low material budget microfabricated cooling devices for particle detectors

P. Petagna & A. Mapelli 1

Low material budget microfabricated cooling devices for particle detectors

P. PETAGNA and A. MAPELLIOn behalf of:

• CERN PH/DT• The NA62 Collaboration• EPFL – LMIS4• EPFL – LTCM• UCL – ELEC/DICE (SOI & MEMS)

30 Sep 2010


Outline of the talk

30 Sep 2010

• Why micro-channel cooling for HEP?• A first application: local cooling for the NA62 GTK• Proposed solution and approach to the problem• Micro-fabrication process• Structural analysis• Thermo-fluid dynamics simulations• First tests on a full-scale prototype• Layout optimization • Next steps and beyond


Why m-channel cooling?

30 Sep 2010

Radiation length (X0): mean distance over which the energy of a high-energy electron is reduced to 1/e (0.37) by bremsstrahlung

1 – Minimization of material budget(Dahl, PDG)

More readily usable quantity: X0 = X0/r [cm]

Cu: 1.436 cmSteel: ~1.7 cmAl alloy: ~8.9 cmTi: 3.56 cmSi: 9.37 cmC6F14 @ -20 C : 19.31 cmK13D2U 70% vf: 23 cmCO2(liquid) @ -20 C: 35.84 cm

Minimize material budget Minimize x[cm] / X0[cm](i.e. use material with high X0 and minimize thickness)

m-channel cooling naturally addresses this issue through the use of Si cooling plate and tiny (PEEK?) pipes in extremely reduced

thickness

Additional

advantage: no

CTE mismatch



30 Sep 2010

Material budget of the CMS Si-strip tracker (10 layers)Material budget of the CMS Si-Pixel tracker (2 layers)

Present LHC large Si trackers (ATLAS and CMS) ~ 2% X0 per layer

SLHC “phase II” upgrade: “significant” reduction needed

Future trackers at ILC ~ 0.1 ÷ 0.2% X0 per layer



30 Sep 2010

2 – Cooling power enhancement

h

f

DNuK

h

Newton’s law for convective heat flux: )( fw TThSQ

Heat transfer coefficient for m-channel system:

Hydraulic diameter ~ 10-4 m or less

Nusselt number = 3.66 ÷ 4.36 for fully developed laminar flow

Fluid thermal conductivity = 0.05 ÷ 0.11 W/mK for low temperature fluids

~103 W/m2K

m-channel cooling: very high heat transfer coefficients (very small Dh possible) and very high heat flux (large S available)



30 Sep 2010

3 – Reduction of DT between heat source and heat sink

With a standard cooling approach, the DT between the module and the fluid ranges between 10 and 20 C (small contact surface + long chain of thermal resistances)

With an integrated m-channel cooling approach, the large surface available for the heat exchange (cold plate vs. cold pipe) and the natural minimization

of the thermal resistance between the source and the sink effectively address the issue of the DT between the fluid and the element to be cooled

Lower temperatures are envisaged for the future Si-trackers at SHLC. This has non-negligible technical impacts on the cooling plants


An example of future potential use

30 Sep 2010

Concept of module for a “level-0 trigger” layer @ SHLC (courtesy of A. Marchioro)

SensorRO chips

m-channel cooling plate Manifolds

Interconnect

A first application: the NA62 GTK

30 Sep 2010 8P. Petagna & A. Mapelli



Vacuum tank

Mag2 Mag3

Mag4Mag1

GTK1 GTK3

GTK2Cedar

selects particleswith 75 GeV/c

seeskaons only

Achromat

250 m

beam: hadrons, only 6% kaons-> only 20% decay in the vacuum tank into a pion and 2 neutrinos -> out of which only 10-11 decays are of interest

straw chambers RICH

hit correlation via matching of arrival times – 100 ps

RICH identifies pions

straw chambersmeasure position

GTK seesall particles



• Sensor & bonds: 0.24% X0 (~200 µm Silicon)

• RO chip: 0.11% X0

(~100 µm Silicon)

•Passive or active cooling plateFinal target: 0.10 – 0.15 % X0

•Priority: minimize X0

•Acceptable DT over sensing area ~ 5 °C•Dimension of sensing area: ~ 60 x 40 mm•Max heat dissipation: ~ 2 W/cm2 •Target T on Si sensor ~ -10 °C

• Support structure outside acceptance region: ~ FREE

3 INDEPENDENT STATIONS:

“SIMPLE” SYSTEM

• 18 000 Pixels / station (300 x 300 mm, 200 mm thick)• 10 ASICS chips bump-bonded to the sensor

Proposed solution


Schematic of the layout of the proposed m-channel cooling plate the coolant will enter and exit the straight channels via manifolds positioned on top and bottom. The channels, distribution manifold and openings for the inlet and outlet connectors are etched into a silicon wafer, which is then coupled to a second wafer closing the hydraulic circuit.

