Thermal Stress Induced Delamination Of Through Silicon Vias In 3 D .

1y ago
6 Views
1 Downloads
740.22 KB
6 Pages
Last View : 30d ago
Last Download : 3m ago
Upload by : Abby Duckworth
Transcription

Thermal Stress Induced Delamination of Through Silicon Vias in 3-D InterconnectsKuan H. Lu, Suk-Kyu Ryu*, Qiu Zhao, Xuefeng Zhang, Jay Im, Rui Huang*, and Paul S. HoMicroelectronics Research Center, University of Texas at Austin, Austin, TX 78758*Department of Aerospace Engineering and Engineering Mechanics, University of Texas at Austin, Austin, Texas 78712garylu@mail.utexas.edugeometry was first studied for three types of TSV structures:(a) full copper filling, (b) annular copper filling, and (c) fullcopper filling with an annular polymer liner between thecopper and the silicon matrix. Here a previous study onthermal stress behavior [3] was extended to calculate thedriving force for interfacial delamination. In addition, weinvestigated the effect of depositing a metal pad on top of theTSV on interfacial delamination. Finally, in the last part, themetallization effect was examined for four TSV materials:copper, aluminum, nickel, and tungsten. The implication onthe thermo-mechnical reliability of TSVs will be discussed.AbstractIn this paper we investigated the interfacial delaminationof through silicon via (TSV) structures under thermal cyclingor processing. First finite element analysis (FEA) was used toevaluate the thermal stresses and the driving force of TSVdelamaination. Then, the modeling results were validated byanalytical solutions of the crack driving force deduced for along crack at the steady state. Both results were found to be ingood agreement at the steady state and together theysuggested a fracture mechanism to account for the TSVdelamination observed. The analytical solution furtherprovided a basic framework for studying the impact ofmaterials, process and structural design on reliability of theTSV structure. In particular, we found that reducing the TSVdiameter yields a definite advantage in lowering the crackdriving force. In addition, annular TSVs and an overlayingmetal pad on a TSV can reduce the crack driving force fordelamination during thermal cycling. Finally, themetallization effect was investigated for four TSV materials:copper, aluminum, nickel, and tungsten. Tungsten was foundto have the smallest crack driving force due to the leastthermal mismatch with the surrounding silicon. The reliabilityimplication was discussed.Crack Driving Force for TSV Delamination: Steady-stateSolutionsThermal stresses can develop around TSVs duringfabrication arising from the thermal expansion mismatchbetween constituent materials. Figure 1 shows the thermalstress inside a copper TSV structure under thermal cyclingfrom FEA simulations, consisting of both shear and normalstresses. Consider first the effect from the shear stresses. TheFEA results reveal a shear stress concentration around theTSV boundary near the wafer surface where the shear stresseshave opposite signs depending on whether a heating or acooling thermal load. The different signs of the shear stresswill drive the copper TSV to “pop up” or to cave in from thewafer surface. In both cases, the stress drives a shear-induceddelamination along the TSV/silicon interface with the samemagnitude of ERR. In contrast, the different signs of thenormal stresses induce distinct effects on interfacialdelamination. Only the tensile normal stress under a negativethermal load can drive the TSV delamination while thecompressive stress under a positive thermal load will notcontribute to the TSV delamination. The stress combinationwill result in delamination under a positive thermal loaddriven mainly by the shear stress near the wafer surface, i.e. aMode II fracture. On the other hand, the delamination under anegative thermal load is driven by both the shear stress andthe radial tensile stress, i.e. a mix-mode fracture, and thedriving force is generally higher. The difference between twoopposite thermal loads is further illustrated in Figure 2.IntroductionThe incorporation of TSVs poses a significant challenge tothermo-mechanical reliability of the 3-D interconnects. Inparticular, the mismatch in coefficients of thermal expansion(CTEs) between the conducting metal in TSV and the siliconmatrix can generate thermal stresses inside and around TSVs[1-3]. Such stresses can be sufficient to degrade theperformance of stress-sensitive devices [4], to induce cohesivecracking in the silicon [3], and to drive interfacialdelamination between the TSV and the silicon matrix [5]. Infact, thermal stress-induced TSV delamination has been foundto be one of the dominant failure modes for 3-D interconnects.During fabrication of 3-D interconnects, TSVs can “pop-up”from the silicon wafer and damage the Back-End-Of-Line(BEOL) structures. Finite element analysis (FEA) has beenapplied to simulate the driving force of TSV delamination [5].In general, the crack driving force was found to increase withthe diameter of TSVs and the circumferential crack length.In this paper, the driving force and the delaminationmechanism for TSV structures were further investigated. Thepaper is organized in three parts.First, for betterunderstanding of the underlying mechanism, the energyrelease rate (ERR) that drives the TSV delamination wasevaluated using analytical solutions deduced for simplifiedTSV structures. The results were supplemented and comparedwith finite element calculations. This was followed by a studyon the impact of materials, process and structural design onthe reliability of the TSV structure. The effect of TSV978-1-4244-6412-8/10/ 26.00 2010 IEEEyCu TSVcross-sectionxShear stress xyNormal stress xxFigure 1. Thermal stress distribution around the TSV boundary nearthe wafer surface.402010 Electronic Components and Technology Conference

