We present a generalized formulation for the treatment of both bending (whispering gallery) loss and scattering loss due to edge roughness in microdisk resonators. The results are applicable to microrings and related geometries. For thin disks with radii greater than the bend-loss limit, we find that the finesse limited by the scattering losses induced by edge roughness is independent of radii. While a strong lateral refractive index contrast is necessary to prevent bending losses, unless the radii are of the order of a few microns, lateral air-cladding is detrimental and only enhances scattering losses. The generalized formulation provides a framework for selecting the refractive index contrast that optimizes the finesse at a given radius.
©2007 Optical Society of America
Photonic devices incorporating microresonators have become increasingly prevalent in recent years. Despite their enormous potential, their widespread applicability remains stifled due to challenges associated with their modeling and fabrication. Because these devices typically achieve performance enhancements through long effective path lengths, minimizing losses is crucial. Primary loss mechanisms in microdisk and microring resonators are generally due to intrinsic bending or whispering gallery radiation and scattering resulting from edge roughness due to imperfect fabrication. It is well known that bending losses increase with decreasing index contrast. Scattering losses resulting from edge roughness however, behave in the opposite manner, increasing with increasing index contrast. In general, modeling this tradeoff requires detailed, numerically-intensive finite element and/or finite difference time domain simulations or cumbersome approximations [1, 2, 3] to be accurate in the design phase. The designer does not have a tool early in the design phase to select the appropriate materials or geometry optimized to target a desired level of performance. In this paper, we implement and combine the results of two powerful analytic methods into a greatly simplified framework. These include solution of the whispering gallery dispersion relation for an infinite dielectric cylinder and a volume current analysis of edge roughness scattering [4, 5, 6, 7, 8, 9] For the design of simpler waveguide geometries, generalized formulations such as that of Kogelnik  have been and continue to be instrumental both in fostering intuition and in making early, high-level design choices. The following derivation proceeds in the same spirit.
2. Review of the optical whispering gallery dispersion relation
The Helmholtz equation in cylindrical coordinates applicable to the axial fields of whispering gallery modes (WGMs) is
The wavevector associated with circulating WGM field solutions of an infinite dielectric cylinder is directed primarily in the azimuthal direction. Wavevector contributions and field variation in the axial direction are generally small. In disks of finite height, the axial variations can be accounted for by using effective index methods. The time-independent modal field distribution for a WGM can thus be separated into a radially-dependent mode profile and an azimuthal phase dependence, Ψz(r)eimφ. Here, Ψz refers to the axial electric (TM) or magnetic (TE) modal field amplitude and m is the azimuthal quantization number. The Bessel equation for the radial field dependence becomes:
Field solutions consist of Bessel functions of the first kind, Jm(k̃ 1 r) inside the disk boundary, (r < R) and Hankel functions of the first kind, Hm (1)(k̃ 2 r) outside the disk boundary, (r > R). Here, a complex propagation constant and frequency k̃j = njω̃/c are introduced for reasons that will become apparent later. Applying the boundary condition that the axial field components be continuous across the interface gives the complete radial dependence:
The radial and azimuthal field components are easily derived from the axial field components by use of Maxwell’s equations for TM modes:
and TE modes:
where Z 0 is the impedance of free space. Satisfying boundary conditions for both the electric and magnetic fields across the interface results in a dispersion relation that can be written respectively for TM and TE modes as:
Multiple radial solutions exist at discrete propagation constant values in the complex plane giving rise to a resonance map. In the absence of gain or extrinsic loss, solutions necessarily involve an imaginary component of the frequency, implying a continuous decay rate of energy confined within the resonator. The complex frequency and decay rate γ are related according to An intrinsic quality factor equal to the characteristic number of optical cycles before confined energy is lost to the radiation continuum may be defined as . This dimensionless parameter generalizes the imaginary part of the solution space. In order to generalize the real part, a normalized radius is defined as X = n 12πR/λ. By specifying only two parameters: the index ratio, n = n 1/n 2 and azimuthal mode number m, one or more radial mode solutions exist; each is characterized by two parameters: X and Qi. Written in normalized form, the dispersion relation becomes:
In order to solve for the complex roots of these equations, a global optimization scheme can be used to minimize the absolute value of each equation over the complex map . Using this method, generalized plots of intrinsic quality factor against normalized radius may be obtained for the WGMs of a dielectric cylinder. Figure 1 displays the bending limited finesse (ℱi = Qi/m) vs. normalized radius for a variety of azimuthal mode numbers and index ratios. These results will later be combined with the results of the normalized volume current formulation for edge scattering.
