Effective medium theory applied to photonic

crystals composed of cubic or square cylinders

Philippe Lalanne

Using the effective medium theory, I interpret the band-gap opening in photonic crystals with simple
geometries as an interference effect between alternating layers of high and low optical indices and
introduce the interesting concept of multidimensional quarter-wave stacks. The interpretation provides
a simple insight into band-gap opening processes. For several simple crystal geometries, I analyze the
variations of the gap width and depth with respect to the light polarization, the incident angle, and
contrast inversion. For two- and three-dimensional structures composed of cubic and square cylinders,
I show that the effective medium theory can be used to predict accurately the gap width, the central
wavelength, and the attenuation at the central wavelength. The validity domain of the effective medium
theory predictions is checked with results from rigorous computations. © 1996 Optical Society of
America
Key words: Effective medium theory, homogenization, artificial medium, photonic crystal.

1. Introduction periodic structures in the optical domain as an effi-

The idea of photonic bands in periodic dielectric cient tool for designing different optical components
structures was suggested some time ago.1,2 The ap- for antireflection-coatings,5–7 quarter-wave plates,8,9
plications of such gaps are numerous and, among and binary diffractive optical elements.10 The EMT
them, the possibility of controlling the spontaneous approximation results in the replacement of a peri-
emission appears promising for a large number of odic structure, which is often analyzed by time-
applications, especially in semiconductor technology. consuming computations, by a homogeneous thin
Photonic crystals are artificial two-dimensional ~2-D! layer. This replacement is also referred to as ho-
and three-dimensional ~3-D! periodic structures with mogenization. The equivalent thin layer has an ef-
a frequency band in which electromagnetic waves are fective index ~or an effective relative permittivity!
forbidden, regardless of propagation direction in that can be defined only by the geometry of the
space. Such crystals were recently the subject of periodic structure. Although the EMT of one-
intense theoretical and experimental efforts.3 Al- dimensional ~1-D! periodic structures is well under-
though a number of complex geometries exhibit a stood11–13 theoretically, much study remains to be
band gap, only a few simple geometries that could be done for 2-D periodic structures.
fabricated at submicrometer sizes have been report- As was noted in an original contribution,1 the idea
ed.4 I am primarily concerned with these poten- of photonic crystals is deeply related to layered struc-
tially manufacturable geometries. tures such as quarter-wave stacks or Fabry–Perot
Effective medium theory ~EMT! can be used to resonators, in which the index of refraction alter-
model the propagation of waves in media with a cor- nates from high to low every quarter-wavelength op-
rugation that is much smaller than the wavelength. tical thickness. In fact in some particular cases,
EMT has been extensively applied to subwavelength photonic crystals can be seen as generalization of
quarter-wave stacks to multidimensional quarter-
wave structures. I attempt to exploit this analogy
with EMT and interpret the opening of a band gap in
The author is with the Institut d’Optique Théorique et Appli-
photonic crystals as an interference effect between
quée, Centre National de la Recherche Scientifique, Unité de Re-
cherche Associée 14, B.P. 147, 91403 Orsay Cedex, France. alternating layers of high and low optical indices.
Received 7 November 1995; revised manuscript received 1 April When EMT is applied to the analysis of 2-D and 3-D
1996. photonic crystals, the resulting homogenized struc-
0003-6935y96y275369-12$10.00y0 ture becomes a layered structure. Although the
© 1996 Optical Society of America analysis of 2-D and particularly 3-D photonic crystals

20 September 1996 y Vol. 35, No. 27 y APPLIED OPTICS 5369

is a difficult and computationally time-demanding
task, the analysis of the homogenized structure is
straightforward as the 2 3 2 matrix formalism used
in thin-film design can be applied. Ideally we would
like the EMT to provide us with some simple and
intuitive rules that can be used to determine the
presence and the size of gaps. Because the EMT can
give only an incomplete model of the solutions to
Maxwell’s equations of complex periodic dielectric
structures, this is a difficult undertaking. However,
we note that, at least for some simple structures, the
EMT gives relevant insight into photonic crystals.
Fig. 1. ~a! 2-D square lattice composed of square cylinders with a
In Section 2, by providing a homogenized represen-
relative permittivity ε2 immersed in a homogeneous medium with
tation of 2-D and 3-D periodic structures through the a relative permittivity ε1. ~b! Equivalent stack of homogeneous
EMT, I first state the equivalence between photonic thin films for a wave propagating in the x direction. ε is the
crystals and thin-film stacks. Although this equiv- effective relative permittivity and depends on the fill factor f, on ε1,
alence is provided only for a simple case, namely, a ε2, and a, according to Eqs. ~1! and ~2!.
2-D photonic crystal composed of square cylinders of
one particular material immersed in another mate-
rial, it is general and can be applied to either a 2-D or
a 3-D periodic structure. Then by exploiting the 2. Multilayer Dielectric Stack Approximation of
analogy between thin-film stacks and photonic crys- Photonic Crystals
tals, I design two 2-D photonic crystals that are built In order to illustrate the methodology used through-
so as to act as multidirectional quarter-wave multi- out this paper, let us consider the simple example of
layer dielectric stacks at a given nominal wavelength. the 2-D photonic structure shown in Fig. 1~a!. It is
These designs are provided for both TE and TM po- composed of an array of square cylinders with rela-
larizations and include a derivation of the volume tive permittivity ε2, immersed in a material of rela-
fraction of the high-index medium. In Section 3 I tive permittivity ε1. We use the Cartesian
check the band-gap properties of the two previously coordinates ~ x, y, z! and denote by L the periods in the
designed photonic crystals by rigorously solving Max- x and y directions. The fill factor f is defined in Fig.
well’s equations. By studying the normal incidence 1. We consider a wave that propagates in the struc-
case in Subsection 3.A, I show that the EMT provides ture along the x direction. A similar discussion
an accurate prediction of the wave attenuation at a holds for a propagation direction along the y axis.
nominal wavelength and of the gap width. It is We denote by l the wavelength of this wave in a
shown that these two main features are both related vacuum. If the wavelength is much larger than the
to the same parameter, the optical index modulation. period L, the EMT can be applied. It turns out that
In Subsection 3.B, nonnormal incident angles are the homogenized photonic structure is a film stack in
studied. The shift ~observed with rigorous compu- which the relative permittivity alternates from ε to
tation! toward small wavelengths of the gap central ε1. This thin-film stack is shown in Fig. 1~b!. The
value for increasing incident angles is fully inter- homogenized relative permittivity ε depends on f, ε1,
preted by the EMT. It is concluded that the EMT is ε2, and on the ratio a 5 Lyl, which characterizes the
a relevant and suitable tool for understanding and deviation from the long-wavelength limit ~a 5 0!. ε
predicting the band-gap existence of photonic crys- is called the effective relative permittivity. If no
tals. In Subsection 3.C, the equivalence between magnetic effects are considered such as in the follow-
photonic crystals and thin-film stacks is tested for ing, the square root of ε is the effective index. Al-
nonsquare lattices with nonsquare filling geometry. though we chose a simple 2-D periodic structure for
The wave attenuation at the central wavelength of a purposes of illustration, the homogenization can be
2-D hexagonal lattice with rod cylinders is predicted applied to a wide variety of structures. In general,
by the EMT and compared with rigorous computa- the homogenized medium is not a simple film stack
tions. In Section 4, I extend the previous quarter- composed of alternating high and low refractive in-
wave stack design of 2-D photonic crystals to 3-D dices. For example, when each inclusion is not com-
photonic crystals with cubic filling geometry in a cu- pletely surrounded by host material but is connected,
bic lattice. Section 5 provides a discussion of some the equivalent homogenized medium can be com-
limitations encountered when one models the band- posed of several ~more than two! thin layers with
gap opening with the EMT. It is shown that not only different refractive indices. Also as is shown in Sec-
must the period-to-wavelength ratio be rather small tion 3, when one considers nonsquare filling geome-
in order to obtain an accurate prediction ~especially try, it turns out that the equivalent homogenized
for 3-D photonic crystals! but also that the connection medium is a rugate stack, a name often given to film
of the electromagnetic field between adjacent grating stacks for which the refractive index continuously
regions plays a crucial role for modeling photonic varies in the direction perpendicular to the stack.
crystals accurately by the EMT. Section 6 concludes Finally note that the methodology is also applicable
the paper. to the design and analysis of 3-D periodic structures,

