Frederik C.
Krebs
*,
Thomas
Tromholt
and
Mikkel
Jørgensen
Risø National Laboratory for Sustainable Energy, Technical University of Denmark, Frederiksborgvej 399, DK-4000 Roskilde, Denmark. E-mail: frkr@risoe.dtu.dk
First published on 4th May 2010
Upscaling of the manufacture of polymer solar cells is detailed with emphasis on cost analysis and practical approach. The device modules were prepared using both slot-die coating and screen printing the active layers in the form of stripes that were serially connected. The stripe width was varied and the resultant performance analysed. Wider stripes give access to higher geometric fill factors and lower aperture loss while they also present larger sheet resistive losses. An optimum was found through preparation of serially connected stripes having widths of 9, 13 and 18 mm with nominal geometric fill factors (excluding bus bars) of 50, 67 and 75% respectively. In addition modules with lengths of 6, 10, 20, 22.5 and 25 cm were explored. The devices were prepared by full roll-to-roll solution processing in a web width of 305 mm and roll lengths of up to 200 m. The devices were encapsulated with a barrier material in a full roll-to-roll process using standard adhesives giving the devices excellent stability during storage and operation. The total area of processed polymer solar cell was around 60 m2 per run. The solar cells were characterised using a roll-to-roll system comprising a solar simulator and an IV-curve tracer. After characterisation the solar cell modules were cut into sheets using a sheeting machine and contacted using button contacts applied by crimping. Based on this a detailed cost analysis was made showing that it is possible to prepare complete and contacted polymer solar cell modules on this scale at an area cost of 89 € m−2 and an electricity cost of 8.1 € Wp−1. The cost analysis was separated into the manufacturing cost, materials cost and also the capital investment required for setting up a complete production plant on this scale. Even though the cost in € Wp−1 is comparable to the cost for electricity using existing technologies the levelized cost of electricity (LCOE) is expected to be significantly higher than the existing technologies due to the inferior operational lifetime. The presented devices are thus competitive for consumer electronics but ill-suited for on-grid electricity production in their current form.
Frederik C. Krebs | Frederik Christian Krebs received his PhD from the Technical University of Denmark in year 2000 and has since then worked in the field of polymer solar cells at Risø National Laboratory for Sustainable Energy. He is currently professor at Risø DTU within areas of research that include new materials with low bandgaps and novel processing capability, large-area processing and manufacture of polymer solar cells, stability and lifetime testing, degradation mechanism studies, outside testing and demonstration. |
Thomas Tromholt | Thomas Tromholt received his MSc in Nanotechnology in 2009 from Aalborg University, Denmark. He is currently pursuing a PhD with a focus on material characterization of organic photovoltaics at Risø National Laboratory for Sustainable Energy. |
Mikkel Jørgensen | Mikkel Jørgensen received his MSc in Chemistry from the University of Copenhagen and his PhD in Organic Chemistry from the University of Copenhagen (1990). He worked as an industrial chemist at NycoMed (1987–1990) during his PhD studies and later as an industrial Chemist at PNA Diagnostics (1990–1993). He then became employed as a senior Scientist at Risø National Laboratory, DTU, Denmark (1994–present). His scientific interests include synthetic chemistry, NMR, chemistry of materials, carrier mobilities in organic materials, energy levels and energy level alignment in organic materials by UPS studies, solar cells, polymers, fluorine chemistry and supramolecular chemistry. |
The purpose of this work was to upscale the manufacture of polymer solar cells from the laboratory level to an industrial level and identify the instruments requirements for complete polymer solar cell module manufacture and describe a complete process enabling manufacture of electrically contacted and characterised polymer solar cell modules that are ready to be implemented in electronic applications. A complete cost analysis detailing processing speed, manual labour and materials cost was also sought so that it is possible to estimate the manufacturing and electricity cost correctly. Finally the capital investment in machinery for the process is detailed.
