Electron Backscatter Diffraction Characterization of Inlaid Cu Lines for Interconnect Applications
D. P. FIELD, T. MUPPIDI, J. E. SANCHEZ, JR.*
School of Mechanical and Materials Engineering, Washington State University, Pullman, Washington; *Unity Semiconductor,Sunnyvale, California, USA
Summary: Automated electron backscatter diffraction(EBSD) techniques have been used to characterize the mi-crostructures of thin films for the past decade or so. The re-cent change in strategy from an aluminum-based inter-connect structure in integrated circuits to one based oncopper has necessitated the development of new fabrica-tion procedures. Along with new processes, complete char-acterization of the microstructures is imperative for im-proving manufacturability of the Cu interconnect lines andin-service reliability. Electron backscatter diffraction hasbeen adopted as an important characterization tool in thiseffort. Cu microstructures vary dramatically as a functionof processing conditions, including electroplating bathchemistry, sublayer material, stacking sequence of sub-layers, annealing conditions, and line widths and depths.Crystallographic textures and grain size and grain bound-ary character distributions, all of which may influencemanufacturability and reliability of interconnect lines, areideally characterized using EBSD. The present discussionpresents some results showing structural dependence uponprocessing parameters. In addition, the authors identify anin-plane orientation preference in inlaid Cu lines {111} nor-mal to the line surface and <110> aligned with the line di-rection. This relationship tends to strengthen as the linewidth decreases.
Key words: thin films, copper, electron backscatter dif-fraction, texture
Aluminum metallization has been widely used for inte-grated circuit (IC) interconnects over the last 40 years,and a large knowledge base has been established on howto control the microstructure to expand manufacturingyield and reliability performance limits. It is well knownthat an aluminum alloy containing a small fraction of cop-per and a microstructure consisting of a uniform bamboograin structure and strong {111} texture is best suited forwithstanding electromigration under high current densities.It is also generally known that certain adhesion layers pro-vide growth conditions for fine grained films that allowsmooth sidewall formation during subsequent metal etch.The shift in strategy from Al to Cu interconnects, and fromconventional metal etching to a damascene, or in-laidmetal, fabrication technique presents new challenges to un-derstand how manufacturing conditions control the mi-crostructure of copper interconnect features. The final mi-crostructure of inlaid copper interconnects forms within theconfined space of the prepatterned dielectric, rather thanwithin a two-dimensional thin film. Epitaxial substrate ef-fects from the trench bottom and sidewalls influence the de-veloping microstructure depending on the interface en-ergy with the barrier or adhesion layer, the feature widthand depth, and the thermally induced stress conditions insubsequent processing. Electron backscatter diffractionhas been used in a number of investigations of thin Cu filmsand lines to determine grain size and crystallographic tex-ture. The present discussion summarizes some of the in-formation gleaned using EBSD about structural evolutionin Cu films and lines. Also, the texture and grain size in sin-gle-level Cu lines as a function of line width are presentedand compared with the structures previously observed inAl lines processed by the more conventional reactive ionetching (RIE) technique.
Materials and Methods
All EBSD measurements and analysis were performedusing a TexSEM Laboratories (Draper, Ut., USA) orienta-tion imaging microscopy (OIM 3.0) system attached to ei-ther a FEI XL-30 FE-SEM or an FEI XL-40 FE-SEM (FEI
This work was supported in part by the Foundation for Advances inMedicine and Science (FAMS, Inc.).
Address for reprints:
David Field201 Sloan HallBox 642920Washington State UniversityPullman, WA 99164-2920, USAe-mail: [email protected]
Company, Hillsboro, Ore., USA). It was previously demon-strated that the lateral spatial resolution of EBSD meas-urements using the column for this scanning electron mi-croscope (SEM) is on the order of 10–20 nm depending onthe material being analyzed (Field 1999, Humphreys et al.1999, Nowell and Field 1998, Troost 1993) and is thus ad-equate for analysis of Cu thin film microstructures.
