Liquid Cooling of High-performance Microcomponents and Microsystems
BY YOGENDRA K. JOSHI
Heat generation rates in electronic components and systems continue to increase, as performance expectations rise and sizes are decreased. Traditionally, the key metric for defining the thermal challenge has been the heat flux, or the rate of heat generation per unit area. With increasing emphasis on shrinking sizes at all levels, the heat generation rate per unit volume is also becoming an important parameter in the search for thermal management (Figure 1).
Interconnect feature sizes are projected to approach 20 nm by the year 2016, as per the International Technology Roadmap for Semiconductors. A continuing increase in current densities will result in volumetric heat generation of 2 ¥ 104 W/mm3, which is one order of magnitude higher than in current high-performance devices and approximately six orders of magnitude higher than the core power density in a boiling water nuclear reactor. This dramatic increase in the volumetric heat generation rates in the interconnects, caused by Joule heating, is projected to ultimately shift the primary source of heat generation from the transistor switching to interconnect dissipation. New ways to address this problem will need to evolve.
Figure 1. Heat generation: from chip to facilities.
At the level of the chip and module, as shown in Figure 1, the volumetric heat generation rates decrease compared to the interconnects. However, the heat fluxes are projected to approach the 100 W/cm2 level for microprocessors. These heat fluxes will be highly non-uniform because of the incorporation of core memory within the central processing unit. This is likely to result in localized spots with heat fluxes of several hundred W/cm2.
A new challenge is also emerging at the size level of the facility housing densely packed arrays of high-performance server racks. The floor-level heat fluxes in such facilities are projected to be two orders of magnitude higher or more than human occupied spaces such as auditoriums or theaters. The current approach for cooling such spaces is to use a raised floor with a plenum to bring in cooling air from a computer room air conditioning unit. Cold air then flows upward through perforated tiles placed in an alternating hot and cold aisle arrangement, so that the hot air exhaust from cabinet rows facing each other does not mix with the cold air coming up through the tiles. In addition to the removal of high heat loads in such rooms, a key concern is the energy efficiency of cooling devices. Since the overall power dissipation in an entire room can approach several MW, such energy savings can be significant.
Cooling air by using heat sinks has been the main technology for thermal management of electronic components. Increasing heat sink size, lack of access to the die for optoelectronic and other mixed technology applications, and concerns about acoustic noise are prompting a serious look at alternate cooling strategies. Liquid cooling offers considerably higher overall heat removal capabilities, as well as the ability to handle large, localized heat generation rates compared to air, due to the higher thermal conductivity and specific heat of liquids. Liquid cooling is already widely used in the form of miniature heat pipes that are a part of nearly all laptop computers. These devices use a closed tube for transport of heat by a carrier fluid from the evaporator to the condenser end. A chip is placed in direct thermal contact with the evaporator, and the condensed working fluid returns to the evaporator through a wick structure lining the walls of the pipe. For high-power dissipations and long distances, the heat pipe is subject to a capillary limit, beyond which a "dry-out" condition may occur.
Miniature Flow Loops for Thermal Management
Another promising technique using liquid cooling involves flow loops. Flow loops allow the flexibility of separating the heat input and heat rejection regions. By miniaturizing the size of a heat exchanger at the chip/package through microfabrication, tight spatial constraints near the chip can be met.
Figure 2. A single-phase, pumped flow loop is shown here with a stacked microchannel heat sink.
Two types of flow loop configurations have recently been implemented. The first uses pumped circulation of the coolant (Figure 2).1 Its stacked microchannel heat exchange unit can be microfabricated of metallic materials such as copper for external attachment, or of silicon for direct chip integration. Conventional high-precision machining techniques such as wire electro-discharge machining and precision sawing may be used for microfabrication in copper. For silicon microfabrication, wet etch and deep reactive ion etching methods are suitable.
Figure 3. Possible applications of two-phase cooling using gravity-driven thermosyphons include computing workstations and servers, as well as electronic modules with multiple heat dissipating devices.
In single-phase microchannel cooling for fully developed flow and heat transfer, the heat transfer coefficient is h a k/D, where k is the thermal conductivity of the coolant and D is the hydraulic diameter of the microchannel. For single microchannels in silicon with water as coolant, heat transfer coefficients of 105 W/m2K and overall thermal resistance of ~ 0.09 K/(W/cm2) have been demonstrated in the literature with an overall pressure drop of ~ 2 bar. Through stacking of multiple layers of microchannels, the pressure drop can be reduced for a given flow rate. In addition, the chip can be made nearly isothermal — even for non-uniform power dissipation — by varying the direction of flow between the various layers of the stack. This avoids an undesirable temperature rise along the flow direction of a single microchannel.
A second type of flow loop is a gravity-driven thermosyphon (Figure 3). The evaporator, which is placed in intimate thermal contact with the heat-generating chip or package, may be miniaturized through the use of a microfabricated structure to enhance boiling. One particular structure used an interconnected network of stacked microchannels, which are effective in initiating and sustaining boiling. The interconnections or pores connect the top and bottom rows of microchannels on each layer of the stack. Under normal operating conditions, placement of the condenser above the evaporator ensured a gravity-driven flow of vapor from the evaporator to the condenser, and a return of the liquid coolant from the condenser to the evaporator. A prototype unit was demonstrated to cool an 85-W microprocessor package in 2001.2
The single-phase cooling scheme has the ability to provide extremely high heat transfer coefficients using coolants such as water. The two-phase cooling scheme offers the advantage of being passive, and does not require an energy input for coolant circulation if a gravity head is available. With imposed circulation, it can also be used for situations where the condenser may need to be placed below the evaporator. The scaling up in the packaging hierarchy and heat removal rates can provide solutions for heat removal in the 100 W/cm2 to 1 kW/cm2 range from the chip to the cabinet using fluid loops.
- X.J. Wei, and Y. Joshi, "Optimization Study of Stacked Microchannel Heat Sinks for Microelectronic Cooling," IEEE Transactions on Components and Packaging Technology, Vol. 26, pp. 55-61, 2003.
- A. Pal, Y. Joshi, M. Beitelmal, C.D. Patel and T. Wenger, "Design and Performance Evaluation of a Compact Thermosyphon," IEEE Transactions on Components and Packaging Technology, Vol. 25, pp. 601-607, 2002.
YOGENDRA K. JOSHI, professor, may be contacted at Georgia Institute of Technology's G.W. Woodruff School of Mechanical Engineering, Atlanta GA 30332; (404) 385-2810; e-mail: firstname.lastname@example.org.