Reliability Studies of SnAgCu BGA Solder Joints on Ni/Cu/Au ...

II. Experimental procedures1. Prototype designThe prototype design and specifications are shown inTable 1. The logic-TEG chips have 400 electrodes for interchipconnections with a 50 µm pitch in an area array at thecenter <strong>of</strong> each chip and 500 external electrodes with a 35 µmpitch with a staggered layout at the periphery <strong>of</strong> the chip. Thememory-TEG chips have 400 electrodes for inter-chipconnections corresponding to the logic electrode layout. Twochips are encapsulated in a fine-pitch <strong>BGA</strong> type package thatmeasures 15 mm × 15 mm and 500 solder balls with a 0.5 mmpitch next to the logic chip. The thickness <strong>of</strong> Ni, Cu and Auis 3.0, 1.5 and 0.05 µm, respectively. 0.300 mm solder ballswere attached to solder pads. The total thickness <strong>of</strong> thepackage including the solder balls is less than 1.0 mm. Thisprototype <strong>of</strong> SMAFTI package and system board wasdesigned to form daisy chain, by which the resistance <strong>of</strong> acomplete circuit can be monitored and recorded.2. Fabrication <strong>of</strong> the prototype <strong>of</strong> SMAFTI packageThe process flow is shown in Fig. 5; a photosensitivepolyimide layer was deposited on a support substrate, and<strong>BGA</strong> pad vias were patterned using photolithography process,in which Cu/Ni <strong>BGA</strong> pad layer was filled. The memory-TEGchips with Sn-Ag electroplated bumps were aligned andbonded to the FTI formed substrate using a flip chip bonder.After reflowing, epoxy based underfill resin which wasdefined as “1 st level underfill resin”, was injected into the gapusing the capillary flow effect, and this resin was cured in anoven. A compression molding process was carried out for theFTI formed silicon wafer. An epoxy based moldingcompound with silica filler was used to encapsulate thememory-TEG chips and FTI surface. After removal <strong>of</strong> thesupport substrate, the surface <strong>of</strong> Cu layer appeared andelectoroless Au was plated on the exposed Cu layer. Thelogic-TEG chips with bumps were aligned and bonded to theFTI. The same type <strong>of</strong> underfill resin used memory-TEGconnection was injected under the logic-TEG chips and cured.Table 1: Specifications <strong>of</strong> the prototype <strong>of</strong> SMAFTI packageDie sizeLogic 5.31 mm x 5.31 mmMemory 7.35 mm x 12.7 mmInter-chip50 µm (area array)connection400 pinExternal electrode 35µm staggered (peripheral)on logic-TEG500 pinPackage typeFine pitch <strong>BGA</strong>Package size15 mm x 15 mm1 st levelunderfill resinType: A, Type: BPin500 pinPitch0.50 mm<strong>BGA</strong> solder SizeΦ0.300 mmjoints Composition Sn-3.0Ag-0.5CuFluxWater solubleType: A, Type: BFig. 5: Fabrication process flowFig. 6: Reflow condition for solder ball attachmentFig. 7: Appearance <strong>of</strong> SMAFTI resin wafer with solder ballsWater soluble flux was printed by screen printer on theNi/Cu/Au <strong>BGA</strong> pad, and Sn-Ag-Cu solder balls were attachedby wafer level solder ball mounter (Athlete FA BM-1000). Amodified screen printing mask was used to place the solderball on the pad. 42,604 balls were attached at one time. Thenthe resin wafer was reflowed in a 7-zone reflow gas-forcedconvectionoven. During reflowing, nitrogen was provided inthe oven to keep the concentration <strong>of</strong> oxygen less than 100ppm. As shown in Fig. 6, reflow peak temperature was 240 ℃and dwell time over 220 ℃ was 58 second. The rate <strong>of</strong>284 2008 Electronic Components and Technology Conference

