Study of bump stresses C Gemme P Netchaeva
Study of bump stresses C. Gemme, P. Netchaeva, L. Rossi, E. Ruscino, F. Vernocchi The aim of this study is to check the effect of stresses on bumps due to thermal cycles in realistic conditions dictated by the mechanical layout. The procedure ôWe simulate the module mechanical structure (carbon-carbon + module + kapton circuit) and we concentrate on Indium bumps. ôWe are interested in “real life” and “unavoidable” cycles [i. e. -20 C, +20 C] and we assume fabrication at room temperature and stress when cold. ôWe want to start and “look” at bumps and their deformation, we then choose a glass tile, measurements will be repeated as soon as “dummy” tiles will be available. ôWe want to maximize the bilaminar effects for a given DT, we then choose a very rigid glue (cyanoacrylate) deposited in a very thin layer. ATLAS Pixel Mechanics meeting at LBNL 13 April 2000 P. Netchaeva, INFN/Genoa 0
Measurement setup (bare module) Tile = 300 mm thick glass with In bumps Chips = bad FE chips with In bumps (550 mm thick). Flipping in Genoa. Measurements at -10 C, 0 C , +20 C and +40 C, T measured with IR thermometer. Measurements at edge of tile (in last double column). X-distance between measured points = 58. 8 mm Displacement measured with microscope. A dozen of cycles [-20 C, +20 C], measured after each cycle in some cases rising to +40 C (90% of time at -20 C, i. e. under stress). Periodically checked bump adhesion ( = measure bump gap with microscope focus): OK ATLAS Pixel Mechanics meeting at LBNL Measurement Rigid glue = cyanoacrylate points x Carbon-carbon 13 April 2000 P. Netchaeva, INFN/Genoa 1
Left edge (low side) Measured at -10 C, +20 C, +40 C the red line represents a fixed position on the chip (recognizable pattern). The distance between the metalization on the glass and this line determines CTE ATLAS Pixel Mechanics meeting at LBNL 13 April 2000 P. Netchaeva, INFN/Genoa 2
Right edge (low side) Measured at 0 C, +20 C, +40 C ATLAS Pixel Mechanics meeting at LBNL 13 April 2000 P. Netchaeva, INFN/Genoa 3
Analysis of results Max displacement measured on one side (for D T = 50 C) is ~7. 5 mm. CTE can be calculated using any pair of temperature measurements, results are all compatible with D(CTE)=4. 8 ± 0. 5 10 -6 C-1(i. e. glass CTE = 4. 0 10 -6 C-1, while silicon is 2. 5 10 -6 C-1; this means that with glass we are exploring an equivalent silicon DT of ~80 C (instead of 50 C as for glass). Bump connectivity can only be controlled by measuring the bump gap. This is constant after each cycle and equal to 8 ± 1 mm (typically ~30 mm when bump is disconnected). Connectivity to be checked with dummy silicon modules (bump chains) after similar cycles. Image focus displaced by ~8 mm. x 100 0 Focused on chip substrate ATLAS Pixel Mechanics meeting at LBNL Focused on under-bump metal 13 April 2000 P. Netchaeva, INFN/Genoa 4
Effect of Kapton circuit (dressed module) After a dozen of cycles of the bare module in about two weeks a kapton circuit has been added. Patterned kapton circuit (50 mm thick, no coverlay) was glued at room temperature with cyanoacrilate. Measured, then cooled to -20 C, then measured again ATLAS Pixel Mechanics meeting at LBNL Kapton circuit cyanoacryl ate 13 April 2000 P. Netchaeva, INFN/Genoa 5
Effect of cooling with kapton Bump detach at 1 st cycle Left +25 C Right +25 C Left -10 C Right 0 C 50 mm apart ATLAS Pixel Mechanics meeting at LBNL 25 mm apart 13 April 2000 P. Netchaeva, INFN/Genoa 6
Back to room temperature All edge modules detach at the 1 st cycle at -20 C. Measuring at room temperature we find chip/substrate distance of 40 mm and 50 mm (left side at -10 C, up and down), 40 mm and 25 mm (right side at 0 C, up and down). The module at 0 C requires ~15 g force applied at the edge of the sensor to bend back to its original shape. One good In bump should resist ~0. 1 g. The module bends back to its original shape when back to room temperature. Left -10 C ATLAS Pixel Mechanics meeting at LBNL Left 25 C 13 April 2000 P. Netchaeva, INFN/Genoa 7
