OUTLINE Inputs to the Tsimulation chips boards crate

OUTLINE • Inputs to the T-simulation: chips, boards, crate mechanics and fan • T-simulation performed into two steps: 1. On large scale: simple AM chip model, but full crate sim, AMB_V 2 - Results 2. On small, deteailed scale: AM 06 package detailed model inserted, AMB_V 3, but simplified crate simulation- Results • Future cooling tests plan and conclusions

Detailed T-simulation (IMEC) for the new AMBSLP to know T on silicon! (a) Optimize chip package and boards dissipation (b) Estimate the temperature in the chip silicon knowing T just outside the package ALL details must be taken into account from package design Substrate RDL to crate, PS, fan design Tall DC-DC converters in the middle Die size Die underfill Heat Sink AMB with four LAMBs model LAMB model Crate & fan model 2

DC-DC 1, 2 V Large Artix FPGAs DC-DC 2, 5 V DC-DC 1 V NOVEMBER 2014 SIMULATION Inputs - AMB_V 2 3

Air flow in the crate simulation Not efficiently cooled slots Bad slots 4

Next to Worse AM pos 5

10 m/s Simulated As uniform 4 m/s A B C D C’ B’ A’ Simulate the Worse case (blue) with Max=3, 2 & 4 m/s 6

AMB_V 2 Full crate sim Simple AM 06 Lower air flux 7

94. 5 o Inside the red box: max T = 94. 5 o Inside the black box: max T = 79 o max in the Bad slots with red arrows (on the left of the box) On the right part of the black box (good slots) T= 61 o, 67 o, 72 o, 73 o, 70 o, 69 o, 75 o and 64 o These are temperatures in good slots where the 8 DC-DC is far.

NEW T-SIMULATION March 2015 1. Improving the AMBSLP layout 2. Detailed model for AM package (see next slides) AMB_V 2 AMB_V 3 3. New inputs to the simulation TOT= 247 W 9

10

11

AM chip + PCB piece With air flow (4 m/s) Results 39, 75 Results 12

55 C 13

99, 4 Worse slot 99 degree on Die is acceptable 14

88 Near to worse-slot 88 degree on Die is acceptable 15

3 slots are even worse than worse simulated case CPU position + 2 SSB Other 3 slots have similar behaviour as worse cases shown by T-sim 2 SSB + 1 AM Worse case shown Max T ~99 ± 5 degrees, acceptable However it is a conservative estimate: 1. Average flux larger than the simulated one 2. The simulation does not allow the x component of air velocity 3. Lowering the fan with respect the crate more uniform flux 16

• The AM 06 package has very low resistance: die T ~ package -PCB T. • Conservative simulation predicts TMax AM 06 die ~99 o ok • Lowering the fan or using a diffusor we can have a more uniform flux of air, uniform temperature, and lower TMax • We will do new cooling tests to see if the T differences between slots detected by the T-sim correspond to the observed differences • We will measure the temperatures for different setups of the free space available in the rack. 17