Micro-sized thin-film solar cells via area-selective electrochemical deposition for concentrator photovoltaics application

Micro-concentrator solar cells enable higher power conversion efficiencies and material savings when compared to large-area non-concentrated solar cells. In this study, we use materials-efficient area-selective electrodeposition of the metallic elements, coupled with selenium reactive annealing, to form Cu(In,Ga)Se2 semiconductor absorber layers in patterned microelectrode arrays. This process achieves significant material savings of the low-abundance elements. The resulting copper-poor micro-absorber layers’ composition and homogeneity depend on the deposition charge, where higher charge leads to greater inhomogeneity in the Cu/In ratio and to a patchy presence of a CuIn5Se8 OVC phase. Photovoltaic devices show open-circuit voltages of up to 525 mV under a concentration factor of 18 ×, which is larger than other reported Cu(In,Ga)Se2 micro-solar cells fabricated by materials-efficient methods. Furthermore, a single micro-solar cell device, measured under light concentration, displayed a power conversion efficiency of 5% under a concentration factor of 33 ×. These results show the potential of the presented method to assemble micro-concentrator photovoltaic devices, which operate at higher efficiencies while using light concentration.

PCE of 15% under 33 × CF 9 , a significant increase from 5% at 1 × CF illumination. The above approaches are an excellent proof-of-concept, but do not achieve materials savings since they all start from large continuous layers. Several bottom-up preparation strategies that reduce consumption of critical raw materials were achieved using laser nucleation and LIFT techniques, reaching PCEs of 3.4% at 20 suns 23,25-27 . Duchatelet et al. 24 reported the fabrication of CIGSe stripes with sub-millimetre width using area-selective electrodeposition (ASED). The deposits were restricted in size by virtue that the Mo back contact consisted of line arrays on a bare glass substrate rather than the normal continuous layers covering the glass. The authors achieved a PCE of 5.3% for the 1 cm long 105 µm wide lines, although they mention the CIGSe annealing process could still be improved since they recognize that temperature gradients caused by cell geometry and the unwanted reaction of selenium on the glass likely forming volatile sodium selenide species was detrimental for high quality absorber layers 28 . The line array design is an excellent way to achieve semi-transparent photovoltaics, but upscaling the electrodeposition to larger sized areas might prove difficult, since there will be considerable voltage drop along each line due to increased resistance of trying to pass current through a narrow line rather that a sheet. Sadewasser et al. 10 demonstrated an alternative ASED approach: CuInSe 2 (CISe) was electrodeposited from a  Table 1.
Here, we present amodified bottom-up ASED approach to forming micro-CIGSe absorber layers, where instead of depositing all the semiconductor elements at once, we deposit the metal elements in two steps, and introduce the chalcogen by reactive annealing. This approach of reactive annealing of electrodeposited metal stacks already led to a PCE of 17.3% for large area CIGSe 26 , which is significantly higher than the 9% reported for the single deposition bath approach 27 . We investigate if the stacked layer approach also translates into higher efficiencies in the micro format. Therefore, we electrodeposited a Cu layer from an aqueous solution, and then an (In + Ga) layer from an ionic liquid into a patterned insulating SiO 2 layer deposited on top of a Mo back contact. An array of recessed circular areas with a diameter of 200 µm exposed the Mo to the electrolyte during electrodeposition to enable the deposition of metal layer islands. The advantage of using the SiO 2 layer is that it protects the underlying Mo layer from being selenized during the annealing process hopefully avoiding increases in the solar cell series resistance due to the formation of MoSe 2 . We investigate two different absorber thicknesses, one around 0.5 µm to see if it is possible to reduce the quantity of CIGSe in the device to the bare minimum, and one of around 1.5 µm to act as a comparison to the normal absorber layer thicknesses of large area films. After reactive annealing, the resulting absorber layers were characterized to determine their composition, phase, morphology, and band gap. Finally, the current-voltage properties of two solar cells formed from the two absorber layers with different thicknesses are investigated over three orders of magnitude light concentration. The thicker device showed an open-circuit voltage of 525 mV under 18 × concentration, which is the highest reported value for any material efficient bottom-up deposition approach for CIGSe.

