Thermal Management of Electronics

Justin A. Weibel Research Associate Professor, School of Mechanical Engineering Associate Director, Cooling Technologies Research Center

Power and Energy, Onshore and Afloat Course Number: CNIT 58100 (CRN: 18530) / IT 58100 (CRN:18534) Nov 20 (3:30-4:45 PM)

Justin A Weibel: [email protected] https://engineering.purdue.edu/CTRC Lecture Topics

Background & Motivation . Thermal Management Landscape . Computing Trends and Challenges . Energy Implications

Thermal Management Basics . Thermal Packaging Architecture . Review of Heat Transport Resistances

Advanced Thermal Management Strategies . Vapor Chamber Heat Spreaders . Microchannel Heat Sinks

Further Textbook Reading

. Heat Transfer Fundamentals . Incropera and DeWitt, Fundamentals of Heat and Mass Transfer . A.F. Mills, Basic Heat and Mass Transfer, 1995 . Electronics Cooling . L.-T. Yeh, Thermal Management of Microelectronic Equipment, 2016 . Heat Pipes: . A. Faghri, Heat Pipe Science and Technology, 1995 . G.P. Peterson, An Introduction to Heat Pipes, 1994 . D. Reay and P. Kew, Heat Pipes, 5th Edition, 2005 . Microchannels . Heat Transfer and Fluid Flow in Minichannels and Microchannels, 2006 . Microchannel Heat Sinks for Electronics Cooling, 2013

Thermal Management Landscape

• Information and communication technology – High-performance computing – Data centers – Telecommunications

• Electrified transportation systems – Hybrid/electric powertrains – Aerospace – Watercraft

• Renewable energy – Smart grid – Energy conversion processes – Photovoltaics – Batteries Computing Trends

National Research Council Study by the Committee on Sustaining Growth in Computing Performance

“Thermal-power challenges and increasingly expensive energy demands pose threats to the historical rate of increase in processor performance.” 2011 NRC/CSTB Study: “The Future of Computing Performance” “…growth in the performance of computer systems will become limited by their power and thermal requirements within the next decade” Heat Flux Challenge Other Challenges

. Multi-core processing: more efficient and lower heat fluxes…. but causes local hot spots that must be accommodated . Thermal management in extremely space constrained devices . System-level volumetric heat generation density . Thermo-mechanical and Electro-thermal co-design constraints . Cost, consumer perception & human factors, market trends, etc.

Challenge of Multiple Scales

Silicon Die (10-2) Heat Sink (10-1)

Package (10-2)

Facility (102)

Transistor (10-9 m)

System (100) Natural Resources Defense Council (NRDC) Data Center Efficiency Assessment, August 2014

© Suresh Garimella IBM Z13 Server

https://apps.kaonadn.net/4882011/product.html#10/1066;C178 Thermal Management Architectures

Primary Heat Transfer Path

. I – Heat Sink . V – Die . II – Thermal Interface Material . VI – Underfill . III – Integrated Heat Spreader . IV – Thermal Interface Material . VII – Package Substrate

Mahajan, “Thermal Interface Materials: A Brief Review of Design Characteristics and Materials”, Electronics Cooling, Feb 2004 Simple Resistance Model

Primary Heat Transfer Path Ta [K] (air temperature)

Rtot [K/W] q [W] (thermal (heat resistance) flow)

Tj [K] Total temperature rise of the (silicon (TTja− ) chip described by the junction- q = temperature) to-ambient thermal resistance Rtot Interfacial Resistances

Interfaces pose large resistances due Primary Heat Transfer Path to microscale surface asperities

1 (∆T ) Rc = = hqc

Dry Contact Conductance

L.-T. Yeh, Thermal Management of Microelectronic Equipment, 2016 Thermal Interface Materials (TIM)

Materials introduced into the air gap Primary Heat Transfer Path to reduce the thermal resistance

RTIM= R bulk ++ RR c12 c

See reference below for calculation of individual resistances

Mahajan, “Thermal Interface Materials: A Brief Review of Design Characteristics and Materials”, Electronics Cooling, Feb 2004 Thermal Interface Materials (TIM)

TIM Type General Characteristics Advantages Disadvantages

Greases Typically silicone based •High bulk thermal •Susceptible to grease pump-out and phase matrix loaded with conductivity separation particles (typically AlN •Thin BLT with minimal attach pressure •Considered messy in a manufacturing or ZnO) to enhance •Low viscosity enables matrix material to environment due to a tendency to migrate thermal conductivity easily fill surface crevices •TIM curing not required •TIM delamination is not a concern Phase Polyolefin, epoxy, low •Higher viscosity leads to increased stability •Lower thermal conductivity than Change molecular weight and hence less susceptible to pumpout greases Materials polyesters, acrylics •Application and handling is easier •Surface resistance can be greater than greases. typically with BN or compared to greases Can be reduced by thermal pre-treatment

