Power. Heat. Heat. heat. are. sink. characterized. highly-compact. with. significantly. decreases the. convective heat. However, as in. are.

Transcription

1 6-8October, Barcelona, Spain Heat Transfer Enhancement in High-P Sinks using Active Reed Technology Heat Pablo Hidalgo, Florian Herrault, Ari Glezer Mark Allen Georgia Institute of Technology, AtlantaGA USA Scott Kaslusky Brian St. Rock United Technologies Research Center, East Hartford, CT06108, USA Abstract Enhanced heat transfer in a high aspect ratio channel that models a segment of a novel, high-performance air-cooled heat-exchanger Two innovative features that enable this system system for high-power electronics is characterized. are integration of a centrifugal blower-diffuser with a high density finned heat-sink, microfabricated active elements that provide small-scale heat transfer enhancement at reduced air rates. The present investigation focuses on heat transfer fluid mechanics that are associated with small-scale motions induced by a piezoelectric vibrating that is integrated within a mm-scale channel. Thesee time- at channel surfaces mixing of rmal boundary layers with core. High-magnifica ation particle image periodic small-scale motions enhance convective heat transfer velocimetry (PIV) measurements are used to characterize interaction of vortical structures shed by with channel wall induced small-scale motions. Performance enhancement by actuation is quantified in terms of increased power dissipation over a range of rates compared to baseline in absence of. It is demonstrated that channel s coefficient of performance can be increased by a factor of 1..4 while accounting for power to changes in channel pressure drop. I. INTRODUCTION Modern, high-power heat sinks are characterized by highly-compact designs as shown, for example, in Fig. 1. In this configuration, a centrifugal blower-diffuser is integrated with a finned heat sink for high fin density. In this design, heat sink fins form diffuser channels additional splitter fins are introducedd as diffuses radially outward to maximize heat transfer area enhance convective heat transport. Ambient air enters centrifugal blower throughh a venturi that also servess as a hollow shaft for blower motor, air through blower is directed through curved diffuser channels. However, as in most compact high power heat sink designs forced convection heat transfer within high aspect ratio fin channels is typically limited by available air volume rate consequently by low channel Reynolds number. This limitation is manifested by two closely coupled stages of heat transport. First, local heat transfer from fin surface is limited by temperature gradient within a thin rmal boundary layer. Second, overall heat transport for a given mass rate is governed by average temperature rise of core. These deficiencies are commonly overcome by increased volume rate with significant concomitant increase in required blower power. Blower Motor Active Spiral Reeds Fin-D Diffuser Fig. 1.High-power heat sink with integrated blower, diffuser active s for enhanced heat transfer. Enhanced local global heat transport within heat sink channels at higher rates (or channel Reynolds numbers) are typically associated with cross mixing that is induced presence of unsteady, small-scale enhancement of heat transfer within heat sink channels by motions. The present work focuses on active deliberately inducing small-scale Reynolds numbers. The presence of motions within core at low channel se small-scale motions disrupts momentumm rmal boundary layers enhances rmal mixing within core, reby significantly decreases fincontrolled, small-scale motions are induced by time- to-air heat transfer resistance. In present experiments, periodic motion of miniature piezoelectric cantilever vibrating s that are placed within mm-scale heat sink channels so that ir planform surfaces are parallel to channel walls (as shown schematically in Fig. 2). The s are driven at resonance period of actuation frequency is smaller than characteristic time of flight through channel. The pressure drop that is caused by presence of s within channels can be partially offset by thrust produced by. The motion of blade induces time-periodic shedding of organized vortical structures (as shown schematically in Fig. 2) turbulent-like small-scale motions that would

2 6-8October, Barcelona, Spain normally be present at much higher Reynolds numbers. II. EXPERIMENTAL SETUP These small-scale motions enhance heat transfer The present investigation is conducted in a modular, high coefficient at fin surfacess as well as mixing of this heated aspect ratio channel test section measuring 2.5 x 27.4 x 60 air with core channel to overcome local mm. The test section is preceded by a 260 mm long settling global limits of forced convective heat transport in section having same cross section, an upstream baseline channel in absence of. Intrafin contraction (contraction ratio of 28) to ensure fully- because of ir low form factor, low power requirements, section is driven by a regulated air supply monitored by a vibrating s are attractive for integration in heat sinks developed, spanwise-uniform (Fig. 3) ). The contraction microfabrication compatibility. precision meter ( rate is measured to