Structure evolution in inlaid Cu lines is generally ac-cepted to be relatively independent of that observed in theCu overburden film (Besser et al. 2001). Because of this,interrogation of line microstructure requires either crosssectioning to reveal the through-thickness structure, orplan-view analysis after over polishing to remove the rem-nant structure from the overburden. Electron backscatterdiffraction analysis of the Cu lines discussed in the pres-ent work was performed on plan-view sections after aslight over polish during the chemical-mechanical pla-narization (CMP) process. The observed structures possessa distinct character that is a function of line width, indi-cating that the overburden microstructure did not signifi-cantly influence the observed structures.
Results
Previous investigations into the microstructure and tex-ture of Cu films and lines by EBSD showed a strong de-pendence of texture and grain size distribution on platingbath chemistry and line height to width ratio (Besser et al.2001, Field et al. 2001, Vanasupa et al. 1999). Copperlines were manufactured by the damascene technique inSiO2 using a 30 nm thick physical vapor deposited (PVD)aN barrier layer and a PVD copper seed layer. Copperfilms were deposited with three different electroplatingchemistries (commercially available solutions, herein de-noted A, B, and C) to form inlaid Cu line structures with0.3 µm width and 0.5 µm height. All samples were allowedto self anneal at room temperature for 5 days before CMP.The resulting textures and grain sizes in the lines were rad-ically distinct from one another.
Figure 1 contains representative orientation images fromself-annealed 0.3 µm lines for each of the deposition con-ditions investigated. The images were taken from speci-mens manufactured using plating processes A, B, and Cfrom left to right across Figure 1. These orientation imagesare shaded with crystallite lattice poles aligned with thespecimen normal orientation according to the key pre-sented in the unit triangle as part of Figure 1. (All orienta-tion images shown in this paper follow this convention.) Itis apparent that individual grains from the structure ob-tained using process A span the entire width of the trench,while those in the process C structure generally span onlyhalf the width, leaving a grain boundary seam down thecenter of the line. The majority of the grains in the processB line span the width of the trench, but a significant frac-tion of those observed do not. Scanning electron mi-croscopy cross-sectional analysis of these lines shows that
310 Scanning Vol. 25, 6 (2003)
the trenches were completely filled in all cases, with noegregious voids near the trench floors.
Grain size analysis was performed by orientation imag-ing for all specimens. In the orientation imaging study, agrain is defined as a set of contiguous measurementsbounded by a region of misorientation >2° from point topoint over the area scanned. Electromigration is influencedby grain size because the grain boundaries offer a path ofrapid diffusion to accommodate migrating atoms, in com-parison with the bulk lattice. Coherent twin boundaries donot offer a path of rapid diffusion similar to that expectedfrom a random, high angle boundary and should thereforenot be included in grain size determinations for this pur-pose. For those 0.3 µm wide lines shown in Figure 1, theaverage length of grain along the lines were 0.98, 0.26, and0.16 µm when twin boundaries were not included as grainboundaries, for the A, B, and C processes, respectively.
Crystallographic textures were measured from all linesusing the data obtained during orientation imaging. Thepresent study included four to six scans of the 0.3 µm linesthat were 15–30 µm in length, giving between 60 and 200grains for each condition. To obtain a statistically reliablemeasure of such textures from individual measurementswould require on the order of 5,000 individual crystallitelattice orientations (Engler et al. 1999). Obviously, thestatistics of the determination are poor for such weakly tex-tured films, but the results point to important conclusions.Figure 2a–c contains {111} pole figures showing the in-lineCu textures from the A, B, and C bath chemistries, re-spectively. The directions identified on the pole figures are
FIG. 1 Test structures for 0.3 µm wide A, B, and C bath chemistriesfrom left to right. The orientation color key is also shown.
SW for trench sidewall and LD indicating the line direc-tion. The intensity scale of the pole figures shows a max-imum (black) at five times random. Processes A and B re-sult in very weak textures. Process A shows no identifiabletrend in texture, while process B contains a small fractionof {111} planes aligned with the trench sidewalls. ProcessC results in a texture of near six times random, with the pre-ferred orientation being the {111} plane aligned with thetrench sidewalls.
