Upconversion in solar cells
© van Sark et al.; licensee Springer. 2013
Received: 25 September 2012
Accepted: 4 January 2013
Published: 15 February 2013
Skip to main content
© van Sark et al.; licensee Springer. 2013
Received: 25 September 2012
Accepted: 4 January 2013
Published: 15 February 2013
The possibility to tune chemical and physical properties in nanosized materials has a strong impact on a variety of technologies, including photovoltaics. One of the prominent research areas of nanomaterials for photovoltaics involves spectral conversion. Modification of the spectrum requires down- and/or upconversion or downshifting of the spectrum, meaning that the energy of photons is modified to either lower (down) or higher (up) energy. Nanostructures such as quantum dots, luminescent dye molecules, and lanthanide-doped glasses are capable of absorbing photons at a certain wavelength and emitting photons at a different (shorter or longer) wavelength. We will discuss upconversion by lanthanide compounds in various host materials and will further demonstrate upconversion to work for thin-film silicon solar cells.
Attaining high conversion efficiencies at low cost has been the key driver in photovoltaics (PV) research and development already for many decades, and this has resulted in a PV module cost of around US$0.5 per watt peak capacity today. Some commercially available modules have surpassed the 20% efficiency limit, and laboratory silicon solar cells are getting closer and closer  to the Shockley-Queisser limit of 31% for single-junction silicon cells . However, a fundamental issue is that conventional single-junction semiconductor solar cells only effectively convert photons of energy close to the bandgap (Eg) as a result of the mismatch between the incident solar spectrum and the spectral absorption properties of the material . Photons with energy (Eph) smaller than the bandgap are not absorbed, and their energy is not used for carrier generation. Photons with energy (Eph) larger than the bandgap are absorbed, but the excess energy Eph – Eg is lost due to thermalization of the generated electrons. These fundamental spectral losses are approximately 50% . Several approaches have been suggested to overcome these losses, e.g., multiple stacked cells , intermediate bandgaps , multiple exciton generation , quantum dot concentrators [8, 9], and spectral converters, the latter being down- and upconverters [10, 11] and downshifters [12, 13]. In these so-called third- or next-generation PV concepts [14, 15], nanotechnology is deemed essential in realizing most of these concepts .
Spectral conversion aims at modifying the incident solar spectrum such that a better match is obtained with the wavelength-dependent conversion efficiency of the solar cell. Its advantage is that it can be applied to existing solar cells and that optimization of the solar cell and spectral converter can be done separately. Different types of spectral conversion can be distinguished: (a) upconversion, in which two low-energy (sub-bandgap) photons are combined to give one high-energy photon; (b) downshifting or luminescence, in which one high-energy photon is transformed into one lower energy photon; and (c) downconversion or quantum cutting, in which one high-energy photon is transformed into two lower energy photons. Downshifting can give an efficiency increase by shifting photons to a spectral region where the solar cell has a higher quantum efficiency, i.e., basically improving the blue response of the solar cell, and improvements of up to 10% relative efficiency increase have been predicted . Up- and downconversion, however, are predicted to be able to raise the efficiency above the SQ limit [10, 11]. For example, Richards  has shown for crystalline silicon (c-Si) that the potential relative gain in efficiency could be 32% and 35% for downconversion and upconversion, respectively, both calculated for the standard 1,000-W/m2 air mass (AM) 1.5 solar spectrum.
Research on spectral conversion is focused on organic dyes, quantum dots, lanthanide ions, and transition metal ion systems for up- and downconversion [13, 17, 18]. An upconversion layer is to be placed at the back of the solar cells, and by converting part of the transmitted photons to wavelengths that can be absorbed, it is relatively easy to identify a positive contribution from the upconversion layer, even if the upconversion efficiency is low. In contrast, proof-of-principle experiments in solar cells are complicated for downconverters and downshifters because of the likelihood of competing non-radiative processes. These downconverters and downshifters have to be placed at the front of the solar cell, and any efficiency loss will reduce the overall efficiency of the system.