Methods
Fabrication of SiO 2 template patterns. Micro-patterned substrates were prepared according to the procedure described by Sadewasser et al. 10 In brief, 1 mm thick soda lime glass (SLG) is covered by a 500 nm Mo layer deposited by sputtering and after that, a 2 μm thick SiO 2 layer deposited by plasma enhanced chemical vapour deposition. Direct-write laser lithography is used to define the desired hole pattern. The SiO 2 is then etched down to the Mo layer by reactive ion etching and residuals are removed by plasma etching. An overview of the resulting pattern and a representation of its cross-section are depicted in Fig. 2.
Absorber preparation and characterization. Cu(In,Ga)Se 2 is grown by electrodeposition of the metals and subsequent reactive annealing in a Se-containing atmosphere 29 . The metal electrodeposition is split into two: first the deposition of Cu from an aqueous bath and second the co-deposition of In and Ga from a deep eutectic solvent. The Cu plating bath is prepared by dissolving CuSO 4 ·7H 2 O (99.99%, Alfa Aesar), NaOH (semiconductor grade, Sigma-Aldrich), sorbitol (99%, Sigma-Aldrich) in ultra-pure water, in order to obtain a dark blue solution with the concentrations of 3.0, 0.2, and 0.1 M, respectively. Prior to deposition, 70 µl of Empigen was added to the solution at room temperature. The substrates were placed onto a rotating disk electrode (RDE) as a working electrode with a Ag/AgCl electrode used as reference and a Pt wire used as counter electrode. The potentiostatic electrodeposition of Cu was carried out at room temperature (E dep = − 1.15 V vs Ag/AgCl), with a RDE speed of 200 RPM. The deposition is stopped when the desired cut-off charge density of is reached. The substrate is subsequently washed, dried, and stored under nitrogen until the next deposition step.
In and Ga are electrodeposited on top of the Cu layer inside a nitrogen-filled glove box. The deep eutectic solvent for the plating consists of urea (99%, Sigma-Aldrich), and choline chloride (> 98%, Sigma-Aldrich), mixed in a (1:2 molar ratio) at 60 °C. InCl 3 (anhydrous 99.999%, Alfa Aesar) and GaCl 3 (ultra-dry 99.999%, Alfa Aesar) are then dissolved in the deep eutectic solvent at the same temperature, in order to obtain concentrations of 50 mM and 25 mM, respectively. For this deposition, a Ag pseudo reference electrode is used, whilst the other electrodes remain the same as previously. The electrodeposition is carried out at 60 °C (E dep = − 1.2 V vs Ag), with a RDE speed of 300 RPM. Charge density cut-offs and targeted stoichiometry are described in Table 2 for the different thickness samples. The resulting metallic micro-deposits are then washed multiple times in air with absolute ethanol and deionized water. From here on, the two samples are labelled S500 and S1500 respectively, where the number represents the thickness of the absorber layer, as expected from the deposition charges (see Table 2).
The Cu-In-Ga micro-deposits, S500 and S1500, are converted into Cu(In 1−x Ga x )Se 2 via reactive annealing using a rapid thermal processor. In the presence of 80 mg of Se (99.999%, Alfa Aesar) they are first kept at Top-view scanning electron microscopy (SEM) micrographs were taken with a Hitachi SU-70 and energy dispersive X-ray spectroscopy (EDX) measurements were collected with an Oxford INCA X-MAX analyser coupled to the SEM. The lamellas used for qualitative EDX analysis were prepared using a dual beam Focus Ion Beam Scanning Electron Microscopy (FIB-SEM) equipped with a STEM detector at 30 kV. Photoluminesence spectra of the absorber layers were measured with a home-built setup using a 660 nm laser excitation, a grating monochromator and an InGaAs detector array. Raman spectroscopy was performed using a 633 nm laser excitation in combination with a 2,400 lines/mm grating with a Renishaw inVia micro-Raman spectrometer.