Why kapton circuit so nasty? Kapton circuit contracts considerably with a -40 C D T, it bends the 300 mm thick glass and applies a shear/peeling force to the bumps. The tabulated CTE is 16. 8 10 -6 C-1(Cu), 30 -80 10 -6 C-1(kapton). We can measure the patterned kapton CTE by comparison of thermal expansion between the 300 mm thick glass (CTE=4 10 -6 C-1) and the kapton. The kapton is glued to the glass on one (short) side only. The glass is thermally coupled to a sizeable Cu mass to allow the measurement at ~constant temperature. The kapton will be forced to stay in a plane by a 1 mm thick glass placed on it (but free to move). (glass/kapton)glue thermal grease Close point (a) Far point (b) Cu thermal mass ATLAS Pixel Mechanics meeting at LBNL 13 April 2000 P. Netchaeva, INFN/Genoa 8
Kapton contraction +25 C Even with small magnification one can see the relative shift of the far point (b). The distance between the far point and the glue joint is ~60 mm ATLAS Pixel Mechanics meeting at LBNL 0 C 13 April 2000 P. Netchaeva, INFN/Genoa 9
Comparison with glass (at point b) The lines have been drawn along the pixel columns (metalized dots) and compare glass and kapton contraction. b: +25 C The shift is 60± 5 mm over ~60 mm (i. e. one part in 103 for 25 degrees change). CTE(patterned kapton) =(44± 4) 10 -6 C-1 b: 0 C i. e. the patterned kapton CTE is dominated by the kapton itself. ATLAS Pixel Mechanics meeting at LBNL 13 April 2000 P. Netchaeva, INFN/Genoa 10
What to do? Stress on bumps should be minimized (long term effects), this can be obtained either by : looser bond between kapton and silicon, or smaller CTE substrate material (upilex? ), or stronger FE/sensor bond (e. g. add a glue bond) Bilaminar effect C-C/silicon seems not to be a problem. . We start investigating how to include a glue bond to strengthen the FE/sensor link in the critical edge region. We need a bond able to stand >100 g of force. ATLAS Pixel Mechanics meeting at LBNL 13 April 2000 . Monolithic module P. Netchaeva, INFN/Genoa 11
Trials with glue on glass samples Dow Corning 734 RTV (tens. strength= 16 Kg/cm 2, 321% elongation, CTE~300 10 -6 C-1, er~2. 5, dielectric rigidity~18 KV/mm), and Epotek 353 ND (flex. strength= 750 Kg/cm 2, CTE=56 10 -6 C-1, viscosity = 2000 cps) have been used on glass samples bumped on bad electronics, both glues are fluid and “electronics compatible”. A minimal quantity is applied on two edges (parallel to columns), it enters in between chip and glass by capillary effect. Dow Corning enters <0. 5 mm, Epotek fills almost everything. Dow ATLAS Pixel Mechanics meeting at LBNL Epotek 13 April 2000 P. Netchaeva, INFN/Genoa 12
Electronic effects (Epotek 353) Small amount of Araldit 353 has been put on glass sample and on a single chip with FE_C (GE_C_7). In both cases glue is deposited at the Eo. C logic edge (the only accessible for the single chip once mounted on board). Easy to see where the glue entered in the single chip assembly: large threshold changes Glue enters as expected (~3 -4 mm) Eo. C logic is here ATLAS Pixel Mechanics meeting at LBNL 13 April 2000 P. Netchaeva, INFN/Genoa 13
Other electronics effects Raw data ATLAS Pixel Mechanics meeting at LBNL Noise 13 April 2000 P. Netchaeva, INFN/Genoa 14
Conclusions Reinforcing glue (if any) should not enter between FE and sensor but be confined to periphery of the last chip (not more than last column). This is possible with UV curing glue NEA 123 (tens. strength =245 kg/cm 2; CTE~10 -4 C-1, 60% elongation before breaking, dielectric rigidity= 980 V/25 mm, er=4). NEA 123 Three samples have been glued at periphery with NEA 123 tested to measure the detaching force (~1000 g, fatigue test is ongoing). The capillary effect should be taken into account for the module-to-support glue too. Detailed simulation (including bumps) should be done to explain all described phenomena and guide us in glue choices. ATLAS Pixel Mechanics meeting at LBNL 13 April 2000 P. Netchaeva, INFN/Genoa 15
- Slides: 16