Chapter 5
SHE on Unstructured Grids

Up to now, the SHE equations have been derived and investigated on the continuous level. Generic discretization schemes were considered in the previous chapter and a system matrix compression scheme was developed. It is thus time to consider the discretization of the equations, which is the focus of this chapter. To make the chapter as self-contained as possible, the necessary prerequisites about the box integration scheme and the generation of suitable unstructured grids are presented first.

5.1 The Box Integration Scheme

The box integration scheme is a widely used discretization scheme, particularly for semiconductor device simulation and for fluid dynamics. Outside the field of microelectronics it is typically called finite volume method  [464]. The main advantage of the box integration scheme is its local conservation property, thus it is particularly suited for the discretization of conservation equations such as the drift-diffusion equations. The box integration can be derived in different ways. In this section a rather direct approach is taken, while a mathematically cleaner approach can be found in  [4], where connections with the finite element method are more pronounced.

In the following, the box integration scheme is explained by example of the discretization of the Poisson equation on the domain Ξ  . Consider

Δψ = f in Ξ n ,
ψ = 0 on Ξ ,
where homogeneous Dirichlet boundary conditions are assumed for simplicity. Given a tessellation         N
B = (Bi )i=1   of the domain Ξ  into small boxes Bi  , a local integral form of the Poisson equation is
BiΔψdV = BifdV on Ξ , ∂BiΞψdA = 0 .
Using Gauss’ Theorem, the problem is recast into
∂Biψ ndA = BifdV on Ξ , BiΞψdA = 0 ,
where n denotes the unit normal vector to the boundary of the box in outward direction.

Let the connecting line [xi,xj]  of the box centers xi  and xj  of the boxes Bi  and Bj  be perpendicular to the box boundary, cf. Fig. 5.1. A first-order approximation of the normal derivative of the potential is a differential quotient along the connecting line [xi,xj]  of length di,j  . Thus, one obtains the discrete local conservation equations

∑  ψj-−-ψiA   = f (x )V  ,                        (5.1)
     di,j    i,j      i  i
 j
where Vi  denotes the volume of the box Bi  and the interface area Ai,j  is zero if the boxes Bi  and Bj  do not have a common boundary. Note that the transition from the continuous to the discrete formulation requires an approximation of fluxes through the box boundaries. There exists a large number of different flux approximation schemes, an overview can be found e.g. in the textbook of LeVeque  [64].


PIC

Figure 5.1: Schematic of a mesh for a box integration scheme. The length between vertices xi and xj  is dij  , while the area of the interface between boxes Bi  and Bj  is Aij  .


To summarize, the main ingredients of a box integration scheme are as follows:

5.2 Construction of Boxes

Two families of tessellations B of the domain Ξ  are frequently used with a box integration scheme. The first family is referred to as vertex-centered, where the unknowns and the boxes Bi  are associated with the vertices of another mesh. The second family is based on a cell-centered approach, where the unknowns are directly associated with the cells of the mesh (e.g. triangles, tetrahedra, hexahedra), which also constitute the boxes B
  i  . Other formulations such as cell-vertex methods exist  [4], but are not further addressed in the following. The respective preference of a certain family is often based on the availability of good flux approximation schemes for the problem at hand. Due to the wide-spread use of vertex-centered box integration schemes in the field of semiconductor device simulation, the focus in this section is on this first family. In particular, the box integration scheme for the SHE equations presented in the next section is based on a vertex-centered construction.

While a cell-centered box integration scheme can be carried out on virtually any mesh, the vertex-centered method is a-priori based on a different mesh and the boxes in B need to be constructed first. The following two constructions are common  [4]:


PIC
(a) Circumcenter method (Voronoi diagram).
  
PIC
(b) Barycenter method.

Figure 5.2: Two methods for the construction of a dual box tessellation from a triangulation.


In order to employ a box integration scheme, there is usually no need to set up the boxes explicitly. As motivated in Sec. 5.1, it is usually sufficient to have the box volumes Vi  and the interface areas Ai,j  available. For the discretization of an additional advective term, the box volume fraction Vi,j  associated with the interface Ai,j  is also required, cf. Fig. 5.3. Since the computation of these values for triangular and tetrahedral meshes is only poorly documented in the relevant literature, the respective algorithms are detailed in the following.