within 1.5%). The settling section is instrumented with multiple pressure ports along around channel. There are two interchangeable test sections. The first test section is designed for detailed measurements of pressure distributions Fig. 2. Schematic rendition of within a channel cooled by a while second test section includes removable sidewalls vibrating. that are integrated with heaters as shown in Fig. 4. To The effectiveness of induced small scale motions for reduce blockage effects due to, side walls of enhanced heat transfer in low-reynolds number s, that test section were designed with slight spanwise indentations are characteristic of air cooling applications, was originally (having a maximum depth of 460 m on each side) opposite demonstrated using syntic jet actuators (e.g., by direct to mounting post of (cf. Fig. 5). Static pressure impingement, Kercher et al. [1] Pavlova et al. [2], or by measurements within test settling sections were integration with a heat sink, Mahalingam Glezer obtained using a high precision alcohol manometer coupled [3],Gerty et al. [4-5], Kota et al. [6]).The utility of with a 48-port pressure switch. Fig. 3 shows a CAD model piezoelectrically vibrating s for external cooling of low- or of assembled test bed including inlet contraction, power electronic packages (e.g., for notebook computers settling section, test section, a section of test cell phones) was investigated by A ikalin et al. [7] section that was used for pressure measurements. A ikalin et al.[8],who also noted lower power consumption reduced noise compared to external cooling by conventional fans. A ikalin et al. [7] investigated cooling performance of a piezo-fan on a heat sink used in conventional cell phones laptop computers reported better than 100% enhancement of heat transfer coefficient compared to natural convection. In cooling of notebook computers, piezo-fans were effective in alleviating Fig. 3. a) Channel test set up assembly, b) side wall of test section hot spot areas that were not adequately cooled by for pressure measurements showing mounting slots for post spanwise indentation for alleviation of blockage. system s conventional fan. In a later experimental numerical investigation, A ikalin et al. [8] considered rmal performance enhancement by a jet-like The test section that is used for heat transfer streaming motion of a piezo-fan operated next to a constant measurements is shown in Fig. 4. The heater side walls are heat flux source compared to natural convection cooling. each comprised of microfabricated spanwise copper The authors reported sensitivity of rmal serpentine trace (having characteristic resistance of 40 performance to frequency deviation from piezo-fan s that is depositedd on a silicon plates. Each individual heater resonance frequency, distance from heat source, is controlled using a high precision current source allowing oscillation amplitude (which depends on blade resolution of power dissipation to within20 mw. length). DOE analysis showed an enhancement of Individual spanwise windings of serpentine heater are convective heat transfer coefficient of up to 275%. In an tapped for (spanwise-averaged) temperature measurements experimental numerical investigation of cooling (with resolution that is better than 0.1 o C) using Joule characteristics of piezoelectric fans Wait et al. [9] showed heating of copper windings. Each channel side wall that associated with higher resonance modes can consists of two adjacent silicon heater plates each measuring lead to enhanced mixing improved cooling at 42.4 x 30.41mm. The heaters are mounted on a ceramic expense of significantly higher power consumption. Finally, frame with a sealed air gap on backside of frame for Gerty [10] investigated small-scale heat transfer rmal insulation (only one heater is shown in Fig. 4 for enhancement by a that is integrated within a heated clarity). The width of channel is set by placing pairs of channel. This work demonstrated effectiveness of this polycarbonate spacers that form top bottom walls of approach for cooling of compact electronic packages in channel allow for rapid assembly disassembly. absence presence of a core motivated The spacer pairs are used to clamp mounting frame present work. at top bottom ends of channel using a machined recess (in present configuration, upstream end of is 10 mm downstream of channel inlet). ce channel was assembled attached to settling section,

3 6-8October, Barcelona, Spain entire test section was rmally insulated fiberglass wool. using with distance from tip of (normalized by channel width w). This figure demonstrates importance of rmal spreading within walls ( or fins) substrate. The effects of spreading are evident in two salient features. First, presence of leads to a significant improvement in heat transfer coefficient (HTC) a decrease in wall temperature upstream of. The second ( perhaps more important) effect is domain of influence of owing substrate conductivity of silicon (150 W/ /m o K). Both features are consequence of Biot number of silicon ( ratio of substrate conductive rmal resistance to convective rmal resistance at Fig. 4. CAD drawings of exploded assembled views of heat wall). The Biot number of silicon heaters is 0.048, (for transfer