3. Review of the volume current method formulation
The volume current method is used to determine edge scattering losses associated with microdisk resonators in the whispering gallery mode regime. Only the fundamental (lowest order) whispering gallery mode characterized by a single radial lobe is considered. We assume a time dependence of exp(-iωt). A dielectric perturbation on the disk edge may be written as a spatially dependent permittivity distribution Δε(r′,z′,φ′). The dielectric perturbation introduces dipole currents which contribute to outwardly radiated (scattered) fields. The expressions for the perturbed current densities manifest themselves in the form of surface-parallel and surface-perpendicular contributions.
The surface-parallel current contribution is readily derived starting from the curl of the Maxwell equation:
By definition, the permittivity gradient is oriented normal to the interface; taking the line integral from just below to just above the interface eliminates the curl:
The surface-perpendicular current contribution is readily derived starting from the continuity relation in the absence of free charges. In combination with the expanded Maxwell equation for the divergence of the displacement vector:
Again, the permittivity gradient is oriented normal to the interface; taking the integral eliminates the divergence:
The current density associated with boundary-continuous parallel electric and perpendicular displacement fields in the presence of the dielectric perturbation is thus given as:
The dielectric perturbation at the disk boundary can be written as a radial step variation at the interface between the two dissimilar refractive indices (n 1 for the core, and n 2 for the cladding),
The interpretation is that the field just inside the interface is perturbed by a lower permittivity when the radial variation is negative and the field just outside the interface is perturbed by a higher permittivity when the radial variation is positive.
4. Spectral density formulation for edge roughness
The mode is primarily affected by perturbations along the direction in which the propagation vector is dominant - here azimuthal. Moreover, etch processes tend to deliver uniform corrugations along height, z. This justifies a decomposition of the radial variation into a Fourier series expansion of corrugation harmonics of azimuthal quantization number, M
The amplitude of each corrugation harmonic may be determined from an experimental measurement of the corrugation. Typically, Gaussian statistics apply as in Borselli  where the roughness can be characterized by two parameters: the rms value of the roughness,σ and its correlation length, Sc. Here, the correlation function for edge roughness, C(s) is related to the measured variation in radius, ΔR(s) with respect to arc length, s at the disk edge over a suitable measured total arc length S meas.
The value of the correlation function at zero is equal to the mean squared roughness,C(0) = σ 2. A Gaussian correlation function can be defined as:
When written in this form, the correlation length is within a factor of of a full width at half maximum (FWHM) definition. The result of further manipulations is also cleaner, hence the motivation. The spectral density is equal to the Fourier transform of the correlation function which is a Gaussian function of spatial frequency variable fs,
If the entire circumference of a disk were to be mapped, a Fourier series representation with harmonics of integer azimuthal quantization numbers would emerge naturally. In practice, it is not often feasible to measure the entire circumference, thus the amplitude coefficients of the Fourier series expansion of the corrugation must be extrapolated from the limited data. To obtain those amplitudes, the spatial frequency variable is expressed as fs = M/2πR and the spectral density is integrated around each integer M.
For most cases of interest, in comparison to the disk circumference, the correlation length is very small Sc/2πR < < 1 allowing the final (trapezoidal) approximation to the integral to hold.
