Today's industry standard solar cell for concentrator applications is a triple-junction cell made of Ga1−xInxP (1.8–1.9 eV), Ga1−yInyAs (1.3–1.4 eV), and Ge (0.7 eV). This type of solar cell has reached record efficiencies up to 41.6% under concentrated sunlight illumination and is produced by several companies [1-3]. In the case of the lattice matched Ga0.49In0.51P/Ga0.99In0.01As/Ge material combination, the Ge bottom junction is known to generate a large excess current, which limits the device performance. Furthermore, the Ge cell suffers of a comparably low current and voltage performance. More advanced triple-junction solar cells have therefore used III–V compounds with bandgaps of around 1.0 eV instead of Ge. Examples are dilute nitride semiconductors like GaInNAs  or inverted metamorphic GaInAs . These solar cells reach efficiencies up to 44.4 % at 302 sun concentration [6, 7]. The performance can be further improved by adding additional junctions. The theoretical optimum for a four-junction solar cell under AM1.5d spectral conditions is reached for a set of declining bandgap energies with 1.9, 1.4, 1.0, and 0.5 eV. In 2011, a combination close to this theoretical optimum was realized with a GaInP/GaAs/GaInNAs/Ge four-junction cell,  but the performance of this device was not satisfactory due to insufficient quality of the dilute nitride material grown by metal-organic-vapor-phase-epitaxy. More promising results were reported for inverted metamorphic devices  reaching AM0 (space solar spectrum) one-sun efficiencies up to 33.6%. In this configuration, two lattice-mismatched Ga1−xInxAs bottom junctions are combined with a Ga0.5In0.5P/GaAs top tandem by using buffer layers to relax the crystal lattice. Dislocations cannot be completely avoided in this configuration, and this leads to constraints in device performance. Wafer bonding is another technology to combine lattice-mismatched materials without creating dislocations. Two crystal structures are brought closely together forming atomic bonds at the interface [10, 11]. This technology requires two epitaxial structures with low surface roughness and a specific surface preparation of the materials but offers the advantage of combining the best defect-free semiconductors in a multi-junction solar cell device. Wafer bonding has been used by several authors in the past [12-15], but the present work, for the first time, presents results of a four-junction device, which is at the level of the best triple-junction solar cells. This is achieved by a Ga0.51In0.49P/GaAs top tandem bonded to a (Ga0.16In0.84)(As0.31P0.69)/Ga0.47In0.53As bottom tandem solar cell (Figure 1). This combination is close to the ideal bandgap combination for a four-junction device and allows one to reach excellent material quality for all junctions.
The epitaxial growth of the III–V compound semiconductor layer structure was performed by metal-organic vapor phase epitaxy in an Aixtron 2800-G4 TM reactor with an 8 × 4-inch configuration at Fraunhofer ISE. Two separate growth runs led to an inverted Ga0.51In0.49P/GaAs top tandem cell on GaAs and an upright (Ga0.16In0.84)(As0.31P0.69)/Ga0.47In0.53As bottom tandem cell structure on InP. The diameter of the substrates was 100 mm in both cases. The overall layer structure for the four-junction solar cell contains the four pn-junctions, and an additional number of barrier and tunnel diode layers between the subcells. All junctions were grown lattice matched to the underlying substrate, which avoids formation of dislocations and ensures outstanding material quality. Highly doped n-type layers of GaAs and InP with ND > 1018 cm−3 were used at the bond interface. The high doping level is beneficial for a low resistance of the bond interface. This is especially important for the application of solar cells under concentrated sunlight illumination where typically current densities of 5–10 A/cm2 are flowing through this device.
The top and bottom tandem solar cells were polished to obtain a root mean square surface roughness below 0.3 nm, which was confirmed by atomic force microscopy measurements on a typical area of 5 × 5 µm2. The two wafers were then cleaned and brought into close contact, and the wafer bond was initiated. After post-bond annealing, stable bonds are formed between the atoms on both sides of the interface. All process steps for the bonding were prepared by the Soitec and CEA-LETI team in a class 100 cleanroom facility using proprietary SmartCut technology experience.
It is important to avoid contamination of the wafers before the bonding as particles may cause large bonding defects. Figure 2 shows a scanning acoustic microscope image of the bonded solar cell wafer pair. Two circles with a clear white contrast in the microscope image are likely due to particles formed during epitaxy or handling of the wafers. Most of the wafer surface was successfully bonded, and scanning electron microscopy images of a solar cell cross-section reveal no indication of defects at the interface (Figure 2 (right)). The conductivity of the bond has been analyzed and a low resistance of <10 mOhm cm2 was confirmed, suitable for the operation of the solar cell under high concentration.
After the wafer bonding, the GaAs substrate was removed by wet chemical etching, and ohmic front and back side contacts were applied at Fraunhofer ISE. The front metal contacts had a finger width of 5 µm, resulting in low shadowing losses. A MgF2/TaOx two layer anti-reflection coating was evaporated onto a 25 nm thick AlInP window layer forming the front-side passivation of the Ga0.51In0.49P top cell. All solar cells were finally separated by a wet chemical mesa etch. Several hundred solar cell devices were processed on a full 100 mm wafer. The area of the cells was varying between 5.2 mm2 and 4 cm2.
Quantum efficiencies  and IV characteristics were measured in the Fraunhofer ISE CalLab using a spectrally adjustable solar simulator to measure the IV characteristics under one-sun AM1.5d (1000 W/m2, 25 °C) standard test conditions. The setup of the solar simulator and the test procedure are described in more detail in reference . The performance of the devices under concentration was determined by the use of a Xe-flash simulator with adjustable distance between the flash bulb and the measurement plane. Because of the small area of the concentrator solar cells, the efficiency depends significantly on the correct measurement of the mesa area. This was done with high precision using a LEXT laser confocal microscope.