5370 APPLIED OPTICS y Vol. 35, No. 27 y 20 September 1996

as shown in Section 4. The effective permittivity ε where fTE and fTM are the fill factors for the TE and
can be expanded as a power series of the period-to- TM polarizations, respectively. From Eq. 3~b!, aTE
wavelength ratio and for TE polarization ~electric and aTM, the period-to-wavelength ratios for TE and
field parallel to the z direction! is given by11 TM polarizations, are straightforwardly derived as
follows:
p2
ε 5 ε0 1 @ f ~1 2 f !~ε2 2 ε1!#2a2 1 O~a4!, (1)
3 aTE 5 0.410, aTM 5 0.444. (5)
where ε0 is the average relative permittivity ε0 5
ε2f 1 ε1~1 2 f !. Similarly for TM polarization ~elec- Using Eqs. 1~a! and 1~b!, one can determine that the
tric field parallel to the y direction! effective indices nTE and nTM equal 1.566 and 1.287,

F G
respectively. Since the EMT results are also avail-
~ε2 2 ε1! ε0 2
2
1 p2 able for arbitrary profiles, the previous design is not
ε5 1 f ~1 2 f ! a 1 O~a4!, (2)
a0 3 ε2ε1 a03 restricted to the simple geometry of Fig. 1~a!. As is
shown in Subsection 3.C and Section 4, it can be
where a0 is the arithmetic average of the inverse applied to the analysis of more complex 2-D geome-
relative permittivities a0 5 fyε2 1 ~1 2 f !yε1. Sim- tries and to 3-D structures. However, the design of
ilar closed forms were also derived for other periodic multidirectional quarter-wave stacks is restricted to
1-D structures with arbitrary index profiles.14,15 simple cubic structures with filling geometries com-
The only restriction is that the periodic structure is posed of lamellar structures.
invariant along the y direction; these structures are To verify if the EMT and the homogenization meth-
often referred to as lamellar gratings. Also note the odology can be efficient tools for modeling photonic
drastic simplification provided by the EMT, reducing crystals, in the following we compare rigorous com-
the analysis of the complex diffraction problem of Fig. putation results and EMT predictions. All the nu-
1~a! to the simple analysis of a stack of thin layers. merical rigorous results reported in this paper were
To take advantage of the homogenized representa- obtained with rigorous coupled-wave analysis16
tion, we now proceed with the design of a photonic ~RCWA!. Our computer code is based on the eigen-
crystal that acts as a quarter-wave stack. The crys- problem recommended in Ref. 17 for TM polarization
tal structure is considered to be the same as that of 1-D gratings and the reflection–transmission ma-
shown in Fig. 1~a!. For two given permittivities ε1 trix technique18 as an efficient tool for solving multi-
and ε2, the only two free parameters for the design grating diffraction problems. On the other hand, the
are the fill factor f and the period L of the structure. analysis of homogenized multilayer stacks is
Quarter-wave stacks play a crucial role in thin-film straightforward and can be performed using a tech-
design because they are known to provide the largest nique based on 2 3 2 matrices to relate the electro-
reflectivity for a given index modulation. Conse- magnetic fields at each boundary.19
quently, a 2-D or a 3-D periodic structure designed to
act as quarter-wave stacks in several directions are 3. Two-Dimensional Periodic Structures
likely to provide strong wave attenuations and large First we consider the diffraction problem of Fig. 2,
band gaps. This is of particular importance when where a finite spatial extent of the 2-D periodic struc-
one considers crystals with a finite spatial extent. ture of Fig. 1 is inserted between two semi-infinite
The homogenized film stack of Fig. 1~b! is a quarter- media of relative permittivities ε1. For a plane wave
wave stack at nominal wavelength l if incident with an angle of incidence u ~nonconical
ÎεfL 5 ly4, (3a)
mounting!, Fig. 2 defines a diffraction problem that
can be solved using standard rigorous techniques.
Îε1~1 2 f !L 5 ly4. (3b)
In Subsection 3.A, we study the case of normal inci-
dence ~u 5 0!. We show that the homogenized mul-
We note that, for a wave that propagates in the y tilayer structure provides a good model of the wave
direction, Eqs. ~3! remain valid. Consequently, Eqs. attenuation in the periodic structure. Nonnormal
~3! characterize 2-D periodic structures that act as incidences are investigated in Subsection 3.B for TE
quarter-wave stacks in both the x and y directions. polarization. Rigorous computation results show
These structures are referred to as multidirectional that, for TE polarization, the periodic structure de-
quarter-wave stacks in the following. Expressing L signed as a quarter-wave stack in Section 2 @see Eqs.
as a function of l and f with Eq. 3~b! and then sub- ~4! and ~5!# is a 2-D periodic crystal. I also show that
stituting into Eq. 3~a!, we obtain a polynomial in f, the EMT provides an accurate model of the band gap
with the coefficients depending on only ε1 and ε2. It over the whole domain of incident angles. In Sub-
can easily be verified that a root of the polynomial section 3.C I discuss the EMT for use in modeling a
larger than 0 and smaller than 1 exists for any rela- more complex 2-D hexagonal geometry composed of
tive permittivities ~ε1, ε2! and that the root is inde- cylinders of high optical index immersed in a low
pendent of the nominal wavelength. For ε1 5 1 and optical index medium. The homogenized medium,
ε2 5 4, we find which is then a rugate film stack, is shown to provide
a good approximation of the wave attenuation in the
fTE 5 0.390, fTM 5 0.437, (4) principal direction of the lattice.

