Warpage Simulation And DOE Analysis With Application In Package-on .

Table 4: Six EMC materials for PoP bottom packageEMCE@25 re 10: Quarter FE model of 15x15mm PoP bottomTable 5: Parameters and naming convention used insimulation and experimentNaming ConventionDie SizeDie ThicknessB: 8.3x8.3mmB: 100µmA DS: 4.4x4.1mmS: 50µmE.g. C-B-S represents a leg using mold compound C, die size of8.3x8.3mm and die thickness of 50µmMold Cpd1.50E-011.00E-015.00E-020.00E 0E-01Simulation (mm)Measurement (mm)Figure 11: Warpage Correlation (Room TemperatureWarpage)1.50E-011.00E-015.00E-020.00E 0E-01-1.50E-01Simulation (mm)Measurement (mm)Figure 12: Warpage Correlation (Reflow TemperatureWarpage)4. DOE Analysis of PoPFigure 13 is a schematic picture of the bi-materialcantilever beam. Based on Timoshenko beam theory, theclose-form mathematical solution to the warpage of thisbi-material cantilever beam under temperature change isEquation 2. With this equation, one can easily calculatethe impact of various material and dimensional changeson warpage. However, actual package is far morecomplex than a bi-material beam. Therefore, it isimpossible to quantify the impact of design and materialimpact based on close-form theoretical equation.DOE analysis, based on different sampling methodssuch as full factorial, Taguchi array, Box-Behnken,Central Composite Design and Latin Hypercube, providesa systematic way of data analysis using the results of acertain number of experimental legs. Coupled withcomputer simulation, DOE analysis can quickly help us toquantify the relationship between input parameters andoutput responses and identify how and how much thechange in inputs affect the output. There is a number ofcommercial software available in the market forsimulation-based DOE analysis such as Modelfrontier,Optimus and VisualDoC. But the current study Mechanical, EXCEL and JMP software.ANSYS/Mechanical is the FEA simulation code, where ashort APDL code is programmed to execute all the legsrequired by a DOE sampling method such as full factorialused in this paper. It should be noted that althoughANSYS/Mechanical itself contains a full factorial DOEanalysis function, such function is limited to 7 inputparameters and 2 level analysis only. Therefore, it isimpossible to put a medium-level setting of a parameter tosee whether there is any non-linear effect and to studymore than 7 DOE parameters. EXCEL is programmedusing macro to import the result files generatedsystematically from ANSYS/Mechanical. The dataimported by EXCEL is then easily copied and pasted intoJMP software for full factorial DOE analysis. JMP itselfis a powerful statistical analysis and DOE analysis toolused widely in manufacturing and quality management.With the combination of above three software, noadditional cost is needed to purchase specialized softwareto perform DOE analysis. It should also be highlightedthat the DOE analysis is not confined to full factorialsampling method only. Instead, every DOE samplingmethod supported by JMP can be easily programmed inANSY APDL to perform the analysis.Previously studied PoP is analyzed with this method.A full factorial analysis with 6 parameters as listed inTable 6 is performed. Good fit is achieved betweensimulation data and DOE predicted data as shown inFigure 14. Pareto plot that can tell us the quantitativeeffect of each DOE parameter is shown in Figure 15. It isseen that CTE of EMC have the most significant impacton warpage and increase in EMC CTE will shift thewarpage curve upwards. The descending order of otherparameters is die size, die thickness, EMC modulus and

CTE of substrate. From Figure 16, we can see that thereis no significant interaction among the six DOEparameters. Apart from sensitivity analysis, anotheradvantage of DOE analysis is that a polynomialregression model can also be obtained to replacesimulation for warpage calculation. This polynomial isuseful especially in the early stage of packagedevelopment as design and material inputs can just beplugged in to quickly calculate the warpage withoutactual simulation. Seven EMCs whose material propertiesfall in the DOE analysis range and are listed in Table 7are plugged into actual ANSYS simulation and also theapproximation model. We can see from Figure 17 that theapproximation model can generate almost the sameresults as actual simulations. With help of theapproximation model, a plot showing the impact of EMCselection on warpage as shown in Figure 18 can bequickly drawn. This plot is helpful in EMC selection.Engineers can quickly tell from the plot which EMC cangive better warpage performance.Figure 15: Pareto plot of different design and materialparametersFigure 13: Bi-layer cantilever beam modelt t1 t 2warpage L2[q E1E2p t1t23(1 p) 2 (α 2 α 1 )(T T0 )4t 3(1 p) 2 (1 pq )( p 2 1 / pq )](Eq.2)Figure 16: Interaction plot of different design andmaterial parametersTable 7: EMC properties of E LTable 6: DOE parameters for PoP bottom packageAve CTEof Core(ppm/K)12ppm/K14ppm/K16ppm/KRTModulus e CTEof EMCppm/K501004.128.32132435RomTemp Eof EMC(MPa)200002600032000EMCE@25 on vs. Approximation Model from DOE AnalysisWarpage (mm)-0.06-0.08-0.1-0.12-0.14-0.16From Package Center to CornerFigure 14: Simulation vs. Approximation 3370601212.-0.041.-0.0200H SimulatedH PredictedI SimulatedI PredictedJ SimulatedJ PredictedK SimulatedK PredictedL SimulatedL PredictedM SimulatedM PredictedN SimulatedN PredictedFigure 17: Comparison of simulation results and thosecalculated from derived approximation model

