Graphene could be used in a number of ways to dissipate heat. For instance, it would work better than carbon nanotubes for making heat-dissipating, electrically conducting interconnects between transistors. Used as a heat spreader in a 3-D chip, it could be paired with thermal vias made of carbon fiber or carbon nanotubes. In situations where graphene's high electrical conductivity would prove a problem, it could be combined with an unconventional electrical insulator like synthetic diamond. If necessary, graphene can be converted into an electrical insulator by irradiating it with low-energy electrons.
In short, incorporating graphene into chip designs could yield devices that are faster, less noisy, and run cooler. Of course, the usefulness of graphene—or, for that matter, of diamond or carbon nanotubes—will depend not just on their physical characteristics but also on their cost and their compatibility with existing chip-manufacturing technology. It's also possible that the high thermal conductivity of graphene may decrease when it is embedded between layers of other materials or when the number of atomic layers gets too high. The exact extent of such changes is a matter of ongoing research. Still, the results so far provide reasons to be optimistic about graphene.
Managing heat in a nanoscale device becomes even more complex as chip designers continue to boost IC speeds while shrinking transistor channels. The relentless push of Moore's Law, which holds that the number of transistors per chip doubles roughly every 18 months, has led to ever higher speeds and drive currents and ever smaller chip features.
The speed and current limits of a chip are proportional to the electron mobility of the semiconductor materials used in its construction. In silicon at room temperature, electron mobility is mainly limited by phonons, which cause the electrons to scatter. A higher electron mobility usually corresponds to weaker electron-phonon interactions and thus less heat generation.
But efforts to increase electron mobility—and thus the speed and current a device can handle—can end up degrading the device's ability to dissipate heat. One approach, already used in conventional chips, is to create strain in the atomic lattice of the semiconductor substrate by forming it out of two materials with slightly mismatched lattice spacings. Often a combination of silicon and silicon germanium is used, the latter having slightly larger spacing between atoms.
A device that includes silicon germanium can have serious problems dissipating heat. That's because the thermal conductivity of SiGe, and of other alloys, is an order of magnitude smaller than that of its constituent semiconductors. So having a layer of SiGe in a device makes it more difficult for heat to escape from the transistor channel to the heat sink. In nanoscale chip architectures, problems caused by overheating in the transistor channels can cancel out any gains in mobility achieved by straining the lattice.
There may be ways to both enhance electron mobility and improve heat removal by devising ways to control the flow of phonons in the device. Such methods are termed phonon engineering. Our theoretical work suggests that sandwiching silicon nanowires or ultrathin silicon films between layers of diamond can increase electron mobility if the interface is good. Diamond not only has a high thermal conductivity, it is also acoustically harder than silicon (acoustical hardness is a product of a material's sound velocity and its mass density) and can be exploited to modify the phonon dispersion within the material. The acoustical mismatch between the silicon and diamond acts to partially suppress the phonons and electron-phonon scattering that would otherwise slow electrons down. Initial experimental results are encouraging, but the quality of the interface continues to be an issue. The rapid progress in depositing synthetic diamond films on silicon achieved in industry and research labs gives us hope that composite wafers with optimized phonon properties may soon become a reality. In other studies, we are investigating the usefulness of graphene for redirecting and controlling the phonons.
What other ways are there to deal with excessive heat? One promising new technology is thermoelectric cooling, either of the entire chip or of local hot spots. A thermoelectric cooler is essentially a heat pump that transfers heat from one side to the other when current is run through it. By design, it has no moving parts or liquid components and can be very robust and compact. Such devices are already being used to cool solid-state lasers and LEDs—and beer: The same basic technology can be found in those portable coolers that plug into the cigarette lighter of your car.
The main problem with thermoelectric cooling of ICs is finding a good material to use. One of the best bulk thermoelectric compounds is bismuth telluride, but it can't be integrated with silicon using conventional chip-fabrication technology. So investigators are seeking thermoelectric materials that are more compatible with manufacturing techniques but still retain their ability to wick away heat. Recent reports suggest that silicon nanowires or tiny blobs of silicon germanium and silicon (called quantum dots) may work.