Device preparation and characterization. The prepared micro-absorbers were etched in a 5 wt% KCN aqueous solution for 30 s, after which CdS was deposited on top via chemical bath deposition 29 . Intrinsic ZnO and Al-doped ZnO were deposited via magnetron sputtering 30 and individual 200 µm diameter holes were electrically isolated by mechanical scribing.
Current-voltage (JV) curves were either measured under AM1.5 illumination using a AAA class solar simulator or at various light concentrations (spanning from 0.04 × to 78 × CF) under a 660 nm continuous wave laser. Shunt resistance (R shunt ) and series resistance (R series ) were extracted following the Hegedus-Shafarman analysis 31 . External quantum efficiency (EQE) measurements were performed in a home-built setup calibrated with silicon and InGaAs photodiodes. All measurements were recorded at room temperature.

Results
Absorber formation and morphology. Figure 3a shows a micrograph of the electrodeposited Cu in a 200 µm diameter hole in the SiO 2 layer. The Cu layer appears smooth and continuous throughout the hole, and the SiO 2 layer appears undamaged by the electrodeposition process. Figure 3b shows a micrograph of the subsequently electrodeposited In-Ga layer with an EDX elemental map over-lay on half the micrograph showing Si in red, Cu in green and In in blue. Here, 5-10 µm wide In islands are uniformly distributed on the surface of www.nature.com/scientificreports/ the Cu layer. Figure 1 of the supplementary information shows a cross-section of both S500 and S1500 before selenization. In both cases a continuous Cu-In-Ga layer is observed on top of the Mo, which stops at the edge of the SiO 2 template. Figure 3c shows the CIGSe micro-absorber layer S500 prepared from the Cu-In-Ga precursor after reactive annealing in a Se-rich atmosphere. This layer appears smooth and covers the whole surface except for a small region indicated by the red circle. Absorber S1500 appears rougher (Fig. 3d) than S500 but covers the entire surface of the hole. For both samples, the high magnification micrographs reveal pinholes in the absorber layer and apparent grain sizes on the order of 1-2 µm (Fig. 3e,f).
Absorber characterization. In the following section the absorber composition, the phases present, and the band gap are probed by EDX, Raman and photoluminescence spectroscopies. Raman spectroscopy is preferred over X-ray diffraction to confirm the presence of the chalcogenide Cu(In,Ga)Se 2 due to the more localized nature of the measurement technique.  Table 3. The underlying atomic percentage values are given in the supporting information (Table S1). Here, the analysis software assumes that the elements are distributed uniformly throughout the depth of analysis, although this is unlikely to be the case since Ga often segregates towards the back of the film, as is observed for CIGSe prepared from metal stacks 32 . S500 shows a CGI ratio of 0.93, similar to what was aimed at from the In-Ga electrodeposited onto Cu using 0.45 C cm −2 (EDX generated image shows Si in red, Cu in green and In in blue), (c,e) S500 after reactive annealing with Se, (d,f) S1500 after reactive annealing with Se. Table 3. CGI and GGI ratios obtained from EDX measurements. EDX was measured in a 100 µm square area inside the micro-absorber. *Determined GGI for S1500 was within the margin of error. . Surprisingly, for S1500, no measurable Ga content was found (the EDX detection limit is around 1%). This result indicates a dependence of the composition of the absorber on the electrodeposition conditions. Furthermore, it points to a change in deposition behaviour when using microelectrodes compared to (large) centimetre sized electrodes, most likely due to an added diffusion effect from the formation of hemispherical diffusion layers in said microelectrodes [33][34][35] .
