VAPOR-VENTING, MICROMACHINED HEAT EXCHANGER FOR ...
Proceedings of IMECE2007 2007 ASME International Mechanical Engineering Congress and Exposition November 11-15, 2007, Seattle, Washington, USA
IMECE2007-42553 VAPOR-VENTING, MICROMACHINED HEAT EXCHANGER FOR ELECTRONICS COOLING Milnes P. David Department of Mechanical Engineering, Stanford University, MERL 247 - Microscale Heat Transfer Lab, Stanford, CA 94305 [email protected]
Carlos Hidrovo Department of Mechanical Eng., Stanford University
Tarun Khurana Department of Mechanical Engineering, Stanford University, Room 101, Bldg 530, Stanford, CA 94305 [email protected]
Beth L. Pruitt Department of Mechanical Eng., Stanford University
ABSTRACT The increasing complexity of modern integrated circuits and need for high-heat flux removal with low junction temperatures motivates research in a wide variety of cooling and refrigeration technologies. Two-phase liquid cooling is especially attractive due to high efficiency and low thermal resistances. While two-phase microfluidic cooling offers important benefits in required flow rate and pump size, there are substantial challenges related to flow stability and effective superheating. This work investigates the use of hydrophobic membrane to locally vent the vapor phase in microfluidic heat exchangers. Previous work has demonstrated selective venting of gas in microstructures and we extend this concept to twophase heat exchangers. This paper details the design, fabrication and preliminary testing of the novel heat exchanger. Proof-of-concept of the device, carried out using an isothermal air-water mixture, found the air-mass venting efficiency exceeding 95%. Two-phase, thermal operation of the heat exchanger found the pressure-drop to be smaller compared to a two-phase, non-venting model. The paper also includes a discussion of design challenges such as membrane leakage and optical inaccessibility. The favorable results demonstrated in this first-generation, vapor-venting, micromachined, heat exchanger motivates further study of this and other novel microstructures aimed at mitigating the negative effects of phase-change. With continued research and optimization, we believe two-phase
Kenneth E. Goodson Department of Mechanical Eng., Stanford University
cooling is a viable solution for high heat flux generating electronics. INTRODUCTION The ITRS roadmap [1], Fig. 1, helps predict the power dissipation and junction temperature requirements in next generation IC’s. As the complexity of IC’s increase, power dissipation continues to rise while the junction temperature needs to be held below 90 oC. The total thermal resistance, Eq. 1, must reduce.
Rth =
∆T q dissipated
=
T j − Ta q dissipated
(1)
Cooling technologies such as fans (Rth ~ 10-100 K/W) and heat pipes (Rth ~ 1-10 K/W) [2] are reaching the limit of their cooling capacity. The future technology roadblock has spawned considerable research into alternate cooling strategies such as thermoelectric [3,4], micro refrigeration [5], micro-jet [6-8] and microchannel liquid cooling. Microchannel heat exchangers using pumped liquid have many advantages: low thermal resistances due to large wetted areas and high heat transfer coefficients, small form factors and simple fabrication. Single-phase liquid cooling using microchannels [9-11] can dissipate large heat fluxes but at the cost of higher pumping power. As power budgets for cooling shrink, two-phase liquid cooling [12-15] is more
1
Copyright © 2007 by ASME
advantageous. Lower pumping power is required to dissipate the same heat flux due to the large enthalpy of vaporization. Microchannel systems using refrigerants or water at subambient pressures can achieve uniform junction temperatures below 90 oC.
A) 102 100
Power (W)
200
98 150
96
100
Cost Performance High Performance Junction Temp
50 0
94 92 90
Junction Temperature (C)
250
88
two detrimental effects: an increase in the pumping power and a rise in the saturation, and thus junction temperatures. Further problems include flow instabilities [16-17], as well as dry-out. Dry-out greatly lowers the local heat transfer coefficient and raises the junction temperature. The solution to vapor entrapment, first proposed by Zhou et al. [18], is to locally vent the vapor formed in the heat exchanging microchannels into a separate vapor-carrying channel. Venting the vapor reduces the two-phase frictional pressure drop to the order of a single-phase flow. The smaller pressure drop reduces pumping power and ensures little change in saturation conditions. To determine the performance improvement we simulated two-phase flow through a microchannel with, and without, vapor venting. The simulation was carried out on a 10 mm long microchannel with a hydraulic diameter of 100 µm and uniform heat dissipation of 100 W/cm2. Flow rate was set at 0.05 ml/min (103 kg/s/m2), outlet pressure at 101 kPa and inlet liquid temperature at 25 o C. The results of the simulation, Fig. 2, show a dramatic improvement in pressure drop in the channel as well as lowered saturation pressure on vapor venting.
2005 2008 2011 2014 2017 2020
Year Two-phase
B) Singl e-pha se
Thermal Resistance (K/W)
0.70
Cost Performance High Performance
0.60 0.50 0.40 0.30
Temperature profile
Pressure profile
0.20 0.10 0.00 2005
2008
2011
2014
2017
2020
Year FIG. 1: ITRS Roadmap 2006 showing A) the rise in power dissipation and reduction of junction temperature and B) the reduction in thermal resistance for high performance and cost performance IC’s However, industry has been slow to adopt a two-phase, microchannel heat exchanger as a reliable cooling solution due to serious drawbacks. The confined geometry of the microchannels causes the vapor formed during phase change to rapidly fill and ‘plug’ the channels. This added friction significantly raises the pressure-drop in the channel, producing
Fig. 2: Simulations of pressure drop and fluid temperature in a channel with no venting (solid line) and complete venting (dashed line) show improved performance on venting The simulation uses a 1-D forward discretization scheme with geometry, inlet pressure, inlet temperature, flow rate and power dissipation profile specified. At the inlet, the liquid is single phase and we use a single-phase pressure drop and heat transfer model with friction factor and Nusselt number appropriate for rectangular channels. As energy is added along the channel, the liquid temperature rises until it reaches the saturation point for the given local pressure. At this point the simulation switches to the Lockhart & Martinelli two-phase
2
Copyright © 2007 by ASME
separated flow pressure drop model with laminar liquid and laminar vapor flow and the Lazarack & Black heat transfer model. The simulation continues till the end of the channel, with local vapor quality determined using an enthalpic balance and a Matlab based steam table [19]. All fluid properties are re-calculated for each cell based on the local temperature and pressure. In the case of complete venting, the vapor is removed as it is formed in each cell, and the total enthalpy and mass flow rate adjusted to satisfy conservation of mass and energy. For this simulation, we neglected solid conduction and environmental losses. Experimentally, previous work has demonstrated gas separation from liquids, both actively [20] and passively [21], in micro-devices. Designed for biological, chemical and fuel cell applications, these devices and are not optimized for use as heat exchangers. Therefore, we designed and fabricated a vapor-venting device specifically for use as a heat exchanger.
DESIGN Early in the design process, we generated several alternate designs incorporating different material combinations, which we then weighed using a Quality Function Deployment (QFD) analysis. Lamers et al. discuss this analysis in detail [22]. The QFD analysis suggests the development of a copper-polymer heat exchanger for commercial applications and a siliconpolymer heat exchanger for research applications due to easier metrology and greater design flexibility. Once we had chosen the materials, we could then design the specific components such as channels, heaters and sensors. In heat exchanger channel design, the two key goals are minimizing the thermal resistance, Eq. 1, and maximizing the power ratio, Eq. 2.
Power Ratio =
DEVICE DESIGN AND FABRICATION Based on overall dimensional, temperature and heat flux requirements, we proposed the silicon heat exchanger shown in Figure 3. A hydrophobic porous membrane ensures only vapor can vent into the vapor channels while liquid does not leak through. The first generation device design includes integrated resistive heaters to simulate an IC producing 200 W/cm2 of heat flux and resistive temperature sensors, accurate to 1 oC, to study junction temperatures and heat transfer coefficients. The following sections discuss the details of the design that led to this structure and the subsequent fabrication process.