Finally, the electroluminescence (EL) emission spectrum of the four solar cell junctions at 298 K was measured with four individual mini spectrometers which are coupled to the solar cell via optical fibers. The EL signal of the two high bandgap junctions was investigated with Si detector arrays. Two additional GaInAs detector arrays were used for the measurement of the low bandgap subcells. The procedure of determining the individual subcell voltages from the EL spectrum at different current densities follows the references [18, 19].
3 RESULTS AND DISCUSSION
Solar cell structures for the InP-based and GaAs-based tandem devices were grown separately and investigated before joining them together as a four-junction cell. Epitaxy process parameters were optimized to achieve high quantum efficiencies in all subcells and to ensure high peak tunnel current densities >20 A/cm2 for the tunnel diodes. Figure 3 shows the absolute external quantum efficiency (EQE) of the best wafer bonded GaInP/GaAs//GaInAsP/GaInAs four-junction solar cell device. The current under the AM1.5d ASTM G173-03 spectrum is limited by the GaInP top subcell with a calibrated photocurrent density of 12.42 mA/cm2 (see IV curve in Figure 4). The other subcell currents are calculated from the EQE to be 12.60 mA/cm2 for GaAs, 12.54 mA/cm2 for GaInAsP, and 13.69 mA/cm2 for the GaInAs bottom cell. This leads to an excess current density of 10% for the bottom GaInAs cell. The junction thicknesses and bandgap energies may be controlled more precisely in the future to adjust for this remaining current mismatch, which leads to losses due to the series-connection of the subcells. Furthermore, the EQE of the GaAs second subcell can be increased if the tunnel diode layers above the cell are exchanged from the current p-Al0.3Ga0.7As and n-GaAs to a more transparent set of semiconductors.
The EQE of the GaInAsP junction shows a significant overlap with the GaAs subcell in the wavelength range between 700 and 900 nm. This confirms the high transparency of the bond interface, an important advantage of direct wafer bonding, which does not require additional materials for the bonding and therefore leads to zero absorption and low interface reflection due to the smooth transition between GaAs and InP with similar refractive indices. Also, the high EQE of the GaInAsP cell indicates excellent crystal quality, which is not degraded by the bonding.
The IV characteristics of the best GaInP/GaAs//GaInAsP/GaInAs concentrator solar cell have been determined as a function of concentration. Figure 4 shows the calibrated IV curve at one-sun AM1.5d ASTM G173-3 together with values for the short-circuit current density Jsc, the open-circuit voltage Voc, fill factor FF, and efficiency η. The fill factor increases under concentration (Figure 5) and peaks at 100 suns before series-resistance losses start to dominate at high current densities. The highest conversion efficiency is reached at 297 suns concentration with a value of 44.7 ± 3.1%. The efficiency remains high in a broad concentration range from 100 to 1000 suns with a measured value of 42.6 ± 3.0% at 962 suns. No indication of a breakdown of the tunnel diodes is observed at high concentration levels.
The voltages of the four subcells have been analyzed by EL spectroscopy. Figure 6 shows the calculated open-circuit voltages as a function of the current densities. The maximum efficiency is reached at a concentration of 297 suns, which is equivalent to a current density of 3694 mA/cm2. Under these conditions, the subcell voltages are determined to be 1.584 V (GaInP), 1.192 V (GaAs), 0.892 V (GaInAsP), and 0.500 V (GaInAs). The difference Eg-qVoc between the bandgap energy of the semiconductor and the product of elementary charge q times the open-circuit voltage of the device is a good indicator for material quality and the ideality of the diodes. The GaInP top cell shows the largest offset in voltage of 296 mV, followed by the GaInAs bottom cell with 240 mV and the two GaAs and GaInAsP middle cells with 228 mV. There is still potential to improve the quality of these top and bottom junctions in the future.
This paper presents results of a wafer bonded four-junction solar cells with a peak efficiency of 44.7% at 297 suns concentration. Wafer bonding allows the defect-free combination of two GaAs-based and InP-based tandem solar cell structures with a lattice mismatch of 3.7 %. More than 90% of the substrate area with a diameter of 100 mm was bonded successfully and showed no indication of interface defects such as voids. The peak external quantum efficiency was above 80% for all subcells. Notably, the high voltage and current of the GaInAsP junction directly below the bond interface confirms the excellent transparency of the bond and the achievement of low bulk defect densities. The solar cells can be operated up to high concentration levels of several thousand suns without any indication of tunnel diode breakdown, and the high performance of 42.6% efficiency at 962 suns is further indication for the low resistance of the bond interface. The solar cell structure still offers significant potential for improvements as some of the largest losses are due to an absorbing AlGaAs/GaAs tunnel diode between the GaInP and GaAs subcell, a current mismatch of 10% and a low voltage of the GaInP top cell.
The authors like to acknowledge device processing and characterization by M. Scheer, K. Wagner, A. Panamatschuk, E. Fehrenbacher, T. Gandy, K. Mayer, I. Semke, and E. Schäffer. We also like to thank Emmanuelle Lagoutte (CEA-LETI), Jocelyne Wasselin (Soitec), Chantal Arena (Soitec), and Fred Newman (Soitec) for very valuable discussions. Soitec and CEA-LETI would like to acknowledge support from the French Environment and Energy Management Agency (ADEME) through the “Investissement d'Avenir” program GUEPARD.