Matthew R.
Leyden
,
Yan
Jiang
and
Yabing
Qi
*
Energy Materials and Surface Sciences Unit (EMSS), Okinawa Institute of Science and Technology Graduate University (OIST), 1919-1 Tancha Onna-son, Okinawa 904-0495, Japan. E-mail: Yabing.Qi@OIST.jp
First published on 8th July 2016
Metal organic halide perovskites are promising materials for solar cells with a maximum certified efficiency of 22.1%. However, there are only a handful of reports on larger area modules, where efficiencies drop with increasing use of the active area. Chemical vapor deposition (CVD) is a technology used in many industrial applications demonstrating potential for scale up. We used a CVD process to fabricate MAPbI3 and FAPbI3 based solar cells with power conversion efficiencies (PCEs) up to 15.6% (MAI, 0.09 cm2) and 5 × 5 cm modules with 9.5% (FAI, 5-cell modules, total active area 8.8 cm2) and 9.0% (FAI, 6-cell modules, total active area 12 cm2). To further investigate scaling issues, we fabricated modules using an established MAPbI3 solution process, and demonstrated maximum PCEs of 18.3% (MAI, 0.1 cm2), 14.6% (MAI, 1 cm2 single cells), and 8.5% at 5 × 5 cm (MAI, 6-cell module, total active area 15.4 cm2). The solution processed cells performed better than CVD cells when comparing PCEs determined from J–V measurements, but the steady state power of solution processed solar cells decreased quickly with increasing area. This decrease in power was correlated with rapid heating of the solar cells under 1 sun illumination, with a pronounced drop in performance at the phase transition temperature of MAPbI3. In contrast, FAPbI3 CVD grown solar modules maintained much of their PCEs transitioning from J–V measurements to the steady state operating conditions (1 sun), suggesting that the FAI based CVD process may outperform MAI based solution processed modules when scaled up to practical sizes.
Chemical vapor deposition (CVD) is a process widely used in industry, and is one of the many ways demonstrated for vapor deposition of perovskites.14 Vapor deposited perovskites can be formed under high15–18 and low vacuum conditions. In this work CVD is categorized as a non-directional, low to atmospheric pressure process. There are a number of studies using CVD to fabricate perovskite solar cells, such as low vacuum single step processing,19 sequential deposition processing,20–25 and growth under atmospheric conditions.26 Perovskite modules grown by CVD have already been demonstrated with a PCE per active area of 6.2%, with an active area of 8.4 cm2.8 CVD offers potential for batch processing for higher throughput, low fabrication cost, flat uniform perovskite films over large areas, and semi-transparent solar cells.20,22 In this work we investigated the scalability of the CVD process by fabricating both single solar cells with different sizes and solar modules. We also compared these cells and modules to solution processed counterparts. Using CVD we demonstrated a small area methyl ammonium iodide (MAI) based champion cell with a PCE of 15.6% (active area = 0.09 cm2), which is the highest reported efficiency for CVD based perovskite solar cells. In addition we made formamidinium iodide (FAI) based modules on 5 × 5 cm substrates with champion PCEs per active area of 9.5% and 9.0% with active areas of 8.8 cm2 and 12 cm2, respectively. These solar modules are used as the primary reference points in this work. CVD processed modules were made with a larger active area of 15.4 cm2 and demonstrated a PCE per active area of 5.8%, but were not used as the reference points for this work. The TiO2 layer of these modules was patterned incorrectly, creating electrical shorting pathways between cells within the module that should be electrically isolated. This issue significantly limited solar module performance and complicated the comparison to other modules (discussed further in the Experimental section).