One potentially important piece of information obtainedfrom spatially specific orientation measurements is that ofmisorientations between neighboring crystallites. Specif-ically, the fraction of twin boundaries in a Cu line may beof some importance in characterizing the microstructure forinterconnect applications. All specimens analyzed containa relatively high fraction of Σ3 boundaries as determinedby misorientation. These are the annealing twin boundariesprevalent throughout all structures and typical of face-cen-tered cubic (FCC) metals with low-stacking fault ener-gies. The fraction of grain boundaries classified as twinboundaries by orientation imaging for these structures are0.85, 0.49, and 0.27 for the self-anneal processes afterdeposition using A, B, and C bath chemistries, respec-tively.
In a separate experiment, Cu lines were prepared fromdamascene trenches in TEOS- based oxide consisting of anumber of 180 × 180 (µm)2 sections, each with a differentline width ranging from 0.20 to 10 µm and 0.5 µm deep.The metallization process consisted of 300 Å PVD Ta bar-rier layer + 800 Å PVD Cu seed layer and electroplated Cufill. The spacing between the lines was not the same in allsections and varied with the width of the lines and the sec-tion in use; however, this is assumed to have no effect in
D. P. Field et al.: EBSD characterization of Cu lines 311
our present study. The wafers were annealed at 400ºC andthe overburden was removed by chemical mechanical pol-ishing (CMP) to reveal the Cu line structure. The cross-sec-tional SEM images indicated no voids or observable defectsin the lines along the thickness.
The lines from these different sections were analyzedusing orientation imaging to determine the texture andgrain size. The analytical conditions were 25 keV acceler-ating voltage using a step size of 50 nm. The sectionsscanned were 20 × 15 (µm)2 to obtain about 100,000 datapoints for each region. These data were used to calculatethe grain size and the pole figures to determine the mi-crostructure and preferred orientation of the grain structurein the lines.
The orientation maps for some the Cu lines used in theexperiment are shown in Figure 3, along with the orienta-tion color key. The results of the average grain diameter,excluding and including twins as separate grains, and thetwin boundary fraction for the various line widths is shownin Table I. The maps and the grain size values clearly showa distinct difference between the wider and the narrowerlines. The narrower lines, that is, those <1 µm wide, werecompletely bamboo structured with the grains spanningthrough the line width, while the lines with the largerwidths had a multicrystalline nature. The average grain size(diameter), as is expected in this case, increases with in-creasing line width. The average grain diameter value in thewider lines was in the order of the grain sizes for the blan-ket films with a similar thickness. The constraint imposedby the sidewall on the grain growth is likely the reason forthe decrease in the average grain sizes in the narrow lines.
Crystallographic textures were calculated from the ori-entation measurements using a discrete binning procedure,as described by Matthies and Vinel (1999) with a bin sizeof 5º. The {111} and {110} pole figures were calculatedand compared for the different line widths. Figure 4 con-tains representative {111} and {110} pole figures for linesof 0.20 µm and 2.8 µm width. For all line widths, thegrains predominantly had a {111} out of plane orientation,as generally observed in electroplated blanket films. The{111} out of plane component is explained by the surfaceenergy minimization criteria in FCC blanket films. In lines,if the plating of the Cu takes place from the bottom of the
FIG. 2 (111) pole figures from the 0.3 µm lines shown in Figure 1for room temperature processes A, B, and C. The gray scale shading(white to black) indicates 0–5 times random for all plots. LD=line di-rection, SW=side wall.
TABLE I Grain size as a function of width for analyses includ-ing twins as distinct grains and not including twins
Line width Average grain dia (µm), Average grain dia (µm), (um) with twins without twins
trench and not on the sidewall, then the argument of sur-face and interface energy is still valid giving a (111) out ofplane texture in the grains. The {111} orientation intensi-ties varied with the line width, showing that the side walland stress differences become more of a factor with the de-creasing width. A plot showing the maximum {111} in-tensity as a function of line width is shown in Figure 5. Thechart shows a U-shaped curve with a minimum obtainedaround 1 µm.
