thermal performance and key challenges for future...

Proceedings of IPACK2005 ASME InterPACK '05

July 17-22, San Francisco, California, USA

IPACK2005-73242 DRAFT

THERMAL PERFORMANCE AND KEY CHALLENGES FOR FUTURE CPU COOLING TECHNOLOGIES

Ioan Sauciuc, Ravi Prasher, Je-Young Chang, Hakan Erturk,

Gregory Chrysler, Chia-Pin Chiu, Ravi Mahajan

Intel Corporation 5000 W. Chandler Blvd., Chandler, AZ, USA, 85226

[email protected]

Proceedings of IPACK2005ASME InterPACK '05

July 17-22, San Francisco, California, USA

IPACK2005-73242

ABSTRACT Over the past few years, thermal design for cooling microprocessors has become increasingly challenging mainly because of an increase in both average power density and local power density, commonly referred to as “hot spots”. The current air cooling technologies present diminishing returns, thus it is strategically important for the microelectronics industry to establish the research and development focus for future non air-cooling technologies.

This paper presents the thermal performance capability for enabling and package based cooling technologies using a range of “reasonable” boundary conditions. In the enabling area a few key main building blocks are considered: air cooling, high conductivity materials, liquid cooling (single and two-phase), thermoelectric modules integrated with heat pipes/vapor chambers, refrigeration based devices and the thermal interface materials performance. For package based technologies we present only the microchannel building block (cold plate in contact with the back-side of the die). It will be shown that as the hot spot density factor increases, package based cooling technologies should be considered for more significant cooling improvements.

In addition to thermal performance, a summary of the key technical challenges are presented in the paper.

NOMENCLATURE A Area (m2) COP coefficient of thermal performance CFD Computational Fluid Dynamics cp specific heat (J/kg-K) DFjc Density factor from junction to case (1/cm2) Dh hydrodynamic diameter of the channel (m) harea equivalent heat transfer coefficient of the fins based on heat sink base area (W/ m2°C)

HX remote heat exchanger h average heat transfer coefficient (W/ m2°C) k thermal conductivity (W/m-K) m& mass flow rate (kg/s)

hDNu Nusselt number for the rectangular channel subject to

constant heat flux boundary condition Q power (W) Rjc Normalized Thermal Impedance from junction to case (ºC-cm2/W) T temperature (°C) Ta Ambient temperature local to the heat sink (ºC) Tc Package case temperature (ºC) TDP Thermal Design Power (W) TIM Thermal Interface Material impedance (°C cm2 /W) Tj Junction or die temperature (ºC) Tj, max Junction temperature at the hottest point of the die (ºC) Ts Heat sink temperature (ºC) Tx Temperature at any location on the thermal solution (ºC) ZT Thermoelectric Figure of Merit

Greek symbols

η fin efficiency Ψcs Thermal Resistance from case to sink (ºC/W) Ψjc Thermal Resistance from junction to case (ºC/W) Ψjs Thermal Resistance from junction to sink (ºC/W) Ψsa Thermal Resistance from sink to ambient (ºC/W) ψx,y thermal resistance (°C/W) between point x and y at

non-uniform heat flux θx,y thermal resistance (°C/W) between point x and y at

uniform heat flux Subscripts

1 Copyright © #### by ASME1 Copyright © 2005 by ASME

a ambient ad adiabatic temperature cp cold plate f fluid fin fins HEX heat exchanger hp Heat Pipe in liquid in j junction max maximum MCA Metrology Capability Analysis out liquid out s sink INTRODUCTION

Thermal design for cooling microprocessors has become a major challenge for the computer technologies as the silicon technologies continue to evolve rapidly. In addition, local power densities are more difficult to be managed, thus making thermal management more challenging. Many technologies [1-20] ranging from air cooling to refrigeration have been proposed for CPU cooling in the last years but their cooling significance was not well quantified in relationship to peak power density cooling, therefore there is a need to focus electronics cooling research and development on the most promising and cost effective technologies. As well known, the primary objective of thermal management in microelectronics applications is to keep the die junction temperature (Tj) below acceptable limits in order to ensure device performance and reliability. In the early stages of the thermal solution design a thermal engineer is more concerned with what-if scenarios such as whether it makes sense to use one technology against the other and make the necessary trade-offs. Typically the time available to thermal designers to perform design sensitivity studies and determine the technology capability envelope is small and complete complex modeling for various configurations and scenarios is not possible. For example the commercially available thermal tools cannot perform the necessary complete fluidic modeling of heat pipes, vapor chambers or two-phase forced convection systems especially at non-uniform heat flux. In addition, an extreme research focus on thermal interfaces has been observed in the last years, but these improvements present diminishing returns. A better return on investment can be obtained if the cooling industry will better understand the contribution of each component to the total thermal resistance.

One classification for the CPU cooling technologies depends on the main heat transfer function. Thus, two major building blocks are considered: the “Air Side Forced Convection” and the “Spreading” building blocks. The “Air Side Forced Convection” building block references to the possible improvements in the interaction between the ambient air and the cooling device. The “Spreading” building block refers to

2

the materials and devices present between the heat source and the ambient air.