Cells and modules | Cell size (cm2) | Normalized Voc | Normalized Jsc (mA cm−2) | Fill factor | Efficiency% (steady state) |
---|---|---|---|---|---|
Champion CVD cell (MAI) | 0.09 | 1.06 | 21.7 | 68 | 15.6 |
Champion solution cell (MAI) | 0.10 | 1.07 | 22.4 | 76 | 18.3 |
Champion solution cell (MAI) | 1.0 | 1.12 | 20.5 | 68 | 14.6 (13.7) |
Champion CVD cell (FAI) | 2.0 | 1.02 | 19.5 | 53 | 10.4 |
Average CVD cell (FAI, spiro) | 2 | 0.97 ± 0.05 | 17.1 ± 1.3 | 47 ± 6 | 7.8 ± 1.3 |
Average CVD cell (FAI, P3HT) | 2 | 0.93 ± 0.03 | 20.3 ± 1.1 | 41 ± 3 | 7.8 ± 0.9 |
Champion CVD module (FAI) | 8.8 | 0.98 | 16.9 | 57 | 9.5 (7.6) |
Champion CVD module (FAI) | 12.0 | 0.94 | 17.8 | 54 | 9.0 (8.8) |
Champ. solution module (MAI) | 12.0 | 0.84 | 19.9 | 43 | 7.1 (7.0) |
Champ. solution module (MAI) | 15.4 | 0.98 | 19.0 | 46 | 8.5 (4.0) |
CVD module (FAI) | 15.4 | 0.77 | 15.2 | 49 | 5.8 |
When scaling to larger areas, the most pronounced drop in performance resulted from the reduction in fill factor and short circuit current (Jsc). The open circuit voltage (Voc) also decreases slightly as the area is increased, but not as significantly as the fill factor and Jsc. The decrease in the fill factor can be explained by an increase in series resistance due to a longer transport distance through fluorine doped tin oxide (FTO), or poor contact between cells. This is an important issue, and can be solved with optimized module fabrication and design. For example, fabricating narrower cells reduces the travel distance through the FTO, and therefore lowers series resistance. The decrease in Jsc is more intrinsic to the perovskite growth process and likely reflects the average performance. Indeed, we see that the average Jsc for the champion modules (16.9 mA cm−2 and 17.8 mA cm−2) is closer to the average of 2 cm2 cells (17.1 mA cm−2) than to the champion 2 cm2 cell (19.5 mA cm−2). We also demonstrated that poly(3-hexylthiophene-2,5-diyl) (P3HT) can be used successfully as a hole transport layer for large area cells with comparable efficiency.
The results in Table 1 suggest that the solution processed cells have higher nominal PCEs (determined from conventional J–V measurements) when comparing similar area devices. However, this trend changed after switching from conventional J–V measurements to steady state operating conditions at the maximum power point. The steady state power is lower than the peak power reported by J–V curves for both CVD and solution methods, but the difference is more pronounced in the solution processed modules.
A previous report demonstrated an FAI based cell prepared by CVD with an active area of 0.04 cm2 to have a nominal PCE of 14.2% and a steady state power of 11.3%.25 This difference between power measured by J–V curves and power measured at the steady state can become increasingly significant as areas become larger. To illustrate such an area dependence, we compared the champion CVD grown solar modules to a solution processed single solar cell and then to a solution processed solar module with a larger active area (Fig. 2). The solution processed champion cell had an active area of 1 cm2 and a nominal PCE of 14.6%. The steady state PCE of this cell was 13.7% and maintained much of its PCE after many hours (approximately 38 h) of operation (Fig. S3†). The peak steady state PCE of the 12 cm2 active area CVD module was 8.8% measured after 30 h of operation. This 40 h steady state measurement was periodically interrupted to measure J–V curves and redefine the maximum power point (Fig. 2b). In contrast to the solution processed cells or CVD modules the steady state power of the solution processed modules slowly decreased with time (Fig. 2a). From this figure we can see that the solution module achieved a peak power of ∼7.9%, but decayed to 7% after 3 hours of continuous operation (6.5%, after 8 h). Both 12 cm2 modules are possibly limited to about ∼9% by series resistance (discussed in the ESI†). The PCE decay for solution processed modules was even more rapid for cells with high use of active area, possible reasons for which are discussed below.
A larger active area module using a 5 × 5 cm substrate and an active area of 15.4 cm2 was made by the Lewis base adduct solution method and had a nominal PCE of 8.5% (nominal power = 131 mW), but a steady state PCE of only 4% (steady state power = 62 mW). This cell was the best of 9 modules using the same process. This module compared poorly to the lower active area champion CVD grown modules with steady state power 66 mW (8.8 cm2 steady state PCE = 7.6%) and 106 mW (12 cm2 steady state PCE = 8.8%), as shown in Fig. 2b. One possible cause for this drop in efficiency is related to substrate heating, which is discussed later. The decrease could also be related to difficulties in applying the Lewis base adduct solution method on larger substrates. The champion solution processed 1 cm2 sample has the same substrate size (15 mm × 15 mm; see Fig. 1c and S2† for specific dimensions) and deposition conditions as the highest performing cell in this work (18.3% PCE, 0.1 cm2). This is mentioned to highlight processing differences between small cells (active area ≤ 1 cm2) and larger modules.