The {110} pole figure of the 0.2 µm lines shows anonuniform in-plane intensity distribution and a maximum{110} orientation along the line length. Figure 6 containsinverse pole figures giving the pole orientation intensityaligned with the line directions. These texture plots showa {110} maximum along the line length, showing the pre-ferred in-plane texture component. These texture values in-creased with the decreasing line width, and the highestvalue of 4× random intensity was observed in the case ofthe narrowest lines, 0.2 µm wide. This result confirms theincreasing effect of the sidewall on texture evolution as linewidth decreases and height to width aspect ratio increases.
Discussion
Orientation dependence in thin films is controlled byprocessing conditions including sublayer material, thick-ness, and stacking sequence. Preferred orientation devel-
FIG. 3 Orientation maps of the copper lines obtained using orientation imaging microscopy. The color code indicates the orientation normalto the surface.
0.20 µm
Boundary levels: 15°4.5 µm = 90 steps IPF [001]
0.36 µm
Boundary levels: 15°4.5 µm = 90 steps IPF [001]
3.64 µm
Boundary levels: 15°4.5 µm = 90 steps IPF [001]
5.04 µm
Orientation color key
111
101001
Boundary levels: 15°4.5 µm = 90 steps IPF [001]
0.99 µm
Boundary levels: 15°4.5 µm = 90 steps IPF [001]
111
LD
101
LD
0.20 µm lines
2.80 µm lines
SW
111
LD
101
LD
FIG. 4 (111) and (110) pole figures for the 0.20 and 2.8 µm widelines showing (111) out of plane and, for the narrow line, (110) in-plane textures. LD= line direction, SW= side wall.
SWmax=16.20
8.600
5.592
3.637
2.365
1.538
1.000
0.650
oping during deposition, recrystallization, and grain growthis generally considered to be either surface energy domi-nated for thin films or strain energy dominated for thickerfilms (Thompson and Carel 1995). When line microstruc-tures are considered, additional energetic and physicalconstraints control texture evolution within the line. The in-laid Cu line is surrounded on three sides by a barrier layermaterial. The constraint associated with energy minimiza-tion for these interfaces could be more dominant than sur-face energy considerations for the free surface. In addition,grain boundary energy minimization could play a role in
texture development during grain growth since the grainboundary area will tend to minimize, creating a bamboo-type grain morphology. Sanchez and Besser (1998) andHau-Riege and Thompson (1999) discussed structure evo-lution and the energetics associated with grain growth forfree-standing and inlaid line microstructures. Two linewidth-dependent phenomena were observed in the tex-tures of the Cu lines. First, the {111} texture strength de-creased with line width down to about 1 µm, then beganto increase as line width further decreased to 0.2 µm. Sec-ond, an in-plane orientation preference was observed forthe most narrow lines.
Decreasing {111} texture strength as the line width nar-rows can be attributed to the increasing influence of thesidewalls. For wide lines, the metal is essentially in a stateof biaxial stress, and the free-surface and bottom interfaceenergies of the lines dominate grain growth and texture de-velopment. The low {111} surface energy dominatesgrowth and results in a preferred {111} fiber texture withlittle influence from sidewall nucleated growth. As the linewidth decreases, there is increased likelihood that the side-walls will contribute significantly. When the height-to-width ratio of the lines becomes >1, the interface energyassociated with the sidewalls could become a dominant fac-tor in structure evolution and a {111} sidewall texture candevelop (Besser et al. 2001, Lingk et al. 1999). There was
D. P. Field et al.: EBSD characterization of Cu lines 313
FIG. 5 Chart showing the change in the (111) peak intensity withvarying line width.
FIG. 6 The inverse pole figure texture plots with line length as the reference direction showing a preferred orientation of (110) along the linelength.
