First Thermal Estimates for Serial Powering at Chip/Module Level Yadira Padilla Joe Conway, Charlie Strohman, Jim Alexander, Anders Ryd, Julia Thom Cornell University Jorgen Christiansen CERN July 21, 2016 Outline • • • • • • • Shunt and LDO power dissipation thermal issues Chip and Dee Geometry Assumptions and Thermal properties Nominal Case: Operation Max case (1.5X): Operation Worst case (1.5X): Shut down Summary and comments 7/21/2016 2 Shunt and LDO power dissipation thermal issues 7/21/2016 3 Shunt-LDO power dissipation • Nominal operation average IC current: 2 x 0.8A (analog and digital) @ 1.2v – What if this can become 2x • Start-up • Mis-configuration • SEU Needs to inject enough current to get system started – What if this can become 0. • Powering down analog FEs • No clock to digital All power will have to be “burned” by SLDO • Nominal shunt current: ~25% – Total current ~2 x 1A – Note: This is what allows serial power to run with constant currents despite load variations • LDO minimum voltage: 0.3volt @ 1A (lower would be nice) – Depends on current • Optimizing injected current (controller by external PS) – Must assure correct function in all cases and no “melt-down” – Can be adjusted on-line • Optimizing Shunt-LDO (SLDO): – Resistance (defined by on-module resistor and possibly on-line) 4 Shunt-LDO regulator Classical shunt regulator: Inject current into voltage “clamps”/shunts + LDO • Problematic current sharing: Variations in clamp voltages and parasitic resistance • Coupling between analog and digital supply voltage via common shunt ATLAS outer tracker developed different schemes to cope with this but this work is now abandoned (decided for DC/DC power) Resistive SLDO: Make it look like well controlled resistors • Well defined current sharing between resistors • Multiple “independent” shunt-LDOs per chip with different output voltages • Multiple chips in parallel • Improved decoupling between analog and digital • (Increased power losses) How to make shunt-LDO appear resistive ? ~4 chips per module Optimal: Compromise between the two ? 5 Thermal issues • Pixel array: 1.7cm x 1.7cm, ~uniform analog and digital power – Nominal operation power: 1.445W (0.7225 W digital + 0.7225 W Analog) – Maximum: 2.023W – Minimum power: 0W • 0.5A shunt-LDO building block – Shunt (NMOS): 5 transistors of 130µm x 30µm = 650um x 30um (yellow) – LDO (PMOS): 5 transistors of 85µmx30µm = 425um x 30um (red) – Power dissipated in thin top layer of Si (few um). • Shunt-LDO’s at bottom of chip – – – – • Analog: 4 blocks of 0.5A = Max 2A Digital: 4 blocks of 0.5A = Max 2A Interleaved to have better heat distribution (we may even be forced to distribute this within the pixel array in case of serious thermal problems) Heat removal from back side of pixel chip (initially forget about bump bonded sensor on top) – Initial trial with uniform heat removal across whole chip – Cooling tube below shunt-LDOs + cooling tube middle pixel array with some kind of heat distribution layer in-between 6 Chip and Dee Geometry 7/21/2016 7 Odd Dee Even Dee Chip and dee geometry Small Disc • Each Z-layer consists of 2 Dees, one with the 1st and 3rd rings of modules (Odd Dee) and the other with the 2nd and 4th rings (Even Dee) • The Dees are carbon fiber(CF) sandwich structures with CO2 cooling tubes embedded in thermally conductive foam with CF face sheets on either side • Pixel modules are populated on both sides of the Dee to create a hermetic layer • The Dees will also host the LpGBTs and other electrical accessories in high-radius locations Pixel chip HDI Si Sensor 8 Thermophysical properties Thermal Conductivity [W/m-K] Characteristic Length [mm] 148.0 0.100 3.8 0.080 Carbon Fiber Tencate K13C2V 620.0 0.250 Permanent Glue 3M2216 0.2 0.020 2803.0 3.500 Thermal Grease TC5022 4.0 0.050 Tube Stainless steel 13.8 1.025 Conductive Thermal Pathways ROC, Shunt, LDO Silicon Phase Change Glue Laird Film Carbon Foam Allcomp K9 CO2 Refrigerant Temperature = -30oC Heat transfer coefficient = 5000 W/m2-K Power “distribution” on chip (Max power: 2.023W) Nominal power: 1.445W Min power: 0W Pixel array 17 mm X 17mm ~19mm Shunt-LDOs: Worst case Max power: 6.6W Model assumptions • Convert conductivity for thermal grease and glues into a thermal conductance value between contact regions so that glues and films are not modeled as solid objects – hLaird film = 0.0475 W/mm2-K – h3M2216 = 0.0100 W/mm2-K – hTC5022 = 0.0800 W/mm2-K • Use constant heat transfer coefficient and constant temperature for CO2 • Thermophysical properties are homogenous throughout each body and not temperature dependent • All heat is transferred to CO2 refrigerant • One half of a Dee ring will be analyzed • Uniform analog and digital power • LDO, Shunt dissipation does not change with pixel array size (same for 20mm2 and 17mm2 Nominal case: Operation 7/21/2016 12 Nominal case: operation • Pixel array: 2 W*(172/202)=1.445 W – Uniformly distributed