Table 8: Details of the 15x15mm PoP top package understudyWarpage @ Pkg Corner (mm)0.20.150.10.050-0.0513 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35-0.1PoPTopPackage-0.15-0.2Die-0.25EMC CTE (ppm/k)EMC Modulus: 20000, Sub CTE: 14EMC Modulus: 32000, Sub CTE: 14EMC Modulus: 26000, Sub CTE: 14DAFigure 18: EMC properties on warpageA similar DOE analysis is also conducted for PoP toppackage of which details are listed in Table 7. However, itis found that the obtained approximation model givespoor fit as shown in Figure 19. It is therefore very likelythat some nonlinear effect exists as certain parameterchanges from low level to medium level and to high level.Further scrutiny also confirmed that there are some stronginteraction effects among mold cap thickness, diesize/thickness and substrate core thickness as shown inFigure 20. With those nonlinear effects and stronginteractions, special attention must be paid when doingdata analysis. The approximation model derived fromsuch case should not be used as the quick calculation toolany more as large deviation is obvious. Therefore, ifapproximation model is to be derived, it’d better to havedifferent DOE analysis for each combination of mold capthickness, die size/thickness, and core thickness or adifferent DOE sampling method should be tried.Data analysis to scrutinize the die size effect isperformed for the two nominal PoP top package designshighlighted in blue in Table 8. Warpage vs. Die Size plotis given in Figure 21 for die thickness of 50µm and100µm. Contrary to previous finding on PoP bottompackage where increase in die size will indirectly reduceEMC volume and shift the warpage curve downwards, itis found this time that increase in die size will further shiftthe warpage curve upwards instead of downwards. This isan important finding and is also later confirmed byphysical measurement of engineering samples. Onepossible explanation is, as both mold cap and substratecore become so thin that die starts to play the dominantrole in determination of warpage. Since silicon die is avery stiff material, increase in die size will make it moredifficult to shift the warpage curve downwards.As PoP top package is going for thinner mold cap andsubstrate core to redude the overall PoP height, attentionmust be paid when specifying the design guideline onwarpage reduction. We cannot use the same designguidelines that related to die size/thickness in both PoPtop and bottom package. A detailed DOE analysis shouldbe conducted each time when developing a new PoP toppackage.EMCDimension opperMaterial Properties1520.060.110.160.210.012Figure 19: Simulation vs. Approximation modelFigure 20: Interaction plot of different design andmaterial parameters

Nominal PoP Top Package Design (Die Thickness: 0.05mm)Warpage (mm)0.250.20.150.10.050Die Size4x4mm8x8mm12x12mm(A)Nominal PoP Top Package Design (Die Thickness: 0.1mm)Warpage (mm)0.250.20.150.10.050Die Size4x4mm8x8mmthe use of DOE analysis, impact of a series of design andmaterial parameters on warpage can be efficientlyanalyzed. The combination of ANSYS/Mechanical,EXCEL and JMP plus some programming providespowerful and low-cost tool for simulation-based DOEanalysis. Design and material selection guidelines can beobtained from such simulation-based DOE analysis.Warpage calculator based on approximation modelderived from simulation-based DOE analysis gives quickand reasonably accurate estimates of warpage.12x12mm(B)Figure 21: Plot showing that die size increase results inwarpage curve going further upwards5. Case StudyIn this case study, a new stack-die PoP top package asillustrated in Figure 22 failed end-of-line visualmechanical inspection with excessive crying modewarpage as shown in Figure 23. Severaldesign/layout/material changes were proposed in an effortto reduce the warpage or in other words to shift thewarpage curve upwards from excessive negative to lessnegative or slightly positive. However, the physical DOEmatrix was too large and costly to run. Therefore, FEAsimulation was requested to reduce the physical DOEmatrix.Due to half symmetry of the package structure, 3Dhalf model as shown in Figure 24 was setup. By using theestablished modeling method and DOE analysis, it wasquickly found that while design/layout changes do notaffect the warpage too much, switching to certain moldcompound material can significantly reduce the warpage.The five mold compound candidates are listed in Table 9.From Figure 25, we can see that the use of moldcompound S, which has the largest average CTE, cansignificantly shift the warpage curve upwards or reducethe crying mode warpage. Subsequent physical DOE trialrun with the simulation-selected mold compoundconfirmed the simulation finding. Therefore moldcompound S is used as the standard BOM for PoP toppackage.6. Summaries and ConclusionValidated FEA simulation for warpage is a valuabletool in package development and failure analysis. With7. Future WorkMaterial properties obtained from different supplierswere measured using different methods. It is thereforebetter to perform material characterization in-house usinga unified method. Cure shrinkage is an important factoraffecting warpage, future work will also performcharacterization work on cure shrinkage of different EMCfrom different suppliers. With detailed in-house materialtesting data and cure shrinkage data, current warpagesimulation should be progressing more towards aquantitative method.AcknowledgementsThe authors would like to thank UTAC managementteam for their support on this project. The engineeringsample build work by Ravi, Kriangsak, Danny and Bimboare greatly appreciated. The review work by UTAC paperreview panel is also acknowledged.Figure 22: Schematic cross-section picture of current14x14mm PoP top packageFigure 23: FA picture showing the large crying modewarpageTable 9: Details of five EMC for warpage problemsolvingEMCOPQRSE@ 0220882α1α2Tg89911133236363954135135143143150