PIC

Figure 5.3: Schematic of a triangular Delaunay mesh and its dual Voronoi diagram, where the box containing vertex xj  is highlighted. The Voronoi volume fraction Vi,j  and the interface area Ai,j  are associated with each edge [xi,xj]


Since Voronoi quantities are required based on connections between xi  and xj  , it is sufficient to provide an algorithm for the computation of the respective quantities for each edge [xi,xj]  . The Voronoi information for the full domain is then obtained by an iteration over all edges of the domain.

Algorithm 3 (Computation of Voronoi quantities for a triangular Delaunay mesh). Requirements: Triangular Delaunay mesh with all circumcenters in the interior of the mesh
Input: edge [xi,xj]

Return Ai,j,di,j,Vi,j  .

The box volumes Vi  for the box associated with vertex xi  is obtained by a summation over all box fractions, i.e.      ∑
Vi =   j Vi,j  . The three-dimensional equivalent is slightly more involved:

Algorithm 4 (Computation of Voronoi quantities for a tetrahedral Delaunay mesh). Requirements: Triangular Delaunay mesh with all circumcenters in the interior of the mesh
Input: edge [x ,x ]
  i  j

Return Ai,j,di,j,Vi,j  .

Note that the point p is used for the computation of the polygon defined by the circumcenters. Again, the box volumes Vi  are obtained by a summation over all Vi,j  related to a vertex xi  .

It has to be noted that the Delaunay property does not ensure that the circumcenters of all cells are inside the mesh, i.e. located inside or at the boundary of any cell of the mesh. Consequently, cells with regular shapes are required near the mesh boundary in order to fulfill the requirements of the two algorithms above, cf. Fig. 5.4. If this criterion is not strictly fulfilled, the system matrix S will be disturbed by, say, a matrix D . Similar to the analysis of round-off errors, the difference of the solution y of the perturbed system with respect to the true solution x is bounded by

∥y-−-x∥-       ∥D-∥
  ∥x∥   ≤ κ (S )∥S ∥ ,                          (5.3)
where κ(S )  denotes the condition number of S and suitable vector and matrix norms are chosen  [27]. Consequently, violations in the range of round-off errors can be justified from a mathematical point of view. Higher violations are acceptable if the condition number is known to be small and only low accuracy in the solution vector is required. In practice, however, even much stronger violations than those justified from the analytical bound (5.3) can result in insignificant errors in the solution vector due to high regularity of the underlying formation. Nevertheless, the regularity of the mesh near the boundary is particularly important for semiconductor device simulations with dominant current flow near the surface, as it is the case e.g. in a MOSFET.


PIC

Figure 5.4: Flat triangles at the domain boundary can lead to circumcenters outside the simulation domain. As a consequence, the volumes and interface areas of dual boxes extend outside the simulation domain (hatched area), leading to systematic errors in the assembled system of linear equations.


The generation of Delaunay triangulations of good quality for arbitrary domains is a challenging task already in two dimensions  [94], and particularly in three dimensions  [17]. The meshes used throughout this thesis are generated by Netgen  [91], which is capable of generating two- and three-dimensional Delaunay triangulations. A further discussion of the peculiarities of Delaunay mesh generation and refinement is, however, beyond the scope of this thesis.

The SHE method requires an energy coordinate in addition to the spatial coordinate. In principle, one may generate a mesh directly for the (x,H )  -space, but this is impractical for two reasons. First, the author is not aware of the existence any four-dimensional mesh generators capable of handling complex domains. Second, the essence of the H  -transform in Sec. 2.3 is the possibility to align the grid with the trajectory of electrons in free flight given by planes of constant total energy H  . Consequently, the embedding of a mesh from the spatial domain to the (x,H )  -space by a tensorial prolongation is of advantage. Given discrete total energies H1  < H2 < ...< HNH   , every vertex x in the n  -dimensional mesh is first embedded in the n+ 1  -dimensional space at location (x,H1 )  . The full n + 1  -dimensional mesh of prismatic cells is then obtained by repeatedly shifting the three-dimensional mesh along the energy axis to obtain the points (x, H )
    2  , (x,H3 )  , ...  , (x,HNH )  . The procedure is illustrated in Fig. 5.5 for the case of a spatially two-dimensional mesh. Clearly, the energy spacing does not need to be equidistant. Moreover, the computation of Voronoi quantities for the mesh in (x,H )  -space is obtained from Voronoi information in x -space by a multiplication of the energy spacing.