In order to explore the lateral homogeneity of the absorber layers, the top-view EDX composition is mapped with 1 µm resolution and the resulting CGI ratios are shown as contour plots (Fig. 4). S500 has a relatively uniform composition with no obvious chemical gradients or large-scale inhomogeneity. The film is mostly in the stoichiometric range with some randomly scattered areas of a few microns in size with CGI values between 0.6 and 0.8, and a handful of micron-sized spots that are copper rich with CGI values around 1.5. For S1500, the absorber layer is macroscopically inhomogeneous with large areas that are extremely Cu poor with CGI ratios from 0.3 to 0.7, and other large areas near stoichiometry. Like S500, there are also a few Cu-rich spots. We hypothesize that the lateral variations in CGI observed for S1500 are because In islands nucleated on the Cu layer with lower density during the In/Ga electrodeposition process, leading to an over deposition of In in certain areas compared to other areas. This might also explain the higher roughness observed for S1500 (Fig. 3d) since taller In islands would lead to a thicker absorber layer growing out of the island during reactive annealing. Figure 5 displays the Raman spectroscopy analysis of the prepared absorbers. Cu(In 1−x Ga x )Se 2 shows a dominant peak at 174 cm −1 for CuInSe 2 and at 184 cm −1 for CuGaSe 2 , corresponding to the main vibration mode A 1 (vibration of Se while Cu, In/Ga remain at rest) 36 . Point measurements are taken across a line over the surface of the absorber samples with a 20 µm spacing ( Supplementary Information Fig. S3) and the resulting spectra are plotted against the measurement displacement in contour plots (Fig. 5a,c). The average spectrum is plotted in Fig. 5b,d. For S500, an intense peak at 174 cm −1 is observed for all measurements, as indicated by the red vertical line in the contour plot. For S1500 the previously described mode is observed, with an additional mode at 151 cm −1 observed in some spots, attributed to the In rich ordered vacancy compound (OVC) CuIn 5 Se 8 37,38 . The occurrence of the OVC in some spots, as shown by the contour plot, agrees well with the very low Cu content on several spots measured by spatially resolved EDX (Fig. 4). Low intensity modes in the range of 211-228 cm −1 are also observed for both samples and are attributed to B 2 -E modes of CIGSe 36 . These modes are more sensitive to the GGI ratio than the A 1 mode. A shift of the B 2 -E modes of S500 to higher wavenumbers (213-228 cm −1 ) when compared to S1500 (211-226 cm −1 ) indicates the presence of Ga in S500, as predicted by the GGI determined from EDX measurements 39 .
Photoluminescence spectra measured with a 660 nm laser excitation for both samples are presented in Fig. 6. Cu(In 1−x Ga x )Se 2 has a band gap around 1.0 eV for CuInSe 2 and 1.7 eV for CuGaSe 2 , with band gap values varying almost linearly with composition 17 . For S500, a peak at 1.02 eV is observed, corresponding to the band-to-band recombination at room temperature, which gives an approximation of the band gap (E g ). A band gap value of 1.02 eV for a slightly Cu poor absorber layer with CGI of 0.93 likely indicates the presence of a small amount of Ga, in agreement with the EDX measurements, since a slightly Cu-poor absorber without Ga has an expected band gap of around 0.99 eV 40 . However, a broad sub band gap peak around 0.8 eV is also observed, indicating the presence of a deep level defect. Previously, the presence of such a peak has been observed in Cu rich absorber layers 41 , although this may also indicate impurities from the electrodeposition process. For S1500 a single peak centred at 0.99 eV is observed, corresponding to the main band-to-band transition. This lower band gap corresponds well to a Cu-poor sample without any Ga 40 , in agreement with the EDX measurements. To summarize the PL measurements, S1500 appears to be purely CuInSe 2 whilst S500 likely contains a small amount of Ga and shows the presence of a deep defect.  Figure 7 displays FIB lamella micrographs and qualitative elemental mapping profiles by EDX. The stacked architecture of the devices Mo/ Absorber/CdS/ZnO can be observed in both samples. For S500 (Fig. 7a) the thickness of the absorber layer fluctuates between 100 and 900 nm. S1500 (Fig. 7b) is thicker in the range 2 μm, as expected from the higher   www.nature.com/scientificreports/ cut-off charges used during deposition. Due to the destructive lamella preparation process, some Cu bright inclusions and voids can be observed in the absorber layers. Additionally, the Cu signal is present in all layers for both samples, due to contamination caused by the copper grid used as support for the sample. The deposited CdS and ZnO window layers are conformal with estimated thicknesses of 50 nm and 500 nm, respectively. Some Cd appears to have inter-diffused into the absorber layer during cell processing which has previously been observed 42 .
