A

BOUNDARY ELEMENT METHOD

A.1 Introduction

The Boundary Element Method (BEM) is a numerical method to solve the boundary value problem by the boundary integral equation (BIE) method. Since the potential problem was first formulated in terms of a direct BIE and solved numerically by Jaswon [A.I] 30 years ago, extensive research effort has been made in BIE formulations of problems in mechanics and their numerical solution schemes. The direct formulation of BIE for the elastostatic problem was first presented by Rizzo [A.2] following the work of Jaswon. Later, the formulation was extended successfully by Cruse and Rizzo [A.3] and Cruse [A.4] to the elastodynamics. The name Boundary Element Method appeared in the late 70's in an attempt to make an analogy with the finite element method (FEM). Today, the BIE/BEM has gained a number of applications in many fields of computational mechanics, such as the fields of wave prop­agation, heat transfer, diffusion and convection, fluid flow, fracture mechan­ics, electrical problems, geomechanics, plates and shells, inelastic problems, contact problems, design sensitivity and optimization and inverse problems [A.5-A.14].

As a method based on transferring the governing differential equations into integral equations over the boundary, in linear elastostatics, the BEM leads to an integral constraint equation over the surface of the body, relating boundary tractions (pressures) to boundary displacements. Then the algebraic system of equations is solved for the unknown displacements and tractions (pressures) on the boundary. In this appendix we discuss the general formulation of the CBIE in linear elastostatics briefly. Detailed information about the formulation can be found in references [A.15-A.18].

298 A BOUNDARY ELEMENT METHOD

A.2 Governing Differential Equation of Elastostatics

Employing the usual tensor notation, where a repeated subscript denotes sum­mation over its range and a comma denotes partial differentiation, we can write the governing differential equation of equilibrium for an infinitesimal element in a elastic body as:

(1ij,j + Ii = 0 (A.I)

in which (1 ij are the stress components and Ii the body force vector.

Hooke's law for a homogeneous isotropic material states:

Ev ECij

(1ij = (1 _ 2v) (1 + v) 8ijCkk + 1 + v (A.2)

where Cij" E and v are the strain, Young's modulus and Poisson's ratio, respectively and 8ij is the Kronecher delta symbol.

The strains and displacements are related by:

1 c-- - -(u- -+U--) '3 - 2 -,3 3,- (A.3)

Substituting Eqs. A.2 and A.3 into Eq. A.I, we can express the equilibrim equation for as a function of the displacements components:

1 2fi(1-v) Ui,jj + 1 _ 2v Uj,ji + E = 0 (AA)

This is called the Navier equation. It is the governing differential equation of elasticity which has to be solved subject to certain boundary conditions. After solving this equation, the unknown tractions (pressures) and displacements on the boundary can be obtained.

A.3 Betti's Reciprocal Theorem and Kelvin's Solution

Betti's general method of integrating the equations of elasticity is one of the classical methods of solving the Navier equation AA. Consider two equilibrium states of a 2-D elastic body, one with displacements, Ui, due to a body force

A.3 Betti's Reciprocal Theorem and Kelvin's Solution 299

Ii. and surface tractions ti, and the other with displacements, u~, due to a body force f: and surface tractions t~. Then the Betti's reciprocal theorem for a two-dimensional domain D with boundary L, as shown in Figure A.1, is given by:

J tiu~dL + J fiU~dD = J t~uidL + J fIui dD (A.5) L D L D

L o

Boundary L

Fig. A.!. Two dimensional elastic body D with boundary L

One of the fundamental solutions of Navier equation is the classic Kelvin solu­tion which is the displacement field u(x) due to an i-directional concentrated unit force F within an elastic body. For plane strain problems [A.1-A.2], the Kelvin solution is:

(1+v) [ 1 ar ar] Uij = 47rE(1 _ v) (3 - 4v) In ;:Oij + ax; aXj (A.6)

T. .. =- 1 {g~[(1-2V)Oij+2:;;:;;]} 'J 47r(1- v)r +(1- 2v)(n·'!!!:" - n·'!!!:") 'ax; J ax;

(A.7)

300 A BOUNDARY ELEMENT METHOD

where Uij and Tij represent the jth directional displacement and traction at any point p caused by a ith directional unit load F applied at point Po, r the distance between Po and p, and ni the directional cosines of the normal n, Fig. A.I.