There is no need for setting up the mesh in (x, H )  -space explicitly, because the additional H  coordinate can also be handled implicitly. Since the coupling of different energy levels is local in space, it is more memory efficient to stick with the spatial mesh and to store the total energy values in an array with, say, NH   elements. The respective array index i  can then be used to identify e.g. the vertex (x,Hi )  , even though only the vertex x is explicitly stored. Since this approach leads to better compatibility with other macroscopic transport models, which do not require an additional energy coordinate, this implicitly tensorial representation is used in ViennaSHE. The price to pay is an additional effort required for for a visualization of the computed distribution function in one and two spatial dimensions, because the mesh in (x, H )  -space needs to be made available then.


PIC

Figure 5.5: A three-dimensional prismatic mesh can be obtained from a two-dimensional mesh by appropriate translations. The two-dimensional cells are then facets of the three-dimensional prism. Here, the remaining facets are obtained by joining corresponding vertices.


5.3 Box Integration for SHE

With the discussion of the box integration scheme in Sec. 5.1 and the computation of Voronoi information from Sec. 5.2, the discretization of the SHE equations (2.34) and (2.35) can now be tackled. For the sake of clarity, the SHE method for spherical energy bands in steady-state is considered. The general case is obtained without additional difficulties.

Due to the use of MEDS for stabilization, cf. Sec. 2.3, two different sets of equations (2.34) and (2.35) need to be considered. The even-order equations are discretized directly on the boxes Bi  of the underlying Voronoi tessellation B , while the odd-order equations are associated with the dual tessellation B˜ centered at the interfaces between the boxes, cf. Fig. 5.6. Note that the set  ˜
B is obtained by taking the union over all adjoint boxes Bi,j  of all boxes Bi  . The volume of adjoint the boxes       ˜
Bi,j ∈ B overlapping the neighboring boxes Bi ∈ B and Bj ∈ B is given by 2Vi,j  , cf. Fig. 5.3 and Sec. 5.2. Consequently, the even-order expansion coefficients are associated with the vertices of the mesh, while the odd-order expansion coefficients are associated with the edge midpoints. More precisely, the discretization of the expansion coefficients reads

             |B |NH
f   (x,H ) = ∑  ∑  α     χ   (x)χ  −  + (H) ,    l even ,        (5.4)
 l,m          i=1 n=1  i,n;l,m  Bi    [H n,Hn]
              ˜
            ∑|B |N∑H
fl,m (x,H ) =       βi,n;l,mχ ˜Bj(x)χ[H −n,H+n](H) ,    l odd ,         (5.5)
             j=1 n=1
where χBi  denotes the indicator function of the box Bi  , χ [H−n ,H+n ]   is the indicator function of the energy interval for the NH  discrete energies Hn  with lower and upper bound H −
  n  and H+
  n  , and B˜ refers to the set of adjoint boxes ˜B
 i  with indicator function χ
 ˜Bi  . The unknowns of the resulting linear system of equations after discretization are the coefficients αi,n;l,m  and βi,n;l,m  , where the βi,n;l,m  are eliminated in a preprocessing step of the linear solver as discussed in Sec. 4.3.


PIC

Figure 5.6: Schematic view of the box discretization for SHE. Even-order unknowns are associated with vertices (filled circles, blue box Bi ∈ B ), while odd-order unknowns are associated with edges (open circles, green box       ˜
Bi,j ∈ B ).