The usefulness of down- and upconversion and downshifting depends on the incident spectrum and intensity. While solar cells are designed and tested according to the ASTM standard , these conditions are rarely met outdoors. Spectral conditions for solar cells vary from AM0 (extraterrestrial) via AM1 (equator, summer and winter solstice) to AM10 (sunrise, sunset). The weighted average photon energy (APE)  can be used to parameterize this; the APE (using the range 300 to 1,400 nm) of AM1.5G is 1.674 eV, while the APE of AM0 and AM10 are 1.697 and 1.307 eV, respectively. Further, overcast skies cause higher scattering leading to diffuse spectra, which are blue-rich, e.g., the APE of the AM1.5 diffuse spectrum is calculated to be 2.005 eV, indeed much larger than the APE of the AM1.5 direct spectrum of 1.610 eV. As downconversion and downshifting effectively red-shift spectra, the more relative energy an incident spectrum contains in the blue part of the spectrum (high APE), the more gain can be expected [12, 23]. Application of downconversion layers will therefore be more beneficial for regions with high diffuse irradiation fraction, such as Northwestern Europe, where this fraction can be 50% or higher. In contrast, solar cells with upconversion (UC) layers will be performing well in countries with high direct irradiation fractions or in early morning and evening due to the high air mass resulting in low APE, albeit that the non-linear response to intensity may be limiting. Up- and downconversion layers could be combined on the same solar cell to overcome regionally dependent efficiencies. Optimization of either up- or downconversion layers could be very effective if the solar cell bandgap is a free design parameter.
In this paper, we focus on upconversion materials for solar cells, in particular for thin-film silicon solar cells. We describe the present state of the art in upconversion materials and application in solar cells.
Upconversion was suggested by Bloembergen  and was related to the development of infrared (IR) detectors: IR photons would be detected through sequential absorption, as would be possible by the arrangement of energy levels of a solid. However, as Auzel pointed out, the essential role of energy transfer was only recognized nearly 20 years later . Several types of upconversion mechanism exist, of which the addition de photon par transferts d’energie or, in English, energy transfer upconversion mechanism is the most efficient; it involves energy transfer from an excited ion, named sensitizer, to a neighboring ion, named activator . Others are two-step absorption, being a ground-state absorption followed by an excited-state absorption, and second-harmonic generation. The latter mechanism requires extremely high intensities, of about 1010 times the sun’s intensity on a sunny day, to take place  and can therefore be ruled out as a viable mechanism for solar cell enhancement.
Upconverters usually combine an active ion, of which the energy level scheme is employed for absorption, and a host material, in which the active ion is embedded. The most efficient upconversion has been reported for the lanthanide ion couples (Yb, Er) and (Yb, Tm) . The first demonstration of such an upconversion layer was reported by Gibart et al.  who used a GaAs cell on top of a vitroceramic containing Yb3+ and Er3+: it showed 2.5% efficiency under very high excitation densities.
Strümpel et al. have identified the materials of possible use in up- (and down-) conversion for solar cells . In addition to the NaYF4:(Er,Yb) phosphor, they suggest the use of BaCl2:(Er3+,Dy3+) , as chlorides were thought to be a better compromise between having a low phonon energy and a high-excitation spectrum, compared to the NaYF4[31, 32]. These lower phonon energies lead to lower non-radiative losses. In addition, the emission spectrum of dysprosium is similar to that of erbium, but the content of Dy3+ should be <0.1% to avoid quenching [25, 26].
NaYF4 co-doped with (Er3+, Yb3+) is, to date, the most efficient upconverter [27, 33], with approximately 50% of all absorbed NIR photons upconverted and emitted in the visible wavelength range. However, the (Yb, Er) couple is not considered beneficial for upconversion in c-Si cells as silicon also absorbs in the 920- to 980-nm wavelength range. These phosphors can be useful for solar cells based on higher bandgap materials such as the dye-sensitized solar cell (DSSC) or Grätzel cell , a-Si(Ge):H, or CdTe. Different mechanisms are responsible for the upconversion luminescence. The Yb3+ ion has only one excited state and is an ideal sensitizer for Er3+ because of the relatively high oscillator strength of the 2F7/2 → 2F5/2 transition and the fact that Er3+ has a state with similar energy (4I11/2) which is populated by energy transfer from Yb3+ (see Figure 2). Population of the first excited state of Er3+ (4I11/2) is therefore directly proportional to the incoming light intensity. When upconversion is the main route, energy transfer from the first excited state (4I11/2) to the second excited state (4F7/2) follows. After some small energy-relaxation steps, emission is observed from the 4S3/2, 2H11/2 (green), and 4F9/2 (red) states. The 4F9/2 can also be reached after energy transfer from the 4I13/2 state.
where Pin is the incoming light intensity and ISCUC is the photogenerated short-circuit current increase due to upconversion in the solar cell. As a result, for current increase due to upconversion, a quadratic power dependence on the concentration factor is expected.