Another classification for the CPU cooling technologies depends on the area of the application. Thus from the chip manufacturer point of view, the cooling technologies are classified as: package based (included by the package manufacturer) and enabling (usually added on top of the package by the Original Equipment Manufacturers – OEMs-). As most researchers will agree, the further away the cooling device is located from the hot spot the less contribution it has to the total thermal stack up has. This paper will show that as the hot spot density increases, package based cooling technologies should be considered to accommodate future CPU requirements.

The authors consider that the “Spreading” building block is the most promising for future improvements of the overall thermal cooling solutions. For the “Spreading” Building Block this paper classifies the main technologies in two main sub-categories:

(A) Technologies which increase the “effective” thermal conductivity for materials or devices compared to conventional designs. This category includes thermal interface materials, diamond, graphite materials and vapor chambers/heat pipes spreaders. Also technologies which help to spread the heat to a large remote heat exchanger (i.e. single or two-phase liquid cooling) are included here.

(B) Refrigeration Technologies (i.e. thermoelectric or vapor compression refrigeration).

In the enabling area a few key main technologies are considered and they will be discussed: liquid cooling (single and two-phase), thermoelectric integrated with heat pipes/vapor chambers, and refrigeration based devices. For the package based technologies, we will focus on the performance of the single phase micro channel based device, since this looks most promising to further improve cooling. This is because it eliminates several interfaces and provides the ability of having a large remote heat exchanger.

By using the density factor metric, this paper will show that at high hot spots densities the cooling technologies need to merge in a total single solution combining both the package and the enabling cooling solutions. At lower hot spot densities some enabling technologies may have better advantages. However, most of these technology options present some high risk items and key challenges which are not solved as of today. It is expected that the present work will significantly help the industry to focus their research in addressing these challenges and improve the return on investment.

Also, this paper addresses and summarizes the cooling capability of most promising technologies and determines their contribution to the total thermal resistance, all within a

2 Copyright © #### by ASME Copyright © 2005 by ASME

reasonable range of boundary condition encountered in CPU cooling.

DEFINITIONS AND METRICS

Thermal Resistance Definition

There are two major CPU architectures today No-Lidded- (Architecture I- Figure 1) and Lidded(Architecture II- Figure 2).

Ta (ambient)

Ts (sink)

Tj (junction) js

sa

ja

Figure 1 Architecture I - Package with Exposed Die

and Heat Sink

Figure 2 Architecture II - Lidded Package and Heat Sink

This paper will focus on the lidded package (Architecture II). The major metrics used in CPU cooling is the thermal resistance (Ψx,y) which is given by:

( )TDP

TT yxyx

−=Ψ , (1)

Where x and y can represent different points on the solution for Architecture II. In other words it is a measure of the resistance to heat flow between the two specified points. The most used resistances in electronic cooling are Ψj,a(junction-to-ambient) , Ψc,a (case-to-ambient) and Ψs,a (sink-to-ambient or heat sink resistance). A typical Architecture II test vehicle may have a 30 mm x 30mm heat spreader attached to a 10 mm x 10 mm die.

Density Factor Definition

In addition to the above metrics, Torresola et. al [21] have proposed the density factor (DF) definition as a significant metric in understanding the cooling capability for different thermal solutions used in CPU cooling. This metric can be used to quantify the impact of non-uniform die heating (so called “hot spots” ) on thermal management difficulty. The advantage

j

Ta

Ts (sink)

Tc (case)

Tj

c

sc

3

of using this metric is its ability to provide a better “apples-to-apples” comparison of the impact of different power maps and die sizes on thermal management. The junction-to-case Density Factor (DFjc) parameter can be used to quantify the impact of die size and die power maps on the thermal resistance. In this case, the junction temperature (Tj) and the case temperature (Tc) are the most convenient temperatures to use. This simple metric is defined as the ratio of thermal resistance to thermal impedance.

Thus the junction-to-case density factor is defined as :

jc

jcjc R

DFΨ

≡ (2)

Thermal impedance, Rjc, is defined as the thermal resistance normalized by die area when the die is uniformly powered.

COOLING BUILDING BLOCKS

Air Side Cooling Building Block

This section is focusing on determining the focus parameters for air cooling extensions. For this building block the most common metric used to compare the thermal performance of heat sinks is the sink to ambient thermal resistance as defined by Equation (2), Ψs,a (sink-to-ambient or heat sink resistance).

Resistance

Parameter Variation[deg.C/W]

Heat Source edge[mm](10,30 -0.08

Footprint edge [mm](50,100) -0.059

Base Conductivity [W/mK](180,390) -0.048

Fin Gap [mm](0.9,1.6) 0.036

Total Height [mm](40,80) -0.026

Velocity between plates [m/s](3,6) -0.016

Base Thickness [mm](4,10) -0.016

Fin conductivity [W/mK](180,390) -0.015

ResistanceParameter Variation

[deg.C/W]Heat Source edge[mm](10,30 -0.08

Footprint edge [mm](50,100) -0.059

Base Conductivity [W/mK](180,390) -0.048

Fin Gap [mm](0.9,1.6) 0.036

Total Height [mm](40,80) -0.026

Velocity between plates [m/s](3,6) -0.016

Base Thickness [mm](4,10) -0.016

Fin conductivity [W/mK](180,390) -0.015

Figure 3: Design of Experiment (DOE-2)- Parallel Plate Heat sink

Sauciuc et al. [10] presented a summary of statistical Design of Experiments (DOE) used to determine how to best minimize the sink to ambient resistance [Figure 3]. The parameters used are self explanatory with their range of variation listed between the brackets. The vertical pareto chart presented in Figure 3 shows the magnitude of the parameter influence. The values showing the resistance variation show the average thermal resistance reduction over the range of boundary conditions considered. It can be seen that the major focus areas to extend air cooling are: increase the heat source size (die size), increase


the heat sink volume with larger fans and use of copper base heat sinks. Other parameters (i.e. fin conductivity) have secondary importance.