5.3.1 Discretization of the Even-Order Equations

Following the box integration procedure from Sec. 5.1, integration over energy from Hn−  to H+n  and over a box Bi  given on the spatial grid leads to

∫ H+n ∫
        ∇x ⋅jl′,m′fl′,m′ − F ⋅Γ l′,m ′fl′,m′dV dH
 H−n   Bi      l,m             l,m
               ∑  ∫ H+n ∫  [
         = -1--             Zl,mσ η(x, H ± ℏω η,H)f0,0(x, H ± ℏωη)Z0,0(H ± ℏωη)
           Y0,0 η   H−n   Bi
                                                                ]
                         − fl,mZl,m ση(x,H, H ∓ ℏω η)Z0,0(H ∓ ℏω η)dV dH  .
(5.6)

The first term on the left hand side is transformed to a surface integral using Gauss’ Theorem, and then the integrals on the left hand side are decomposed into the individual intersections with the neighboring adjoint boxes Bi,j  :

            +[                                                   ]
   ∑    ∫ H n  ∫        l′,m′              ∫           l′,m′
           −           jl,m  fl′,m′ ⋅ni,jdA −         F ⋅Γl,m  fl′,m ′dV   dH
Bi,j of Bi H n  ∂Bi∩Bi,j                    Bi∩Bi,j
            1  ∑  ∫ H+n∫  [
        =  ----            Zl,mσ η(x, H ± ℏωη,H )f0,0(x,H  ± ℏωη,t)Z0,0(H ±  ℏωη)
           Y0,0  η  H −n   Bi
                                                                ]
                         − fl,mZl,m ση(x, H, H ∓ ℏωη)Z0,0(H ∓ ℏωη) dV dH  ,
(5.7)

where ni,j  denotes the outward pointing unit vector with respect to Bi  and perpendicular to the common interface of Bi  and Bj  . Since for even l  the coupling terms  l′,m′
jl,m  and  l′,m′
Γl,m  take according to Thm. 1 nonzero values for odd  ′
l only, and the scattering operator exclusively couples terms of the same order (l,m )  , the discretization (5.5) can directly be inserted:

   ∑            [        ∫ H+n                     ∫ H+n         ]
        βi,n;l′,m ′ Ai,jni,j ⋅    jl′,m′dH  − Vi,jF|B  ⋅      Γ l′,m′dH
Bi,j of Bi                H −n  l,m              i,j   H−n   l,m
                          ∫  +
                  --1-∑     Hn [
                = Y0,0     H−   Zl,mσ η(x, H ± ℏω η,H )αi,n±n ′η;0,0Z0,0(H ± ℏωη)
                       η    n                                             ]
                                 − αi,n;l,mZl,m ση(x,H, H ∓ ℏω η)Z0,0(H ∓ ℏω η)VidH  ,
(5.8)

where  l′,m′
jl,m  ,  l′,m ′
Γl,m  as well as the scattering rates are assumed to be homogeneous over the device, the force F is approximated by a constant vector F i,j  on Bi,j  and where n′η  refers to a suitable index shift such that Hn+n ′η ≃ Hn + ℏω η  . Typically, discrete energies are chosen as fractions of ℏω η  such that Hn + ℏ ωη  is again a discrete energy. It is noteworthy that the approximation of the force term by a constant vector over Bi,j  is mostly a matter of convenience. One may also use more accurate representations, but the price to pay is increased computational effort for the evaluation of the integral over Bi,j  . A dimensional splitting has been used by Hong et al.  [42], which can be seen in this context as considering only the projection F ⋅ni,j  for the integration over Bi,j .

A slight rearrangement for better exposition of the unknowns αi,n;l,m  and βi,n;l,m  yields

               [         ∫  +                   ∫   +        ]
  ∑                        Hn  l′,m ′               Hn   l′,m ′
        βj,k;l′,m′ Ai,jni,j ⋅ H−n  jl,m dH  − Vi,jF i,j ⋅ H −n Γ l,m dH
Bi,j of Bi
                            Vi ∑  ∫ H+n
               = αi,n±n′η;0,0----      − ση(x,H  ± ℏωη,H )Zl,mdH
                           Y0,0  η  H n
                             V  ∑  ∫ H+n
                   − αi,n;l,m --i-        σ η(x, H,H  ∓ ℏωη)Z0,0(H  ∓ ℏωη)dH  .
                            Y0,0  η  H −n
(5.9)

