21
Finite Element Modelling of Photonic Crystals Ben Hiett J Generowicz, M Molinari, D Beckett, KS Thomas, GJ Parker and SJ Cox High Performanc e Computing

Finite Element Modelling of Photonic Crystals Ben Hiett J Generowicz, M Molinari, D Beckett, KS Thomas, GJ Parker and SJ Cox High Performance Computing

Embed Size (px)

DESCRIPTION

High Performance Computing The Physics… F Solve Maxwell Equations with periodic boundary conditions F Numerical method based on Finite Element Method

Citation preview

Page 1: Finite Element Modelling of Photonic Crystals Ben Hiett J Generowicz, M Molinari, D Beckett, KS Thomas, GJ Parker and SJ Cox High Performance Computing

Finite Element Modelling of Photonic Crystals

Ben HiettJ Generowicz, M Molinari, D Beckett,

KS Thomas, GJ Parker and SJ Cox

HighPerformance Computing

Page 2: Finite Element Modelling of Photonic Crystals Ben Hiett J Generowicz, M Molinari, D Beckett, KS Thomas, GJ Parker and SJ Cox High Performance Computing

HighPerformance Computing

Photonic Crystals Photonic Crystals: the presence

of ‘photonic band gaps’ Huge potential in a range of

applications. Highly efficient narrow-band

lasers, integrated optical circuits, high-speed optical communication

networks. Hence a need for accurate and

efficient modelling.

Page 3: Finite Element Modelling of Photonic Crystals Ben Hiett J Generowicz, M Molinari, D Beckett, KS Thomas, GJ Parker and SJ Cox High Performance Computing

HighPerformance Computing

The Physics… Solve Maxwell Equations with periodic

boundary conditions

Numerical method based on Finite Element Method

0)(

)()()(

1 2

rH

rHrHr c

Page 4: Finite Element Modelling of Photonic Crystals Ben Hiett J Generowicz, M Molinari, D Beckett, KS Thomas, GJ Parker and SJ Cox High Performance Computing

HighPerformance Computing

Advantages of FEM

Simple and intuitive representation of the photonic crystal structure

Accurate expression of the sharp discontinuities in dielectric constant

0

)sin()(n

nn nxAxf

12

1

Page 5: Finite Element Modelling of Photonic Crystals Ben Hiett J Generowicz, M Molinari, D Beckett, KS Thomas, GJ Parker and SJ Cox High Performance Computing

HighPerformance Computing

Advantages of FEM Adaptive mesh refinement

allows improvement of solution in specific areas of high relative error

The global matrices comprising the eigenvalue problem are SPARSE. Hence the method scales (almost) linearly in terms of computation and memory requirements

Visualisation of a Global Matrix (non-zero elements

highlighted)

Page 6: Finite Element Modelling of Photonic Crystals Ben Hiett J Generowicz, M Molinari, D Beckett, KS Thomas, GJ Parker and SJ Cox High Performance Computing

HighPerformance Computing

Domain DiscretisationDomain Discretisation

Selection of Interpolation Function

Derivation of the Elemental Equations

Matrix Assembly

Solution of the Eigensystem

Visualisation

Unit cell Periodically tiled unit cells

Page 7: Finite Element Modelling of Photonic Crystals Ben Hiett J Generowicz, M Molinari, D Beckett, KS Thomas, GJ Parker and SJ Cox High Performance Computing

HighPerformance Computing

Interpolation Function

Linear, quadratic, cubic… Trade off between

computation/memory cost and solution accuracy

Domain Discretisation

Selection of Interpolation Function

Derivation of the Elemental Equations

Matrix Assembly

Solution of the Eigensystem

Visualisation

x

y

y

1

2

3 1y 2y

3y

x

y

1

2

3

3y

1y 2y

Page 8: Finite Element Modelling of Photonic Crystals Ben Hiett J Generowicz, M Molinari, D Beckett, KS Thomas, GJ Parker and SJ Cox High Performance Computing

HighPerformance Computing

Elemental EquationsDomain Discretisation

Selection of Interpolation Function

Derivation of the Elemental Equations

Matrix Assembly

Solution of the Eigensystem

Visualisation

0)(

)()()(

1 2

rH

rHrHr c

BkA λ)(

dxiiA ljjl )()( kk

dxxB ljjl )(

Page 9: Finite Element Modelling of Photonic Crystals Ben Hiett J Generowicz, M Molinari, D Beckett, KS Thomas, GJ Parker and SJ Cox High Performance Computing