Silver Ink | Max PCE (%) | Mean PCE (%) | Mean Isc/mA | Mean Voc/V | Mean FF (%) |
---|---|---|---|---|---|
Toyo UV | 2.20 | 1.95 (±0.19) | 61.8 (±2.4) | 7.75 (±0.30) | 39.0 (±2.0) |
Toyo Solvent | 0.77 | 0.65 (±0.08) | 40.7 (±5.8) | 5.38 (±0.46) | 28.6 (±1.5) |
Spraylat XCMS-016 | 1.00 | 0.86 (±0.09) | 44.8 (±2.5) | 6.12 (±2.5) | 30.1 (±1.1) |
Spraylat XCMD-022 | 0.83 | 1.05 (±0.07) | 44.8 (±2.4) | 6.87 (±0.39) | 32.7 (±1.2) |
Spraylat XCMS-031 | 1.17 | 1.11 (±0.10) | 53.7 (±2.7) | 6.23 (±0.21) | 31.9 (±1.0) |
Dupont 410 | 1.73 | 1.64 (±0.05) | 54.7 (±1.2) | 7.78 (±0.07) | 37.1 (±0.8) |
Spraylat XPVS-145 | 0.70 | 0.60 (±0.07) | 41.0 (±4.5) | 4.77 (±0.35) | 29.8 (±1.2) |
Dupont 5000 | 1.01 | 0.96 (±0.03) | 40.6 (±3.9) | 7.18 (±0.4) | 31.8 (±0.96) |
Dupont 5007E | 1.29 | 1.13 (±0.10) | 52.2 (±3.5) | 6.53 (±0.25) | 31.9 (±1.32) |
Dupont 5025 | 1.59 | 0.93 (±0.44) | 45.6 (±8.5) | 5.96 (±1.77) | 31.7 (±3.0) |
Fig. 1 The substrate (PET) is shown along with the position and order of the layers. ITO (1), ZnO (2), Active (3), PEDOT (4) and silver (5). The view is along the striped pattern (above). The finally connected module is shown schematically (below) as three serially connected stripes where the active areas and passive areas are highlighted. |
In principle the striped pattern could be endless but we chose to pattern the ITO in stripes that extend for the length of the typical module as shown in Fig. 2. Register and cutting marks were printed along the web along with lock holes that enable registration and juxtaposition of subsequent layers with respect to the first ITO layer during later printing and for the final cutting of the modules. The lengths of the stripes were 200, 225 and 250 mm (shown for 225 mm in Fig. 2).
Fig. 2 The substrate (PET) is shown with the etched ITO pattern and register marks. The 16 ITO stripes each measure 225 × 13 mm and are spaced along the web direction by 25 mm (top). The silver print for three of the different module lengths are shown overlaid on the ITO pattern (middle). The 225 mm long modules (middle left), 100 mm long modules (middle center) and 60 mm long module (middle right) are shown. The printed silver pattern is shown for the modules with 9, 13 and 18 mm wide ITO (bottom). |
In the case of the modules that were shorter than the ITO length in the direction of the web those were printed so that they overlapped with the ITO but maintained a repetition gap of 25 mm in order to enable cutting.
The drying temperature in all experiments was 140 °C. The ZnO solution was slot-die coated at a speed of 2 m min−1 with a wet layer thickness of 3.125 micron and a dry layer thickness of 23 nm. The active layer was coated at a speed of 1.4 m min−1 with a wet layer thickness of 4.85 micron and a dry layer thickness of 127 nm and the PEDOT:PSS was coated at a speed of 0.3 m min−1 with a wet layer thickness of 75 μm and a dry layer thickness of 20 μm (Fig. 3). The slow web speed during coating PEDOT:PSS is due to the short available drying length of 1 m. To ensure proper drying of the thick PEDOT:PSS layer only slow speeds could be employed.