Al 2O3fillers •No cure required •Requires attach pressure to increase thermal •Delamination is not a concern effectiveness hence can lead to increased mechanical stresses

Gels Al, Al 2O3 , Ag particles •Conforms to surface irregularity before •Cure process needed in silicone, olefin cure •Lower thermal conductivity than matrices that require •No pump out or migration concerns grease curing •Lower adhesion than adhesives; delamination can be a concern Adhesives Typically Ag particles in •Conform to surface irregularity before •Cure process needed a cured epoxy matrix cure •Delamination post reliability testing is a concern •No pump out •Since cured epoxies have high post cure •No migration modulus, CTE mismatch induced stress is a concern Mahajan, “Thermal Interface Materials: A Brief Review of Design Characteristics and Materials”, Electronics Cooling, Feb 2004 Principles of Conduction

Primary Heat Transfer Path Fourier’s Law (1D):

dT Area, Ac dx dT q= − kA q c dx x

x1 L x2 k – thermal conductivity [W/mK] is a 1D Conduction Resistance material property A – cross sectional area perpendicular to ∆T c q = heat flow Rcond,1 D dT/dx is the rate of change of temperature with distance (K/m) L Rcond,1 D = kAc

L.-T. Yeh, Thermal Management of Microelectronic Equipment, 2016 Integrated Heat Spreader Resistance

Heat flow is not 1D when heat sink is larger Primary Heat Transfer Path than the chip area; ‘constriction’ resistance associated with heat spreading to different area.

 The spreading resistance can be determined from the AAps− πλkA R+ tanh( t ) R = po following set of parameters: c π 1+ πλkA R tanh( t ) k AAps po • footprint or contact area of the heat source, As 3 • footprint area of the heat sink base-plate, Ap π 2 1 thickness of the heat sink base-plate, t λ = + • AA • thermal conductivity of the heat sink base-plate, k ps • average heat sink thermal resistance, R0 Principles of Convection

Primary Heat Transfer Path Newton’s Law of Cooling:

L q= hA() T − T 1 s ∞ h= ∫ hdx L 0 h – heat transfer coefficient [W/m2K] is not a material property, but depends on the flow conditions and geometry

y y T∞

T∞ u(y)

U∞

x Ts Principles of Convection

Primary Heat Transfer Path

L.-T. Yeh, Thermal Management of Microelectronic Equipment, 2016 Internal Flow (Constant Surface Temperature):

q=∆= hA() Tlm mc p ( T out − T in ) Ts

(TTs− out) −−( TT s in ) ∆=Tlm T ln[(TTs−− out) /( TT s in ) ] Tm(x) hPx Tx( )=+− T ( T T )exp m s m, in s  x mc p in out Fin Heat Sink Resistance

Assumptions Primary Heat Transfer Path • Uniform heat sink base temperature • Fully developed flow between fins (conservative) • Uniform flow • Ambient air temperature constant around fins

q=ηo hA t() ∆ T lm

1 G Rt = ηothA

Tm H h

Ts Fin Heat Sink Resistance

Assumptions Primary Heat Transfer Path • Uniform heat sink base temperature • Fully developed flow between fins (conservative) • Uniform flow • Ambient air temperature constant around fins

k Nu h = fluid dc G Nu – nondimensional Nusselt number is a constant for fully developed laminar flow d - Hydraulic diameter [m] is the c Tm characteristic dimension H h 4GH d = c 2HG+ Ts G – fin gap; H – fin height Fin Heat Sink Resistance

Total Heat Sink Resistance : Primary Heat Transfer Path ()TT− R = sa HS q

q=ηo hA t() ∆= T lm mc p ( T out − T a )

(TTs− out) −−( TT s a ) ∆=Tlm ln[(TTs−− out) /( TT s a ) ] η hA T=−− T( TT )exp −ot out s s a  mc p

()TTsa− 1 RHS = = q η hA −−ot Determines the required mc p 1 exp flow rate of air mc p Simple Resistance Model

Primary Heat Transfer Path Ta [K] (air temperature)

q [W] Rtot [K/W] (heat (thermal flow) resistance)

Tj [K] Total temperature rise of the (silicon (TTja− ) chip described by the junction- q = temperature) to-ambient thermal resistance Rtot Advanced Thermal Management Strategies

Justin A Weibel: [email protected] https://engineering.purdue.edu/CTRC Next-Gen Thermal Management

Naval vessels and aircraft are Northrop equipped with increasingly Grumman AN/APG-81 advanced electronic systems that AESA Radar require high-density cooling strategies to enable reductions in size, weight, and power without Active electronically scanned arrays (AESA) house hundreds of compromising performance. high-power amplifiers (HPA); next-gen GaN HPA devices require transformative high-heat-flux embedded cooling strategies.