HighPerformance Computing

Matrix AssemblyDomain Discretisation

Selection of Interpolation Function

Derivation of the Elemental Equations

Matrix Assembly

Solution of the Eigensystem

Visualisation

1

23

Global matrix assembly via local to global node mapping of elemental matrices

Page 10: Finite Element Modelling of Photonic Crystals Ben Hiett J Generowicz, M Molinari, D Beckett, KS Thomas, GJ Parker and SJ Cox High Performance Computing

HighPerformance Computing

Solution of the Eigensystem

Computationally Expensive (~95%) Needs to be efficient Sub-space iterative technique Only compute eigenvalues of

interest (lowest) Exploit similarity of adjacent

solutions Search a larger sub-space to

improve convergence

Domain Discretisation

Selection of Interpolation Function

Derivation of the Elemental Equations

Matrix Assembly

Solution of the Eigensystem

Visualisation

BkA λ)(

Page 11: Finite Element Modelling of Photonic Crystals Ben Hiett J Generowicz, M Molinari, D Beckett, KS Thomas, GJ Parker and SJ Cox High Performance Computing

HighPerformance Computing

VisualisationDomain Discretisation

Selection of Interpolation Function

Derivation of the Elemental Equations

Matrix Assembly

Solution of the Eigensystem

Visualisation

Page 12: Finite Element Modelling of Photonic Crystals Ben Hiett J Generowicz, M Molinari, D Beckett, KS Thomas, GJ Parker and SJ Cox High Performance Computing

HighPerformance Computing

Convergence Does the method work

in practice? Converges to analytic

solution for free space Verified against other

structures in the literature

Compared with experimentMesh Granularity (no. of elements)

Erro

r

Page 13: Finite Element Modelling of Photonic Crystals Ben Hiett J Generowicz, M Molinari, D Beckett, KS Thomas, GJ Parker and SJ Cox High Performance Computing

HighPerformance Computing

Triangular Lattice Spectra

Filling fraction = 80%

Rod dielectric constant = 12.25

Unit Cell for

calculation

Periodic repeat of unit cell

Page 14: Finite Element Modelling of Photonic Crystals Ben Hiett J Generowicz, M Molinari, D Beckett, KS Thomas, GJ Parker and SJ Cox High Performance Computing

HighPerformance Computing

Computational Efficiency Can get more accurate solution by

Using a finer mesh Using a higher order interpolation function

A compromise is necessary

Page 15: Finite Element Modelling of Photonic Crystals Ben Hiett J Generowicz, M Molinari, D Beckett, KS Thomas, GJ Parker and SJ Cox High Performance Computing

HighPerformance Computing

Which is Best ?

First order is cheap but inaccurate

Higher order gives better accuracy for same compute time

Increa

sed Accu

racy

Reduc

ed Tim

e

Page 16: Finite Element Modelling of Photonic Crystals Ben Hiett J Generowicz, M Molinari, D Beckett, KS Thomas, GJ Parker and SJ Cox High Performance Computing

HighPerformance Computing

Interesting geometries Exhibit large photonic

band gaps Substrate material can

have low dielectric constant

Practical to manufacture 12-fold symmetric

quasicrystals…

Page 17: Finite Element Modelling of Photonic Crystals Ben Hiett J Generowicz, M Molinari, D Beckett, KS Thomas, GJ Parker and SJ Cox High Performance Computing

HighPerformance Computing

Quasicrystal Spectra

Filling fraction = 28% Rod dielectric constant = 12.25

Page 18: Finite Element Modelling of Photonic Crystals Ben Hiett J Generowicz, M Molinari, D Beckett, KS Thomas, GJ Parker and SJ Cox High Performance Computing

HighPerformance Computing

Quasicrystal Spectra (2)Filling fraction = 28%

Rod dielectric constant = 4.0804

Page 19: Finite Element Modelling of Photonic Crystals Ben Hiett J Generowicz, M Molinari, D Beckett, KS Thomas, GJ Parker and SJ Cox High Performance Computing

HighPerformance Computing

Quasicrystal with Defect

Filling fraction = 28% Rod dielectric constant = 12.25

Page 20: Finite Element Modelling of Photonic Crystals Ben Hiett J Generowicz, M Molinari, D Beckett, KS Thomas, GJ Parker and SJ Cox High Performance Computing

HighPerformance Computing

Conclusion Finite Elements have advantages

simple and intuitive crystal representation Dielectric discontinuities are modelled accurately Resultant eigenvalue problem is SPARSE

Can use FEM modelling to tune photonic crystal properties

Also have fully 3D extension

Page 21: Finite Element Modelling of Photonic Crystals Ben Hiett J Generowicz, M Molinari, D Beckett, KS Thomas, GJ Parker and SJ Cox High Performance Computing

HighPerformance Computing

Further Information

E-mail: [email protected]

Web: www.photonics.n3.net

HighPerformanceComputingCentre