## Abstract

The propagation of the long-range surface plasmon-polariton mode in waveguides comprised of a curved thin metal film of finite width embedded in a homogeneous background dielectric is described. The curve and its mode are modelled in cylindrical coordinates using a rigorous vectorial numerical method and an absorbing boundary condition is applied on the radiating side of the bend. From the results obtained, it is confirmed that long-range structures are not incompatible with bending and that reasonably small radii of curvature can be used.

© 2006 Optical Society of America

## 1. Introduction

The design of curved waveguides is of central importance in integrated optics [1] since achieving efficient directional change is necessary in numerous structures such as offsets (e.g.: S-bend), splitters and combiners (e.g.: Y-junction), Mach-Zehnder interferometers (e.g.: back-to-back Y-junctions) and couplers (e.g.: coupled S-bends). Achieving an efficient directional change implies minimising for instance the area or path length required for the curve while minimising its insertion loss. The circular curved waveguide is a suitable structure for achieving directional change and its analysis as a dielectric structure has been the subject of intense research for at least a few decades [2].

Recently a straight thin metal film of finite width surrounded by a dielectric has been characterised theoretically and proposed as the foundation waveguide for a new integrated optics technology [3]. This symmetric “metal stripe” waveguide supports as a fundamental mode, a long-range (low-loss) surface plasmon-polariton (LRSPP) wave, identified as the ${\mathit{\text{ss}}}_{b}^{\mathit{0}}$ mode [3]. The existence of this mode and its properties have been verified experimentally [4–6]. While numerous elements and devices incorporating curved metal stripes propagating the long-range ${\mathit{\text{ss}}}_{b}^{\mathit{0}}$ mode have been demonstrated [7–12], no report addressing the modal analysis of such curves has appeared in the open literature. As is well understood, the loss and confinement of the ${\mathit{\text{ss}}}_{b}^{\mathit{0}}$ mode vanish as the metal vanishes [3], and therefore so does its ability to propagate around curves. Clearly then, it is important to accurately model curved metal stripes in order to determine the onset of significant curvature induced radiation and to optimise structures incorporating curves. This paper deals with the theory and modelling of such curves, and particularly, on the propagation characteristics of the long-range ${\mathit{\text{ss}}}_{b}^{\mathit{0}}$ mode therein.

## 2. Method

The curved waveguide of interest is shown schematically in cross-sectional view in Fig. 1(a) and in top view in Fig. 1(b), along with the cylindrical coordinate system (ρ, ϕ, z) used for its analysis. The metal film of thickness *t* and width *w* has an equivalent complex permittivity *ε*
_{2} and is embedded in an infinite background dielectric of permittivity *ε*
_{1}. The circular curve has a radius of curvature *r*_{0}
that is invariant with *ϕ* so the problem is two-dimensional and the structure is completely characterised via the modes that it supports along the propagation direction which is *ϕ*. The straight waveguide is obtained in the limit *r*_{0}
→ ∞.

Compared to the straight waveguide, bending induces: (i) radiation loss due to the finite speed of light in the cladding medium, (ii) modal offset (towards the outside of the curve) and modal distortion leading to a transition loss at the junction between waveguides of different radii of curvature, and (iii) changes in the effective index of the mode. These effects become more pronounced as the radius of curvature decreases, and are generally deleterious, with effects (i) and (ii) leading to higher insertion loss and effect (iii) leading to phase distortion in interferometric structures for example. However, reducing the radius of curvature reduces the size of components, which is generally desirable. Fortunately, all of these effects (i) - (iii) can be accurately quantified via modal analysis of the curve, so appropriate compensation and trade-off can made in the design of a component.

The modes supported by the curved waveguide must satisfy the time-harmonic (*e*
^{+jωt}) Maxwell’s equations for source-free isotropic media [13]:

with *ε*_{r}
being a function of *ρ* and *z*. For convenience of analysis, an electric Hertzian vector potential **∏**_{e}
is introduced and its curl defined on the basis of the null divergence of **H** and the vector identity ∇ ∙ ∇ × **A** = 0 (where **A** is an arbitrary vector) as:

Using this definition and manipulating Maxwell’s equations (1)-(4) in the usual way [13] yields the vector wave equation governing **∏**_{e}
:

and the following relationship between **E** and **∏**_{e}
:

where *β*
_{0} = √*μ*
_{0}
*ε*
_{0} is the phase constant of free-space. The modes supported by the curved waveguide propagate along the *ϕ* axis, so a mode exhibits an *e*
^{-jβϕϕ} = *e*
^{-jneffβ0r0ϕ} dependency for propagation along + *ϕ*, where *n*_{eff}
is the complex effective refractive index of the mode. The loss of the mode a is given in dB by:

for a curve subtending an angle of *θ* in radians, with *r*_{0}
in m and *β*_{0}
in m^{-1}.