defects such as missing balls was 170 ppm. The appearance <strong>of</strong>SMAFTI resin wafer with solder balls attached is shown inFig. 7. Water soluble flux was cleaned by de-ionized waterwhile the resin wafer was cut into the single pieces by theautomatic dicing saw.3. Board level assembly and reliability testBefore board level assembly, SMAFTI packages wereaged at 125 ℃ for 10 hours, and stored at 30 ℃/ 70 % for 168hours. Then they were mounted on FR-4 printed circuit boardafter printing <strong>of</strong> no-clean Sn-Ag-Cu solder paste and werereflowed twice in a 12-zone convection oven. As shown inFig. 8, reflow peak temperature was 250 ℃ and dwell timeover 220 ℃ was 46 second. After reflowing, epoxy basedunderfill resin that was defined as “2 nd level underfill resin”was injected between SMAFTI package printed circuit board,and cured in an oven.Board level reliability tests such as thermal cyclic test(TCT), cyclic bending test, and drop test were performed. Inorder to obtain the fatigue life <strong>of</strong> the solder joints, SMAFTIpackage and the printed circuit board were designed to formdaisy chain, by which the resistance <strong>of</strong> a complete circuit canbe monitored and recorded.TCT temperature ranged from -40 °C to 125 °C, and both<strong>of</strong> the soak time at -40 ℃and 125 °C were 10 minuets. Bothramp up and ramp down rates were about 15 °C /min. Thecyclic bending test was under the condition <strong>of</strong> 20 mmdeflectionand 1 second /cycle frequency. Drop height used inthis test was 1,000 mm.Table 2: Board level assembly conditionLayerFR-4 6layerSystemPad structureCu OSP/ Non soldermask definedboardSize 40mm x 110 mmThicknesst = 1.00 mm<strong>Solder</strong> paste<strong>SnAgCu</strong> (no-clean)t = 0.12 mmPre condition 125 0 C/10h+30 0 C/70%RH/168h2 nd levelunderfill resinType:A,Type:BIII. Results and discussionFig. 9: SEM image <strong>of</strong> <strong>BGA</strong> ball joint (×300)CuSnAuAgFig. 10: EPMA image <strong>of</strong> <strong>BGA</strong> ball joint (×300)No.3 <strong>Solder</strong>No.2 (Cu,Ni) 6Sn 5No.1 Cu 6Sn 5Fig. 8: Reflow condition for board level assembly2µmNo.4 Ni layerFig. 11: SEM image <strong>of</strong> IMC layer (×5000)285 2008 Electronic Components and Technology Conference

Cu Sn No.1 Ni,CuSnNo.3Fig. 12: Results <strong>of</strong> EDX analysis (×5000)No.2No.41. Cross section <strong>of</strong> solder jointsCross sectioning analysis was performed on the solderjoint <strong>of</strong> SMAFTI package. Sn-Ag-Cu solder was attached onthe Ni/Cu/Au pad shown in Fig. 3. Fig. 9 shows a SEM image<strong>of</strong> a cross-section <strong>of</strong> a <strong>BGA</strong> solder joint <strong>of</strong> SMAFTI packageafter fabrication. The distributions <strong>of</strong> Cu, Sn, Au and Agobserved by EPMA can be seen in Fig. 10. The SEM image<strong>of</strong> IMC (Intermetallic compound) layer is shown in Fig. 11,whose EDX analysis results are described in Fig. 12.There was only Ni detected as shown in No.4 <strong>of</strong> Fig. 12, sothis layer in No.4 <strong>of</strong> Fig. 11 was the Ni layer <strong>of</strong> the Ni/Cu/Aupad. There was only Sn detected as shown in Fig. 12, whichshowed No.3 in Fig. 11to be the solder.During reflowing, almost all Au whose thickness was 0.5µm was dissolved into the solder uniformly as shown in Fig.10. The thickness <strong>of</strong> Cu layer was about1.5 µm, and wasconsumed entirely to form Cu 6 Sn 5 layer, whose meanthickness became 2.1µm after reflowing. And there was no Nior Au detected in this IMC layer as shown in No.1 <strong>of</strong> Fig. 12.There was another IMC layer between Cu 6 Sn 5 and Nilayer, which consisted <strong>of</strong> Cu, Ni and Sn as shown in No.2 <strong>of</strong>Fig .12. During reflowing, after the consumption <strong>of</strong> Au andCu, Ni layer below Cu formed thin (Cu,Ni) 6 Sn 5 layer[8,10,11]. Since Cu layer was situated between solder and Ni,this layer decreased the degradation <strong>of</strong> Ni layer during liquidphase reaction with the solder.When the Sn-Ag-Cu solder is attached on the Ni/Au padas shown in Fig. 4, (Cu,Ni,Au) 6 Sn 5 forms as IMC layer.When the solder is attached on the Cu-OSP or the Cu-SOPpad, Cu 