(a) (b)

Figure 4 – (a) Angled Fin Dense Plates and (b) Heat Pipe Remote Cooling

Air Cooling Extension: Architecture II Test Vehicles

0.000

0.050

0.100

0.150

0.200

0.250

High Density Fins Angled Fin Dense PlatesHeatsink

Remote Heat PipeCooling

Sink

to A

mbi

ent R

esis

tanc

e (o C

/W)

Figure 5– Air Cooling Building Block - Technology Performance

Based on the design of experiments presented in Figure 3 a few reference samples have been fabricated. Figure 4 shows the photos of two technologies which extend air cooling: the “Angle Fin Dense Plates” and the conventional “Heat Pipe Remote Cooling”. Figure 5 shows a summary of the cooling performance for these technology options. It can be seen that simple technologies ranging from high fin density to heat pipes with large remote heat sinks could potentially decrease the sink to ambient thermal resistance to about 0.18 ºC/W for Architecture II Products. The constraints encountered in microprocessor cooling have been taken into consideration and more details are obtainable from Sauciuc et. al [10;11]. These studies showed that air cooling improvements present diminishing returns and other technologies should be the focus of the research and development.

High Thermal Conductivity Materials

Significant improvements in heat sink performance may be realized if high conductivity materials are used for the heat sink base. Unfortunately, the conventional materials with very high conductivity (i.e. diamond) are very expensive. Some anisotropic materials (i.e. graphite) have been proposed in the last years.

4

Thermal Conductivities and Heat Sink Resistance

0.15

0.2

0.25

0.3

200 400 600 800 1000 1200 1400 1600 1800 2000

In-plane Conductivities, kx and kz (W/m-K)

Hea

tsin

k R

esis

tanc

e (C

/W)

ky = Thermal conductivity normal to the base

ky = 50W/m-K

ky = 100W/m-K

ky = 360W/m-K

Copper

Thermal Conductivities and Heat Sink Resistance

0.15

0.2

0.25

0.3

200 400 600 800 1000 1200 1400 1600 1800 2000

In-plane Conductivities, kx and kz (W/m-K)

Hea

tsin

k R

esis

tanc

e (C

/W)

ky = Thermal conductivity normal to the base

ky = 50W/m-K

ky = 100W/m-K

ky = 360W/m-K

Copper

Figure 6 - Particular Study - Anisotropic Materials Effect on Sink to Ambient Resistance

Figure 6 presents the results of a study of anisotropic materials. The materials proposed for this investigation have a very high in-plane thermal conductivity (x and z directions), with a low thermal conductivity in the normal direction (y). Graphite is one example of a material with this property. For this particular study, the normal thermal conductivity (ky) was varied between 50 W/mK to 360 W/mK. The anisotropic heat sink (80mm x 80 mm x 60 mm) was compared with a heat sink made of copper for an Architecture II type test vehicle. Both heat sinks have same geometry and are using same fan and more details are presented in [2;10;11]. It can be seen that, for significant improvements, ky should as high as possible, even for high in-plane conductivities. To the best knowledge of the authors, current anisotropic materials have normal conductivities less than 100W/mK. Therefore, more research needs to be done in this area, to increase ky in values larger than copper’s conductivity. In addition, these materials tend to be cost prohibitive for the electronics cooling market. Therefore, it can be concluded that so far it is very difficult to find cost-effective materials to improve the spreading resistance of the heat sink base. This is a key challenge for this building block. However we did not determine if there are particular geometries where it make sense to use the current anisotropic materials.

Passive Phase Change (liquid/vapor) Spreaders

Another way to improve spreading is by using two-phase (liquid/vapor) devices. Despite many experimental and theoretical studies in heat pipes, vapor chambers and thermo siphons [10;13;21-37], it is arguable that the fundamental theory in heat pipes and vapor is well behind the experimental work, mainly in explaining heat transfer phenomena in the wick structure and in the evaporator section. Therefore most of the heat pipe/vapor chamber/thermosiphon designs are based on empirical experimental data and “know how” of each heat pipe/vapor chamber manufacturer.

Due to the large amount of work in boiling/phase change over the last decades we defined the reasonable phase change limit

kx

kz

kykx

kz

ky


based on experimental data only. Although some references [38] suggest the possibility of achieving heat transfer coefficients in excess of 100,000 W/m2K none of our testing could confirm this in case of Architecture II test vehicles. Other parameters such as heat carrying capacity and the burnout flux are not discussed here since the first objective is to reduce thermal resistance.

Figure 7. Schematic of the evaporator section for a thermo siphon

The most used metrics in phase change devices is the evaporator resistance as defined in equation (3) (which is the major contributor to the total resistance of the phase change device). The location of the sink and adiabatic temperatures are shown in Figure 7.