Doublesticky tape
Liquid channel
q dissipated
(2)
Pump Power
Assuming the vapor phase vents away, we can simplify the design analysis by using single-phase pressure drop and heattransfer correlations for laminar, thermally fully developed flow. This leads to the following dependencies for thermal resistance, Eq. 3, and power ratio, Eq. 4:
Rth ∝
1 1 1 , , , Dh Q pump k fluid Asurface
Power Ratio ∝
Vapor bubbles forming and venting
1 Q pump
, k,
1
η
, Asurface , D h3 ,
(3)
1 L
(4)
Vapor channel Glass Porous, hydrophobic membrane
Epoxy Heat flux
Silicon Oxide
Silicon
Liquid inlet
Aluminum temperature sensors
Aluminum heater
Liquid outlet
Vapor outlet
Fig. 3: Schematic of proposed vapor venting heat exchanger
3
Copyright © 2007 by ASME
The flow-rate, Qpump, thermal conductivities, k, and fluid viscosity, η, are set by the pump and materials leaving the hydraulic diameter, Dh, wetted surface area, Asurface, and channel length, L, to be specified. Large surface area and short channels are obvious benefits. Typically, the designer chooses Dh based on relative importance of Power Ratio and Rth to an application. Since the effect of geometry on venting behavior is a key question in this work, we designed a variety of channel geometries. Geometries considered in our work includes 400 µm, 200 µm and 100 µm wide parallel channels, 200 µm wide single-channel serpentines and 200 µm square pin-fin arrays in both parallel and staggered form. The vapor channels need to
be as large as possible to maintain small pressure drops and avoid reverse venting. However, the depth and thus Dh of the vapor channel is limited by the materials available to construct the walls of the vapor channel. We then designed the heaters and sensors considering performance specifications, lithographic resolution, current density limits and available spatial footprint. For the vapor separation layer we obtained PTFE membranes with 0.22 µm pores and Nylon membranes with 0.1 µm pores from GE Osmonics and Sterlitech. PTFE and Nylon membranes are hydrophobic and have hightemperature stability and the small pores minimize liquid leakage through the membrane.
A)
Growth of 0.25 µm of oxide on 400 µm double polished Si wafer, followed by sputtering of 0.5 µm of Al
B)
Once heaters and sensors have been patterned (Mask 1) and dry-etched, 0.5 µm of LTO is deposited to protect the metal structures
C)
Backside LTO is patterned (Mask 2) and then pad-etched to allow wire bonding. Frontside oxide is stripped for dry etching.
D)
Front-side liquid channels are then patterned (Mask 3) and dry-etched followed by backside pattern (Mask 4) and dry-etch to create ports and release device
E)
2.5 µm of SU-8 is then contact printed on front-side using a SU-8 coated transfer wafer
F)
Membrane is attached and the SU-8 UV exposed and hard cured at 150 oC
G)
Double sticky mylar sheet is laser ablated to form vapor channels and attached to top of membrane followed by a glass slide to cap the device.
Porous membrane (50 µm)
SiO2 (0.25 + 0.5 µm)
SU-8 (2.5 µm)
Double-sticky mylar (50 µm)
Aluminum (0.5 µm)
Glass slide (2 mm)
Fig. 4: Process flow for fabrication of the device
4
Copyright © 2007 by ASME
A)
B) 7 mm
Heaters
Temperature sensor
Vapor ports
Thermal isolation trench
Liquid ports
Fig. 5: Images showing (A) backside and (B) front-side of the device after release from the wafer and before membrane installation
Surfboards
FABRICATION We generated a four-mask process to fabricate our device. Figure 4 describes the fabrication scheme. Figure 5 shows both sides of a 200 µm wide serpentine channel device after step D in Fig. 4 was completed. The central, operational region of the device has an area of 0.75 cm2. The central region consists of four heaters and sixteen distributed temperature sensors on the back and micro-channels on the front. The extensions on either side of the center are for fluidic interfacing. The thermal isolation trench is necessary to localize the heat in the central region and minimize conduction losses to the extensions. After the last step in Fig. 4 is complete, surfboards are attached and the device wire-bonded, resulting in the final product shown in Fig. 6. The current yield of devices was poor due to equipment issues and fabrication difficulties in steps D and E in Fig. 4. We expect the yield to improve in future runs.
Glass slide
Fig. 6: Completed, wire-bonded device before integration of fluidic connectors
EXPERIMENTAL ISOTHERMAL AIR-WATER TEST The heat exchanger was first tested using an isothermal, air-water flow to prove venting does occur and to quantify the venting efficiency of the device. Using an air-water mixture separates out any thermal interactions in the venting behavior and simplifies the measurement of venting efficiency. Venting efficiency is defined as the ratio of mass of air vented to the mass of air input to the device. Using the setup in Fig. 7, we measured the venting efficiency of a 400 µm wide (Dh = 250 µm), parallel-channel device with a PTFE membrane.
PTFE Membrane
First, we pumped DI water at 0.5 ml/min through the device. Once steady state conditions were obtained, we switched on the syringe-pump and pumped air at 1 ml/min. This results in the formation of intermittent slugs of air and water downstream of the mixing junction. Once steady state conditions were established for the mixed flow, we collected data from the pressure sensor and the four IR slot sensors. At the completion of data collection, we switched off the airflow and pumped only DI water through the device to remove any remaining air bubbles in the junction, tubing and device. We then repeated the process for the next set of water and air flowrates.
5
Copyright © 2007 by ASME
Syringe pump containing air
Upstream IR Slot Sensors
Vapor channel Liquid channel
Downstream IR Slot Sensors
0.5 um filter PC with DAQ
Pi Peristalsis pump
DI Water
Water output
Fig. 7: Experimental setup to quantify vapor venting efficiency of the device. IR slot sensors used to measure void fraction and interface velocities from which air volume and mass is calculated
We calculate the air volumes by measuring the interface velocities and void fraction of air over a given period using the two pairs of IR slot sensors. We can then calculate the air mass using the ideal gas law with room temperature properties. A summary of the results is provided in Table 1. Table 1: Summary of measured venting efficiency for various combinations of air and water flow rates.
AQ ml/min 0.4 0.6 0.8 1 2 2 2 2
LQ ml/min 0.5 1 1 1 0.5 0.7 1 2
Venting Eff. % 100.00 100.00 100.00 100.00 99.95 99.77 99.75 99.97
Under all flow conditions we tested, almost all the air vents through the membrane. Measurement of inlet air mass at air flow-rates of 2 ml/min is difficult using the current setup. When an air-slug vents in the device, the flow resistance in the fluidic circuit drops, causing a sudden acceleration in the flow. The calculations overestimate the mass of air entering the
device because the higher velocity is not constant over a single air-slug and the venting non-periodic due to the differing sizes of the air slugs. Downstream measurements are accurate as the flow is unaffected by the upstream venting and flow rate is constant. Using visual estimates of the average size of an air slug when pumped at 2 ml/min, we calculated a very conservative estimate of air mass venting efficiency of 95%. Venting isn’t perfect at the higher air flow-rate of 2 ml/min because the higher flow velocity during venting results in insufficient time for the larger air slug to completely vent. In summary, even with a very conservative estimate of venting efficiency at the higher air flow-rates the message is clear – the heat exchanger has the capacity to vent a large fraction of air in an air-water mixture. THERMAL, PHASE-CHANGE TEST Having verified the ability to vent air out of an air-water mixture still left the open question of whether the device can vent under operating conditions. To test this, we attached the same 400 µm wide parallel channel device tested above to the setup shown in Fig. 8. We connected the two downstream heaters on the device to 0-25 V DC power supplies and the sensors to the PC-based DAQ board via a signal-conditioning box. We also attached the 0-5 psig, Omega pressure sensor to the DAQ board for simultaneous measurements. We pumped house air at 5 psi (34 kPa) through the vapor side of the device to facilitate removal of condensed and leaked water. Liquid water can block the vapor channel and reduce venting.