Fig. 3 The slot-die coating of ZnO nanoparticles (left), P3HT:PCBM (middle) and PEDOT:PSS (right). The wet films are shown above and the corresponding dried films are shown below. |
Fig. 4 Screen printing of silver using a Klemm line as seen from the un-winder (top left) and the same device type (12 × 18 mm stripes) after having printed the silver (top right). Flat bed screen printing using a flat bed screen printer from Alraun (bottom left) showing 16 × 13 mm stripes on the re-winder (bottom right). |
Fig. 5 Laminator comprising un-winder, edge guide and cutting table, laminator, laminate un-winder, longitudinal cutting knifes and re-winder (left). A photograph of the edge guide system (right). |
After application of the adhesive onto the barrier foil it was laminated onto the un-encapsulated solar cell material. Typically the lamination was first carried out on the back side of the device to avoid excess handling of the sensitive active layers in the device. In the final step barrier material was laminated onto the front side of the device.
Once the module to be tested has been positioned vacuum is applied to the vacuum table to fix the module over the contacts. Contact is made by pneumatic cylinders that force contact between a conducting strip on the vacuum table and the device. It is possible to employ both top and bottom contacting schemes in the system. The IV measurements are then carried out using a Keithley 2400 sourcemeter. The computer program allows for tracing multiple curves, annealing the device and a set of criteria can be set determining when the next module should be tested. For each device a report is generated in Excel format including the data which may include the latest IV curve, a photograph of the device and annealing behaviour of Isc, Voc, FF, PCE, Rs and Rsh as a function of time. In addition, a summary report for the entire roll is generated. This enables the quick identification of devices on the roll that behaves abnormally. A schematic drawing and a photograph of the roll-to-roll IV tester is shown in Fig. 6. The lamp source is a Steuernagel KHS1200 providing approximately AM 1.5 G and 1000 W m−2 at the position of the solar cell module. The temperature of the devices during testing is 72 °C. The typical scheme for characterisation when aiming at carefully establishing that the device exhibits the desired operational stability was to record IV curves continuously from −20 to +20 V for a set period of time (2–20 min) and continue to the testing of the next module unless the change in performance (dPCE/dt) was larger than 0.01%. After completion of the testing the data was carefully analysed (vide supra).
Fig. 6 The roll-to-roll IV tester is shown schematically (left). The setup consists of a roll-to-roll system (Alraun) that can be controlled via a PC through a custom made hardware control (R2R Control). Solar cell modules are automatically positioned under an AM 1.5 sun simulator (Steurnagel LichtTechnik). Pneumatic actuators (not shown) bring the solar cell in contact with leads running to an SMU that performs IV scans. In annealing mode IV scans are repeated until a stable PCE is reached and the software then instructs the R2R control to advance the roll to the next module. A photograph along the roll of solar cells is shown from the un-winding side (right). The metering wheel and positioning camera can be seen in the upper part of the image and the pneumatic contacting cylinders can be seen under the solar simulator in the middle of the image. |
Fig. 7 The simple sheeting machine employed in this study comprising un-winder, transport roller camera, knife cutter and conveyor belt (left). A photograph of the Prym button contact machine (right). |
Fig. 8 The power that can be extracted from a module aperture as a function of stripe width and the nominal power conversion efficiency obtained for the modules with 9, 13 and 18 mm wide stripes. All devices were prepared using the same silver ink (PV410). |
Ideally the stripe should be as narrow as possible to minimize the Ohmic loss and as wide as possible to maximize the active area. Since the serial connection of stripes takes up some of the active area (Fig. 1) the increased performance due to narrowing of the stripe is quickly lost due to the inactive area from the serial interconnections. In Fig. 1 the gaps between the ITO stripes are shown along with the positioning of the different layers to avoid short circuit and to achieve interconnection. Ideally the gap between the ITO stripes which is necessary to electrically isolate the individual stripes should be as small as possible. Also the gap between the slot-die coated stripes should be as small as possible. In practical terms it is very difficult to slot-die coat a narrower gap than 1 mm at low speeds (<60 m min−1) due to both the mask and the coating behaviour of liquids and the aperture loss from a gap larger than 3 mm are too excessive. During this work gaps of 1, 2 and 3 mm have been employed and 2 mm is the gap width that gives the largest degree of control. It is however highly feasible to employ 1 mm gap but it challenges the skill of the experimenter doing the coating. The optimum device geometry thus includes the physical properties of the materials (sheet resistivity, conductivity, optical transmission), the tolerances that can be handled (masks, printing resolution) and the skill of the experimenter (technical yield). From the experiments carried out during these trials a stripe width of 13 mm and a gap of 2 mm was the best compromise between active area, technical yield and device performance as shown in Fig. 8. Further work thus concentrated on 13 mm wide ITO stripes with a 2 mm gap between the stripes. The results have been detailed in the following.