This lower total power despite having a larger active area was surprising, and raised concerns that perhaps substrate heating was limiting the performance.28 The MAI based perovskite will undergo a phase transition at elevated temperatures (54–57 °C) from the tetragonal to the cubic phase,29,30 and this phase transition is believed to cause reduced performance.28 We observed that the solution processed cell heated up to 54 °C within a few minutes of testing under 1 sun illumination and rapidly dropped in PCE, but would regain efficiency when removed from the solar simulator and allowed to cool. However, this behaviour is not completely due to the phase transition of the perovskite film (Fig. S4†). Decay in performance is observed from heating at temperatures below the transition temperature as well, and loss in performance cannot be exclusively explained by the transition. It is possible that heating accelerates a decay process that is significant in solution processed large-area modules, but less significant in small-area cells. Shunt pathways are reported to cause degradation,31 and the use of large modules provides more possible pathways than smaller substrates.
To demonstrate more clearly the impact of the phase transition, the intensity of the solar simulator was increased to 1.5 sun, allowing the module to rapidly heat beyond the transition temperature (Fig. 3a). The temperature was monitored in situ by using a thermocouple placed on the underside at the center of the module. A phase transition was indeed observed at ∼52 °C by a latent heating rate and an increase in the decay rate of the steady state current. Note that there may be a lag between the temperature of the thermocouple and that of the perovskite film, explaining the lower than expected transition temperature. Operating in the cubic phase not only caused a rapid decay in performance, but also appeared to increase the dark current, evident from the large negative power when the light was turned off. A similar increase in the dark current was also observed when operating at 1 sun intensity (Fig. S4†). The dark current partially recovered to a lower value if the cell was allowed to cool and given sufficient time before the next measurement. Based on this observation it is possible that the perovskite slowly returns to the tetragonal phase, allowing the cell to partially recover performance. This residual material in the cubic phase may explain why the second phase transition in Fig. 3a is less distinct.
When the FAI based CVD grown perovskite solar module was measured at 1.5 sun (Fig. 3b), there was no evidence of performance decay with increasing temperature. This is reasonable, because there is no phase transition for FAPbI3 in the temperature range of solar cell operation. However, the FAI based CVD grown perovskite solar module had a pronounced dependence on light soaking, which is discussed in more detail below. These comparative results suggest that the CVD growth of the FAI based perovskite maintains a better steady state PCE going from small to large areas than the MAI based Lewis base adduct solution method. It is possible that the solution process may require further optimization for large scale substrates, such as improving the module patterning process to reduce the opportunities for shunt pathways, modification of anti-solvent deposition conditions, or use of FAI instead of MAI.32
To decouple the effects of temperature from possible light induced effects, an additional control experiment was performed on the MAPbI3 solution processed module by recording dark current during heating of the module on a hotplate (Fig. 3c). Here the hotplate was ramped up to 62 °C at a constant rate of 2°C min−1, at the approximate maximum power point (3.9 V). In this figure we can see that the dark current increases with increasing temperature, with a sharp increase in the current and noise at the phase transition temperature (∼57 °C). After 5 min at 62 °C the module is removed from the hotplate and allowed to cool to room temperature, and the dark current remains at a high level, which is similar to that in Fig. 3a. We see from Fig. 3c that the dark current stops rising once the cell is removed from heat. This suggests that the increasing dark current is not due to voltage induced ion migration alone. However ion migration could be accelerated by heating, or by the transition to the cubic phase.
In Fig. 4a cells from the 8.8 cm2 champion CVD module are added in series one at a time to show the additive nature in Voc. To our surprise, the weak performing Jsc of cell #1 (15.9 mA cm−2) did not seem to rigidly limit the Jsc of the entire module (16.9 mA cm−2). This observation led us to consider whether possibly perovskite modules are tolerant to partial shade, similar to reports on electrolyte based dye sensitized solar cells. Dye sensitized solar cells can be modelled as a two-diode system, where the Jsc of a module may be minimally impacted by a weak performer.33 To test this hypothesis, a second perovskite module was used and 3 of its 5 cells were covered with semitransparent tape. Fig. 4b shows the cells, uncovered and with 1 layer and 2 layers of tape. Both Jsc and Voc are strongly impacted by the shadowing of cells in a module. From these partial shade experiments it is clear that perovskite modules are not especially tolerant to partial shade, and behave more like silicon solar cells,34 than dye sensitized solar cells. This means that perovskite modules should have bypass diodes, allowing them to pass current through a cell with a lower photocurrent, and subsequently tolerate partial shade. An alternative explanation of the behavior seen in Fig. 4a can be given by a combination of a low shunt resistance and low recombination resistance of the weak performing cell 1. A low shunt resistance can pass the excess current supplied by the higher performing cells.