6
Copyright © 2007 by ASME
Microscope 5 psi, pressurized air inlet
Vapor channel Liquid channel 0.5 um filter
Pi DI Water
PC with DAQ
Liquid output
Peristalsis pump
Signal Cond.
Vapor output
DC Power Supply
Fig. 8: Setup used to carry out thermal test with phase change in device temperature sensors. We believe the problem stems from processing and wire-bonding issues. We will investigate these issues in detail before the next round of fabrication.
0.5 ml/min 4.00 3 ml/min P drop/P ini
We pumped DI Water at 0.5 ml/min through the device for a few minutes until steady state conditions were established. We then slowly raised the power input to both heaters in equal steps, starting at 0 V. At each step rise, we allowed the device to operate for a few minutes to ensure steady state behavior before collecting data from the temperature and pressure sensors. We continued ramping the power until boiling occurred. Venting in the membrane along with vapor flow in the vapor-side outlet tubing indicated the onset of boiling. We raised the power a minimum of 30% beyond this point to obtain higher quality flows and to gather data beyond the onset of boiling. We then switched off the power and continued pumping water to cool the device back to room temperatures, before repeating the experiment at another flow rate. The experimental and simulated results for two water flow rates, 0.5 ml/min and 3 ml/min is shown in Fig. 9. Power input is non-dimensionalised with the theoretical power required to boil water, calculated as ρQflowCp∆T with a ∆T of 80 oC. Pressure drop is non-dimensionalised with the pressure drop for single-phase flow with no heating. The maximum total power output by the heaters is 4.5 W and 21.5 W for the 0.5 ml/min and 3 ml/min case respectively. The plot clearly shows a much lower pressure drop in the channel than expected from simulations. In summary, preliminary data collected from a 400 µm wide, parallel-channel device shows improved pressure-drop trends. We will quantify this improvement once a statistically significant amount of data is collected. Temperature data has not been included here due to poor performance of the
0.5 ml/min SIM 3 ml/min SIM
2.00
0.00 0
0.2 0.4 0.6 0.8
1
1.2 1.4 1.6 1.8
Total Actual Power / Theoretical Power
Fig. 9: Comparison of experimental data with simulations of two-phase flow in a non-venting device show much smaller pressure drops for both 0.5 ml/min and 3 ml/min liquid flow-rates
7
Copyright © 2007 by ASME
A)
B)
2V, 0.1 W
5V, 0.58 W
C)
13 V, 3.3 W
50 um
Non-ideal regions on membrane leads to droplet growth
Dry-out of membrane due to boiling and venting
Fig. 10: Images of same location on membrane taken using a microscope while increasing power to the heaters. A) and B) show leakage of water through non-ideal regions of membrane and C) shows vaporization of droplets and venting of vapor CHALLENGES AND SOLUTIONS LEAKAGE Leakage through the PTFE membrane, as seen in Fig. 10, increases with power input to the heaters until the membrane temperature is high enough to boil the wicked water away. This leakage necessitates the use of pumped air to remove the water droplets forming on the vapor side. Pumping air is an impractical final solution, as the ultimate goal is to operate the heat exchanger in a closed loop with a single pump. Wicking also complicates device modeling, which must now include phase change and two-phase flow within the membrane. To understand this leakage we consider the expression for leakage pressure through a pore [21], obtained from the Young-Laplace equation, and given as:
Leakage Pressure, ∆P =
4σ cos(180 − θ adv,max ) d
(5)
where σ is the surface tension of the fluid at some temperature, θadv,max the advancing angle of the fluid in the membrane material and d the diameter of the pore. Using characteristic values for PTFE and water at room temperature, with σ = 0.072 N/m, θadv,max = 129o and d = 0.22 µm, the leakage pressure is approximately 820 kPa. However, the liquid pressure in the channels is less than 30 kPa at the highest tested flow rate of 3 ml/min. σ, θadv,max and/or d must deviate from standard values, as listed below, resulting in a much lower leakage pressure. i)
σ is inversely proportional to the temperature and reduces to 0.06 N/m at 100 oC. The reduction in surface tension is insufficient to be the sole culprit of leakage, but does explain the increase in leakage with heater power.
ii) Reduction of θadv,max on PTFE due to surface or water contamination. Previous work documents the loss of hydrophobicity in PTFE [23] and Nylon [24] when immersed in saline solutions at various temperatures. iii) Defects in the membrane as well as the natural distribution of pore sizes in a mesh structure such as PTFE results in some fraction of pore-diameters larger than 0.22 µm. We believe that leakage occurs due to some convolution of these three non-idealities. Only a small fraction of the membrane area need be non-ideal to cause leakage. In Fig. 10.A, wetting initially occurs in distinct, darker, patches, over which larger droplets form as leakage increases as seen in Fig. 10.B. Using higher quality membranes with smaller pore size distribution and smaller pores can reduce leakage. Examples include PTFE membranes with smaller pores, porous glass and porous silicon. This solution requires design optimization, as the smaller pores reduce venting efficiency due to higher pressure-drop across the membrane. Porous glass and silicon have the added disadvantage of requiring low-energy coatings to ensure hydrophobicity. Using pure, filtered DI water is also important along with taking greater care to not contaminate or damage the membrane before device integration. OPTICAL ACCESS In the current design, the PTFE membrane is opaque, limiting the ability to study the flow regimes and venting mechanism in the liquid channels. This lack of information hinders both development of an appropriate two-phase model and optimization of liquid channel geometry to maximize venting efficiency. Using a quartz substrate, instead of silicon, allows optical access to the liquid channels through the backside of the device. Similarly, optical access from the front is possible using membranes such as porous polycarbonate and porous
8
Copyright © 2007 by ASME
glass. The fabrication of a quartz structure for the liquid channels is straightforward, but channel geometries are limited by isotropic etching. The low thermal conductivity of quartz would also make high heat flux studies challenging. Porous polycarbonate and glass require surface treatments to be hydrophobic and have only limited optical transparency. FUTURE WORK The first priority is to perfect the current fabrication process to maximize the yield and obtain devices with fully functional temperature sensors. Given a large number of fully functional devices, we can then carry out a detailed, statistically significant, fluid and thermal study. The study would quantify the performance improvement, provide a better understanding of the role of channel geometry and assist in the development of a two-phase, 2D, model of the flow in the device. We also expect the study to assist in the development of the next generation device that incorporates some of the proposed solutions to leaking and lack of optical access. Figure 11, illustrates one such example of a potential next generation device. Unlike the current stacked layer design, this structure provides excellent optical access to all parts of the structure. The design also allows fabrication of advanced porous structures such as CNT forests, ZnO nanotube forests and porous silicon within the device. CNT and ZnO nanostructures possess superhydrophobicity [25-28] with contact angles in excess of 140o, though CNT nanostructures require additional surface treatment for stable superhydrophobicity [27]. Nanotube forests can have various pore sizes, i.e. distance between adjacent nanotubes, based on growth conditions. Advantages of this design include: i) low leakage with the help of superhydrophobic nm-pore membranes, ii) ability to engineer membranes to specifications and iii) excellent optical access that would assist in detailed studies of venting. However, this design is not suitable for commercial development due to cost of membrane fabrication and loss of heat transfer area.
