transfer is inefficient, but its so thin as to be negligible.

And no boundary layer forms (well, it does but it is reduced by a factor of 10) on the fins because they are rotating. The equations for fluid dynamics are quite different between an inertial reference frame and a rotating one. Basically, the fluid cannot settle into little pockets because the (fictional) centripetal force is pushing it outwards along the fin channels.

Re:Still has a boundary layer.

[idontgno]

Good Lord. Have your psychiatrist adjust your dosage.

As is the case in a conventional "fan-plus-heat-sink" CPU cooler, the heat load is placed in thermal contact with the bottom surface of an aluminum base plate that functions as a heat spreader. As in a conventional CPU cooler, this heat spreader plate is stationary. In a conventional CPU cooler, the top surface of the heat spreader base plate is populated with fins. In the air bearing heat exchanger, instead of having fins, the top of the heat spreader base plate is simply a flat surface.

The âoeheat-sink-impellerâ (the finned, rotating component) consists of a disc-shaped heat spreader populated with fins on its top surface, and functions like a hybrid of a conventional finned metal heat sink and an impeller. Air is drawn in the downward direction into the central region having no fins, and expelled in the radial direction through the dense array of fins. A high efficiency brushless motor mounted directly to the base plate is used to impart rotation (several thousand rpm) to the heat-sink-impeller structure. The bottom surface of this rotating disc-shaped heat spreader is flat, such that it can mate with the top surface of the heat spreader plate described above.

During operation, these two flat surfaces are a separated by a thin (~0.03 mm) air gap, much like the bottom surface of an air hockey puck and the top surface of an air hockey table. This air gap is a hydrodynamic gas bearing, analogous to those used to support the read/write head of computer disk drive (but with many orders of magnitude looser mechanical tolerances).

Heat flows from the stationary aluminum base plate to the rotating heat-sink-impeller through this 0.03-mm-thick circular disk of air. As shown later in Figure 18, this air-filled thermal interface has very low thermal resistance and is in no way a limiting factor to device performance; its cross sectional area is large relative to its thickness, and because the air that occupies the gap region is violently sheared between the lower surface (stationary) and the upper surface (rotating at several thousand rpm). The convective mixing provided by this shearing effect provides a several-fold increase in thermal conductivity of the air in the gap region.

TL;DR version: Stationary heat spreader surface on top of the IC. Teensy tiny air gap, small enough to permit heat transfer while functioning as an air bearing between heat spreader and... the next part, a heat-absorbing rotary impeller which pulls heat through the air gap into its fins, which are in turn cooled by air flow caused by centrifugal acceleration of the air through the rotating impeller assembly (squirrel-cage-fan style).

I'm not gonna pretend that there's no boundary-layer effect over the impeller blade surfaces, but I expect it'll be less than the effect caused by the common "push air down into the cooler and have it decelerate and turn 90 degrees to exit" cooler. Flow-through coolers would be more efficient than that, but air still has to decelerate through the cooler, whereas this impeller cooler makes the air accelerate during the cooling action. That might make a difference.

How well do bearings conduct heat?

The generic answer is "depends on thickness of air bearing surface (i.e., how big of an air gap), coverage area of bearing surface (i.e., is the heat spreader the size of the entire impeller, or just the small central portion of it), and the rotational speed of the rotating part on the other side of the gap -- moderate rotation speeds, in the 2k to 10krpm range, make the air in the gap turbulent and sheared rather than laminar, forcing mixing and heat transfer.

WTF happened to /.

Well, in this case, an actual scientific research article of relatively high coolness and technical merit leaked past the editors. I understand how this could be upsetting to most slashbots, given the novelty and rarity of this type of thing. Certainly, t

Re:Still has a boundary layer.

[JustinOpinion]

The article is wrong to suggest that the boundary layer disappears for moving surfaces. Fan blades do indeed get dust on them. However the actual work, described in the technical report, makes it clear that they are not claiming an elimination of boundary layer effects, merely a reduction of the boundary layer thickness:

This rotating heat exchanger geometry places the thermal boundary layer in an accelerating frame of reference. Placing the boundary layer in this non-inertial frame of reference adds a new force term to the Navier-Stokes equations, whose steady state solution governs the functional form of the heat-sink-impeller flow field [Schlichting, 1979]. At a rotation speed of several thousand rpm, the magnitude of this centrifugal (in the frame of reference of the boundary layer) force term is as such that as much as a factor of ten reduction in average boundary layer thickness is predicted [Cobb, 1956]. Unlike techniques such as air jet impingement cooling, the mechanism for boundary layer thinning in the air bearing heat exchanger does not rely on a process that entails dissipation of significant amounts of energy, nor is the boundary layer thinning effect localized in a small area. Rather, the centrifugal force generated by rotation acts on all surfaces simultaneously, and all portions of the finned heat sink are subject to the resulting boundary layer thinning effect. For the limiting case of flat rotating disk, an exact solution of the Navier-Stokes equation is possible and indicates that the magnitude of the boundary-layer thinning effect is constant as a function of radial position.

(Emphasis added.)

How well do bearings conduct heat?

Again, the technical document makes it clear that the rotating heat sink is not coupled via a bearing to the surface it's cooling. Rather there is a very thin layer of air separating them. Naively one might think that this layer of air (generally a poor heat conductor) would become limiting, and there would be poor heat transfer from the hot plate to the rotating heat sink. However they address this:

Heat flows from the stationary aluminum base plate to the rotating heat-sink-impeller through this 0.03-mm-thick circular disk of air. As shown later in Figure 18, this air-filled thermal interface has very low thermal resistance and is in no way a limiting factor to device performance; its cross sectional area is large relative to its thickness, and because the air that occupies the gap region is violently sheared between the lower surface (stationary) and the upper surface (rotating at several thousand rpm). The convective mixing provided by this shearing effect provides a several-fold increase in thermal conductivity of the air in the gap region.

So, basically by keeping the air gap very thin (30 microns), and by substantially shearing/mixing this thin air disk, its thermal conductivity can be sufficient to transfer heat up into the rotating fins. Overall a rather clever design.

WTF happened to /.

I agree a lot of junk gets posted to Slashdot. But in this case, a link was actually provided to a good technical document that answers many questions, provides schematics, and shows graphs of various performance measures.