Equations (5)-(7) are solved numerically for any mode of interest using the Method of Lines (MoL) formulated in cylindrical coordinates [14,15]. An absorbing boundary condition positioned on the radiating side of the curve (right side in Fig. 1(a)) is required and used [16,17]. The formulation of the MoL for curves is vectorial and well suited to this waveguide problem since it makes use of a 1-D discretisation along the *ρ* axis and analytical forms along *z*. The analytical forms make it easy to handle either a very large or an infinitesimal metal thickness *t*, while the width of the metal *w* and the radius of curvature *r*_{0}
set the discretisation. Using the MoL, the complex effective index *n*_{eff}
of a mode, along with the spatial distribution of its six field components, can be determined.

## 3. Results

The formulation and software developed [15] were validated by: (i) reproducing the 1-D field distributions and radiation losses computed for dielectric curves using the MoL [14] (the radiation losses reported therein were also validated experimentally [18]); (ii) comparing favourably with the radiation losses computed for dielectric curves using the finite element [20] and finite difference [21] methods (the radiation losses computed therein were also validated experimentally [19]); (iii) reproducing the transition losses computed using the finite-difference method [21] for a straight dielectric waveguide end-coupled to a curved one as a function of *r*_{0}
; (iv) reproducing the phase and attenuation constant versus thickness *t* (and the field distribution at *t* = 100 nm) of the ${\mathit{\text{ss}}}_{b}^{\mathit{0}}$
, ${\mathit{\text{as}}}_{b}^{\mathit{0}}$
, ${\mathit{\text{aa}}}_{b}^{\mathit{0}}$
, ${\mathit{\text{sa}}}_{b}^{\mathit{0}}$
modes for straight 1 μm wide metal stripe waveguides [3] in the limit *r*_{0}
→ ∞; and (v) making comparisons with experiment for S-bends fabricated using thin narrow Au stripes in SiO_{2} and propagating the long-range ${\mathit{\text{ss}}}_{b}^{\mathit{0}}$
mode at λ_{0} = 1550 nm [11].

In order to remain consistent with the results reported in Ref. [3] for the straight metal stripe, a set of curves were analysed at the same wavelength of *λ*_{0}
= 633 nm and assuming the same materials: an Ag film having *ε*_{r,2}
= - 19 -*j*0.53 surrounded by a dielectric of relative permittivity *ε*_{r,1}
= 4. The radius of curvature *r*_{0}
is variable and the cross-sectional dimensions (*w*, *t*) of the film considered are: *w* = 0.5, 1 μm with *t* = 10, 11, 12, 13, 14, 15 nm. The highest loss, highest confinement waveguide among this set is the *w* = 1 μm, *t* = 15 nm structure, for which the attenuation of the long-range ${\mathit{\text{ss}}}_{b}^{\mathit{0}}$
mode is 26 times lower than that of the surface plasmon supported by the corresponding single interface (compare the ${\mathit{\text{ss}}}_{b}^{\mathit{0}}$
mode for *w* = 1 μm, *t* = 15 nm with the ${\mathit{\text{ss}}}_{b}^{\mathit{0}}$
mode for t → ∞ in Fig. 2(b) of Ref. [3]). Thus, the *ss _{b}^{0}*
mode remains long ranging for all (

*w*,

*t*) combinations considered.

Figure 2 shows in Part (a) the total insertion loss (IL) of the ${\mathit{\text{ss}}}_{b}^{\mathit{0}}$
mode for *θ*= 90°, *w* = 1 μm and *t* = 15 nm as a function of *r*_{0}
. The total insertion loss is comprised of the propagation and radiation losses. One immediately notes the existence of an optimum radius of curvature *r*_{0,opt}
where the insertion loss is minimised (*IL*_{min}
), occurring in this case at *r*_{0,opt}
~ 130 μm. An optimum radius exists because the waveguide includes an absorbing medium (the metal). For *r*_{0}
> *r*_{0,opt}
the insertion loss is dominated by the propagation loss, which increases with *r*_{0}
due to the increasing arc length of a fixed angle curve (say 90°), while for *r*_{0}
< *r*_{0,opt}
radiation loss dominates and increases with decreasing *r*_{0}
. Part (b) gives the radiation loss component, which is obtained by repeating the computations with Im{*ε*_{r,2}
} = 0. It is noted that some radiation loss indeed occurs at the optimum radius of *r*_{0,opt}
~ 130 μm. Part (c) plots the Re{*n*_{eff}
} of the ${\mathit{\text{ss}}}_{b}^{\mathit{0}}$
mode of the curve as a function of *r*_{0}
, showing that Re{*n*_{eff}
} converges to the value of the corresponding straight case as *r*_{0}
→ ∞, and diverges as *r*_{0}
→ 0. This divergence stems from the definition of *n*_{eff}
for the curve. Consider that *n*_{eff}*β*_{0}
gives the phase constant in rad/m of the mode propagating along the arc *r*_{0}*ϕ*. Since the mode peak shifts outward as *r*_{0}
→ 0, then *n*_{eff}*β*_{0}
must increase if the phase along *r*_{0}*ϕ* is to be the same as the phase of the mode peak, leading to the observed divergence as *r*_{0}
→ 0. The difference between the effective indices of the curved and straight metal stripes seems to follow an ${r}_{\mathit{0}}^{\mathit{-}\mathit{2}}$
dependence as in dielectric
waveguides [22]. Part (d) summarizes *r*_{0,opt}
and *IL*_{min}
for all combinations (*w*, *t*) considered and shows a clear trend: the higher confinement, higher attenuation structures (e.g.: *w* = 1 μm, *t* = 15 nm) provide a lower *IL*_{min}
and a smaller *r*_{0,opt}
.

