The design goals of many new projects include miniaturization, increased performance, longer battery life, and silent operation. All of these goals run counter to effective thermal management, and designers are facing the prospect of sacrificing performance to handle the excessive heat that modern ICs generate. Cooling problems plague all electronic systems, from the simplest portable devices to complex board-level systems. Although mechanical-packaging engineers typically handle thermal analysis on larger projects, most design teams rely on rules of thumb, experience with similar projects, guidelines from board or chassis manufacturers, simple thermal analysis, and the old standby—trial and error.

Much of this thermal consternation is due to the effects of Moore's Law: the exponential growth in the number of transistors on a chip. Over the last 30 years, ICs have progressed from 10,000-nm resolutions with 3000 transistors to less than 100-nm resolutions with hundreds of millions of transistors. With these added transistors, designers have increased IC performance by adding parallel execution paths and therefore increased power dissipation. The power consumption in CMOS ICs is proportional to fCV², where C is the sum of all transistor gate, drain, and interconnect capacitances; f is the frequency of operation; and V is the core voltage. It is easy to see that, as the number of transistors and the frequency increase, the power climbs. Although reducing the voltage saves power, leakage current through ultrathin gate structures also contributes to overall power consumption. This leakage current is nonlinear and becomes significant at less than 1V. Figure 1 shows the relentless growth of power density in each successive microprocessor generation.

Along with shock, vibration, and humidity, temperature is the major source of environmental stress in electronic systems. Temperature can reduce performance by lowering output-voltage swings, reducing switching speeds, lowering noise margins, and generally reducing overall signal quality. In addition to the performance hit, temperature stresses also reduce system reliability. Mechanically, high temperature can lead to wire-bond failures, die fractures, and increased corrosion. Excessive heat may also overstress electrical components, causing gate-oxide breakdown, electromigration, ion diffusion, and, ultimately, failure. The key to improving reliability and maintaining performance is to ensure that each component in the system is operating within its temperature parameters.

Designers have many options for dissipating the heat generated by electronic circuits. Convection, a passive form of electronic cooling, transfers heat by airflow due to temperature gradients; in other words, hot air rises. This type of cooling is most common for small, portable devices, such as cell phones and PDAs, and you can augment it by increasing the surface area of dissipating devices with a heat sink. For simple systems, designers can estimate the thermal performance directly from component-data-sheet parameters. The most common measure of package thermal performance is θ_JA, the thermal resistance measured or modeled from junction to ambient. You use θ_JA to measure the temperature difference between the component and the ambient atmosphere when the component has consumed 1W of power.

If you know the thermal resistance and the operating power, P, of the component, you can approximate the junction temperature, T_J, from T_J=T_A+θ_JA*P.

Data sheets include another parameter, θ_JC, the thermal resistance from junction to case, which may be more useful when the component is cooled by forced air, heat sink, or liquid cooling. In this case, T_J=T_C+θ_JC*P, where T_C is the temperature of the case or packaging. The efficiency of convection cooling improves when you supplement the air movement with a fan. Although forced-air cooling has for years served the electronics industry, it is now stretched to its limit, with each new high-performance design calling for increased heat-sink area and higher airflow rates. With the notable exception of PC/104, most bus-based systems include forced-air cooling, and their specifications include guidelines for maximum dissipation per plug-in module.

As the number of components on a board or system grows, designers need an automatic analysis technique to detect thermal stress. The Betasoft-Board package from Dynamic Soft Analysis is an example of software specially designed for board-level thermal analysis. It outputs board temperature and gradient maps, component and junction temperatures, and the amount those temperatures exceed their respective limits (Figure 2). The software models boards with as many as 1500 components per side and performs 3-D modeling of the complex flow and thermal fields based on heat conduction, convection, and radiation.