De Wild et al. recently applied a commercially available upconverter, Gd2O2S:Yb3+, Er3+, in which Yb3+ absorbs light around 980 nm and Er3+ emits in the visible spectrum (400 to 700 nm) . These absorption and emission wavelengths are very suitable for use with wide-bandgap solar cells, such as single-junction a-Si:H, as the absorption edge of a-Si:H is between the wavelengths for absorption and emission. Furthermore, the spectral response is very high in that emission range. The dominant upconversion mechanism in Gd2O2S:Yb3+, Er3+ is energy transfer upconversion.
Nanocrystals of NaYF4:Er3+, Yb3+ also show upconversion. An advantage of using nanocrystals is that transparent solutions or transparent matrices with upconverting nanocrystals can be obtained. Recent reviews on upconverting nanoparticles summarize the status of a variety of upconverter materials that are presently available as nanocrystals, mostly phosphate and fluoride nanocrystals [38, 39]. However, a problem with upconversion nanocrystals is the lower upconversion efficiency . There is a clear decrease in efficiency with decreasing size in the relevant size regime between 8 and 100 nm, which is probably related to surface effects and quenching by coupling with high-energy vibrations in molecules attached to the surface.
Upconversion systems consisting of lanthanide nanocrystals of YbPO4 and LuPO4 have been demonstrated to be visible by the naked eye in transparent solutions, however at efficiency lower than that of solid-state upconversion phosphors . Other host lattices (NaXF4, X = Y, Gd, La) have been used, and co-doping with Yb3+ and Er3+, or Yb3+ and Tm3+ appeared successful, where Yb3+ acts as sensitizer. Nanocrystals of <30 nm in size, to prevent scattering in solution, have been prepared, and they can be easily dissolved in organic solvents forming colloidal solutions, without agglomeration. Further efficiency increase is possible by growing a shell of undoped NaYF4 around the nanocrystal; in addition, surface modification is needed to allow dissolution in water, for use in biological labeling.
Porous silicon layers are investigated for use as upconverter layers as host for rare-earth ions because these ions can easily penetrate the host due to the large surface area and porosity. A simple and low-cost dipping method has been reported , in which a porous silicon layer is dipped into a nitrate solution of erbium and ytterbium in ethanol (Er(NO3)3:Yb(NO3)3:C2H5OH), which is followed by a spin-on procedure and a thermal activation process at 900°C. Excitation of the sample at 980 nm revealed upconversion processes as visible and NIR photoluminescence is observed; co-doping of Yb with Er is essential, and doping only with Er shows substantial quenching effects .
Finally, sensitized triplet-triplet annihilation (TTA) using highly photostable metal-organic chromophores in conjunction with energetically appropriate aromatic hydrocarbons has been shown to be another alternative upconversion system [43, 44]. This mechanism was shown to take place under ambient laboratory conditions, i.e., low-light-intensity conditions, clearly of importance for outdoor operation of solar cells. These chromophores (porphyrins in this case) can be easily incorporated in a solid polymer such that the materials can be treated as thin-film materials . A problem with TTA upconverters is the spectral range. No efficient upconversion of NIR radiation at wavelengths beyond 800 nm has been reported which limits the use to wide-bandgap solar cells [37, 46].