Figure 24: Quarter FE model of 14x14mm PoP topWarpage 9.764247.89395.89694544.4841.2.-2.00E-02900.00E 01Along Package DiagnalFigure 25: Plot showing the improvement of crying modewarpage with the use of mold compound SReferences1. Akito Yoshida et al, “A Study on Package StackingProcess for Package-on-Package (PoP)”, Proceedingsof ECTC2006, pp.825-830.2. A. Mertol, “Application of the Taguchi Method on theRobust Design of Molded 225 Plastic Ball Grid ArrayPackages”, IEEE Trans. Comp. Packag., Manufact.Technol. B, vol. 18, pp. 734-743, Nov. 19953. A. Mertol, “Optimization of High Pin Count Cavity-upEnhanced Plastic Ball Grid Array (EPBGA) Packagesfor Robust Design”, IEEE Trans. Comp. Packag.,Manufact. Technol. B, vol. 20, pp. 376-388, Nov.19974. A. Mertol, “Application of the Taguchi Method toChip Scale Package (CSP) Design”, IEEE Trans. Adv.Packag., vol. 23, 2000, pp. 266-2765. A. Dasgupta, M. G. Pecht, B. Mathieu, “Design-ofexperiment Methods for Computational ParametricStudies in Electronic Packaging, Finite Element inAnalysis and Design”, 30 (1998), pp.125-1466. G. Q. Zhang, J. H. J. Janssen, L. J. Ernst, J. Bisschop,Z. N. Liang, F. Kuper, and R. L. gofElectronic Packaging Using Philips’ OptimizationStrategy”, IMAPS2000, USA7. S. Stoyanov, C. Balley and M. Cross, “OptimizationModeling for Flip Chip Solder Joint Reliability,Soldering & Surface Mount Technology”, 14/1, 2002,pp. 49-588. B. Vandevelde, E. Beyne, G. Q. Zhang, Jo F. J. M.Caers, D. Vandepitte and M. Baelmans, “SolderParameter Sensitivity for CSP Life-Time PredictionUsing Simulation-Based Optimization Method”, IEEEtrans. on Electronics Packag. Manufact., vol. 25, 4,October 20029. B. Vandevelde, E. Beyne, G. Q. Zhang, Jo Caers, D.Vandepitte, M. Baelmans, “Parameterized Modelingof Thermo-mechanical Reliability for CSPAssemblies”, Journal of Electronic Packaging, vol.125, December 200310. W. D. van Driel, G. Q. Zhang, J. H. J. Janssen, L. J.Ernst, “Reponse Surface Modeling for NonlinearPackaging Stresses”, Journal of Electronic Packaging,vol. 125, December 200311. W. D. van Driel, G. Q. Zhang, J. W. C. de Vries, M.Jansen, L. J. Ernst, “Virtual Prototyping andQualification of Board Level Assembly”, Proceedingsof EPTC2003, December 200312. S. G. Jagarkal, M. M. Hossain, D. Agonafer, M. Lulu,S. Reh, “Design Optimization and Reliability of PWBLevel Electronic Package”, Proceedings ofITHERM’04, June, 2004 Las Vegas, USA13. C. C. Lee, S. M. Chang, K. N. Chiang, “Design ofDouble Layer WLCSP Using DOE with FactorialAnalysis Technology”, Proceedings of EPTC2004,December 200414. A. A. Giunta, “Use of Data Sampling, SurrogateModels, and Numerical Optimization in EngineeringDesign”, AIAA 2002-0538, American Institute ofAeronautics and Astronautics15. Wei Sun et al, “Simulation-Based DesignOptimization of Solder Joint Reliability of WaferLevel Copper Column Interconnectsz”, Proceedingsof ECTC2006, pp.444-450.16. W. H. Zhu et al, “Cure shrinkage characterization andits implementation into correlation of warpagebetween simulation and measurement”, Proceedingsof EUROSime2007.

Figure 7: Simulated warpage curve along package half-diagonal (10x10.5mm wCSP) Figure 8: Shadow Moiré measurement @ 25 C for 10x10.5mm wCSP Figure 9: Shadow Moiré measurement @ 215 C for 10x10.5mm wCSP 3. Warpage Modeling of PoP Bottom Package In this part, simulation is used extensively in a PoP bottom package development. The purpose is to .