Outdoor measurements of a photovoltaic system using diffractive spectrum-splitting and concentration Open Access

Despite the wide spectral content of sunlight (with most of the power in the range of $∼$350 nm to $∼$2000 nm), efficiency of single-junction solar cells is limited due to thermalization and non-absorption losses.^1,2 Photons that have energy below the bandgap of the absorber are not absorbed. This is termed as the non-absorption loss. On the other hand, if the photon energy is higher than the absorber bandgap, the excess energy is lost to heat (referred to as the thermalization loss). According to Polman and Atwater,² thermalization and non-absorption account for at least 40% of the drop in efficiency. One way to overcome these loss mechanisms is to use multi-junction solar cells.^3–5 In a traditional multi-junction configuration, solar cells of different bandgaps are stacked on top of each other in descending order of bandgaps, i.e., the highest bandgap device on the top and the lowest at the bottom. As a result, light first strikes the highest bandgap device, which absorbs the high-energy photons and the low-energy photons are absorbed in the subsequent low bandgap cells. Multi-junction solar cells require lattice-matched growth, which limits the choice of materials. Furthermore, since the sub-cells are connected in series, performance of the device is limited by the sub-cell generating the lowest current. Multi-junction architecture can also be realized by mechanical stacking, for example, by physical wafer bonding^6,7 or by using organic adhesives.⁸ Performance of these devices is constrained by the current-matching requirement. Use of insulating organic adhesive can somewhat mitigate this issue but bring other disadvantages such as poor optical transmission and thermal conduction between the sub-cells.⁹ 

An alternative is to use spectral-splitting optical elements to split the incident sunlight spatially and to use separate cells with bandgaps matched to the spectral bands.^10,11 This technique allows the independent operation of sub-cells, and eliminates the lattice-matching and current-matching requirements. Several spectrum-splitting techniques have been presented in the literature for photovoltaic application.^12–29 The most common approach is to use dichroic filters which allow certain bandwidth of incident light to pass while reflecting the rest thus splitting the spectrum into two spectral bands.^12–21 These dichroic mirrors can be arranged in different configurations to achieve spectrum splitting. A photovoltaic (PV) system consisting of two tandem cells (GaInP/GaAs and GaInAsP/GaInAs) and a single junction cell (Si) along with dichroic prism, lens array and concentrators was described by Barnett et al.¹³ 42.7% efficiency was reported for this system considering ideal dichroic splitting optics. Another architecture consisting of dichroic mirrors, lenses and concentrators with two tandem cells (GaInP/GaAs and GaInAsP/GaInAs) achieved 38.5% module efficiency.¹⁴ 34% efficiency was demonstrated for a system in which one tandem cell (GaInP/GaAs) and two single junction cells (Si and GaSb) with dichroic filters on top of them were arranged in parallelepiped configuration.¹⁵ Polyhedral specular reflector design was proposed for ultra-high efficiency modules.¹⁶ In this design, seven subcells can be arranged in a parallelepiped configuration. With moderate concentration, these designs can lead to >50% module efficiency. Holographic elements have also been a popular choice for photovoltaic spectrum splitting.^22–25 Holographic elements can be designed to diffract incident light at particular angle depending on the wavelengths thus in effect achieving spectrum splitting. A PV module consisting of reflection holograms with GaAs and Si cells has been proposed.²³ Overall system efficiency was calculated 27.94% with the holographic filter. Another design consisting of a set of twelve volume holographic gratings with four subcells predicted module efficiency of 37.1% at 672X concentration.²⁴ Spectrum splitting using prisms has also been implemented.^26,27 In this case, dispersion causes incident lights to bend at different angles leading to spectrum splitting. Polycarbonate based prismatic lens has been demonstrated to simultaneously concentrate and split incident spectrum.^26,27

Spectrum splitting techniques are particularly useful for space photovoltaics.¹⁰ A concentrating-spectrum splitting system for orbital power generation was proposed by Onffroy, et al.²⁸ The design consisted of several dichroic mirrors to achieve multiple spectral bands. Michel, et al. presented a blazed diffraction grating based spectrum splitting technique for space applications.²⁹ An important consideration is that the spectrum splitting components need to be simple (preferably planar), compact and lightweight so that they can be transported and assembled easily. Furthermore, the improved performance offered by these additional components should be commensurate with the additional cost. This implies that the components should be inexpensive to manufacture and have high broadband efficiency. These conditions are more stringent when spectrum splitting approaches are to be incorporated in terrestrial applications, since space applications usually allow higher cost compared to terrestrial ones.