Parts (a) and (b) of Fig. 3 show normalised contours of the Re{*E*_{z}
} of the ${\mathit{\text{ss}}}_{b}^{\mathit{0}}$
mode (*w* = 1 μm, *t* = 15 nm) for *r*_{0}
= 1 m and *r*_{0}
= *r*_{0,opt}
respectively, the 1 m radius being used to model the waveguide as straight (*r*_{0}
→ ∞). Part (c) shows normalized distributions of the Re{*E*_{z}
} along a horizontal cut immediately above the metal film (*w* = 1 μm, *t* = 15 nm) for three radii of curvature: *r*_{0}
= 1 m, 100 μm and 50 μm. Two principal effects are noted from these plots as *r*_{0}
decreases: (i) the mode peak shifts from the center towards the outside of the bend, and (ii) the field becomes oscillatory along the outside of the bend. The first effect causes a transition (coupling) loss to occur at the junction between waveguides of differing radii of curvature, say curved and straight, as can be appreciated by comparing the contours in Part (a) with those of Part (b). This transition loss can be computed via an overlap integral on the participating normalised mode fields, bearing in mind that the curved mode is a radiation mode [11]. Part (d) summarizes the transition loss for curves designed at *r*_{0,opt}
and joined to their corresponding straight waveguide (*r*_{0}
→ ∞) for all combinations (*w*, *t*) considered. A clear trend is apparent from Part (d): the lower confinement, lower attenuation structures (e.g.: *w* = 0.5 μm, *t* = 10 nm) provide a lower transition loss. This trend leads away from the trend observed Fig. 2(d), however, transition losses can be substantially eliminated by simply laterally offsetting the curved and straight waveguide sections.

## 4. Conclusions

The main conclusions drawn from this work regarding the long-range ${\mathit{\text{ss}}}_{b}^{\mathit{0}}$ mode propagating along curves in metal stripe waveguides are that: (i) long-range structures are not incompatible with bending, (ii) the higher attenuation, higher confinement structures considered have a lower minimum insertion loss at a smaller optimal radius of curvature, (iii) reasonably small radii of curvature can be used, and (iv) the effects caused by bending are the same as those encountered in conventional dielectric waveguides. The analysis technique discussed in this paper could also be used to study bends in other surface plasmon waveguide structures [23,24].

Note added in proof: A paper was recently published on the modelling of surface plasmon waveguides, including bend structures, using a 3-D electromagnetic field solver [25].

## References and links

**1. **S. E. Miller, “Integrated Optics: An Introduction,” Bell Syst. Tech. J. **48**, 2059–2069 (1969).

**2. **E.A.J. Marcatili, “Bends in Optical Dielectric Guides,” Bell Syst. Tech. J. **48**, 2103–2132 (1969).

**3. **P. Berini, “Plasmon polariton waves guided by thin lossy metal films of finite width: bound modes of symmetric structures,” Phys. Rev. B **61**, 10484–10503 (2000). [CrossRef]

**4. **R. Charbonneau, P. Berini, E. Berolo, and E. Lisicka-Shrzek, “Experimental observation of plasmon-polariton waves supported by a thin metal film of finite width,” Opt. Lett. **25**, 844–846 (2000). [CrossRef]

**5. **R. Nikolajsen, K. Leosson, I. Salakhutdinov, and S.I. Bozhevolnyi, “Polymer-based surface-plasmon-polariton stripe waveguides at telecommunication wavelengths,” Appl. Phys. Lett. **82**, 668–670 (2003). [CrossRef]