For the device prepared with S500 (Fig. 7a), the In map shows an apparent homogeneous distribution in the absorber layer. The Se map shows its presence not only in the absorber layer but clearly extending into the Mo bottom layer, indicating the formation of an interfacial MoSe 2 layer during the reactive annealing step, despite the relatively low annealing temperature of 450 °C. Ga is weakly detected in the absorber layer, again in agreement with the measured top view EDX ( Table 1). The qualitative cross-sectional localized measurement appears to show Ga spread equally through the depth of the absorber layer. However, a Ga-ion beam is used for the milling process that creates the lamellas observed, causing contamination of this element in other layers and hinders an accurate estimation of the distribution of this element over the depth of the absorber. Its contaminating presence can also be observed in the ZnO layer. For the device prepared with S1500 (Fig. 7b), a similar description of the elemental profiles is observed. The main difference lies in the lack of measurable Ga signal in the area of the absorber layer, as expected from the GGI measured from the top view EDX. The Se signal seems to extend both into the Mo bottom layer and partially to the top CdS layer, indicating a diffusion of this element into both layers. To summarize the cross-sectional measurements, both samples show the presence of a MoSe 2 layer, with S500 showing a small amount of Ga and being significantly thinner than S1500.

Device optoelectronic characterization. To initially assess the micro-solar cells, current-voltage (JV)
curves and EQE spectra were measured (Fig. 8) and a summary of the solar cell parameters extracted by the Hegedus-Shafarman method 31 is given in Table 4. Devices made from the absorbers S500 and S1500 have power conversion efficiencies of 1.4 and 3.7%, and open-circuit voltages of 144 and 331 mV, respectively, at 1 sun (AM1.5) illumination. The V oc are low compared to state-of-the-art low bandgap CIGSe devices which achieve 577 mV 43 . There are several reasons for this. For S500 the low V oc could be attributed to back surface recombination since there is a high recombination velocity at the CIGSe/Mo interface and electron hole pairs would be generated near this interface. Indeed, the thicker S1500, where electron hole pairs would be generated far from the back interface, does have a higher V oc than S500. However, a normal large area CIGSe solar cell made using similar conditions as S1500 had a V oc of 427 mV. The key problem appears to be that the micro-solar cells suffer from low shunt resistances, R shunt , of 24 and 97 Ω cm 2 for S500 and S1500, respectively. For efficient devices, R shunt is typically one order of magnitude larger than measured here. A low R shunt is caused by shorting between the back contact and the front contact which may be due to pinholes or conductive secondary phases in the absorber layer itself, a too thin absorber layer, or delamination of the SiO 2 interface at the CIGSe interface. As shown in SI Fig. 2 we discount delamination at the SiO 2 interface as the cause of a low R shunt , since the cross section of both samples clearly shows the interface to be intact, and the absorber layer clearly separates the conducting layers. Most likely, the low R shunt is caused by the presence of pinholes (Fig. 3e,f), and the presence of Cu-rich areas www.nature.com/scientificreports/ which are known to be conductive (Fig. 4). These pinholes and secondary phases were not observed in previous work using the same electrodeposition methodology on large area substrates 29 . Nonetheless, the V oc for both devices reported here is significantly higher than the previously reported 94 mV for a CISe device using a similar SiO 2 template 10 . Indeed as hypothesized in the introduction, as for large area solar cells, the electrodeposition route of stacked metal precursor layers leads to better opto-electronic properties compared to the deposition of all elements simultaneously. Surprisingly, the short-circuit current (J sc ) for both devices is quite similar, despite the differences in their thickness, which suggests current collection is not improved by increasing thickness of the absorber layer.