5. Far field scattered power
Returning now to the electrodynamics of scattering, the vector potential in the far-field consists of the volume-integrated current density vector with a retardation phasor term to account for coherent interaction among the current density elements,
Here, the volume integral is represented in cylindrical coordinates appropriate to the geometry of the circulating mode while the far-field scattering direction is represented in spherical coordinates (see Fig. 2). The angle cosine between the current density element and the observation point is expanded in spherical coordinates as
For most geometries in which the disk height is smaller than the radius d < < R the small polar angle approximations about θ= 90° can be made,
Incorporating these approximations results in a volume integral written completely in cylindrical coordinates
Note that the integrated vector potential will be azimuthally independent due to the inherent symmetry of the geometry yet retain a polar dependence. In the far-field, the electric and magnetic fields and Poynting vector are expressed in terms of the vector potential as
The power radiates as transverse electromagnetic waves into all angles of the far-field. The scattered power per unit solid angle thus consists of only polar and azimuthal contributions,
where the radiation vector N = 4πr A/μ 0 is introduced for convenience . The total scattered power results from the solid angle integral:
Finally, the loss per unit length attributed to scattering is directly related to the scattered power according to:
where Pg is the power in the guided mode.
6. TM scattering losses
For TM WGM modes, the modal electric field is perpendicular to the plane of the disk (directed along the z-axis) and thus does not give rise to radiated fields polarized in the azimuthal direction. Thus, the radiation vector only possesses a polar component (the projection of the z-component),
The resulting scattered field retains the orthogonality of the corrugation harmonics; thus, the integral for the radiation vector can be treated separately for each harmonic and later summed incoherently. The modal field is continuous across the interface greatly simplifying the result. Incorporating the unit step perturbation as a limit in the integral results in:
Because the perturbation is small ΔRM < < R and localized to the disk edge surface all radial variables are replaced with the nominal disk radius and the integral is collapsed.
Implementing the identity:
and retaining only the square modulus of the polar radiation vector component yields:
Next the scattered power is calculated assuming that the field and corrugation are z-independent:
The calculation of the z integral is straightforward and results in a polar sinc pattern. Incorporating both this and the Gaussian correlation function, the expression for the scattering loss becomes,
7. TE scattering losses
The edge scattering loss derivation for TE WGM modes is considerably more complicated. First, the modal electric field lies in the plane of the disk with both radial, Er and azimuthal, Eφ components. While the azimuthal component may be negligible for very low index contrast microresonators, in general it can be quite strong and cannot be neglected. Second, each of these components couples to both polar and azimuthal components of the radiated fields. Third, discontinuities in the planar electric field components exist unless the roughness is locally flat . Fortunately, this approximation is valid for typical (shallow) roughness distributions where the corrugation depth is much smaller than the correlation length (σ < < Sc). For a treatment of deep perturbations, see Johnson .
The modal field amplitudes projected in cartesian coordinates are
The radiation vector will consist of both polar θ and azimuthal φ components each arising from these modal field components.
For convenience, field projection variables are defined:
resulting in a compact expression for the radiation vector components
The integral for the radiation vector is treated separately for each harmonic. Furthermore, for convenience, the radial and azimuthal field contributions can be treated separately and summed later. Incorporating the unit step perturbation as a limit in the integral results in:
Because the perturbation is small ΔRM < < R and localized to the disk edge surface, all radial variables can be replaced with the disk radius to collapse the integral.
