Authored by: Nelson J. Gernert, Vice President of Engineering and Technology, Thermacore Inc., Lancaster, Pa.
Computers Beat the Heat, tinyurl.com/2f7zkh2
Heat pipes are basically vacuum-sealed tubes. Yet they keep high-power electronics cooler than traditional heat sinks.
Power semiconductors are encountering the same thermal control situations experienced by the microprocessor industry in the early 1990s. As microprocessor speed increased, so did the amount of heat generated by the microprocessor chip. Depending on the particular application, the heat released ranged from 10 to 150 W. For cost savings, this heat is typically rejected to ambient air. Conventional cooling approaches, such as extruded heat sinks, placed under constrained conditions were insufficient to meet the ever-growing cooling needs of the micro-processor. The computer industry found a solution using heat pipes. Today, heat pipes are used extensively for cooling microprocessors in laptops, desktop computers, high-performance servers, and workstations.
In its simplest sense, a heat pipe is a heat mover or spreader. It takes heat from a source, such as power semiconductors, and moves or spreads it to a region where it’s more readily dissipated. The heat pipe moves this heat with a minimal drop in temperature. A typical heat pipe is a sealed and evacuated tube that contains a porous wick structure and a very small amount of working fluid. The porous wick is typically a sintered-powder metal that lines the internal circumference of the tube. The central core of the tube is left open to permit vapor flow.
Each heat pipe has three sections: an evaporator, an adiabatic, and a condenser. As heat enters the evaporator section, it is absorbed by vaporizing the working fluid. The generated vapor travels down the center of the tube through the adiabatic section to the condenser section where the vapor condenses, giving up the latent heat acquired during vaporization. The condensed fluid is returned to the evaporator section by gravity or by capillary action in the porous wick structure. Heat-pipe operation is completely passive and continuous. This makes heat pipes quite reliable as there are no moving parts to fail.
In comparison to microprocessors, the heat released from large power semiconductors, such as IGBTs, diodes, or thyristors, can range into multiple kilowatts — a factor of 10 to 30× more heat. As electronic packaging trends towards smaller and lighter assemblies, ridding multiple kilowatts of heat from power semiconductors to ambient air becomes more of a challenge.
For example, two groups of six IGBTs each generate 6 kW of heat. One group is attached to a state-of-the-art aluminum extruded heat sink while the other group uses a heat-pipe heat-sink assembly. Both heat sinks operate under the same performance conditions and were modeled using Mentor Graphics FloTherm computational fluid-dynamics (CFD) software. The large extruded heat sink was 42-in. long × 24-in. wide × 3-in. high. Correspondingly, the heat-pipe assembly was 42-in. long × 11-in. wide × 9-in. high. Airflow rate and temperature at the cooling inlet was 600 cfm at 40°C. The goal was to create a small, air-cooled, and lightweight heat-sink package that kept the IGBTs within their operating temperature limits.
The alternative is to use a liquid-pump loop system that ultimately transfers the heat to air. Most companies try to avoid this scenario due to reliability, maintenance, and cost issues.
The aluminum heat-sink profile is a state-of-the-art large-profile extruded heat sink with the highest aspect ratio available. Consequently, the implied advantage is improved thermal control. The heat-sink profile measures 24-in. wide × 42-in. long × 3 in. high. The fin pitch is 2.5 fins/in. Each fin was 0.08-in. thick with a base thickness of 0.67 in. The weight and volume of this heat sink is 151 lb and 3,024 in.3, respectively.
The heat-pipe heat sink uses standard 0.75-in.-diameter heat pipes embedded in an aluminum plate under the power semiconductors. The pipes extend from the plate to a remote fin stack. The 11 heat pipes absorb heat from the electronics and transport it to the plate fins. The fins are cooled by forced convection. The aluminum mounting plate is 17-in. long × 12-in. wide × 0.98-in. thick. The fin stack is 19.3-in. long × 10.8-in. wide × 9-in. deep. The plate fins are 0.02-in. thick with a fin pitch of 10 fins/in. This heat sink weighs 70 lb and occupies an overall volume of 2,200 in.3
Heat-sink CFD analysis
Six 5 × 5-in. IGBTs, handling 1 kW each, were applied to each heat sink abutting the others in a two by three array. A 0.003-in. layer of interface material (with a k =1 W/m‑K) was assumed between the IGBT array and the heat sink. In each case, a 40°C ambient temperature and a volumetric flow of 600 cfm was used. The air was fully ducted through each heat sink.
The IGBTs have a maximum junction temperature of 125°C and a package thermal resistance of 0.04°C/W. Using this information, the case temperature, Tcase, for the IGBTs is 125°C − 40°C = 85°C. If the incoming ambient air is 40°C, the remaining temperature drop to dissipate the heat is Tcase − Tambient air = 85°C − 40°C = 45°C. This information was used as input for the CFD analysis. In all situations, the model options selected included steady turbulent flow and conduction, negligible radiation heat transfer and negligible buoyancy effects. The air properties varied with temperature.