(3)

(a) (b)

Figure 8. Photos of the (a) Conventional Uniformed Porous Layer (UPL) (b) Sintered structures SEM Photos of the Micro Porous Layer ( MPL) boiling structures

Several types of optimized boiling structures were tested using Architecture II type test vehicles. Figure 8(a) shows the conventional sintered boiling surfaces and figure 8(b) shows more developed MPL surfaces [44]. The comparison data is presented in figure 9 which also include other types of boiling surfaces: sintered (A&B types) and wire mesh (water was used as working fluid for all type of experiments). The ambient

TDPads

EvapTTθ −

≈

temperature for all tests was room temperature of about 23-25 °C nad this may translate in operating vapor temperatures of about 35-45 °C .

0

0.02

0.04

0.06

0.08

0.1

0.12

A B

Wire Mesh UPL MPL

θ EV

AP

[o C/W

]

Figure 9. Thermosyphon test data with different boiling/phase change surfaces using Architecture II with 120W

The evaporator resistance was determined at a horizontal orientation for the boiling surface. It can be seen that the evaporator resistance for most of the surfaces was about 0.1 ºC/W using Architecture II Test Vehicles and we consider this to be the practical performance extension for the phase change devices. Also, we would like to emphasize that experiments must be done in order to establish the evaporator performance for each particular application and other parameters such as heat carrying capacity and the burnout flux should also be considered.

TEC Based Cooling Devices (enabling)

Another technology to be discussed is a cooling apparatus with no moving parts using conventional TEC combined with conventional spreading devices (Vapor Chamber, Thermosiphon or Heat Pipe). Figure 10 shows the integration of TEC with other components.

5 Copyright © #### by ASME 5 Copyright © 2005 by ASME

Spreading Device

TECs

Heat Sink

TIM B

TIM A

Figure 10. TEC Based Concept Thermal Solution

The spreading device conducts the heat to the cold side of a conventional thermoelectric device. Several TEC modules are used to introduce a negative (effective) resistance into the chain of thermal resistances, thus decreasing the operating temperature of the spreading device.

One measure of performance of a TEC is the COP (coefficient of performance). It is the ratio of two heat flow terms as give by equation (4):

TEC

CPU

PQ

COP = (4)

0

0.02

0.04

0.06

0.08

0.1

0.12

0.14

0.16

0.18

0.2

2 2.5 3 3.5 4 4.5 5

COP

θsa [

o C/W

]

TDP=100WTDP=120WTDP=140WTDP=160WNo-TEC20 Sample Mean

Figure 11. TEC Based devices cooling capability using heat pipes for lidded package using a heat sink of 80x120x27 mm3 with 0.3 mm thick fins spaced 1.3 mm apart; Tamb=23oC, θhp≈0.1oC/W, thermal grease of RTIM=0.4 oCcm2/W. 20 sample data, tested at 120W with COP=3.

Figure 11 presents the model and test data showing the thermal capability for a remote 40 mm tall heat sink using TEC based thermal solution using heat pipes for spreading. The analytical model is based on the estimation of individual component thermal performance (i.e. spreading, fins resistance, etc.) and than integrated into one dimensional analytical model. These details have been already discussed by Sauciuc et. al [12;13]. The horizontal line at just over 0.18 oC/W thermal resistance without the TECs represents the conventional parallel plate air cooled heat sink (see Air Side Cooling Building Block). It can be seen from this figure that significant thermal improvements over conventional air cooling can be obtained even at 160 W and COP=3. For the case studied here, the thermal improvement (reduction in effective heat sink resistance) for this solution range from about 0.06 ºC/W to 0.11 ºC/W for IHS type processors. This graph also shows a single data point, represented by the yellow diamond, which is the measured mean of 20 samples. This measurement reconfirms the TEC capability and model prediction. The key challenges for this building block are cost reductions, improved reliability and improved thermoelectric materials.

Thermal Interface Building (Block)

The thermal interface material (TIM) performance and the bond line thickness (BLT) is usually measured in an ASTM-based [40] material tester [Figure 12]. Other researchers [41;42;43] have already presented different thermal interface materials but how significant are the overall junction to ambient improvements was not addressed.

Figure 12: ASTM-based Material Tester

The above method is used by many TIM manufacturers to estimate the performance of TIM and to develop new formulations of TIMs. Some advantages of the material tester include the controlled co-planarity between surface of interest,

Temperature

Dis

tanc

e

θTIM

Copper block(Cold side)

Copper block(Hot side)

TIM

Surface roughness

(exaggerated)

Bulk TIM resistance (for BLT)

Thermocouple locations

Rc = θTIM - (Bulk TIM resistance)

Power input, P

Tcold,int

Thot,int


the ability to measure the TIM thickness, the ability to apply an accurate pressure on the TIM, and a controlled uniform temperature at the TIM-surface interface.

Table 1: TIM performance data from Material Tester measured at 0.621 MPa loading on TIM

Material θTIM (oC-cm2/W) BLT (µm)

Grease 1 0.02 7.6

Grease 2 0.06 17.8

Grease 3 0.16 7.6

Gel 1 0.05 7.6

*PCM 1 0.08 27.9

*PCM 2 0.11 48.3

*PCM 3 0.11 7.6

*PCM 4 0.18 101.6

Foil 1 0.16 50.8

Foil 2 0.11 10.2

Solder 1 0.11 127

Solder 2 0.18 127 *PCM: Phase Change Material Table 1 shows the TIM performance and BLT measured in an ASTM-based material tester. The test data range from 0.02 oC-cm2/W to 0.2 oC-cm2/W. This table shows that the thermal performance of TIMs can be pushed to very low levels so any more future improvements will bring diminishing returns. One factor to focus for the industry: the thermal interfaces can degrade with usage. Degradation in TIM performance depends on usage temperature, the time of usage, mechanical loading, and material properties. Since all factors are not well understood, TIM degradation should be characterized using empirical methods.