The role of defects and coating errors cannot be ruled out and thus the longer the stripe the larger the chance of defects. It was thus expected that all modules should exhibit roughly the same average performance but that the spread in performance would be larger for the smaller modules. An additional complication for the 60 and 100 mm modules are that they share the same ITO and PEDOT:PSS stripe and therefore the active area had to be carefully masked to avoid light harvesting from areas outside the nominal module aperture. The data has been summarized in Fig. 9 and Table 2 where it is evident that the larger modules have a lower average PCE but a large spread due to a large spread in the short circuit current. The smaller modules have a smaller spread.
Fig. 9 I sc versus PCE for the three different module lengths employed with 13 mm wide ITO stripes and a 2 mm gap. |
Module Size | Max PCE (%) | Mean PCE (%) | Mean Isc/mA | Mean Voc/V | Mean FF (%) |
---|---|---|---|---|---|
Large (360 cm2) | 1.69 | 1.18 (±0.13) | 185.8 (±13.0) | 7.83 (±0.21) | 36.7 (±1.7) |
Medium (160 cm2) | 1.36 | 1.22 (±0.11) | 81.0 (±5.8) | 6.81 (±0.17) | 34.8 (±1.0) |
Small (96 cm2) | 2.00 | 1.79 (±0.09) | 60.1 (±5.1) | 7.56 (±0.12) | 37.9 (±2.1) |
Fig. 10 AFM images of the active layer surface post production and after delamination at the active layer–PEDOT:PSS interface. The images are 100 × 100 μm (insets are 5 × 5 μm). The surface of the active layer beneath printed silver was visibly different and with a roughness of 40 nm (left). The surface underneath PEDOT:PSS without the print was smoother with a roughness of 25 nm (middle). The devices that were pre-washed with isopropanol during coating of PEDOT:PSS were much smoother underneath the printed silver with a roughness of 20 nm (right). |
It was found that the devices that were pre-washed with isopropanol during coating of the PEDOT:PSS gave rise to a much smother film at the active layer-PEDOT:PSS interface after printing silver on top. We thus ascribe the roughness to be induced by the solvents in the silver ink and to be responsible for the short circuits. The isopropanol pre-washing was found to alter the surface morphology and to enable printing of the silver without inducing short circuits. In Fig. 11 and Table 1 we show results from printing using different commercially available screen printable silver inks. It should be emphasized that all the inks are highly performing and all give rise to the specified performance in terms of conductivity, drying, printability, ITO contacting etc. The variation in performance has to be viewed in the context of this process that specifically requires printing on top of a PET/ITO/ZnO/P3HT:PCBM/PEDOT:PSS composite film.
Fig. 11 I sc versus PCE for ten different commercial silver inks tested on 96 cm2 area modules. |
The main reasons for employing the silver back electrode in the form of a silver grid was to allow for semi-transparency, reduction of cost, enabling faster drying and to minimize the influence of diffusion of solvents from the silver ink into the active layers. It should be emphasized that the use of a silver grid and consequent achievement of semi-transparency is possible due to the reasonable conductivity of the PEDOT:PSS layer and it would not be possible to employ a silver grid in case of hole-transporting oxides (i.e. V2O5, WO3, MoO3) back electrodes where a fully covering back electrode is required.
