200 likes | 206 Views
This presentation discusses the value and methodology of thermal simulations for the LASP test board, which is used for the upgrade of the ATLAS Liquid Argon Calorimeter in the ATLAS detector at CERN. The simulation results demonstrate the impact of component placement and heatsink design on temperature and airflow management. The ANSYS Icepack software is used for the simulations.
E N D
LASP TeStBoardCOOLINGSIMULATIONS – FIRST APPROACH Mohamed ZAYED Presented by Jean-pierre Cachemiche on behalf of the ATLAS Liquid Argon Calorimeter group TWEPP 2019 2-6 September, 2019 – Santiago De Compostela - Spain
OUTLINE • Context • Value • Method • Simulation results • Simulations versus measurements • Conclusion
CONTEXT The LAr Signal Processor (LASP) board Readout of the ATLAS LAr Calorimeter – Upgrade Phase 2 ATLAS Lar Calorimeter – Electronics block diagram - Phase 2
CONTEXT: LASP Test Board • 2 High-end FPGAs – Stratix 10 (Intel/Altera) • High data exchange rate: 1Tb/s per FPGA • Parallel processing : on-the-fly energy and time calculation of calorimeter signals
CONTEXT: MAIN CONSTRAINTS / KEY NUMBERS • Power estimation per FPGA 120 W • Projection from a similar design – LATOME phase 1 upgrade • INTEL/ALTERA Application Note : AN-767 • Designing for Stratix 10 Deviceswith Power in Mind • Coolingcapacitylimitation of verticallycooled ATCA shelves in LHC racks : • Maximum power dissipation of 350W per main blade • Board to boardspacing ~ 30 mm • Strong limitation on heatsinksheight • Optical transceivers (SAMTEC Fireflys) operationaltemperature • Max 50 °C
CONTEXT: Technology • ATCA technology • Complete processor : 200 ATCA boards • ~ 60 000 opticalfibers (in/out) 6 FAN’s Extraction 6 FAN’sIntake Overallairflow = 1180 CFM (2000 m^3/h)
VALUE HOW USEFUL ARE THERMAL SIMULATIONS ? 3 EXAMPLES • Component placement • MainlyFPGAs and opticaltransceivers • Avoidhair-dryereffect • Avoid hot zones • Heatsink design • Shape • Thickness • Number and dimension of fins • Airflow management • Study and understandairflowpath • Optimizeairflow use MX SX HOTSWAP J1
METHOD • Using the software ANSYS – ICEPACK (one of the mostpowerfultools) • Input Parameters : • Geometry of card • Geometry and type of radiators • Materials • Position of components • Ambient temperature • Temperature and airflow speed • Power dissipated by each component • Resultsfrom simulation • Temperature Graphs • Airflow Graphs
METHOD: Ansys Icepack main features • Software thatprovidespowerful solutions for coolingelectronics • It uses the ANSYS Fluent Fluid Dynamics (CFD) solver for fluid and thermal flow analysis • ANSYS Icepack CFD solver uses ANSYS Electronics Desktop (AEDT) and Graphical User Interface (GUI) • Icepack’ssolverperforms conduction, convection, radiation and heattransferanalyzes • It has manyadvancedfeatures for modelinglaminar and turbulent flows • The software provides a libary of fans, heatsinks and materials to provide solutions to the frequentcoolingproblems of electronics
SIMULATION RESULTS: Based on “first approach” placement FPGA MX = 68°C 120 W FPGA SX = 68°C 120 W Gives an order of magnitude on FPGAstemperature and airflowpath
SIMULATION RESULTS: Heat sink modelling • Heatsink dimensions 140 mm x 90 mm x 30 mm Impact on FPGA temperature ? • Variable = Number or Thickness of fins Airflow circulation and Heat dispersion are concurrent effects Vertical fins –length: 140mm –thickness: 0,3 mm OPTIMAL VALUE The optimal value is not alwayseasy to machine ! Copper heatsinkfrom Dynatron 120mm x 82mm x 27mm 0,6 mm thickness per fin Vertical fins –length: 140mm – 55 fins – 0,45mm beetween fins Helpsmake a decision on heatsink design
SIMULATION RESULTS: LASP Test Board Final Placement LASP Geometry FPGA Mx (130W) withitssink Voltage Regulator FireFly’s hot-swap DC/DC Memory card Voltage Regulator FPGA Sx (130W) withitssink • This placement isalreadyiterated • Placement takes component functionality and coolingrequirementsintoaccount
SIMULATION RESULTS: ANSYS boardmodelling Ambient Temperature = 22°C LASP ANSYS 400Wdissipated PCB Heatsink MX FPGA MX130 W 1W/FireFly – 24 W Memory Heatsink SX Capacitors FPGA SX 130 W hot-swap – 10 W DC/DC – 80W 4 Fan – 25CFM
SIMULATION RESULTS: ANSYS Heatsinkmodelling • Pure Copper SX/MX Heatsink • Base : 3 mm • Fin : 2 mm thickness x20 • 68 mm width • 111 mm length • 20 mm height • Conductivepaste : Silicon LASP ANSYS – Isometricview • AluminumFireflyHeatsink • - Base : 1 mm • Height : 12 mm • Heatsink dimensions: availablespace. • Cheaper and lesschallengingthan water cooling
SIMULATION RESULTS: FPGAs temperature Simulation – FPGA SX / MX – Heatsink SX / MX • Trends of simulation is nice • The MX is hotter than SX, why ? • The DC/DC component dissipate 80 W • The hot-swap component dissipate 10 W • The AirFlow carry the heat of thesetwo components to the MX • Heatisalsodispersed in the PCB by components Heatsink MX 48°C FPGA MX 49°C HeatsinkSX 43°C FPGA SX 44°C 4 Fan – 25CFM
SIMULATION RESULTS: Fireflys temperature and airflow Specification by CERN is OK Fireflys Temperature 40°C – 47 °C 4 Fan – 25CFM • Good air circulation in heatsink • Good DC/DC cooling 4 Fan – 25CFM
SIMULATION RESULTS: PCB temperature Simulation – Temperature Influence on PCB • DC/DC – 60°C • Good temperature for a component without a heatsink 49°C • Hot-swap – 50°C • Increased temperature due to DC/DC but stillreasonable 44°C • PCB • Average temperature PCB = 42 °C 50°C 60°C 4 Fan – 25CFM
SIMULATIONS VERSUS MEASUREMENTS: Phase 1 LDPB processor configuration How good are Ansys simulations ? • Simulation of 4 LATOME mezzanine cardsaligned in vertical position like on Phase-1 LArProcessor Board (LDPB) • Verysimplistic configuration (FPGA environment not takenintoaccount) • Just to check if tendencies are Ok • Simulation results are compared to measurementsdone @ CERN
SIMULATIONS VERSUS MEASUREMENTS: Phase 1 LDPB configuration results • ANSYS simulation vs measurements 77°C 68°C 57°C 40°C Without Extraction / 3 FAN – 70 CFM Extremely simplified SIMULATION RESULTS AND MEASUREMENTS SHOW THE SAME TENDENCIES
CONCLUSION • The ANSYS-Icepack Software has been used to simulate the thermal effects on the LASP Test Board for the LArcalorimeter for Phase 2 upgrade • The simulation washelpful to : • Guide the placement of components • Help define the shape and fins of heatsinks • Understand the coolingairflowacross the board • Absolute temperature values of simulation are to betakenwith caution due to imperfect/incompletemodelling • But comparative simulations are veryhelpful • Offers a point of comparisonwhen the boardbecomesavailable