6 Sn 5 forms. In this study, the IMC layer between theSn-Ag-Cu solder and the Ni/Cu/Au pad was investigated, andfound to be form Cu 6 Sn 5 and thin (Cu,Ni) 6 Sn 5 .2. The results <strong>of</strong> impact shear studyThe solder joint integrity <strong>of</strong> SMAFTI against impact loadwas investigated by way <strong>of</strong> the impact shear testing. Theresults were compared with those <strong>of</strong> FP<strong>BGA</strong> whose pad metalstructure was Ni/Au, Cu-OSP and Cu-SOP. At the shearspeed <strong>of</strong> 0.1 mm/s, the bond tester (Dage Series 4000) wasused, and the impact shear tester (Dage 4000-HS) wasintroduced to quantify the solder joint integrity for impactload as shown in Fig. 13, whose test speed ranged from 10mm/s to 4000 mm/s [7]. Not only shear load but also solderjoint energy was plotted, and discussed which was superior topresent the solder joint integrity. 30 points <strong>of</strong> solder jointswith Ni/Au and Ni/Cu/Au pad were tested for impact shearNiSnstudy in each condition, while 20 points <strong>of</strong> solder joints withCu-SOP and Cu-SOP pad were testedThe results <strong>of</strong> the impact shear load as a function <strong>of</strong> shearspeed are plotted in Fig. 14. The horizontal axis is the shearspeed that is showed in logarithm, and the vertical axis is theshear load. Although the test conditions were the same amongthe pad surface finishes, the plots in horizontal axis arestaggered intentionally so that they can be seen clearly. Thetop <strong>of</strong> the error bar means the maximum value, while thebottom means the minimum.The shear strength change <strong>of</strong> solder joints with Ni/Auagainst the shear speed was different from those with Cu-OSP, Cu-SOP and Ni/Cu/Au. The mean strengths <strong>of</strong> solderjoints with Ni/Au were about 4 N at all speeds, while thehigher the shear rate became, the higher the shear strengthwith Cu-OSP, Cu-SOP and Ni/Cu/Au grew. The defection <strong>of</strong>shear strength with Ni/Au was the largest <strong>of</strong> all <strong>of</strong> these padmetal surface finishes, and the minimum shear strength withNi/Au was the weakest <strong>of</strong> all at the speed <strong>of</strong> 10 mm/s and100mm/s. On the other hand, although the mean shear loadswith Ni/Cu/Au seemed to be lower than those <strong>of</strong> the otherpads, the deflections were the smallest <strong>of</strong> all, which indicatesthat solder joints with Ni/Cu/Au have good durability againstimpact loads.The changes <strong>of</strong> the solder joint energy against the shearspeed are plotted in Fig. 15, where the horizontal axis is theshear speed showed in logarithm, and the vertical axis is theshear load. The plots in horizontal axis are staggered, too.As the shear speed grew faster, the solder joint energieswith Ni/Au, Cu-SOP and Cu-SOP decreased gradually. Andthe solder joint energies with Ni/Cu/Au showed the sameinclinations as the others. On the whole, the mean solder jointenergies with Ni/Au seemed to be the smallest <strong>of</strong> all.However, the defections were the smallest and the minimumsolder joint energies seemed to be the highest <strong>of</strong> all the padmetal surface finishes, which indicates the solder joints withNi/Cu/Au have good durability against impact loads.From the results shown in Fig. 14 and Fig. 15, it seems thatthe solder joints with Cu 6 Sn 5 IMC layer have better jointreliabilities against the impact load than those with(Cu,Ni,Au) 6 Sn 5 . It indicates that the concentration <strong>of</strong> Cu thatdiffuses into the solder changes the hardness <strong>of</strong> the solderbulk, and the solder joint integrity <strong>of</strong> IMC layer is better thanthat <strong>of</strong> (Cu,Ni,Au) 6 Sn 5 . And the small amount <strong>of</strong> Au into thesolder may change the solder joint integrity <strong>of</strong> Ni/Cu/Aubetter than Cu-OSP and Cu-SOP.The relationship between the shear load and the solder jointenergy is considered and described in Fig. 16, where thehorizontal axis is the displacement <strong>of</strong> the shear tool, and thevertical axis is the shear load. F bulk and F interfacial is defined asthe shear loads <strong>of</strong> bulk solder failure mode and interfacialfailure mode, respectively. E bulk and E interfacial are defined asthe solder joint energies which were the integrals <strong>of</strong> shearload at the length, l.E = ∫F(l) dl286 2008 Electronic Components and Technology Conference