The silver inks that generally performed well were those with a high silver content (80–86% wt/wt) and those that did not perform so well generally had a lower silver content and consequently more solvent. Also the high-silver-content inks generally had a higher viscosity resulting in slower diffusion of solvents out of the ink. Generally the time from printing to drying was kept as short as possible (<1 min) to avoid excessive diffusion of solvents into the active layer. One of the tested inks employed UV curing and was as such solvent-free. This ink performed exceptionally well and we ascribe this advantage to the fact that the silver is mixed with a UV-cross linkable prepolymer. A secondary advantage of this ink was that it is not soluble in solvents after curing in contrast to all the other inks.
The front side of the solar cell was fully laminated with barrier foil and the backside (the printed side) was laminated with material having a width such that some of the printed silver electrodes were exposed. This enabled roll-to-roll characterisation of the solar cells as shown in Fig. 6. This proved to be an excellent tool in several aspects. When roll coating it is very easy to adjust or deliberately vary a coating parameter during coating and since each module on the roll has a specific label it is easy to probe the effect of the change or alteration when finally characterising the whole roll. Typical variables are coating speed, wet thickness, drying temperature etc. This proved to be a powerful tool in the optimisation scheme. An example has been outlined in Fig. 12 where the PCE for the modules has been plotted as a function of the number on the roll. It is the 60 mm modules spaced by 25 mm. A variation in the wet thickness of the active layer is clearly reflected as a step in performance. The roll-to-roll characterisation can thus be used as a quality control tool and in the context of research and development for process optimisation. The roll-to-roll characterisation of the solar cells could also be used for annealing the modules while characterising such that a consistent comparison between the modules could be made. The devices printed with UV-curable silver ink were very close to fully operational “as prepared” due to UV-activation of the ZnO during UV-curing whereas the all the modules prepared using thermally dried silver inks required light soaking for some time before the optimum performance was reached. To further explain this, some experiments were carried out to highlight the dependence of UV light and oxygen on the initial device performance. The encapsulation of the devices employs a multilayer barrier film with a UV filter having a cut-off at around 380 nm. Without the UV filter the devices work immediately but will operate for a significantly shorter time due to the UV-induced degradation. In the absence of UV the devices initially work poorly as illustrated in Fig. 13 where continuous IV scans are shown as a function of time. The devices however gradually improve over the course of 15–30 min.
Fig. 12 A plot of the module PCE along a roll of 60 mm modules. The plot shows the spread in performance that amounts to around ±10% around the average value. Upon changing the wet thickness of the active layer a step change in performance is observed. |
Fig. 13 Evolution of the IV curves of a freshly prepared device after encapsulation. The illumination was continuous (1000 W m−2, AM 1.5 G, 72 ± 2 °C). The device improves over the course of 15 min from 0.5% to 1.06% power conversion efficiency. |
In order to explain the behaviour we performed experiments with and without UV light and found that the interaction with oxygen corroborates well with the reported photoconductivity of ZnO.47 We propose that the interactions shown in Fig. 14 are in place which is in agreement with earlier observation made on memory devices based on ZnO.47 The initially prepared ZnO device thus has oxygen adsorbed on the surface of the ZnO nanoparticles that act as traps for electrons leading to poor transport.
Fig. 14 Proposed mechanism and interaction of ZnO nanoparticles with oxygen and UV light. The charges denoted as ± are bound charges whereas the encircled charges are mobile. The initially doped state can become de-doped by application of a voltage or by application of UV light. The oxygen present will gradually re-absorb on the surface and re-dope the ZnO nanoparticles. |
In the presence of UV light the direct bandgap excitation of the ZnO will lead to free carriers and immediate conductivity. The insertion of a UV filter at this point will lead to poor conductivity again. During longer UV irradiation the oxygen is gradually de-doped by some of the holes generated in the ZnO and the conductivity becomes permanent. The desorption of oxygen by direct excitation using UV light is thus a kinetically slow process. This observation was found to have significant impact on the speed at which the completed modules can be annealed and characterised (vide supra).
