Download GAB TDR Section 9.4

subsection FEA and thermo-mechanical performance for petals and staves; deformations.
*** This section refers to the earlier section 9.1.4 ”thermal performance requirements” that needs extensive
updating (wrt LoI) and would need expanding to include the petal. Are those details in the right place?
*** ‘deformations’ might refer to: Static Deflections (Gravity, Loading by services); Vibration; Coupled ThermalMechanical (cool-down, partial powering, time/luminosity-dependent power loads, TID annealing ..); Creep,
Moisture, Magnetic field etc effects. There is not an appreciable body of FEA studies of such effects (but should check
again with S.Yang and R.Nickerson et al.).
*** Figures for this section generally need to be re-made so as to be consistent with TDR guidelines.
subsubsection: Thermal Performance
Detailed three dimensional thermal FEA models have been developed
{ref. Abaqus}
that predict the temperature distributions across the sensor, readout chips and local support structures. These
address major concerns during normal operation such as thermal runaway headroom (section n
{ref. subsection Requirements and Overview, subsubsection Thermal Performance }
above), sensor leakage current (hence shot noise) and cooling requirements, that depend in turn on radiation
damage to both the sensor and readout chips. They also allow temperature estimates outside normal operation, for
instance during ‘beam off’ conditions and initial detector commissioning at elevated temperature. The models
concentrate on evaluating the conductive thermal paths to the cooling pipes and heat removal by the evaporating
CO2: heat exchange with ambient (by convection and radiation) has been considered, but in most situations is a
negligibly small effect. The FEA predictions will be verified by comparison with stave and petal ‘thermo-mechanical’
structures currently under construction.
FEA models of the barrel stave and end cap petal differ mainly due to their geometry and readout chip arrangements
(fig n).
{figure: FEA model regions} .
Since the stave has a predominantly periodic structure, it is sufficient to model a short segment at its end, where
there is a coincidence of slightly higher radiation level, thermal load from the end-of-structure chips and reduced
cooling efficiency (due to the insulating break). Separate models are required for the inner radius (short strip) and
outer radius (long strip) barrels. By contrast, the end cap petal spans the radial extent of the strip tracker with an
irregular module geometry: here the FEA model is constructed to describe the full petal. All petals are of the same
design, and the FEA is usually run for conditions expected at the end of the tracker, where the predicted radiation
level is highest.
Thermal Conductivities
Thermal properties used as input to the FEA are summarised in Table n.
{Table: Thermal Conductivity Input}
Apart from minor variations they are the same for the Stave and the Petal, in particular along the conduction path
between sensor and coolant as described in section
{ref. subsection Requirements and Overview, subsubsection Thermal Performance } .
Evaporative CO2 cooling is simulated as a convective film between the pipe wall and fluid, at its evaporation
temperature. The convective heat transfer coefficient (htc) varies around the cooling loop and is computed using the
stand-alone package CoBra [ ] {ref. CoBra} , for the appropriate stave or petal thermal loads. The cooling pipe inner
diameter is currently assumed to be 2mm, although calculations are in hand to optimise this.
Conductivity values for the foam and co-cured facing have been measured by ATLAS institutes and are so far
consistent with manufacturers’ specifications. Thermal conductance through the various layers, of which the most
crucial is the interface between the graphitised foam and cooling pipe, have also been measured. Where thermal
performance is sensitive to these they will be monitored as part of the QA/QC process.
Read-out chip temperatures depend on the thermal path through the printed circuit boards. In particular, the DCDC
converter (that steps down the input voltage to power the ABC and HCC chips) is a localised source of power
dissipation that is particularly important at the TID peak: care is being taken to model the structure of the pcb (and
compare with measurements) so as to correctly predict the DCDC chip temperature.
Certain material and design variants are under study and their effect simulated in the FEA models; these include the
use of UV cure glue for chip attachment (for increased assembly speed) and a reduction of the glue area between
sensor and bus (in order to reduce its mass).
Thermal Loads - performance at the TID peak .
