Experimental demonstration of broadband absorption enhancement in partially aperiodic silicon nanohole structures

We report the design, fabrication, and optical absorption measurement of silicon membranes patterned with partially aperiodic nanohole structures. We demonstrate excellent agreement between measurement and simulations. We optimize a partially aperiodic structure using a random walk algorithm and demonstrate an experimental broadband absorption of 4.9 times that of a periodic array.

Nanophotonic light-trapping structures 1, 2 hold promise to reduce the material cost and improve the efficiency of solar cells. Most of the light-trapping architectures investigated so far consist of perfectly periodic arrays of sub-wavelength features. Recently, however, alternative lighttrapping schemes have been proposed based on aperiodic photonic patterns. Among these proposals, several have investigated structures with complete aperiodicity 3-6 of either random 3,4,6 or deterministic 5 nature. Such structures were found to improve solar absorption, with the additional benefit of performance insensitivity to incidence angle and polarization. However, accurate full-field electromagnetic modeling of realistic, three-dimensional, completely aperiodic structures remains unfeasible, severely limiting their design and optimization. Computational work has instead focused on partially aperiodic structures [7][8][9][10][11][12][13][14][15] : such structures are aperiodic on smaller length scales, and periodic on a larger scale (typically on the order of a micron). The large-scale periodicity provides a finite computational cell for calculations, making such structures highly amenable to simulation. Partially aperiodic structures also show strong broadband absorption enhancement relative to their periodic counterparts. Moreover, machine based optimization techniques make it possible to deliberately engineer the partially-aperiodic structure 16 , dramatically improving theoretical light-trapping performance 9,11,14,17 .
In this paper, we present direct experimental verification of broadband absorption enhancement in partially aperiodic dielectric light-trapping structures. We choose a silicon membrane patterned with nanoholes as our experimental platform. Systematic large-scale numerical simulations show that partially aperiodic nanohole patterns outperform periodic nanohole arrays, given the same hole size and filling ratio. We then use a random-walk algorithm to identify an optimized aperiodic configuration that maximizes solar absorption. Periodic, random, and optimized aperiodic structures were fabricated and optically characterized. Our 3 measurement results confirm the accuracy of our simulations and verify the theoretically predicted increased absorption. We experimentally demonstrate a 4.9 times increase in absorption in the 600nm -1000 nm range for the optimized aperiodic structure, compared to a periodic array.
Figs. 1(a) and (b) illustrate the geometry of the structures under study. A free-standing silicon nanomembrane with thickness t=310 nm is patterned with either periodic ( Fig.1 (a)) or partially aperiodic ( Fig. 1(b)) nanohole patterns. The periodic pattern is a square lattice of nanoholes with We first carried out systematic numerical simulations to predict the effect of partially aperiodicity on broadband absorption. We use the transfer matrix method (TMM) 18 to calculate the absorptance spectra of various structures at normal incidence. Absorptance spectra for two orthogonal polarizations were averaged to obtain the average absorptance for randomly polarized incidence light. We use the tabulated optical constants for crystalline silicon 19 for our simulations.
In order to quantify the broadband optical absorption performance in the solar spectrum, we define a figure of merit (F.O.M.) as follows: A super cell periodicity of 600 nm was found to be optimal. This periodicity coincides with the optimal lattice constant for a perfectly periodic nanohole array with the same filling ratio (optimization not shown). For the same super cell periodicity, we observe that different nanohole configurations can have significantly different absorption.
In order to fully exploit partial aperiodicity for absorption enhancement, we carried out random walk optimizations 9 for every super cell periodicity, starting with the same initial periodic 6 configuration described above. The optimization was performed for 100 iterations for each super cell periodicity, subject to the same constraint that the inter-hole distance is larger than 40 nm. Drawing on the above analysis, we focus on partially aperiodic structures with a super cell periodicity of 600 nm for our experimental demonstration. The patterns chosen for fabrication are shown in Fig. 2(a -c). Fig. 2(a) illustrates the periodic nanohole array. Fig. 2(b) is the partially aperiodic pattern with a F.O.M. closest to the mean value of our 100 randomly generated configurations. We refer to this configuration as the average aperiodic pattern. Fig.   2(c) shows the optimized aperiodic structure.