In Fig. 16, F interfacial is as high as F bulk , while E interfacial ismuch smaller than E bulk . It indicates that solder joint energycan describe the solder joint integrity more clearly than shearload.15F interfacialF bulkHeight:50µmShear Speed:10 – 4000mm/sDisplacementImpact shear load [N]105E interfacialF(l):Shearload at lFig. 13: The image <strong>of</strong> impact shear testingE bulkShear Load [N]<strong>Solder</strong> Joint Energy [mJ]8765432102.52.01.51.00.50.1mm/s10mm/s100mm/s1000mm/sNi/AuCu-OSPCu-SOPNi/Cu/Au10 -1 10 0 10 1 10 2 10 3 10 4Shear Speed [mm/s]Fig. 14: Shear load versus shear speedNi/AuCu-OSPCu-SOP4000mm/sNi/Cu/Au10 mm/s 100 mm/s 1000 mm/s 4000 mm/s0.010 1 10 2 10 3 10 4Shear Speed [mm/s]Fig. 15: <strong>Solder</strong> joint energy versus shear speed00 200 400 l600 800 1000 1200Displacement <strong>of</strong> the shear tool [um]Fig. 16: The relationship between shear loadand solder joint energy3. Thermal warpage and solder surface condition<strong>BGA</strong> type package sometimes causes de-wetting issue <strong>of</strong>solder joints at board level assembly process. There are someroot causes <strong>of</strong> this failure induced by the design <strong>of</strong> thepackage. One is the thermal warpage change <strong>of</strong> <strong>BGA</strong> packageduring reflowing. Another one is the oxidation or thepollution <strong>of</strong> solder ball surface [12]. In order to care thisfailure, thermal warpage and solder surface condition wereevaluated.As SMAFTI package shown in Fig. 2 consists <strong>of</strong> Si dice,molded resin, and very thin FTI layer, the main factor thatrules the warpage is the thermal characteristic caused by thesematerials. Fig. 17 shows the thermal warpage change againstthe temperature <strong>of</strong> SMAFTI package, which was measuredusing Shadow moiré method. According to the result, thewarpage change during reflowing was less than 20 µm, andthe maximum warpage was 47µm, which are small enoughcompared with the height <strong>of</strong> solder ball. Higher volume <strong>of</strong> theSi and lower volumes <strong>of</strong> interposer and molded resin thanthose <strong>of</strong> standard FP<strong>BGA</strong> seems to make the warpage smaller.245 0 CTemperature [K]38 0 C160 0 C29μm220 0 CDefinition <strong>of</strong> warpage41μm60 0 C47μm40μm47μmReflow time [s]Fig. 17: Thermal warpage change <strong>of</strong> SMAFTI package-+287 2008 Electronic Components and Technology Conference

Relative Concentration [%]Thickness <strong>of</strong> Oxidation Layer [nm]CPeak <strong>of</strong> OOSnThickness <strong>of</strong> Oxidization Layer½peak <strong>of</strong> OSputtering Time [min]Fig. 18: Depth pr<strong>of</strong>ile <strong>of</strong> Auger electron microscopy15105Flux AFlux BN = 3 points00 48 96 144 192 240 288 336Exposure Time at 30 o /70%RH [h]Fig. 19: Thickness change <strong>of</strong> oxidation layer at 30°C /70 %2. Board level reliability testAs mentioned in previous section, 1 st level underfill resintype “a” and “b” were used to encapsulate flip chip solderjoints in the SMAFTI package. Another underfill resin type“A” and “B” were used to encapsulate the <strong>BGA</strong> solder jointsbetween the package and the printed circuit board. Wedefined this underfill resin as “2 nd level underfill resin”.Therefore, there were 4 types <strong>of</strong> specimens evaluated in thisstudy changing underfill resins. The results <strong>of</strong> the board levelreliability test are shown in Table 3.The results <strong>of</strong> thermal cyclic test are described in Fig. 19.For the test vehicle with 1 st level underfill resin “a”, and 2 ndlevel underfill resin “A”, the first failure occurred at 1006cycles, and 0.1 % Weibull life <strong>of</strong> thermal cycle was 296cycles. For the vehicle with 1 st level underfill resin “a” and 2 ndlevel underfill resin “B”, the first failure occurred at 955cycles, and 0.1% Weiblull life was 51 cycles. For the vehiclewith 1 st level underfill resin “b” and 2 nd level underfill resin“A”, the first failure and 0.1 % life was 931 and 306 cycles,respectively, while for the vehicle with 1 st level underfill resin“b” and 2 nd level underfill resin “B”, 1406 and 989 cycles,respectively. These results indicate that the material properties<strong>of</strong> both <strong>of</strong> 1 st level underfill resin and 2 nd level underfill resinare very important for thermal cycle reliability, and thatoptimized resins can ensure sufficient reliability for this newinterconnection technology.Table 3: Results <strong>of</strong> board level reliability test <strong>of</strong> SMAFTIUnderfillTCT Cyclic DropResin[cycle] Bending TestYield1 st 2 nd FirstTest (Min) (Min)Ffailure (0.1%) [cycle] [cycle]a A Initial 1006 296 >20000 >10a B failure 955 51 >20000 >10b A 0/108 931 306 >20000 >10b B pcs 1406 989 >20000 >10The surface <strong>of</strong> the solder balls were controlled by flux andwashing condition. To prevent the solder surface pollution,the water used in the washing machine was not recycled, butdrained directly.The water soluble flux was chosen to reduce the thickness<strong>of</strong> oxidation layer. The information <strong>of</strong> the solder surface wasobtained by Auger electron microscope. Fig. 18 shows thedepth pr<strong>of</strong>ile <strong>of</strong> the solder surface, where the horizontal axisis the sputtering time and the vertical axis is the relativeconcentration <strong>of</strong> each element. The sputtering rate convertedto SiO 2 was 11nm/min. As shown in Fig. 18, 1/2 peak methodwas applied to quantify the thickness <strong>of</strong> oxidation layer. Theevaluation result was described in Fig. 19, where thehorizontal axis is the exposure time conditioned at 30 °C and70 %RH humidity in a humidity chamber. Using flux type: B,the thickness <strong>of</strong> oxidation layer could be kept thinner, andthere was no initial failure during assembly, which is shownin Table 3.F(t)99%95%90%60%50%40%10%5%1%0.1%Weibull Plot1stUF:type"a"/2ndUF:type"A"1stUF:type"a"/2ndUF:type"B"1stUF:type"b"/2ndUF:type"A"1stUF:type"b"/2ndUF:type"B"1 10 100 1000 10000cycleFig. 20: Weibull distributions for temperature cycledcomponents288 2008 Electronic Components and Technology Conference

OpenMemory-TEGLogic-TEGPackage SideBoard Side100µmFig. 21: Failure location and mode <strong>of</strong> solder jointsThe failure mode was a crack in solder joints. The location<strong>of</strong> the solder joint failure is shown in Fig. 20, where the redpins in the Figure show the cracks. All <strong>of</strong> the cracks occurredwithin the area <strong>of</strong> memory-TEG. It is obvious that duringTCT test, solder joints under the memory chip have muchstress.The results <strong>of</strong> cyclic bending and drop tests were shown inTable 3. The minimum life <strong>of</strong> cyclic bending test and droptest were more than 20,000 cycle and 10 cycles, respectively,whatever the type <strong>of</strong> the underfill resin was used.Accordingly, if the suitable underfill resins are chosen, thefeasibility <strong>of</strong> the fundamental reliability <strong>of</strong> the structure issuccessfully demonstrated to meet production qualityrequirements.IV. ConclusionsA novel <strong>BGA</strong> pad metal structure, Ni/Cu/Au