Here I assume that the chip power dissipation is given in earlier sections. e.g. Section 8.2 “1.5V/4.0 A (max).”
Thermal loads are simulated for power dissipation by the readout electronics {ref. subsection Hybrid Design and
Prototypes}, which is relatively independent of temperature, and for the sensor leakage power, that increases
exponentially with temperature, as described in section {ref. subsection Requirements and Overview, subsubsection
Thermal Performance } . Sensor power dissipation is simulated as Joule heating, by assigning an electric potential
difference across the sensor thickness together with a specific electrical conductivity: the conductivity is input as a
table (for interpolation) that describes the expected leakage power dissipation with temperature. Hence the FEA
correctly accounts for temperature and power variation across the sensor area (albeit not through its thickness).
At locations where radiation levels are highest, ionisation damage to the CMOS read-out electronics is expected to
peak early in the life of the detector, when the sensors are relatively un-damaged and leakage currents are
consequently low. However, the read-out heat itself is predicted to result in large temperature gradients and
elevated chip temperatures, particularly in the regions of the DCDC converters. Figure n {figure: T plots at TID peak}
shows predicted temperature distributions based on the present, limited understanding of the ionising dose effect:
detailed studies are in progress, to quantify the irradiation damage and to understand convective corrections and
design strategies that will reduce its impact.
End of life: Thermal Runaway.
The danger of thermal runaway is greatest at the end of operation where both the leakage current and required
bias voltage are a maximum. By that stage, at the locations most vulnerable to runaway the excess read-out power
due to ionisation damage will have reduced to a negligible level. At the maximum expected sensor power (allowing
a factor x2 on predicted fluence) the FEA finds stable temperature solutions for both the stave and the petal sensors
(Fig. n){figure: Sensor T plots at 3000fb-1}. Temperature variations are smaller here than at the TID peak and
although localised peaks are evident, runaway will occur only if the temperature dependence of heat generated over
an area comparable to the sensor exceeds the thermal conductance of the local support. FEA results for the
evaporation temperature crtitical for runaway are given in table n. { ref. Thermal Performance Summary Table}.
The behaviour of the average temperature of a sensor can be estimated quite accurately from an analytic model in
terms of just two thermal resistance parameters derived from the FEA {ref. Beck&Viehhauser }. This allows a simple
prediction of quantities of interest, such as runaway headroom, shot noise and required cooling power. It will be
further used to understand the combined effects of ionising dose and fluence as these vary with integrated
luminosity and location on the detector.
References
Abaqus: FEA software used here was Abaqus (a product of Dassault Systemes Simulia Corp. RI 02909-2499, USA).
CoBra: Design Considerations of Long Length Evaporative CO2 Cooling Lines, Bart Verlaat and Joao Noite, GL-209,
10th IIF/IIR Gustav Lorentzen Conference on Natural Working Fluids, Delft, The Netherlands, 2012.
Beck&Viehhauser: Analytic model of thermal runaway in silicon detectors, G.Beck and G.Viehhauser,
Nucl.Instr.Meth. A618, 131 (2010).
Figure captions
FEA model regions: FEA model regions. (left) the short strip stave model that simulates the extreme 20cm at the
end of stave and (right, on a reduced scale) the full end cap petal model.
T plots at TID peak: Preliminary temperature distribution predictions at the TID peak for a) the petal surface and b)
the extreme sensors of the stave. The petal result is overly pessimistic, since it assumes peak readout power
simultaneously at all radii.
Sensor T plots at 3000fb-1 (note that the petal plot here assumes no safety factor on predicted fluence. It would be
helpful if Afroditi could supply a T plot for SF=2, for consistency with the stave. Also: should identify which face of the
petal?): FEA sensor temperature distributions at 3000fb-1, assuming sensor bias voltage of 500V and CO2
evaporation temperature of -30C, for the petal located and for the two extreme sensors of the inner barrel stave.
Thermal Performance Summary Table
Inner Radius Barrel Stave
End-cap petal (at highest Z))
Maximum temperature variation
across any sensor at TID peak.