Micro channel Cooling Building Block (package)

Various innovative designs on the use of micro channel heat sinks for CPU cooling applications have appeared during the past decade, as dramatic improvement in the techniques of micro fabrication has made the technology more feasible. Current priority is to develop a single-phase cooling system, as the technology is more mature in the industry. For the package based technologies (Architecture I- Figure 1) we propose to have a micro channel cold plate in contact with the back-side of the die. The cold plate design includes a set of parallel micro channels with a width from tens to hundreds of microns. These micro channels provide extended surface area for heat transfer to the coolant. Due to their extremely small hydrodynamic diameter the cooling performance is very good. The micro channel packaged heat sink was is based on the concept that the micro channels provide high surface area to volume ratios and

7

also serve to enhance the heat transfer coefficient significantly in the laminar flow regime where heat transfer coefficient is inversely proportional to hydrodynamic diameter. As a rule of thumb, all channels with characteristic dimensions between 1 µm and 1 mm are typically called micro channels. Figure 12 shows the micro channel cold plate combined with a pump and a remote heat exchanger.

Cooling fan

Pump

Processor Cold plate

Water flow

Heat out

Microchannelcold plate ToutTin

Tj,max

Tamb

HXToutTin

Figure 12. Schematic of the Micro-Channel Cooling System (Architecture I)

The thermal resistance of a micro-channel cooling system is defined in equation 5 below. The model prediction of the thermal capability for the micro channel cold plate is plotted in Figure 13 for different selected fluid viscosities. It can be seen that within reasonable pressure drop for the pump, the junction to inlet resistance can be decreased to about 0.1 ºC/W which is considered a practical limit for a 10 mm x 10 mm test die (commercial available CFD model for a single phase fluid).

TDPTT

R fluid,inmaxj,ij

−=− nlet (5)


0.00

0.05

0.10

0.15

0.20

0.25

0.30

0.35

0 50 100 150 200 250 300 350 400 450 500

Presure drop (kPa)

R j-in

let (

o C/W

)

µ= 0 .5 3 7

µ=1 .5

µ =2 .2 9

Unreasonable bearing forces/ Pump Size

cPcP

cP

Figure 13. Extension of the micro channel resistance vs. pressure drop

The key challenge for micro channel development is to develop a pump which uses the working fluid as lubricant for the bearing. Most likely water can not be used as a coolant due to freezing issues whereas traditional antifreeze coolants such as Propylene Glycol (50%) have very high viscosity and low thermal conductivity (50% of water). Therefore, there is also a need to develop alternate antifreeze coolants.

Single Phase/ Two Phase Active (enabling)

Except for TEC based devices all future enabling cooling technologies may require some fluid moving device (i.e pump or compressors). Most of the current development for active devices is focused on single phase liquid cooling, two-phase liquid cooling and vapor compression refrigeration.

The definition of the cold plate resistance is similar to the microchannel cold plate resistance with the only change is substituting the Tj with Ts, for simplicity we considered:

TDPTT ins

cp−

=θ (6)

The test data for optimized single phase liquid cooling (LC) using Propylene Glycol as working fluid and a two-phase forced convection (2PFCC) using some refrigerant is presented in Figure 14. It can be seen that these particular optimized single phase has a smaller resistance at lower TDP but the two-phase devices are becoming better at larger TDPs. The fact that the two-phase resistance is lower at larger TDP is explained by the fact that boiling improves with the increase in the heat flux.

0.06

0.07

0.08

0.09

0.1

0.11

70 80 90 100 110 120 130

Q [W]

θC

P [o C/

W]

2PFCC

LC

Figure 14. Typical Cold Plate Resistance Variation for Architecture II at 120W heat load

Key Challenges for the micro channel is to develop a long life bearing.

Refrigeration Building Block (enabling)

A compact miniature compressor system that uses R-134a as a refrigerant, may be the ultimate method for cooling high power density electronic components. Brown at al. [20] shows the performance of such a refrigeration system developed together with Intel Corporation [Figure 15].

Compressor

Cold Plate & Resistive Heater

Condenser Airflow

Figure15 Proposed refrigeration system (Courtesy of Kryotech)

Figure 15 shows the proposed refrigeration system. The system has the same basic components as a standard vapor-compression system; compressor, evaporator, condenser and expansion valve, but are using a new rotary compressor measuring 3.5 inches high and 2.5 inches in diameter. The entire refrigeration system measures 16 inches long, 7 inches wide and 3.5 inches high.


Figure 16. Refrigeration Performance

Figure 16 shows COP and θsa data as the condenser cooling fan speed (air flow rate) is varied. The data shows a gradual decrease in θsa with increased condenser fan speed on the order of 11%. The cold plate temperature (evaporator) decreases slightly and then holds a relatively constant value around 11oC. The system COP is shown to decrease slightly with increasing condenser fan speed. This is a result of increasing compressor operating temperature for this testing scenario.