Porous membrane
Vapor channel
Glass cover
Liquid channel
Fig. 11: Schematic of proposed side-by-side structure with either a surface treated porous silicon, surface treated CNT forest or ZnO nanotube forest separating the two channels SUMMARY Two-phase, microchannel based heat exchangers have significant advantages for use in high-heat flux electronics
cooling and represents an important area of research and development. Commercial adoption has been slow due to the problems associated with vaporization of the working fluid. Motivated by patent work, simulations and previous work in gas venting, we designed and fabricated a vapor venting, microchannel heat exchanger. Vapor venting occurs through a porous, hydrophobic, membrane sandwiched between the liquid-carrying and vapor-carrying channels. Venting the vapor helps mitigate some of the problems associated with vapor entrapment in the microchannels. We carried out a proof-of-concept experiment on a 400 µm wide, parallel-channel device, using an air-water mixture under isothermal conditions. The air mass venting efficiency exceeded 95% for airflow rates of 2 ml/min and 100% for lower airflow rates. Venting efficiency drops at the higher air flow-rate due to high velocities coupled with large air slugs that result in incomplete venting. We also ran thermal, phasechange tests at flow rates of 0.5 ml/min and 3 ml/min and the results show a much lower pressure drop compared to flow simulations of a non-venting device under similar operating conditions. Challenges in the current design include liquid leakage through the membrane and lack of optical access to the liquid channels. To reduce leakage, membranes with smaller pore sizes and tighter pore size distributions must be used. Examples include PTFE with smaller pore sizes, porous glass and porous silicon. Care must be taken to use pure filtered DI water and not contaminate or damage the membranes is any manner during integration into the device. Optical access is achieved by fabricating the liquid channels on a quartz substrate instead of silicon or by using semi-transparent membranes such as polycarbonate or porous glass. Finally, we proposed a next generation device design where the microchannels and membrane are side-by-side instead of stacked. This design provides the best optical access, low leakage, and flexibility to engineer membrane characteristics. Potential membrane materials include porous silicon and superhydrophobic CNT or ZnO nanotube forests. In conclusion, this preliminary study proves that vapor venting in heat exchangers is both possible and provides measurable benefits. This success motivates the overall goal of this project: to develop a reliable, closed-loop, two-phase heat exchanger for use in cooling of next generation, high heat-flux generating electronics. ACKNOWLEDGEMENTS The author would like to acknowledge the support of the MARCO Interconnect Focus Center, one of five research centers funded under the Focus Center Research Program, a Semiconductor Research Corporation program. The author also appreciates support from the Semiconductor Research Corporation through Task 1445 as well as the Stanford Graduate Fellowships. Fabrication was performed in the Stanford Nanofabrication Facility, a part of the National
9
Copyright © 2007 by ASME
Nanotechnology Infrastructure Network, and supported by the National Science Foundation under Grant ECS-9731293. REFERENCES [1] ITRS Roadmap 2006 Update, http://public.itrs.net/, Aug. 2007. [2] Kandlikar, S. G., et al. (eds.), Handbook of Phase Change: Boiling and Condensation, Philadelphia: Taylor & Francis, 1999, Chp. 16, pp. 405-409. [3] Cancheevaram, J., et al., “Performance of Integrated thinfilm thermoelectrics in cooling “hot-spots” on microprocessors – Experimental setup and results,” Mat. Res. Soc. Symp. Proc., Vol. 793, 2004, pp. S8.18.1-S8.18.7. [4] Chen, C., Shakouri, A., “Heat transfer in nanostructures for solid-state energy conversion,” J. of Heat Transfer, Vol. 124, No. 2, 2002, pp. 242-252. [5] Phelan, P. E., “Current and future miniature refrigeration cooling technologies for high power microelectronics,” IEEE Transaction on Components and Packaging Technologies, Vol. 25, No. 3, 2002, pp. 356-365. [6] Wang, E. N., et al., “Micromachined jets for liquid impingement cooling of VLSI chips,” JMEMS, Vol. 13, No. 5, 2004, pp. 833-842. [7] Kercher, D. S., et al., “Microjet cooling devices for thermal management of electronics,” IEEE Transactions on Components and Packaging Technologies, Vol. 26, No. 2, 2003, 359-366. [8] Brunschwiler, T., et al., “Direct liquid jet impingement cooling with micron sized nozzle array and distributed return architecture,” The Tenth Intersociety Conference on Thermal and Thermomechanical Phenomena in Electronics Systems, ITHERM ‘06, San Diego, CA, 30 May - 2 June, 2006, pp. 196-203. [9] Qu, W., Mudawar, I., “Experimental and numerical study of pressure drop and heat transfer in a single-phase microchannel heat sink,” Int. J. of Heat and Mass Transfer, Vol. 45, No. 12, 2002, pp. 2549-2565. [10] Kandlikar, S., Grande, W., “Evaluation of single phase flow in microchannels for high heat flux chip cooling – thermohydraulic performance enhancement and fabrication technology,” Heat Transfer Engineering, Vol. 25, No. 8, 2004, pp. 5-16. [11] Garimella, S., Singhal, V., “Single phase flow and heat transport and pumping considerations in microchannel heat sinks,” Heat transfer engineering, Vol. 25, No. 1, 2004, pp. 1525. [12] Jiang, L., Wong, M., Zohar, Y., “Forced Convection Boiling in a Microchannel Heat Sink,” JMEMS, Vol. 10, No. 1, 2001, pp. 80-87. [13] Zhang, L. et al., “Measurements and Modeling of TwoPhase Flow in Microchannels with Nearly Constant Heat Flux Boundary Conditions,” JMEMS, Vol. 11, No. 1, 2002, pp. 1219.
[14] Qu, W., Mudawar, I., “Flow boiling heat transfer in twophase micro-channel heat sinks--I. Experimental investigation and assessment of correlation methods,” International Journal of Heat and Mass Transfer, Vol. 46, No. 15, July 2003, pp. 2755-2771. [15] Hestroni, G., et al., “A uniform temperature heat sink for cooling of electronic devices,” International Journal of Heat and Mass Transfer, Vol. 45, No. 16, July 2002, pp. 32753286. [16] Kennedy, J. E., et al., “The onset of flow instability in uniformly heated horizontal microchannels,” J. of Heat Transfer, Vol. 122, No. 1, 2000, pp. 118-125. [17] Wu, H. Y., Cheng, P., “Boiling instability in parallel silicon microchannels at different heat flux,” Int. J. of Heat and Mass Transfer, Vol. 47, No. 17-18, 2004, pp. 3631-3641. [18] Zhou, P., Goodson, K. E., Santiago, J., U.S. Patent 6,994,151 B2, 2006. [19] Brissette, F., Matlab Steam Table, 2005, Montreal, Canada. [20] Yang, Z., et al., “A prototype of ultrasonic microdegassing device for portable dialysis system,” Sensors and Actuators A, Vol. 95, 2002, pp. 274-280. [21] Meng, D.D. et al, “A Degassing plate with hydrophobic bubble capture and distributed venting for microfluidic devices,” J. of Micromech. Microeng., Vol. 16, 2006, pp. 419424. [22] Lamers, T. L., et al., “Application of a modified quality function deployment method for MEMS,” 2007 ASME IMECE, Seattle WA, November 11-15, 2007, Paper IMECE2007-42374. [23] Syed, H. Z., Hackam, R., “Effects of water salinity, electric stress and temperature on the hydrophobicity of polytetrafluoroethane”, IEEE Conference on electrical insulation and dielectric phenomena, 1998 – Annual Report, Atlanta GA, Oct. 25-28, 1998, pp. 100-103. [24] Tokoro, T., Hackam, R., “Loss and recovery of hydrophobicity, surface energies, diffusion coefficients and activation energy of nylon”, IEEE Transactions on dielectrics and electrical insulation, Vol. 6, No. 5, 1999, pp. 754-763. [25] Feng, L., et al., “Superhydrophobicity of Nanostructured Carbon Films in a wide range of pH values,” Angew. Chem. Int. Ed., Vol. 42, 2003, pp. 4217-4220. [26] Li, S., et al., “Superhydrophobicity of large area honeycomb like aligned carbon nanotubes,” J. Phys. Chem. B, Vol. 106, 2002, pp. 9274-9276. [27] Lau, K., et al., “Superhydrophobic Carbon Nanotube Forests,” Nano Letters, Vol. 3, No. 12, 2003, 1701-1705. [28] Feng, X., et al., “Reversible Super-Hydrophobicity to Super-Hydrophilicity transition of aligned ZnO nanorod films,” J. Am. Chem. Soc., Vol. 126, 2004, 62-63.