Implementing the identity
and retaining only the square modulus of the radiation vector components yields:
Assuming that the field and corrugation are z-independent, the scattered power is given as:
The calculation of the z integral is straightforward and again, results in a polar sinc pattern. Incorporating both this and the Gaussian correlation function, the expression for the scattering loss results, here split into azimuthal and radial field contributions:
8. Normalized formulation for edge scattering losses
The expressions for the edge scattering loss scale with the inverse fourth power of the wavelength, typical of scattering processes. They also predict that the loss is strongly dependent on index contrast in proportion to at least the square of the permittivity difference. Finally, the expression is insensitive to disk height d when the height exceeds the wavelength as we will show later. The quality factor can be expressed in normalized units, from which useful limiting approximate forms can be derived. The quality factor is given by the radians per cycle divided by the fractional loss per cycle. Its association with the scattering loss is given by Qs -1 = αsR/m,
These expressions can be written in a compact form by defining normalized units:
where X, ξ, ℓc respectively are normalized quantities representing the radius X = n 12πR/λ, roughness ξ = n 1 σ/λ, and correlation length ∓c = n 1 Sc/λ. The superscript symbol p refers to the polarization state (TM, TEr, or TEφ). The quantity Γpz = Pc/Pg is the ratio of power vertically confined to total power guided or simply the vertical confinement factor. The usual confinement factor for a planar waveguide can be applied here when using the effective index method. This requires solution of the simple planar slab waveguide dispersion relation (kz vs. ω) as in the Kogelnik formulation . The quantity Γpr is the edge confinement factor and is defined as the ratio of the intensity at the disk edge, (∣E(R)∣2/2Z 0) to the characteristic intensity of the mode, (2kPc/d). This requires solution of the simple infinite cylinder whispering gallery mode dispersion relation (X,Qi vs. m,n). The radial mode profile is normalized to a power per unit disk height of Pc/d by integrating the azimuthal component of the Poynting vector along the radial dimension from the disk center out to the radiation boundary (Rr = mλ/2πn 2). Finally, a geometric factor resulting from the sum/integral of the far-field scattering pattern into azimuthal and polar angles is defined, where refers to the associated polar integral.
There are useful limits to consider: thick and thin cylinder. In the limit of a thick cylinder (d >> λ, but where d << R such that the small polar angle approximation, Eqn. 37 still holds), the scattered radiation is directed outward at θ = 90°. The following expressions result for the polar integrals where the derivation is assisted by making the change of variables d cos θ/λ ≡ τ′)
In the limit of a thin cylinder (d << λ), the following expressions result for the polar integrals:
The polar integrals for scattering as a function of corrugation order are plotted for comparison in Fig. 3. These display characteristics similar to those in Rabiei . The scattering as a function of corrugation order can be expressed as scattering as a function of local azimuthal coordinate, Δφ through a modified grating equation,
If the correlation length is small such that the width of the spectral density of the roughness distribution (Xλ/n 1 Sc) is much wider than that of the polar integral (2X/n), then the summation term is greatly simplified. For typical fabrication processes, the relation (Sc << λ/2n 2) is generally the case.
This allows the further reduction of the parameter to simple limiting values:
Here, a normalized thickness has been defined as δ = d/λ. Prior work  neglected cross-term radiation vectors Nθ,ϕ and Nϕ,r resulting in geometric factors of (4/3,2/3,2). The corrected set of geometric factors incorporates the cross-terms and a neglected factor of 1/2:
As a final useful approximation, the edge confinement factor may be approximated in the high index contrast limit as Γr ≈ 1/X. This simple inverse radius dependence is found for both TM and TE when the radial and azimuthal contributions are summed. It is dominated by the azimuthal component in the high index contrast limit validating approximations made in Little . For a breakdown of these contributions, see Fig. 4. Because the edge confinement is inversely related to the normalized disk radius, it directly cancels the increased circumferential path length per round trip. Ultimately this leads to a edge scattering limited finesse (ℱs = Qs/m) that is independent of radius. Incorporating all these approximations results in simple forms for the edge scattering limited finesse. For thick, vertically extended (d > λ) microresonators, the expressions for TM and TE differ slightly:
However, for thin, thumbtack-like (d < λ) microresonators, they are equivalent:
The dependence of the fractional loss per round trip (inverse of finesse) on the index contrast, roughness, and edge confinement is consistent with prior models of scattering losses in straight waveguides [1, 2] and microracetracks .
