page 1
page 2
page 3
page 4
page 5
page 6
page 7
page 8
page 9
page 10
page 11
page 12
page 13
page 14
page 15
page 16
page 17
page 18
page 19
page 20
page 21
page 22
page 23
page 24
page 25
page 26
page 27
page 28
page 29
page 30
page 31
page 32
page 33
page 34
page 35
page 36
page 37
page 38
page 39
page 40
page 41
page 42
page 43
page 44

overcome the predicted performance and costproblems of future IC fabrication. The ITRSroadmap predicts 3D integration as a key tech-nology to overcome this so-called 'wiring crisis'and the solution will most likely be based onTSV technology. The most promising 3D integration schemescurrently under consideration involve the ver-tical stacking of integrated circuits and otherdevices. These schemes vary in their details butall must solve two central problems: how tobond the integrated layers together and howto create electrical connections among them.Bonding and TSV technologies each have theirown unique set of considerations which oftencenter around how the structure will hold upduring subsequent processing, such as theaddition of another layer:·Will the stresses induced by additional ther-mal processing cause debonding or shifting ofthe existing bonds? ·Will the stress and strain cause cracks ordelamination in the TSVs? ·What are the best materials and processesto use to minimize these negative effects?PLASMA FOCUSED ION BEAMFIB systems, which use an ion beam to cut andimage cross sections through subsurface struc-tures with nanoscale precision and imagingresolution, have long been a mainstay of phys-ical analysis for integrated circuits. Althoughthe structures used in 3D integration can beexpected to decrease in size as the technolo-gies evolve, they are much larger than thedimensions of the transistors and intercon-nects used in current integrated circuits, andthe cutting speed of FIBs designed for ICs isgenerally inadequate for TSVs and bondingstructures. A typical 10 ?m ??10 ?m IC cross-section requires the removal of 1000 ?m3ofmaterial and takes a few minutes. A 100 ?m ??100 ?m TSV cross-section requires the removalof 1,000,000 ?m3of material and would takemost of a day with conventional FIB.The Vion PFIB system (FEI Company, Hills-boro, Oregon, USA) uses an inductively cou-pled plasma source [1-3] (Figure 1) to providematerial removal rates 20??faster than con-ventional FIBs that use liquid metal ion sources(LMIS). A LMIS is essentially a point source 50nm in diameter with a low angular intensity.The Vion system's plasma source is larger, 15?m, but has a much higher angular intensity.Because of its small virtual size, the LMIS is easyto focus into a small spot at low beam currents,but at beam currents above 10 nA sphericalaberration effects severely degrade perfor-mance. The plasma source can deliver currentsin excess of a ?A (>20??greater than a typicalLMIS based system) while still maintaining awell focused beam. Since material removalrates are primarily a function of beam current,the PFIB has an advantage of 20??or moreover conventional FIB at high currents, whilestill preserving excellent milling precision andimaging resolution at low beam currents.The xenon ion beam emitted by the plasmasource has high sputtering yield, high bright-ness and low energy spread. In addition, byintroducing various gases, the PFIB can selec-tively etch specific materials or deposit pat-terned conductors and insulators (similar toconventional FIB systems). The plasma sourcealso offers the potential to use different ionspecies to enhance performance in specificapplications.CURTAININGThe difference in FIB milling rates of the vari-ous materials present in a device (Cu, Si, Sn,dielectrics, polyimides and mold compounds)can cause 'curtaining' when milling cross-sec-tions. This milling artifact can make detailedFigure 2: Curtaining artifacts (upper left), caused by variations in milling rate for different materials, can be effectively suppressed (right) by rocking the sampleto mill in a sequence of alternating angles (lower left).MICROSCOPY AND ANALYSISNOVEMBER 102011Figure 1b: The PFIB maintains excellent spot size performance over a broad range of beam currents. Figure 1c: At high beam currents the PFIB can remove material twenty times faster than a liquid metal ion source.bc

PFIB INMICROELECTRONICSanalysis of the structures difficult or evenimpossible. Figure 2 shows typical curtaining effects onthe silicon substrate as well as the TSV itself,caused by milling through the overlying roughpoly crystalline metal film. These curtainingeffects can be effectively suppressed by rock-ing the sample during the FIB milling process.Milling in a sequence of alternating incidenceangles creates a clean cross section free of cur-taining artifacts without the need for time-consuming low current polish steps.EXAMPLES OF APPLICATIONS OFPLASMA FOCUSED ION BEAM Through Silicon ViasTSVs are themselves subject to a number ofeffects that can result in defects and failures.For example, the large differences in thermalexpansion between copper via fill and the sur-rounding silicon substrate can cause crackingwithin the copper and delamination from thevia sidewall during thermal processing. 'Key-holing' results from incomplete filling of vias(Figure 3).Solid Liquid Interdiffusion BondingOne of the most difficult issues to address isthe behavior of bonds between chips duringsubsequent processes (Figure 4). For example,it is critical that a bond between the first chipsin the stack not be disturbed by the subse-quent bonding of an additional chip. Solid-liquid-interdiffusion (SLID) [4] is a uniquedirect metal bonding technology that avoidsremelting of existing bonds during the forma-tion of new bonds by using high melting inter-metallic phases. During bond formation, solidmetal diffuses into the liquid phase of a lowermelting metal resulting in high melting pointfinal phase that remains solid during subse-quent bond forming processes.Anisotropic Conductive AdhesivesAnisotropic conductive adhesives (ACA) can beused [5] to bond wafers together physicallyand electrically using an organic bonding com-pound (benzocyclobutene, BCB) filled with 4-?m sized metal covered polymer spheres(MPS). The BCB assures mechanical strengthwhereas the MPS provide the required electri-cal conductivity at interconnection points. Theconcentration of MPS must be high enough toensure good electrical contact betweenopposed pads and at the same time lowenough to guarantee electrical insulationwhere pads are not present. To study the bonding in detail, samples werecleaved, then milled with the plasma-FIB toreveal the bonding region and finallyinspected with plasma-FIB imaging. Theplasma-FIB milling speed makes it possible toprepare the sample (~200 ??50 ??600 ?m3material removed) within 30 minutes. Themetal layer covering the polymer spherescould be observed at the bonding interfacewith sufficient resolution to estimate both thelocal MPS density and their compression statebetween the bond pad metal layers. In Figure5 the bonding process is illustrated in the topFigure 4: The void between these pads is the result of an incomplete SLID bonding process. The various intermetallic phases are clearly visible above, below andto the right of the void. MICROSCOPY AND ANALYSISNOVEMBER 201111Figure 3: (a, b) Differing thermal expansionbetween copper via fill and silicon substrate caused delamination shownin this via before (a) and after (b)annealing. (c) Keyholing occurred when this viawas not filled completely with tung-sten.caCu Cu3SnCu6Sn5FIB debrisb