Upconversion in solar cells was calculated to potentially lead to a maximum conversion efficiency of 47.6%  for nonconcentrated sunlight using a 6,000-K blackbody spectrum in detailed-balance calculations. This optimum is reached for a solar cell material of approximately 2-eV bandgap. Applied on the back of silicon solar cells, the efficiency limit would be approximately 37% . The analysis of the energy content of the incident AM1.5G spectrum shows that cells with an upconverter layer would benefit from an extra amount of 35% light incident in the silicon solar cell . An extension to the models described above was presented in a study by Trupke et al. , in which realistic spectra were used to calculate limiting efficiency values for upconversion systems. Using an AM1.5G spectrum leads to a somewhat higher efficiency of 50.69% for a cell with a bandgap of 2.0 eV. For silicon, the limiting efficiency would be 40.2% or nearly 10% larger than the value of 37% obtained for the 6,000-K blackbody spectrum . This increase was explained by the fact that absorption in the earth’s atmosphere at energies lower than 1.5 eV (as evident in the AM1.5G spectrum) leads to a decrease in light intensity. Badescu and Badescu  have presented an improved model that takes into account the refractive index of solar cell and converter materials in a proper manner. Two configurations are studied: cell and rear converter, the usual upconverter application, and front converter and cell (FC-C). They confirm the earlier results of Trupke et al.  in that the limiting efficiency is larger than that of a cell alone, with higher efficiencies at high concentration. Also, the FC-C combination, i.e., upconverter layer on top of the cell, does not improve the efficiency, which is obvious. Further, building on the work by Trupke et al. , the variation of refractive indexes of cell and converter was studied, and it was found that the limiting efficiency increases with the refractive index of both cell and upconverter. In practice, a converter layer may have a lower refractive index (1.5, for a transparent polymer: polymethylmethacrylate (PMMA) ) than a cell (3.4). Using a material with a similar refractive index as the cell would improve the efficiency by about 10%. Finally, a recent study on realistic upconverter and solar cell systems, in which non-ideal cell and upconverters were considered, corroborates the above findings . In this study, non-ideal absorption and radiative recombination, as well as non-radiative relaxation in the upconverter, were taken into account. Atre and Dionne also stressed that thin-film PV with wide-bandgap materials may benefit the most from including upconverters .
The first experiment in which an upconversion layer was applied on the back of solar cells comprised an ultrathin (3 μm) GaAs cell (bandgap 1.43 eV) on top of a 100-μm-thick vitroceramic containing Yb3+ and Er3+: it showed 2.5% efficiency upon excitation of 256-kW/m2 monochromatic sub-bandgap (1.391 eV) laser light (1 W on 0.039-cm2 cell area) as well as a clear quadratic dependence on incident light intensity. An efficiency of the solar cell of 2.5% was obtained even though the excitation wavelength (891 nm) is not resonant with the absorption peak of Yb3+ (approximately 980 nm), leading to inefficient upconversion. Secondly, the design was such that not all emitted photons were directed to the solar cell.
Richards and Shalav  showed upconversion under a lower excitation density of 2.4 W/ cm2 reaching 3.4% quantum efficiency at 1,523 nm in a crystalline silicon solar cell with NaYF4 doped with Er3+ as upconverter. This was for a system optimized for the wavelength of 1,523 nm. Intensity-dependent measurements showed that the upconversion efficiency was approaching its maximum due to saturation effects [51, 52]. Under broadband excitation, upconversion was shown for the same system by Goldschmidt et al.  reaching an upconversion efficiency of 1%. Since c-Si has a rather small bandgap (1.12 eV), transmission losses due to the low energy photons are not as high as for wider bandgap solar cells. Hence, the efficiency gain for larger bandgap solar cells is expected to be higher. Upconversion of 980-nm light was also demonstrated in DSSCs [54, 55] and of 750-nm light in ultrathin (50 nm) a-Si:H solar cells in 2012 . In the latter proof-of-principle experiment, for the first time, an organic upconverter was applied.
The photogenerated current could be increased by 40% if the spectral response was sustained at high level up to the bandgap cutoff at 700 nm and by even more if light with wavelengths λ > 700 nm could be more fully absorbed. These two effects can be achieved with the upconversion layer, combined with a highly reflecting back contact. While the upconversion layer converts sub-bandgap photons to ‘super’-bandgap photons that can thus be absorbed, a non-conductive reflector is a much better alternative than any metallic mirror, thus sending back both the unabsorbed super-bandgap photons as well as the upconverted super-bandgap photons into the cell. It is commonly estimated that the stabilized efficiency of the approximately 9% cell can be enhanced to approximately 12%. Besides a-Si, a material denoted as protocrystalline Si could be used; this is an amorphous material that is characterized by an enhanced medium-range structural order and a higher stability against light-induced degradation compared to standard amorphous silicon. The performance stability of protocrystalline silicon is within 10% of the initial performance; its bandgap is slightly higher than that of amorphous silicon.