In practice it is often found that TE operation results in higher losses [5, 8]. This has been attributed to a variety of interpretations such as a higher field strength at the sidewall and arguments involving incomplete derivations of geometric factors. The derivation presented in this paper shows that neither is the case. Rather, it is likely that other mechanisms are responsible for the discrepancy such as the lower vertical confinement factors associated with TM operation. The validity of this approximation for a thin disk is shown by plotting the bending and edge scattering limited finesse as a function of normalized radius in Fig. 5. The field strengths at the disk edge and bending loss contributions were obtained by numerical solution of the dispersion relation for whispering-gallery modes of a cylinder. Figure 6 displays the TM and TE finesse curves for two different values of roughness across a variety of refractive index ratios. Note that at small radii, the advantage of high indices to combat bending loss is apparent as higher indices leaded to higher finesse. For larger radii however, where bending loss is insignificant, the ordering of the curves interchanges where edge scattering limited operation becomes prevalent. Figure 7 displays the tradeoff from a different perspective, plotting the finesse versus refractive index ratio for a few radii (parameterized by m). With increasing refractive index, bending losses decrease while edge scattering losses increase. The tradeoff is thus clear: if index ratio is a design variable, one desires enough to negate the effects of bending loss - but only just enough - as any more leads to increased edge scattering loss in a practical device.
In conclusion, we have presented a generalized formulation incorporating both bending and edge scattering loss limited finesse (and quality factor) in microdisk resonators. The results are also applicable to microrings, microracetracks, and microtoroids. In this formulation, bend loss limited operation is completely specified by a refractive index ratio and an azimuthal quantization number (or alternatively, a normalized radius). The formulation predicts an edge scattering limited finesse that is independent of disk radius and polarization in the thin disk regime. It is completely specified by the same index ratio, along with a product of two normalized roughness parameters, the normalized disk height and the vertical confinement factor. Effectively, only three bundled parameters are thus required to completely specify these primary loss mechanisms for all lowest order whispering gallery modes. We expect that the generalized expressions should be widely applicable both in predicting the impact of fabrication imperfections and/or in selecting the lateral index contrast for optimizing the performance of microresonators.
We would like to thank Ellen Chang for her contributions to this work. This work was supported by the Laboratory for Physical Sciences, College Park, MD.
References and links
1. P. K. Tien, “Light waves in thin films and integrated optics,” Appl. Opt. 10, 23952419, (1971). [CrossRef]
2. R. J. Deri and E. Kapon, “Low-Loss III-V Semiconductor Optical Waveguides.” IEEE J. Quantum Electron. , 27, 626–640, (1991). [CrossRef]
3. V. Van, P. P. Absil, J. V. Hryniewicz, and P. T. Ho, “Propagation loss in single-mode GaAs-AlGaAs microring resonators: measurement and model,” J. Lightwave Technol. 19, 1734–1739, (2001). [CrossRef]
4. M. Kuznetsov and H. A. Haus, “Radiation loss in dielectric waveguide structures by the volume current method,” IEEE. J. Quantum Electron. QE–19, 1505–1514, (1983). [CrossRef]
7. M. Borselli, K. Srinivasan, P. E. Barclay, and O. Painter, “Rayleigh scattering, mode coupling, and optical loss in silicon microdisks,” Appl Phys Lett. 85, 3693–3695, (2004). [CrossRef]
9. P. Rabiei, “Calculation of Losses in Micro-Ring Resonators With Arbitrary Refractive Index or Shape Profile and Its Applications,” J. Lightwave Technol. 23, 1295–1301, (2005). [CrossRef]
11. J. E. Heebner, R. W. Boyd, and Q. Park, “SCISSOR Solitons & other propagation effects in microresonator modified waveguides,” J. Opt. Soc. Am. B. 19, 722–731, (2002). [CrossRef]
12. S. G. Johnson, M. Ibanescu, M. A. Skorobogatiy, O. Weisberg, J. D. Joannopoulos, and Y. Fink, “Pertubation theory for Maxwell’s equations with shifting material boundaries,” Phys Rev E. 65, 066611, (2002). [CrossRef]
13. S. G. Johnson, M. L. Povinelli, M. Soljacic, A. Karalis, S. Jacobs, and J. D. Joannopoulos, “Roughness losses and volume-current methods in photonic-crystal waveguides,” Appl Phys B. 81, 283–293, (2005). [CrossRef]