Compact cooling devices are taking shape to deal with the next generation of computer chips. One of the research projects, conducted at the University of Maryland under initial sponsorship from several private companies and federal government laboratories, studied liquid cooling. In order to avoid the design complexities associated with direct liquid cooling, and to make the device of near-term applicability to systems designers, the research team at Maryland decided to use indirect liquid cooling. The university researchers focused on the use of two phase thermosyphons to meet these requirements. Researchers conceptualized a two-chamber, closed-loop device with an evaporator chamber at the chip and a condenser some distance away connected through tubing. The working fluid tested in laboratory experiments was the dielectric coolant PF 5060 made by 3M Co. The University of Maryland and Hewlett-Packard team selected two test beds to evaluate the performance and ease of integration of these devices within existing high-performance computing systems.
The year is 2005. Imagine getting into your computer-controlled hybrid electric vehicle for your half-hour morning commute to work. You park the quiet, pollution-free car and get into your sleek, smart office building using your magnetic entry card.
As you enter your office, you have the day’s work cut out for you. The complex, three-dimensional computational fluid dynamics and heat transfer design problem you have been working on for the past two weeks must be finished by the end of the day. You have a design review team from the customer coming early next week for a meeting. You type the final set of input data into your 3.5-GHz multiprocesssor desktop computer and go out to get a cup of coffee. You come back in a few minutes and find pungent smoke coming from the vents of the computer box. Your worst nightmare has come true.
Far-fetched? Not really, according to the International Technology Roadmap for Semiconductors. This forecast of the international semiconductor business predicts that thermal dissipations for high-performance microprocessors are projected to approach the 160-watt threshold within the next five years. The resulting heat fluxes (heat dissipation rates per unit surface area) are projected to be over 100 W/cm2 in certain regions of a microprocessor chip.
Yogendra Joshi is a professor at the George W. Woodruff School of Mechanical Engineering of the Georgia Institute of Technology in Atlanta.
Today s trusted workhorses of the electronics cooling technology are fan-cooled heat sinks. Over the past decade, their sizes and coolant air velocities have increased significantly to meet power dissipation requirements. Further dramatic improvements in their performance appear infeasible.
High-performance thermal management alternatives are currently under development to handle the steady state and transient heat loads of next-generation electronic systems at a number of universities, industries, and government laboratories. Much of this activity is being sponsored by the Defense Advanced Research Projects Agency.
Liquid Cooling of Electronics
One of the research projects, conducted at the University of Maryland under initial sponsorship from several private companies and federal government laboratories, studied liquid cooling.
Liquids are significantly better heat transfer media than air, because their thermal conductivity and thermal capacity are higher. The use of liquids for immersion cooling of microwave tubes dates back to the early 1940s and, more recently, has been realized in the cooling of high-performance supercomputers. As mainframe computers gave way to personal computers, liquid cooling seemed to be an idea whose time had come and gone.
However, with relentless increases in device integration and operating frequency, heat dissipation requirements are now back to the mainframe era. As in the 1980s, implementation of direct immersion liquid cooling brings considerable complexity in design and an associated increase in costs.
A much more appealing alternative for the designer is to use liquids for the indirect cooling of electronics. An example of this technology is the heat pipe used in the cooling of laptop computers. With the continuing need for higher heat-removal capability, and longer distances within many electronic systems between the processor and the ultimate heat-removal location, the heat pipe performance may become inadequate. That results in dryout, or the inability of the fluid to continue circulating in the heat pipe.
In order to avoid the design complexities associated with direct liquid cooling, and to make the device of near-term applicability to systems designers, the research team at Maryland decided to use indirect liquid cooling. The intended use was for cooling individual high-power chips or multiple chips, such as the central processing unit of high-performance computers. To facilitate its easy integration into electronic products, the cooling method had to be simple to implement. Since the incorporation of pumps or fans brings in associated reliability concerns, the system had to be flexible enough to allow entirely passive operation, although a fan could be added to augment performance or to reduce the size of the overall device. Another requirement was that the device be a retrofit option to augment thermal management of already fielded systems.
The university researchers focused on the use of two-phase thermosyphons to meet these requirements. Researchers conceptualized a two-chamber, closed-loop device with an evaporator chamber at the chip and a condenser some distance away connected through tubing. The working fluid tested in laboratory experiments was the dielectric coolant PF 5060 made by 3M Co.
The dielectric nature of the fluid ensured that even in the unlikely event of leakage, the coolant would not short the electronics. The saturation temperature, or boiling point, of the liquid at atmospheric pressure was 56°C, which ensured that the maximum chip temperature could be maintained in the 70° to 90°C range, often used in the thermal design of electronics.
Additional control over the boiling temperature of the coolant could be achieved by varying the operating pressure. The evaporator was intended to be attached directly to the heat-generating chip or package, much like a conventional heat sink.
One of the key challenges for the researchers was to miniaturize the device. Typical electronic cooling applications place stringent space restrictions near chips. The evaporator, therefore, had to be more compact than a typical heat sink. The second challenge was to initiate the boiling predictably and at temperatures close to the saturation value.
The highly wetting nature of these dielectric fluids, due to their low surface tension, results in the flooding of naturally occurring cavities on a nominally smooth surface. It is the trapping of air or vapor in these cavities that results in the initiation of the highly efficient boiling mode of heat transfer.