De Wild et al.  have demonstrated upconversion for a-Si cells with NaYF4 co-doped with (Er3+, Yb3+) as upconverter. The upconverter shows absorption at 980 nm (by the Yb3+ ion) leading to efficient emission of 653- (red) and 520- to 540-nm (green) light (by the Er3+) after a two-step energy transfer process. The narrow absorption band around 980 nm for Yb3+ limits the spectral range of the IR light that can be used for upconversion. An external quantum efficiency of 0.02% at 980-nm laser irradiation was obtained. By using a third ion (for example, Ti3+) as a sensitizer, the full spectral range between 700 and 980 nm can be efficiently absorbed and converted to red and green light by the Yb-Er couple. A transition metal ion such as Ti3+ incorporated in the host lattice absorbs over a broad spectral region and transfers the energy to a nearby Yb3+ ion through a dipole-dipole interaction [27, 31]. The resulting light emission in the green and red regions is very well absorbed by the cell with very good quantum efficiency for electron–hole generation.
Concentrated broadband light excitation has recently been used to study two types of bifacial a-Si:H solar cells that were made with and without Gd2O2S:Er3+, Yb3+ upconverter attached at the back of the cells . The upconverter powder mixture was applied to the rear of the solar cells by first dissolving it in a solution of PMMA in chloroform, after which it was drop cast. Two types of p-i-n a-Si:H solar cells were made: one on Asahi-textured SnO2:F glass and one on flat ZnO:Al 0.5% superstrate. The efficiency obtained for the cells is 8% for textured and 5% for flat solar cells, both without a back reflector. Backside illumination yields an efficiency of 5% for textured solar cells and 4% for flat solar cells. With illumination from the back, the efficiency is lower because the generation profile is reversed within the cell, and thus, the photogenerated minority carriers have to travel the largest mean distance, rather than the majority carriers. The spectral response measured through the n-layer shows a quantum efficiency of 0.7 for both textured and flat solar cells at 550 nm; the spectral response at 660 nm is lower, i.e., 0.4 for textured cells and 0.15 for flat cells. The transmission for 900 to 1,040 nm was 40% to 45% for the textured solar cells and between 60% and 80% for the flat solar cells. The thickness of the i-layer was chosen such that an interference maximum occurs at 950 nm, increasing the transmission at this wavelength. As a result, more light can be absorbed by the upconverter layer in the case of the flat solar cell configuration. Concentration levels of up to 25 times were reached using near-infrared light from a solar simulator.
Monochromatic laser light with wavelength at 981 nm and a power density of 0.2 W/cm2 was used for textured solar cells and yielded a current density of 0.14 mA/cm2 for the upconverter solar cells and 0.04 mA/cm2 for the reference solar cells. Evidently, the contribution of sub-bandgap absorption is much smaller using monochromatic laser light. The current due to the upconverter is comparable to the current measured under 20 sun: approximately 0.1 mA/cm2 (see Figure 6). This is remarkable in two ways. First, the results are in contrast with previously reported experiments with broadband excitation of c-Si solar cells , where the current under broadband excitation was much smaller than that under laser light excitation. However, in , another upconverter was applied (NaYF4) and different processes occur in the upconverter, namely excited state absorption. In the upconverter in this work (Gd2O2S), energy transfer upconversion is the main upconversion path, and the broadband absorption of Yb3+ may increase the transfer between Yb3+ and Er3+.
Second, the power that is absorbed by Yb3+ is 3.44 mW/cm2, which yields a broadband power density of 70 mW/cm2 under a concentration of 20 sun. This is three times less than the power density of the laser. A large difference here is that for broadband illumination, a 900-nm-long pass filter was used. Therefore, light of the solar simulator extends to further than 1,600 nm; thus, also the 4I13/2 state of Er3+ is excited directly. Addition of other paths that lead to upconverted light may contribute to the current. These paths may be non-resonant excited-state absorption between the energy levels of Er3+ or three-photon absorption around 1,540 nm at the 4I13/2 state of Er3+ (see Figure 2). Direct excitation of the 4I13/2 state of Er3+ followed by excited-state absorption from 4I13/2 to 2F9/2 results in a visible photon around 650 nm, while three-photon absorption around 1,540 nm results in emission from the 2F9/2 state too. Wavelengths required for these transitions are around 1,540 and 1,200 nm, which are present within the broad excitation spectrum. Contribution of these upconversion routes increases the emission and thereby the current in the solar cells.
