pubs.acs.

org/NanoLett

Toward the Lambertian Limit of Light

Trapping in Thin Nanostructured Silicon Solar
Cells
Sang Eon Han, and Gang Chen*
Department of Mechanical Engineering, Massachusetts Institute of Technology, 77 Massachusetts Avenue,
Cambridge, Massachusetts 02139, United States

ABSTRACT We examine light trapping in thin silicon nanostructures for solar cell applications. Using group theory, we design surface
nanostructures with an absorptance exceeding the Lambertian limit over a broad band at normal incidence. Further, we demonstrate
that the absorptance of nanorod arrays closely follows the Lambertian limit for isotropic incident radiation. These effects correspond
to a reduction in silicon mass by 2 orders of magnitude, pointing to the promising future of thin crystalline silicon solar cells.

KEYWORDS Light trapping, group theory, symmetry breaking, optical absorption, Lambertian limit

C urrently crystalline silicon (c-Si) yields the highest the c-Si wafer thickness can be reduced from the current
efficiency in single junction solar cells and takes up 100-300 µm to 1-3 µm, significantly reducing the materi-
∼80% of the solar cell market. A challenge in c-Si als cost.
solar cells is that, due to its poor absorption, c-Si wafers are Various schemes have been studied to enhance optical
typically 100-300 µm thick and account for ∼40% of the absorption in Si thin films including plasmonic scattering,6
total module cost.1 Because of this limitation, alternative diffraction gratings,7-10 and nanorod arrays.11-14 However,
materials with stronger absorption such as amorphous previous work has not demonstrated how to approach the
silicon (a-Si), cadmium telluride (CdTe), or copper indium Lambertian limit when the film thickness is a few microme-
gallium diselenide (CIGS) are used for thin film solar cells. ters. Even at normal incidence, the absorption enhancement
However, solar cells based on these materials have not
factor could exceed the Lambertian limit over very limited
achieved efficiencies as high as c-Si. Moreover, some of them
spectral ranges.8-10 Typically, absorption decreased at other
have issues of material abundancy and toxicity. If c-Si with
angles of incidence so that angle-averaged spectrum was far
a thickness of a few micrometers can be made to absorb as
below the Lambertian limit. The limit was broken only when
efficiently as thick Si wafers, a significant cost reduction is
the film thickness was a few nanometers.15 In this case, only
expected.1
the fundamental limit was considered, which is reached for
Lambertian light trapping sets the thermodynamic limit
weak absorbers. Here, using group theory, we consider how
of maximum absorption in a slab of thickness in the geo-
the Lambertian limit can be realized in thin structures with
metric optics regime.2-4 This is attained when the surface
absorptance comparable to that of thick wafers.
is ideally rough such that the incident light is randomly
scattered. Optical path length can be increased by a factor We begin our discussion by considering the properties
of 4n2 compared to flat surfaces in the limit of weak of gratings. In general, the following three conditions will
absorption, where n is the refractive index of the film.1 yield high-performance gratings. First, two-dimensional grat-
Conventionally, the 4n2 enhancement is known as the ings are preferred to one-dimensional ones, since two-
Lambertian limit or the Yablonovitch limit. If some optical dimensional gratings offer more diffraction channels.8 Sec-
elements are used that ideally allow light within an ac- ond, gratings couple well to incident light when they are
ceptance angle 2θ, the thermodynamic limit increases to tapered because the optical density varies gradually.16 When
4n2/sin2 θ.4,5 This indicates that the Lambertian limit is the used on the front surfaces, they can also decrease reflec-
thermodynamic limit only for isotropically incident radiation tion.17 Third, the grating period should be slightly smaller
which corresponds to θ ) π/2. For Si with a refractive index than the typical wavelength. If the period is too small, light
∼3.5, the 4n2 limit means that the path length can be does not sense the structural details and optical diffraction
increased by a factor of ∼50. If this limit can be reached, is weak. When the period is larger than the wavelength,
reflection orders higher than the zeroth order are possible,
leading to a decrease in absorption. In this case, light is not
* Corresponding author, [email protected].
Received for review: 08/23/2010
confined well in the film but leaks through the higher-order
Published on Web: 10/06/2010 channels.15

© 2010 American Chemical Society 4692 DOI: 10.1021/nl1029804 | Nano Lett. 2010, 10, 4692–4696
 As an example, we consider pyramid structures with
square bases. Prior studies of submicrometer pyramid
structures considered relatively thick wafers.18 Here we
calculated absorptance for thickness ∼2 µm using the
transfer matrix method19 and experimental values for the
dielectric function.20 The pyramid structures were optimized
to obtain the maximum ultimate efficiency.12,21 The ultimate
efficiency η is defined as the maximum efficiency of a
photovoltaic cell as the temperature approaches 0 K when
each photon with energy greater than the band gap produces
one electron-hole pair