10
Copyright © 2007 by ASME
IMECE2007-42553 VAPOR-VENTING, MICROMACHINED HEAT EXCHANGER FOR ELECTRONICS COOLING Milnes P. David Department of Mechanical Engineering, Stanford University, MERL 247 - Microscale Heat Transfer Lab, Stanford, CA 94305 [email protected]
Carlos Hidrovo Department of Mechanical Eng., Stanford University
Tarun Khurana Department of Mechanical Engineering, Stanford University, Room 101, Bldg 530, Stanford, CA 94305 [email protected]
Beth L. Pruitt Department of Mechanical Eng., Stanford University
ABSTRACT The increasing complexity of modern integrated circuits and need for high-heat flux removal with low junction temperatures motivates research in a wide variety of cooling and refrigeration technologies. Two-phase liquid cooling is especially attractive due to high efficiency and low thermal resistances. While two-phase microfluidic cooling offers important benefits in required flow rate and pump size, there are substantial challenges related to flow stability and effective superheating. This work investigates the use of hydrophobic membrane to locally vent the vapor phase in microfluidic heat exchangers. Previous work has demonstrated selective venting of gas in microstructures and we extend this concept to twophase heat exchangers. This paper details the design, fabrication and preliminary testing of the novel heat exchanger. Proof-of-concept of the device, carried out using an isothermal air-water mixture, found the air-mass venting efficiency exceeding 95%. Two-phase, thermal operation of the heat exchanger found the pressure-drop to be smaller compared to a two-phase, non-venting model. The paper also includes a discussion of design challenges such as membrane leakage and optical inaccessibility. The favorable results demonstrated in this first-generation, vapor-venting, micromachined, heat exchanger motivates further study of this and other novel microstructures aimed at mitigating the negative effects of phase-change. With continued research and optimization, we believe two-phase
Kenneth E. Goodson Department of Mechanical Eng., Stanford University
cooling is a viable solution for high heat flux generating electronics. INTRODUCTION The ITRS roadmap [1], Fig. 1, helps predict the power dissipation and junction temperature requirements in next generation IC’s. As the complexity of IC’s increase, power dissipation continues to rise while the junction temperature needs to be held below 90 oC. The total thermal resistance, Eq. 1, must reduce.
Rth =
∆T q dissipated
=
T j − Ta q dissipated
(1)
Cooling technologies such as fans (Rth ~ 10-100 K/W) and heat pipes (Rth ~ 1-10 K/W) [2] are reaching the limit of their cooling capacity. The future technology roadblock has spawned considerable research into alternate cooling strategies such as thermoelectric [3,4], micro refrigeration [5], micro-jet [6-8] and microchannel liquid cooling. Microchannel heat exchangers using pumped liquid have many advantages: low thermal resistances due to large wetted areas and high heat transfer coefficients, small form factors and simple fabrication. Single-phase liquid cooling using microchannels [9-11] can dissipate large heat fluxes but at the cost of higher pumping power. As power budgets for cooling shrink, two-phase liquid cooling [12-15] is more
1
Copyright © 2007 by ASME
advantageous. Lower pumping power is required to dissipate the same heat flux due to the large enthalpy of vaporization. Microchannel systems using refrigerants or water at subambient pressures can achieve uniform junction temperatures below 90 oC.
A) 102 100
Power (W)
200
98 150
96
100
Cost Performance High Performance Junction Temp
50 0
94 92 90
Junction Temperature (C)
250
88
two detrimental effects: an increase in the pumping power and a rise in the saturation, and thus junction temperatures. Further problems include flow instabilities [16-17], as well as dry-out. Dry-out greatly lowers the local heat transfer coefficient and raises the junction temperature. The solution to vapor entrapment, first proposed by Zhou et al. [18], is to locally vent the vapor formed in the heat exchanging microchannels into a separate vapor-carrying channel. Venting the vapor reduces the two-phase frictional pressure drop to the order of a single-phase flow. The smaller pressure drop reduces pumping power and ensures little change in saturation conditions. To determine the performance improvement we simulated two-phase flow through a microchannel with, and without, vapor venting. The simulation was carried out on a 10 mm long microchannel with a hydraulic diameter of 100 µm and uniform heat dissipation of 100 W/cm2. Flow rate was set at 0.05 ml/min (103 kg/s/m2), outlet pressure at 101 kPa and inlet liquid temperature at 25 o C. The results of the simulation, Fig. 2, show a dramatic improvement in pressure drop in the channel as well as lowered saturation pressure on vapor venting.
2005 2008 2011 2014 2017 2020
Year Two-phase
B) Singl e-pha se
Thermal Resistance (K/W)
0.70
Cost Performance High Performance
0.60 0.50 0.40 0.30
Temperature profile
Pressure profile
0.20 0.10 0.00 2005
2008
2011
2014
2017
2020
Year FIG. 1: ITRS Roadmap 2006 showing A) the rise in power dissipation and reduction of junction temperature and B) the reduction in thermal resistance for high performance and cost performance IC’s However, industry has been slow to adopt a two-phase, microchannel heat exchanger as a reliable cooling solution due to serious drawbacks. The confined geometry of the microchannels causes the vapor formed during phase change to rapidly fill and ‘plug’ the channels. This added friction significantly raises the pressure-drop in the channel, producing
Fig. 2: Simulations of pressure drop and fluid temperature in a channel with no venting (solid line) and complete venting (dashed line) show improved performance on venting The simulation uses a 1-D forward discretization scheme with geometry, inlet pressure, inlet temperature, flow rate and power dissipation profile specified. At the inlet, the liquid is single phase and we use a single-phase pressure drop and heat transfer model with friction factor and Nusselt number appropriate for rectangular channels. As energy is added along the channel, the liquid temperature rises until it reaches the saturation point for the given local pressure. At this point the simulation switches to the Lockhart & Martinelli two-phase
2
Copyright © 2007 by ASME
separated flow pressure drop model with laminar liquid and laminar vapor flow and the Lazarack & Black heat transfer model. The simulation continues till the end of the channel, with local vapor quality determined using an enthalpic balance and a Matlab based steam table [19]. All fluid properties are re-calculated for each cell based on the local temperature and pressure. In the case of complete venting, the vapor is removed as it is formed in each cell, and the total enthalpy and mass flow rate adjusted to satisfy conservation of mass and energy. For this simulation, we neglected solid conduction and environmental losses. Experimentally, previous work has demonstrated gas separation from liquids, both actively [20] and passively [21], in micro-devices. Designed for biological, chemical and fuel cell applications, these devices and are not optimized for use as heat exchangers. Therefore, we designed and fabricated a vapor-venting device specifically for use as a heat exchanger.
DESIGN Early in the design process, we generated several alternate designs incorporating different material combinations, which we then weighed using a Quality Function Deployment (QFD) analysis. Lamers et al. discuss this analysis in detail [22]. The QFD analysis suggests the development of a copper-polymer heat exchanger for commercial applications and a siliconpolymer heat exchanger for research applications due to easier metrology and greater design flexibility. Once we had chosen the materials, we could then design the specific components such as channels, heaters and sensors. In heat exchanger channel design, the two key goals are minimizing the thermal resistance, Eq. 1, and maximizing the power ratio, Eq. 2.
Power Ratio =
DEVICE DESIGN AND FABRICATION Based on overall dimensional, temperature and heat flux requirements, we proposed the silicon heat exchanger shown in Figure 3. A hydrophobic porous membrane ensures only vapor can vent into the vapor channels while liquid does not leak through. The first generation device design includes integrated resistive heaters to simulate an IC producing 200 W/cm2 of heat flux and resistive temperature sensors, accurate to 1 oC, to study junction temperatures and heat transfer coefficients. The following sections discuss the details of the design that led to this structure and the subsequent fabrication process.