20 September 1996 y Vol. 35, No. 27 y APPLIED OPTICS 5371

Fig. 2. Diffraction problem used to check the validity of the
quarter-wave photonic crystal design. The periodic structure is
the same as in Fig. 1~a! and has an infinite spatial extent in the y
direction. L denotes the number of grids, each composed of one
grating and one thin film. The fill factor and the period-to-
wavelength ratio are given by Eqs. ~4! and ~5!. ε2 5 4 and ε1 5 1.

A. Normal Incidence
According to Eqs. ~4! and ~5! the diffraction geometry
of Fig. 2 is equivalent to a quarter-wave stack at
normal incidence. The transmittance T ~transmit-
ted energy flux over incident energy flux ratio! of a
stack with L grids is simply given by20

T512 Un1 2 ne
n1 1 ne
U
2

with ne 5 SD
n
n1
2L

n1 , (6)

where n is the effective index equal to nTE or nTM for

TE or TM polarization, respectively. Fig. 3. ~a! Logarithm of transmittance T as a function of the
Figure 3 compares numerical results obtained with number L of grids for the diffraction problem of Fig. 2 ~ε2 5 4 and
RCWA and with Eq. ~6!. The logarithm of the trans- ε1 5 1!. The solid curves correspond to the EMT approximation of
mittance is plotted as a function of the number L of Eq. ~6!. Circles are numerical results obtained with the RCWA.
grids. The solid curves are transmittances given by In this example, u 5 0, l 5 1, and the fill factors f and the period-
Eq. ~6!, and the circles represent RCWA results. We to-wavelength ratios a are given in Eqs. ~4! and ~5!, respectively.
~b! The dashed curves and the pluses were obtained with Eq. ~6!
note that the EMT prediction is accurate especially
and the RCWA computation for the contrast-inversion problem ~ε2
for TE polarization. The exponential attenuation of 5 1 and ε1 5 4!. In that case, fTE 5 0.44, aTE 5 0.284, and nTE 5
the transmittance clearly indicates that the wave 1.57 for TE polarization; fTM 5 0.37, aTM 5 0.33, and nTM 5 1.29
cannot propagate in the crystal at normal incidence. for TM polarization.
Moreover, although the designs are almost identical
for TE and TM polarizations ~ fTE ' fTM and aTE '
aTM!, we note that the wave attenuation is much
larger for TE than for TM polarization. This differ- logarithm of the transmittance for the contrast-
ence is due to the effective index modulation being inversion case. The pluses were obtained with
larger for TE than for TM polarization. RCWA, and the dashed curves were derived using Eq.
It is also interesting to study the effect of a contrast ~6!. We note again excellent agreement between the
inversion, in other words, to consider a periodic struc- EMT prediction and RCWA results for TE polariza-
ture composed of square cylinders with low permit- tion. For TM polarization, the EMT prediction is
tivity ε1 inserted in a homogeneous medium of high more accurate than before. This can be explained by
relative permittivity ε2. Using Eq. ~1! and solving the fact that the period-to-wavelength ratio for the
for a quarter-wave stack condition, we found that fTE contrast-inversion case ~aTM 5 0.33! is much smaller
5 0.44, aTE 5 0.284, and nTE 5 1.57 for TE polariza- than the period-to-wavelength ratio of Eqs. ~5! ~aTM 5
tion. Similarly, for TM polarization we found that 0.444!, and, of course, EMT approximations are much
fTM 5 0.37, aTM 5 0.33, and nTM 5 1.21. So in this more valid at small a values. We also note that in
inversion case the homogenized medium contains L the contrast-inversion case, the attenuation is much
grids, each composed of two layers of optical indices larger for TM than for TE polarization. This can
n2 5 2 and n 5 nTE or nTM. Figure 3~b! shows the again be explained by the EMT, considering that the

5372 APPLIED OPTICS y Vol. 35, No. 27 y 20 September 1996

Fig. 4. Wavelength dependence of the transmittance of the six- Fig. 5. Enlarged view of Fig. 4 centered around the nominal
grid stack shown in Fig. 2 for TE polarization and normal inci- wavelength. The solid and dashed curves are those of Fig. 4, and
dence. The solid and dashed curves correspond to the RCWA and the dash– dot curve corresponds to a more accurate EMT model of
EMT results. The EMT model assumes an effective index of n 5 the six-grid stack with an effective index n varying with the wave-
1.566 independent of the wavelength. length according to Eq. ~1!.