By utilizing Kelvin's concentrated load solution as the prime states in Eq. A.5, an integral equation is obtained, which relates the displacements at an interior point Po, to the displacements and tractions on the surface:

Ui(PO) = - J Tij (Po, p)Uj (P)dL + J Uij (po,p)tj (p)dL + J Uij (Po, p)!J(p)dD L L D

(A.8) Note that because f' = F = 1 is an i-directional concentrated force at Po, there is

J fIui(P)dD = 1 x Ui(PO) = Ui(PO) D

(A.9)

in Eq. A.5. This result is known as Somigliana's identity for displacements and is a continuous representation for the displacements at any interior point Po in domain D.

A.4 Boundary Integral Equation

The Boundary Integral Equation is obtained by taking the point Po to the boundary. Po could be located at a smooth surface or at the intersection of several smooth surfaces, that is, at an edge or corner on the boundary. Let us employ Eq. A.8 for the domain L + L'1/ as shown in Fig. A.2:

Ui(PO) + J Tij(Po,p)uj(p)dL

L+L"

= J Uij(Po,p)tj(p)dL + f Uij(Po,p)!J(p)dD L+L" D+D"

(A.lO)

where D is the domain with the boundary L, D'1/ is the exclusion area of a circle with radius 'fJ centered at Po and boundary L'1/. Let 'fJ~ 0, the following boundary equation is derived:

A.5 Numerical Formulation of BIE 301

Cij(pO)Uj(pO) = J [Uij(pO,P)tj(p) - Tij(pO,p)Uj(p)] dL+ J Uij (Po, p)/j (p)dD L D

(A.ll) where Cij(po) is a 2x2 matrix depending on the smoothness of the curve L at the point Po. Note that Cij(Po)=O when i 1= j.

Body 0

Fig. A.2. Elastic body D with Po on the boundary L

The above equation is the boundary constraint equation relating surface trac­tions to surface displacements, and includes the effect of body forces. If the body force at any place is zero, Eq. A.ll can be written as:

Cij(pO)Uj(po) = J [Uij(po,p)tj(p) - Tij(pO,P)Uj(p)] dL L

A.5 Numerical Formulation of Boundary Integral Equation

(A.12)

In general, the analytical solution to the boundary integral equation A.12 for bodies of complicated shape is difficult to obtain and a numerical method which reduces the integral equation to linear algebraic equations has to be used. In order to solve the integral equation numerically, the boundary will be discretized into a series of elements (Here isoparametric elements are used. For detailed information on isoparametric elements, please review books on FEM Or BEM.) over which displacements and tractions are written in terms of their values at a series of nodal points. Writing the discretized form of A.12

302 A BOUNDARY ELEMENT METHOD

for every nodal point, a system of linear algebraic equations is obtained. Once the boundary conditions are applied the system can be solved to obtain all the unknown values and consequently an approximate solution to the boundary value is obtained.

Let us consider how Eq. A.12 can be discretized to find the system equations. Assume for simplicity that the boundary of the body involved is divided into N linear elements (from LlLl to LlLN) as shown in Fig. A.3. There are N nodes from Nl to NN on the boundary when N linear elements are used.

o x

I I

Node Po

Fig. A.3. An elastic body with disretized boundary L

Then, Eq. A.12 may be written approximately as:

N N

Cij(po)Uj(Po) = L J Uij (Po, p)tj (p)dL - L J Tij(PO,p)uj(p)dL k=lLlLk k=lLlLk

(A.13) There are now two types of integrals to be carried out over the elements LlLk ,

i.e.,

A.5 Numerical Formulation of BIE 303

1 J Uij(Po,p)tj(P)dL(P) = J Uij(po,p)tj{e)J{e)~ ~Lh -1

1 J Tij(PO,P)Uj(P)dL(P) = J Tij (Po, p)Uj(e)J(e)de (A.14)

~L/c -1

where e is the natural coordinate corresponding to point p and J (e) is the Jacobian function [A.20j. We assume that the linear element ilL" has two nodes N" and Nk+1 respectively. Then the value of Uj and tj at any point p on the element can be defined in terms of their nodal values and two linear interpolation functions (also called as shape functions) !P1 and !P2, which are given in terms of the natural coordinate e, i.e.

tj(e) = !P1(e)tj + !P2(e)tj+1

Uj(e) = !P1(e)uj+1 + !P2(e)uj+1 (A.15)

where tj, uj, tj+1 and uj+1 are the nodal tractions and displacements in the jth direction respectively.

Substitution of Eq. A.15 into Eq. A.14 yields:

J Uij (Po, p)tj (P)dL = [J Uij(PO,p)!p1(e)J(e)~] tj ~Lh -1

+ [1 U'; (po ,P)!I', (e)J(e)d{] If' (A.16)

J Tij(PO,p)Uj(P)dL = [J Tij(Po,p)!p1(e)J(e)~] uj ~L/c -1

+ [1 T,; (po ,p)!I', (e)J(e)d{] U;+' (A.17)

Applying the above procedure to every element, Eq. A.13 can be written as:

304 A BOUNDARY ELEMENT METHOD

where t~ and u~ is the jth directional traction and displacement on node N k ,

and G~ and Ht are values evaluated directly or indirectly through Eq. A.14 or Eqs. A.16 and Eq. A.17. These integrals can be calculated using numerical integration formula such as ordinary Gauss-quadrature rules when Po is not on element LlLk:

1 f Uij(Po,p)tj(p)dL = f Uij (Po,p)tj (e)J(e)d{ ilL" -1

M

= L Uij(PO,Pm)tj(Pm)J(em) m=1

1 f Tij(Po,p)Uj(p)dL = f Tij (Po,p)Uj (e)J(e)d{ ilL" -1

M

= L Tij(Po,Pm)Uj(Pm)J(em) m=1

(A.19)

where e is the natural coordinate corresponding to integration point Pm, J(e) the Jacobian function and M the number of Gaussian integration points.

In the following two situations the ordinary Gauss-quadrature applied to Eq. A.19 cannot evaluate the integral accurately and efficiently.

i. The Po is on the element. In this case, the integral will become a singular integral (Note the singular terms l/r and In(llr) in Eqs. A.6 and A.7) and special methods have to be used. Fortunately mature methods such as the rigid body translation method and special logarithm Gaussian quadrature [A.2l] can be employed to deal with them respectively.

ii. The Po is not on the element but close to the element. In this case, the integrals will become nearly singular and the ordinary Gauss-quadrature is no longer practical since a huge number of integration points is needed. The evaluation of the nearly singular integral is the most important work in BEM modeling of thin structures. Most of the earlier methods are either inefficient or cannot provide accurate results when the ratio of the po-element distance to the element size is smaller than 10-2 • They are not

A.6 System Equations 305

very useful in dealing with thin structures, such as thin films or coatings, with thickness to length ratio in the micro- or nano-scale. In [A.19], a semi-analytical method which can evaluate such integrals accurately and efficiently is discussed in detail.

After handling the above integrals properly and assuming that Po is the kth node on the boundary L, then Eq. A.IS can be written as:

Cij (k)U1 = [Gtj,G~j,G~j, .... G~j' ... G~-l,G~][tJ' tltJ, .... tf-l,tfJT [ l' 2' 3' k' N-1' N'][ 1 2 3 N-1 N]T -Hij,Hij,Hij,· .. Hij ... Hij ,Hij Uj,Uj.,Uj, .... Uj ,Uj

(A.20) where Cij(k) = Cij(Po). Assembling Eq. A.20 generates the following equa­tion:

[G1 G2 G3 Gk GN- 1 GN] [t1 t2 t3 tN- 1 tN]T ij , ij , ij'···· ij'··· ij , ij j' j.' j,.... j , j - [HI H2 H3 Hk H N- 1 HN][ 1 2 3 N-1 N]T - ij' ij' ij,··· ij··· ij , ij Uj , Uj., Uj, .... Uj 'Uj

(A.2I)

where sub-matrix Ht = Ht; - Cij(k). Note that Gij and Hij are 2x2 matrix in two-dimensional elasticity.