The steady-state ERR solutions apply to a TSV structurefor a long, circumferential crack with its crack tip away fromthe wafer surface. At the onset of spontaneous crackpropagation, the ERR will be lower than the steady-stateERR. To illustrate this behavior, the non-steady state ERRsfor short cracks with increasing crack length was calculatedusing FEA simulations. The results are superimposed to thesteady-state solution under a thermal load of -250oC (Figure4).Cu TSV, T -250oCCoolingFigure 2. TSV delamination under opposite thermal loads.Energy Release Rate (J/m2)18The analytical solutions for steady-state TSV delaminationhave been derived for both the cooling and heating conditions[6]. Consider an isolated, infinitely long TSV embedded in aninfinite matrix with a circumferential crack propagating alongthe axial direction under a thermal load. The steady-stateERR, Gss, can be expressed as follows:E T D f2(1)4(1 v )2864251015202530Crack Length ( m)Figure 4. Energy release rate vs. crack length at circumference of CuTSV for various TSV diameters. Dash lines indicate the steady-stateERR solution (ΔT -250 ºC).oAnnular Cu TSV, T -250 C2Rate GSS (J/m )Energy Release3530252015100.2Ratio0.4of inner- to 0.6outer-d3530252020100.8iamete1.0r, 0( mfr,300D5)5040ete8(1 v )Fully filled Cu TSV, CoolingFigure 5. Plot of steady-state ERR, Eq. 3, as functions of and theouter-diameter of TSV ( T -250oC).151050eter,Df ( m)Figure 4 shows a good agreement between the steady-stateERR solution and the FEA simulations at a long crack length 2Df. The figure also shows that the ERR increases with thecrack length and TSV diameter, and can exceed 10 J/m2 for aTSV with 20 µm diameter under a -250oC of thermal load.40530020-200-150)-1000TSV T ( oC10diam2)Rate G SS (J/m100(2)where E, , T, Df and v represents Young’s modulus, CTEmismatch, thermal load, TSV diameter and Poisson’s ratio,respectively. The elastic mismatch between TSV and thematrix was neglected here for simplicity. The analyticsolutions indicate that the steady-state ERR is proportional tothe diameter of TSV, and scales with the square of the thermalload and the square of CTE mismatch. Comparing betweenEq. 1 and 2, the steady-state ERR in the cooling process is 35% higher than that in the heating process under the samethermal load. It implies that the TSV delamination is morelikely to occur during the cooling processes. Figure 3 depictsthe steady-state ERR for delamination of copper TSVs as afunction of the TSV diameter and the thermal load.Energy ReleaseDf 10 umDf 5 um120E T D f 1 v Heating: G SS 14Dash Lines:Steady-State ERRd iamCooling: G SS Df 20 umDf 15 um16TSVHeatingTSV Filling Structures: Annular and Lined TSVsThe steady-state solution for annular TSV delaminationhas been derived as well for the cooling conditions [6]. ByFigure 3. Crack driving force for TSV delamination under cooling.412010 Electronic Components and Technology Conference