30
25
20
15
10
5
00 1 2
Line width (µm)
111
Pea
k in
tens
ity (
x ra
ndom
)
3 4 5 6
max=4.88
4.000
3.031
2.297
1.741
1.320
1.000
0.758
111
1010010.20 µm
111
1010010.26 µm
111
1010010.36 µm
111
1010010.55 µm
111
1010010.70 µm
111
1010010.99 µm
no evidence of sidewall-oriented textures in any of thelines investigated in this study, indicating again that addi-tional factors influence texture evolution and grain growthin electroplated inlaid Cu lines (such as bath chemistry andsublayer material and thickness). The observed increase intexture strength at the most narrow line widths is also anindication that sidewall nucleation and grain growth in-fluenced by the sidewall interface energy is not a dominantmechanism of texture development.
In addition to potential sidewall nucleation, the stress-state in the Cu line changes to one of tri-axial stress whenthe height-to-width ratio is near ≥1.0. Because of the phys-ical constraints on three sides of the lines, this effect con-tinues to increase as the line width narrows. Another struc-tural feature observed from the EBSD imaging of the Culines is the relatively high frequency of twin boundaries inlines having widths near 1 µm. Figure 7 shows the fractionof twin boundaries as a function of line width. In this case,the twins were identified as those having misorientationswithin 2° of the ideal twin misorientation (60º about a<111> axis). It is apparent by comparison of Figures 5 and7 that the high twin fractions exist in line widths where theweaker textures were present. It should be pointed out thatthis observation was not an anomaly that could be attrib-uted to inhomogeneity of the microstructure. The weakertextures and higher fractions of twin boundaries were con-sistently present in lines of approximately 1 µm in widthas observed over several different and widely separated re-gions. When a large number of twin boundaries exists in amicrostructure, there is a concomitant weakening of the tex-ture. A single grain twinned repeatedly over the varioustwin variants, and including twins of twins, results in a rel-atively random distribution of orientation components afteronly three twinning generations (Humphreys and Hatherly1995). Therefore, the weaker texture in the regions wherea large number of twin boundaries exists is not surprising.The growth of annealing twin is dependent on several fac-tors, including stress in the material, but is not generallywell understood. It is apparent from the data presentedherein that, for the given processing conditions, twinboundaries are most likely to develop in lines having widthsnear one micron. Since these lines were side by side withthose of other widths, it is reasonable to conclude that thedifferences in stress state are responsible for the higher frac-tion of twin boundaries; these boundaries, in turn, are re-sponsible for the weaker textures observed.
The observed in-plane orientation preference for thenarrow inlaid Cu lines shows that the annealed structure inthe line, for the processing conditions used, tends to a{111} surface normal and a [110] direction along the linelength. The {111} surface orientation is the low-energy sur-face in FCC materials and provides evidence of bottom-upgrowth of the Cu during electroplating. Side-wall growthwould result in a preferred orientation having the {111}pole aligned with the trench side wall. The [110] in-planeorientation component must be a result of either interfaceor grain boundary energy minimization. Interface energy
314 Scanning Vol. 25, 6 (2003)
minimization would require that the {211} plane alignedwith the trench sidewalls is the low-energy orientationplane for Cu grains in contact with the Ta barrier (forplanes normal to the {111} surface orientation). Grainboundary energy minimization would require that the{110} plane aligned with the grain boundary plane is thelow-energy position for [111] tilt boundaries. While inter-face energy data are not available, the energy of free sur-faces shows that the {112} plane is the lowest energy planenormal to the {111} surface orientation (Sundquist 1964).
Perhaps not coincidentally, a similar preferred structurewas observed in heavily annealed, free-standing, narrow Allines. Figure 8 is adapted from Field and Wang (1998),showing the preferred structure in the Al lines. The narrowline Cu microstructures discussed in this paper tend to thesame orientation preference as shown for the narrow Allines. This comparison suggests that the {211} plane in the{111} oriented Cu grains creates a low energy interfacewith the Ta barrier in a manner similar to the Al grainswhere it is presumed that free surface energy minimizationresults in the observed texture.
FIG. 8 Schematic showing the preferred orientation for both free-standing narrow Al lines (Field and Wang 1998) and the narrow Culines analyzed in this investigation.
FIG. 7 Twin boundary fraction in the Cu lines as a function of linewidth.