A.6 System Equations

By allowing every node (from N 1to NN; see Fig. A.3) on the boundary L to be Po, N equations like A.2I can be obtained. Assembling them together in matrix form gives the global system equations [H][U]= [G][T], i.e.

306 A BOUNDARY ELEMENT METHOD

=

(A.22)

The vectors U and T in above equation represent all the values of nodal displacements and tractions before applying boundary conditions. These con­ditions can be introduced by rearranging the columns in H and G, passing all unknowns to a vector X on the left hand side. This gives the final system of equations, i.e.

[AjX = [Bj (A.23)

All boundary values will be fully determined by solving the above system. Furthermore, by applying the boundary values into Eq. A.8, the displacement value in any domain point can be obtained.

A.7 BEM VS. FEM

For computational modeling of elastic bodies/structures, besides BEM, an­other conventional method is the finite element method (FEM) [A.20j. In comparison with FEM, BEM has the following advantages:

A.S References 307

i. For the analysis of thin structures, FEM is usually based on plate and shell theories. However, most plate and shell theories are based on various as­sumptions about the geometry, loading and deformation of the structure, and therefore the accuracy and reliability of the FEM for thin structures is in doubt. This is especially true for the stress analysis of thin structures since plate and shell models can not predict the normal stresses (contact stresses) accurately. Without the above assumptions, BEM is in general more accurate in stress analysis of thin structures such as the silicon wafer in CMP [A.19].

ii. In BEM only the boundary of the body f structure needs to be meshed while in FEM the whole body needs to be meshed. Therefore, the memory requirement of BEM is lower and the computation time is less.

iii. In BEM, remeshing is minimal when the geometry of the body fstructure is changed. For instance, when the thickness of the wafer is changed, there is no need to remesh the boundary of the wafer but only to change the size of the boundary elements on the left and right sides of the wafer. In FEM, however, the whole body/domain needs to be remeshed. Therefore, BEM is more convenient than FEM on problems of geometry optimizations.

A.8 References

[A.l] M. Jaswon, "Integral equation methods in potential theory, I." Proc. Roy. Soc. Ser. A., Vol. 275, pp. 23-32, 1963.

[A.2] F. J. Rizzo, "An integral equation approach to boundary value prob­lems of classical elastostatics," Quarterly of Applied Mathematics, Vol. 25, pp. 83-95, 1967.

[A.3] T. A. Cruse and F. J. Rizzo, "A direct formulation and numerical solution of the general transient elastodynamic problem I," Journal of Math. Analysis and Applications, Vol. 22, p. 244, 1968.

[AA] T. A. Cruse, "A direct formulation and numerical Solution of the gen­eral transient elastodynamic problem II," Journal of Math. Analysis and Applications, Vol. 22, pp. 341, 1968.

[A. 5] R. D. Ciskowski and C. A. Brebbia, Boundary Element Methods in Acoustics, Elsevier Applied Science, New York, U. S. A., 1991.

[A.6] L. C. Wrobel and C. A. Brebbia, Boundary Element Methods in Heat Transfer, Elsevier Applied Science, New York, U. S. A., 1992.

[A.7] P. K. Banerjee and L. Morino, Boundary Element Methods in Non­linear Fluid Dynamics, Elsevier Applied Science, New York, U. S. A., 1990.

[A.8] T. A. Cruse, Boundary Element Analysis in Computational Fracture Mechanics, Kluwer Academic Publishers, Boston, U. S. A., 1988.