**6. **P. Berini, R. Charbonneau, N. Lahoud, and G. Mattiussi, “Characterization of long-range surface-plasmon-polariton waveguides,” J. Appl. Phys. **98**, 043109 (2005). [CrossRef]

**7. **R. Charbonneau, P. Berini, E. Berolo, and E. Lisicka-Shrzek, “Long-range plasmon-polariton wave propagation in thin metal films of finite-width excited using an end-fire technique,” Proc. SPIE **4087**, 534–540 (2000). [CrossRef]

**8. **T. Nikolajsen, K. Leosson, and S.I. Bozhevolnyi, “Surface plasmon polariton based modulators and switches operating at telecom wavelengths,” Appl. Phys. Lett. **85**, 5833–5836 (2004). [CrossRef]

**9. **R. Charbonneau, N. Lahoud, G. Mattiussi, and P. Berini, “Demonstration of integrated optics elements based on long-ranging surface plasmon polaritons,” Opt. Express **13**, 977–984 (2005) http://www.opticsexpress.org/abstract.cfm?id=82563 [CrossRef] [PubMed]

**10. **A. Boltasseva, T. Nikolajsen, K. Leosson, K. Kjaer, M. S. Larsen, and S. I. Bozhevolnyi, “Integrated Optical Components Utilizing Long-Range Surface Plasmon Polaritons,” J. Lightwave Technol. **23**, 413–422 (2005). [CrossRef]

**11. **R. Charbonneau, C. Scales, I. Breukelaar, S. Fafard, N. Lahoud, G. Mattiussi, and P. Berini, “Passive integrated optics elements based on long-range surface plasmon-polaritons,” J Lightwave Technol. **24**, 477–494 (2006). [CrossRef]

**12. **H. S. Won, K. C. Kim, S. H. Song, C.-H. Oh, P. S. Kim, S. Park, and S. I. Kim, “Vertical coupling of long-range surface plasmon polaritons,” Appl. Phys. Lett. **88**, 011110 (2006). [CrossRef]

**13. **R.E. Collin, *Field theory of Guided Waves* (IEEE Press, Piscataway, New Jersey, 1991).

**14. **R. Pregla, “The Method of lines for the analysis of dielectric waveguide bends,” J. Lightwave Technol. **14**, 634–639 (1996). [CrossRef]

**15. **J. Lu, “Modelling optical waveguide bends and application to plasmon-polariton waveguides,” M.A.Sc. Thesis, University of Ottawa, Ottawa, Canada, 2003.

**16. **T. G. Moore, J. G. Blaschak, A. Taflove, and G. A. Kriegsmann, “Theory and application of radiation boundary operators,” IEEE Trans. Antennas Propag. **36**, 1797–1812 (1988). [CrossRef]

**17. **R. Pregla, “MOL-BPM Method of lines Based Beam Propagation Method,” in *Progress in Electromagnetics Research Vol. 11*, J. A. Kong, Ed. (EMW Publishing, Cambridge, Mass., 1995).

**18. **R. J. Deri and R. J. Hawkins, “Polarization, scattering, and coherent effects in semiconductor rib waveguide bends,” Opt. Lett. **13**, 922–924 (1988). [CrossRef] [PubMed]

**19. **M. W. Austin, “GaAs/GaAlAs Curved rib waveguides,” IEEE J Quant. Elect. **18**, 795–800 (1982). [CrossRef]

**20. **T. Yamamoto and M. Koshiba, “Numerical analysis of curvature loss in optical waveguides by the finite-element method,” J. Lightwave Technol. **11**, 1579–1583 (1993). [CrossRef]

**21. **S. Kim and A. Gopinath, “Vector analysis of optical dielectric waveguide bends using finite-difference method,” J. Lightwave Technol. **14**, 2085–2092 (1996). [CrossRef]

**22. **E. F. Kuester and D. C. Chang, “Surface-wave radiation loss form curved dielectric slabs and fibers,” IEEE J. Quant. Elect. **11**, 903–907 (1975). [CrossRef]

**23. **B. Steinberger, A. Hohenau, H. Ditlbacher, A. L. Stepanov, A. Drezet, F. R. Aussenegg, A. Leitner, and J. R. Krenn, “Dielectric stripes on gold as surface plasmon waveguides”, Appl. Phys. Lett. **88**, 094104 (2006). [CrossRef]

**24. **J.-C Weeber, M. U. Gonzälez, A.-L. Baudrion, and A. Dereux, “Surface plasmon routing along right angle bent metal strips”, Appl. Phys. Lett. **87**, 221101 (2005). [CrossRef]

**25. **A. Degiron and D. Smith, “Numerical simulations of long-range plasmons”, Opt. Express **14**1611–1625 (2006) http://www.opticsexpress.org/abstract.cfm?id=88076 [CrossRef] [PubMed]