∫0λ I(λ)A(λ) λλg dλ

g FIGURE 1. Absorptance of a pyramid structure. Red solid line
represents absorptance of a pyramid structure. To smooth out the
η) (1) narrow peaks, the spectrum was averaged over ∆λ /λ2 ∼ 0.05 µm-1.
∫0∞ I(λ)A(λ) dλ Black dashed line represents the Lambertian limit of a c-Si film of
equal thickness. The pyramid height and the length of its base are
566 and 800 nm, respectively. A 90 nm Si3N4 layer (red) is coated
on the pyramids. The c-Si film (gray) thickness is 2.012 µm and a
SiO2 layer (blue) of thickness 717 nm is placed between the Si and
where I is the AM1.5G solar spectrum,22 A the absorptance, the Ag back reflector (green).
λ the wavelength, and λg the wavelength corresponding to
the band gap. Throughout this work, the absorptance cal-
culations were performed for unpolarized light. Because of
surface recombination issues, we fixed the ratio of the height
to the base length of the pyramid to 12 which corresponds
to the same surface area increase of 3 times over a flat
film as in conventional c-Si solar cells. In all structures in this
work, a SiO2 film was placed between the Si layer and a Ag
back reflector, and a Si3N4 antireflection layer of thickness
90 nm was coated on the Si structures. A number of
absorptance peaks occurred in the spectra corresponding to
waveguide resonances in the film.4,8,9 The optimum period
of the pyramid structure was found to be 800 nm, which
corresponds to the typical wavelength according to the third
condition given above. The optimized absorptance spectrum FIGURE 2. (a) (left) Unit cell of the pyramid structure. The blue line
is the 4-fold rotation axis. (right) Top view of the pyramid unit cell.
at normal incidence is compared to the Lambertian limit in Red dotted line represents mirror planes. (b) Top view of a skewed
Figure 1. Overall, the absorptance of the pyramid grating is pyramid. (c, d) Hz-field profile of the modes in a pyramid structure
not very far from the Lambertian limit. The performance of on a square lattice (c) and in structures with a 6-fold rotation axis
and six mirror planes (d), where the z axis is normal to the paper.
this structure is significant considering two points. First, an + and - represent the sign of the Hz-field and letters denote the
appreciable amount of light is absorbed in Ag at long irreducible representations of the electric field. The E representation
wavelengths and only absorption in Si is accounted for in (c), and the E1 and E2 representations (d) are doubly degenerate.
Only the E (c) and E1 (d) representations couple to light.
the calculations, while the Lambertian limit assumes a
perfect metal. Second, the low reflectance of the pyramid
structure at short wavelengths approaches the Lambertian be the operator that performs mirror reflection on a Bloch
limit which assumes perfect antireflection. field in a structure with mirror symmetries. When it is
Although the pyramid structure is promising, we note that operated on a Bloch vector field k with a wave vector k that
its absorptance is below the Lambertian limit over most of lies in the mirror plane, the resulting field satisfies the
the spectrum even at normal incidence. To overcome this, equation26,27
we consider the symmetry properties of the pyramid struc-
ture. The point group symmetry of a unit cell of this structure PMk ) (k (2)
contains a 4-fold rotation axis and four mirror planes (see
Figure 2a). This symmetry remains the same when the cell
is arranged in a square lattice as in our case. In general, Thus, the field can be either symmetric (+) or antisymmetric
symmetries result in degeneracy of the modes. More im- (-) under mirror reflection. Let the incident light be linearly
portantly, the presence of mirror planes results in certain polarized with an electric field Exx̂, where x̂ is a unit vector
modes not coupling to incident light.23-25 To see this, let PM in the x direction. Mirror reflection operation on this field