0

20

40

60

80

100

120

140

160

180

200

20 25 30 35 40 45

Evaporating temperature, Tevap (°C)

CO

P ( -

)

Figure17 COP versus evaporating temperature for R-134a at condensing temperature of 45 °C and constant cooling capacity of 100 W

Trutassanawin et al. [45] presented a detailed study to investigate the thermal resistance capability at high evaporating temperatures. As expected, the COP of the system increases with an increase of the evaporating temperature as shown in Figure 17 and approaches infinity at the limit of an evaporation temperature of 45 ºC, in which case the evaporation temperature is equal to the condensation temperature. At this , no mechanical power is needed to run the system and the system would operate as a heat pipe. This shows that there may be very significant benefits for having refrigeration systems at high evaporator temperatures. The key challenges for this

0

2

4

6

8

10

12

14

0.01 0.011 0.012 0.013 0.014 0.015 0.016

Condenser Fan Air Flow Rate (kg/s)

CO

P &

Eva

pora

tor

Tem

pera

ture

(o C)

-0.16

-0.14

-0.12

-0.1

-0.08

-0.06

-0.04

-0.02

0

sa (o C

/W)

COPCold Plate TempTheta-sa

building block are: cost, bearing reliability and operation at all orientations. Although refrigeration may look promising as an option, the maturity of development on low compressor size, operation at all orientations, water condensation and cost are major high risks items. This leaves micro channel based technologies as one of the future major focus packaged based technology.

Summary - Future Technology Focus and Challenges

0%

50%

100%

150%

200%

250%

DF=1.5 DF=3 DF=4.5 DF=6 DF=7.5Density Factor [1/cm2]

Coo

ling

Lim

it Im

prov

emen

ts (

%).

Enabled Liquid Cooling

Enabled TEC Based Devices

Enabled Refrigeration

MicroChannel (SP) Die Level

Figure 18. Cooling Limit Improvements (reference conventional air-cooling)

Based on the DFjc defined at the beginning of this paper and using the data from each building block the proposed technologies are compared against conventional air cooling (Figure 18).

−

≡

coolingair

Ψ

coolingair

Ψconsideredtechnology

Ψ

sImprovmentLimitCooling

ja

jaja (7)

It can be seen that at low DFs the enabling technologies provide significant improvements, however as the DF increases the difference between technologies is not so significant with the microchannel equaling the enabling refrigeration (theta_sa for refrigeration is Ψsa = 0 ºC/W in order to avoid condensation). There are key high risk concerns associated with all of the future technologies since most of them are using rotation devices which need to ensure high package reliability. The low size of the devices imposes significant larger forces on bearings reducing the lifetime of the pumps.


Bearing/PumpPackage Reliability

& Service

MEDIUM-HIGH

LOW

HIGH

Yes

High

Single Phase MicroChannel

Compressor Size

CondensationOperation at all orientations

Power Cycling Effect on Reliability

Bearing/Pump Reliability and

Service

N/AKey Concerns

HIGH

LOW-MEDIUM

MEDIUM-HIGH

Yes

High


LOW

LOW

LOW

No

Low

Air Cooling

NoYesForm factor Changes needed

LOWMEDIUM-HIGHReliability (risk level)

MediumLow-MediumCooling Performance

MEDIUM

LOW


MEDIUM

MEDIUM

Enabled TEC Based

Devices

Cost

Electrical Power Consumption

Technology

Bearing/PumpPackage Reliability

& Service

MEDIUM-HIGH

LOW

HIGH

Yes

High

Single Phase MicroChannel

Compressor Size

CondensationOperation at all orientations

Power Cycling Effect on Reliability

Bearing/Pump Reliability and

Service

N/AKey Concerns

HIGH

LOW-MEDIUM

MEDIUM-HIGH

Yes

High


LOW

LOW

LOW

No

Low

Air Cooling

NoYesForm factor Changes needed

LOWMEDIUM-HIGHReliability (risk level)

MediumLow-MediumCooling Performance

MEDIUM

LOW


MEDIUM

MEDIUM

Enabled TEC Based

Devices

Cost

Electrical Power Consumption

Technology

Table 2. Qualitative comparison for different future CPU cooling technologies

Table 2 presents a comparison of different technologies proposed and the key concerns associated with it. It can be seen that most of the promising technologies are using some active (rotating device) thus the bearing reliability may be the most significant challenge for all of these technologies. In addition, refrigeration does not work at all orientations which makes it not applicable for all market segments (i.e. mobility). The designer has the challenge of integration of the proposed concepts and to make the necessary cost tradeoffs using Figure 18 and Table 1.

Conclusions

This paper addresses uses a “building block” approach to quantify thermal performance and determine the key challenges for future enabling and package based cooling technologies using a range of “reasonable” boundary conditions. Model and test data shows that the thermal resistances of the cold plates for both liquid cooling and phase-change device can drop to approximately 0.08-0.1 oC/W. For the TEC based passive devices values lower than Ψsa =0.1 oC/W for powers up to 120-130 Watts at a high coefficient of performance (COP=3) can be obtained. The enabled vapor compression refrigeration cooling test data shows that the thermal resistances of the entire system can be negative and can drop to approximately Ψsa =– 0.1 oC/W, but, in order to avoid condensation, for this study we only consider that Ψsa down to 0 oC/W. Using reasonable boundary conditions for the micro channel cooling, we have shown more than 100% improvements at low non-uniformity heating and more than 50% at high non-uniformity heating compared to conventional air cooling. For particular sets of reasonable boundary conditions and hot spot density factor, the cooling capability of different technologies may significantly increase if the industry and academia will further focus and solve the key challenges presented in the paper.