index modulation for TM polarization ~n2 5 2, nTM 5 the nominal wavelength. The most striking feature
1.21! is much larger than for TE polarization ~n2 5 2, of Fig. 4 is the existence of a large gap, which, ac-
nTE 5 1.57!. cording to the quarter-wave stack design, is almost
For large L values ~ne .. 1 or ne ' 0! the logarithm centered at the nominal wavelength of l 5 1. We
of the transmittance T given by Eq. ~6! can be rewrit- also observed that the EMT provides a good approx-
ten as imation of the transmittance over a large spectral
domain from 0.6 to 1.35. For wavelengths smaller
log~T! 5 c 2 LA, (7) than 0.6, RCWA results present strong variations
where c is a constant and A is a dimensionless coef- that are not predicted by the EMT for which two
ficient equal to log @1 1 ~n 2 n1!yn1#2. Equation ~7! smaller gaps centered around 0.44 and 0.54 appear
provides a linear relationship between the logarithm as discussed in Section 5. Figure 5 shows an en-
of the transmittance and the number of grids. In the larged view of Fig. 4, centered around the nominal
following, A is referred to as the wave attenuation per wavelength of l 5 1. As in Fig. 4, the solid and
grid. In practice, crystals of infinite spatial extent dashed curves represent the numerical values of the
are not realistic, and the attenuation per grid deter- transmittance obtained with RCWA and EMT, re-
mines how much a wave attenuates when it propa- spectively. The dash– dot curve corresponds to a
gates in a specific direction of the photonic crystal. more accurate approximation obtained with the EMT
Large A values are obtained for large relative index by taking into account the wavelength dependence of
modulation ~n 2 n1!yn1. the effective index according to the dispersion rela-
In Figs. 3~a! and 3~b! the largest difference between tion of Eq. ~1!. Note that excellent agreement be-
RCWA results and the EMT prediction occurs for TM tween RCWA and EMT results can now be obtained,
polarization and for the periodic structure composed and that the EMT provides an accurate prediction of
of high optical index square cylinders immersed in a the band-gap width and depth. Figure 6 corre-
low optical index medium. In this case, the wave sponds to the TM polarization case of a six-grid stack.
attenuations per grid obtained with the EMT and Here again, a band gap centered around the nominal
RCWA are 0.48 and 0.40, respectively ~16% relative wavelength is obtained. Note that the band-gap
difference!. We conclude that the EMT provides a width is smaller than for the TE polarization case.
reasonably good estimate of the wave attenuation per Although an important disparity exists between
grid for the four cases. RCWA transmittance computation results and the
In Fig. 4 we plotted the transmittance for different EMT, we conclude that the EMT provides a fairly
wavelengths that vary from 0.42 to 1.35. The nu- accurate prediction of the band-gap edges and of its
merical results were obtained for normal incidence, general shape.
for TE polarization, and for the diffraction problem of In thin-film design, the width of the reflectance
Fig. 2 with six grids ~L 5 6!. The dashed and solid band of a high reflector is defined by the wavelengths
curves represent numerical results obtained with the between which the reflectance increases as layers are
EMT and RCWA, respectively. For the EMT, the added. In photonic crystal terminology, this reflec-
effective index nTE was assumed to be independent of tance band corresponds to the gap that holds for a
the wavelength and equal to 1.566, the index value at crystal of infinite spatial extent. So the reflectance

20 September 1996 y Vol. 35, No. 27 y APPLIED OPTICS 5373

Fig. 6. Wavelength dependence of the transmittance of the six- Fig. 7. Transmittance of the six-grid stack shown in Fig. 2 for TE
grid stack shown in Fig. 2 for TM polarization and normal inci- polarization and for two angles of incidence, u 5 py12 and u 5 py6.
dence. The solid and dashed curves correspond to RCWA and The solid, dashed and dash– dot curves are the same as in the
EMT results. The EMT model assumes an effective index of n 5 captions for Figs. 4 and 5.
1.287 independent of the wavelength.

result was previously derived by the authors of Ref.

band of thin-film stacks can be used simply as a mea- 22 from an electromagnetic field analysis.
sure of the photonic gap that exists for one direction
of wave propagation. If a quarter-wave reflector is B. Nonnormal Incidence
designed at a nominal wavelength l, the wavelengths In Subsection 3.A we considered only the existence of
of the reflectance band edges are21 a gap for wave propagation along the x direction. To
determine if a band gap exists, it is necessary to study
ledge 5 ly~m 6 D!, (8) the transmittance for different incident angles u.
where m is an odd integer equal to the order number Because of symmetry, however, it is possible to re-
of the stack, and strict the analysis to u varying between 0 and py4.

S D
For that, we note that the transmittances for positive
2 nH 2 nL and negative angles of incidence are identical and
D5 arcsin , (9) that the analysis to u varying between py4 and py2
p nH 1 nL
corresponds to incident angles u9 varying from 0 to
where nH and nL are the high and low optical indices 2py4, with u9 defined relative to the y direction.
of the stack, respectively. Applying Eqs. ~8! and ~9! In Figure 7 the transmittance of the six-grid stack
to the case of the photonic crystal depicted in Fig. 2 of Fig. 2 is shown for TE polarization and for two
and assuming that the relative index modulation angles of incidence, u 5 py12 and u 5 py6. As be-
~n 2 n1!yn1 is weak, for the gap width Dl related to fore, the solid and dashed curves were obtained with
a wave that propagates in the x direction we obtain RCWA and the EMT model with an effective index n

S D
assumed constant, n 5 1.566. The dash– dot curve
4 n 2 n1 shows the results obtained with the wavelength ef-
Dl 5 l arcsin , (10)
p n 1 n1 fective index dependence given by Eq. ~1!. For the
two angles of incidence a gap exists that is centered
where n is the effective index equal to nTE and nTM for on shorter wavelengths as u increases. We note that
TE and TM polarizations. To derive Eq. ~10!, m was the EMT provides an excellent approximation over
considered to be equal to 1 and D was supposed to be the whole spectral domain of interest. The gap shift
much smaller than 1. According to Figs. 5 and 6, Eq. toward shorter wavelengths can be simply under-
~10! can be used to predict the bandwidth with rela- stood with thin-film stacks. This simply results
tive errors of 1% for TE and 10% for TM polarizations. from the fact that the optical path difference
Equations ~10! and ~7! show the importance of the
relative index modulation for determining the gap f 5 4pnidi~cos uiyl! (11)
width and the attenuation per grid. Also, as can be
seen from Figs. 3, 4, and 6, structures with high- in the ith layer of thickness di and optical index ni is
index inclusions in a low-index medium are prefera- reduced by a cosine factor that takes into account the
ble for TE polarization, whereas contrast-inversion tilted incidence. It follows that the wavelength for
cases are more suitable for TM polarization. This which the minimum reflectivity is observed is smaller
results directly from the fact that large relative index than the nominal wavelength by approximately the
modulation can be obtained with low-index inclusions same cosine factor. Figure 8 shows the transmit-
for TM and with high-index inclusions for TE. This tance for u 5 py4. A gap exists for wavelengths