The final goal is to have both wafers in silicon bonded together by fusion bonding to produce a monolithic cooling element

An alternative design, in case of technical difficulties with the fusion bonding process, relies on a flat Pyrex cover 50 µm thick anodic-bonded to the silicon wafer carrying the hydraulic circuit. On top of this flat plate, an additional silicon frame (surrounding the beam area) will be again anodic-bonded. In this way the global structure of the cooling wafer will be symmetric, the effects of coefficient of thermal expansion (CTE) mismatching between silicon and Pyrex will be minimized and the same resistance to pressure and manipulation as in the baseline case will be attained

Approach to the problem


Take advantage of recent results obtained in two different fields of development:• m-channel cooling devices have started to be actively studied for future

applications for high power computing chips or 3D architectures. • Thin and light m-fluidic devices in silicon are largely in development for bio-

chemical applications.

Anyway for the first case, where the power densities are extreme, the mass of the device (hence its material budget) is an irrelevant parameter. In the second case the typical values of the flow rate and pressure are much lower. Furthermore, the presence of a low temperature fluid and possibly of a high radiation level is unique to the HEP detector case.

dedicated R&D is nevertheless unavoidable for the specific application

under study.

Approach to the problem


The procedure followed to tackle the different challenges and to converge in a limited time on a single device satisfying all the requirements is to move in parallel along different lines of R&D in a “matrix” approach, where the intermediate results of one line are used to steer the parallel developments.

Fabrication technique

studies

Thermo-fluid dynamic

simulations

Numerical structural

simulations

Experimental tests

Common specs

Possible

layouts

Optimal

layout

Pressure

limits

m-fabrication process


START: Czochralski silicon wafer polished on both sides (4′′ diameter, 380 μm thick, 0.1-0.5 ohm-cm p-type).

(a) A layer of 1 µm of oxide (SiO2) is grown on both sides of the wafer

(b) Clariant AZ-1512HS photoresist is spin coated on one side of the wafer at 2000 rpm and lithography is performed to obtain an image of the channels in the photoresist

(c) Dry etching of the top layer oxide is used to transfer the micro-channels pattern

(d) A second lithography is performed with frontside alignment to image two fluid transfer holes, 1.4 mm diameter, for fluid injection and collection from the two manifolds.

(e) Deep Reactive Ion Etching (DRIE) is used to partially etch the access holes down to 280 µm

(f) The photoresist is stripped in Microposit Remover 1165 at 70°C

(g) and DRIE is used to anisotropically etch 100 μm deep channels separated by 25 µm wide structures in silicon

(h) Subsequently the oxide layers are removed by wet etching in BHF 7:1 for 20 min at 20°C

(i) At present, the processed Si wafer and an unprocessed Pyrex wafer (4” diameter and 525 µm thick) are then cleaned in a Piranha bath (H2SO4 + H2O2) at 100°C and anodic bonding is performed to close the channels with the Pyrex wafer

m-fabrication process


Scanning Electron Microscope image of the cross-section of 50 x 50 mm channels etched

in silicon bonded to a Pyrex wafer

Finally, PEEK connectors (NanoPort® assemblies from Upchurch Scientific) are aligned, together with a gasket and a preformed adhesive ring to the inlet and outlet on the silicon and clamped. They undergo a thermal treatment at 180°C for 2 hours to develop a complete bond between the connectors and the silicon substrate.

The anodic bonding is performed at ambient pressure and T is raised to 350°C then lowered to 320°C. At this stage a constant voltage of 800 V is applied between the Si and Pyrex wafer.

In the final production both the processed and the unprocessed wafers will be in 525 µm thick silicon.

The bonded wafer undergoes a further processing: this includes a final local etching to obtain a thinner region in the beam acceptance area

The resulting wafer is diced according to alignment marks previously etched in Si to obtain a cooling plate with precise external references for integration into the electromechanical assembly

1 mm

....