channel. reference, an aluminum channel would have a Biot number of 0.03), refore rmal characteristics of The piezoelectric used in present experiments channel walls are similar to characteristicss of (Fig. 5) is 19 mm long 25 mm wide include a 100 conventional heat sinks. m thick plastic blade. The 7 m thick PZT plate covered nearly 50% of length of plastic blade actuator is mounted on a 229 m stainless steel post support. The resonance frequency of is approximately 850Hz its maximum peak-to-peakk tip displacement is 1.4 mm (approximately 60% of channel width) ). IV. Fig. 5. The piezoelectric HEAT TRASNFER MEASUREMENTS The cooling performance of s is evaluated using two sets of measurements. In first set, power to side wall heaters is invariant over a range of rates, heat transfer within channel is assessed for baseline in absence of, n in presence of active. Reed activation resultss in a reduction in wall temperature (compared to baseline ) measured differences in wall temperature yields local heat transfer enhancement. The second set of measurements is used quantify increment in heat dissipation thatt can be achieved in presence of. Two comparisons are made. First, for a given channel rate, dissipated heater power in presence of actuation is increased until channel wall temperature is approximately equal to wall temperature of baseline (in absence of ). Second, for increased heat dissipation in presence of actuation for a given rate, rate of baseline configuration is increased until power dissipation matches -enhanced dissipation at lower rate. As noted above, first set of heat transfer experiments are conducted using invariant heat dissipation (W) for several rates. The streamwise variation of heat transfer enhancement is shown in Fig. 6 in terms of variation of ratio of Nusselt numbers = Nu REED ON /Nu REE ED OFF Fig. 6. Nusselt number enhancement at 30LPM ( ), 40LPM ( ) 75LPM ( ). The variation of channel pressure drop with rate streamwise variation of static pressure are shown in Fig. 7a b, respectively (in terms of pressure coefficient. These data are shown for baseline (in absence of ), in presence of active. Even thoughh channel walls are indented to alleviate blockage effects by, re is still a significant increase in pressure drop (compared to baseline ) in presence of active. As shown in Fig. 7a, increase in pressure drop diminishes with rate (it about 150% at 40 LPM 80% at 75 LPM). As noted above, blockage is caused by its maximum thickness whichh is approximately 20% of width of channel. The present work has demonstrated that wall indentation can significantly mitigate increase in pressure drop that effectiveness of indentation can be furr improved by additional refinement. The streamwise variations of static pressure within channel at Q = 60 LPM in absence presence of actuation are shown in Fig. 7b. The pressure upstream of tip ( tip is 29 mmm downstream of channel s inlet) is significantly higher than in channel segment downstream of tip due to blockage. While actuation results in a small increase of pressure upstream of tip, downstream of, streamwise pressure gradient changes, re

4 6-8October, Barcelona, Spain is a local increase in pressure. This rise in pressure increase in dissipated heat while keeping power downstream of which is produced by thrust is not unchanged. Under se conditions, pressure rise in sufficient to overcome pressure rise due to blockage. a longer channel at Q = 30 LPM is four times lower than at 60 LPM (for a given channel cross section working fluid Fig. 7. a) Overall pressure drop in baseline channel ( ) in indented channel with active ( ); b)streamwise variation of static pressure along indented channel at 60LPM with( ) without ( ) actuation The overall rmal enhancement by in channel is characterized in terms of several primary parameters including: fluid power ( ), power (which is held invariant at 20 mw), total fluid power (fluid power plus power) average HTC ( Table I). These results show that for a given dissipated power ( W) HTC at Q = 30 LPM with actuation (about 160 W/m 2 o K) is comparable to HTC of baseline at over twice rate (Q = 75 LPM). This is noteworthy because total power at Q = 30 LPM is 38 mw while power at Q = 75 LPM is 110 mw or nearly 2.9 times higher. Furrmore, preliminary observations indicate that a reduction of 25% in power results in a trivial loss in cooling effectiveness, but can save 13% of total power at Q = 30 LPM. LPM TABLE I RESULTS SUMMARY HEAT TRANSFER AND POWER INPUT TEST Reed Mode Heaters (W) Fluid Reed Total Avg. HTC (W/m 2 K) Thermal enhancement is also measured by comparing rmal performance at equal total fluid power for baseline in presence of active actuator. In Table II rmal performance of two s that have approximately same total fluid power is compared. The total fluid power at Q = 30 LPM with active (38 mw) is almost same as at Q = 60 LPM in absence of (i. e., baseline, 41 mw). For se operating conditions, average HTC increases by 42% with actuation at half rate. As shown in Fig. 8, average rmal resistance ( ) of actuated decreases by 21% ( maximum reduction in is 55%). These data indicate that for same operating conditions channel can be significantly lengned with a comparable Fig. 8. Comparison of rmal resistances of baseline