In terms of stability the devices perform quite well both during storage and operation which has enabled round robin studies between 18 different laboratories,25 sharing of cells between laboratories but also demonstration projects,19 prototyping and product simulation (vide supra). The operational stability of 16 stripe 225 mm devices and 8 stripe 250 mm modules are shown in Fig. 15 under continuous illumination at two different incident light intensities and temperatures. In all cases the performance initially drops but in the cases of lower light intensity and lower temperature the performance increase over 200 h followed by a relatively steady level for 300 h with a slow decline. In the case of the higher intensity and higher temperature there is a short period with a slight increase or a steady performance followed by an exponential decay over the remaining 450 h. In cases where a particular device exhibits degradation in performance it is normal to quote some measure of the lifetime or stability. OPVs are known to exhibit many different degradation curves. When plotting the performance as a function of time under continuous illumination the most typical examples of degradation behaviour include:
1) An initial increase or decrease in performance followed by a decay that may be linear or exponential
2) Linear or exponential decay from the very beginning
3) A fast initial decay followed by stabilisation and a much slower decay
4) Complex initial behaviour followed by any of the above
5) Any of the above with a sudden catastrophic failure
The time from the start of the measurement to the point at which the device yields 80% of the initial performance (often termed T80) or the time to the point at which the device yields half the initial performance (often termed half life or T50) are measurable parameters that in principle can be used to compare device stability and operational lifetime. It is however debatable where the starting point should be taken and the most rational choice is possibly that T80 (or T50) is quoted as the time from the maximum achievable performance. It is obvious that unless the actual degradation is shown from the very start of device life (i.e. from when freshly prepared) it gives little meaning to compare different device geometries. As an example a cell that drops linearly to half its value in 100 h and then remains within 20% of that value for 10000 h would have a T80 of 40 h if measuring form the beginning. If however the supplier quotes the performance as about half the initial performance then the T80 would be 10000 h. In the end the people applying the solar cell will decide what the real T80 is. When looking at Fig. 13 at the devices tested at 280 W m−2 and 45 ± 5 °C it is clear that the devices drop in performance during the first 5 h of operation by a factor of three. The T80 would in this case be around 1 h. The device then increases steadily for 200 h and then starts to degrade slowly. It is clear that the device exhibits quite stable performance from this point onwards and would present a T80 of around 300 h. The question is whether one could expect the user to wait the 200 h it takes to reach the specified performance. The best solution to this answer is quite possibly to solve the technical problems before attempting practical use of the modules presented in Fig. 15. In any case the use of OPV can be expected to require new ways of thinking about application and specification of performance.
Fig. 15 Operational stability of two 16 stripe 225 mm modules (280 W m−2, AM 1.5 G, 45 ± 5 °C) and two 8 stripe 250 mm modules (1000 W m−2, AM 1.5 G, 85 ± 5 °C) tested for 480 h. The two modules that were adjacent on the roll (CR 126 881/882, 280 W m−2, AM1.5G) exhibit quite similar performance and operational lifetime behavior while two modules that were at opposite ends of the roll (CR 79 26/198, 1000 W m−2, AM1.5G) exhibit some variation in operational lifetime and performance. |
Fig. 16 Examples of the scale on which modules have been prepared. Modules from about 50 × 50 cm2 (left) to 10 × 10 cm2 (right) are shown. The devices span from single stripe cells to serially connected modules where the smallest module has 2 serially connected stripes and the largest has 38 serially connected stripes. Stripe widths from 5 to 18 mm are shown. |
In Fig. 16 several of the module sizes tested are shown and they range from 50 × 50 cm2 to 10 × 10 cm2 and comprise from single stripes through 2 to 38 serially connected stripes in stripe widths from 5 to 18 mm. Experience have shown that there are no technical differences between the various modules except one of increased cost during experimentation for the larger modules. The typical webs speeds that have been involved are 1–2 m min−1 and in total Risø DTU have prepared more than 1000 square metres of OPV in 2009 alone on the small equipment described in the following.