Perovskite solar cells grown by CVD exhibit a pronounced sensitivity and improved performance from light soaking similar to many other reports on perovskite solar cells. This can be explained by light induced self-poling of the perovskite layer, causing migration of ions within the perovskite film.35 Light induced poling appears to be present in the solution processed MAI cells as well as FAI based CVD processed cells, but the time scale of solution processed cells seems much shorter. Differences could be due to the speed of ion migration in the material (FAPbI3vs. MAPbI3), and how far the ions migrate, and the structure of the film. Other possible explanations for this behavior include UV illumination creating photoactive surface states, or shifting of the conduction band edge in TiO2.36 The effects of light soaking are more noticeable as the area is increased. Small cells often reached full efficiency after minutes of light soaking, while larger cells could take up to an hour or more before reaching a stable efficiency. Light soaking impacted dark current measurements, indicating that the effect of light soaking is long lasting and is present even after the light is turned off (Fig. 3b and 4c).
Fig. 5 shows J–V curves from champion small area cells for both CVD and solution processing, in addition to EQE curves showing approximate agreement between the integrated current and Jsc measured from J–V curves. However, as shown in the previous figures, J–V curves only offer an indication of how a cell may perform during steady state conditions. Also, larger area modules may behave differently from their small area counterparts.
These titanium oxide layers were then patterned by photolithography and etched with a timed dry etch ICP process (Oxford instruments, 6 min, 100 w RF, 200 w ICP, 16 sccm CF4, 4 sccm Ar, at a pressure of 40 mTorr), largely removing the TiO2, while leaving a continuous layer of underlying FTO. The photoresist is then removed by using boiling acetone, followed by UV-ozone treatment. A photograph of the resulting film can be seen in the ESI (Fig. S5†). This etching process was sufficient to eliminate the contact between cells caused by compact layer TiO2 in the small modules, but for large modules with narrow spacing between cells there remained conductive paths (100 ± 80 kohm). At the time of fabrication, this was thought to be sufficient insulation, but this conductivity appeared to limit the Voc obtained in CVD grown perovskite large modules. This problem was addressed by increasing the spacing between FTO cells to 1 mm (12 cm2 active area). With increasing spacing it is possible to pattern TiO2 deposition with a shadow mask (Fig. S5†), eliminating the need for photolithography and plasma processing.
To fabricate the mesoporous layer for solution processed cells and modules 0.15 M titanium diisopropoxide dis(acetylacetonate) (Sigma Aldrich) in butanol was spun onto substrates at 2800 rpm and dried at ∼120 °C, followed by a 0.1 g mL−1 solution of titanium dioxide nanoparticles in butanol at 2000 rpm. The film is then annealed at 550 °C. These substrates were then treated with 20 mM TiCl4 in DI water for 10 min at 70 °C, followed by a second anneal at 500 °C. These titanium oxide layers were then patterned by photolithography and etched with a timed ICP process, in the same way as performed for compact TIO2 by spray pyrolysis. No conductive path was observed between cells in the large modules after patterning with the ICP, suggesting that the TiO2 formed by solution deposition is less dense and continuous than by spray pyrolysis. This could also be related to dark current measurements and the formation of shunt pathways, which show more dark current than cells made by spray pyrolysis (Fig. 3 and S6†). For this reason, the compact layer of the 12 cm2 active area solution module was made by spray pyrolysis.
External quantum efficiency (EQE) measurements were performed on an Oriel IQE 200 in DC mode. AC mode EQE measurements at 30 Hz were lower than expected and the integrated current did not match Jsc. This is likely related to light soaking effects mentioned in Fig. 4c. Continuous light may be needed for perovskite solar cells to function at full efficiency. The AC mode can still be used with some success at high frequencies (Fig. S6†).
Footnote |
† Electronic supplementary information (ESI) available. See DOI: 10.1039/c6ta04267h |
This journal is © The Royal Society of Chemistry 2016 |