5374 APPLIED OPTICS y Vol. 35, No. 27 y 20 September 1996

 Fig. 9. 2-D circular rod profiles on a 2-D triangular lattice with an
hexagonal symmetry. Six grids are represented.

mal symmetry in all directions of space. The second

choice concerns the filling geometry of the lattice that
can obviously provide an infinite number of possibil-
ities. Spheres, cubes, rectangles, Gaussian spheres,
and so on have excited the imagination of theorists.4
Fig. 8. Transmittance of the six-grid stack shown in Fig. 2 for TE So the simple cubic lattice with a cube-filling geom-
polarization and for u 5 py4. The solid and dashed curves are the etry as studied in Subsection 3.B cannot be seen as
same as in the caption for Fig. 4.
representative of the geometry complexity encoun-
tered in photonic crystal design. In this subsection I
show how the EMT can be applied to more complex
larger than 0.78 and smaller than 1.05. However geometries. However, despite all the advances in
the EMT approximation does not provide an accurate material growth techniques, we note that the simple
model of the gap over the whole spectral domain of cubic 2-D lattice discussed above is already a chal-
interest. For wavelengths smaller than 0.78, RCWA lenging objective for visible light applications, even
results exhibit strong modulations of the transmit- with the most modern facilities.
tance that almost reach 100% at some particular Let us now consider the 2-D lattice depicted in Fig.
wavelengths. A similar surprising transmittance 9. This kind of structure was investigated by sev-
behavior was previously encountered at small wave- eral authors,23,24 and the computations indicate a
lengths and for normal incidence ~see Fig. 4!. This band gap in the case of a circular rod profile for either
difference between the EMT prediction and RCWA TE or TM polarization. More recently, the electro-
results is discussed in Section 5. From Figs. 5, 7, magnetic properties of this structure were studied in
and 8 and from the symmetry of the structure, it can great detail. In Ref. 25 the author considers that the
be concluded that the multidirectional quarter-wave rods are composed of a medium of high index n2 and
structure of Fig. 2 is a 2-D photonic crystal for TE are immersed in a homogeneous medium of low index
polarization. The photonic band gap found with rig- n1. Let us consider a wave that propagates in the x
orous computation spreads over the spectral interval direction. When one models the lattice of Fig. 9 with
~0.85, 1.05!. This shows up the relevancy of the the EMT, two problems appear. First, as is dis-
quarter-wave design of Section 2 and the benefit of cussed in Section 5, the hexagonal structure cannot
using the EMT as an approximate theory for studying be taken into account and the EMT model cannot
and designing photonic crystals. We also note that reflect the high degree of symmetry of the hexagonal
the EMT accurately predicts the gap width and depth lattice. This difficulty is intrinsic and precludes a
over the whole band-gap domain. solution. Second, the filling geometry ~cylinders!
Finally we note that special care has to be taken cannot be modeled as simply as cubic geometries ~la-
when the EMT is applied to model band gaps of 2-D mellar structures!. A solution consists in approxi-
photonic crystals for TM polarization. In that case, mating each cylinder grating by a stack of thin films
the zeroth-order effective index does not depend only or more ideally by a gradient-index film. As shown
on the period-to-wavelength ratio but also on the an- in Fig. 10, the cylindrical rod structure can be ap-
gle of incidence u ~see Refs. 11 and 14!. To our proximated with many lamellar grating layers hav-
knowledge, a second-order EMT has not been yet ing different filling factors. This technique has been
derived, and this is the reason I have not provided a used for designing antireflection coatings with sub-
comparison between RCWA and the EMT for TM wavelength gratings of continuous profile relief. Si-
polarization and nonnormal incidence. nusoidal,26 quintic,5 and triangular27 profiles were
investigated by several authors. For the specific
C. Noncubic Lattices and Noncubic Filling Geometries case of a multilevel relief profile, the homogenized
In the search for a crystal that exhibits a photonic medium is a film stack. Each vertical film corre-
band gap, many different possibilities were investi- sponds to a distinct level of the surface relief profile
gated during the past few years. Two levels of along the vertical direction ~see Fig. 10!. For a con-
choice are involved. The first concerns the Bravais tinuous profile that probably provides a better ap-
structure. Face-centered cubic lattices were more proximation of the rod structure, the homogenized
extensively studied, probably because of their maxi- medium is a graded-index film. With the 2 3 2 ma-

20 September 1996 y Vol. 35, No. 27 y APPLIED OPTICS 5375

Fig. 10. Homogenized medium associated with the 2-D hexagonal structure of Fig. 9. The figure shows a rough approximation composed
by a stack of four thin films.

trix method we used, 1000 film layers were sufficient 4. Three-Dimensional Periodic Structures
to provide a good approximation of the cylindrical rod I now extend the quarter-wave stack design of 2-D
shape. To compare EMT predictions with the rigor- periodic structures ~see Section 2! to the case of 3-D
ous computation results obtained in Ref. 25, we chose periodic structures. Let us again consider a simple
the same parameters as in Ref. 25: n1 5 1, n2 5 2, cubic lattice with a cubic filling geometry. One pos-
a period of unity length, and a rod diameter of 0.4. sible representation of the lattice elementary cell is
Figure 11 shows a comparison between the EMT and shown in Fig. 12 where ε1 and ε2 are the relative
the rigorous results of Ref. 25. The decimal loga- permittivities of the cubic medium and of the sur-
rithm of the transmittance ~called the attenuation in rounding material, respectively. In the following we
Ref. 25! that differs by a log~10! factor with the pa- consider two structures: the case with ε1 5 1 and ε2
rameter A of Eq. ~7! is plotted as a function of the 5 13 and the contrast-inversion case with ε1 5 13 and
number L of grids. Results were obtained for TE ε2 5 1. The wave is assumed to propagate perpen-
polarization and normal incidence. The solid curve dicularly to the cubic faces. According to Fig. 1, the
represents results derived with the EMT model and homogenized structure is a film stack composed of
the circles represent the numerical results obtained two materials with relative permittivities ε1 and ε,
in Ref. 25 ~see the numerical values of the central with ε as the effective permittivity that depends on ε1,
column of Table 1!. For small L values, a discrep- ε2, f, and a. For 2-D periodic structures, no closed
ancy exists between rigorous and EMT results. form of the effective permittivity has been derived, so
However for L larger than 10, we note that the EMT we write
prediction is much more accurate. This indicates
that the homogenized structure provides a good ε 5 h0 1 h2~Lyl!2, (12)
model of the interference effect between the cylinder
gratings of the hexagonal lattice. where h0 and h2 are the zeroth-order and second-
order effective permittivity coefficients. For a peri-
odic structure based on the elementary cell of Fig. 12
to act as a quarter-wave stack, Eqs. ~3! have to be
satisfied. From Eqs. ~3a!, ~3b!, and ~12!, it can easily
be shown that the fill factor f must satisfy