We use electron beam lithography combined with inductively coupled plasma reactive ion etching (ICP-RIE) to transfer the nanohole patterns into the device layer of a silicon-on-insulator (SOI) wafer (SOITEC). Every pattern was defined within a circular region with a diameter of 50 µm. After the pattern formation, a 6 mm by 6 mm silicon membrane area was defined by standard UV photolithography and ICP-RIE. The center square part of the silicon membrane with dimensions of 500 µm by 500 µm was aligned to overlap the nanohole patterns, while the rest of the membrane was patterned with access holes to facilitate the wet etching of the buried oxide (BOX) layer. Finally, the silicon membrane was released from the handle silicon wafer in 49% hydrofluoric acid (HF) and wet-transferred 21 to an oxidized silicon wafer. The device area was centered over a perforated window in the silicon wafer to obtain a free-standing membrane.  average (e), and optimized (f) partially aperiodic nanohole patterns. Image processing software 22 revealed that the differences in silicon filling ratio between these patterns are below 2%.  lens (f=30mm, N.A.=0.39). The collected light was focused into multimode fibers connected to the spectrometer for analyzing both the reflectance R exp (ω) and transmittance T exp (ω) spectra. The experimental absorptance spectrum A exp (ω) can then be determined by A exp (ω) = 1-R exp (ω)-T exp (ω). We used a silver mirror as the reflection reference while the transmission reference was air.
We simulate the fabricated structures assuming a silicon membrane thickness of 328 nm and a uniform nanohole diameter of 110nm. The thickness was determined by fitting the measured optical transmittance spectrum of an unpatterned area, and the nanohole size was found to give the best agreement between simulation and measurement. These values are slightly different from those used for our calculations for Fig. 1(c). Nonetheless, simulations with these experimentally determined structural parameters give the same trends as the original simulations.
We focus on the spectral data between 600 nm to 1000 nm. For wavelengths below 600 nm, light can be diffracted from the aperiodic structures, leading to incomplete collection by our finite NA lenses. We thus focus on the 600 nm to 1000 nm range, where accurate measurements of optical absorptance can be obtained.
In Fig. 3, we plot the simulated (left column) and measured (right column) spectra. Overall, we observe excellent agreement between simulation and measurement. The experimental spectra tend to be smoother with broader features, which can be attributed to the non-uniformity in the size and shape of nanoholes throughout each pattern 23 , as well as the finite angular spread of the incidence beam 24 .
We first examine the perfectly periodic array (red lines). We observe characteristic Fabry-Perot fringes in both transmittance and reflectance spectra, with a few guided resonance induced features 25 in the short wavelength range. An experimental value of 5.6% was obtained for the F.O.M. between 600 nm and 1000 nm, based on the data in Fig. 3(f). The average aperiodic structure (blue lines) has a similar reflectance to the periodic structure (Figs. 3(a-b), but strongly This represents a broadband absorption enhancement 4.79 times higher than a solid film with the same thickness, with only 75.5% of the silicon volume, and 4.9 times higher than a periodic array with the same silicon volume.
In order to qualitatively explain the improvement in optical absorption from periodic, to average, to optimized aperiodic structures, we examine the spatial Fourier spectra. The FFTs of the dielectric structure are shown in Fig. 4(d-f), and the corresponding real space configurations are again shown in Fig. 4(a-c) for reference. The amplitudes of the Fourier components are normalized to the "DC" component of the Fourier transform. The Fourier transform of the periodic square array is also a square lattice, featuring discrete, well-defined peaks with a 200 nm periodicity. Such small length scales can only support the excitation of a few guided resonance modes in the silicon nanohole array for wavelengths above 600 nm, consistent with the spectra shown in Fig. 3(c). As a result, the absorption performance is poor. In contrast, in partially aperiodic configurations, the randomization of the nanohole positions within a super 11 cell generates non-zero spatial frequencies between (2π/200) nm -1 and (2π/600) nm -1 , corresponding to effective periodicities of 200 nm and 600 nm, respectively. This redistribution of the Fourier components facilitates the coupling between the incident light and in-plane optical resonance modes for wavelength above 600 nm, leading to many more resonance-induced features in the spectra (see Fig. 3(c)) and significantly enhanced broadband optical absorption.
Compared to the average aperiodic structure, the optimized structure further suppresses the spatial frequencies corresponding to a 200 nm effective periodicity, while clearly enhancing the spatial frequencies corresponding to a beneficial 600 nm effective periodicity and enabling even better light-trapping. In summary, we demonstrate both numerically and experimentally that silicon membranes patterned with partially aperiodic nanohole structures yield higher broadband optical absorption than those with simple periodic patternings. In particular, optimized aperiodic structures perform significantly better (4.9 times in the current study) than periodic ones, though they both contain the same total amount of absorbing material.
Since the random-walk optimization algorithm used in this paper is not guaranteed to find the global optimum structure, our results may be understood as a conservative estimate on performance: further optimization may yield aperiodic structures with even higher broadband absorption. Moreover, we expect that optimization starting from a periodic array with larger lattice constant (600 nm) may yield even more efficient absorbers. We expect that the fundamental optical physics demonstrated here may contribute to the development of highly efficient, thin film photovoltaic devices.