Doublesticky tape
Liquid channel
q dissipated
(2)
Pump Power
Assuming the vapor phase vents away, we can simplify the design analysis by using single-phase pressure drop and heattransfer correlations for laminar, thermally fully developed flow. This leads to the following dependencies for thermal resistance, Eq. 3, and power ratio, Eq. 4:
Rth ∝
1 1 1 , , , Dh Q pump k fluid Asurface
Power Ratio ∝
Vapor bubbles forming and venting
1 Q pump
, k,
1
η
, Asurface , D h3 ,
(3)
1 L
(4)
Vapor channel Glass Porous, hydrophobic membrane
Epoxy Heat flux
Silicon Oxide
Silicon
Liquid inlet
Aluminum temperature sensors
Aluminum heater
Liquid outlet
Vapor outlet
Fig. 3: Schematic of proposed vapor venting heat exchanger
3
Copyright © 2007 by ASME
The flow-rate, Qpump, thermal conductivities, k, and fluid viscosity, η, are set by the pump and materials leaving the hydraulic diameter, Dh, wetted surface area, Asurface, and channel length, L, to be specified. Large surface area and short channels are obvious benefits. Typically, the designer chooses Dh based on relative importance of Power Ratio and Rth to an application. Since the effect of geometry on venting behavior is a key question in this work, we designed a variety of channel geometries. Geometries considered in our work includes 400 µm, 200 µm and 100 µm wide parallel channels, 200 µm wide single-channel serpentines and 200 µm square pin-fin arrays in both parallel and staggered form. The vapor channels need to
be as large as possible to maintain small pressure drops and avoid reverse venting. However, the depth and thus Dh of the vapor channel is limited by the materials available to construct the walls of the vapor channel. We then designed the heaters and sensors considering performance specifications, lithographic resolution, current density limits and available spatial footprint. For the vapor separation layer we obtained PTFE membranes with 0.22 µm pores and Nylon membranes with 0.1 µm pores from GE Osmonics and Sterlitech. PTFE and Nylon membranes are hydrophobic and have hightemperature stability and the small pores minimize liquid leakage through the membrane.
A)
Growth of 0.25 µm of oxide on 400 µm double polished Si wafer, followed by sputtering of 0.5 µm of Al
B)
Once heaters and sensors have been patterned (Mask 1) and dry-etched, 0.5 µm of LTO is deposited to protect the metal structures
C)
Backside LTO is patterned (Mask 2) and then pad-etched to allow wire bonding. Frontside oxide is stripped for dry etching.
D)
Front-side liquid channels are then patterned (Mask 3) and dry-etched followed by backside pattern (Mask 4) and dry-etch to create ports and release device
E)
2.5 µm of SU-8 is then contact printed on front-side using a SU-8 coated transfer wafer
F)
Membrane is attached and the SU-8 UV exposed and hard cured at 150 oC
G)
Double sticky mylar sheet is laser ablated to form vapor channels and attached to top of membrane followed by a glass slide to cap the device.
Porous membrane (50 µm)
SiO2 (0.25 + 0.5 µm)
SU-8 (2.5 µm)
Double-sticky mylar (50 µm)
Aluminum (0.5 µm)
Glass slide (2 mm)
Fig. 4: Process flow for fabrication of the device
4
Copyright © 2007 by ASME
A)
B) 7 mm
Heaters
Temperature sensor
Vapor ports
Thermal isolation trench
Liquid ports
Fig. 5: Images showing (A) backside and (B) front-side of the device after release from the wafer and before membrane installation
Surfboards
FABRICATION We generated a four-mask process to fabricate our device. Figure 4 describes the fabrication scheme. Figure 5 shows both sides of a 200 µm wide serpentine channel device after step D in Fig. 4 was completed. The central, operational region of the device has an area of 0.75 cm2. The central region consists of four heaters and sixteen distributed temperature sensors on the back and micro-channels on the front. The extensions on either side of the center are for fluidic interfacing. The thermal isolation trench is necessary to localize the heat in the central region and minimize conduction losses to the extensions. After the last step in Fig. 4 is complete, surfboards are attached and the device wire-bonded, resulting in the final product shown in Fig. 6. The current yield of devices was poor due to equipment issues and fabrication difficulties in steps D and E in Fig. 4. We expect the yield to improve in future runs.
Glass slide
Fig. 6: Completed, wire-bonded device before integration of fluidic connectors
EXPERIMENTAL ISOTHERMAL AIR-WATER TEST The heat exchanger was first tested using an isothermal, air-water flow to prove venting does occur and to quantify the venting efficiency of the device. Using an air-water mixture separates out any thermal interactions in the venting behavior and simplifies the measurement of venting efficiency. Venting efficiency is defined as the ratio of mass of air vented to the mass of air input to the device. Using the setup in Fig. 7, we measured the venting efficiency of a 400 µm wide (Dh = 250 µm), parallel-channel device with a PTFE membrane.
PTFE Membrane
First, we pumped DI water at 0.5 ml/min through the device. Once steady state conditions were obtained, we switched on the syringe-pump and pumped air at 1 ml/min. This results in the formation of intermittent slugs of air and water downstream of the mixing junction. Once steady state conditions were established for the mixed flow, we collected data from the pressure sensor and the four IR slot sensors. At the completion of data collection, we switched off the airflow and pumped only DI water through the device to remove any remaining air bubbles in the junction, tubing and device. We then repeated the process for the next set of water and air flowrates.
5
Copyright © 2007 by ASME
Syringe pump containing air
Upstream IR Slot Sensors
Vapor channel Liquid channel
Downstream IR Slot Sensors
0.5 um filter PC with DAQ
Pi Peristalsis pump
DI Water
Water output
Fig. 7: Experimental setup to quantify vapor venting efficiency of the device. IR slot sensors used to measure void fraction and interface velocities from which air volume and mass is calculated
We calculate the air volumes by measuring the interface velocities and void fraction of air over a given period using the two pairs of IR slot sensors. We can then calculate the air mass using the ideal gas law with room temperature properties. A summary of the results is provided in Table 1. Table 1: Summary of measured venting efficiency for various combinations of air and water flow rates.
AQ ml/min 0.4 0.6 0.8 1 2 2 2 2
LQ ml/min 0.5 1 1 1 0.5 0.7 1 2
Venting Eff. % 100.00 100.00 100.00 100.00 99.95 99.77 99.75 99.97
Under all flow conditions we tested, almost all the air vents through the membrane. Measurement of inlet air mass at air flow-rates of 2 ml/min is difficult using the current setup. When an air-slug vents in the device, the flow resistance in the fluidic circuit drops, causing a sudden acceleration in the flow. The calculations overestimate the mass of air entering the
device because the higher velocity is not constant over a single air-slug and the venting non-periodic due to the differing sizes of the air slugs. Downstream measurements are accurate as the flow is unaffected by the upstream venting and flow rate is constant. Using visual estimates of the average size of an air slug when pumped at 2 ml/min, we calculated a very conservative estimate of air mass venting efficiency of 95%. Venting isn’t perfect at the higher air flow-rate of 2 ml/min because the higher flow velocity during venting results in insufficient time for the larger air slug to completely vent. In summary, even with a very conservative estimate of venting efficiency at the higher air flow-rates the message is clear – the heat exchanger has the capacity to vent a large fraction of air in an air-water mixture. THERMAL, PHASE-CHANGE TEST Having verified the ability to vent air out of an air-water mixture still left the open question of whether the device can vent under operating conditions. To test this, we attached the same 400 µm wide parallel channel device tested above to the setup shown in Fig. 8. We connected the two downstream heaters on the device to 0-25 V DC power supplies and the sensors to the PC-based DAQ board via a signal-conditioning box. We also attached the 0-5 psig, Omega pressure sensor to the DAQ board for simultaneous measurements. We pumped house air at 5 psi (34 kPa) through the vapor side of the device to facilitate removal of condensed and leaked water. Liquid water can block the vapor channel and reduce venting.