16ε12~1 2 f !4 5 f 2@16ε2~1 2 f !2h0 1 h2#. (13)

Note that h0 and h2 depend on f. We call u~ f ! and
v~ f ! the left and right terms of Eq. ~13!, respectively.

Fig. 11. Decimal logarithm of the transmittance as a function of

the number L of grids in the periodic structure of Fig. 9. Circles
represent results from rigorous computation ~see Table 1 in Ref. Fig. 12. Elementary cell of a simple cubic lattice with cubic filling
23!. The solid curve represents results obtained by EMT model- geometries. The corresponding 3-D crystal is composed of cubes
ing, when 1000 films are used to approximate the cylindrical rod of size fL and relative permittivity ε2, immersed in a medium of
shape. relative permittivity ε1.

5376 APPLIED OPTICS y Vol. 35, No. 27 y 20 September 1996

Fig. 13. u~ f ! and v~ f ! as functions of the fill factor f, ε1 5 13 and Fig. 15. Same as Fig. 14 except that ε1 5 13 and ε2 5 1.
ε2 5 1.

ture at normal incidence. The incident and sub-

In Ref. 15 the authors derive formulas to compute h0 strate media have an optical index equal to 1, and the
and h2 for arbitrary symmetric structures. The nu- five-grid stack is composed of ten alternating quarter-
merical evaluation requires the inversion of an infi- wave layers with a homogeneous thin film of relative
nite matrix, which is truncated in practice. In Fig. permittivity ε2 just above the incident medium. The
13, u~ f ! and v~ f ! are shown for ε2 5 1 and ε1 5 13 RCWA results ~solid curve! confirm the existence of a
~low-index material inserted in a high-index medi- large gap centered on the nominal wavelength ac-
um!. For the numerical computation of h0 and h2, cording to the quarter-wave stack design, and the
2061 orders were retained for the matrix inversion EMT prediction ~dashed curve! agrees well with rig-
problem.15 The two curves intersect for f 5 0.56. orous computation on a large spectral domain from
As in the 2-D case of Section 2, a solution for f inde- 0.6 to 1.4.
pendent of L and l exists for any values ε1 and ε2. For ε1 5 1 and ε2 5 13, the same procedure can be
This can be understood by noting that u~ f ! monoton- used. The intersection of curves u~ f ! and v~ f ! oc-
ically decreases from a positive value to zero in the curs for f 5 0.43. The period L then equals 0.44.
interval @0; 1#, whereas v~ f ! is positive and equals Figure 15 shows a comparison between the RCWA
zero for f 5 0 and 1. For a nominal wavelength l results and the EMT prediction. The agreement is
equal to 1, the period L and the effective index n are, only qualitative. RCWA computations confirm the
respectively, 0.158 and 2.85, according to Eqs. ~3a! existence of a gap centered at a wavelength slightly
and ~3b!. Figure 14 shows a comparison between larger than the nominal wavelength. Moreover, the
the RCWA and the EMT results for a five-grid struc- EMT overestimates the maximal attenuation at the
gap center, and the EMT does not predict the exis-
tence of two other gaps centered around 0.63 and
0.72. As is discussed in Section 5, the poor agree-
ment of EMT and RCWA results is due to the large
period-to-wavelength ratio 0.44 for this case.
5. Limitations
There are several limitations when one models
band gaps of photonic crystals with the EMT. The
first is due to the EMT itself. For large period-to-
wavelength ratios, the EMT prediction of the effective
index is no longer valid. For example, in Fig. 3
the TM transmittance is better approximated for the
contrast-inversion case ~aTM 5 0.33! than for the
high-index inclusion case ~aTM 5 0.444!. Similarly
for the 3-D photonic structures studied in Section 4,
the EMT prediction is in excellent agreement with
the RCWA results in Fig. 14 ~a 5 0.158!, whereas a
large discrepancy is observed in Fig. 15 ~a 5 0.437!.
Fig. 14. Comparison between RCWA ~solid curve! and EMT In general as was noted in Ref. 15, the EMT provides
~dashed curve! results of the transmittance of a five-grid photonic excellent approximations for TE polarization of 1-D
crystal with the elementary cell of Fig. 12. 81 orders are retained periodic structures over the whole domain of opera-
for RCWA computations; ε1 5 13 and ε2 5 1. tion of the structure as a zeroth-order filter ~no dif-

20 September 1996 y Vol. 35, No. 27 y APPLIED OPTICS 5377

 observation of broad gaps indicate that the gap open-
ing at small wavelengths is most probably due to
some more complex effects than Wood’s anomalies.
We suggest here that frustrated total internal reflec-
tion by tunneling of the first few evanescent orders
inside the homogeneous region plays a crucial role in
the interpretation of the gap opening at small wave-
lengths. To be more specific, we can estimate the
coupling strength of the first evanescent order. For
normal incidence, we denote by kz 5 2p@~1yL!2 2
~n1yl!2#1y2 the module of the z wave-vector compo-
nent of the first evanescent order and by t the nega-
tive exponential of the product of kz by the thickness
~1 2 f !L of the homogeneous regions

t 5 exp 2 kz~1 2 f !L, (14)