Upconversion for solar cells is an emerging field, and the contribution of upconverter research to upconverter solar cell research increases rapidly. However, up to now, only proof-of-principle experiments have been performed on solar cells, mainly due to the high intensities that are deemed necessary. Some routes to enhance absorption are presently being developed, such as external sensitization and plasmonics.
External sensitization can be achieved by, e.g., quantum dots or plasmons. Quantum dots (QDs) can be incorporated in a concentrator plate where the QDs absorb over a broad spectral range in the IR and emit in a narrow line, e.g., around 1,520 nm, resonant with the Er3+ upconversion wavelength. Energy transfer from the QDs to Er3+ in this scheme is through radiative energy transfer. The viability of this concept was proven by Pan et al.  in c-Si solar cells, where a layer with QDs was placed below the upconverter layer. With the QDs, more light was absorbed and upconverted, which was proven by measuring the excitation spectra for the upconverted emission. The increased upconverted emission resulted in higher currents in the solar cell.
More challenging are options to enhance upconversion efficiencies by manipulating emission and excitation processes through plasmonic coupling . The use of plasmonic effects with upconverter materials is a new and emerging field, with many possibilities and challenges. In general, plasmonic resonance can be used in two ways to increase the upconversion efficiency: by enhancing either the absorption strength or the emission strength. When the absorption strength is enhanced, the emission increases with the square of the enhancement in the non-linear regime. In the case of resonance between the plasmon and the optical transition, strong enhancement can be achieved. Recently, Atre et al.  have modelled the effects of a spherical nanocresent consisting of a core of an upconverter material and a crescent-shaped Ag shell. A 10-fold increase in absorption as well as a 100-fold increase in above-bandgap power emission toward the solar cell was calculated. A similar study has been performed using Au nanoparticles . Experimental proof has recently been reported by Saboktakin et al. . A related method is to enhance the absorption strength by nanofocusing of light in tapered metallic structures . At the edges, enhancement has been reported due to focusing of the light in these areas. The other option is enhancing the emission. In this case, the emission of the upconverter is enhanced by nearby plasmon resonances . Since the field enhancement decays away exponentially with the distance to metallic nanoparticle, the upconverter species have to be close to the surface of the nanoparticle to benefit from the field enhancement effects. For organic molecules, this presents no problem because the molecules are small enough to be placed in the field. For lanthanide upconverters, this is more difficult because the ions are typically contained in materials with grain sizes in the micrometer range. However, several groups have managed to make nanosized NaYF4 particles [67, 68]. This offers the possibility of plasmonic enhancement for lanthanide upconverters and decreases the light intensity required for efficient upconversion. Alternatively, upconversion using sensitized triplet-triplet annihilation in organic molecules at moderate monochromatic excitation intensities increases the a-Si:H cell efficiency as well [46, 56].
In this paper, we have briefly reviewed upconversion for solar cells and have presented some relevant experimental results, focusing on the application of lanthanides in combination with wide-bandgap solar cells (a-Si:H). The proof-of-principle experiments that have been performed so far have shown that high intensities are needed to demonstrate upconversion for solar cells. Within the lanthanides, large steps in decreasing the necessary intensity are not expected. In the organic field, there is a rapid decrease in intensity needed for efficient upconversion, while conversion wavelengths are not appropriate yet.
External sensitization using quantum dots or options to enhance upconversion efficiencies by manipulating emission and excitation processes through plasmonic coupling may offer routes for successful upconversion deployment in solar cells. With further developments in these organic molecules, it remains to be seen if lanthanide upconverters, with plasmonic enhancement, or molecules in which TTA can be employed, will be the upconverter material for the future in wide-bandgap solar cells.
average photon energy
dye-sensitized solar cell
external collection efficiency
front converter and cell
The authors gratefully acknowledge Agentschap NL for the partial financial support within the framework of the EOS-NEO Programme as well as the Utrecht University Focus and Mass Programme, Karine van der Werf, Caspar van Bommel, Bart Sasbrink, Martin Huijzer, and Thijs Duindam for the sample preparation and characterization. AM acknowledges the support from the EU-FP7 NANOSPEC Programme (STREP 246200).
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.