As discussed above, concept of using spectrum splitting optics in photovoltaic system is not novel, and there have been several design proposals and demonstrations in the literature. However, all these approaches suffer from one or more drawbacks causing their incorporation into the photovoltaic systems somewhat impractical. For example, dichroic filters are expensive. Furthermore, in order to achieve multiple spectral bands, the solar cells and the dichroic filters need to be arranged in a complex configuration which is undesirable for a PV system. The number of spectral bands are also limited due to the Fresnel reflection losses from multiple surfaces. Holographic diffractive elements offer relatively cheap solution, but they suffer from low diffraction efficiencies and the specific position of the spectral bands are challenging to achieve. Multiple holographic elements are required to achieve efficient broadband spectrum splitting. Refraction based spectrum splitting systems suffer from multiple drawbacks such as inability to scale to large areas, low efficiency, lack of controlling spatial position and band edges of spectral bands.

We previously demonstrated a planar, broadband diffractive-optical element (termed a polychromat) that performs simultaneous spectrum-splitting and concentration.^30,31 We experimentally demonstrated increase in peak power density of 20% and 36% with of two- and three-band spectrum-splitting and concentration, respectively.^31,32 However, these systems were designed for and the experiments were performed with artificial light sources. Here, we present experimental demonstration of a system that achieves two-band spectrum splitting and 3X geometric concentration under the standard AM1.5 spectrum. The solar cells were made of GaInP and CIGS. Our system was assembled and tested under direct sunlight in Salt Lake City, Utah over three separate days in September 2015. We measured an increase of $∼$25% in the peak power density compared to the case without spectrum-splitting and concentration. As far as we are aware, this is the first demonstration of efficient diffractive spectrum-splitting and concentration under ambient sunlight resulting in an increase in the overall PV power output.

The geometry of our system is illustrated in Fig. 1(a). The polychromat diffracts the incident sunlight in such a way that high-energy photons are directed towards the GaInP cell, while the low-energy photons are incident upon the CIGS (labeled C3847) cell. The polychromat also concentrates the light as indicated. The polychromat is pixelated along the X direction and uniform along the Y direction. Thus, spectrum splitting and concentration occur only along the X direction. The width of each pixel is $3μ$m, while the height varies from 0 to $2.5μ$m with up to 61 levels in between (Fig. 1(b)). These values are constrained primarily by the grayscale process used to fabricate the polychromat. The height profile was optimized to maximize the output power of the solar cells. This was done using a modified version of the direct-binary-search (DBS) algorithm combined with a scalar diffraction formulation and an optoelectronic model.^30,31 DBS is an iterative method and it perturbs the pixels of the polychromat before a metric of overall electrical power output is evaluated. The optimized height profile is shown in Fig. 1(c). A magnified view of the leftmost $100μ$m of the polychromat is illustrated in Fig. 1(d). Details of the design process have been described in our previous works.^31,32

The solar cells were selected due to their optimal bandgaps and availability. GaInP cell has a bandgap of 1.8eV, while the bandgap of C3847 cell is 1.13eV. The measured external quantum efficiencies (EQE) of these cells confirm that the absorption of GaInP cell is mainly in the visible region, while that of C3847 extends into the infrared region. Fig. 2 shows the measured EQE plots of the solar cells along with the AM1.5 spectrum. The GaInP cell is 5 mm × 5 mm and the C3847 cell is 4.75 mm × 9 mm. Overall size of one period of the polychromat is 14.25 mm × 9 mm. This period is repeated three times during the fabrication to satisfy the periodic-boundary condition assumed during design.