6
Copyright © 2007 by ASME
Microscope 5 psi, pressurized air inlet
Vapor channel Liquid channel 0.5 um filter
Pi DI Water
PC with DAQ
Liquid output
Peristalsis pump
Signal Cond.
Vapor output
DC Power Supply
Fig. 8: Setup used to carry out thermal test with phase change in device temperature sensors. We believe the problem stems from processing and wire-bonding issues. We will investigate these issues in detail before the next round of fabrication.
0.5 ml/min 4.00 3 ml/min P drop/P ini
We pumped DI Water at 0.5 ml/min through the device for a few minutes until steady state conditions were established. We then slowly raised the power input to both heaters in equal steps, starting at 0 V. At each step rise, we allowed the device to operate for a few minutes to ensure steady state behavior before collecting data from the temperature and pressure sensors. We continued ramping the power until boiling occurred. Venting in the membrane along with vapor flow in the vapor-side outlet tubing indicated the onset of boiling. We raised the power a minimum of 30% beyond this point to obtain higher quality flows and to gather data beyond the onset of boiling. We then switched off the power and continued pumping water to cool the device back to room temperatures, before repeating the experiment at another flow rate. The experimental and simulated results for two water flow rates, 0.5 ml/min and 3 ml/min is shown in Fig. 9. Power input is non-dimensionalised with the theoretical power required to boil water, calculated as ρQflowCp∆T with a ∆T of 80 oC. Pressure drop is non-dimensionalised with the pressure drop for single-phase flow with no heating. The maximum total power output by the heaters is 4.5 W and 21.5 W for the 0.5 ml/min and 3 ml/min case respectively. The plot clearly shows a much lower pressure drop in the channel than expected from simulations. In summary, preliminary data collected from a 400 µm wide, parallel-channel device shows improved pressure-drop trends. We will quantify this improvement once a statistically significant amount of data is collected. Temperature data has not been included here due to poor performance of the
0.5 ml/min SIM 3 ml/min SIM
2.00
0.00 0
0.2 0.4 0.6 0.8
1
1.2 1.4 1.6 1.8
Total Actual Power / Theoretical Power
Fig. 9: Comparison of experimental data with simulations of two-phase flow in a non-venting device show much smaller pressure drops for both 0.5 ml/min and 3 ml/min liquid flow-rates
7
Copyright © 2007 by ASME
A)
B)
2V, 0.1 W
5V, 0.58 W
C)
13 V, 3.3 W
50 um
Non-ideal regions on membrane leads to droplet growth
Dry-out of membrane due to boiling and venting
Fig. 10: Images of same location on membrane taken using a microscope while increasing power to the heaters. A) and B) show leakage of water through non-ideal regions of membrane and C) shows vaporization of droplets and venting of vapor CHALLENGES AND SOLUTIONS LEAKAGE Leakage through the PTFE membrane, as seen in Fig. 10, increases with power input to the heaters until the membrane temperature is high enough to boil the wicked water away. This leakage necessitates the use of pumped air to remove the water droplets forming on the vapor side. Pumping air is an impractical final solution, as the ultimate goal is to operate the heat exchanger in a closed loop with a single pump. Wicking also complicates device modeling, which must now include phase change and two-phase flow within the membrane. To understand this leakage we consider the expression for leakage pressure through a pore [21], obtained from the Young-Laplace equation, and given as:
Leakage Pressure, ∆P =
4σ cos(180 − θ adv,max ) d
(5)
where σ is the surface tension of the fluid at some temperature, θadv,max the advancing angle of the fluid in the membrane material and d the diameter of the pore. Using characteristic values for PTFE and water at room temperature, with σ = 0.072 N/m, θadv,max = 129o and d = 0.22 µm, the leakage pressure is approximately 820 kPa. However, the liquid pressure in the channels is less than 30 kPa at the highest tested flow rate of 3 ml/min. σ, θadv,max and/or d must deviate from standard values, as listed below, resulting in a much lower leakage pressure. i)
σ is inversely proportional to the temperature and reduces to 0.06 N/m at 100 oC. The reduction in surface tension is insufficient to be the sole culprit of leakage, but does explain the increase in leakage with heater power.
ii) Reduction of θadv,max on PTFE due to surface or water contamination. Previous work documents the loss of hydrophobicity in PTFE [23] and Nylon [24] when immersed in saline solutions at various temperatures. iii) Defects in the membrane as well as the natural distribution of pore sizes in a mesh structure such as PTFE results in some fraction of pore-diameters larger than 0.22 µm. We believe that leakage occurs due to some convolution of these three non-idealities. Only a small fraction of the membrane area need be non-ideal to cause leakage. In Fig. 10.A, wetting initially occurs in distinct, darker, patches, over which larger droplets form as leakage increases as seen in Fig. 10.B. Using higher quality membranes with smaller pore size distribution and smaller pores can reduce leakage. Examples include PTFE membranes with smaller pores, porous glass and porous silicon. This solution requires design optimization, as the smaller pores reduce venting efficiency due to higher pressure-drop across the membrane. Porous glass and silicon have the added disadvantage of requiring low-energy coatings to ensure hydrophobicity. Using pure, filtered DI water is also important along with taking greater care to not contaminate or damage the membrane before device integration. OPTICAL ACCESS In the current design, the PTFE membrane is opaque, limiting the ability to study the flow regimes and venting mechanism in the liquid channels. This lack of information hinders both development of an appropriate two-phase model and optimization of liquid channel geometry to maximize venting efficiency. Using a quartz substrate, instead of silicon, allows optical access to the liquid channels through the backside of the device. Similarly, optical access from the front is possible using membranes such as porous polycarbonate and porous
8
Copyright © 2007 by ASME
glass. The fabrication of a quartz structure for the liquid channels is straightforward, but channel geometries are limited by isotropic etching. The low thermal conductivity of quartz would also make high heat flux studies challenging. Porous polycarbonate and glass require surface treatments to be hydrophobic and have only limited optical transparency. FUTURE WORK The first priority is to perfect the current fabrication process to maximize the yield and obtain devices with fully functional temperature sensors. Given a large number of fully functional devices, we can then carry out a detailed, statistically significant, fluid and thermal study. The study would quantify the performance improvement, provide a better understanding of the role of channel geometry and assist in the development of a two-phase, 2D, model of the flow in the device. We also expect the study to assist in the development of the next generation device that incorporates some of the proposed solutions to leaking and lack of optical access. Figure 11, illustrates one such example of a potential next generation device. Unlike the current stacked layer design, this structure provides excellent optical access to all parts of the structure. The design also allows fabrication of advanced porous structures such as CNT forests, ZnO nanotube forests and porous silicon within the device. CNT and ZnO nanostructures possess superhydrophobicity [25-28] with contact angles in excess of 140o, though CNT nanostructures require additional surface treatment for stable superhydrophobicity [27]. Nanotube forests can have various pore sizes, i.e. distance between adjacent nanotubes, based on growth conditions. Advantages of this design include: i) low leakage with the help of superhydrophobic nm-pore membranes, ii) ability to engineer membranes to specifications and iii) excellent optical access that would assist in detailed studies of venting. However, this design is not suitable for commercial development due to cost of membrane fabrication and loss of heat transfer area.