In the case of spherical energy bands, the integrals on the left hand side can be rewritten using (4.2) and (4.3) as

∫ H+                 ∫ H+
    njl′,m′dH  = al′,m′   n vZdH   ,                   (5.10)
 H −n  l,m         l,m   Hn−
∫ H+n                 ∫ H+n
     Γ l′,m′dH = bl′,m ′     -Z--dH  .                  (5.11)
 H−n   l,m         l,m   H−n  ℏ|k|
The integrals on the right hand side can be calculated analytically for the parabolic dispersion relation (3.2) and the nonparabolic model (3.3), cf.  [42]. When using the full-band modification (3.20), the integral (5.11) simplifies to an evaluation of the full-band data at H −
  n  and   +
H n  .

5.3.2 Discretization of the Odd-Order Equations

The discretization of the odd-order projected equations (2.35) is carried out using essentially the same steps as for the even-order equations, but differs in a few technical details. Integration over the energy interval from   −
H n  to   +
Hn  and over an adjoint box Bi,j  overlapping the boxes Bi  and Bj  leads to

∫ H+ ∫
   n     jl′,m′⋅∇  f ′ ′ + F ⋅Γˆl′,m ′f ′ ′dV dH
 H−n   Bi,j l,m     x l,m        l,m  l,m
                   ∫ H+  ∫   [
          = -1--∑      n      Z   σ (x,H ± ℏω  ,H )f   (x, H ± ℏω )Z   (H ± ℏω  )
            Y0,0 η   H−n   Bi,j  l,m  η          η    0,0         η   0,0        η
                                                                   ]
                            − fl,mZl,m ση(x,H, H ∓ ℏω η)Z0,0(H ∓ ℏω η)dV dH  .
(5.12)

Assuming  l′,m ′
jl,m  to be piecewise constant in each of the adjoint boxes Bi,j  , application of Gauss’ Theorem to the first term, and splitting the second integral into the two overlaps with Bi  and Bj  leads to

∫ H+n [∫         ′ ′             ∫         ′ ′
              jll,,mm fl′,m′ ⋅ndA +         jll,m,m fl′,m ′ ⋅ndA  ,
 H−n    ∂Bi,j∩Bi                   ∂Bi,j∩Bj
      ∫            ′ ′          ∫            ′ ′       ]
    +         F ⋅ ˆΓ ll,m,m fl′,m ′dV +      F  ⋅ ˆΓ ll,,mm fl′,m ′dV dH
        Bi,j∩Bi                    Bi,j∩Bj
      ∑   ∫ H+n ∫   [
= --1-              Zl,m ση(x,H  ± ℏωη,H )f0,0(x,H ± ℏ ωη)Z0,0(H ± ℏ ωη)
  Y0,0 η   H−n   Bi,j
                                                            ]
                     − fl,mZl,mση(x,H, H  ∓ ℏωη)Z0,0(H  ∓ ℏωη) dV dH .
(5.13)

For odd l  , the terms jl′,m′
 l,m  and   ′ ′
ˆΓ l,m
 l,m  are nonzero for even l′ only. Inserting the expansion (5.4) on the left hand side and (5.5) on the right hand side, one arrives at

             [                                        ]
        ∫ H+n  ∫         l′,m ′        ∫           l′,m ′
αi,n;l′,m′   −           jl,m  ⋅ndA +         F ⋅ ˆΓ l,m dV  dH
         Hn    ∂Bi,j∩B[i               Bi,j∩Bi                  ]
               ∫ H+n  ∫         l′,m ′        ∫           l′,m ′
     + αj,n;l′,m ′  −            jl,m  ⋅ndA +         F ⋅Γˆl,m dV   dH
                Hn    ∂Bi,j∩Bj               Bi,j∩Bj
            2Vi,j ∑  ∫ H+n
 = βi,n±n ′η;0,0-----        Zl,mση(x, H ± ℏωη,H )Z0,0(H  ± ℏωη)dH
             Y0,0  η  H −n
                  ∑  ∫ H+n
     − βi,n;l,m2Vi,j         Zl,mσ η(x, H,H  ∓ ℏωη)Z0,0(H  ∓ ℏωη)dH  ,
              Y0,0  η   H−n
(5.14)

where the scattering rates and the density of states are assumed to be independent of the spatial variable. Since the current density expansion terms  l′,m′
jl,m  are also assumed to be constant within the box Bi,j  , an application of Gauss’ Theorem shows