To investigate current collection efficiency, the relative external quantum efficiency (EQE) of the devices was measured (Fig. 8b). EQE spectra are discussed relatively and presented with arbitrary units normalized to the respective maximum because the probe beam size was on the order of the size of the device, so we cannot confidently assume that all photons hit the absorber area. For both spectra, a sudden break at 800 nm can be observed due to the switching of a light source in the setup which caused a slight deviation of the beam position. The spectra show some parasitic absorption from the CdS window layer between 400 and 580 nm and reach their maximum at around 600 nm. For higher wavelengths, a continuous drop in the EQE response is observed. The poor carrier collection at long wavelengths is indicative of a short minority carrier transport length. Additionally, long wavelength losses for S500 could be attributed to an incomplete absorption since its thickness is well below a micron in some parts (Fig. 7a). The band gap values are estimated from the maximum of the first derivative of the EQE spectra at long wavelengths. The obtained values of 1.02 and 0.98 eV for S500 and S1500, respectively, fit very well with the peak maxima in the PL spectra (Fig. 6).

Device light concentration series.
To study the photovoltaic performance under concentrated light, the devices were illuminated with a 660 nm laser over three orders of magnitude of intensity. According to Eq. (2), there should be a logarithmic increase of the V oc with illumination intensity, leading to an increase in efficiency. A concentration factor (CF) representing light intensity was defined as a ratio between the J sc obtained in the power series and the J sc obtained in the solar simulator (calibrated for 1 sun) 44 . The solar cell parameters η, V oc , FF, R shunt , R series , and their dependence on illumination intensity are plotted (Fig. 9). For comparison, the data points from the AM1.5 solar simulator JV curves are added.
As expected from Eq. (2), starting from below 0.1 × CF an initial increase in efficiency for both devices is observed with increasing illumination intensity, mainly due to the increase in V oc and in FF (Fig. 9a-c). Maximum PCE of 4.7% at 33 × CF for S500 and 5.2% at 3 × CF for S1500 are reached. This means there is an increase in efficiency for both devices when comparing with the values obtained with AM1.5 in the solar simulator of 1.4 and 3.5%, respectively. For higher illumination intensities, a decrease in efficiency is observed that can be explained  www.nature.com/scientificreports/ by a combination of the decrease of both FF (Fig. 9c) and R shunt (Fig. 9d). For the intensity range explored, V oc continuously increases, with a slightly reduced slope at higher illumination intensities, similar to the observations by Paire et al. 19 for co-evaporated CIGSe micro-cells. On the other hand, Heidmann et al. 45 and Ringleb et al. 46 reported a decrease of V oc at around 30-40 suns for CISe micro-solar cells prepared via bottom-up laser nucleation and LIFT techniques. In our case, an average rate of increase of around 127 and 150 mV dec −1 for S500 and S1500, respectively, is determined. This is nearly twice as high as the 82 mV dec −1 predicted in Fig. 1 for a record CIGSe solar cell with a diode factor (A) of 1.38. Our results indicate A = 2.2 (S500) and A = 2.5 (S1500), which is too high for efficient solar cells. Additionally, the measured V oc under AM1.5 illumination is lower than the one measured with 660 nm light which will be discussed in more detail below. Overall, a maximum V oc of 525 mV for S1500 was achieved under a concentration factor of 18 × , which is the highest reported value for micro-cells prepared via bottom-up approaches and under light concentration like the one presented in this work. The R shunt obtained for the devices measured under the 660 nm laser and the AM1.5 illumination show different values. This means there is an apparent dependence of R shunt on the illumination intensity and wavelength. For S500, a slightly larger R shunt value of 24 Ω cm 2 is determined from the AM1.5 curve when compared to the closest value in the power series (14 Ω cm 2 ) and a higher FF value is observed. For S1500 however, a lower R shunt value for the AM1.5 curve is found with a corresponding lower FF value. This implies a strong dependence of the FF on the R shunt which is expected and well described by Green 47 .