© 2010 American Chemical Society 4693 DOI: 10.1021/nl1029804 | Nano Lett. 2010, 10, 4692-–4696
about the yz plane results in -Exx̂. Thus, the incident electric
field is antisymmetric under the mirror reflection and the
symmetric electric fields in the structure cannot be excited
by the incident wave, leading to a decrease in absorption.
Group theory predicts26 that, in our example, the modes
can be classified into five categories according to their
symmetries as shown in Figure 2c where the magnetic fields
in the z direction are schematically drawn. The correspond-
ing electric fields parallel to the xy plane can be obtained by
taking the curl of the magnetic fields according to the
Maxwell’s equation. Among the five, only one doubly de-
generate mode (E) has the electric fields that are antisym-
metric under mirror reflection about the xz or yz plane and
can couple to light. For another example, a hexagonal lattice
where the unit cell contains a cylindrical or conical object
has six mirror planes. The number of possible symmetries
of modes in this case is six and the electric fields of only one
of them (E1) are antisymmetric under mirror reflection about
the xz or yz plane as shown in Figure 2d. Thus only this
degenerate mode can couple to light. If all the possible
symmetries of modes occurred equally frequently over a
range of spectrum, only a sixth of the modes would couple
to light, resulting in a small number of absorptance peaks.
From the above observations, we note that the mirror
symmetries need to be broken to improve absorption. One
way to break symmetry is to destroy the periodic arrange-
ment.28 However, in this case the number of reflection
channels increases at wavelengths where only zeroth order
reflection is possible at perfect periodicity. This effect will
decrease absorption. Thus, we distort the shape of the
objects in the unit cell.25,29 For example, we consider a
skewed pyramid with a rectangular base as shown in Figure
2b. When this nonsymmetric shape is arranged in a hex-
agonal lattice, no mirror planes exist. We optimized this
structure to obtain the maximum efficiency. The ratio of the FIGURE 3. (a) Absorptance of a skewed pyramid structure at normal
incidence (red solid line) and averaged over all directions of
pyramid height to the longer side of its base was kept to incidence (blue solid line). Angle-averaged absorptance of the
1/21/2 so that the surface area was approximately twice that pyramid structure in Figure 1 is also shown (green dashed line). To
of a flat surface. The optimized absorptance is shown in smooth out the narrow peaks, the spectrum was averaged over ∆λ
/λ2 ∼ 0.05 µm-1. The black dashed line represents the Lambertian
Figure 3a. We observe that the absorptance of these skewed limit of a c-Si film of equal mass. The skewed pyramid height and
pyramids exceeds the Lambertian limit at normal incidence the longer side of its base are 636 and 900 nm, respectively. A 90
over a broad range of the spectrum. The calculated ultimate nm Si3N4 layer (red) is coated on the skewed pyramids. The c-Si film
(gray) thickness is 2.012 µm and a SiO2 layer (blue) of thickness 741
efficiency was slightly higher than the limit. Figure S1 in the nm is placed between the Si and the Ag back reflector (green). (b)
Supporting Information shows that absorption increases as Angular distribution of the ultimate efficiency of the skewed
the pyramids are more skewed. Therefore, the effect of pyramids. The radial axis represents the polar angle. (c) The first
Brillouin zone of a hexagonal lattice. The letters represent various
symmetry-breaking on absorption is significant. Note that symmetry points.
this effect is related to optical phase. Because of the lack of
structural symmetry, the optical phase of any mode cannot the angle-averaged absorption of the skewed pyramid grat-
be symmetric under mirror reflection and all modes couple ing is less than but close to the limit at long wavelengths and
to light. Thus this effect is different from absorption en- higher than nonskewed pyramids in Figure 1 over most of
hancement by symmetry breaking of macroscopic struc- the spectrum. The angular distribution of the ultimate ef-
tures30 where phase information is absent. ficiency in Figure 3b shows that, in general, the absorptance
Although the absorptance is high at normal incidence, we decreases at large polar angles.
need to consider absorptance averaged over all incident We note from Figures 1 and 3a that the symmetry
angles to make a fair comparison to the Lambertian limit breaking is not as effective at off-normal incidence as at
which is for isotropic incident radiation. Figure 3a shows that normal incidence. This can be explained by considering

© 2010 American Chemical Society 4694 DOI: 10.1021/nl1029804 | Nano Lett. 2010, 10, 4692-–4696
 FIGURE 5. Ultimate efficiency of various structures as a function of
their effective thickness. The effective thickness is the thickness of
a film that has the same Si mass as the structures. For the rod
structure, the actual length of the rods is ∼4 times its effective
thickness. Lines with filled dots are for normal incidence and three
open dots are for angle-averaged response. Black, red, and green
open dots represent pyramids, skewed pyramids, and rods, respec-
tively. The structural parameters of pyramids, skewed pyramids, and
rods are the same as in Figures 1, 3, and 4, respectively. The blue
line represents an unstructured c-Si film with a Ag back reflector
and a Si3N4 antireflection coating of thickness 61 nm. The gray area
is the range of ultimate efficiency corresponding to an unstructured
film of 100-300 µm.