The filling of the available cavities with the liquid can raise the temperature at which a continuous boiling begins by as much as 20° or 30°C, allowing the chip to overheat to unacceptably high temperatures.
These challenges were addressed through the use of a boiling enhancement structure employed inside the evaporator chamber. The copper structure, consisting of a three-dimensional interconnected network of microchannels, was soldered to the inside surface of the evaporator bottom. Each layer of the structure was 1.15 mm thick and had an array of parallel channels on each face with widths of 310 micrometers and heights of 0.55 mm, spaced 0.48 mm apart.
Each layer of the structure has an array of parallel channels on each face. The channels on one face run orthogonally to those on the other. They are cut deeper than halfway through the material so they form pores where they intersect. These pores play a central role by providing engineered cavities for vapor generation and departure, resulting in a very predictable start of boiling.
A prototype using copper contained six layers, each 1x1 cm. A less conductive material, such as silicon, would perhaps not be effective beyond four layers.
Such structures are known to result in high heat dissipation rates (above 100 W/cm2 for dielectric coolants) in boiling in large enclosures measuring 9 x 3 x 14 cm.
The University of Maryland studied the structure in a much smaller space, 4.5 x 4.5 x 2 cm, and found that its thermal performance remained almost unaltered, even under tight confinement. Laboratory tests of such thermosyphons revealed that heat removal rates of 75 W/cm2 with PF 5060 were feasible with entirely passive means. The condenser size could be reduced if it was cooled by forced air convection.
While the thermosyphon device worked effectively in a university laboratory, the transfer of this technology to industry was critical to the success of the project. Up to this point in the investigation, testing used heaters that simulated the temperatures of computer chips in operation.
Enter Hewlett-Packard. Jointly, the University of Maryland and Hewlett-Packard team selected two test beds to evaluate the performance and ease of integration of these devices within existing high-performance computing systems. The first test bed was an HP Vectra VL 800 desktop computer using a 1.5-gigahertz Intel Pentium 4 chip, and the other was a high-performance HP L-Class server, with a four-way PA RISC 8600 processor.
The desktop computer used a fan-cooled heat sink, called Turbo Cooler, 6.9 cm in diameter and 4.7 cm high, along with an overall system fan for thermal management. The server used tower heat sinks, which consisted of vertical heat pipes cooled with a fin array. A set of fans at the inlet and outlet of the system enclosure operated in a push-pull mode to create forced air convection.
The desktop test bed was selected with two key considerations in mind. First, the next generation of this machine is likely to have single-chip power dissipation in the 100-W range, requiring innovative thermal management solutions. Second, noise concerns made the elimination of the fan on the heat sink a priority, something easily achieved by the thermosyphon.
The server was chosen to demonstrate the use of thermosyphons for a multiprocessor machine.
Hewlett-Packard wanted to be convinced that the solution worked and that it could be mass-produced. While a university laboratory could demonstrate the concept and optimize performance, a thermal management products company was needed for possible mass production. That role was filled by Thermacore Inc. of Lancaster, Pa.
The research team now included the author, at the time a professor with the University of Maryland; his students and research associates; Monem Beitelmal and Chandrakant Patel of Hewlett-Packard Research Laboratories in Palo Alto, Calif.; and Todd Wenger and Jon Zuo of Thermacore.
The team collectively designed thermosyphons for the two computers. The thermosyphons were to be drop-in replacements for existing thermal management devices. The evaporators would be clamped down on the chips, much like the conventional heat sinks. The condensers would be cooled by the system fans. All structural elements of the thermosyphon were made out of copper for ease of assembly. Proper charging of the system with the coolant was critical for getting the best performance.
In order to benchmark the effect of the working fluid, the team also decided to build some prototype units that used water.
Demonstrating the Prototypes
The day of reckoning came in May 2001, when the prototypes had to be demonstrated to DARPA. This was done with the desktop Hewlett-Packard Vectra computer. The thermosyphon for the Vectra included an evaporator of 3.2 x 3.2 x 2.9 cm and a condenser 2.6 x 8.2 x 7.5 cm. The condenser was cooled by the system fan of 9.2 x 9.2 cm. A computer program was used that would exercise the processor to dissipate any fraction of its maximum power of 85 W
Using deionized water as the working fluid, the temperature of the evaporator base plate bottom was measured at 57°C under a chip power of 85 W and a local ambient temperature of 23°C. Since the specification called for a maximum case temperature of the central processing unit chip below 70°C, the thermosyphon was considered a great success.
At the same chip power, the PF 5060 working fluid resulted in a higher temperature of 101°C at the evaporator base plate bottom. Efforts are under way to see if this can be reduced by more careful removal of absorbed air, which tends to limit the condenser performance. The use of PF 5060 is highly desirable in applications at lower temperatures, where freezing may be a concern with water. The thermosyphon provided considerable flexibility in routing and placement of electronics and also eliminated a noisy fan.
Efforts are now under way to further optimize the device by focusing on the condenser design. In addition, a number of fundamental investigations are being done to understand the operation of the boiling enhancement devices.
The alliance is also involved in developing and implementing microfabricated cooling devices for wearable computers.