over pixel array of 17x17 mm • Analog shunt-LDO: 0.5 W – 4 shunts each burning 0.2 W/4 = 0.05 W – 4 LDOs each burning: 0.3 W/4 = 0.075 W • Digital shunt-LDO: 0.5 W – 4 shunts each burning 0.2 W/4 = 0.05 W – 4 LDOs each burning: 0.3 W/4 = 0.075 W • Total power per chip 2.445 W • Thermal issues (Total power 117.36W): 1. Temperature gradient over array less than ~5oC • Analog front-ends quite temperature sensitive. Acceptable gradient to be verified with circuit simulations) 2. Hottest spot on chip (shunt-LDO) temperature 13 Nominal case: results • Location ΔT [oC] Quarter D 10.5 ROC Pixel area 2.5 LDO, Shunt, Periphery 3.7 • • Max Temperature -18.78oC Min Temperature -29.32oC Resistance due to glues keeps heat on chip increasing temps on chip Temp distribution on pixel array dependent on cooling line position Conclusion: Hottest spot on LDO (Below 1,414oC melting point of silicon), ΔT of pixel array about 2.5oC 7/21/2016 14 Max case (1.5x): Operation 7/21/2016 15 Max case (1.5xnominal power): operation • Pixel array: 2.023 W (When power-up or if mis-configured) – Uniformly distributed over pixel array of 17x17 mm • Analog shunt-LDO: 1.9 W – 4 shunts each burning 0.4 W/4 = 0.1 W – 4 LDOs each burning: 1.5 W/4 = 0.375 W • Digital shunt-LDO: 1.9 W – 4 shunts each burning 0.4 W/4 = 0.05 W – 4 LDOs each burning: 1.5 W/4 = 0.375 W • Total power 5.823 W. • Thermal issues (Power 269.9 W): – Pixel chip does not need to be operational but it must be guaranteed that the chip does not “melt” – Hottest spot on chip (shunt-LDO) temperature 16 Max case (1.5xnominal power): results 2 1 Location ΔT [oC] Quarter D 31.0 ROC Pixel area 6.9 LDO, Shunt, Periphery 17.0 Max Temperature 2.57oC Min Temperature -28.43oC • Temperatures much higher in chip due to higher loads by the shunts and LDOs • Resistance of glues is still an issue • Conclusion: Hottest spot on LDO 2.6oC (Below 1,414oC melting point of silicon), ΔT of pixel array about 2.5oC 7/21/2016 17 Worst case (1.5x): shut-down 7/21/2016 18 Worst case (1.5x): shut-down • Pixel array: 0 W (Pixel array shut-down. Power dissipated by on-chip shunt) – Uniformly distributed over pixel array of 17x17mm • Analog shunt-LDO: 3.3 W – 4 shunts each burning 1.8 W/4 = 0.45 W – 4 LDOs each burning: 1.5 W/4 = 0.375 W • Digital shunt-LDO: 3.3 W – 4 shunts each burning 1.8 W/4 = 0.45 W – 4 LDOs each burning: 1.5 W/4 = 0.375 W • Total power 6.6 W • Thermal issues (413.9 W): – Hottest spots on chip (shunt or LDO) temperature • Chip should not “melt” • Thermal induced bending of module • Other ? 19 Worst case: results 2 1 Location ΔT [oC] Quarter D 44.4 ROC Pixel area 5.1 LDO, Shunt, Periphery 26.0 Max Temperature 16.77oC Min Temperature -27.58oC • Temperatures much higher in chip due to higher loads by the shunts and LDOsTemp distribution on pixel array dependent on cooling line position • Conclusion: Hottest spot on LDO 16.77oC (Below 1,414oC melting point of silicon), ΔT of pixel array about 5.1oC 7/21/2016 20 Scenario: normal > normal X1.5 > shutdown 16.77oC -18.78oC Worst case Nominal case X1.5 Failure would not cause chips to melt Nominal case 7/21/2016 21 Glue and thermal grease effect • Assuming worst case scenario with heat loads • Assuming a perfect world, no glue or thermal resistance from thermal grease • Highest temperature achievable would be 2.74oC (16.77oC if thermal resistance due to glue and grease) due to better temperature distribution 7/21/2016 22 Comments and Conclusions 7/21/2016 23 Comments and pending questions • Chips will not melt even at worst case • Material selection key to thermal dissipation • Resistances due to proper contact (glue and thermal grease) key in removing heat from LDO and shunts; still limited by properties of remaining materials • More detailed analysis needed as chip design and Dee materials are finalized • Thermal analysis on chip will move to Andy Jung ([email protected]) from Purdue University, with Cornell as secondary check 7/21/2016 24 Thank you! 7/21/2016 25 Additional slide: Thermal loads applied Nominal Power Area Heat Flux [W/mm2] Pix array 1.44500 289.00000 0.005000000 Analog Shunts 0.05000 0.01950 2.564102564 Analog LDO 0.07500 0.01275 5.882352941 Digital Shunt 0.05000 0.01950 2.564102564 Digital LDO 0.07500 0.01275 5.882352941 Power Area Heat Flux [W/mm2] Pix array 2.02300 289.00000 0.007000000 Analog Shunts 0.10000 0.01950 5.128205128 Nominal X 1.5 Analog LDO 0.37500 0.01275 29.411764706 Digital Shunt 0.05000 0.01950 2.564102564 Digital LDO 0.37500 0.01275 29.411764706 Worst case CO2 Refrigerant Temperature = -30oC Heat transfer coefficient = 5000 W/m2-K hLaird film = 0.0475 W/mm2-K h3M2216 = 0.0100 W/mm2-K hTC5022 = 0.0800 W/mm2-K Power Area Heat Flux [W/mm2] Pix array 0.00000 289.00000 0.000000000 Analog Shunts 0.45000 0.01950 23.076923077 Analog LDO 0.37500 0.01275 29.411764706 Digital Shunt 0.45000 0.01950 23.076923077 Digital LDO 0.37500 0.01275 29.411764706 7/21/2016 26
© Copyright 2026 Paperzz