19 C
29 C
CO2 evaporation temperature that
will induce runaway at End of Life
-17 C
-22 C
Summary of FEA thermal performance predictions.
Other Table captions:
Thermal Conductivity Input (attractive though this table is, it does not conform to the format prescribed in the TDR
guidelines): Thermal Conductivity values used in the FEA models.
Back-up / miscellaneous material
Runaway curves? ..TC,crit for Stave and Petal: -17/-22C for safety factor x2!
Runaway sensitivity to (e.g. halving) K values… Module glue pattern…
Shot noise.
Long strip modules (inc higher shot noise)?
Stephanie’s CFD of buoyancy around the barrel?
Deformations: I don’t know of any FEA – and little more than ¾-pt bend, R.N’s creep….
DaveL slides: Bend test, Resonant frequency; DCDC stave edge supported deflections, Stave thermal image.
(Dave:) For the most part these FEAs do not exist I think. I vaguely remember vibrational mode studies by
Stephanie.The gravitational sag of the stave core is pretty straightforward, but of the stave with modules is difficult.
Same for vibration, etc.
It depends on the support and the sensor to stave epoxy. I think your thermal simulations are the most useful
studies done up until now.
(RichardN:) I think I had a similar request from Ian Wilmut, I haven’t don’t anything about it yet, but can certainly
supply information. The main problem is that right now we have results which basically say we have a problem. Not
sure that is what we want in a TDR Anyway, I have this on my list and will produce something
Investigation of detailed dependencies/ value of the analytic approximation.
- sensitivity to material conductivities…
…Leakage current annealing?
…Reverse annealing?
U-bend FEA is OK! (+ Stave T asymmetry reduces away from end as CO2 feed/return approach same conditions).
[… it should be noted that thermal conduction between adjacent modules (along the beam direction for the stave and
the radial direction for the petal) is small in comparison with conduction to the cooling pipe: this weak thermal
coupling between ..?????????? can be treated approximate modularity is useful when evaluating cooling
requirements]
AfroK’s FEAST2 v4 has:
- at TID peak: R3 sensor variation +3.4 to < -25 i.e. > 28.4 degrees, R5 Feast +75C (cf my 19 degrees and +57C).
- at 3000fb-1 (SF=1): R3 sensor Tmax = -16C, R3 Feast -3C.
Note that in the petal FEA the foam is tied to the pipe i.e. negligible interface resistance.
Afroditi’s summary:
Based on the conditions described above, the Petal will be safely away from thermal runaway at the end of the run
(3000fb-1). The coolant temperature at thermal runaway is estimated ~ -15C.
The TID case is also examined, where the chip FEAST2 temperature will be at maximum around 75 C. The maximum
temperature of the silicon is ~ +3 C at the TID. There is big temperature gradient in the hybrid below the DCDC
converter, looking at a cross section of the Petal.
Shot Noise (emld,TonyA, 21 April): Is there an official figure for the S:N required for the strip tracker? (I appreciate
there are other parameters such as time walk, that don't belong in this section). Given your knowledge of the CCE,
what should we assume for the maximum acceptable shot noise?
10:1 is the target. The importance on the shot noise really depends on the assumed noise
.. fix to the FE works (both the increase from the correct polarity and from the ASIC irradiation). And we remove the
extra noise we see between modules and single chips.
Assuming all the module noise is understood and match test chips, Jan brings the performance of correct polarity to
the same value, removes the irradiation effect and the current A12A, the normal S:N is as low as 15:1 on the
outside. The highest allowable shot will be different on each layer, but 700 e- Shot looks ok for all layers. It would
be 3200 nA per channel.
As shot adds in quadrature, the other noise sources if not removed will be dominant realitive to shot at this level so
it seems like a reason (able..?)
Running temperatures for that current gets -4 for the ring 4, -2 for ring 5 and warmer than 0 for the rest. I would
guess the runaway condition will be hit first. It was the case for LHCb. Cheers, Tony