Fig. 16. Wavelength dependence of the transmittance of the pho- where t is the residual amplitude of an incident wave
tonic crystal shown in Fig. 2 for TE polarization and normal inci- with unit amplitude after tunneling through a thick-
dence. All the results were obtained with RCWA: solid curve,
ness ~1 2 f !L. So t characterizes the coupling
L 5 1; dotted curve, L 5 2; dash– dot curve, L 5 3; dashed curve,
L 5 6.
strength between adjacent gratings, which depends
on the period-to-wavelength ratio. Numerically for
L 5 0.41, f 5 0.39, and n1 5 1, we found that t is
always larger than 2%, equals 5% for l 5 0.66, and
fracted waves!. For TM polarization of 1-D periodic reaches 10% for l 5 0.51. We conclude that the two
structures, slight deviations are observed near the gaps observed for small wavelengths in Fig. 16 belong
cutoff wavelength, and, for 2-D periodic structures, to a spectral domain in which the coupling strength is
the EMT approximation is valid only at small larger than 10%. Moreover, according to Fig. 4, dis-
period-to wavelength ratios, typically less than 0.2. crepancies between the EMT predictions and the
The poor performance of the EMT at small wave- RCWA results appear for wavelengths smaller than
lengths has important consequences. For example, 0.65, that is, for a coupling strength larger than 5%.
in Figs. 4 and 15 two gaps are not predicted at wave- The tunneling effect can also be invoked to explain
lengths smaller than the nominal wavelength. For the difference between the RCWA results and the
the periodic structure shown in Fig. 9, a similar effect EMT prediction observed in Fig. 8 for u 5 py4. In
was reported by the authors of Ref. 24. The opening that case, the observation of unpredictable resonance
of these unpredicted gaps was interpreted as a con- effects at larger wavelengths ~0.77 instead of 0.6! is
sequence of Wood’s anomalies. In Fig. 16 we show due simply to the fact that, as the angle of incidence
the transmittance of the 2-D periodic structure of Fig. u increases, the spectral domain of operation of the
2 as a function of the wavelength for several numbers gratings as zeroth-order filters shifts toward larger
of grids, namely, L 5 1, 2, 3, and 6. For L 5 1 ~solid wavelengths. In other words, the module z wave-
curve!, the transmittance is almost constant over the vector component of the first negative order decreases
whole spectral domain of interest, except that a as the angle u increases. So tunneling effects can be
strong resonance effect with a nearly 100% variation observed at larger wavelengths and appear in the
of the transmittance is observed for l 5 0.42. This band gap predicted by the EMT model. When one
resonance effect ~Wood’s anomaly! is well understood uses homogenization techniques to design or study
as resulting from the excitation of leaky waveguide photonic crystals, the computation of parameter t can
modes of a grating surrounded by two homogeneous be used for estimating quickly the accuracy that can
media of lower optical indices.28 The existence of be expected with the EMT. According to the previ-
the two gaps at small wavelengths for L . 1 is not ous examples, weak mismatches can be expected for
due simply to the excitation of leaky waveguide t larger than 5%, and t values larger than 10% are
modes or to an inaccurate EMT prediction of the ef- likely to indicate that the EMT predictions are not
fective index. The first reason is that the resonance valid.
width of leaky waveguide modes is deeply related to Finally there is another limitation of the EMT
the grating-index modulation,29 and that the two for modeling photonic crystals that is intrinsic to the
gaps observed at small wavelengths in Fig. 16 are methodology used. To illustrate our purpose
much broader than a narrow leaky waveguide reso- straightforwardly, let us consider the 2-D chessboard
nance. Moreover unlike the zero transmittance gen- structure of Fig. 17 that represents a 2-D version of
erally observed at the peak resonance of leaky the 3-D structure originally discussed by Yablono-
waveguide modes, the attenuation at the central vitch.1 It is composed of an array of parallelepipeds
wavelengths of the two gaps monotonically increases of high index n2 immersed in a medium of low index
with the number L of grids. This attenuation de- n1. Each layer thickness is supposed to be ly2~n1 1
pendence on the number of layers and the surprising n2!. It is straightforward to see that for a wave that