A second example detailed the manufacture and integration of discrete electronic components in a single application for the lighting Africa initiative.48,49 The purpose of the first demonstration was to charge a battery for powering a radio at a music festival and the second demonstration was intended to provide inhabitants in developing countries with light without the need for kerosene lamps. Both demonstrations were carried out in public and the cost structure have enabled the analysis of how well the technology performs in the context of electricity production50,51 and more notably allowed for the identification of where investments in efforts to reduce the cost are best placed.48,50,51 In the case of the solar hat19 that was prepared entirely using screen printing with the consequent limitations on performance gave an electricity cost of 4500 € Wp−1. In the case of the lamp for the lighting Africa initiative a practical electricity cost of 35 € Wp−1 was achieved. When looking at the solar cells prepared in this study where the cost was further reduced through optimisation of the usage of ink and by purchasing larger quantities of materials at lower cost a learning curve can be plotted as shown in Fig. 17.
Fig. 17 Learning curve for the manufacturing cost in € Wp−1 for R2R manufacture polymer solar cells based on ProcessOne (ref. 21). The lowest achievable cost with P3HT:PCBM and ITO using ProcessOne is estimated to be around 5 € Wp−1 shown as a broken line. Note that the first point is based on the solar hat (ref. 19) which differs slightly from ProcessOne with respect to materials and printing technology. |
The lowest electricity cost that can be achieved when manufacturing modules based on ProcessOne using P3HT:PCBM on ITO is estimated to be ca. 5 € Wp−1. The lowest cost reached to date with the equipment described, the current usage of materials and the materials cost on the scale of these experiments is 8.1 € Wp−1. It should be noted that the investment in man power and processing time already contribute very little (<20%) to the total cost which is essentially limited to the cost of materials. Further reductions have to come from: elimination of the expensive components such as ITO, an increase of the processing speed (e.g. by more than a factor of 10) and possibly from lower costs for PEDOT:PSS and the active materials through significant upscaling on the part of chemical suppliers, and possibly through elimination of one adhesive layer (Table 3). It is clear that new high-performing active materials could enter Table 3 and give significantly lower cost in € Wp−1 provided that they have a similar cost to the P3HT:PCBM composite. It should also be noted that the manufacturing cost shown in Table 3 excludes the cost associated with characterisation of the final module. The cost of characterisation using the equipment described in section 2.6 is quite low and around 0.005 € per IV curve based on the power consumption of the R2R characteriser. The costs per kWh of electricity and the typical positioning time including recording of the IV curve are respectively 2.2 kW, 0.265 € and 30 s. The annealing time required for ProcessOne however requires long testing times if a particular performance is to be guaranteed. With a typical annealing time of 20 min the extra cost of characterization is 0.194 € module−1. While this is not prohibitive in terms of cost it limits the production capacity to the production speed to the annealing time. The typical total manufacturing time is in the range of 42–95 s per module. This manufacturing time is the total manufacturing time starting from a PET substrate fully covered with ITO. It is clear that the annealing time exceeds the manufacturing time by a factor of approximately 20. With the currently accessible lifetimes for OPV however the LCOE is not even close to being competitive as detailed above whereas the cost of electricity for freshly prepared cells is likely to be competitive with inorganic solar cells on a small scale. This shows that low cost manufacture of OPV with a competitive performance is possible while work on improving the operational stability must be carried out to decrease the LCOE. When looking at the capital investment in manufacturing equipment this is generally viewed as low for the OPV technology and range as low as zero in cases where existing equipment can be used. This was the case for the solar hat19 that demonstrated manufacture using existing industrial equipment and infrastructure albeit with low performance. It is of some interest to show what the capital investment in equipment is to enable manufacture on the scale described here. All the equipment described above has been summarized in Table 4. The cost for the machinery and the investment in infrastructure (in a Danish setting) are shown. The machine park quoted is enough to enable full R2R manufacture and characterisation of the polymer solar cell modules described here. The realistic production capacity with the speeds achieved here is optimistically about 100 metres per day (assuming one 8 h shift). This gives an annual production capacity of around 20000 metres equivalent to 80000 modules described in Table 3 or 50 kWp assuming a technical yield of 95%. With the ∼530 k€ investment in equipment shown in Table 4 it is unlikely that this is financially viable at an electricity cost of 8 € Wp−1 and it is almost certain that one would be required to offer the OPVs at a significantly higher cost to enable a viable business structure. This still requires that customers are available in the competitive and relatively small thin film PV market. The above assumes that the R2R characterisation is not required as this would reduce the annual throughput to about 26000 of the modules described in Table 3 assuming that the R2R characteriser spends 20 min on each module and runs 24 h a day which is realistic.