Betasoft-Board interfaces with several board-layout programs for automatic transfer of component-placement and parameter information. Advanced features include multilayer boards, irregular shapes, bolt-on heat sinks, conduction cooling through wedge locks, or sealed compartments. The flow field can be natural or forced-convection, and heat exchangers can cool closed systems. The software models the effects of gravity, air pressures, and flow directions. You can also attach heat sinks, heat pipes, chip fans, and conduction pads to components. You can transfer the junction temperatures calculated by the Betasoft-Board analysis into reliability software to improve the reliability predictions.

Fluent Inc offers electronics-cooling-simulation software based on computational-fluid dynamics for mechanical-packaging engineers. Its Icepak software allows an engineer to build a computer model of a product or system and then virtually prototype and test it under real-world conditions (Picture). Icepak accepts input from standard CAD-analysis tool sets and includes a downloadable library of fans, heat sinks, and IC packages for manual model building. The latest version includes a new user environment that features a model manager, advanced object wizards, alignment tools, and four-window simultaneous views. The thermal model in the photo on pg 49 is of a 1U server cooled using heat sinks and two blowers at the back of the chassis. The colors correspond to surface temperature, and the flow ribbons indicate the pattern air takes coming through the inlet vents, over and around the components, angled DIMMs, heat sinks, power supplies, and baffles before entering the blowers and exiting the system. Icepak is available for Unix and Windows platforms as an annually renewable license. A typical entry-level price is $20,000, which includes the software, training, and unlimited technical support.

Although you may have a well-designed system that reacts thermally as predicted by manual or automatic analysis, several situations may add undue stress and cause unexplained failures. For example, consider a board-based system that was designed and analyzed with a full complement of plug-in boards but now has empty slots due to changed requirements or degraded operation.

The resulting air leakage through the empty slots upsets the laminar flow of air inside the card cage and reduces the airflow over the remaining boards. Card-cage manufacturers generally offer baffle boards that you can install in an unused slot to both close the front panel and block the airflow to the unused slot. You may encounter other thermal problems in board-based systems, such as loading one area in the chassis with all the high-power boards instead of evenly distributing the heat producers. Board-layout problems may also create thermal shadows where a tall component sits ahead of a short component in the air stream. A turbulent eddy often forms over the short component, trapping the heated air and causing severe overheating.

Thermal-induced failures can also result from an unmanaged power-down sequence. Most systems are designed with a single power switch that simultaneously removes voltage from the power supply, the boards, and the fans. If the system happens to be running near its upper thermal limit, and airflow is halted, the internal temperature will continue to rise after shutdown even though the components are no longer powered. Because of the component density, normal convection cannot remove the remaining heat, and elevated temperatures may last 10 to 20 minutes, possibly damaging the IC-die level, packaging, and solder joints. Solutions are to simply to add a time delay to allow fans to operate for a period after power-down or to add a separate switch to control air circulation. You can easily evaluate the thermal performance of a forced-air system by measuring the temperature in the intake and exhaust air stream at each board position. You should measure temperatures at several points along the board to check for hot spots or uneven airflow. A difference of 10ºC or greater between the intake and exhaust temperatures may indicate the need for increased airflow.

Even if a design calls for a high-power processor or system on chip, advanced techniques are available to cool a hot spot. For example, Thermacore offers a line of passive heat-dissipation devices based on heat-pipe technology. A heat pipe consists of a vacuum-tight envelope, a wick structure, and a working fluid. The heat pipe is evacuated and then backfilled with a small quantity of working fluid, just enough to saturate the wick. The atmosphere inside the heat pipe is set near the equilibrium between liquid and vapor. Heat entering at the evaporator upsets this equilibrium, generating vapor at a slightly higher pressure. This higher pressure vapor travels to the condenser end, where the slightly lower temperatures cause the vapor to condense, giving up its latent heat of vaporization. The condensed fluid is then pumped back to the evaporator by the capillary forces developed in the wick structure. This continuous cycle transfers large quantities of heat with low thermal gradients.