The effect of a compliant liner on the TSV delaminationwas also investigated. The cross-section of the lined TSVmodel used is depicted in Figure 6b. The liner is assumed tobe BCB polymer. The thickness of the liner ranges from0.5 m to 2 m while the inner diameter is fixed at 20 m.Delamination is assumed to occur at the interface between thecopper and the BCB liner. The crack driving force for threelined TSVs were simulated and compared with the fully filledcopper TSV, as shown in Figure 8. At a crack length 10 m,the ERRs among three lined structures are similar and alllower than the fully filled TSV. However, the ERRs of thelined TSVs can be higher than the fully filled TSV at a shortercrack length at below 5 m. This can be attributed to a largerout-of-plane displacement in the BCB liner near the wafersurface. Although a soft polymer liner can reduce the thermalstress in the silicon matrix [3], it may influence the drivingforce for interfacial delamination in two ways: higher ERRthan fully filled TSV for a shorter crack and lower ERR for alonger crack.ignoring the elastic mismatch between TSV and the matrix forsimplicity, the steady-state ERR, Gss, can be expressed asE T D f2G SS 4(1 v ) 1 2(3)where represents the ratio of the inner- to the outerdiameters of an annular TSV. Figure 5 depicts the steady-stateERR as a functions of the ratio and the TSV diameter.According to Equation 3, ERR will decrease significantlywith decreasing metal thickness in the annular TSV. For avery thin layer of metallization, delamination energy canreduce to nearly zero. This suggests that the driving force forTSV delamination can be effectively reduced using an annularstructural design.The steady-state ERR solution for annular structures wascompared with FEA simulations. The cross-section of theannular TSV model used in the calculation is depicted inFigure 6a. The inner diameters of copper TSVs range from5 m to 15 m while the outer diameter is fixed at 20 m. Thenon-steady state ERR for copper TSV delamination wassimulated for a -250oC of thermal load. Figure 7 shows a goodagreement between the steady-state solution and the FEAsimulations for a long crack length. The results also indicatethat the larger the inner diameter, the shorter the crack lengthto reach the steady-state.(a) Annular TSVEnergy Release Rate (J/m2)121086Fully filled TSV 20 umBCB liner 0.5 umBCB liner 1 umBCB liner 2 um420(b) TSV w. liner0Figure 6. Cross-section of TSV structures considered in FEAsimulations: (a) annular TSV and (b) TSV with liner. The yellowarrows indicate circumferential crack propagation from the wafersurface.51015202530Crack Length ( m)Figure 8. Comparison between BCB lined and fully filled TSVs fora fixed diameter of 20 m.Outer dia. 20 m, T -250oC14Energy Release Rate (J/m2)Df 20 m, T -250oC14Dash Lines:Steady-State ERR121086(a) Vertical crack4Fully filled TSV 20 umAnnular: inner dia. 5 umAnnular: inner dia. 10 umAnnular: inner dia. 15 um2Figure 9. Schematics of various interfacial crack geometriesconsidered.TSV Delamination under Overlaying Metal PadSimplified TSV structures have been discussed in theprevious sections. The actual TSVs integrated in 3-Dinfrastructures are much more complicated. To add morecomplexity to the model, a TSV with a “nail head” wasanalyzed (Figure 9). The overlaying metal pad is required to00510152025(b) Horizontal crack30Crack Length ( m)Figure 7. Simulated ERR for annular TSVs with various innerdiameters and fixed outer diameter.422010 Electronic Components and Technology Conference