[A.9] S-K. Hsu, "Boundary element modeling of electric and displacement fields in two dimensional piezoelectric material," M.S. Thesis, Dept.

308 A BOUNDARY ELEMENT METHOD

of Mechanical and Aerospace Engineering, Case Western Reserve Uni­versity, Cleveland, OH, U. S. A., 1994.

[A. 10] W. S. Venturini, Boundary Element Method in Geomechanics, Springer­Verlag, New York, U. S. A., 1983.

[A.ll] D. E. Beskos, Boundary Element Analysis of Plates and Shells, Springer­Verlag, New York, U. S. A., 1991.

[A.12] G. Karami, A Boundary Element Method for Two Dimensional Con­tact Problems, Springer-Verlag, New York, U. S. A., 1989.

[A.13] J. C. F. Telles, The Boundary Element Method for Inelastic Problems, Springer-Verlag, New York, U. S. A., 1983.

[A.14] D. B. Ingham and L. C. Wrobel, Boundary Integral Formulations for Inverse Analysis, Computational Mechanics Publication, Boston, U. S. A., 1997.

[A.15] P. K. Banerjee, The Boundary Element Methods in Engineering, McGraw­Hill, New York, U. S. A., 1994.

[A.16] C. A. Brebbia and S. Walker, Boundary Element Techniques in Engi­neering, Newnes-Butterworths, London, U. K., 1980.

[A.17] C. A. Brebbia and J. Dominguez, Boundary Elements: an Introductory Course, McGraw-Hill, New York, U. S. A., 1989.

[A.18] J. H. Kane, Boundary Element Analysis in Engineering Continuum Mechanics, Prentice-Hall, New Jersey, U. S. A., 1994.

[A.19] J. F. Luo, Y. J. Liu and E. J. Berger, "Analysis of two dimensional thin structures (from micro- to nano-scales) using the boundary element method," Computational Mechanics, Vol. 22, pp. 404-412, 1998.

[A.20] R. D. Cook, D. S. Malkus and M. E. Plesha, Concepts and Applications of Finite Element Analysis, John Wiley & Sons, New York, U. S. A., 1989.

[A.21] D. V. Griffiths and I. M. Smith, Numerical Methods for Engineers: A Programming Approach, CRC Press, Boca Raton, Florida, U. S. A., 1991.

Index

Active abrasive, 67, 97 Active abrasive number, 67, 97, 102 Active abrasive particles, 67 Active abrasive size, 97 Effective contact area, 99 'Swarm' simulation, 288

Abrasive particle size control, 290 Abrasive particle, pad asperity and

wafer interactions, 53 Abrasive size distribution, 68, 97 Abrasive weight concentration, 115 Average size of active abrasives, 66 Average size of abrasives, 99

BEM,297 Boundary conditions in CMP, 244 Boundary element method, 297 Boundary element model, 177, 264 Boundary integral equation, 300 Bulk Pad, 156

Chemical-mechanical planarization, 1 Chemical-mechanical polishing, 1 CMP,1 CMP configureations, 257 CMP input parameters, 7 CMP output parameters, 8 Comprehensive CMP models, 28 Constant element, 240 Consumable related parameters, 7 Consumables, 54 Contact area, 61 Contact mechanics model, 178

Contact pressure, 62, 99 Cook's model, 18 Copper damascene process, 3 Copper dishing, 148 Copper line width, 153 Copper thinning, 148 Coupling effects of slurry chemi­

cals, abrasive particle size and wafer-pad contact area, 115

Damascene process, 147 Design for manufacturability, 16, 289 Die-Scale Wafer Topography, 153 Dishing, 148, 164, 197 Dishing as a function of pattern density

and line width, 197 Displacement condition, 244 Dual damascene process, 4 Dummy filling, 39, 229

Edge effects, 42, 266 Effective pattern density, 32, 153, 221 Effective pattern density window, 35 Effective stiffness of pad asperity layer,