Grayscale lithography was used to fabricate the polychromat. In this process, a commercially available photoresist, Shipley 1813 was first spin-coated on a 3” × 3” RCA-cleaned glass substrate at 1000 rpm for 60s followed by soft baking in an oven at 110° C for 30 minutes. The optimized design height profile (see Fig. 1(c)) was converted to a grayscale map using grayscale-calibration data. This grayscale map was used to expose the sample using the Heidelberg $μ$PG 101 pattern generator, which works by direct-laser-writing. The sample was then developed in AZ 1:1 developer for 90s. In grayscale patterning of positive photoresist, larger exposure dose (or larger grayscale) leads to shallower features after development. An optical micrograph of the polychromat is shown in Fig. 3(a). To estimate the fabrication errors, a surface profiler (Tencor P20h) was used to measure heights along the X direction of the polychromat. Comparison between the design and measured heights show an average error of $∼$50 nm, which is sufficient to generate excellent optical performance based on our previous study of fabrication-error tolerance.^31,32 In the future, the polychromat can be readily replicated by imprinting techniques with high fidelity for mass production.³³ 

The optimized polychromat is meant to efficiently manipulate the sunlight intensity distribution in the spatial-spectral domain. Therefore, it is essential to measure the spatial-spectral response of the polychromat. We first illuminated the polychromat by a collimated beam from a broadband supercontinuum source (NKT Photonics). A multimode fiber was placed at the image plane of the polychromat, which was at a distance d =163 mm (see Fig. 1(a)). The fiber tip was mounted on a motorized stage and the other end was connected to a spectrometer (Ocean optics Jazz). The fiber tip was scanned for 14.25mm along the X direction in the image plane and the spectrum was stored at each location. These measurements were used to create the spatial-spectral intensity map shown in Fig. 3(b). We also simulated the spatial-spectral map of the designed polychromat at the same distance d, shown in Fig. 3(c). Excellent agreement between measurement and simulation is noted. The optical efficiency of a particular band can be defined as the ratio of the integrated power incident upon that band to the total incident power from the sun. Measured and simulated optical efficiencies as a function of wavelength are shown in Fig. 3(d). They match quite well. The decrease of optical efficiency in measurement is primarily due to fabrication error of the polychromat (⁠$∼$50nm) and the finite aperture of the multimode filter tip used in measurement. Nevertheless, this fabrication error poses trivial impact on the ultimate performance of the system, as described below.

Electrical measurements were performed under ambient sunlight on three separate days with mostly clear skies: September 18, 20 and 23, 2015. Fig. 4(a) shows a photograph of the measurement setup. An optical rail was placed on an adjustable mount. Two filter holders were used to hold the polychromat and the solar cells. To make the incident light normal to the polychromat, a simple alignment procedure was developed using a square-array mask and a target screen as shown in Fig. 4(b). The shadow of the mask pattern was aligned to the target pattern on the screen. After alignment, the solar cells were placed at a distance of 163mm from the polychromat. The solar cells were positioned at appropriate bands of the image plane. Current-voltage characteristics were measured for each cell with and without the polychromat. For the reference, we used the measurements without the polychromat, but with the light passing through unpatterned glass substrate to account for any Fresnel reflections. The results obtained on September 18 are plotted in Fig. 5. Short-circuit current densities of GaInP and C3847 cells were increased by 36.79% and 13.46%, respectively. Open circuit voltages were increased by 1.11% and 0.25%, respectively. Power densities were also calculated from the current voltage measurements. Increase in peak power densities were 39.46% and 12.84%, respectively. The overall increase was 25.80% compared to the performance without the polychromat. Measurements from all three days are summarized in Table I. The principle of power density enhancement in this photovoltaic system primarily comes from two factors: (1) more photons over a broader spectrum are harnessed due to the introduction of the lower bandgap, which essentially increases current; (2) utilization rate of photon energy is increased due to the introduction of higher bandgap, which increases output voltage. These two factors effectively mitigates the aforementioned non-absorption loss and thermalization loss, respectively.

It is important that the solar cells are aligned with the appropriate bands in the image plane of the polychromat as accurately as possible. To investigate the effect of relative shift between the solar cells and corresponding spectral bands on the power boost, we performed a simulation study that is summarized in Fig. 6(a) and (b). Positive alignment errors were defined as the shift of the solar cells to the right (+X direction) with respect to the corresponding spectral bands (Fig. 6(a)). Net power boost decreases for shifts in both direction (Fig. 6(b)). Power boost reduces to zero at $∼$3.75 mm and $∼$3 mm for positive and negative shifts, respectively.