at 60LPM ( ) with a having similar total fluid power in presence of an active at 30 LPM ( ). LPM TABLE II RESULTS SUMMARY HEAT TRANSFER AND POWER INPUT TEST Heaters (W) Fluid Reed 20 0 Total Avg. HTC (W/m K) For two conditions shown in Table II (i.e., similar total power in presence of an active at Q = 30 LPM, for baseline at Q = 60 LPM), wall temperature T when is active is lower than for baseline at a higher rate. It is refore desirable to determine by how much dissipated power in actuated at Q = 30 LPM can be increased to reach a wall temperature that is similar to that of baseline at Q = 60 LPM. Fig. 9 shows that nominal difference in wall temperature between two s is approximately 6 o C. In order to (approximately) equalize wall temperatures between two s, dissipated power is increased by about 25% to almost 15 W. These measurements are conductedd over a range of rates in presence of actuation such that wall temperature matches corresponding baseline wall temperature. It is found that dissipated heat can be increased up to 20 W at 30 LPM, 19W at 40 LPM, 18W at 60 LPM 17W at 75 LPM, which correspond to increases of 66%, 58%, 50% 42%, in dissipated heat, respectively. The variation of rmal resistancee of channel with total fluid power is computed for several rates of baseline with actuation (Fig. 10). In absence of actuation When actuation is applied re is a clear offset that indicates a decrease in rmal resistance an increase in fluid power. As noted above, increase in fluid power can be mitigated by adjusting wall indentation refore, it should be possible to realize a channel configuration with an integrated that decreases rmal resistance of system withoutt a significant increase in total fluid power

5 6-8October, Barcelona, Spain investment. Furrmore, as noted in discussion in The COP is inversely proportional to clearly sensitive to connection with Fig. 9, actuation enables significantly total fluid power. higher heat dissipation at same total fluid power than baseline at significantly higher rates. Fig. 9. Streamwisee variation of wall temperatures of enhanced at 30 LPM ( ) of baselinee channel at 60LPM ( ) for similar total fluid power. Fig. 10.Variation of rmal resistance with total fluid power for several rates with ( ) without ( ) actuation. Finally, rmal enhancement that is enabled by actuation is quantified using coefficient of performance (COP), which is ratio of dissipatedd heat total fluid power invested. Fig. 11 shows variation of COP of channel with rate. Threee measurement sets are considered. The first set is variation of COP for baseline (in absence of ) for constant heat dissipation ( W). The second set is corresponding COP with actuation where at each rate dissipated heat is increased until surface temperature within channel matches temperature of corresponding baseline. Finally, third set of measurements compares COP of actuated at Q = 30 LPM while dissipating 20W to COP of baseline for which rate is increased until heat dissipation is also 20 W at same wall temperature as for -actuated at 30 LPM. These heat dissipation conditions are matched (in terms of channel s wall temperature) at Q = 72 LPM. It is important to note that under se matched rmal dissipation conditions COP of actuated is 535 while corresponding COP of baseline is 220 which is 1.4 times less efficient than -enhanced configuration. Fig. 11. Variation of COP with volume rate for baseline channel for a power dissipation of W ( ), 20W ( ) - to enhanced channel at a wall temperatures that are approximately equal baseline channel for each rate ( ). V. PIV MEASUREMENTS The field in cross stream center-span (zz = 0) plane within channel is measured at 30 LPM (Re w = 2200) using high magnificationn particle image velocimetry in baseline in presence of active. The measurement domain extends 4 channel widths (10 mm) downstream of tip of magnification is 3μm/pixel. The effect of on is demonstrated using phase-averaged cross stream distributions of spanwise vorticity concentrations velocity vectors at two phases during motion of (having period T) as shown in Fig.. At t = T tip of is 0.47 mm above (Fig. a) 0.5 mmm below (Fig. b) channel s centerline. These distributions show CW (red) CCW (blue) vorticity concentrations that are advected past s blade (above below centerline) along with corresponding vorticity concentrations within boundary layers on top bottom surfaces. These images also show induced changes in speed of core as moves closer to ways from channel walls. Perhaps most salient feature of se data is clear evidence of time-periodic disruption of within wall boundary layers indicating enhanced heat transfer mixing with cooler core. The time-periodic motions that are induced by lead to formation shedding of counter rotating vortical structures that interact with disrupt boundary layers on channel s walls. Fig. 13a b are cross stream maps of time-averaged turbulent kinetic energy (TKE) within same streamwise domain in absencee presence of actuation. In absence of actuation (Fig. 13a), TKE is relatively low (considering channel Reynolds number). However, when is actuated (Fig. 13b) ree is a remarkable enhancement in TKE whichh is indicative of enhanced small-scale motions within channel refore strong mixing with core.

