1、AN-04-9-2 AThermal Bus System for Cooling Electronic Components in High-Density Cabinets Michael Wilson, Ph.D. Associate Member ASHRAE Jonathan Wattelet, Ph.D. Kevin Wert, Ph.D. ABSTRACT The increasing power densi of electronic systems is reaching the point at which it is no longer possible to adequ
2、ately cool the individual components by direct air cool- ing. Successful cooling applications for high-power electron- ics must deal with thermal issues at the device, device cluster, PCB brinted circuit board), subassembly, rack, and cabinet. A thermal bus was designed to collect waste heat from in
3、di- vidual or groups of components and passively transport that heat to a more favorable location where it can be physically removed by forced air or to a location inside or outside the cabinet where it can be transferred to an external cooling circuit or sink. This concept is fully compatible with
4、current data center cooling architectures such as hot aislekold aisle. Extension to anticipated next-generation architectures, such as pumped liquid cooling, is relatively straightforward. INTRODUCTION Figures 1 and 2 show the trends that are driving the ther- mal problems in telecommunication cabin
5、ets (Telecommuni- cations OEM 2001; TI1 2001). Many years ago, electronic components were adequately cooled by the natural circulation of air over the components. As components became more powerful (Figure i), fans were required to circulate air, and heat sinks were added to critical components to p
6、rovide adequate cooling. Today, there is often insufficient space to mount a large enough heat sink directly on the component, and heat pipes are often employed to transport heat to an area in the card or chassis where a larger heat sink can be placed. However, power densities are now approaching th
7、e point where an adequate amount of air cannot be supplied in the vicinity of the circuit card. Consider Figure 2. The downward- sloping line shows the trend of decreasing case temperature that is demanded by increasing clock speed and heat flux. The upward-sloping line shows the trend of rising enc
8、losure temperature that is caused by the increasing heat dissipation in the enclosure. The difference between the two lines indicates the “thermal budget” available for the design of the cooling deviceshystems. As the two lines converge, the thermal budget shrinks, presenting increasing challenges t
9、o the ther- mal system design. As the thermal budget approaches zero, alternative technologies, such as pumped single- and two- phase loops or refrigeration systems, will have to be used to provide the necessary cooling. Clearly, finding effective solu- tions to these thermal problems will become a
10、major constraint on the reduction of cost and time-to-market, two governing factors between success and failure in commercial electronics sales. THE THERMAL BUS CONCEPT These challenges in electronics cabinet thermal manage- ment are driving a holistic approach to thermal solutions. Successful cooli
11、ng applications must deal with thermal issues at the device, device cluster, PCB (printed circuit board), subassembly, rack, cabinet, and data center levels. This paper describes how the thermal problems at each of these levels, as weil as their interdependence, can be considered as part of an integ
12、rated thermal architecture for thermal management of high-powered electronics. This thermal architecture concept is illustrated in Figure 3, which shows a schematic diagram of heat flow paths in a typical electronic enclosure cooling system (Phillips 2000). Heat is generated within individual compon
13、ents (Level I) and can be transported via various inter- mediate pathways (Levels 2-6) to some ultimate sink, such as the data center cooling system (Level 7). Table 1 summarizes Michael Wilson is a senior research engineer and Jonathan Wattelet is the manager of advanced product research at Modine
14、Manufacturing Company, Racine, Wisc. Kevin Wert is the manager for technology development at Thermacore International, Inc., Lancaster, Penn. 02004 ASHRAE. 567 100 F P t c I 10 f 2 - - t L 1980 1985 1990 1995 2000 2005 2010 Year t n* generatiun Figure 1 Increasing heat dissipation in telecommunicati
15、ons cabinets. P o the definitions of thermal problems at each level. Also included in Table 1 is an assessment of the techniques that are currently used at each level and a description of thermal archi- tecture in the most general sense. To illustrate the concept, consider Figure 4, which shows the
16、layout of a server box (Computer OEM 2001). At the lowest level, Level 1 (within the processor package), the heat generated at the junctions spreads across three layers of mate- rials before reaching the outer surface of the package lid: the silicon material of the die, the epoxy interface between t
17、he die and the lid, and the AlSiC lid. The circuit board in this case plays a minor role in cooling the processors, since the majority of the heat goes upward into the package lid. Therefore, Levels 2 and 3 are not relevant to this particular case. The heat sinks on top of the processors provide the
18、 interface between the components (Level 1) and the airflow (Level 4), functioning as the direct thermal link (Level 6). The lower power components on the board also dissipate their heat into the airflow. Level 5 (system heat exchanger) is not present since the airflow is discharged directly into th
19、e room air. Level 7 in this case is the air-conditioning system that provides thermal control for the room where the server resides. This case is perhaps the simplest example of a thermal architecture. As the number and variety of the components increase, the thermal architecture can become far more