FIGURE 4. (a) Absorptance of the nanorod structure with skewed in Figure 4a. Moreover, the angle-averaged absorptance
pyramids on its top at normal incidence (red solid line) and averaged
over all directions of incidence (blue solid line). The pyramid height closely follows the Lambertian limit. We also note that, at
and the longer side of the base are 1125 and 450 nm, respectively. long wavelengths, the angle-averaged absorptance is even
The length of the rod of rectangular cross section is 6.508 µm. A 90 higher than the Lambertian limit over a limited spectral
nm Si3N4 layer (red) is coated on the tapered rods. A SiO2 layer (blue)
of thickness 953 nm is placed between the Si rod and the Ag back range, which is not possible for macroscopic rod structures
reflector (green). To smooth out the narrow peaks, the spectrum was where the density of photonic states is not modified by the
averaged over ∆λ /λ2 ∼ 0.05 µm-1. The black dashed line represents structures.4 For this rod structure, the angular distribution
the Lambertian limit of a c-Si film of equal Si mass. The equivalent
film thickness is 1.768 µm. (b) Angular distribution of the ultimate of the ultimate efficiency in Figure 4b shows that the polar
efficiency of the nanorods. The radial axis represents the polar angle. angle dependence is weaker than the skewed pyramids on
a film. Because the angular distribution still bears some
resemblance to 6-fold rotation symmetry, the symmetry
symmetries at various directions. In Figure 3c, the first
Brillouin zone is drawn for a hexagonal lattice. According to breaking might be pursued further.
group theory, the mirror reflection operations that leave a To appreciate the absorption enhancement of the struc-
wave vector unchanged within reciprocal lattice vectors tures discussed above, we calculated their ultimate efficien-
constitute the symmetry group for the field having the wave cies as a function of the Si layer thickness. Even though
vector.26 For example, the mirror plane at the T (Σ) point is angle-averaged ultimate efficiencies are of interest, computa-
the xz (yz) plane and the mirror planes at the M point tions for all directions require a long time. Thus we focus on
are the xz and yz planes. At normal incidence of the Γ point, normal incidence and refer to the angle-averaged response for
the symmetry group is the same as the lattice and six mirror the selected thicknesses of the structures in Figure 5. Figure 5
planes exist. In general, symmetry is the highest at the Γ shows that to obtain the ultimate efficiency of a flat Si wafer
point and is lowered at other directions. Specifically, the of thickness 100-300 µm, films of only 2-4 µm are needed
number of mirror planes at the T (or Σ), M, and K points is for the skewed pyramids and the Si mass equivalent to 1-3
one, two, and three, respectively, and an arbitrary point is µm films are needed for nonsymmetric nanoroad structures
nonsymmetric. Thus, breaking the structural symmetry (see the gray area). Moreover, the nonsymmetric rod array
is expected to be most effective at normal incidence but is achieves efficiencies higher than the Lambertian limit at
still advantageous near various high symmetry points. normal incidence over all thickness ranges of interest.
To increase the angle-averaged absorption further, we Therefore, the simple technique of symmetry breaking
consider a nonsymmetric nanorod array which has skewed makes it possible not only to realize the Lambertian limit in
pyramids at its top as shown in Figure 4a (inset). With this thin Si films but also to reduce the cost of c-Si solar cells
structure, the ultimate efficiency is above the Lambertian significantly by decreasing the c-Si mass by 2 orders of
limit of a film of equal c-Si mass at normal incidence as seen magnitude.

