27 August 2013

Wax cooling can massively speed up processors for short bursts

Researchers at the University of Michigan have turned to wax to help push microprocessors to the limit. The wax can efficiently absorb heat as it changes phase and begins to melt. In recent tests, they boosted an Intel Core i7 microprocessor — which normally runs at around 10 watts — up to 50 watts.

As mentioned in a recent post by Wired, the researchers believe they could probably cycle the chip up to 100 watts, at least for brief periods of time. At 54 degrees Celsius, local hot spots on the chip will cause the overlying wax to melt. We should note here that while it may not be recommended, a stock-clocked chip should be able to run close to double that (80-85 degrees C), so for this scheme to become the norm, materials with a higher melt temperature might be preferred.

It is not clear how melting is sensed here, but if we draw an analogy from boiling water, large amounts of heat can be absorbed during a phase change without any rise in temperature. When a critical point is reached, the chip either needs to run at a slower speed, or power down a percentage of its transistors. Since the unique mechanisms of heat propagation work so efficiently during a phase change, the need to sense temperature at many locations to protect the chip can be reduced. The idea of using wax as a cooling machine is nothing new — it was used in several Apollo missions back when we had the space program to cool batteries.

Running chips in this kind of burst mode is inevitable. In simpler systems, like motors or jet engines, this is the standard operating procedure. For example, you might continuously dump 500 amps into a starter motor for several seconds, but if you try that continuously, your windings will burn up. For more complicated motor applications, like those involving stepper or servo motors, the voltage and amperage rating of the windings can be exceeded — and resulting torques increased — by rapidly pulsing the power. Similarly for jets, if you run your turbine at take-off power too long, you will melt your blades. If you keep the afterburners on too long, the nozzle would melt. The point here in these examples is that without chopping your system power, not only will your design be grossly suboptimal, but the possible existence of a compact, affordable footprint for your larger machine concept is a failure.
A chip with little divots that can be filled with a cooling solution like wax.

Phase change materials actually have many potential overlapping applications in computing. In some cases, they might actually be capable of performing some rudimentary computations themselves. The so-called “spin-glass” models, in which elements interact with their immediate neighbors in various ways to adopt larger scale order (the same as alignment in magnetic materials), in fact formed the original inspiration for successful neural network models. An even more esoteric (and unproven) application for fatty-material-based phase change is in the insulating sheath of axons. The myelin here is composed largely of lipids critically poised at the fluid transition temperature. When a signal goes down the axon, depending on the physical integrity of the myelin, some of the transmission energy is in fact carried by the myelin.

It may be a little while before your smartphone employs wax for cooling. If thermoelectric energy harvesting is made more efficient, it may even be longer. Nobody wants to see precious battery power being used to melt wax, but it may turn out that it actually makes good sense.

Now read: New material identified by US Navy could revolutionize computer chip heat dissipation.


Post a Comment

Get every new post delivered to your Inbox.


Copyright © 2015 Tracktec. All rights reserved.

Back to Top