Another important consideration is the impact of the angle of incidence of sunlight on the polychromat. The polychromat was designed assuming normal incidence. To investigate the impact of oblique incidence, we considered incident angles (from normal) of -1.4° to 1.1° and calculated the diffracted field and the corresponding power boost of the solar cells for each angle. Fig. 6(c) illustrates the percent change in power boosts for different incident angles. Net power boost reduces to zero at $∼$1.35° and $∼$1.05°. Note that the results look similar to the ones presented in Fig. 6(b). This is due to the fact that with the change of the angle of incidence, the diffraction pattern undergoes a lateral shift (in X direction), which is equivalent to misalignment of solar cells with respect to the spectral bands. The angle of incidence can be approximately related to the lateral misalignment $Δx$ by $θ=Δx/d$⁠, where d is the gap between the polychromat and the solar cell plane (see Fig. 1(a)). In general, such polychromat based spectrum splitting system would require solar tracking (in this case single axis tracking) to maintain high power boost. By reducing the distance, d, it is possible to reduce the required precision of tracking. In order to reduce d, while maintaining concentration, we will require higher diffractive power, which in turn requires smaller polychromat pixels. This can be improved in the future with better manufacturing processes.

The polychromat was designed for the standard AM1.5 spectrum.³⁴ However it is important to know the effect of variation in spectral irradiance on the performance of the system. We firstly obtained spectral irradiance data (1) for different hours of two separate days (21st of June and December, 2015) and (2) for hour 12, day 15 of each month over a year using the solar spectrum calculator developed by PV Lighthouse.³⁵ Input parameters for this calculator are given in Table II. We simulated the net power boost of GaInP-C3847 configuration for each spectrum keeping all other parameters the same as were used for designing the original polychromat for the AM1.5 spectrum. Although the open circuit voltages and fill factors of the two cells change with spectrum, we neglected this in our simulation for simplicity. The results are shown in Fig. 7. The change in net power boost, compared with the reference case without spectrum-splitting and concentration, varies by less than 2% over the course of the daylight hours for both days considered (Figs. 7(d), (e), (g), (h)). Over the course of the year, the power boost from the CIGS cell changes by up to 10% (Fig. 7(f)). However, the net power boost of the system is dominated by the GaInP cell, which does not change much. As a result, the change in the net power boost (at hour 12, day 15 of the month) over the course of the year is less than 1% (Fig. 7(i)). These results indicate that the polychromat-based spectrum splitting is robust to small changes in the incident spectrum over time.

It has been found that degradation in CIGS cells primarily results from increase in series resistance.^36,37 This, in turn reduces the fill factor. To analyze the effect of fill factor loss on polychromat performance, we reduced the fill factors of GaInP and CIGS cells by 0 to 10% and calculated the net power boost for each fill-factor combination. The results are summarized in Fig. 8. No change in net power boost was found if the fill factor of both cells reduced by same amount (along the diagonal line in Fig. 8). Net power boost increases if the fill factor of only the CIGS cell reduces and that of the GaInP cell remains unchanged. The opposite happens, if fill factor of GaInP reduces, i.e., net power boost decreases. The net power boost changes by as much as +/- 2% as a result of the degradation of the fill factors.

Here, we show that one can increase the output power from a photovoltaic device by spectrally separating sunlight into two bands and concentrating these bands onto two single-junction solar cells positioned in the same plane via a broadband, planar diffractive optic. Such configuration successfully circumvents the issues of current- and lattice-matching present in traditional multi-junction photovoltaics. Specifically, we used GaInP and CIGS cells with a geometric concentration of 3X. It was experimentally demonstrated an increase in output power of about 25% under ambient sunlight. Our numerical calculation indicates that this power boost is insusceptible to daily and annual variation of solar spectrum. Fill-factor degradation also brings negligible effect on its performances. The principle can clearly be extended to more bands and higher concentrations in order to achieve ultra-high efficiency photovoltaics.³¹ Furthermore, since the polychromat is an ultra-thin planar optic, it can be inexpensively manufactured using imprinting techniques over large areas.³³ 

We would like to thank Daniel Friedman, Kannan Ramanathan and Lorelle Mansfield for providing the GaInP and CIGS cells. We also thank Brian Baker for assistance with fabrication of the polychromat. The project was funded by a DOE Sunshot Grant, EE0005959 and a NASA Early Stage Innovations Grant, NNX14AB13G. R.M. was partially funded by the Utah Science Technology and Research (USTAR) Initiative. We are grateful to Jose Dominguez-Caballero for assistance with the polychromat design, and Brian Baker, Steve Pritchett and Brian van Devener for support in the Utah Nanofabrication facility. RM is the co-inventor of a patent related to this technology. RM is also the co-founder of PointSpectrum Corp., which is commercializing this technology.