6 6-8October, Barcelona, Spain ACKNOWLEDGMENT This work was supported by DARPA s Microsystems Technology ice (MTO) MACE program. Fig..Phase-lockespanwise vorticity concentrations downstream of vibrating. Fig. 13. Time-averaged TKE in absence (a) presence (b) of PIV measurements of cross stream velocity actuation for volume rate of 30LPM. VI. CONCLUSIONS The present experimental investigationn focuses on heat transfer enhancement within a mm-scale straight channel that is characteristic of low-reynolds number s within high-power heat sink channels. Active enhancement of heat transfer within heat sink channels is effected by induced small-scale motions that disrupt momentum rmal boundary layers enhances rmal mixing within channel s core, reby significantly decreasing fin-to-air heat transfer resistance. These motions are induced by time-periodic motion of miniature piezoelectric cantilever vibrating s that are placed within channel so that ir planform surfaces are parallel to channel walls actuation period is smaller than characteristic time of flight through channel. Comparisonss between baseline (in absence of ) in presence of actuation show that at same total fluid power investment, actuation results in a 42% increase in heat transfer coefficient at about half rate. Furrmore, comparisons of two s at a given power dissipation show that in order to achieve similar heat transfer coefficient total fluid power investment in baseline has to be nearly 2.9 times higher. Furrmore, for same wall temperature in two s, coefficient of performance (COP) in presence of actuation is approximately 1.4 times higher than in baseline. The present results indicate that technology has potential to significantly enhance efficiency of conventional, high-power heat sinks without ncreasing ir cooling volume rate or fin density. REFERENCES [1] D.S. Kercher, J. Lee, O. Br, M. G. Allen A. Glezer, Microjet Cooling Devices for Thermal Management of Electronics,, IEEE Transactions on Components Packaging Technologies, vol. 26, No.2, pp , [2] A. Pavlova M. Amitay, Electronic cooling using syntic jet impingement, Journal of Heat Transfer, vol. 8, pp , [3] R. Mahalingam A. Glezer, Design rmal characteristics of a syntic jet ejector heat sink, Journal of Electronic Packaging, vol. 7, pp , [4] Gerty, D., Mahalingam, R., et al. (2006). "Design Characterization of a Heat Sink Cooled by an Integrated Syntic Jet Matrix." Proceedings of ITHERM 2006, May 30-June 2, 2006, San Diego, CA. [5] D. Gerty, D.W. Gerlach, Y..K. Joshi A. Glezer, Development of a prototype rmal management solution for 3-D stacked chip electronics by intervealed solid spreaderss syntic jets, THERMINIC 2007, Budapest, Hungary, September [6] K. Kota, P. Hidalgo, Y..K. Joshi A. Glezer, Thermal management of a 3D chip stack using a liquid interfacee to a syntic jet cooled spreader, THERMINIC 2009, Leuven, Belgium, 7-9 October [7] T. A ikalin, S.M. Wait, S.V. Garimella A. Raman, Experimental investigation of rmal performance of piezoelectric fans, Heat Transfer Engineering, vol. 25, pp. 4-14, [8] T. A ikalin, S.V. Garimella, A. Raman J. Petroski, Characterization optimization of rmal performance of miniature piezoelectric fans, Int. J. Heat Fluid Flow vol. 28, pp , [9] S.M. Wait, S. Basak, S.V. Garimella A. Raman, Piezoelectric fans using higher flexural modes for electronics cooling applications, IEEE Transactions on Components Packaging Technologies, vol. 30, No.1, pp.119-8, [10] D. Gerty, Fluidic-driven Cooling of Electronic Hardware: Part I: Channel Integrated Vibrating Reed; Part II: Active Heat Sink,, PhD Thesis, 2008.