DARPA ‘Thermal Ground Planes’ DARPA ‘ICECool Fundamentals’

Radio Frequency TGP Hierarchical Manifold Microchannel Array (Raytheon, Purdue, GTRI, Thermacore) (Purdue, Raytheon, IBM) Vapor Chamber and Heat Pipe Spreaders

Justin A Weibel: [email protected] https://engineering.purdue.edu/CTRC Principles of Heat Pipe Operation

Passive Operation Low ΔT Operate against gravity

http://www.aavid.com/product-group/heatpipe/operate Reliable Heat Transfer Performance

Evaporator - Qin Condenser – heff &Tamb

. Effective Conductivity . 100,000 W/mK possible . Not an intrinsic property; does not account for operation limits; meant for higher level system models after experimental validation Heat Pipe Transport Limits

Reay and Kew, Heat Pipes (2006) Heat Pipe Transport Limits

Transport What it is Why it happens Possible Solution Limit Sum of gravitational, liquid and Heat pipe input power Modify heat pipe wick vapor flow pressure drops exceed exceeds the design heat Capillary structure design or reduce the capillary pumping head of the transport capacity of the heat power input heat pipe wick structure pipe Typically only a problem at Vapor flow reaches sonic velocity start-up. The heat pipe (choking) when exiting heat pipe Power/temperature carries a set power and the Sonic evaporator resulting in a constant combination, too much power large ∆T self-corrects as heat pipe transport power and at low operating temperature the heat pipe warms up. large temperature gradients Increase vapor core area. High velocity vapor flow prevents Heat pipe operating above Increase vapor space Entrainment condensate from returning to designed power input or at too diameter or operating /Flooding evaporator (liquid tear-off) low an operating temperature temperature

High radial heat flux causes Use a wick with a higher Film boiling in heat pipe film boiling resulting in heat Boiling heat flux capacity or spread evaporator initiates pipe dryout and large thermal out the heat load resistances

Viscous forces prevent vapor flow Heat pipe operating below Increase heat pipe in the heat pipe, vapor pressure recommended operating operating temperature or Viscous drop can’t exceed absolute vapor temperature, operating near find alternative working pressure freezing point fluid Adapted from S. D. Garner, Heat Pipes for Electronics Cooling Applications (1996) Capillary Limit

Pclvg≥∆ PPP +∆ +∆

. At steady operation, the following pressures balance

. Capillary pumping pressure, Pc

. Liquid pressure drop, ΔPl

. Vapor pressure drop, ΔPv

. Gravitational pressure drop, ΔPg

. There is a maximum capillary pressure that can be developed by

a given liquid-wick combination. The required PC must never exceed PC,max that can be provided by the wick. Else, dryout is caused in the evaporator. Fluid Figures of Merit

. Neglecting gravitational and vapor pressure drop

Pclvg≥∆ PPP +∆ +∆ . The maximum heat transport becomes ρσ h     fg KA 2 Qmax =    µ Lreff  eff  . Fluid property dependent Figure of Merit

Mills (1995)

ρσ h M = fg µ Microchannel Heat Sinks

Justin A Weibel: [email protected] https://engineering.purdue.edu/CTRC Why Microchannels?

• Amongst the most effective of the high-heat-flux heat transfer technologies • Flow boiling operation has promise for: (1) lower pumping power, (2) improved fluid temperature uniformity, and (3) higher heat transfer coefficients (compared to single-phase liquid cooling) • Why now? new manufacturing techniques and emerging application needs

Microchannel Test Chip High-Speed Visualization of Flow Boiling in a Microchannel A Simple Calculation

As revealed by simple scaling, large convection coefficients are associated with small channel dimensions (h ∝ 1/d).

Example: Consider a microchannel of width 60 µm and depth 300 µm (i.e.,

Dh = 100 µm). With a uniform wall temperature assumption, in fully developed laminar flow, the NuDh is 4.8. Even for a dielectric liquid with k = 0.06 W/mK,

2 h = (NuDh kf) / Dh = 2880 W/m K

For water (k ≈ 0.6 W/mK) the value of h is 10 times higher Single-Phase Heat Transfer