Material | Material cost/€ | Processing cost/€ | Total/€ |
---|---|---|---|
Barrier | 0.4575 | 0.03173 | 0.4892 |
Pressure sensitive adhesive | 0.1918 | 0.03173 | 0.2236 |
PET-ITO | 2.6077 | 0.21111 | 2.8188 |
ZnO | 0.0582 | 0.16667 | 0.2249 |
P3HT-PCBM | 0.4492 | 0.16667 | 0.6159 |
PEDOT:PSS (EL-P 5010) | 0.2311 | 0.16667 | 0.3978 |
Silver (PV410) | 0.4120 | 0.16667 | 0.5787 |
Total | 4.4078 | 0.9412 | 5.3491 |
Equipment | Machine cost/€ | Infrastructure/€ | Total/€ |
---|---|---|---|
a Includes both hot air dryer and UV-curing oven. b Includes edge guiding system, humidity sensors and corona treater. c Includes solar simulator (KHS1200), source meter (Keithley 2400), power meter, control electronics and pneumatic control. | |||
R2R Screen printer (Alraun)a | 126400 | 35800 | 162200 |
R2R Slot-die Coater (SCM)b | 152000 | 18500 | 170500 |
R2R Etching machine (Asia) | 42500 | 9500 | 52000 |
R2R Laminator (GM) | 35800 | 0 | 35800 |
R2R Characteriser (Risø DTU)c | 72850 | 0 | 72850 |
R2R Sheeter (GM) | 28900 | 0 | 28.900 |
Contactor (Prym) | 8000 | 0 | 8000 |
Total | 466450 | 63800 | 530250 |
It is possible to heat treat the solar cells by passage of the roll through an oven at temperatures of 120–140 °C and thus reduce the required annealing time to 1–2 min. It is however associated with some problems of deformation of the foil since the barrier material is not heat stabilized and when heating a multilayer foil where some layers deform upon heating the shape of the final modules is not predictable.
Fig. 18 Photographs of various OPV demonstration projects. The solar hat (ref. 19) is shown (top left), the lamp for the Lighting Africa initiative (top right), the mobile curtain Suntiles™ that was designed by Astrid Krogh and displayed publically at the Danish Design Centre during autumn/winter 2009/2010 (bottom left) and the grid connected OPV modules (panel size: 1 m × 1.7 m) mounted on a solar tracking station (bottom right). |
Finally the complete manufacture of a low cost flexible and very light weight LED based lamp that can be recharged by sunlight for the Lighting Africa initiative.48,49 The practical results from the demonstration projects have convincingly shown that OPVs can be manufactured for specific purposes on a reasonable scale and they work well. These experiences have also revealed that there is still a large room for improvement in terms of developing design tools such that the application is well matched to the capacity of the solar cell in addition to improving the materials and constitution of the technology. Significant problems are the mechanical stability which is a relatively unexplored area that present significant problems such as delamination at interfaces and mechanically stable electrical contacts.
This journal is © The Royal Society of Chemistry 2010 |