5378 APPLIED OPTICS y Vol. 35, No. 27 y 20 September 1996

 mogeneous regions surrounding isolated inclusions
was suggested to be responsible for the EMT–RCWA
discrepancies. In Section 5 I proposed a simple cri-
terion that defines the spectral domain for which ho-
mogenization techniques can or cannot be used for
modeling photonic crystals. I believe that this study
helps to clarify some aspects of the complex electro-
magnetic process that results in band-gap openings.
Fig. 17. 2-D version of the 3-D chessboard structure originally
proposed by Yablonovitch.1 Gray areas represent high-index me- When this study was completed, the author was a
dia. visiting scientist at the Institute of Optics, University
of Rochester. He is pleased to acknowledge the Di-
rection Générale de l’Armement ~DGA! for partial
financial support under Direction de la Recherche et
des Etudes Techniques-DGA contract 94-1123. He
is grateful to P. Chavel and G. M. Morris for fruitful
discussions.
References and Notes
1. E. Yablonovitch, “Inhibited spontaneous emission in solid-
state physics and electronics,” Phys. Rev. Lett. 58, 2059 –2062
~1987!.
Fig. 18. Two different structures that cannot be distinguished 2. S. John, “Strong localization of photons in certain disordered
from the chessboard striker of Fig. 17 by homogenization. dielectric superlattices,” Phys. Rev. Lett. 58, 2486 –2489
~1987!.
3. See, for example, recent special issues on photonic band struc-
propagates in the x or y direction, the homogenized tures: J. Mod. Opt. 41~2! ~1994! and J. Opt. Soc. Am. B 10~2!
~1993!.
structure associated with the chessboard structure is
4. J. W. Haus, “A brief review of theoretical results for photonic
simply a homogeneous medium. Similarly, for a band structures,” J. Mod. Opt. 41, 195–207 ~1994!.
wave that propagates along the vertical direction, the 5. W. H. Southwell, “Pyramid-array surface-relief structures pro-
two crystals of Fig. 18 and the chessboard crystal of ducing antireflection index matching on optical surfaces,” J.
Fig. 17 are not distinguished by the EMT. We con- Opt. Soc. Am. A 8, 549 –553 ~1991!.
clude that the EMT cannot model lateral shifts ~along 6. R. Brauer and A. Bryngdahl, “Design of antireflection gratings
the y direction in Fig. 17! that exist between adjacent with approximate and rigorous methods,” Appl. Opt. 33, 7875–
layers.30 The lack of accurately taking into account 7882 ~1994!.
lateral shifts through homogenization is an impor- 7. E. B. Grann, M. G. Moharam, and D. A. Pommet, “Artificial
tant limiting factor that has to be considered when uniaxial and biaxial dielectrics with use of two-dimensional
subwavelength binary gratings,” J. Opt. Soc. Am. A 11, 2695–
complex structures are homogenized.
2703 ~1994!.
8. R. C. Enger and S. K. Case, “Optical elements with ultrahigh
6. Conclusion
spatial-frequency surface corrugations,” Appl. Opt. 22, 3220 –
I interpreted band-gap openings in photonic crystals 3228 ~1983!.
as resulting from interference effects between alter- 9. D. C. Flanders, “Submicrometer periodicity gratings as artifi-
nating layers of high and low optical indices. This cial anisotropic dielectrics,” Appl. Phys. Lett. 42, 492– 494
was accomplished by exploiting the equivalence be- ~1983!.
tween subwavelength periodic structures and homo- 10. F. T. Chen and H. G. Craighhead, “Diffractive phase elements
geneous thin films with the EMT. For 2-D and 3-D on two-dimensional artificial dielectrics,” Opt. Lett. 20, 121–
123 ~1995!.
periodic structures, we designed and tested periodic
11. S. M. Rytov, “Electromagnetic properties of a finely stratified
structures that act as quarter-wave stacks. By com- medium,” Sov. Phys. JETP 2, 466 – 475 ~1956!.
paring EMT predictions with RCWA results, we 12. R. C. McPhedran, L. C. Botten, M. S. Craig, M. Nevière, and D.
showed in a number of cases that the band-gap width Maystre, “Lossy lamellar gratings in the quasi-static limit,”
and the attenuation per grid in the spectral band-gap Opt. Acta 29, 289 –312 ~1982!.
domain can be accurately predicted by homogenized 13. G. Bouchitté and R. Petit, “Homogenization techniques as ap-
structures. This provides us with an intuitive in- plied in the electromagnetic theory of gratings,” Electromag-
sight into band-gap opening processes of simple pho- netics 5, 17–36 ~1985!.
tonic crystals, such as those composed of cubic, 14. J. M. Bell, G. H. Derrick, and R. C. McPhedran, “Diffraction
square, or rod cylinders of one relative permittivity gratings in the quasi-static limit,” Opt. Acta 29, 1475–1489
~1982!.
immersed in a homogeneous medium of another rel-
15. Ph. Lalanne and D. Lemercier-Lalanne, “On the effective me-
ative permittivity. Two limitations were observed dium theory of subwavelength periodic structures,” J. Mod.
and identified. For large period-to-wavelength ra- Opt., to be published.
tios, the homogenized structures do not provide a 16. M. G. Moharam, E. B. Grann, D. A. Pommet, and T. K. Gay-
reliable model especially when 3-D photonic crystals lord, “Formulation for stable and efficient implementation of
are studied. Moreover frustrated total internal re- the rigorous coupled-wave analysis of binary gratings,” J. Opt.
flection by tunneling of evanescent orders inside ho- Soc. Am. A 12, 1068 –1076 ~1995!.

20 September 1996 y Vol. 35, No. 27 y APPLIED OPTICS 5379

17. Ph. Lalanne and G. M. Morris, “Highly improved convergence 25. D. Maystre, “Electromagnetic study of photonic bandgaps,”
rate of the coupled-wave method for TM polarization,” J. Opt. Pure Appl. Opt. 3, 975–993 ~1994!.
Soc. Am. A 13, 779 –784 ~1996!. 26. Y. Ono, Y. Kimura, Y. Otha, and N. Nishida, “Antireflection
18. M. G. Moharam, D. A. Pommet, E. B. Grann, and T. K. Gay- effects in ultrahigh spatial-frequency holographic relief grat-
lord, “Stable implementation of the rigorous coupled-wave ings,” Appl. Opt. 26, 1142–1146 ~1987!.
analysis for surface-relief gratings: enhanced transmittance 27. D. H. Raguin and G. M. Morris, “Antireflection structured
matrix approach,” J. Opt. Soc. Am. A 12, 1077–1086 ~1995!. surfaces for the infrared spectral region,” Appl. Opt. 32, 1154 –
19. M. Born and E. Wolf, Principles of Optics, 6th ed. ~Macmillan, 1167 ~1993!.
New York, 1964!, Chap. 1, p. 55. 28. S. T. Peng, T. Tamir, and H. L. Bertoni, “Theory of periodic
20. P. Yeh, Optical Waves in Layered Media ~Wiley, New York, waveguides,” IEEE Trans. Microwave Theory Tech. MTT-23,
1988!, Chap. 5, p. 102. 123–133 ~1977!.
21. Ref. 20, Chap. 7, p. 144. 29. J. Saarinen, E. Noponen, and J. Turunen, “Guided-mode res-
22. R. D. Meade, A. M. Rappe, K. D. Brommer, and J. D. Joanno- onance filters of finite aperture,” Opt. Eng. 34, 2560 –2566
poulos, “Nature of photonic band gap: some insights from a ~1995!.
field analysis,” J. Opt. Soc. Am. B 10, 328 –332 ~1993!. 30. In fact, as was confirmed by RCWA, the structure of Fig. 17,
23. M. Plihal and A. A. Maradudin, “Photonic band structure of similar to its corresponding 3-D version, does not present any
two-dimensional systems: the triangular lattice,” Phys. Rev. gap in the x and y directions. In that particular case, the
B 44, 8565– 8571 ~1991!. EMT does not fail, but, in general, the conclusion derived here
24. R. D. Meade, K. D. Brommer, A. M. Rappe, and J. D. Joanno- holds. For a discussion regarding the 3-D chessboard crystal
poulos, “Existence of a photonic bandgap in two dimensions,” see E. Yablonovitch, “Photonic crystals,” J. Mod. Opt. 41, 173–
Appl. Phys. Lett. 61, 495– 497 ~1992!. 194 ~1994!.

5380 APPLIED OPTICS y Vol. 35, No. 27 y 20 September 1996