Reference Correlations Conditions Valid range

Geometry Flow regime Kays and Crawford 2 3 45 (1) Rectangular Fully developed flow Nu fd =−+8.235( 1 1.883/α 3.767 / α − 5.814 / α + 5.361/ αα − 2 / ) (Both hydrodynamically and Re < 2200 thermally) Incropera and DeWitt µ 0.14 Circular Simultaneously developing Sieder-Tate correlation 1/3 f (2) Nu =1.86( Re PrDL / )  flow Re < 2200 µw Stephan and Preußer 1.33 Circular Simultaneously developing 0.7 < Pr < 7 Stephan correlation 0.0677( Re PrDL / ) (3) Nu =3.657 + 0.3 flow or 1+ 0.1Pr( ReDL / ) (Constant wall temperature) Re Pr D/L < 33 for Pr > 7 Stephan and Preußer 1.33 Circular Simultaneously developing 0.7 < Pr < 7 Stephan correlation 0.086( Re PrDL / ) (4) Nu =4.364 + 0.83 flow or 1+ 0.1Pr( ReDL / ) (Constant wall heat flux) Re Pr D/L < 33 for Pr > 7 Incropera and DeWitt 0.8 Circular Hydrodynamically Hausen correlation 0.19( Re PrDL / ) (5) Nu =3.66 + 0.467 developed, but thermally Re < 2200 1+ 0.117( Re PrDL / ) developing laminar flow (Constant wall temperature) 1/3 Shah and London  DD  Circular Hydrodynamically 1.953 Re Pr Re Pr≥ 33.3 developed, but thermally −  LL  (6) Nu =  developing laminar flow  DD (Constant wall heat flux) 4.364+< 0.0722Re Pr Re Pr 33.3  LL Kacac et al. µ 0.14 Circular Hausen correlation 2/3 1/3 2/3 f (7) Nu =0.116( Re −+ 125) Pr 1(DL / )  µ Transitional flow 2200 < Re < w 10000 Incropera and DeWitt Dittus-Boelter correlation Nu = 0.023 Re0.8 Pr 1/3 (8) Circular Fully developed turbulent flow Re > 10000 Incropera and DeWitt Petukhov correlation ( f /8) RePr (9) Circular Fully developed turbulent Nu = 1/2 2/3 flow Re > 10000 Kf+−12.7( /8) ( Pr 1) 900 0.63 K =+−1.07 Re 1+ 10Pr

Gnielinski Gnielinski correlation ( f 8)( ReD -1000) Pr (10) Circular Both the transitional and 3000 < Re < 5 Nu= 6 12 23 fully developed turbulent × 1+− 12.7( f 8) (Pr 1) 10 flows Single-Phase Conclusions

• Conventional theory offers reliable predictions for the flow characteristics in microchannels • Heat transfer measurements are conducted with an emphasis on correctly matching the entrance and thermal conditions when comparing to conventional correlations developed for large-sized channels • Numerical predictions obtained based on a classical, continuum approach with carefully matched entrance and boundary conditions are in good agreement with the data Two Phase Flow in Microchannels

. Flow boiling is attractive because it . can dissipate higher heat fluxes . requires lower coolant mass . has more uniform wall temperature

. Flow boiling mechanisms in microchannels are not fully understood . Most available flow maps based on macroscale channel flow are not applicable . Results obtained for single microchannel cannot be simply extrapolated for multiple microchannels . Lack of generally accepted models/correlations

Liu et al., IJHMT 48, 2005; JHT 127, 2005 Two Phase Flow in Microchannels

. In-situ measurement of local wall temperature & flux, high-speed Flow Boiling Regimes in Microchannels visualizations Bubbly . Flow regime determination and regime- based modeling for dielectric fluids . Instability, hysteresis, surface roughness, geometry, operating condition effects Alternating Churn/Annular Microchannel Test Chip

Inverted Annular

Chen, Garimella, IEEE CPT 2007; IJMF 2006 Channel width: 100-5850 μm Bertsch, Groll, Garimella, IJHMT 2008; NMTE 2008 Channel depth: 100-400 μm Harirchian, Garimella, IJHMT 2008, 2010; IJMF 2008, 2009; JEP 2010 Hydraulic diameter: 100-750 μm Mass flux: 225-1450 kg/m2s Patel, Harirchian, Garimella, IPACK 2011 Flow Regime Prediction and Models

Comprehensive Flow Regime Map

Confined Slug . G – x and jf – jg flow regime maps 1100-12 Churn/Confined Annular depend on channel dimensions Bubbly Churn/Wispy-Annular . Comprehensive four-region flow Churn/Annular regime map represents data for a wide range of channel e R .

dimensions, mass fluxes, and heat l -3 11000 fluxes B . Validation of flow-regime based theoretical heat transfer and pressure drop models against Bl= 0.017( Bo0.4× Re -0.3 )

empirical data -4 0.5 1100-1 Bo×= Re 160

101 102 103 104 (Bo)0.5.Re

“Microchannels”: Confined Flow Regimes “Macrochannels”: Unconfined Flow Regimes Harirchian, Garimella, IJHMT 2010, IJMF 2012 Current Challenges/Needs

 Embedded sensors for system control  Measurement techniques for characterization of thin liquid films that govern heat transport in small channels  Development of numerical methods to allow tractable direct simulation of flow boiling in small channels  Electro-thermal co-design for development of high- performance, intrachip heat sink technologies  Characterization of the impact of static/dynamic instabilities and flow maldistribution on performance