To battle the heat that high-performance systems generate, many manufacturers of extended-temperature boards have turned to conduction cooling to help tame thermal problems. Thermal planes within a board and conductive bars across and along the edge of the board shunt heat away from high-power components to the chassis and eventually to the outside air. Vista Controls' PPC G4C is a recent example of a single-board computer incorporating conduction cooling (Figure 3). The 6U Compact-PCI board incorporates a PowerPC 7457 processor along with dual Gigabit Ethernet ports. Additionally, the board offers dual 64-bit, 66-MHz PMC sites, 2 Mbytes of external L3 cache, 512 Mbytes to 1 Gbyte of onboard SDRAM, 64 kbytes of nonvolatile RAM, and 128 Mbytes of program flash. The PPC G4C is available in extended-temperature, rugged air-cooled, and conduction-cooled versions. The board is priced at $7100.

Several new technologies on the horizon offer alternative techniques for cooling electronic systems. For example, SynJets, or synthetic jet-ejector arrays, developed by the Georgia Institute of Technology's School of Mechanical Engineering consist of a diaphragm mounted within a module that has one or more orifices. Electromagnetic or piezoelectric drivers cause the diaphragm to vibrate 100 to 200 times per second. The rapid cycling of air into and out of the module creates pulsating jets that you can direct to the precise locations that require cooling. SynJets produce two to three times as much cooling as conventional fans with two-thirds less energy input. With no friction parts to wear out, SynJet cooling modules are much smaller than fans, and you can mount them directly within the cooling fins of heat sinks (Figure 4). Although the jets move much less air than similarly sized fans, the turbulent and pulsating airflow breaks up thermal boundary layers.

VIDA (vibration-induced droplet atomization), another advanced cooling technology from Georgia Tech, uses atomized liquid coolants to carry heat away from electronic components. VIDA uses high-frequency vibration produced by piezoelectric actuators to create sprays of tiny cooling liquid droplets inside a closed cell attached to an electronic component. The droplets form a thin film on the heated surface, allowing thermal energy to be removed by evaporation. The heated vapor then condenses, either on the exterior walls of the cooling cell or on tubes carrying liquid coolant through the cell. The liquid is then pumped back to the vibrating diaphragm for reuse. To date, researchers have been able to cool about 420W/cm² and ultimately expect to increase that figure to 1000W/cm². Georgia Tech has licensed both of these cooling technologies to industry, and they should soon be available.

Another advanced cooling technology, developed at Stanford University's Mechanical Engineering Department, uses common materials to produce a noiseless, closed-loop, active-cooling system for high-power ICs. The university has now licensed the technology to Cooligy. The active-microchannel system employs a fluid pumped in a sealed cooling loop. A heat collector is attached to the chip to absorb heat that hot spots generate. The heat travels a small distance into fluid flowing through the microchannels in the collector, 20 to 100 microns wide each, which transport the heat away from the chip to a radiator, where the heat is exhausted to the outside air (Figure 5). The fluid then travels through Cooligy's solid-state electrokinetic pump to complete the cooling loop. Using fluid to transfer heat means that the cooling system can pump large amounts of heat away from the chip to any location the system designer chooses. Cooligy's electrokinetic pump is based on an interaction between a fluid and glass. The walls of a fluid-filled glass tube carry a negative charge that is balanced by positive ions in the fluid accumulating near the walls of the tube. When you apply an electric field along the length of the tube, the excess positive ions near the tube's wall move parallel to the wall and push the fluid through the tube. The core of Cooligy's pump is a glass disk with millions of paths through which fluid is pumped using this electrokinetic effect.

Because your customers will always want the latest technology to squeeze the best performance from a system, you can expect that each new project will have thermal issues. To avert these temperature problems, IC-design and heat-analysis software may become standard tools for future development projects. As technology continues to keep pace with Moore's Law, we must learn to live with its inevitable side effect: heat.