Cooling processor hot spots

Many high-performance electronic systems require a cooling mechanism to keep sensitive devices, such as microprocessors, from overheating. Cooling mechanisms are becoming more critical as their integrated density- and current-voltage-handling capability continues to increase. A device can shed heat by radiation, convection, or conduction. In radiation, the device sheds heat as electromagnetic radiation. Radiation can occur in a vacuum and does not rely on gas, fluids, or other substances surrounding a heat sink. The material and the color of the device affect the heat-radiation efficiency. Radiation accounts for a small portion of the dissipated heat in a high-performance processor cooler.

Unlike radiation cooling, convection cooling relies on a viscous substance, such as a gas or a fluid, to transport the heat away from a solid surface. The heat from the solid molecules transfers to the molecules of gas or fluid surrounding the device, and the movement of the gas or fluid transports the heat away from the device. If the movement of the gas or the fluid is forced, such as by a fan or a pump, the cooling mechanism is relying on forced convection. Forced convection, such as with a fan and heat-sink combination, is a primary mechanism for heat dissipation for high-performance microprocessors.

The heat sink in the forced-convection example relies on conduction for transferring heat away from the processor. Conduction is the exchange of kinetic energy between molecules that are in direct physical contact with each other. The more energetic molecules from the operating processor device collide with the less energetic molecules of the heat sink, resulting in the exchange of heat. The heat sink provides—by virtue of its geometry, such as a set of fins—a greater surface area to shed heat by convection to the surrounding air or fluid and to effect the removal of the heat from the system.

As high-performance processors continue to migrate to higher clock rates and higher density process technologies, the peak power density for processor devices continues to increase as well. This increasing peak power density poses cost and reliability challenges as the size of the heat sink increases relative to the heat source and the necessary volumetric airflow increases to maintain the device operating temperature below a threshold. Fluid-cooling mechanisms can dissipate more heat to maintain better temperature consistency with less flow volume but do so with increased complexity because they require well-sealed piping and may require pumps.

Heat pipes and microchannel cooling are two heat-transfer mechanisms that do not dissipate significant amounts of heat but can effectively transfer large amounts of hot-spot heat to where heat dissipation is more appropriate and cost-effective. Remote heat removal enables a wider range of heat-sink geometries for heat dissipation. Laptop computers have for years successfully used heat pipes for cooling; microchannel cooling has only recently entered the commercial arena.

Heat pipes are passive devices that consist of a container, a wick, and working fluid (Figure 1). The container is a sealed, hollow tube that isolates the working fluid from the outside environment, can maintain a pressure differential across its walls, and enables heat transfer to move from and into the working fluid. A porous lining inside the container is called the wick or capillary structure. The wick enables capillary pressure to transport the working fluid from the condenser to the evaporator ends of the container. The working fluid for electronics cooling is often deionized water, ammonia, acetone, or methanol, depending on the operating-temperature range, and it resides within the wick and container.

One end of the pipe, the evaporator, attaches to the heat source. As the heat enters the evaporator, it causes the working fluid to boil and turn to vapor. The evaporating fluid saturates the hollow center of the wick and creates a pressure gradient along the pipe. The pressure gradient forces the vapor to spread through the middle, or adiabatic, section of the pipe toward the condenser, or cold end, of the pipe. Condensation of the vapor occurs wherever the temperature drops below the temperature at the evaporation area. You can attach heat sinks, fins, and heat-collection blocks to heat pipes via soldering, brazing, or epoxy bonding or by expansion of the heat-pipe tubing into the sink, fins, or block. Attaching a heat sink to the pipe causes the condensation to occur at that point in the pipe.

As the vapor condenses back to a fluid state, gravity or capillary action within the wick transports the fluid back to the evaporator area, completing an operating cycle. When the evaporator is elevated higher than the condenser, the capillary action must work against gravity to pump the fluid back to the evaporator. Sintered wicks can return the working fluid to the evaporator via capillary action. Screen and grooved wick structures cannot return the working fluid against gravity if the angle is greater than –5%. The pore radius of the wick structure determines how much the heat pipe can operate against gravity.