form electrical interconnects. The overlaying flange of metalpad is expected to constrain the crack propagation along thevertical interface, and thus reduce the ERR for TSVdelamination.The TSV delamination under an overlaying metal pad waspreviously examined [5, 6]. Consider a vertical,circumferential crack propagating downward underneath arigid metal pad (Figure 9a). The delamination is mainly drivenby a negative thermal load because the crack driving forcegenerated by a positive thermal load was found to be oneorder of magnitude smaller. An analytic solution for the crackdriving force under a negative thermal load has been derived[6]. The steady-state ERR, Gss, can be expressed as follows:GSS has to be removed using a chemical-mechanical polishing(CMP) process after annealing. The copper overburden isrigid and can keep the TSV/matrix interfaces fromdelaminating. If the overburden is fully removed, the ERR forTSV delamination will increase by 35% as mentionedearlier. Without the protection of the overlaying copper film,such TSV structure needs to go through another thermalprocess to form the metal contact on top, which can furtherincrease the risk of delamination. Therefore, a metallizationprocess without fully depleting the copper overburden isexpected to mitigate the type of TSV delamination along theaxial direction as discussed here.Another type of TSV delamination can occur with thecrack propagating horizontally at the interface between thesilicon top surface and the flange part of the nail head (Figure9b). Such delamination is mainly driven by a positive thermalload. Under a positive thermal load, the TSV tends to extrudeout of the silicon matrix due to the larger thermal expansionof copper, resulting in a peeling stress at the horizontalinterface. The peeling stress can drive crack propagation fromthe inner- to the outer-portion of the flange.Based on the geometry in Figure 9b, two axial-symmetricFEA models were built to evaluate the crack driving force forhorizontal cracks under a 250oC of thermal load. FEAsimulations were performed separately for both the outwardand the inward circumferential cracks, and the results areshown in Figure 11. For the outward crack, the ERR reaches amaximum at 2 m away from the edge of the TSV. As thecrack tip extends away from the edge of the TSV, the crackdriving force gradually decreases to zero. On the other hand,the ERR for the inward crack generally increases as the cracktip approaches the edge of the TSV. According to oursimulations, the ERR for the horizontal cracks under heatingis much smaller than the vertical crack under cooling. Theresult suggests that the delamination is more likely to initiatefrom the vertical interface during thermal cycling.2Em T Df (1 v f )3 (1 2v f )(1 D2 ) (1 vm )(1 v f )2 (1 D )2 4(1 v f )(1 2v f )(1 D ) (1 vm )(1 D ) 2 (4)where D is the first Dundurs’ parameter and the subscripts mand f stands for the matrix and the fiber, respectively. If theelastic mismatch between the TSV and the matrix isneglected, Equation 4 can be simplified to become Equation2. FEA simulations were performed to validate the analyticalsolution. An axial-symmetric FEA model was built accordingto the geometry in Figure 9a, and the crack driving force wassimulated as a function of the vertical crack length for a 20 mcopper TSV under a -250oC thermal load. The calculated ERRwas plotted and compared with another 20 m copper TSVwithout the nail head, as shown in Figure 10. The results showa good agreement between the analytical solution and theFEA simulations for the steady state with a long crack length.An overlaying metal pad with the thickness of a quarter ofTSV diameter is shown rigid enough to restrict the crackopening and to reduce the ERR for TSV delamination by 30%.Df 20 m, T -250oCDf 20 m, T 250oC1.412Outbound crackInbound crack1.210Energy Release Rate (J/m 2)Energy Release Rate (J/m2)1486With Cu Pad (FEA)Steady State SolutionWithout Cu Pad (FEA)Steady State Solution4200102030401.00.80.60.40.2500.0Crack Length ( m)0Figure 10. Effect of an overlaying Cu pad on ERR for delaminationat the vertical interface ( T -250oC).20406080Crack Length ( m)Figure 11. ERR for outbound and inbound crack along the Cuflange/Si interface ( T 250oC).The overlaying metal pad can reduce the risk of TSVdelamination, and the metallization process for TSVs can beengineered accordingly. For instance, an overburden ofcopper layer always comes with the electroplating process andBoth the horizontal and the vertical interfaces will have tobe damaged for the “nail-head” TSV to pop up. A possible432010 Electronic Components and Technology Conference