Extrusion CFD analysis
An idealized “wind tunnel” was constructed around the extrusion for the domain/boundary conditions. The upstream and downstream surfaces were open for passage of air through the domain.
A uniform 600-cfm flow of air at 1 bar and 40°C was specified on the upstream surface. The remaining sides were made to coincide with the corresponding surfaces of the extrusion. These faces were given a symmetrical boundary condition.
For gridding the domain was mathematically divided into 478 × 79 × 11 cells. Five cells were used to resolve the flow profile between each pair of fins. The extrusion base thickness was also divided into five cells.
Using these parameters, the maximum temperature attained under the IGBTs was 130°C, well beyond the temperature limitations of the IGBTs. Therefore, it was determined that the extruded heat sink cannot meet the necessary thermal performance.
Heat-pipe CFD analysis
The heat-pipe assembly was comprised of two sub-models: the fin pack and the heat-input IGBT mounting plate. In addition, each heat pipe was constructed from two components: a high conductivity (k = 50,000 W/m‑K) cuboid that represented the high effective pipe conductance along the pipe length and a “thin” enclosure to coincide with the boundaries of the high-k cuboid. The effective thickness and thermal conductivity of this enclosure were chosen to represent the thermal impedance associated with radial transfer of heat into or out of the pipe across the pipe/fin or plate interface, pipe wall, and liquid-saturated wick structure.
The fin submodel was solved first: Each heat pipe was assumed to be carrying an equal share of the total heat load. This submodel yielded the operating temperature of each heat pipe. These temperatures then served as boundary conditions to the mounting plate submodel.
Again, an idealized wind tunnel was constructed around the fin pack for the domain/boundary model. The upstream and downstream surfaces were open for the passage of air through the domain. A uniform 600-cfm flow of air at 1 bar and 40°C was specified on the upstream surface. The remaining sides were made to coincide with the corresponding surfaces of the fin pack. These faces were given a symmetrical boundary condition.
For gridding the fin pack domain was divided into 46 × 43 cells in the plane of each fin. Five cells were included between each pair of fins.
IGBT mounting-plate submodel
The domain boundaries coincided with the boundaries of the chill plate. A symmetry condition was applied to all boundary surfaces, save the one through which the heat pipes would pass. On this surface, temperature boundary conditions were applied to each heat pipe based on the results of the fin submodel.
In this case, though, the domain was divided into 46 × 43 × 9 cells.
The maximum temperature under the IGBTs was forecast to be 80°C, well under the maximum temperature limits. The CFD analysis indicates that the heat-pipe assembly achieves the desired cooling results.
The heat-pipe assembly was tested and the results compared to the CFD predicted results. It was not necessary to test the extruded heat sink because Thermacore, under numerous occasions, has shown that its modeling approach for extrusions yields results within 5% accuracy.
In the heat-pipe assembly test, six heater blocks simulated the high-power IGBTs. Thermocouples mounted under each heater block measured the temperature of the plate. Other thermocouples were mounted in the airflow stream for calorimetric purposes. A steady-state heat load of 6 kW and 40°C ambient conditions were maintained. Airflow through the fin stack was achieved with two 30-cfm, 7.87-in.-diameter by 2.76-in. brushless, 48-Vdc fans. The acoustical noise rating associated with these fans was 57 dBA. This is under the Belcore specification for acoustical noise suppression of 65 dBA.
At these test conditions, the average measured block temperature was 76°C. This yields a thermal resistance value for the entire heat sink of 0.006 °C/W. In comparison, the CFD results for the heat-pipe assembly agreed to within 5Vo with the measured results. Consequently, the test results verified the validity of the modeling approach.
The heat-pipe heat-sink assembly effectively dissipated the 6 kW of heat while maintaining the IGBTs under rated temperature limits. In comparison, the extrusion was not able to meet the requirements. The entire IGBT mounting plate for the heat-pipe assembly was uniform in temperature displaying only an approximate 10°C variation. Any temperature variation was associated with the location of the heat pipe within the fin stack as the cooling air warmed passing through the stack. Uniform IGBT cooling is an important parameter when considering the effect of temperature on the quality of the output waveform.
The heat pipe system weighed 70 lb making it 81 lb lighter than the extrusion. It also consumed a volume of 2,200 in.3 while the extrusion had a volume of 3,024 in.3 Overall, the heat pipe unit occupied 27% less volume.
Heat-pipe technology lets designers reject multiple kilowatts of heat from power semiconductors directly to ambient air. This is an important conclusion considering the potential alternative is a liquid pumped loop system that has inherent long-term reliability (leaks), maintenance (pump failure, fluid cleanliness, filtering), and corresponding cost issues.
Finally, fan or blower noise is becoming an issue with cooling of power semiconductors. The heat-pipe assembly used a dual-fan pack solution that achieved an acoustic level of 57 dBA, well under the 65-dBA ratings listed in the Belcore specification for acoustical noise suppression.
1. Dunn, P.D., Reay, D.4., Heat Pipes, Fourth Edition, Pergamon Press, Copyright 1994 Elsevier Science Ltd.