0.45
0.40
0.35
0.30
0.25
0.20
0.15
0.10
0.05
0.000 1 2
Line width (µm)
Twin
bou
ndar
y fr
actio
n
3
(111)
(112)(110)
4 5 6
Acknowledgments
One author, DPF, acknowledges Advanced Micro De-vices for financial support provided to Washington StateUniversity in support of this work.
References
Besser PR, Zschech E, Blum W, Winter D, Ortega R, Rose S, Her-rick M, Gall M, Thrasher S, Tiner M, Baker B, BraeckelmannG, Zhao L, Simpson C, Capasso C, Kawasaki H, Weitzman E:Microstructural characterization of inlaid copper interconnectlines. J Elect Mats 30, 320–330 (2001)
Engler O, Jura J, Matthies S: Influence of Number and Arrangementof Single Orientation Measurements on their Statistical Rele-vance. In Proc Twelfth Intl Conf on Textures of Materials (Ed.Szpunar JA). NRC Research Press, Ottawa (1999) 68–73
Field DP, Sanchez JE, Besser PR, Dingley DJ: Analysis of grainboundary structure in Al-Cu interconnects. J Appl Phys 82,2383–2392 (1997)
Field DP, Wang P-H: Development of a preferred grain boundarystructure in Al-Cu interconnects. In Boundaries and Interfacesin Materials: The David A. Smith Symposium (Eds. Pond RC,Clark WAT, King AH, Williams DB). Minerals, Metals and Ma-terials Society (1998) 323–328
Field DP: QC for IC interconnects? Vac Thin Films 2,1, 36–40 (1999)Field DP, Dornisch D, Tong HH: Investigating the microstructure-re-
liability relationship in Cu damascene lines. Scripta Mater 45,1069–1075 (2001)
D. P. Field et al.: EBSD characterization of Cu lines 315
Hau-Riege SP, Thompson CV: Modeling of texture evolution in cop-per interconnects annealed in trenches. Scripta Mat 41, 403–408(1999)
Humphreys FJ, Hatherly M: Recrystallization and related annealingphenomena. Pergamon 358–359 (1995)
Humphreys FJ, Huang Y, Brough I, Harris C: Electron BackscatterDiffraction of Grain and Subgrain Structures–resolution con-siderations. J Microsc 195, 212–216 (1999)
Lingk C, Gross ME, Brown WL: X-ray diffraction pole figure evi-dence for (111) sidewall texture of electroplated Cu in submi-cron damascene trenches. Appl Phys Lett 74, 682–684 (1999)
Matthies S, Vinel GW: On some methodical developments concern-ing calculations performed directly in the orientation space.Mat Sci Forum 157–162, 1641–1646 (1999)
Nowell MM, Field DP: Texture and grain boundary structure de-pendence of hillock formation in thin metal films. Mat Res SocSymp Proc 516, 115–121 (1998)
Sanchez JE Jr., Besser PR: Modeling microstructure development intrench-interconnect structures. Proc Intl Interconnect Technol-ogy Conference IITC98, 247–249 (1998)
Sundquist BE: A direct determination of the anisotropy of the surfacefree energy of solid gold, silver, copper, nickel, and alpha andgamma iron. Acta Metall Mater 12, 67–74 (1964)
Thompson CV, Carel R: Texture Development in PolycrystallineThin Films. Mats Sci Eng B32, 211–219 (1995)
Troost KZ: Towards submicron Crystallography in the SEM. PhilipsJ Res 47, 151–163 (1993)
Vanasupa L, Pinck D, Joo Y-C, Nogami T, Pramanick S, Lopatin S,Yang K: The impact of linewidth and line density on the textureof electroplated Cu in damascene-fabricated lines. ElectrochemSolid-State Lett 2, 275–277 (1999)
![ELECTRON BACKSCATTER DIFFRACTION CRYSTAL … · 2014. 8. 8. · Electron backscatter diffraction (EBSD) measurement [1,2] is generally very useful in analyzing crystal morphology](https://static.documents.pub/doc/80x56/6020ac1488b59757b674b100/electron-backscatter-diffraction-crystal-2014-8-8-electron-backscatter-diffraction.jpg)