mechanism is as follows. First, the vertical interface betweenthe TSV and the silicon matrix is debonded duringtemperature cycles. Once the vertical interface is debonded,the copper via is easier to stick out of the wafer upon heating,resulting in an increased crack driving force on the horizontalinterface. After repeated thermal cycles, both interfaces can besufficiently delaminated to induce the TSV pop-up from thewafer. This process is illustrated in Figure 12.the crack driving force alone. The interfacial adhesionbetween the metallization material and the barrier layer ofTSVs also play an important role in controlling thedelamination. In this case, copper material can develop moreplastic deformation compared with tungsten. Therefore,copper can absorb energy by plastic deformation near thecrack tip and increase the fracture resistance to crackpropagation. This interesting and important aspect of fracturecharacteristics for TSV will be further investigated.Table 1. Thermo-mechanical properties of Al, Cu, Ni, W, and Si.(a)(b)(c)Figure 12. Illustration of TSV pop-up process: (a) Vertical interfacedebonding under cooling, (b) Horizontal interface debonding underheating, (c) TSV pop-up after repeated thermal cycles.Effect of TSV Metallization on DelaminationCurrent TSVs in the 3-D infrastructure are fabricated withvarious metals, such as copper, tungsten, and nickel. Thedifference in thermo-mechanical reliability among thesematerials is of interest. Based on the steady-state ERRequations, a metal with a low modulus and a low CTE isfavored for reducing the driving force for delamination.Unfortunately, a metal with low modulus usually comes witha high CTE, and vice versa. The thermo-mechanicalproperties of four commonly used metals are listed in Table 1,in which aluminum has the smallest modulus while tungstenhas the smallest CTE. To compare these four materials, theERRs for TSV delamination have been calculated for thesame thermal load, and the results are plotted in Figure 13.MaterialCTE(ppm/oC)Young’sModulus 4001300.350.350.310.280.28SummaryThe characteristics of thermal stress-induced TSVdelamination were investigated using analytic solutions andFEA simulations. The driving force for TSV delaminationwas found to be different depending on whether a positive ora negative thermal load. Under a positive thermal load, thefracture mode was primarily a shear mode (mode II)delamination. On the other hand, the fracture mode under anegative thermal load was a mixture of peeling (mode I) andshear (mode II). Based on the analytic solutions, the drivingforce for TSV delamination can be reduced by scaling downthe diameter of TSVs and by using lower CTE metals. TSVswith various filling structures were compared. Annularstructure exhibited lower ERRs than the fully filled TSVstructure, and an overlaying metal pad can constrain the crackopening and thus reduce the ERR for delamination along theaxial direction. The effect of metallization on TSVdelamiantion was investigated using four materials:aluminum, copper, nickel and tungsten. Tungsten exhibitedthe least ERR due to the smallest CTE mismatch with thesilicon substrate. Nevertheless, the thermo-mechanicalreliability of TSV structures also depends on other factorssuch as processing temperature, interfacial adhesion, and theplasticity of the materials. These effects will be furtherinvestigated.Figure 13. Steady-state ERR: TSVs filled with four kinds of metals.AcknowledgmentsThe authors want to thank Semiconductor ResearchCorporation for the financial support and Dr. Philip Garrou ofMicroelectronic Consultants of North Carolina for helpfuldiscussion.Figure 13 indicates that the driving force for TSVdelamination significantly decreases with reduced CTE of themetal. This can be attributed to the fact that ERR isproportional to the square of the CTE mismatch between themetal and the silicon matrix. Given the same thermal load andTSV dimension, tungsten, which has the lowest CTE amongthe four materials, exhibits the least ERR for delamination.However, the tendency to delamination cannot be judged byReferences1. Suhir, E. et al, “Disc-like Copper Vias Fabricated in aSilicon Wafer: Design for Reliability,” Proc 58thElectronic Components and Technology Conf, Orlando,FL, May 2008, pp. 1664-1666.2. Chen, Z. et al, “Thermo-mechanical Characterization ofCopper Filled and Polymer Filled TSVs ConsideringNonlinear Material Behaviors,” Proc 59th Electronic442010 Electronic Components and Technology Conference