A variation on this theme is to attach thermoelectric coolers to the chip's heat sinks instead of relying on air to carry away the heat. In January, a team led by researchers at Intel demonstrated the first chip-scale thermoelectric refrigerator.
Another approach is to introduce a liquid cooling system comparable to that in your car's engine or what was used in the early Cray supercomputers, which were cooled by Freon circulating through pipes. For state-of-the-art ICs, it may be possible to cool individual heat-dissipating elements on the chip using liquids. Liquid cooling avoids the thermal resistance that occurs at the interface of two solids. Exactly how to manage the microscale (if not nanoscale) plumbing remains an area of active research.
Efforts also continue to improve the thermal interface materials between the chip and the heat sink. The right kind of interface material can greatly improve heat removal and reduce the thermal resistance between the chip and its packaging. Common interface materials consist of an oil or grease embedded with ceramic or metal particles. But here again, fillers made from graphene look attractive. Even a tiny amount of graphene can lead to a substantial increase in thermal conductivity.
Although researchers are optimistic about these recent developments in thermal management, it will take years until any of the new materials and designs find their way into commercial chips. So you may be wondering what you can do in the meantime to keep your computer from overheating. Here's one easy fix: Open up your machine and remove any dust, which can accumulate and prevent heat from dissipating properly. Or follow my wife's example and aim a good strong fan at your computer. With any luck, that will tide you over until the more elegant solutions I've described here finally arrive.
This article originally appeared in print as "Chill Out".
About the Author
Alexander A. Balandin is chairman of materials science and engineering at the University of California, Riverside. In "Chill Out", he describes new materials and designs to keep chips cool. He notes that people often confuse "photons," which are quanta of light, with "phonons," which measure crystal lattice vibrations and are the focus of Balandin's research. Once, an overeager assistant painstakingly changed phonon to photon on each flyer for a lecture Balandin was giving, assuming it was a typo.
To Probe Further
For more about Alexander A. Balandin's work on thermal management in electronic devices, see the Web site of the Nano-Device Laboratory at the University of California, Riverside: http://ndl.ee.ucr.edu. His work on phonon engineering is supported by the U.S. Air Force's Office of Scientific Research. Balandin is also associated with two multi-institutional research consortia that are investigating materials and designs for future electronics: the Center on Functional Engineered Nano Architectonics (http://www.fena.org) and the Interconnect Focus Center (http://www.ifc.gatech.edu).
Heating Up Graphene
Illustration by Bryan Christie Design
Measuring the thermal conductivity of something with a thickness of just one atomic layer is tricky. At the University of California, Riverside, we approached the problem this way: First, we prepared samples to measure, each consisting of a long graphene flake suspended across a trench in a silicon wafer and attached to heat sinks. We then heated the graphene flake with a laser. The heat wave propagated from the middle of the graphene to the heat sinks.
To measure the temperature at the center of the hot spot, we came up with an unconventional use for a micro Raman spectrometer. Ordinarily, Raman spectroscopy is an optical technique used to identify materials. The Raman spectrum of graphene has a clear peak, referred to as a G peak; the spectral position of the peak depends on the temperature of the sample. So by measuring the exact position of the G peak, we were able to use our spectrometer as a thermometer.
Only a small portion of the laser light ended up dissipating in the graphene. Most of the light passed through it and was reflected back. We determined the fraction of the light dissipated in the sample by comparing the Raman intensity of the graphene with that of bulk graphite. Knowing the temperature rise, the dissipated light power, and the geometry of the graphene flake, we then determined the graphene’s thermal conductivity. The measured values exceeded 3000 watts per meter per kelvin near room temperature and depended on the size of the graphene flake. We learned later that our results agreed with physicist Paul G. Klemens’s predictions years earlier. -Alexander A. Balandin
To return to the article