182 Effects of abrasive size distribution, 97 Effects of oxide and copper removal

mechanism on erosion and dishing, 200

Effects of pad properties on erosion and dishing, 204

Effects of pressure on dishing and erosion, 206

Electronics computer aided design, 39

310 Index

Erosion, 148, 164

Feature pattern density, 34 Feature- and die-scale modeling, 11,287 Feature- and die-scale models, 32 Feature-die-wafer scale integration, 280 Feature-scale model for standard-cell

(hierarchical) layouts, 220 Feature-scale wafer topography, 152 Formulations of material removal rate,

71 Four-parameter model, 156 Framework of the integrated CMP

model, 45

Green CMP, 290

ILD IIMD Planarization, 2 Integrated CMP model, 45

Kaufman's model, 26, 118

Lateral directional die-feature scale integration, 220

Layout pattern density, 34 Line width, 153 Line width dependency of topography

evolution, 176 Linear elasticity model, 154 Linear viscoelastic pad assumption, 164 Linear viscoelasticity model, 156 Low-k materials, 3

Machine configuration related parame­ters,8

Material removal mechanism in CMP, 53

Material removal rate, 17, 102 Material removal regions, 115 Maxwell element model, 156 Micro-injection molding of polishing

pad topography, 290 Mini-filling, 39 Mini-variation, 39 MIT model, 32 MRR, 17, 53, 71

Normal distribution of abrasive particle size, 67

Overpolishing stage, 163 Oxide erosion, 148, 163, 193 Oxide erosion as a function of pattern

density and line width, 193 Oxide erosion as a function of time, 195

Pad asperities, 61 Pad asperity layer, 156, 177 Pad asperity layer deformation, 177 Pad contact density, 223 Pad hardness, 65 Pad surface, 61 Pad Topography, 61 Particle-scale modeling, 10, 285 Particle-scale models, 17 Pattern density, 153 Pattern density effects, 32 Pattern density window, 221 Physical models, 38 Pitch width, 153 Planarization length, 154, 221 Plastic deformations over wafer-particle

and pad-particle interfaces, 63 Polishing pad-wafer interactions, 22 Pressure condition, 245 Pressure distribution, 261 Pressure distribution models based on

semi-solid-solid and solid-fluid­solid contact, 42

Pressure distribution models based on solid-solid contact, 41

Preston's equation, 18, 53, 83

Quadratic element, 242

Retaining ring, 265 Revised Preston's equation, 18 Reynold's equation, 270 Rigid flat punch indentation model, 261

Semi-two dimensional model, 176 Shallow trench isolation, 6 Single damascene process, 4 Sliding indentation, 64 Slurry abrasive concentration, 70 Slurry chemical-polishing pad interac-

tions,27 Slurry chemical-slurry particle

interactions, 27 Slurry chemical-wafer interactions, 25

Slurry particle-polishing pad interac­tions,23

Slurry particle-wafer surface interac-tions, 18

Solid-fl~d-solid contact, 269 Solid-solid contact mode, 56, 261 Stable contact, 99 Step height evolution model, 148 Step height reduction model, 36 Stribeck curve, 269 Surface quality, 17

Thinning, 148 Three-dimensional CMP models, 38 Three-dimensional feature-scale

modeling, 237 Topography evolution model, 148 Transition of three stages, 186 Two-dimensional boundary element

model,177 Two-dimensional contact mechanics

model,178

Index 311

Velocity distribution, 257 Vertical directional die-feature scale

integration, 214 Voigt-Kelvin element model, 156

Wafer hardness, 65 Wafer related parameters, 7 Wafer-abrasive-pad contact, 98 Wafer-pad contact at die-scale, 166 Wafer-pad contact at the feature-scale,

159 Wafer-scale modeling, 12, 287 Wafer-scale modeling of CMP, 255 Wafer-scale models, 41 Weight density function, 35, 221 WIDNU,32 With-in die non-uniformity, 32 With-in wafer non-uniformity, 41, 255 WIWNU, 41, 255

Zhao and Shi's model, 22, 54,83