3.4.5.6.Components and Technology Conf, San Diego, CA, May2009, pp. 1374-1380.Lu, K. H. et al, “Thermo-Mechanical Reliability of 3-DICs containing Through Silicon Vias,” Proc 59thElectronic Components and Technology Conf, San Diego,CA, May 2009, pp. 630-634.Thompson, S. et al, “Uniaxial-Process-Induced StrainedSi: Extending the CMOS Roadmap,” IEEE Trans.Electron Devices, Vol. 53, No. 5, May 2006.Liu, X. et al, “Failure Mechanisms and Optimum Designfor Electroplated Copper Through-Silicon Vias (TSV),”Proc 59th Electronic Components and Technology Conf,San Diego, CA, May 2009, pp. 624-629.Ryu, S.-K. et al, “The Effect of Near-Surface ThermalStresses on Reliability of Through-Silicon-Vias (TSVs) in3-D Interconnects,” in preparation.452010 Electronic Components and Technology Conference

Eq. 1 and 2, the steady-state ERR in the cooling process is 35% higher than that in the heating process under the same thermal load. It implies that the TSV delamination is more likely to occur during the cooling processes. Figure 3 depicts the steady-state ERR for delamination of copper TSVs as a function of the TSV diameter and the thermal .

Related Documents:

Figure 8 shows Tool Maker’s microscope with which delamination was measured. Figure:-8. Schematic view of delamination factor and a view of tool makers’ microscope. Delamination is commonly classified as peel-up delamination at the twist drill entrance and pushdown d

Delamination is one of the most common defects found in fiber-reinforced composite laminates due to their weak transverse tensile and inter-laminar shear strengths [1]. . scattering and energy leakage of guided waves for the detection and sizing of delamination-type defects [3-7]. By interaction of guided waves with delamination, waves are .

1.4 importance of human resource management 1.5 stress management 1.6 what is stress? 1.7 history of stress 1.8 stressors 1.9 causes of stress 1.10 four major types of stress 1.11 symptoms of stress 1.12 coping with stress at work place 1.13 role of human resource manager with regard to stress management 1.14 stress in the garment sector

can be encountered when testing delamination fracture toughness. Keywords: composite, delamination, fracture toughness, test technique, carbon-carbon SYMBOLS a Crack length, in. b Specimen width, in. c MMB apparatus lever length, in. h Half thickness of the specimen, in. h D Thickness of doubler plate, in. l Inner half span length in 4ENF test, in.

Keywords: Thermal barrier coatings; Delamination; Thermal gradients; Cracking; Environmental degradation 1. Introduction The maximum temperature capability of thermal barrier systems used in gas turbines is often limited by deposits of calcium-magnesium-alumino-silicate (CMAS) [1–3]. These deposits melt and wet the yttria-stabilized zirconia .

1. Stress-Strain Data 10 2. Mohr Coulomb Strength Criteria and 11 Stress Paths 3. Effect of Different Stress Paths 13 4. Stress-Strain Data for Different Stress 1, Paths and the Hyperbolic Stress-Strain Relationship 5. Water Content versus Log Stress 16 6. Review 17 B. CIU Tests 18 1. Stress-Strain Data 18 2.

2D Stress Tensor x z xx xx zz zz xz xz zx zx. Lithostatic stress/ hydrostatic stress Lithostatic stress Tectonic stress Fluid Pressure-Hydrostatic-Hydrodynamic Lithostatic Stress Due to load of overburden Magnitude of stress components is the same in all

In the English writing system, many of the graphemes (letters and letter groups) have more than one possible pronunciation. Sometimes, specific sequences of letters can alert the reader to the possible pronunciation required; for example, note the letter sequences shown as ‘hollow letters’ in this guide as in ‘watch’, ‘salt’ and ‘city’ - indicating that, in these words with .