Interestingly, for the device prepared with absorber S500, which showed lower estimated thickness and more evidence of pinholes, R shunt decreases by one order of magnitude over the three orders of magnitude of illumination probed compared to three orders of magnitude change for S1500. This means that the device prepared with absorber S1500 reaches a peak efficiency at a lower concentration factor than S500, due to its steeper decrease of R shunt with illumination intensity (Fig. 9d). On the other hand, any possible temperature increase from the increasing illumination intensity does not appear to have a strong dependence on R series , as it is stable over the range studied (Fig. 9d), as predicted by Sadewasser et al. 11 .
In order to better understand the observed impact of illumination wavelength on the solar cell parameters, JV curves were taken for both devices in the solar simulator with the inclusion of a 610 nm long pass optical filter R shunt S500 power series R series S500 AM1.5 R series S500 power series R shunt S1500 AM1.5 R shunt S1500 power series R series S1500 AM1.5 R series S1500 power series R (Ω cm www.nature.com/scientificreports/ between the light source and the cells (Fig. 10). This filter only allows transmission of wavelengths longer than the cut-on wavelength of 610 nm, essentially illuminating the micro-concentrator devices with red light (like the 660 nm laser used for the power series). The long pass filter reduces the J sc of both devices as expected since the blue part of the spectrum is cut out. For both devices, an increase in V oc is observed, with S500 gaining about 30 mV, similar in magnitude to the gain of about 60 mV observed for the 660 nm laser illumination. Also, R shunt increases with the introduction of the long pass filter for both devices as also observed during laser illumination. This can be explained by the photoconductivity of CdS which is known to increase under blue light illumination for some devices 48,49 . S500 has a high number of pinholes (Fig. 3c,e) which were filled with CdS during device processing. If now the conductivity of CdS is increased, those filled holes create a direct shunt path between front and back contact. Thus, explaining the higher V oc under red light illumination.

Conclusion
An overall analysis of the process of area-selective electrodeposition of a CIGSe absorber layer into micropatterned substrates, the assembly of micro-concentrator photovoltaic devices and the characterization of the absorber and device are presented in this work. Metallic stacks consisting of copper, indium and gallium were electrodeposited into microelectrodes of 200 µm diameter. Evidence of a change of deposition behaviour between microelectrodes and large area electrodes was observed, as controlling the local and global composition proved to be difficult. Ongoing experimental work focusses on this as the opto-electronic properties of CIGSe crucially depend on the composition. The metallic stacks were annealed at a relatively low temperature of 450 °C to form mostly the chalcopyrite phase, as well as the OVC compound CuIn 5 Se 8 , for Cu-poor compositions. We showed that a stacked metal precursor approach gave higher open circuit voltages compared to the "all elements being deposited at once" approach previously used for these SiO 2 templates, which is the same result as found for large area electrodeposition. Under one-sun illumination the thinnest absorber layer device collected a similar amount of current as compared to the thickest absorber layer but had a significantly lower open circuit voltage presumably caused by a shunt resistance four times lower than the thickest absorber layer device. Under increased illumination intensity, illumination dependent shunt resistance is identified as the main problem that is limiting device efficiency increase. To improve the performance of these devices, their low shunt resistance and the light dependent shunt resistance must be ameliorated, by making the absorber layers pinhole free, and more tightly control the composition and homogeneity of the absorber layers. Nonetheless, these devices achieve the highest V oc , under light concentration conditions, of a bottom-up micro-CIGSe device published in the literature. Individual micro-concentrator devices showed an increase from 2.2% at 1 × CF to 4.8% at 33 × CF and also from 4.8% at 1 × CF to 5.2% at 3 × CF under 660 nm illumination, clearly demonstrating the potential of area-selective electrodeposition as a material saving technique for the preparation of micro-concentrator solar cells.