REFERENCES

[1] Chu, R.C., Simons, R.E., Ellsworth, M.J., Schmidt, R.R., Cozzolino, V., 2004, “Review of Cooling Technologies for

10

Computer Products,” Device and Materials Reliability, IEEE Transactions, Vol.4, Issue: 4, pp. 568 - 585.

[2] Sauciuc, I., Chang, J., Chau, D. G. Chrysler, R. Prasher, H. Erturk, 2004, “Novel Thermal Technologies for CPU Cooling Applications”, IATTJ Journal.

[3] M.Saini, R.L Webb, 2003, “Heat rejection limits of air cooled plane fin heat sinks for computer cooling Components and Packaging Technologies,” IEEE Transactions on Packaging Technologies, Volume: 26 , Issue: 1.

[4] Kitajo, Y., Takeda, Y., Kurokawa, Y., Ohta, T., Mizunashi, H., “Development of a high performance air cooled heat sink for multichip modules,” Semiconductor Thermal Measurement and Management Symposium, Feb. 1992

[5] Tasaka, M., Shinohara, K., Hayashi, C., Ka-shima, S., Koyama, K., 1998, “Cooling performance of heat sinks with corru-gated-fins,” ITHERM '98.

[6] Shaukatullah, H., 1999, “Bibliography of liquid cooled heat sinks for thermal enhancement of electronic packages,” Semiconductor Thermal Measurement and Management Symposium.

[7] Upadhya, G.., Zhou, Peng., Hom, J., Goodson, K., Munch, M., 2004, ”Electro-kinetic micro channel cooling system for servers,” ITHERM, vol.1, pp.:367 – 371.

[8] Vogel, M.R., 1994, “Liquid cooling performance for a 3-dimensional multichip module and miniature heat sink,” Proceedings of 10th IEEE/CPMT, 1-3 Feb. 1994.

[9] Bierschenk, J., Johnson, D., 2004, “Extending the limits of air cooling with ther-moelectrically enhanced heat sinks,” D.;Thermal and Thermomechanical Phenomena in Electronic Systems, 2004. ITHERM '04. The Ninth Intersociety Conference on , 1-4 June 2004

[10] Sauciuc, I., Chrysler, G., Mahajan, R., Szleper, M., 2003, “Air-cooling Extension – Performance Limits for Processor Cooling Applications”- Proceedings of the 19th SEMI-THERM, San Jose, California, USA, 2003.

[11] Sauciuc, I., Chrysler, G., Mahajan, R., Prasher, R., 2002, “Spreading in Heat Sink Base: Phase Change Systems or Solid Metal??”, IEEE Special Transactions on Components and Packaging Technologies, December 2002, Volume 25, Number 4, page 621-629.

[12] Sauciuc, I, Erturk, H., Chrysler, G., Mahajan, R., 2004, “Thermoelectric and Phase Change Integration Building Block with Application to CPU cooling”, Intel Technology Symposium, September 2004, San Franscisco.

[13] Sauciuc, I., Ikeda, M., Erturk, H., Kimura, Y., Chrysler, G., Noda, H., Mahajan, R., Nakamura, T., 2004, “Integration of Thermoelectric and Phase Change (liquid/vapor) devices with application to CPU cooling”, IMAPS-ATW, October 2004, Paolo Alto.

[14] Xie, H., Aghazadeh, M., Lui, W and Haley, K., 1996, “Thermal Solution to Pentium Processors in TCP in Notebooks and Sub-Notebooks”, IEEE, Transaction on Components and Packaging and Manufacturing Technology, Part A, Vol. 19, No. 1, Mar. 1996, pp 201-210.


[15] Maveety, J.G.; Jung, H.H., 2002, “Heat transfer from square pin fin heat sinks using air impingement cooling Components and Packaging Technologies,” IEEE Transactions , Volume: 25 , Issue: 3 , Sept. 2002

[16] Zheng, W., Xu, G., 1996, “Influence of air heat sink geometry on the performance of thermoelectric cooling modules Thermoelectric,” Fifteenth International Conference on , 26-29 March 1996

[17] Kim, S.J., 2004, “Analysis Methods for Thermal Optimization of Micro channel Heat Sink,” Heat Transfer Engineering, Vol. 25(1), pp.37-49, 2004.

[18] Kleiner, M.B.; Kuhn, S.A., Haberger, K., 1995, “High performance forced air cooling scheme employing micro channel heat exchangers,” Components Packaging, and Manufacturing Technology, Part A, IEEE Transactions, Volume: 18 , Issue: 4 , Dec. 1995.

[19] Radmehr, A.; Patankar, S.V., 2004, “A flow network analysis of a liquid cooling system that incorporates micro channel heat sinks”, Thermal and Thermomechanical Phenomena in Electronic Systems, 2004 ITHERM, The Ninth Intersociety Conference on , 1-4 June 2004 Pages:714 - 721 Vol.1 1

[20] Brown, M.F., Chrysler, G.M., Maveety, J.G., and Sanchez, E.A., 2001, "Thermal Management for Electronics Cooling using a Miniature Compressor," M-CALC III, Third Workshop on Military & Commercial Applications for Low-Cost Cryocoolers, San Diego, CA.