30 mm

Structural analysis


“Sacrificial” samples with different manifold width are produced and brought to collapse by gradually increasing pressure under a high speed camera in order to determine the limit pressure and the exact breaking mechanics. 60

3

0.05 0.025

varying width

1 – Experiments

Structural analysis


2 – Numerical simulations vs. tests

Yield stress ~25 MPa [ICES 2009]

A simplified ANSYS 2D parametric model has been developed and calculations are checked against experimental results in order to validate the model for further forecasts, including the effect of wall thinning or of geometrical variations

0.0 0.5 1.0 1.5 2.00

20

40

60

80 Pyrex rupturePower (Pyrex rupture)Connector de-tachment

Manifold width (mm)

Pint

(bar

)

Structural analysis


0 0.5 1 1.5 20

50

100

150

200

250

300

350

400

450

Rupture Pressure for Silicon Cover (165 Mpa)

tp=50µ

tp=200µ

tp=350µ

tp=525µ

Manifold Width (mm)

Pint

(Bar

)

0.0 0.5 1.0 1.5 2.00

20

40

60tp=50µtp=200µ tp=350µ tp=525µ Power (tp=525µ )

Manifold width (mm)

Pint

(bar

)

Rupture Pressure for Pyrex Cover (25 Mpa)

3 – Extrapolations

Thermo-fluid dynamics simulations


The choice of the cooling fluid circulating in the micro-channels has naturally been oriented towards perfluorocarbon fluids (CnF2n+2), which are widely used as coolant medium in LHC detectors. They exhibit interesting properties for cooling applications in high radiation environment such as thermal and chemical stability, non-flammability and good dielectric behaviour. In particular C6F14 is liquid at room temperature and is used as single phase cooling fluid in the inner tracking detectors of CMS.

Properties C6F14 @ -25°CDensity r [kg/m3] 1805Viscosity n [10-7 m2/s] 8.2Heat capacity cp [J/(kg K)] 975Thermal conductivity l [10-2 W/(m K)] 6.275

Based on the properties of C6F14, a mass flow of 7.325*10-3 kg/s is required to extract the heat dissipated

by the readout chips (~32 W) with a temperature difference of 5K between the inlet and outlet temperature

of the coolant

1 10 4 2 10 4 3 10 4 4 10 42 10 3

4 10 3

6 10 3

8 10 3

0.01

0.012

0.014

mpt 90mm b( )

b

The results from the analytical calculations performed indicate that the suited range of the micro channel geometry is the following:• Width: between 100 mm and 150 mm• Height: between 80 mm and 120 mm• Fin width: between 25 mm and 75 mm• Between 300 and 500 channels to cover the area

Flow rate attained with 2 bar Dp vs. channel width for a fixed height of 90 mm

First tests on a full-scale prototype


1mm

Inlet

Outlet

manifold depth 100mm

• Channel cross section 100mm x 100mm• Power density 1 W/cm2 (50% nominal)• Mass flow 3,66 x 10-3 kg/s (50% nominal)• Inlet temperature 18 C• Outlet pressure 1bar• Laminar flow

Test sample and numerical model



Simulated vs. experimental pressure drop



Thermal visualizationIN

OUT

Thermograph before injection

Thermograph at injection

IN

OUTThermograph after few seconds of coolant circulation

Heat load simulated by a Kapton heater of suited resistance and geometrical dimension



Steady state DT between inlet and surface probes

1 2 3

4 5 6

4

5

6

321

Layout optimization


Outlet

Wedged manifold, depth 150mm, 280mm and 400mm

Optimized geometry for uniform and minimal DPCFD models of the geometry presently under tests have been successfully validated. Further

optimization of the manifold geometry and of the channel cross section can then be performed through CFD analysis in order to reduce the amount of samples to be produced for testing purposes

Inlet


Layout optimization

30 Sep 2010

Effect of inlet manifold geometry on DP

Rectangular manifold, 1 mm wide, 100 mm thick, central inlet & outlet

Wedged manifold, 1.6 mm Max width, 150 mm thick, opposed inlet & outlet



Layout optimization


Layout optimization


two inletstwo inlets

two inletstwo inletstwo inlets

Summary table

Next steps and beyond


Immediate future1. Perform full-scale thermal tests in cold (vacuum vessel)

2. Define the details and properties of the Si-Si fusion bonding process (industrial partnership), fix the final thickness and verify with a new series of tests

3. Complete the detailed study of the integration in the GTK module

1. Study m-channels in combination with CO2 evaporative cooling

2. Challenge the system aspects for larger and more complex detectors (e.g. ATLAS IBL? CMS PIX? LHCb VeLo?)



Next year

Two-phase CO2 vs. single phase C6F14: DP and DT in a 50 x 50 mm channel

Two-phase flows comparison: DP and DT in a 50 x 50 mm channel plate under the same heat and mass flow for CO2 , C3F8 and C2F6



Sensor

Chips

REMOVE

Embedded m-channels!

A long-term dream?

low material budget microfabricated cooling devices for particle detectors

Documents

channel cooling approach

microchannel cooling

thicknessmchannel cooling

cooling plants

local cooling

standard cooling approach

use of si cooling plate

channel system