20、 complicated, and all seven levels may exist in the same system. The remainder of this paper introduces a thermal bus concept based on the integratiodassembly of passive heat transport “building block“ technologies. The proposed concept serves as Level 4 andor Level 6 in the integrated ther- mal arc
21、hitecture illustrated in Figure 3. Waste heat is collected from individual or groups of components and is passively transported to a more favorable location where it can physi- cally be removed by forced air or to a location inside or outside the cabinet where it can be transferred to some external
22、cool- ing circuit or sink. The thermal bus concept is fully compatible 1 yt generution IY89 I u94 2000 90 ? -i 9- H =Ar/Yowcr -0 -* )* Loner Junction Temperuture o 60 ;- - e! 50 ; a 40 *a. Therma- Chip to heat pipe interface I 2.5 2.5 core International Incorporated. Heat pipe Thermal connector Loop
23、 thermosyphon 15.0 Liquid-cooled condenser 17.0 DISCUSSION Christian Belady, Hewlett Packard, Richardson, Tex.: The temperature drop across the thermal connector is low. What special things have you done for the interface to get you there? Using a liquid-cooled condenser, the thermal resistance is c
24、ooled solution. Another advantage of liquid cooling is that it uses less space, allowing for more circuit cards. This solution allows for significantly lower thermal resistances without the worry of having plumbing joints within the electronics cabi- Michael Wilson: Our calculations assumed a Contac
25、t reSiS- circumference times its length. This resulted in a calculated value of about 1.2 K. The 2 K is therefore conservative. David Copeland, Senior Thermal Engineer, Fujitsu Labo- ratories of America, Sunnyvale, Calif.: Power density trends show rapid increase from 1997-2005, then much 0.0486 “Cm
26、, which is almost halfofthe resistance ofthe air- tance of0.5 K per W/cm2. The area for heat pipes was ha1f its net. SUMMARY AND DISCUSSION Heat rejection from electronic components to ambient room air can present many engineering challenges in high- density cabinets due to the premium on space. A t
27、hermal bus is implemented to transfer the heat from the component level to a location where more surface area for heat rejection is available. Significant reductions in thermal resistance can be achieved by using the thermal bus over current extrusion tech- nology. The thermal bus can be implemented
28、 so that it conforms with current data center cooling techniques, but it is also flexible so that it can be implemented with a liquid cool- ing strategy. REFERENCES Computer OEM. 2001. Communication with Thermacore International Incorporated. Patel, C.D., C.E. Bash, C. Belady, L. Stahl, and D. Sulli
29、van. 2001. Computational fluid dynamics modeling of high compute density data centers to assure system inlet air specifications. Proceedings of the IPACK 2001, pp. 1-9. Phillips, A.L. 2000. Thermal bus for high power enclosures, Thermacore internal document. Sullivan, R.F. 2000. Alternating cold and
30、 hot aisles provides more reliable cooling for server farms. The Uptime Institute. Retrieved May 5, 2003 from the World Wide Web: . Telecommunication OEM. 2001. Communication with Ther- macore International Incorporated. TTI. 2001. GR-3028-CORE Thermal management in tele- communications central offi
31、ces. Telcordia Technologies Generic Requirements, Issue 1. Telcordia Technologies Incorporated. weaker rate of increase. Trends predicted seem to be fairly correct for 1998-2003 product releases. Do we still support predicted trends after 2005? Wilson: When telecomm went from 1 st to 2nd generation,
32、 the amplifier efficiency increased, so the waste heat trend line jumped down. With 3rd generation (3G), a higher linearity is required, which means less of the dynamic range of the ampli- fiers can be used. This essentially throws away the ends of the amplifier response, and dramatically increases
33、waste heat. This accounts for the big jump in the 3rd generation trend line. Of course, with the implosion of telecomm, the roll out of 3G is much slower than anticipated, and advanced 3G is even slower. We would assume that this curve is somewhat dated, and that advances will be slowed until teleco
34、mm becomes highly profitable. You might note that the graph in the presen- tation ends at 2005. Rebecca Perry, Engineer, Sun Microsystems, San Diego, Calif.: Is it possible to design multivendor products with consistent building-block-like cooling characteristics (not just airflow direction) to redu
35、ce complexity to the end user and make product integration in a data center easier and more reli- able? Wilson: With a simple flat-plate-type interface, a cooling sink would be built into the back plane of the cabinet. Individual chassis would have a thermal extension that would mate with these sink
36、s well outside the envelope of the individual chassis. It is very doable, but does require setting some standards. Reinhard Seidl, Mechanical Engineer, Taylor Engineer- ing, LLC, Alameda, Calif.: How far along are the fluid cool- ing systems mentioned at the meeting? Are there actual systems using d
37、ielectric fluids/water/DX cooling currently in use, or are these all at the prototype stage (not looking at obso- lete supercomputer systems that were water cooled)? Wilson: There are demonstration systems that have been- and are-operating. It should be noted that the components of the systems in the paper are in practical everyday use. ASHRAE Transactions: Symposia 573