[21] Torresola, J., Chiu, C.P., Chrysler, G., Grannes, D., Mahajan, R., Prasher, R., Watwe, A., “Density Factor Approach to Representing Impact of Die Power Maps on Thermal Management” IEEE Transactions on Advanced Packaging, (in press).

[22] Antonenko et al., 1990, " Mechanism of Vapour Phase Propagation at boiling inside the Heat Pipe Wick", Proceedings of 7th International Heat Pipe Conference , Minsk.

[23] Chen X. Q., Zhang, Z., Ma, T., 1992, "Natural Convection Boiling Heat Transfer in a Two- Phase Closed Thermosyphon", Proceedings of 8th International Heat Pipe Conference , Beijing .

[24] Cheng X., Zhang Z., Ma T., 1992, “Heat transfer Correlation of Evaporator Section in a Two-Phase Thermosyphon”, Proceedings of 8th International Heat Pipes Conference, Beijing.

[25] Collier, J.,” Convective Boiling and Condensation” , Mc Graw Hill , London, 1972.

[26] Dobran , F., 1985, " Steady State Characteristics and Stability Thresholds of a Closed Two - Phase Thermosyphon", Int. J. Heat Mass Transfer, vol. 28, pp. 949- 957.

[27] Dunn, P. D., and Reay , D. A., " Heat Pipes", 4th ed., Pergamon Press, Oxford, United Kingdom, 1994.

[28] Faghri, A., "Heat pipe science and technology" , Taylor and Francis, Washington, US, 1995.

11

[29] Grob, U., 1990, "Pool boiling heat transfer inside of a two-phase closed thermosiphon", Proceedings, 9th. Int. Heat Transfer Conf., Jerusalem.

[30] Imura, H., Yoshida, M., and Ippohshi, S., 1992, "The visual observation of the flow patterns in two-phase single and double-tube thermosiphons", Proceedings of 8th International Heat Pipe Conference, Beijing .

[31] Jialun , H., Tongze, M., 1992, " Investigation of Boiling Liquid Pool Height of Two- Phase Closed Thermosyphon", Proceedings of the 8th International Heat Pipe Conference, Beijing.

[32] Kaminaga, F., et al., 1992, " Study on Boiling Heat transfer Correlation in a Closed Two- Phase Thermosyphon ", Proceedings of the 8th International Heat Pipe Conference, Beijing.

[33] Kovalev , S.A., Solovev, S.L., Ovodkova, 1990, " Evaporation Heat Transfer theory in Heat Pipes and thermosiphons" , Proceedings of 7th International Heat Pipe Conference , Minsk.

[34] Masaka , M., et al., 1990, " Boiling heat transfer of a liquid film formed in the evaporator of a two-phase closed thermosiphon" , Heat transfer- Japanese Research, vol. 19 n5, pp. 416-428.

[35] Niro, A., and Beretta, G.P., 1983, "Boiling regimes in a closed thermosiphon", Int. J. Heat and Mass Transfer, Vol. 33, No. 10.

[36] Nukiyama, S., 1934, “Maximum and minimum values of heat transmitted from metal to boiling water under atmospheric pressure”, J.Soc. Mech. Eng. Japan, Vol.37, p. 367.

[37] Sauciuc I., Akbarzadeh A., Johnson P. W., 1995, “Characteristics of two-phase closed thermosiphons for medium temperature heat recovery applications ”, Heat Recovery Systems & CHP, Vol. 15, No. 7, pp. 631-640.

[38] Gross, U., 1992 , "Falling film evaporation inside a closed thermosiphon", Proceedings of 8th International Heat Pipe Conference , Beijing.

[39] Zuo, Z.J. et al., 2002, “An Integrated thermal architecture for thermal management”, THERMES Proceedings, pp. 325.

[40] “Standard Test Method for Thermal Transmission Properties of Thin Thermally Conductive Solid Electrical Insulation Materials,” ASTM D5470-93.

[41] Russell, A., Chiu, C.P, 1998, “A Testing Apparatus for Thermal Interface Materials,” Proceedings of 1998 International Symposium on Microelectronics, Page 1036 – 1041.

[42] Solbrekken, G.L., Chiu, C.P., Byers, B., Reichebbacher, D., 2000, “The development of a tool to predict package level thermal interface material performance,” 7th Intersociety Conference on Thermal and Thermomechanical Phenomena in Electronic Systems, ITHERM 2000, Vol. 1, Page 48 – 54.

[43] Chiu, C.P, Solbrekken, G.L., Young, T.M., 2000, “Thermal modeling and experimental validation of thermal interface


performance between non-flat surfaces,” 7th Intersociety Conference on Thermal and Thermomechanical Phenomena in Electronic Systems, ITHERM 2000, Vol. 1, Page 52 – 62.

[44] Liter, S. and Kaviany, M., 2001, “Pool Boiling CHF Enhancements by modulated porous layer: theory and

12

experiment”, International Journal in Heat and Mass Transfer, vol. 44, pp. 4287-4311.

[45] Trutassanawin S., Groll E., Garimella, S., “Thermodynamic Analysis of Miniature-Scale Vapor Compression System at High Evaporation Temperatures”, Internal Report, December 2003


thermal performance and key challenges for future...

Documents