Porous membrane
Vapor channel
Glass cover
Liquid channel
Fig. 11: Schematic of proposed side-by-side structure with either a surface treated porous silicon, surface treated CNT forest or ZnO nanotube forest separating the two channels SUMMARY Two-phase, microchannel based heat exchangers have significant advantages for use in high-heat flux electronics
cooling and represents an important area of research and development. Commercial adoption has been slow due to the problems associated with vaporization of the working fluid. Motivated by patent work, simulations and previous work in gas venting, we designed and fabricated a vapor venting, microchannel heat exchanger. Vapor venting occurs through a porous, hydrophobic, membrane sandwiched between the liquid-carrying and vapor-carrying channels. Venting the vapor helps mitigate some of the problems associated with vapor entrapment in the microchannels. We carried out a proof-of-concept experiment on a 400 µm wide, parallel-channel device, using an air-water mixture under isothermal conditions. The air mass venting efficiency exceeded 95% for airflow rates of 2 ml/min and 100% for lower airflow rates. Venting efficiency drops at the higher air flow-rate due to high velocities coupled with large air slugs that result in incomplete venting. We also ran thermal, phasechange tests at flow rates of 0.5 ml/min and 3 ml/min and the results show a much lower pressure drop compared to flow simulations of a non-venting device under similar operating conditions. Challenges in the current design include liquid leakage through the membrane and lack of optical access to the liquid channels. To reduce leakage, membranes with smaller pore sizes and tighter pore size distributions must be used. Examples include PTFE with smaller pore sizes, porous glass and porous silicon. Care must be taken to use pure filtered DI water and not contaminate or damage the membranes is any manner during integration into the device. Optical access is achieved by fabricating the liquid channels on a quartz substrate instead of silicon or by using semi-transparent membranes such as polycarbonate or porous glass. Finally, we proposed a next generation device design where the microchannels and membrane are side-by-side instead of stacked. This design provides the best optical access, low leakage, and flexibility to engineer membrane characteristics. Potential membrane materials include porous silicon and superhydrophobic CNT or ZnO nanotube forests. In conclusion, this preliminary study proves that vapor venting in heat exchangers is both possible and provides measurable benefits. This success motivates the overall goal of this project: to develop a reliable, closed-loop, two-phase heat exchanger for use in cooling of next generation, high heat-flux generating electronics. ACKNOWLEDGEMENTS The author would like to acknowledge the support of the MARCO Interconnect Focus Center, one of five research centers funded under the Focus Center Research Program, a Semiconductor Research Corporation program. The author also appreciates support from the Semiconductor Research Corporation through Task 1445 as well as the Stanford Graduate Fellowships. Fabrication was performed in the Stanford Nanofabrication Facility, a part of the National
9
Copyright © 2007 by ASME
Nanotechnology Infrastructure Network, and supported by the National Science Foundation under Grant ECS-9731293. REFERENCES [1] ITRS Roadmap 2006 Update, http://public.itrs.net/, Aug. 2007. [2] Kandlikar, S. G., et al. (eds.), Handbook of Phase Change: Boiling and Condensation, Philadelphia: Taylor & Francis, 1999, Chp. 16, pp. 405-409. [3] Cancheevaram, J., et al., “Performance of Integrated thinfilm thermoelectrics in cooling “hot-spots” on microprocessors – Experimental setup and results,” Mat. Res. Soc. Symp. Proc., Vol. 793, 2004, pp. S8.18.1-S8.18.7. [4] Chen, C., Shakouri, A., “Heat transfer in nanostructures for solid-state energy conversion,” J. of Heat Transfer, Vol. 124, No. 2, 2002, pp. 242-252. [5] Phelan, P. E., “Current and future miniature refrigeration cooling technologies for high power microelectronics,” IEEE Transaction on Components and Packaging Technologies, Vol. 25, No. 3, 2002, pp. 356-365. [6] Wang, E. N., et al., “Micromachined jets for liquid impingement cooling of VLSI chips,” JMEMS, Vol. 13, No. 5, 2004, pp. 833-842. [7] Kercher, D. S., et al., “Microjet cooling devices for thermal management of electronics,” IEEE Transactions on Components and Packaging Technologies, Vol. 26, No. 2, 2003, 359-366. [8] Brunschwiler, T., et al., “Direct liquid jet impingement cooling with micron sized nozzle array and distributed return architecture,” The Tenth Intersociety Conference on Thermal and Thermomechanical Phenomena in Electronics Systems, ITHERM ‘06, San Diego, CA, 30 May - 2 June, 2006, pp. 196-203. [9] Qu, W., Mudawar, I., “Experimental and numerical study of pressure drop and heat transfer in a single-phase microchannel heat sink,” Int. J. of Heat and Mass Transfer, Vol. 45, No. 12, 2002, pp. 2549-2565. [10] Kandlikar, S., Grande, W., “Evaluation of single phase flow in microchannels for high heat flux chip cooling – thermohydraulic performance enhancement and fabrication technology,” Heat Transfer Engineering, Vol. 25, No. 8, 2004, pp. 5-16. [11] Garimella, S., Singhal, V., “Single phase flow and heat transport and pumping considerations in microchannel heat sinks,” Heat transfer engineering, Vol. 25, No. 1, 2004, pp. 1525. [12] Jiang, L., Wong, M., Zohar, Y., “Forced Convection Boiling in a Microchannel Heat Sink,” JMEMS, Vol. 10, No. 1, 2001, pp. 80-87. [13] Zhang, L. et al., “Measurements and Modeling of TwoPhase Flow in Microchannels with Nearly Constant Heat Flux Boundary Conditions,” JMEMS, Vol. 11, No. 1, 2002, pp. 1219.
[14] Qu, W., Mudawar, I., “Flow boiling heat transfer in twophase micro-channel heat sinks--I. Experimental investigation and assessment of correlation methods,” International Journal of Heat and Mass Transfer, Vol. 46, No. 15, July 2003, pp. 2755-2771. [15] Hestroni, G., et al., “A uniform temperature heat sink for cooling of electronic devices,” International Journal of Heat and Mass Transfer, Vol. 45, No. 16, July 2002, pp. 32753286. [16] Kennedy, J. E., et al., “The onset of flow instability in uniformly heated horizontal microchannels,” J. of Heat Transfer, Vol. 122, No. 1, 2000, pp. 118-125. [17] Wu, H. Y., Cheng, P., “Boiling instability in parallel silicon microchannels at different heat flux,” Int. J. of Heat and Mass Transfer, Vol. 47, No. 17-18, 2004, pp. 3631-3641. [18] Zhou, P., Goodson, K. E., Santiago, J., U.S. Patent 6,994,151 B2, 2006. [19] Brissette, F., Matlab Steam Table, 2005, Montreal, Canada. [20] Yang, Z., et al., “A prototype of ultrasonic microdegassing device for portable dialysis system,” Sensors and Actuators A, Vol. 95, 2002, pp. 274-280. [21] Meng, D.D. et al, “A Degassing plate with hydrophobic bubble capture and distributed venting for microfluidic devices,” J. of Micromech. Microeng., Vol. 16, 2006, pp. 419424. [22] Lamers, T. L., et al., “Application of a modified quality function deployment method for MEMS,” 2007 ASME IMECE, Seattle WA, November 11-15, 2007, Paper IMECE2007-42374. [23] Syed, H. Z., Hackam, R., “Effects of water salinity, electric stress and temperature on the hydrophobicity of polytetrafluoroethane”, IEEE Conference on electrical insulation and dielectric phenomena, 1998 – Annual Report, Atlanta GA, Oct. 25-28, 1998, pp. 100-103. [24] Tokoro, T., Hackam, R., “Loss and recovery of hydrophobicity, surface energies, diffusion coefficients and activation energy of nylon”, IEEE Transactions on dielectrics and electrical insulation, Vol. 6, No. 5, 1999, pp. 754-763. [25] Feng, L., et al., “Superhydrophobicity of Nanostructured Carbon Films in a wide range of pH values,” Angew. Chem. Int. Ed., Vol. 42, 2003, pp. 4217-4220. [26] Li, S., et al., “Superhydrophobicity of large area honeycomb like aligned carbon nanotubes,” J. Phys. Chem. B, Vol. 106, 2002, pp. 9274-9276. [27] Lau, K., et al., “Superhydrophobic Carbon Nanotube Forests,” Nano Letters, Vol. 3, No. 12, 2003, 1701-1705. [28] Feng, X., et al., “Reversible Super-Hydrophobicity to Super-Hydrophilicity transition of aligned ZnO nanorod films,” J. Am. Chem. Soc., Vol. 126, 2004, 62-63.
10
Copyright © 2007 by ASME