Heat pipes can operate as low as 40°C, but, at that temperature, they can carry only a fraction of the heat they can carry at 0°C and higher. The upper temperature limits of heat pipes depend on the fluid used, and 60 to 80°C is the average limit. Water at atmosphere boils at 100°C. Water-based heat pipes can operate at less than 100°C because the inside of the pipe is not at atmospheric pressure. The internal pressure of the heat pipe is the saturation pressure of the fluid at the fluid temperature. Therefore, the fluid in the heat pipe boils at any temperature above its freezing point. A water-based heat pipe is under partial vacuum at room temperature (20°C), and the heat pipe boils as soon as heat is input.

Heat pipes do not have a set thermal conductivity because of the two-phase heat-transfer mechanism. The effective thermal conductivity improves with the length of the pipe because the thermal gradient remains close to the same for different-length pipes for the same heat load. A heat pipe's effective thermal conductivity also changes based on the amount of power transfers and the sizes of the evaporator and condenser. Heat pipes range from 3 mm to 5/8 in. in diameter and from inches to several feet in length, and they can carry as much as 1000W.

You can bend and form heat pipes at manufacture into a variety of fixed configurations. They require no maintenance, and there are no moving parts. In a properly designed heat pipe, the working fluid is completely contained within the capillary wick structure and is at less than one atmosphere. Therefore, if you introduce a leak into the heat pipe, air leaks into the pipe rather than fluid's leaking out. The fluid then slowly vaporizes as it reaches its atmospheric boiling point.

David Tuckerman, PhD, and Fabian Pease, PhD, both Stanford professors, published in 1981 that microchannels etched into silicon could remove heat densities as high as 1000W/cm². A microchannel system consists of high-aspect-ratio channels that are etched into a small piece of silicon (Figure 2). The channels are each 20 to 100 microns wide—about the width of a human hair—and three to 10 times as tall as they are wide. An electrokinetic pump forces the fluid in a sealed loop through the channels to transfer heat conducted from the heat collector to the radiator where the heat convects to the air.

A microchannel system is similar to a heat-pipe system in that it has no moving parts and relies on fluid in a sealed system to transfer heat from a source to a remote dissipation point. A microchannel system differs from a heat pipe in that the electro-kinetic pump is an active device; the fluid does not go through a phase change, and it relies on a physical loop of the channels to complete the thermal cycle (Figure 3). The fluid is basically ionized water with some additives to prevent biological growth.

The heat generated at the processor hot spot travels approximately 1.5 mm from where the heat originates to the walls of the microchannels. The heat from the walls of the microchannels conducts a small distance into the fluid before being transferred to the radiator, because the heat-transfer coefficient varies inversely with the width of the channel. As the microchannels get narrower, the walls of the channels stay cooler and more completely heat the layer of fluid as it travels through the collector. Therefore, unlike large cold plates with macroscale channels, microchannels neither require nor create turbulent flow to mix the hot fluid next to the walls with the cooler fluid in the center of the channel.

The thickness of the walls between the microchannels can vary at the time of construction to direct and expose more fluid over the hot spots while less fluid flows over cooler spots. At the processor hot spot, the channel walls would be at their thinnest; the walls would be thicker at the cooler spots, where less heat dissipation is necessary. Varying the microchannel-wall thickness can also support flexible routing through the radiator for more heat-dissipation options.

The core element of the microchannel system is the electrokinetic pump; it has no moving parts. It is a glass disk with micron-sized pores. When you apply an electric field across the pump, it causes the electrokinetic effect, which is an interaction between glass and fluid that allows positive ions to accumulate near the glass walls (Figure 4). This effect causes the fluids and the positive ions in them to be pumped through the disk. Current implementations of the pump can deliver 1.3 atmospheres of pressure or flow rates greater than 33 ml/minute, using a porous, sintered-glass cylinder with a 40-mm diameter and 1-mm thickness to support cooling heat loads on the order of 100W. The microchannel system has been demonstrated on current high-performance processors that reach 300 to 400W/cm². Hence, the theoretical limit of 1000W/cm² represents a margin for meeting future cooling challenges.

Author Information

You can reach Technical Editor Robert Cravotta at 1-661-296-5096, fax 1-661-296-1087, e-mail rcravotta@edn.com.