∫                    ∫
         l′,m′                  l′,m ′
 ∂B  ∩B jl,m  ⋅ndA  +  ∂B  ∩B  jl,m  ⋅ndA
   i,j  i   ∫            i,j  j     ∫
         =         jl′,m′ ⋅nj,idA +         jl′,m′ ⋅ni,jdA ,
            ∂Bj∩Bi,j  l,m            ∂Bi∩Bi,j  l,m
(5.15)

which allows to replace the integration over the boundary of Bi,j  by two integrations over the interface of the boxes B
  i  and B
 j  . This leads with constant force approximations F
  i,j  over the box Bi,j  to

        ∫ H+n [    l′,m′                 l′,m′]
αi,n;l′,m ′  −   Ai,jjl,m  ⋅nj,i + Vi,jF i,j ⋅ ˆΓ l,m dH
         Hn   ∫   +
                H n [    l′,m ′               ˆ l′,m′]
     + αj,n;l′,m′   −  Ai,jjl,m  ⋅ni,j + Vi,jF i,j ⋅Γ l,m dH
                Hn  ∫  +
            2Vi,j∑     Hn
 = βi,n±n′η;0,0Y0,0     H−n  ση(x,H  ± ℏωη,H )Zl,mdH
                  η      +
             2Vi,j ∑  ∫ Hn
     − βi,n;l,m Y----      − ση(x,H, H ∓  ℏωη)Z0,0(H  ∓ ℏωη)dH  .
               0,0  η  Hn
(5.16)

The integrals over the coupling coefficients can again be further simplified as in (5.10) and (5.11).

5.4 Results

A comparison of the number of vertices required for a MOSFET and for a FinFET using structured and unstructured grids is given in the following. A comparison for a one-dimensional  +   +
n nn   -diode is omitted since there the structured and unstructured meshes coincide.

The schematic MOSFET layout shown in Fig. 5.7 consists of 1028  nodes. A much more aggressive coarsening could have been applied to the left and the right of the device, as well as deep inside the bulk of the device (bottom). Nevertheless, the structured grid shown at the right, which has the same grid spacing as the triangular mesh in the channel region, consists of 1594  nodes, which consequently leads to about 50  percent more unknowns. Note that the high resolution inside the channel induces an increased solution deep in the substrate, because so-called hanging nodes1 are prohibited for the box integration scheme. Therefore, a lower resolution, which would be typically sufficient deep in the substrate, cannot be obtained there. For demonstration purposes, only a coarse mesh has been chosen at the contacts. A finer grid at the contacts would result again in additional grid nodes in the bulk for the structured grid.

For a fully three-dimensional layout such as that of a FinFET, the difference in the number of grid points between structured and unstructured grids becomes much larger. Since the current flow is predominantly near the surface of the fin, a coarser mesh can be chosen in the center of the fin when using unstructured grids. At the transition from the source to the channel and particularly from the channel to the drain, a fine resolution is necessary in order to account for the high electric fields and the heated carriers in the latter case. The sample meshes of a FinFET depicted in Fig. 5.8 show that with only 4838  nodes a fine mesh can be obtained in the channel and towards the drain region, while a structured grid with comparable resolution in the channel leads to 27456  nodes, hence the difference is a factor of six. This difference mostly stems from the spurious high resolution of the structured grid in the source and drain regions.


PIC

Figure 5.7: Comparison of a triangular mesh of a MOSFET and a structured grid with the same resolution inside the channel. While the triangular grid consists of 1028  points, the structured grid leads to 1594  grid points.



PIC

Figure 5.8: Comparison of the tetrahedral unstructured grid used for the simulation of half of a tri-gate transistor (no gate, oxide and body shown), and a structured grid with the same mesh size inside the channel. While the tetrahedral grid consists of 4838  points, the structured grid is made up of 27456  points, thus leading to considerably higher computational costs. For simplicity, only a course mesh has been chosen around the contacts.