© 2010 American Chemical Society 4695 DOI: 10.1021/nl1029804 | Nano Lett. 2010, 10, 4692-–4696
 To conclude, we investigated light-trapping properties of (6) Stuart, H. R.; Hall, D. G. Appl. Phys. Lett. 1996, 69, 2327.
periodic nanostructures in the wave optics regime. Using (7) Sheng, P.; Bloch, A. N.; Stepleman, R. S. Appl. Phys. Lett. 1983,
43, 579.
tapered two-dimensional gratings that are nonsymmetric, (8) Bermel, P.; Luo, C.; Zeng, L.; Kimerling, L. C.; Joannopoulos, J. D.
we obtained absorptance close to the Lambertian limit at Opt. Express 2007, 15, 16986.
normal incidence. Further, we demonstrated that rod arrays (9) Mallick, S. B.; Agrawal, M.; Peumans, P. Opt. Express 2010, 18,
5691.
with nonsymmetric tapered tops can exhibit absorptance (10) Zhou, D.; Biswas, R. J. Appl. Phys. 2008, 103, No. 093102.
close to the Lambertian limit even when averaged over all (11) Hu, L.; Chen, G. Nano Lett. 2007, 7, 3249.
directions of incidence. This effect amounts to the reduction (12) Han, S. E.; Chen, G. Nano Lett. 2010, 10, 1012.
of the crystalline Si wafer thickness by 2 orders of magnitude (13) Kelzenberg, M. D.; Boettcher, S. W.; Petykiewicz, J. A.; Turner-
Evans, D. B.; Putnam, M. C.; Warren, E. L.; Spurgeon, J. M.; Briggs,
so that thin crystalline Si nanostructures, including films, can R. M.; Lewis, N. S.; Atwater, H. A. Nat. Mater. 2010, 9, 239.
have comparable absorption as thick wafers. Hence, these (14) Garnett, E.; Yang, P. Nano Lett. 2010, 10, 1082.
structures have great potential to make high efficiency thin (15) Yu, Z.; Raman, A.; Fan, S. Proc. Nat. Acad. Sci. U.S.A. 2010, doi:
10.1073/pnas.1008296107.
crystalline Si cells. The design principles presented in this (16) Chutinan, A.; Kherani, N. P.; Zukotynski, S. Opt. Express 2009,
work are not restricted to crystalline Si but can be applied 17, 8871.
to any kind of solar cell. (17) Sai, H.; Fujii, H.; Arafune, K.; Ohshita, Y.; Kanamori, Y.; Yugami,
H.; Yamaguchi, M. Jpn. J. Appl. Phys. 2007, 46, 3333.
(18) Sai, H.; Kanamori, Y.; Arafune, K.; Ohshita, Y.; Yamaguchi, M.
Acknowledgment. This work was supported by NSF Prog. Photovoltaics 2007, 15, 415.
through UC Berkeley SINAM. This research was supported (19) Bell, P. M.; Pendry, J. B.; Martı́n-Moreno, L.; Ward, A. J. Comput.
in part by the National Science Foundation through TeraGrid Phys. Commun. 1995, 85, 306.
resources provided by Purdue University under Grant Num- (20) Handbook of Optical Constants of Solids; Palik, E. D., Ed.; Aca-
demic: Orlando, FL, 1985.
ber TG-PHY100046.
(21) Shockley, W.; Queisser, H. J. J. Appl. Phys. 1961, 32, 510.
(22) Air Mass 1.5 Spectra; American Society for Testing and Materials,
Supporting Information Available. Additional informa- http://rredc.nrel.gov/solar/spectra/am1.5/.
tion regarding the effect of pyramid skewness on absorption (23) Robertson, W. M.; Arjavalingam, G.; Meade, R. D.; Brommer,
and the polarization dependence of ultimate efficiency. This K. D.; Rappe, A. M.; Joannopoulos, J. D. Phys. Rev. Lett. 1992, 68,
2023.
material is available free of charge via the Internet at http://
(24) Ochiai, T.; Sakoda, K. Phys. Rev. B 2001, 63, 125107.
pubs.acs.org. (25) Kilic, O.; Digonnet, M.; Kino, G.; Solgaard, O. Opt. Express 2008,
16, 13090.
REFERENCES AND NOTES (26) Inui, T.; Tanabe, Y.; Onodera, Y. Group Theory and Its Applications
(1) Catchpole, K. R.; Polman, A. Opt. Express 2008, 16, 21793. in Physics; Springer-Verlag: Berlin, 1990.
(2) Yablonovitch, J. Opt. Soc. Am. 1982, 72, 899. (27) Joannopoulos, J. D.; Johnson, S. G.; Winn, J. N.; Meade, R. D.
(3) Stuart, H. R.; Hall, D. G. J. Opt. Soc. Am. A 1997, 14, 3001. Photonic Crystals: Molding the Flow of Light, 2nd ed.; Princeton
(4) Agrawal, M. Photonic Design for Efficient Solid-State Energy University Press: Princeton, NJ, 2008.
Conversion Ph.D. dissertation, Department of Electrical Engineer- (28) Chutinan, A.; John, S. Phys. Rev. A 2008, 78, 023825.
ing, Stanford University: Stanford, CA, 2008; Chapter 6. (29) Yu, Z.; Raman, A.; Fan, S. Opt. Express 2010, 18, A366.
(5) Campbell, P.; Green, M. A. IEEE Trans. Electron Devices 1986, ED- (30) Campbell, P.; Wenham, S. R.; Green, M. A. Sol. Energy Mater. Sol.
33, 234. Cells 1993, 31, 133.

© 2010 American Chemical Society 4696 DOI: 10.1021/nl1029804 | Nano Lett. 2010, 10, 4692-–4696