for waferlevel3-D packaging technology, SMAFTI was proposed andits reliability was investigated.The IMC layer between the Sn-Ag-Cu solder ball and theNi/Cu/Au pad formed Cu 6 Sn 5 and thin (Cu,Ni) 6 Sn 5 , where Audissolved into the solder entirely.The solder joint integrity <strong>of</strong> SMAFTI against impact loadwas investigated by way <strong>of</strong> the impact shear testing. Theresults were compared with those <strong>of</strong> FP<strong>BGA</strong> whose pad metalstructure was Ni/Au, Cu-OSP and Cu-SOP. On the whole, thesolder joints with Ni/Cu/Au have good durability againstimpact loads.The warpage change during reflowing was less than 20µm, and the maximum warpage was 47µm, which are smallenough compared with the height <strong>of</strong> solder ball. Usingsuitable flux, the thickness <strong>of</strong> oxidation layer could be keptthinner, and there was no initial failure during assembly.Board level reliability tests such as thermal cyclic test(TCT), bending cyclic test, and drop test were performed. The0.1 % Weibull life <strong>of</strong> thermal cycle was 989 cycles, and thelife cycles <strong>of</strong> more than 20,000 and 10cycles were obtainedfor cyclic bending and drop tests, respectively.AcknowledgmentsThe authors would like to thank Katsuyuki Abe, HajimeYanase, and Junko Yamaki who helped this study. Specialthanks to Yukihiko Kato and Yoshiharu Fujimori for supportin wafer level assembly.References1. Kurita, Y., Soejima, K., et al, “A Novel “SMAFTI”Package for Inter-Chip Wide-Band Data Transfer ”, Proc56 th Electronic Components and Technology Conf, pp.170-179.2. Kurita, Y., Soejima, K., et al, “A Novel “SMAFTI”Package for Inter-Chip Wide-Band Data Transfer ”, Proc57 th Electronic Components and Technology Conf, pp.821-829.3. Ezaki, T., Kondo, K., et al, “A 160Gb/s Interface DesignConfiguration for Multichip LSI,” Proc. <strong>of</strong> IEEEInternational Solid-State Circuits Conference (ISSCC2004), San Francisco, CA, Feb. 2004, pp.140–141, Vol.1.4. Kurita, Y., Maeda, M., et al, “Development <strong>of</strong> COC(Chip-On-Chip) Interconnection Technology,” Proc. <strong>of</strong>12 th Micro Electronics Symposium (MES 2002), Osaka,Japan, Oct. 2002, pp. 51–54.5. Kurita, Y., Soejima, K., et al, “Development <strong>of</strong> High-Density Inter-Chip-Connection Structure Package,” Proc.<strong>of</strong> 15 th Micro Electronics Symposium (MES 2005), Osaka,Japan, Oct. 2005, pp. 189–192.6. Kawashiro, F., Kurita, Y., et al, “Development <strong>of</strong> <strong>BGA</strong>Attach Process with High Density SiP Technology“SMAFTI”,” 13 th Symposium on Micro jointing andAssembly Technology in Electronics (MATE 2007),Yokohama, Japan, Feb. 2007, pp. 49–54.7. Newman, K., “<strong>BGA</strong> Brittle Fracture – Alternative <strong>Solder</strong>Joint Integrity Test Methods”, Proc 55 th ElectronicComponents and Technology Conf, pp. 170-179.8. Zeng, K., and Tu, K. N., “Six cases <strong>of</strong> reliability study <strong>of</strong>Pb-free solder joints in electronic packaging technology,”Materials Science and Engineering, R38 (2002), pp.55-105.9. Amagai, M., Toyoda, Y., et al, “High Drop Test<strong>Reliability</strong> : Lead-free <strong>Solder</strong>s ”, Proc 54 th ElectronicComponents and Technology Conf, pp. 1304-1309.10. Snugovsky, L., Snugovsky, P., et al, “Phase equilibria inSn rich corner <strong>of</strong> Cu-Ni-Sn system,” Material science andtechnology, vol. 22, No. 8(2006), pp.899-90211. Jurenka, C., Kim, J.Y. et al., “Effect <strong>of</strong> the Cu Thicknesson the Stability <strong>of</strong> a Ni/Cu Bilayer UBM <strong>of</strong> Lead FreeMicrobumps During Liquid and Solid State Aging”, 2005Electronic Components and Technology Conference,pp.89-93.12. Grigoriev, A. et al, “Surface oxidation <strong>of</strong> liquid Sn,”Surface Science , Vol. 575 (2005), pp.223-232289 2008 Electronic Components and Technology Conference